ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: wait on ublk_dev_ready() instead of ub->completion
ub->completion is only re-armed by a successful START_USER_RECOVERY. If
the ublk server sends END_USER_RECOVERY without one - e.g. its START
failed with -EBUSY and the error was ignored - the wait is satisfied by
the stale completion of the previous recovery cycle, and the device is
marked LIVE and the requeue list kicked while the FETCH stream is still
running and ubq->canceling is still set. The kick redispatches a
previously requeued request, __ublk_queue_rq_common() sees ->canceling
and parks it again via __ublk_abort_rq(), and after the last FETCH
clears ->canceling nothing ever kicks the requeue list again: the
request is stranded there while holding its tag. If it is the flush
machinery's flush_rq, every subsequent fsync piles up in uninterruptible
sleep and teardown hangs on tag draining. This matches a report of a
lost PREFLUSH with ext4 on top of ublk after daemon crash recovery.
ub->completion is an edge-triggered latch used as a proxy for the level
condition "every queue has fetched all I/O commands", which can regress
(F_BATCH's UNPREP, daemon death) and whose re-arm can be skipped. Drop
it and wait on the real condition instead: the new helper
ublk_wait_dev_ready_and_lock() waits on ublk_dev_ready() via
wait_var_event_interruptible(), woken from ublk_mark_io_ready(), then
re-checks it under ub->mutex, waiting again on regression, and returns
with the mutex held and readiness guaranteed.
Readiness becomes true in the same ub->mutex critical section that
clears the last queue's ->canceling, so END_USER_RECOVERY marks the
device LIVE and kicks the requeue list strictly after ->canceling
clears. The wait stays interruptible, so a server whose daemon died can
still be signalled out. For ublk_ctrl_start_dev() this replaces the
fail-fast -EINVAL on an F_BATCH ready->UNPREP regression with waiting
until the device is ready again.
Reported-by: George Salisbury <gsalisbury@apnic.net>
Fixes: 728cbac5fe21 ("ublk: move device reset into ublk_ch_release()")
Cc: stable@vger.kernel.org
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Link: https://patch.msgid.link/20260719134540.120269-1-tom.leiming@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2026-07-19 08:45:40 -05:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-01 00:57:35 +00:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-13 22:07:11 +08:00
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
2023-05-19 14:50:27 +08:00
|
|
|
|
2026-01-16 22:18:39 +08:00
|
|
|
|
2026-01-08 02:19:31 -07:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:31 -07:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
|
|
|
|
|
2024-07-24 22:33:11 +08:00
|
|
|
|
|
|
|
|
|
2025-04-21 13:59:50 +03:00
|
|
|
|
2025-05-23 00:35:20 +08:00
|
|
|
|
2026-01-13 00:05:01 +02:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
2024-07-24 22:33:11 +08:00
|
|
|
|
2026-04-09 21:30:13 +08:00
|
|
|
|
|
|
|
|
|
2024-07-24 22:33:11 +08:00
|
|
|
|
2025-02-27 14:39:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-22 18:38:17 +08:00
|
|
|
|
2022-07-28 20:39:16 +08:00
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:14 +08:00
|
|
|
|
2022-09-23 23:39:17 +08:00
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
2023-04-18 21:18:10 +08:00
|
|
|
|
ublk: support user copy
Currently copy between io request buffer(pages) and userspace buffer is
done inside ublk_map_io() or ublk_unmap_io(). This way performs very
well in case of pre-allocated userspace io buffer.
For dynamically allocated or external userspace backend io buffer,
UBLK_F_NEED_GET_DATA is added for ublk server to provide buffer by one
extra command communication for WRITE request. For READ, userspace
simply provides buffer, but can't know when the buffer is done[1].
Add UBLK_F_USER_COPY by moving io data copy out of kernel by providing
read()/write() on /dev/ublkcN, and simply let ublk server do the io
data copy. This way makes both side cleaner, the cost is that one extra
syscall for copy io data between request and backend buffer.
With UBLK_F_USER_COPY, it actually becomes possible to run per-io zero
copy now, such as, only do zero copy for big size IO, so it can be
thought as one prep patch for supporting zero copy. Meantime zero copy
still needs to expose read()/write() buffer for some corner case, such
as passthrough IO.
[1] READ buffer in UBLK_F_NEED_GET_DATA
https://lore.kernel.org/linux-block/116d8a56-0881-56d3-9bcc-78ff3e1dc4e5@linux.alibaba.com/T/#m23bd4b8634c0a054e6797063167b469949a247bb
ublksrv loop usercopy code:
https://github.com/ming1/ubdsrv/commits/usercopy
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230519065030.351216-8-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-05-19 14:50:30 +08:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
2024-10-07 12:24:17 -06:00
|
|
|
|
2025-04-21 13:59:50 +03:00
|
|
|
|
2025-05-20 12:54:33 +08:00
|
|
|
|
2025-05-23 00:35:20 +08:00
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
2025-06-20 09:10:03 -06:00
|
|
|
|
2026-01-08 02:19:38 -07:00
|
|
|
|
2026-01-13 00:05:01 +02:00
|
|
|
|
2026-01-16 22:18:44 +08:00
|
|
|
|
2026-01-30 00:19:51 +08:00
|
|
|
|
2026-03-31 23:31:54 +08:00
|
|
|
|
|
|
|
|
|
2022-07-22 18:38:17 +08:00
|
|
|
|
2024-10-07 12:24:14 -06:00
|
|
|
|
2024-10-07 12:24:17 -06:00
|
|
|
|
|
|
|
|
|
2024-10-07 12:24:14 -06:00
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
2025-02-27 18:37:07 +08:00
|
|
|
|
2026-01-08 02:19:31 -07:00
|
|
|
|
|
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
2026-01-16 22:18:36 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-28 12:04:11 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-10-29 09:04:31 +08:00
|
|
|
|
2025-05-20 12:54:33 +08:00
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:36 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:46 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-28 20:39:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-05-20 12:54:32 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-10-09 17:33:18 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:10:05 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-21 09:58:25 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-11-21 09:58:25 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-30 16:52:34 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
|
|
|
|
|
2025-06-20 09:09:56 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:10:05 -06:00
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:39 -07:00
|
|
|
|
2025-06-20 09:10:05 -06:00
|
|
|
|
2025-06-20 09:09:56 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:10:05 -06:00
|
|
|
|
|
|
|
|
|
2025-06-20 09:09:56 -06:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:37 +08:00
|
|
|
|
2025-06-20 09:10:08 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-13 22:07:11 +08:00
|
|
|
|
2025-03-28 13:42:29 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2022-09-23 23:39:16 +08:00
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
2024-10-07 12:24:17 -06:00
|
|
|
|
2023-10-09 17:33:18 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
2026-01-16 22:18:39 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:39 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:39 +08:00
|
|
|
|
|
|
|
|
|
2025-11-01 21:31:18 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-26 17:12:23 +03:00
|
|
|
|
|
|
|
|
|
2023-02-07 23:07:00 +08:00
|
|
|
|
2022-07-21 15:09:12 +02:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-10-09 17:33:19 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
2025-08-08 09:52:15 -06:00
|
|
|
|
ublk: speed up ublk server exit handling
Recently, we've observed a few cases where a ublk server is able to
complete restart more quickly than the driver can process the exit of
the previous ublk server. The new ublk server comes up, attempts
recovery of the preexisting ublk devices, and observes them still in
state UBLK_S_DEV_LIVE. While this is possible due to the asynchronous
nature of io_uring cleanup and should therefore be handled properly in
the ublk server, it is still preferable to make ublk server exit
handling faster if possible, as we should strive for it to not be a
limiting factor in how fast a ublk server can restart and provide
service again.
Analysis of the issue showed that the vast majority of the time spent in
handling the ublk server exit was in calls to blk_mq_quiesce_queue,
which is essentially just a (relatively expensive) call to
synchronize_rcu. The ublk server exit path currently issues an
unnecessarily large number of calls to blk_mq_quiesce_queue, for two
reasons:
1. It tries to call blk_mq_quiesce_queue once per ublk_queue. However,
blk_mq_quiesce_queue targets the request_queue of the underlying ublk
device, of which there is only one. So the number of calls is larger
than necessary by a factor of nr_hw_queues.
2. In practice, it calls blk_mq_quiesce_queue _more_ than once per
ublk_queue. This is because of a data race where we read
ubq->canceling without any locking when deciding if we should call
ublk_start_cancel. It is thus possible for two calls to
ublk_uring_cmd_cancel_fn against the same ublk_queue to both call
ublk_start_cancel against the same ublk_queue.
Fix this by making the "canceling" flag a per-device state. This
actually matches the existing code better, as there are several places
where the flag is set or cleared for all queues simultaneously, and
there is the general expectation that cancellation corresponds with ublk
server exit. This per-device canceling flag is then checked under a
(new) lock (addressing the data race (2) above), and the queue is only
quiesced if it is cleared (addressing (1) above). The result is just one
call to blk_mq_quiesce_queue per ublk device.
To minimize the number of cache lines that are accessed in the hot path,
the per-queue canceling flag is kept. The values of the per-device
canceling flag and all per-queue canceling flags should always match.
In our setup, where one ublk server handles I/O for 128 ublk devices,
each having 24 hardware queues of depth 4096, here are the results
before and after this patch, where teardown time is measured from the
first call to io_ring_ctx_wait_and_kill to the return from the last
ublk_ch_release:
before after
number of calls to blk_mq_quiesce_queue: 6469 256
teardown time: 11.14s 2.44s
There are still some potential optimizations here, but this takes care
of a big chunk of the ublk server exit handling delay.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250703-ublk_too_many_quiesce-v2-1-3527b5339eeb@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-07-03 23:41:07 -06:00
|
|
|
|
|
|
|
|
|
2025-07-13 22:33:56 +08:00
|
|
|
|
2025-08-27 20:15:59 +08:00
|
|
|
|
2025-12-23 11:27:40 +08:00
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
2026-01-13 00:05:01 +02:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:16 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-05-20 12:54:32 +08:00
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
2026-03-31 23:31:53 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
2025-02-27 14:39:15 -08:00
|
|
|
|
2026-01-08 02:19:36 -07:00
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-30 16:52:29 -06:00
|
|
|
|
2026-01-16 22:18:34 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:44 +08:00
|
|
|
|
2026-01-16 22:18:34 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:44 +08:00
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:39 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-30 16:52:29 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:30 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:54 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:53 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:54 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:30 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-21 09:58:27 +08:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:31 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-05-20 14:36:53 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-05-20 14:36:54 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
2024-08-12 09:36:24 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-04-08 10:41:20 +09:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-02-15 08:02:59 +01:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-05 06:22:38 +09:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-08-12 09:36:24 +08:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-08-12 09:36:24 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
2025-05-07 14:04:29 +02:00
|
|
|
|
2024-08-12 09:36:24 +08:00
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-08-12 09:36:24 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-05 06:22:38 +09:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-08-12 09:36:24 +08:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-10 20:43:26 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
2024-08-12 09:36:24 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
2026-05-20 14:36:54 -06:00
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-10 20:43:26 +08:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-05-20 14:36:54 -06:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-02-15 08:02:59 +01:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-10 16:48:36 +08:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:53 -06:00
|
|
|
|
2026-01-16 22:18:46 +08:00
|
|
|
|
2023-05-19 14:50:27 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2023-06-20 20:01:32 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
treewide: Replace kmalloc with kmalloc_obj for non-scalar types
This is the result of running the Coccinelle script from
scripts/coccinelle/api/kmalloc_objs.cocci. The script is designed to
avoid scalar types (which need careful case-by-case checking), and
instead replace kmalloc-family calls that allocate struct or union
object instances:
Single allocations: kmalloc(sizeof(TYPE), ...)
are replaced with: kmalloc_obj(TYPE, ...)
Array allocations: kmalloc_array(COUNT, sizeof(TYPE), ...)
are replaced with: kmalloc_objs(TYPE, COUNT, ...)
Flex array allocations: kmalloc(struct_size(PTR, FAM, COUNT), ...)
are replaced with: kmalloc_flex(*PTR, FAM, COUNT, ...)
(where TYPE may also be *VAR)
The resulting allocations no longer return "void *", instead returning
"TYPE *".
Signed-off-by: Kees Cook <kees@kernel.org>
2026-02-20 23:49:23 -08:00
|
|
|
|
2025-02-28 21:31:48 -07:00
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
2026-01-30 00:19:50 +08:00
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 21:31:48 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:10 +08:00
|
|
|
|
2025-02-28 21:31:48 -07:00
|
|
|
|
2023-01-06 12:17:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 21:31:48 -07:00
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:10 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:31 -07:00
|
|
|
|
|
|
|
|
|
2026-03-13 07:47:01 -07:00
|
|
|
|
2026-01-08 02:19:31 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-04-06 20:40:59 +08:00
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
2026-05-06 16:22:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-30 17:27:50 +08:00
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
2026-05-10 22:48:43 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:09 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-27 18:37:07 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-27 17:51:15 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:31 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-02-15 08:02:59 +01:00
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
2024-02-15 08:02:59 +01:00
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
2025-03-27 17:51:13 +08:00
|
|
|
|
|
|
|
|
|
2025-05-20 12:54:32 +08:00
|
|
|
|
|
|
|
|
|
2025-03-27 17:51:13 +08:00
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:40 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:27 +08:00
|
|
|
|
|
|
|
|
|
ublk: support user copy
Currently copy between io request buffer(pages) and userspace buffer is
done inside ublk_map_io() or ublk_unmap_io(). This way performs very
well in case of pre-allocated userspace io buffer.
For dynamically allocated or external userspace backend io buffer,
UBLK_F_NEED_GET_DATA is added for ublk server to provide buffer by one
extra command communication for WRITE request. For READ, userspace
simply provides buffer, but can't know when the buffer is done[1].
Add UBLK_F_USER_COPY by moving io data copy out of kernel by providing
read()/write() on /dev/ublkcN, and simply let ublk server do the io
data copy. This way makes both side cleaner, the cost is that one extra
syscall for copy io data between request and backend buffer.
With UBLK_F_USER_COPY, it actually becomes possible to run per-io zero
copy now, such as, only do zero copy for big size IO, so it can be
thought as one prep patch for supporting zero copy. Meantime zero copy
still needs to expose read()/write() buffer for some corner case, such
as passthrough IO.
[1] READ buffer in UBLK_F_NEED_GET_DATA
https://lore.kernel.org/linux-block/116d8a56-0881-56d3-9bcc-78ff3e1dc4e5@linux.alibaba.com/T/#m23bd4b8634c0a054e6797063167b469949a247bb
ublksrv loop usercopy code:
https://github.com/ming1/ubdsrv/commits/usercopy
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230519065030.351216-8-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-05-19 14:50:30 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-29 10:29:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-05-20 12:54:32 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: support user copy
Currently copy between io request buffer(pages) and userspace buffer is
done inside ublk_map_io() or ublk_unmap_io(). This way performs very
well in case of pre-allocated userspace io buffer.
For dynamically allocated or external userspace backend io buffer,
UBLK_F_NEED_GET_DATA is added for ublk server to provide buffer by one
extra command communication for WRITE request. For READ, userspace
simply provides buffer, but can't know when the buffer is done[1].
Add UBLK_F_USER_COPY by moving io data copy out of kernel by providing
read()/write() on /dev/ublkcN, and simply let ublk server do the io
data copy. This way makes both side cleaner, the cost is that one extra
syscall for copy io data between request and backend buffer.
With UBLK_F_USER_COPY, it actually becomes possible to run per-io zero
copy now, such as, only do zero copy for big size IO, so it can be
thought as one prep patch for supporting zero copy. Meantime zero copy
still needs to expose read()/write() buffer for some corner case, such
as passthrough IO.
[1] READ buffer in UBLK_F_NEED_GET_DATA
https://lore.kernel.org/linux-block/116d8a56-0881-56d3-9bcc-78ff3e1dc4e5@linux.alibaba.com/T/#m23bd4b8634c0a054e6797063167b469949a247bb
ublksrv loop usercopy code:
https://github.com/ming1/ubdsrv/commits/usercopy
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230519065030.351216-8-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-05-19 14:50:30 +08:00
|
|
|
|
2025-05-20 12:54:32 +08:00
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:27 +08:00
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:40 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-24 12:27:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:27 +08:00
|
|
|
|
2025-06-20 09:09:56 -06:00
|
|
|
|
2023-05-19 14:50:27 +08:00
|
|
|
|
2025-06-20 09:09:56 -06:00
|
|
|
|
2025-06-20 09:10:05 -06:00
|
|
|
|
2023-05-19 14:50:27 +08:00
|
|
|
|
|
|
|
|
|
2025-06-20 09:10:07 -06:00
|
|
|
|
2023-05-19 14:50:27 +08:00
|
|
|
|
2025-06-20 09:10:07 -06:00
|
|
|
|
2023-05-19 14:50:27 +08:00
|
|
|
|
|
|
|
|
|
2025-06-20 09:10:07 -06:00
|
|
|
|
2023-05-19 14:50:27 +08:00
|
|
|
|
2025-09-17 19:49:53 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:46 +08:00
|
|
|
|
2022-07-13 22:07:11 +08:00
|
|
|
|
|
|
|
|
|
2025-07-15 09:42:43 -06:00
|
|
|
|
2025-06-20 09:10:05 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-13 22:34:04 +08:00
|
|
|
|
2022-07-13 22:07:11 +08:00
|
|
|
|
|
|
|
|
|
2022-07-28 20:39:16 +08:00
|
|
|
|
|
|
|
|
|
2023-03-30 19:36:23 +08:00
|
|
|
|
2022-07-28 20:39:16 +08:00
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:40 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-02-23 15:55:38 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-02-23 15:55:38 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2025-03-28 13:42:29 -06:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-11-11 19:07:18 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:38 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-09-17 19:49:38 -06:00
|
|
|
|
2024-11-11 19:07:18 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2024-10-07 12:24:15 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:17 +08:00
|
|
|
|
2024-10-07 12:24:15 -06:00
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:17 +08:00
|
|
|
|
|
|
|
|
|
2024-10-07 12:24:15 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-10-07 12:24:17 -06:00
|
|
|
|
|
|
|
|
|
2024-10-07 12:24:15 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:14 +08:00
|
|
|
|
2024-10-07 12:24:17 -06:00
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:14 +08:00
|
|
|
|
|
|
|
|
|
2024-10-07 12:24:15 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:14 +08:00
|
|
|
|
2024-10-07 12:24:15 -06:00
|
|
|
|
2022-09-23 23:39:14 +08:00
|
|
|
|
|
|
|
|
|
2024-10-07 12:24:17 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-21 15:09:16 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-02-23 15:55:38 +08:00
|
|
|
|
2022-07-21 15:09:16 +02:00
|
|
|
|
|
|
|
|
|
2023-01-31 12:04:46 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-06-08 13:02:55 +02:00
|
|
|
|
2023-01-31 12:04:46 +08:00
|
|
|
|
2023-06-08 13:02:36 +02:00
|
|
|
|
2023-01-31 12:04:46 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-13 00:05:01 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-31 12:04:46 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2023-01-31 12:04:46 +08:00
|
|
|
|
2022-07-21 15:09:16 +02:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:33 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-16 07:02:46 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:33 -07:00
|
|
|
|
2026-04-16 07:02:46 +08:00
|
|
|
|
2026-01-08 02:19:33 -07:00
|
|
|
|
2026-04-16 07:02:46 +08:00
|
|
|
|
2026-01-08 02:19:33 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:26 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:28 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-11-06 10:16:47 -07:00
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:26 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-11-06 10:16:47 -07:00
|
|
|
|
2026-01-08 02:19:33 -07:00
|
|
|
|
2025-11-06 10:16:46 -07:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:37 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-03-30 19:36:22 +08:00
|
|
|
|
|
|
|
|
|
2026-05-13 15:18:46 -06:00
|
|
|
|
2023-03-30 19:36:22 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-05-13 15:18:46 -06:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
2023-03-30 19:36:22 +08:00
|
|
|
|
|
|
|
|
|
2025-11-07 21:23:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2023-03-30 19:36:21 +08:00
|
|
|
|
2025-03-27 17:51:13 +08:00
|
|
|
|
ublk: support user copy
Currently copy between io request buffer(pages) and userspace buffer is
done inside ublk_map_io() or ublk_unmap_io(). This way performs very
well in case of pre-allocated userspace io buffer.
For dynamically allocated or external userspace backend io buffer,
UBLK_F_NEED_GET_DATA is added for ublk server to provide buffer by one
extra command communication for WRITE request. For READ, userspace
simply provides buffer, but can't know when the buffer is done[1].
Add UBLK_F_USER_COPY by moving io data copy out of kernel by providing
read()/write() on /dev/ublkcN, and simply let ublk server do the io
data copy. This way makes both side cleaner, the cost is that one extra
syscall for copy io data between request and backend buffer.
With UBLK_F_USER_COPY, it actually becomes possible to run per-io zero
copy now, such as, only do zero copy for big size IO, so it can be
thought as one prep patch for supporting zero copy. Meantime zero copy
still needs to expose read()/write() buffer for some corner case, such
as passthrough IO.
[1] READ buffer in UBLK_F_NEED_GET_DATA
https://lore.kernel.org/linux-block/116d8a56-0881-56d3-9bcc-78ff3e1dc4e5@linux.alibaba.com/T/#m23bd4b8634c0a054e6797063167b469949a247bb
ublksrv loop usercopy code:
https://github.com/ming1/ubdsrv/commits/usercopy
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230519065030.351216-8-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-05-19 14:50:30 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-03-30 19:36:22 +08:00
|
|
|
|
2023-05-19 14:50:26 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-11-21 09:58:25 +08:00
|
|
|
|
2023-05-19 14:50:28 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-07 21:23:16 -07:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-07-13 22:34:05 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:53 -06:00
|
|
|
|
ublk: support user copy
Currently copy between io request buffer(pages) and userspace buffer is
done inside ublk_map_io() or ublk_unmap_io(). This way performs very
well in case of pre-allocated userspace io buffer.
For dynamically allocated or external userspace backend io buffer,
UBLK_F_NEED_GET_DATA is added for ublk server to provide buffer by one
extra command communication for WRITE request. For READ, userspace
simply provides buffer, but can't know when the buffer is done[1].
Add UBLK_F_USER_COPY by moving io data copy out of kernel by providing
read()/write() on /dev/ublkcN, and simply let ublk server do the io
data copy. This way makes both side cleaner, the cost is that one extra
syscall for copy io data between request and backend buffer.
With UBLK_F_USER_COPY, it actually becomes possible to run per-io zero
copy now, such as, only do zero copy for big size IO, so it can be
thought as one prep patch for supporting zero copy. Meantime zero copy
still needs to expose read()/write() buffer for some corner case, such
as passthrough IO.
[1] READ buffer in UBLK_F_NEED_GET_DATA
https://lore.kernel.org/linux-block/116d8a56-0881-56d3-9bcc-78ff3e1dc4e5@linux.alibaba.com/T/#m23bd4b8634c0a054e6797063167b469949a247bb
ublksrv loop usercopy code:
https://github.com/ming1/ubdsrv/commits/usercopy
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230519065030.351216-8-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-05-19 14:50:30 +08:00
|
|
|
|
|
|
|
|
|
2023-03-30 19:36:22 +08:00
|
|
|
|
2023-05-19 14:50:26 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-21 09:58:25 +08:00
|
|
|
|
2023-05-19 14:50:28 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-16 17:53:44 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-05-20 14:36:54 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-27 17:51:14 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2025-12-12 22:34:15 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-09-17 19:49:53 -06:00
|
|
|
|
2026-01-16 22:18:46 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2023-03-30 19:36:20 +08:00
|
|
|
|
2025-12-12 22:34:15 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-03-30 19:36:20 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-02-07 15:08:38 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
2023-03-30 19:36:20 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2026-03-31 23:31:53 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-09-17 19:49:53 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-12-12 22:34:15 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2026-01-16 22:18:46 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2026-01-16 22:18:46 +08:00
|
|
|
|
2023-03-30 19:36:20 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-12-12 22:34:15 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2025-06-24 18:41:21 +08:00
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:27 +08:00
|
|
|
|
2025-04-30 16:52:34 -06:00
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:27 +08:00
|
|
|
|
2022-07-28 20:39:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-30 16:52:34 -06:00
|
|
|
|
2025-06-24 18:41:21 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-30 16:52:34 -06:00
|
|
|
|
2022-07-28 20:39:16 +08:00
|
|
|
|
2025-09-22 11:02:31 -06:00
|
|
|
|
2022-07-28 20:39:16 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:15 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-10-07 12:24:15 -06:00
|
|
|
|
2022-09-23 23:39:15 +08:00
|
|
|
|
|
|
|
|
|
2025-12-12 22:34:15 +08:00
|
|
|
|
2022-09-23 23:39:15 +08:00
|
|
|
|
|
|
|
|
|
2025-06-20 09:09:56 -06:00
|
|
|
|
2025-11-21 09:58:26 +08:00
|
|
|
|
2025-05-20 12:54:34 +08:00
|
|
|
|
2025-06-20 09:09:56 -06:00
|
|
|
|
2025-05-20 12:54:34 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-21 09:58:26 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-23 21:51:58 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-21 09:58:26 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-23 21:51:58 +08:00
|
|
|
|
2025-11-21 09:58:26 +08:00
|
|
|
|
2026-01-23 21:51:58 +08:00
|
|
|
|
2025-11-21 09:58:26 +08:00
|
|
|
|
|
|
|
|
|
2025-05-20 12:54:32 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-21 09:58:24 +08:00
|
|
|
|
2025-11-21 09:58:25 +08:00
|
|
|
|
2025-05-20 12:54:32 +08:00
|
|
|
|
2025-11-21 09:58:25 +08:00
|
|
|
|
2025-11-21 09:58:26 +08:00
|
|
|
|
|
|
|
|
|
2025-05-20 12:54:34 +08:00
|
|
|
|
2025-12-12 22:34:15 +08:00
|
|
|
|
2025-11-21 09:58:26 +08:00
|
|
|
|
2025-05-20 12:54:32 +08:00
|
|
|
|
2025-05-22 23:20:40 +08:00
|
|
|
|
2025-11-21 09:58:26 +08:00
|
|
|
|
2025-05-20 12:54:32 +08:00
|
|
|
|
|
|
|
|
|
2026-01-23 21:51:58 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-05-20 12:54:32 +08:00
|
|
|
|
2026-01-23 21:51:58 +08:00
|
|
|
|
2025-11-21 09:58:26 +08:00
|
|
|
|
2025-05-20 12:54:32 +08:00
|
|
|
|
2025-11-21 09:58:26 +08:00
|
|
|
|
2026-01-23 21:51:58 +08:00
|
|
|
|
2025-11-21 09:58:26 +08:00
|
|
|
|
|
|
|
|
|
2025-05-20 12:54:32 +08:00
|
|
|
|
|
|
|
|
|
2025-04-30 16:52:31 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:53 +08:00
|
|
|
|
2025-04-30 16:52:31 -06:00
|
|
|
|
2026-03-31 23:31:53 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-30 16:52:31 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-10-31 14:34:30 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-10-31 14:34:30 -06:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2025-04-30 16:52:30 -06:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:13 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
2022-09-23 23:39:13 +08:00
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
2022-09-23 23:39:13 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
2022-09-23 23:39:15 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-03-30 19:36:22 +08:00
|
|
|
|
2022-07-28 20:39:16 +08:00
|
|
|
|
|
|
|
|
|
2025-04-30 16:52:27 -06:00
|
|
|
|
2022-07-28 20:39:16 +08:00
|
|
|
|
|
|
|
|
|
2025-04-30 16:52:32 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-30 16:52:34 -06:00
|
|
|
|
2025-04-30 16:52:32 -06:00
|
|
|
|
|
|
|
|
|
2022-07-28 20:39:16 +08:00
|
|
|
|
|
|
|
|
|
2025-04-30 16:52:31 -06:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2026-05-13 15:18:46 -06:00
|
|
|
|
2026-01-23 21:51:58 +08:00
|
|
|
|
2025-11-21 09:58:26 +08:00
|
|
|
|
|
|
|
|
|
2025-05-20 12:54:32 +08:00
|
|
|
|
2025-11-21 09:58:26 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-05-13 15:18:46 -06:00
|
|
|
|
2026-01-23 21:51:58 +08:00
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-23 21:51:58 +08:00
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-18 09:41:12 +08:00
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-18 09:41:12 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-18 09:41:12 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:40 +08:00
|
|
|
|
|
|
|
|
|
2025-10-31 14:34:30 -06:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
2025-10-31 14:34:30 -06:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-10-31 14:34:30 -06:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-11-21 23:56:45 +08:00
|
|
|
|
2022-10-29 09:04:31 +08:00
|
|
|
|
2025-03-28 12:04:11 -06:00
|
|
|
|
|
|
|
|
|
2022-10-29 09:04:31 +08:00
|
|
|
|
2025-03-28 12:04:11 -06:00
|
|
|
|
|
|
|
|
|
2022-10-29 09:04:31 +08:00
|
|
|
|
|
|
|
|
|
2025-10-31 14:34:30 -06:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
2025-10-31 14:34:30 -06:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-28 12:04:08 -06:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
|
|
|
|
|
2025-10-31 14:34:30 -06:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
2025-03-28 12:04:08 -06:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
|
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
2025-03-28 12:04:10 -06:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
2025-03-28 12:04:10 -06:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
|
|
|
|
|
2023-05-02 10:42:31 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-13 22:33:57 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-02 10:42:31 +08:00
|
|
|
|
2025-07-13 22:33:57 +08:00
|
|
|
|
|
|
|
|
|
2023-05-02 10:42:31 +08:00
|
|
|
|
2025-07-13 22:33:57 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-02 10:42:31 +08:00
|
|
|
|
|
|
|
|
|
2025-04-09 09:14:42 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-08 15:44:43 -06:00
|
|
|
|
2024-10-07 12:24:17 -06:00
|
|
|
|
2022-10-29 09:04:31 +08:00
|
|
|
|
2022-09-23 23:39:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-08 15:44:43 -06:00
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:16 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-04-09 09:14:42 +08:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:42 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-09 09:14:42 +08:00
|
|
|
|
2026-01-16 22:18:42 +08:00
|
|
|
|
|
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
2026-01-16 22:18:42 +08:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
2025-03-27 17:51:11 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
2026-01-16 22:18:42 +08:00
|
|
|
|
2022-09-23 23:39:15 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:42 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-11-21 23:56:45 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:42 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-25 10:25:54 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
2025-07-01 15:23:25 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-25 10:25:54 +08:00
|
|
|
|
|
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
|
|
|
|
|
2025-07-01 15:23:25 +08:00
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
|
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
|
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-14 18:32:01 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-27 17:51:17 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2023-05-02 10:42:31 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:42 +08:00
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-21 09:58:25 +08:00
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:09:56 -06:00
|
|
|
|
|
|
|
|
|
2025-06-20 09:10:05 -06:00
|
|
|
|
2025-04-16 11:54:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-21 15:09:12 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-13 22:33:56 +08:00
|
|
|
|
2022-07-21 15:09:12 +02:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-05-08 20:37:46 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:38 +08:00
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
2025-04-16 11:54:38 +08:00
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
2025-08-08 09:52:15 -06:00
|
|
|
|
2025-07-13 22:33:56 +08:00
|
|
|
|
2025-04-16 11:54:38 +08:00
|
|
|
|
|
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-09 20:14:54 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-03 23:41:08 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-27 20:15:59 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-08-27 20:15:59 +08:00
|
|
|
|
|
|
|
|
|
2025-12-12 17:19:49 -07:00
|
|
|
|
2025-08-27 20:15:59 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-27 20:15:59 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-03 23:41:08 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-08 15:44:43 -06:00
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:40 +08:00
|
|
|
|
|
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-08 15:44:43 -06:00
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-04-16 11:54:38 +08:00
|
|
|
|
|
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
2022-07-21 15:09:12 +02:00
|
|
|
|
2025-08-27 20:15:59 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-11-11 19:07:18 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-10-09 17:33:19 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-10-09 17:33:19 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:38 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:39 -06:00
|
|
|
|
2025-04-16 11:54:41 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:43 +08:00
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:41 +08:00
|
|
|
|
2025-09-17 19:49:39 -06:00
|
|
|
|
2025-04-16 11:54:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:46 +08:00
|
|
|
|
2025-04-16 11:54:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:43 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-04-16 11:54:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-30 16:52:34 -06:00
|
|
|
|
2025-09-17 19:49:39 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2026-01-16 22:18:43 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
ublk: speed up ublk server exit handling
Recently, we've observed a few cases where a ublk server is able to
complete restart more quickly than the driver can process the exit of
the previous ublk server. The new ublk server comes up, attempts
recovery of the preexisting ublk devices, and observes them still in
state UBLK_S_DEV_LIVE. While this is possible due to the asynchronous
nature of io_uring cleanup and should therefore be handled properly in
the ublk server, it is still preferable to make ublk server exit
handling faster if possible, as we should strive for it to not be a
limiting factor in how fast a ublk server can restart and provide
service again.
Analysis of the issue showed that the vast majority of the time spent in
handling the ublk server exit was in calls to blk_mq_quiesce_queue,
which is essentially just a (relatively expensive) call to
synchronize_rcu. The ublk server exit path currently issues an
unnecessarily large number of calls to blk_mq_quiesce_queue, for two
reasons:
1. It tries to call blk_mq_quiesce_queue once per ublk_queue. However,
blk_mq_quiesce_queue targets the request_queue of the underlying ublk
device, of which there is only one. So the number of calls is larger
than necessary by a factor of nr_hw_queues.
2. In practice, it calls blk_mq_quiesce_queue _more_ than once per
ublk_queue. This is because of a data race where we read
ubq->canceling without any locking when deciding if we should call
ublk_start_cancel. It is thus possible for two calls to
ublk_uring_cmd_cancel_fn against the same ublk_queue to both call
ublk_start_cancel against the same ublk_queue.
Fix this by making the "canceling" flag a per-device state. This
actually matches the existing code better, as there are several places
where the flag is set or cleared for all queues simultaneously, and
there is the general expectation that cancellation corresponds with ublk
server exit. This per-device canceling flag is then checked under a
(new) lock (addressing the data race (2) above), and the queue is only
quiesced if it is cleared (addressing (1) above). The result is just one
call to blk_mq_quiesce_queue per ublk device.
To minimize the number of cache lines that are accessed in the hot path,
the per-queue canceling flag is kept. The values of the per-device
canceling flag and all per-queue canceling flags should always match.
In our setup, where one ublk server handles I/O for 128 ublk devices,
each having 24 hardware queues of depth 4096, here are the results
before and after this patch, where teardown time is measured from the
first call to io_ring_ctx_wait_and_kill to the return from the last
ublk_ch_release:
before after
number of calls to blk_mq_quiesce_queue: 6469 256
teardown time: 11.14s 2.44s
There are still some potential optimizations here, but this takes care
of a big chunk of the ublk server exit handling delay.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250703-ublk_too_many_quiesce-v2-1-3527b5339eeb@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-07-03 23:41:07 -06:00
|
|
|
|
2023-10-09 17:33:20 +08:00
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
2023-10-09 17:33:20 +08:00
|
|
|
|
ublk: speed up ublk server exit handling
Recently, we've observed a few cases where a ublk server is able to
complete restart more quickly than the driver can process the exit of
the previous ublk server. The new ublk server comes up, attempts
recovery of the preexisting ublk devices, and observes them still in
state UBLK_S_DEV_LIVE. While this is possible due to the asynchronous
nature of io_uring cleanup and should therefore be handled properly in
the ublk server, it is still preferable to make ublk server exit
handling faster if possible, as we should strive for it to not be a
limiting factor in how fast a ublk server can restart and provide
service again.
Analysis of the issue showed that the vast majority of the time spent in
handling the ublk server exit was in calls to blk_mq_quiesce_queue,
which is essentially just a (relatively expensive) call to
synchronize_rcu. The ublk server exit path currently issues an
unnecessarily large number of calls to blk_mq_quiesce_queue, for two
reasons:
1. It tries to call blk_mq_quiesce_queue once per ublk_queue. However,
blk_mq_quiesce_queue targets the request_queue of the underlying ublk
device, of which there is only one. So the number of calls is larger
than necessary by a factor of nr_hw_queues.
2. In practice, it calls blk_mq_quiesce_queue _more_ than once per
ublk_queue. This is because of a data race where we read
ubq->canceling without any locking when deciding if we should call
ublk_start_cancel. It is thus possible for two calls to
ublk_uring_cmd_cancel_fn against the same ublk_queue to both call
ublk_start_cancel against the same ublk_queue.
Fix this by making the "canceling" flag a per-device state. This
actually matches the existing code better, as there are several places
where the flag is set or cleared for all queues simultaneously, and
there is the general expectation that cancellation corresponds with ublk
server exit. This per-device canceling flag is then checked under a
(new) lock (addressing the data race (2) above), and the queue is only
quiesced if it is cleared (addressing (1) above). The result is just one
call to blk_mq_quiesce_queue per ublk device.
To minimize the number of cache lines that are accessed in the hot path,
the per-queue canceling flag is kept. The values of the per-device
canceling flag and all per-queue canceling flags should always match.
In our setup, where one ublk server handles I/O for 128 ublk devices,
each having 24 hardware queues of depth 4096, here are the results
before and after this patch, where teardown time is measured from the
first call to io_ring_ctx_wait_and_kill to the return from the last
ublk_ch_release:
before after
number of calls to blk_mq_quiesce_queue: 6469 256
teardown time: 11.14s 2.44s
There are still some potential optimizations here, but this takes care
of a big chunk of the ublk server exit handling delay.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250703-ublk_too_many_quiesce-v2-1-3527b5339eeb@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-07-03 23:41:07 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: set canceling flag even when disk is not allocated
ublk_start_cancel() previously bailed out early when ublk_get_disk()
returned NULL, treating it as "our disk has been dead". That is correct
for the post-teardown case, but it also wrongly covers the pre-start
case: ublk_ctrl_start_dev() has not assigned ub->ub_disk yet, while
io_uring is already tearing down the daemon's uring_cmds via
ublk_uring_cmd_cancel_fn().
In that window, the cancel path skips ublk_set_canceling(), so
ubq->canceling stays false, even though ublk_cancel_cmd() goes on to
NULL out every io->cmd. ublk_ctrl_start_dev() then proceeds to set
ub->ub_disk, call add_disk(), and schedule partition_scan_work. When
ublk_partition_scan_work() runs bdev_disk_changed() and the resulting
read reaches ublk_queue_rq() -> ublk_queue_cmd(), the ubq->canceling
check passes and the code dereferences the NULL io->cmd:
BUG: kernel NULL pointer dereference, address: 0000000000000018
RIP: ublk_queue_cmd drivers/block/ublk_drv.c [inline]
RIP: ublk_queue_rq+0x73/0x100
Call Trace:
blk_mq_dispatch_rq_list+0x1c5/0xca0
...
bdev_disk_changed+0x3d4/0x5e0
ublk_partition_scan_work+0x89/0xe0
process_one_work+0x344/0x8a0
Fix it by always setting ub->canceling / ubq->canceling under
cancel_mutex. When the disk is allocated, keep the existing
quiesce/unquiesce dance so the flag is observed across the
ublk_queue_rq() barrier. When the disk is not yet allocated, there is
no request_queue and ublk_queue_rq() cannot be running concurrently, so
simply flipping the flag is sufficient: any subsequent I/O - including
the partition scan started by ublk_ctrl_start_dev() - will see
canceling set and be aborted via __ublk_queue_rq_common().
Fixes: 7fc4da6a304b ("ublk: scan partition in async way")
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Link: https://patch.msgid.link/20260527144042.2095194-1-tom.leiming@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2026-05-27 09:40:42 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: speed up ublk server exit handling
Recently, we've observed a few cases where a ublk server is able to
complete restart more quickly than the driver can process the exit of
the previous ublk server. The new ublk server comes up, attempts
recovery of the preexisting ublk devices, and observes them still in
state UBLK_S_DEV_LIVE. While this is possible due to the asynchronous
nature of io_uring cleanup and should therefore be handled properly in
the ublk server, it is still preferable to make ublk server exit
handling faster if possible, as we should strive for it to not be a
limiting factor in how fast a ublk server can restart and provide
service again.
Analysis of the issue showed that the vast majority of the time spent in
handling the ublk server exit was in calls to blk_mq_quiesce_queue,
which is essentially just a (relatively expensive) call to
synchronize_rcu. The ublk server exit path currently issues an
unnecessarily large number of calls to blk_mq_quiesce_queue, for two
reasons:
1. It tries to call blk_mq_quiesce_queue once per ublk_queue. However,
blk_mq_quiesce_queue targets the request_queue of the underlying ublk
device, of which there is only one. So the number of calls is larger
than necessary by a factor of nr_hw_queues.
2. In practice, it calls blk_mq_quiesce_queue _more_ than once per
ublk_queue. This is because of a data race where we read
ubq->canceling without any locking when deciding if we should call
ublk_start_cancel. It is thus possible for two calls to
ublk_uring_cmd_cancel_fn against the same ublk_queue to both call
ublk_start_cancel against the same ublk_queue.
Fix this by making the "canceling" flag a per-device state. This
actually matches the existing code better, as there are several places
where the flag is set or cleared for all queues simultaneously, and
there is the general expectation that cancellation corresponds with ublk
server exit. This per-device canceling flag is then checked under a
(new) lock (addressing the data race (2) above), and the queue is only
quiesced if it is cleared (addressing (1) above). The result is just one
call to blk_mq_quiesce_queue per ublk device.
To minimize the number of cache lines that are accessed in the hot path,
the per-queue canceling flag is kept. The values of the per-device
canceling flag and all per-queue canceling flags should always match.
In our setup, where one ublk server handles I/O for 128 ublk devices,
each having 24 hardware queues of depth 4096, here are the results
before and after this patch, where teardown time is measured from the
first call to io_ring_ctx_wait_and_kill to the return from the last
ublk_ch_release:
before after
number of calls to blk_mq_quiesce_queue: 6469 256
teardown time: 11.14s 2.44s
There are still some potential optimizations here, but this takes care
of a big chunk of the ublk server exit handling delay.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250703-ublk_too_many_quiesce-v2-1-3527b5339eeb@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-07-03 23:41:07 -06:00
|
|
|
|
|
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
2023-10-09 17:33:20 +08:00
|
|
|
|
|
|
|
|
|
2025-04-25 09:37:40 +08:00
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-04-25 09:37:40 +08:00
|
|
|
|
|
|
|
|
|
2026-05-08 20:37:46 +08:00
|
|
|
|
2025-04-25 09:37:40 +08:00
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
2023-10-09 17:33:20 +08:00
|
|
|
|
|
|
|
|
|
2025-04-25 09:37:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-05-16 00:26:01 +08:00
|
|
|
|
2025-04-25 09:37:40 +08:00
|
|
|
|
|
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
|
|
|
|
|
2026-05-08 20:37:46 +08:00
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
2026-05-08 20:37:46 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2026-05-08 20:37:46 +08:00
|
|
|
|
|
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
|
|
|
|
|
2026-01-30 00:19:50 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-30 00:19:50 +08:00
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
2026-01-30 00:19:50 +08:00
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
|
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
|
|
|
|
|
ublk: speed up ublk server exit handling
Recently, we've observed a few cases where a ublk server is able to
complete restart more quickly than the driver can process the exit of
the previous ublk server. The new ublk server comes up, attempts
recovery of the preexisting ublk devices, and observes them still in
state UBLK_S_DEV_LIVE. While this is possible due to the asynchronous
nature of io_uring cleanup and should therefore be handled properly in
the ublk server, it is still preferable to make ublk server exit
handling faster if possible, as we should strive for it to not be a
limiting factor in how fast a ublk server can restart and provide
service again.
Analysis of the issue showed that the vast majority of the time spent in
handling the ublk server exit was in calls to blk_mq_quiesce_queue,
which is essentially just a (relatively expensive) call to
synchronize_rcu. The ublk server exit path currently issues an
unnecessarily large number of calls to blk_mq_quiesce_queue, for two
reasons:
1. It tries to call blk_mq_quiesce_queue once per ublk_queue. However,
blk_mq_quiesce_queue targets the request_queue of the underlying ublk
device, of which there is only one. So the number of calls is larger
than necessary by a factor of nr_hw_queues.
2. In practice, it calls blk_mq_quiesce_queue _more_ than once per
ublk_queue. This is because of a data race where we read
ubq->canceling without any locking when deciding if we should call
ublk_start_cancel. It is thus possible for two calls to
ublk_uring_cmd_cancel_fn against the same ublk_queue to both call
ublk_start_cancel against the same ublk_queue.
Fix this by making the "canceling" flag a per-device state. This
actually matches the existing code better, as there are several places
where the flag is set or cleared for all queues simultaneously, and
there is the general expectation that cancellation corresponds with ublk
server exit. This per-device canceling flag is then checked under a
(new) lock (addressing the data race (2) above), and the queue is only
quiesced if it is cleared (addressing (1) above). The result is just one
call to blk_mq_quiesce_queue per ublk device.
To minimize the number of cache lines that are accessed in the hot path,
the per-queue canceling flag is kept. The values of the per-device
canceling flag and all per-queue canceling flags should always match.
In our setup, where one ublk server handles I/O for 128 ublk devices,
each having 24 hardware queues of depth 4096, here are the results
before and after this patch, where teardown time is measured from the
first call to io_ring_ctx_wait_and_kill to the return from the last
ublk_ch_release:
before after
number of calls to blk_mq_quiesce_queue: 6469 256
teardown time: 11.14s 2.44s
There are still some potential optimizations here, but this takes care
of a big chunk of the ublk server exit handling delay.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250703-ublk_too_many_quiesce-v2-1-3527b5339eeb@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-07-03 23:41:07 -06:00
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
2025-04-25 09:37:40 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-08 12:45:41 -06:00
|
|
|
|
2022-07-30 17:27:47 +08:00
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
2022-07-30 17:27:47 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
2025-04-25 09:37:40 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:16 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2022-09-23 23:39:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:36 +08:00
|
|
|
|
2022-09-23 23:39:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:36 +08:00
|
|
|
|
2022-09-23 23:39:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:36 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-12-25 19:06:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:16 +08:00
|
|
|
|
2023-10-09 17:33:20 +08:00
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:16 +08:00
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:36 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2024-12-25 19:06:40 +08:00
|
|
|
|
2023-10-09 17:33:20 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
2024-10-07 12:24:16 -06:00
|
|
|
|
|
|
|
|
|
ublk: improve detection and handling of ublk server exit
There are currently two ways in which ublk server exit is detected by
ublk_drv:
1. uring_cmd cancellation. If there are any outstanding uring_cmds which
have not been completed to the ublk server when it exits, io_uring
calls the uring_cmd callback with a special cancellation flag as the
issuing task is exiting.
2. I/O timeout. This is needed in addition to the above to handle the
"saturated queue" case, when all I/Os for a given queue are in the
ublk server, and therefore there are no outstanding uring_cmds to
cancel when the ublk server exits.
There are a couple of issues with this approach:
- It is complex and inelegant to have two methods to detect the same
condition
- The second method detects ublk server exit only after a long delay
(~30s, the default timeout assigned by the block layer). This delays
the nosrv behavior from kicking in and potential subsequent recovery
of the device.
The second issue is brought to light with the new test_generic_06 which
will be added in following patch. It fails before this fix:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 30.0611 s, 0.0 kB/s
DEAD
dd took 31 seconds to exit (>= 5s tolerance)!
generic_06 : [FAIL]
Fix this by instead detecting and handling ublk server exit in the
character file release callback. This has several advantages:
- This one place can handle both saturated and unsaturated queues. Thus,
it replaces both preexisting methods of detecting ublk server exit.
- It runs quickly on ublk server exit - there is no 30s delay.
- It starts the process of removing task references in ublk_drv. This is
needed if we want to relax restrictions in the driver like letting
only one thread serve each queue
There is also the disadvantage that the character file release callback
can also be triggered by intentional close of the file, which is a
significant behavior change. Preexisting ublk servers (libublksrv) are
dependent on the ability to open/close the file multiple times. To
address this, only transition to a nosrv state if the file is released
while the ublk device is live. This allows for programs to open/close
the file multiple times during setup. It is still a behavior change if a
ublk server decides to close/reopen the file while the device is LIVE
(i.e. while it is responsible for serving I/O), but that would be highly
unusual. This behavior is in line with what is done by FUSE, which is
very similar to ublk in that a userspace daemon is providing services
traditionally provided by the kernel.
With this change in, the new test (and all other selftests, and all
ublksrv tests) pass:
selftests: ublk: test_generic_06.sh
dev id is 0
dd: error writing '/dev/ublkb0': Input/output error
1+0 records in
0+0 records out
0 bytes copied, 0.0376731 s, 0.0 kB/s
DEAD
generic_04 : [PASS]
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250416035444.99569-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-04-16 11:54:39 +08:00
|
|
|
|
2024-10-07 12:24:16 -06:00
|
|
|
|
2026-01-09 20:14:54 +08:00
|
|
|
|
2024-10-07 12:24:16 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-05 22:25:30 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:38 +08:00
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
2026-01-23 17:15:43 +08:00
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:38 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2026-04-05 22:25:30 -06:00
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:35 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
|
|
|
|
|
2025-08-08 09:52:15 -06:00
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:38 +08:00
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
2026-04-05 22:25:30 -06:00
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-08 12:45:41 -06:00
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
|
|
|
|
|
ublk: wait on ublk_dev_ready() instead of ub->completion
ub->completion is only re-armed by a successful START_USER_RECOVERY. If
the ublk server sends END_USER_RECOVERY without one - e.g. its START
failed with -EBUSY and the error was ignored - the wait is satisfied by
the stale completion of the previous recovery cycle, and the device is
marked LIVE and the requeue list kicked while the FETCH stream is still
running and ubq->canceling is still set. The kick redispatches a
previously requeued request, __ublk_queue_rq_common() sees ->canceling
and parks it again via __ublk_abort_rq(), and after the last FETCH
clears ->canceling nothing ever kicks the requeue list again: the
request is stranded there while holding its tag. If it is the flush
machinery's flush_rq, every subsequent fsync piles up in uninterruptible
sleep and teardown hangs on tag draining. This matches a report of a
lost PREFLUSH with ext4 on top of ublk after daemon crash recovery.
ub->completion is an edge-triggered latch used as a proxy for the level
condition "every queue has fetched all I/O commands", which can regress
(F_BATCH's UNPREP, daemon death) and whose re-arm can be skipped. Drop
it and wait on the real condition instead: the new helper
ublk_wait_dev_ready_and_lock() waits on ublk_dev_ready() via
wait_var_event_interruptible(), woken from ublk_mark_io_ready(), then
re-checks it under ub->mutex, waiting again on regression, and returns
with the mutex held and readiness guaranteed.
Readiness becomes true in the same ub->mutex critical section that
clears the last queue's ->canceling, so END_USER_RECOVERY marks the
device LIVE and kicks the requeue list strictly after ->canceling
clears. The wait stays interruptible, so a server whose daemon died can
still be signalled out. For ublk_ctrl_start_dev() this replaces the
fail-fast -EINVAL on an F_BATCH ready->UNPREP regression with waiting
until the device is ready again.
Reported-by: George Salisbury <gsalisbury@apnic.net>
Fixes: 728cbac5fe21 ("ublk: move device reset into ublk_ch_release()")
Cc: stable@vger.kernel.org
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Link: https://patch.msgid.link/20260719134540.120269-1-tom.leiming@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2026-07-19 08:45:40 -05:00
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: wait on ublk_dev_ready() instead of ub->completion
ub->completion is only re-armed by a successful START_USER_RECOVERY. If
the ublk server sends END_USER_RECOVERY without one - e.g. its START
failed with -EBUSY and the error was ignored - the wait is satisfied by
the stale completion of the previous recovery cycle, and the device is
marked LIVE and the requeue list kicked while the FETCH stream is still
running and ubq->canceling is still set. The kick redispatches a
previously requeued request, __ublk_queue_rq_common() sees ->canceling
and parks it again via __ublk_abort_rq(), and after the last FETCH
clears ->canceling nothing ever kicks the requeue list again: the
request is stranded there while holding its tag. If it is the flush
machinery's flush_rq, every subsequent fsync piles up in uninterruptible
sleep and teardown hangs on tag draining. This matches a report of a
lost PREFLUSH with ext4 on top of ublk after daemon crash recovery.
ub->completion is an edge-triggered latch used as a proxy for the level
condition "every queue has fetched all I/O commands", which can regress
(F_BATCH's UNPREP, daemon death) and whose re-arm can be skipped. Drop
it and wait on the real condition instead: the new helper
ublk_wait_dev_ready_and_lock() waits on ublk_dev_ready() via
wait_var_event_interruptible(), woken from ublk_mark_io_ready(), then
re-checks it under ub->mutex, waiting again on regression, and returns
with the mutex held and readiness guaranteed.
Readiness becomes true in the same ub->mutex critical section that
clears the last queue's ->canceling, so END_USER_RECOVERY marks the
device LIVE and kicks the requeue list strictly after ->canceling
clears. The wait stays interruptible, so a server whose daemon died can
still be signalled out. For ublk_ctrl_start_dev() this replaces the
fail-fast -EINVAL on an F_BATCH ready->UNPREP regression with waiting
until the device is ready again.
Reported-by: George Salisbury <gsalisbury@apnic.net>
Fixes: 728cbac5fe21 ("ublk: move device reset into ublk_ch_release()")
Cc: stable@vger.kernel.org
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Link: https://patch.msgid.link/20260719134540.120269-1-tom.leiming@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2026-07-19 08:45:40 -05:00
|
|
|
|
2025-04-16 11:54:38 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2023-04-18 21:18:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-05 23:31:42 +08:00
|
|
|
|
2023-04-18 21:18:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-13 22:34:02 +08:00
|
|
|
|
2023-05-19 14:50:25 +08:00
|
|
|
|
2025-11-21 09:58:25 +08:00
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
2025-11-21 09:58:25 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
|
|
|
|
|
2025-11-21 09:58:25 +08:00
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
2025-11-21 09:58:25 +08:00
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:38 +08:00
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-21 09:58:25 +08:00
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
2026-01-16 22:18:38 +08:00
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
2026-01-16 22:18:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-13 22:34:02 +08:00
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
|
|
|
|
|
2025-07-13 22:34:00 +08:00
|
|
|
|
|
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
2023-05-19 14:50:25 +08:00
|
|
|
|
2025-07-13 22:34:00 +08:00
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:25 +08:00
|
|
|
|
|
|
|
|
|
2025-07-13 22:33:59 +08:00
|
|
|
|
|
|
|
|
|
2025-07-13 22:34:00 +08:00
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:25 +08:00
|
|
|
|
|
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
2025-09-17 19:49:48 -06:00
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:48 -06:00
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
|
|
|
|
|
2025-11-21 09:58:25 +08:00
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
2023-05-19 14:50:25 +08:00
|
|
|
|
|
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-27 14:39:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:09:56 -06:00
|
|
|
|
2025-02-27 14:39:15 -08:00
|
|
|
|
2025-06-20 09:10:06 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:10:07 -06:00
|
|
|
|
2025-02-27 14:39:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:43 -06:00
|
|
|
|
2025-09-17 19:49:46 -06:00
|
|
|
|
2025-06-20 09:09:56 -06:00
|
|
|
|
2025-03-01 12:03:16 -07:00
|
|
|
|
2025-02-27 14:39:15 -08:00
|
|
|
|
|
|
|
|
|
2025-03-01 12:03:16 -07:00
|
|
|
|
2025-02-27 14:39:15 -08:00
|
|
|
|
2025-09-17 19:49:44 -06:00
|
|
|
|
2025-04-29 10:29:38 +08:00
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:36 -07:00
|
|
|
|
2025-02-27 14:39:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:10:07 -06:00
|
|
|
|
2025-02-27 14:39:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:10:05 -06:00
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:43 -06:00
|
|
|
|
2025-09-17 19:49:46 -06:00
|
|
|
|
2025-06-20 09:10:05 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:46 -06:00
|
|
|
|
2025-09-17 19:49:43 -06:00
|
|
|
|
2025-06-20 09:10:05 -06:00
|
|
|
|
2026-05-13 15:18:46 -06:00
|
|
|
|
2025-06-20 09:10:05 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-27 14:39:15 -08:00
|
|
|
|
2025-06-20 09:10:02 -06:00
|
|
|
|
2025-03-01 12:03:16 -07:00
|
|
|
|
2025-02-27 14:39:15 -08:00
|
|
|
|
2025-06-20 09:10:02 -06:00
|
|
|
|
2025-04-29 10:29:38 +08:00
|
|
|
|
|
|
|
|
|
2025-03-01 12:03:16 -07:00
|
|
|
|
2025-02-27 14:39:15 -08:00
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:47 -06:00
|
|
|
|
2025-07-13 22:34:03 +08:00
|
|
|
|
2025-09-17 19:49:47 -06:00
|
|
|
|
2025-07-13 22:34:03 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:47 -06:00
|
|
|
|
2025-07-13 22:34:03 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-21 09:58:28 +08:00
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
2025-11-21 09:58:28 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:35 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-21 09:58:28 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:49 -06:00
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
2025-04-16 11:54:35 +08:00
|
|
|
|
2025-11-21 09:58:28 +08:00
|
|
|
|
2025-04-16 11:54:35 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
2025-11-21 09:58:28 +08:00
|
|
|
|
|
|
|
|
|
2026-01-23 17:15:42 +08:00
|
|
|
|
2026-04-05 22:25:30 -06:00
|
|
|
|
2025-04-16 11:54:35 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:50 -06:00
|
|
|
|
2025-07-13 22:34:00 +08:00
|
|
|
|
2025-04-30 16:52:26 -06:00
|
|
|
|
2025-05-05 11:26:23 -06:00
|
|
|
|
2025-04-30 16:52:26 -06:00
|
|
|
|
2025-09-17 19:49:50 -06:00
|
|
|
|
2025-04-30 16:52:26 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:50 -06:00
|
|
|
|
2025-04-30 16:52:26 -06:00
|
|
|
|
|
|
|
|
|
2025-07-13 22:34:00 +08:00
|
|
|
|
2025-04-30 16:52:26 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:51 -06:00
|
|
|
|
2025-07-15 09:42:43 -06:00
|
|
|
|
2025-07-13 22:34:00 +08:00
|
|
|
|
2025-09-17 19:49:51 -06:00
|
|
|
|
2025-07-15 09:42:43 -06:00
|
|
|
|
2025-07-13 22:34:04 +08:00
|
|
|
|
2025-04-30 16:52:26 -06:00
|
|
|
|
|
|
|
|
|
2025-06-24 18:41:21 +08:00
|
|
|
|
|
|
|
|
|
2025-04-30 16:52:32 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-21 09:58:25 +08:00
|
|
|
|
2025-04-30 16:52:32 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-08 09:32:50 -06:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-08-08 09:32:50 -06:00
|
|
|
|
2026-02-18 20:59:30 -08:00
|
|
|
|
|
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2025-11-26 12:48:35 +00:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-08-08 09:32:50 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-24 18:41:21 +08:00
|
|
|
|
2025-06-20 09:09:57 -06:00
|
|
|
|
2025-07-13 22:34:04 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-08-08 09:32:50 -06:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-08-08 09:32:50 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-06-20 09:09:57 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:10:03 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-08 09:32:50 -06:00
|
|
|
|
2025-06-20 09:10:03 -06:00
|
|
|
|
2025-06-20 09:09:57 -06:00
|
|
|
|
2025-08-08 09:32:50 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2025-08-08 09:32:50 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-09-17 19:49:41 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:09:58 -06:00
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:47 -06:00
|
|
|
|
2025-07-13 22:34:03 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
2025-06-20 09:09:58 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:10:03 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:46 -06:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:10:00 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-05-05 11:26:23 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-05-05 11:26:23 -06:00
|
|
|
|
2022-07-28 20:39:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-04-18 21:18:10 +08:00
|
|
|
|
2022-07-28 20:39:16 +08:00
|
|
|
|
|
|
|
|
|
2023-04-18 21:18:10 +08:00
|
|
|
|
2025-02-27 14:39:15 -08:00
|
|
|
|
2025-09-17 19:49:46 -06:00
|
|
|
|
2025-06-20 09:10:05 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-09-17 19:49:50 -06:00
|
|
|
|
2025-04-16 11:54:35 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-08-08 09:32:50 -06:00
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
2025-09-17 19:49:48 -06:00
|
|
|
|
2025-07-13 22:34:04 +08:00
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:51 -06:00
|
|
|
|
2025-07-13 22:34:04 +08:00
|
|
|
|
|
|
|
|
|
2025-08-08 09:32:50 -06:00
|
|
|
|
2025-07-13 22:34:04 +08:00
|
|
|
|
2026-01-16 22:18:46 +08:00
|
|
|
|
2025-07-13 22:34:04 +08:00
|
|
|
|
2025-04-30 16:52:26 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2022-07-28 20:39:16 +08:00
|
|
|
|
2025-06-24 18:41:21 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
2025-09-17 19:49:48 -06:00
|
|
|
|
2025-07-13 22:34:01 +08:00
|
|
|
|
2025-06-24 18:41:21 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-26 12:48:35 +00:00
|
|
|
|
2025-02-25 14:24:55 -07:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
2026-01-08 02:19:36 -07:00
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:09:56 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:46 -06:00
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:10:07 -06:00
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-05-13 15:18:46 -06:00
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:10:07 -06:00
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-10-31 14:34:30 -06:00
|
|
|
|
2023-10-09 17:33:17 +08:00
|
|
|
|
2025-10-31 14:34:30 -06:00
|
|
|
|
|
|
|
|
|
2026-05-01 19:23:12 +08:00
|
|
|
|
2025-02-25 14:24:55 -07:00
|
|
|
|
2026-05-01 19:23:12 +08:00
|
|
|
|
|
|
|
|
|
2025-02-25 14:24:55 -07:00
|
|
|
|
2025-09-22 11:02:31 -06:00
|
|
|
|
2023-10-09 17:33:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: replace monitor with cancelable uring_cmd
Monitor work actually introduces one extra context for handling abort, this
way is easy to cause race, and also introduce extra delay when handling
aborting.
Now we start to support cancelable uring_cmd, so use it instead:
1) this cancel callback is either run from the uring cmd submission task
context or called after the io_uring context is exit, so the callback is
run exclusively with ublk_ch_uring_cmd() and __ublk_rq_task_work().
2) the previous patch freezes request queue when calling ublk_abort_queue(),
which is now completely exclusive with ublk_queue_rq() and
ublk_ch_uring_cmd()/__ublk_rq_task_work().
3) in timeout handler, if all IOs are in-flight, then all uring commands
are completed, uring command canceling can't help us to provide forward
progress any more, so call ublk_abort_requests() in timeout handler.
This way simplifies aborting queue, and is helpful for adding new feature,
such as, relax the limit of using single task for handling one queue.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20231009093324.957829-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-09 17:33:21 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-10-09 17:33:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-06-30 23:18:27 +02:00
|
|
|
|
2026-01-16 22:18:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-06-30 23:18:27 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
|
|
|
|
|
2026-01-23 17:15:43 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
2026-01-23 17:15:43 +08:00
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
2026-01-23 17:15:43 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:47 +08:00
|
|
|
|
2026-01-16 22:18:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-23 17:15:42 +08:00
|
|
|
|
2026-04-05 22:25:30 -06:00
|
|
|
|
2026-01-23 17:15:42 +08:00
|
|
|
|
2026-01-16 22:18:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-06-30 23:18:27 +02:00
|
|
|
|
2026-01-16 22:18:37 +08:00
|
|
|
|
|
|
|
|
|
2026-06-30 23:18:27 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-06-30 23:18:27 +02:00
|
|
|
|
2026-01-16 22:18:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:46 +08:00
|
|
|
|
2026-01-16 22:18:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:46 +08:00
|
|
|
|
2026-01-16 22:18:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:46 +08:00
|
|
|
|
2026-01-16 22:18:38 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:46 +08:00
|
|
|
|
2026-01-16 22:18:38 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:46 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:36 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:44 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:36 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:44 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:44 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-02-18 20:59:30 -08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:44 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:34 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-02-18 20:59:30 -08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:36 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:36 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:36 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:38 +08:00
|
|
|
|
2026-01-16 22:18:36 +08:00
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:36 +08:00
|
|
|
|
2026-01-16 22:18:44 +08:00
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:36 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:34 +08:00
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:35 -07:00
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:35 -07:00
|
|
|
|
2026-01-08 02:19:37 -07:00
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:39 -07:00
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:35 -07:00
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:35 -07:00
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:35 -07:00
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:37 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:35 -07:00
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
2025-09-17 19:49:42 -06:00
|
|
|
|
2026-01-08 02:19:35 -07:00
|
|
|
|
2025-04-29 10:29:37 +08:00
|
|
|
|
2025-09-17 19:49:42 -06:00
|
|
|
|
2026-01-08 02:19:35 -07:00
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
2026-01-08 02:19:35 -07:00
|
|
|
|
2026-01-08 02:19:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
2026-01-08 02:19:39 -07:00
|
|
|
|
2026-05-13 15:18:46 -06:00
|
|
|
|
2026-01-08 02:19:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
2026-01-08 02:19:37 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
2026-01-08 02:19:37 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
2026-01-08 02:19:35 -07:00
|
|
|
|
2026-01-08 02:19:39 -07:00
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:34 -07:00
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:34 -07:00
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:34 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
2026-01-16 22:18:39 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-20 09:09:56 -06:00
|
|
|
|
2025-06-20 09:10:05 -06:00
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
2026-01-16 22:18:39 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
2026-01-16 22:18:39 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:39 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-01 21:31:18 +08:00
|
|
|
|
|
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-10-09 17:33:18 +08:00
|
|
|
|
2022-07-22 18:38:17 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
2025-09-17 19:49:38 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2026-01-16 22:18:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:39 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
2026-01-16 22:18:39 +08:00
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2026-01-16 22:18:39 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2026-01-16 22:18:39 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-11-01 21:31:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-22 18:38:16 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-10-12 10:05:59 -05:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-22 18:38:16 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2022-07-22 18:38:16 +08:00
|
|
|
|
|
|
|
|
|
ublk: speed up ublk server exit handling
Recently, we've observed a few cases where a ublk server is able to
complete restart more quickly than the driver can process the exit of
the previous ublk server. The new ublk server comes up, attempts
recovery of the preexisting ublk devices, and observes them still in
state UBLK_S_DEV_LIVE. While this is possible due to the asynchronous
nature of io_uring cleanup and should therefore be handled properly in
the ublk server, it is still preferable to make ublk server exit
handling faster if possible, as we should strive for it to not be a
limiting factor in how fast a ublk server can restart and provide
service again.
Analysis of the issue showed that the vast majority of the time spent in
handling the ublk server exit was in calls to blk_mq_quiesce_queue,
which is essentially just a (relatively expensive) call to
synchronize_rcu. The ublk server exit path currently issues an
unnecessarily large number of calls to blk_mq_quiesce_queue, for two
reasons:
1. It tries to call blk_mq_quiesce_queue once per ublk_queue. However,
blk_mq_quiesce_queue targets the request_queue of the underlying ublk
device, of which there is only one. So the number of calls is larger
than necessary by a factor of nr_hw_queues.
2. In practice, it calls blk_mq_quiesce_queue _more_ than once per
ublk_queue. This is because of a data race where we read
ubq->canceling without any locking when deciding if we should call
ublk_start_cancel. It is thus possible for two calls to
ublk_uring_cmd_cancel_fn against the same ublk_queue to both call
ublk_start_cancel against the same ublk_queue.
Fix this by making the "canceling" flag a per-device state. This
actually matches the existing code better, as there are several places
where the flag is set or cleared for all queues simultaneously, and
there is the general expectation that cancellation corresponds with ublk
server exit. This per-device canceling flag is then checked under a
(new) lock (addressing the data race (2) above), and the queue is only
quiesced if it is cleared (addressing (1) above). The result is just one
call to blk_mq_quiesce_queue per ublk device.
To minimize the number of cache lines that are accessed in the hot path,
the per-queue canceling flag is kept. The values of the per-device
canceling flag and all per-queue canceling flags should always match.
In our setup, where one ublk server handles I/O for 128 ublk devices,
each having 24 hardware queues of depth 4096, here are the results
before and after this patch, where teardown time is measured from the
first call to io_ring_ctx_wait_and_kill to the return from the last
ublk_ch_release:
before after
number of calls to blk_mq_quiesce_queue: 6469 256
teardown time: 11.14s 2.44s
There are still some potential optimizations here, but this takes care
of a big chunk of the ublk server exit handling delay.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250703-ublk_too_many_quiesce-v2-1-3527b5339eeb@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-07-03 23:41:07 -06:00
|
|
|
|
2022-07-22 18:38:16 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-06-20 20:01:32 +02:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:34 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:10 +08:00
|
|
|
|
2025-02-28 21:31:48 -07:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-30 17:27:50 +08:00
|
|
|
|
2022-07-21 15:09:16 +02:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2022-07-30 17:27:50 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2022-07-30 17:27:50 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2022-07-22 18:38:16 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2026-01-16 22:18:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-22 18:38:16 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 21:31:48 -07:00
|
|
|
|
|
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2025-02-28 21:31:48 -07:00
|
|
|
|
2024-02-23 15:55:38 +08:00
|
|
|
|
2025-02-28 21:31:48 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-14 18:59:52 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: wait on ublk_dev_ready() instead of ub->completion
ub->completion is only re-armed by a successful START_USER_RECOVERY. If
the ublk server sends END_USER_RECOVERY without one - e.g. its START
failed with -EBUSY and the error was ignored - the wait is satisfied by
the stale completion of the previous recovery cycle, and the device is
marked LIVE and the requeue list kicked while the FETCH stream is still
running and ubq->canceling is still set. The kick redispatches a
previously requeued request, __ublk_queue_rq_common() sees ->canceling
and parks it again via __ublk_abort_rq(), and after the last FETCH
clears ->canceling nothing ever kicks the requeue list again: the
request is stranded there while holding its tag. If it is the flush
machinery's flush_rq, every subsequent fsync piles up in uninterruptible
sleep and teardown hangs on tag draining. This matches a report of a
lost PREFLUSH with ext4 on top of ublk after daemon crash recovery.
ub->completion is an edge-triggered latch used as a proxy for the level
condition "every queue has fetched all I/O commands", which can regress
(F_BATCH's UNPREP, daemon death) and whose re-arm can be skipped. Drop
it and wait on the real condition instead: the new helper
ublk_wait_dev_ready_and_lock() waits on ublk_dev_ready() via
wait_var_event_interruptible(), woken from ublk_mark_io_ready(), then
re-checks it under ub->mutex, waiting again on regression, and returns
with the mutex held and readiness guaranteed.
Readiness becomes true in the same ub->mutex critical section that
clears the last queue's ->canceling, so END_USER_RECOVERY marks the
device LIVE and kicks the requeue list strictly after ->canceling
clears. The wait stays interruptible, so a server whose daemon died can
still be signalled out. For ublk_ctrl_start_dev() this replaces the
fail-fast -EINVAL on an F_BATCH ready->UNPREP regression with waiting
until the device is ready again.
Reported-by: George Salisbury <gsalisbury@apnic.net>
Fixes: 728cbac5fe21 ("ublk: move device reset into ublk_ch_release()")
Cc: stable@vger.kernel.org
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Link: https://patch.msgid.link/20260719134540.120269-1-tom.leiming@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2026-07-19 08:45:40 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-08 19:29:26 -06:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2024-02-15 08:02:59 +01:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2024-02-15 08:02:59 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-04-30 15:16:24 -06:00
|
|
|
|
|
|
|
|
|
2024-05-11 08:37:52 -06:00
|
|
|
|
2024-02-15 08:02:59 +01:00
|
|
|
|
2022-07-21 15:09:16 +02:00
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
2024-02-15 08:02:59 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-07-04 14:28:15 +09:00
|
|
|
|
2024-02-15 08:02:59 +01:00
|
|
|
|
|
|
|
|
|
2024-11-08 16:46:51 +01:00
|
|
|
|
2024-02-15 08:02:59 +01:00
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
2024-06-17 08:04:40 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-06-17 08:04:41 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-27 18:37:07 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-27 17:51:15 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:31 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: wait on ublk_dev_ready() instead of ub->completion
ub->completion is only re-armed by a successful START_USER_RECOVERY. If
the ublk server sends END_USER_RECOVERY without one - e.g. its START
failed with -EBUSY and the error was ignored - the wait is satisfied by
the stale completion of the previous recovery cycle, and the device is
marked LIVE and the requeue list kicked while the FETCH stream is still
running and ubq->canceling is still set. The kick redispatches a
previously requeued request, __ublk_queue_rq_common() sees ->canceling
and parks it again via __ublk_abort_rq(), and after the last FETCH
clears ->canceling nothing ever kicks the requeue list again: the
request is stranded there while holding its tag. If it is the flush
machinery's flush_rq, every subsequent fsync piles up in uninterruptible
sleep and teardown hangs on tag draining. This matches a report of a
lost PREFLUSH with ext4 on top of ublk after daemon crash recovery.
ub->completion is an edge-triggered latch used as a proxy for the level
condition "every queue has fetched all I/O commands", which can regress
(F_BATCH's UNPREP, daemon death) and whose re-arm can be skipped. Drop
it and wait on the real condition instead: the new helper
ublk_wait_dev_ready_and_lock() waits on ublk_dev_ready() via
wait_var_event_interruptible(), woken from ublk_mark_io_ready(), then
re-checks it under ub->mutex, waiting again on regression, and returns
with the mutex held and readiness guaranteed.
Readiness becomes true in the same ub->mutex critical section that
clears the last queue's ->canceling, so END_USER_RECOVERY marks the
device LIVE and kicks the requeue list strictly after ->canceling
clears. The wait stays interruptible, so a server whose daemon died can
still be signalled out. For ublk_ctrl_start_dev() this replaces the
fail-fast -EINVAL on an F_BATCH ready->UNPREP regression with waiting
until the device is ready again.
Reported-by: George Salisbury <gsalisbury@apnic.net>
Fixes: 728cbac5fe21 ("ublk: move device reset into ublk_ch_release()")
Cc: stable@vger.kernel.org
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Link: https://patch.msgid.link/20260719134540.120269-1-tom.leiming@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2026-07-19 08:45:40 -05:00
|
|
|
|
2023-07-26 22:45:00 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
ublk: wait on ublk_dev_ready() instead of ub->completion
ub->completion is only re-armed by a successful START_USER_RECOVERY. If
the ublk server sends END_USER_RECOVERY without one - e.g. its START
failed with -EBUSY and the error was ignored - the wait is satisfied by
the stale completion of the previous recovery cycle, and the device is
marked LIVE and the requeue list kicked while the FETCH stream is still
running and ubq->canceling is still set. The kick redispatches a
previously requeued request, __ublk_queue_rq_common() sees ->canceling
and parks it again via __ublk_abort_rq(), and after the last FETCH
clears ->canceling nothing ever kicks the requeue list again: the
request is stranded there while holding its tag. If it is the flush
machinery's flush_rq, every subsequent fsync piles up in uninterruptible
sleep and teardown hangs on tag draining. This matches a report of a
lost PREFLUSH with ext4 on top of ublk after daemon crash recovery.
ub->completion is an edge-triggered latch used as a proxy for the level
condition "every queue has fetched all I/O commands", which can regress
(F_BATCH's UNPREP, daemon death) and whose re-arm can be skipped. Drop
it and wait on the real condition instead: the new helper
ublk_wait_dev_ready_and_lock() waits on ublk_dev_ready() via
wait_var_event_interruptible(), woken from ublk_mark_io_ready(), then
re-checks it under ub->mutex, waiting again on regression, and returns
with the mutex held and readiness guaranteed.
Readiness becomes true in the same ub->mutex critical section that
clears the last queue's ->canceling, so END_USER_RECOVERY marks the
device LIVE and kicks the requeue list strictly after ->canceling
clears. The wait stays interruptible, so a server whose daemon died can
still be signalled out. For ublk_ctrl_start_dev() this replaces the
fail-fast -EINVAL on an F_BATCH ready->UNPREP regression with waiting
until the device is ready again.
Reported-by: George Salisbury <gsalisbury@apnic.net>
Fixes: 728cbac5fe21 ("ublk: move device reset into ublk_ch_release()")
Cc: stable@vger.kernel.org
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Link: https://patch.msgid.link/20260719134540.120269-1-tom.leiming@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2026-07-19 08:45:40 -05:00
|
|
|
|
2026-01-16 22:18:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-21 15:09:16 +02:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2024-02-15 08:02:59 +01:00
|
|
|
|
2022-07-21 15:09:16 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-14 18:59:52 -08:00
|
|
|
|
2022-07-21 15:09:16 +02:00
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
2024-02-15 08:02:59 +01:00
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
2025-12-23 11:27:40 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:07 +08:00
|
|
|
|
2024-02-23 15:55:38 +08:00
|
|
|
|
2023-03-18 22:12:31 +08:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-21 15:09:16 +02:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-30 00:19:51 +08:00
|
|
|
|
|
|
|
|
|
2026-03-08 22:39:02 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-30 00:19:51 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-12-23 11:27:40 +08:00
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
2022-07-21 15:09:16 +02:00
|
|
|
|
2024-12-25 19:06:40 +08:00
|
|
|
|
2022-07-30 17:27:48 +08:00
|
|
|
|
2022-07-21 15:09:16 +02:00
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
2025-04-08 19:29:26 -06:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2022-07-21 15:09:15 +02:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2022-07-21 15:09:15 +02:00
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
|
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2022-07-21 15:09:15 +02:00
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
2022-07-21 15:09:15 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2022-07-21 15:09:15 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2022-07-21 15:09:15 +02:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
2022-07-22 18:38:17 +08:00
|
|
|
|
2022-07-30 17:27:50 +08:00
|
|
|
|
|
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
|
|
|
|
|
2025-04-08 19:29:26 -06:00
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
|
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
2025-06-26 12:20:45 +10:00
|
|
|
|
|
|
|
|
|
2025-06-19 12:10:31 +10:00
|
|
|
|
|
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-10-07 12:24:14 -06:00
|
|
|
|
2024-10-07 12:24:17 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-10-07 12:24:14 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-05-23 00:35:20 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-02 10:42:31 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-10-16 21:48:47 +08:00
|
|
|
|
2023-05-02 10:42:31 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-10-16 21:48:47 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-05-20 12:54:33 +08:00
|
|
|
|
|
|
|
|
|
2024-10-16 21:48:47 +08:00
|
|
|
|
2025-05-20 12:54:33 +08:00
|
|
|
|
|
|
|
|
|
2024-10-16 21:48:47 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:31 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
2023-01-31 12:04:46 +08:00
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2023-10-12 10:05:59 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:10 +08:00
|
|
|
|
2025-02-28 21:31:48 -07:00
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:10 +08:00
|
|
|
|
|
|
|
|
|
2022-07-21 15:09:14 +02:00
|
|
|
|
2026-02-21 17:06:51 -08:00
|
|
|
|
2022-07-21 15:09:14 +02:00
|
|
|
|
|
|
|
|
|
2022-07-22 18:38:16 +08:00
|
|
|
|
2023-10-09 17:33:19 +08:00
|
|
|
|
ublk: speed up ublk server exit handling
Recently, we've observed a few cases where a ublk server is able to
complete restart more quickly than the driver can process the exit of
the previous ublk server. The new ublk server comes up, attempts
recovery of the preexisting ublk devices, and observes them still in
state UBLK_S_DEV_LIVE. While this is possible due to the asynchronous
nature of io_uring cleanup and should therefore be handled properly in
the ublk server, it is still preferable to make ublk server exit
handling faster if possible, as we should strive for it to not be a
limiting factor in how fast a ublk server can restart and provide
service again.
Analysis of the issue showed that the vast majority of the time spent in
handling the ublk server exit was in calls to blk_mq_quiesce_queue,
which is essentially just a (relatively expensive) call to
synchronize_rcu. The ublk server exit path currently issues an
unnecessarily large number of calls to blk_mq_quiesce_queue, for two
reasons:
1. It tries to call blk_mq_quiesce_queue once per ublk_queue. However,
blk_mq_quiesce_queue targets the request_queue of the underlying ublk
device, of which there is only one. So the number of calls is larger
than necessary by a factor of nr_hw_queues.
2. In practice, it calls blk_mq_quiesce_queue _more_ than once per
ublk_queue. This is because of a data race where we read
ubq->canceling without any locking when deciding if we should call
ublk_start_cancel. It is thus possible for two calls to
ublk_uring_cmd_cancel_fn against the same ublk_queue to both call
ublk_start_cancel against the same ublk_queue.
Fix this by making the "canceling" flag a per-device state. This
actually matches the existing code better, as there are several places
where the flag is set or cleared for all queues simultaneously, and
there is the general expectation that cancellation corresponds with ublk
server exit. This per-device canceling flag is then checked under a
(new) lock (addressing the data race (2) above), and the queue is only
quiesced if it is cleared (addressing (1) above). The result is just one
call to blk_mq_quiesce_queue per ublk device.
To minimize the number of cache lines that are accessed in the hot path,
the per-queue canceling flag is kept. The values of the per-device
canceling flag and all per-queue canceling flags should always match.
In our setup, where one ublk server handles I/O for 128 ublk devices,
each having 24 hardware queues of depth 4096, here are the results
before and after this patch, where teardown time is measured from the
first call to io_ring_ctx_wait_and_kill to the return from the last
ublk_ch_release:
before after
number of calls to blk_mq_quiesce_queue: 6469 256
teardown time: 11.14s 2.44s
There are still some potential optimizations here, but this takes care
of a big chunk of the ublk server exit handling delay.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250703-ublk_too_many_quiesce-v2-1-3527b5339eeb@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-07-03 23:41:07 -06:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-04-09 21:30:16 +08:00
|
|
|
|
2025-12-23 11:27:40 +08:00
|
|
|
|
2022-07-21 15:09:14 +02:00
|
|
|
|
2022-07-22 18:38:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-22 18:38:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:24 +08:00
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
2025-06-20 09:10:03 -06:00
|
|
|
|
2026-01-13 00:05:01 +02:00
|
|
|
|
|
|
|
|
|
2022-10-29 09:04:29 +08:00
|
|
|
|
2026-01-16 22:18:44 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-10-29 09:04:29 +08:00
|
|
|
|
2025-04-29 10:29:37 +08:00
|
|
|
|
2025-05-20 12:54:33 +08:00
|
|
|
|
|
|
|
|
|
ublk: support user copy
Currently copy between io request buffer(pages) and userspace buffer is
done inside ublk_map_io() or ublk_unmap_io(). This way performs very
well in case of pre-allocated userspace io buffer.
For dynamically allocated or external userspace backend io buffer,
UBLK_F_NEED_GET_DATA is added for ublk server to provide buffer by one
extra command communication for WRITE request. For READ, userspace
simply provides buffer, but can't know when the buffer is done[1].
Add UBLK_F_USER_COPY by moving io data copy out of kernel by providing
read()/write() on /dev/ublkcN, and simply let ublk server do the io
data copy. This way makes both side cleaner, the cost is that one extra
syscall for copy io data between request and backend buffer.
With UBLK_F_USER_COPY, it actually becomes possible to run per-io zero
copy now, such as, only do zero copy for big size IO, so it can be
thought as one prep patch for supporting zero copy. Meantime zero copy
still needs to expose read()/write() buffer for some corner case, such
as passthrough IO.
[1] READ buffer in UBLK_F_NEED_GET_DATA
https://lore.kernel.org/linux-block/116d8a56-0881-56d3-9bcc-78ff3e1dc4e5@linux.alibaba.com/T/#m23bd4b8634c0a054e6797063167b469949a247bb
ublksrv loop usercopy code:
https://github.com/ming1/ubdsrv/commits/usercopy
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230519065030.351216-8-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-05-19 14:50:30 +08:00
|
|
|
|
2023-04-18 21:18:10 +08:00
|
|
|
|
2026-01-16 22:18:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-29 10:29:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
2025-04-29 10:29:37 +08:00
|
|
|
|
|
|
|
|
|
2023-08-04 13:46:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-22 18:38:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-01 21:31:16 +08:00
|
|
|
|
2022-07-22 18:38:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-01 21:31:16 +08:00
|
|
|
|
2022-07-22 18:38:16 +08:00
|
|
|
|
2025-11-01 21:31:16 +08:00
|
|
|
|
2022-07-22 18:38:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-01 21:31:16 +08:00
|
|
|
|
2022-07-22 18:38:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-01 21:31:16 +08:00
|
|
|
|
|
|
|
|
|
2022-07-22 18:38:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: speed up ublk server exit handling
Recently, we've observed a few cases where a ublk server is able to
complete restart more quickly than the driver can process the exit of
the previous ublk server. The new ublk server comes up, attempts
recovery of the preexisting ublk devices, and observes them still in
state UBLK_S_DEV_LIVE. While this is possible due to the asynchronous
nature of io_uring cleanup and should therefore be handled properly in
the ublk server, it is still preferable to make ublk server exit
handling faster if possible, as we should strive for it to not be a
limiting factor in how fast a ublk server can restart and provide
service again.
Analysis of the issue showed that the vast majority of the time spent in
handling the ublk server exit was in calls to blk_mq_quiesce_queue,
which is essentially just a (relatively expensive) call to
synchronize_rcu. The ublk server exit path currently issues an
unnecessarily large number of calls to blk_mq_quiesce_queue, for two
reasons:
1. It tries to call blk_mq_quiesce_queue once per ublk_queue. However,
blk_mq_quiesce_queue targets the request_queue of the underlying ublk
device, of which there is only one. So the number of calls is larger
than necessary by a factor of nr_hw_queues.
2. In practice, it calls blk_mq_quiesce_queue _more_ than once per
ublk_queue. This is because of a data race where we read
ubq->canceling without any locking when deciding if we should call
ublk_start_cancel. It is thus possible for two calls to
ublk_uring_cmd_cancel_fn against the same ublk_queue to both call
ublk_start_cancel against the same ublk_queue.
Fix this by making the "canceling" flag a per-device state. This
actually matches the existing code better, as there are several places
where the flag is set or cleared for all queues simultaneously, and
there is the general expectation that cancellation corresponds with ublk
server exit. This per-device canceling flag is then checked under a
(new) lock (addressing the data race (2) above), and the queue is only
quiesced if it is cleared (addressing (1) above). The result is just one
call to blk_mq_quiesce_queue per ublk device.
To minimize the number of cache lines that are accessed in the hot path,
the per-queue canceling flag is kept. The values of the per-device
canceling flag and all per-queue canceling flags should always match.
In our setup, where one ublk server handles I/O for 128 ublk devices,
each having 24 hardware queues of depth 4096, here are the results
before and after this patch, where teardown time is measured from the
first call to io_ring_ctx_wait_and_kill to the return from the last
ublk_ch_release:
before after
number of calls to blk_mq_quiesce_queue: 6469 256
teardown time: 11.14s 2.44s
There are still some potential optimizations here, but this takes care
of a big chunk of the ublk server exit handling delay.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250703-ublk_too_many_quiesce-v2-1-3527b5339eeb@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-07-03 23:41:07 -06:00
|
|
|
|
2022-07-22 18:38:16 +08:00
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-02-23 15:55:39 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-02-07 23:07:00 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2023-02-07 23:07:00 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-02-07 23:07:00 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-02-07 23:07:00 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2024-02-23 15:55:39 +08:00
|
|
|
|
2023-07-26 22:45:02 +08:00
|
|
|
|
2023-02-07 23:07:00 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-13 00:05:00 +02:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2026-01-13 00:05:01 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
2025-04-08 19:29:26 -06:00
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
2026-01-14 18:59:52 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-14 18:59:52 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
2022-07-21 15:09:13 +02:00
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:09 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
2025-04-08 19:29:26 -06:00
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:09 +08:00
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
2025-04-08 19:29:26 -06:00
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-04 14:34:26 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
2026-05-06 16:22:38 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-04-06 20:40:59 +08:00
|
|
|
|
2026-05-06 16:22:38 +08:00
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-30 10:14:14 -07:00
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-10-07 12:24:15 -06:00
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-10-07 12:24:17 -06:00
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-10-07 12:24:17 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
2024-10-07 12:24:17 -06:00
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-30 10:14:14 -07:00
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
2025-04-08 19:29:26 -06:00
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: wait on ublk_dev_ready() instead of ub->completion
ub->completion is only re-armed by a successful START_USER_RECOVERY. If
the ublk server sends END_USER_RECOVERY without one - e.g. its START
failed with -EBUSY and the error was ignored - the wait is satisfied by
the stale completion of the previous recovery cycle, and the device is
marked LIVE and the requeue list kicked while the FETCH stream is still
running and ubq->canceling is still set. The kick redispatches a
previously requeued request, __ublk_queue_rq_common() sees ->canceling
and parks it again via __ublk_abort_rq(), and after the last FETCH
clears ->canceling nothing ever kicks the requeue list again: the
request is stranded there while holding its tag. If it is the flush
machinery's flush_rq, every subsequent fsync piles up in uninterruptible
sleep and teardown hangs on tag draining. This matches a report of a
lost PREFLUSH with ext4 on top of ublk after daemon crash recovery.
ub->completion is an edge-triggered latch used as a proxy for the level
condition "every queue has fetched all I/O commands", which can regress
(F_BATCH's UNPREP, daemon death) and whose re-arm can be skipped. Drop
it and wait on the real condition instead: the new helper
ublk_wait_dev_ready_and_lock() waits on ublk_dev_ready() via
wait_var_event_interruptible(), woken from ublk_mark_io_ready(), then
re-checks it under ub->mutex, waiting again on regression, and returns
with the mutex held and readiness guaranteed.
Readiness becomes true in the same ub->mutex critical section that
clears the last queue's ->canceling, so END_USER_RECOVERY marks the
device LIVE and kicks the requeue list strictly after ->canceling
clears. The wait stays interruptible, so a server whose daemon died can
still be signalled out. For ublk_ctrl_start_dev() this replaces the
fail-fast -EINVAL on an F_BATCH ready->UNPREP regression with waiting
until the device is ready again.
Reported-by: George Salisbury <gsalisbury@apnic.net>
Fixes: 728cbac5fe21 ("ublk: move device reset into ublk_ch_release()")
Cc: stable@vger.kernel.org
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Link: https://patch.msgid.link/20260719134540.120269-1-tom.leiming@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2026-07-19 08:45:40 -05:00
|
|
|
|
2023-07-26 22:45:01 +08:00
|
|
|
|
|
|
|
|
|
ublk: have a per-io daemon instead of a per-queue daemon
Currently, ublk_drv associates to each hardware queue (hctx) a unique
task (called the queue's ubq_daemon) which is allowed to issue
COMMIT_AND_FETCH commands against the hctx. If any other task attempts
to do so, the command fails immediately with EINVAL. When considered
together with the block layer architecture, the result is that for each
CPU C on the system, there is a unique ublk server thread which is
allowed to handle I/O submitted on CPU C. This can lead to suboptimal
performance under imbalanced load generation. For an extreme example,
suppose all the load is generated on CPUs mapping to a single ublk
server thread. Then that thread may be fully utilized and become the
bottleneck in the system, while other ublk server threads are totally
idle.
This issue can also be addressed directly in the ublk server without
kernel support by having threads dequeue I/Os and pass them around to
ensure even load. But this solution requires inter-thread communication
at least twice for each I/O (submission and completion), which is
generally a bad pattern for performance. The problem gets even worse
with zero copy, as more inter-thread communication would be required to
have the buffer register/unregister calls to come from the correct
thread.
Therefore, address this issue in ublk_drv by allowing each I/O to have
its own daemon task. Two I/Os in the same queue are now allowed to be
serviced by different daemon tasks - this was not possible before.
Imbalanced load can then be balanced across all ublk server threads by
having the ublk server threads issue FETCH_REQs in a round-robin manner.
As a small toy example, consider a system with a single ublk device
having 2 queues, each of depth 4. A ublk server having 4 threads could
issue its FETCH_REQs against this device as follows (where each entry is
the qid,tag pair that the FETCH_REQ targets):
ublk server thread: T0 T1 T2 T3
0,0 0,1 0,2 0,3
1,3 1,0 1,1 1,2
This setup allows for load that is concentrated on one hctx/ublk_queue
to be spread out across all ublk server threads, alleviating the issue
described above.
Add the new UBLK_F_PER_IO_DAEMON feature to ublk_drv, which ublk servers
can use to essentially test for the presence of this change and tailor
their behavior accordingly.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20250529-ublk_task_per_io-v8-1-e9d3b119336a@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-05-29 17:47:10 -06:00
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
ublk: wait on ublk_dev_ready() instead of ub->completion
ub->completion is only re-armed by a successful START_USER_RECOVERY. If
the ublk server sends END_USER_RECOVERY without one - e.g. its START
failed with -EBUSY and the error was ignored - the wait is satisfied by
the stale completion of the previous recovery cycle, and the device is
marked LIVE and the requeue list kicked while the FETCH stream is still
running and ubq->canceling is still set. The kick redispatches a
previously requeued request, __ublk_queue_rq_common() sees ->canceling
and parks it again via __ublk_abort_rq(), and after the last FETCH
clears ->canceling nothing ever kicks the requeue list again: the
request is stranded there while holding its tag. If it is the flush
machinery's flush_rq, every subsequent fsync piles up in uninterruptible
sleep and teardown hangs on tag draining. This matches a report of a
lost PREFLUSH with ext4 on top of ublk after daemon crash recovery.
ub->completion is an edge-triggered latch used as a proxy for the level
condition "every queue has fetched all I/O commands", which can regress
(F_BATCH's UNPREP, daemon death) and whose re-arm can be skipped. Drop
it and wait on the real condition instead: the new helper
ublk_wait_dev_ready_and_lock() waits on ublk_dev_ready() via
wait_var_event_interruptible(), woken from ublk_mark_io_ready(), then
re-checks it under ub->mutex, waiting again on regression, and returns
with the mutex held and readiness guaranteed.
Readiness becomes true in the same ub->mutex critical section that
clears the last queue's ->canceling, so END_USER_RECOVERY marks the
device LIVE and kicks the requeue list strictly after ->canceling
clears. The wait stays interruptible, so a server whose daemon died can
still be signalled out. For ublk_ctrl_start_dev() this replaces the
fail-fast -EINVAL on an F_BATCH ready->UNPREP regression with waiting
until the device is ready again.
Reported-by: George Salisbury <gsalisbury@apnic.net>
Fixes: 728cbac5fe21 ("ublk: move device reset into ublk_ch_release()")
Cc: stable@vger.kernel.org
Signed-off-by: Ming Lei <tom.leiming@gmail.com>
Link: https://patch.msgid.link/20260719134540.120269-1-tom.leiming@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2026-07-19 08:45:40 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-13 22:33:56 +08:00
|
|
|
|
2024-10-07 12:24:15 -06:00
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
|
|
|
|
|
2024-10-07 12:24:17 -06:00
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-14 18:59:52 -08:00
|
|
|
|
2025-04-16 11:54:38 +08:00
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
|
|
|
|
|
2025-04-16 11:54:37 +08:00
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-08 19:29:26 -06:00
|
|
|
|
2023-06-03 12:06:01 +08:00
|
|
|
|
|
|
|
|
|
2025-02-27 14:39:15 -08:00
|
|
|
|
2023-06-03 12:06:01 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-05 14:31:46 -05:00
|
|
|
|
2025-04-21 13:59:50 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-05 14:31:46 -05:00
|
|
|
|
2025-04-21 13:59:50 +03:00
|
|
|
|
|
|
|
|
|
2026-03-05 14:31:46 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-21 13:59:50 +03:00
|
|
|
|
|
|
|
|
|
2026-03-05 14:31:46 -05:00
|
|
|
|
2025-04-21 13:59:50 +03:00
|
|
|
|
2026-03-05 14:31:46 -05:00
|
|
|
|
2025-04-21 13:59:50 +03:00
|
|
|
|
2025-05-23 00:35:20 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 22:18:44 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-05-23 00:35:20 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-03 23:41:08 -06:00
|
|
|
|
2025-05-23 00:35:20 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk: speed up ublk server exit handling
Recently, we've observed a few cases where a ublk server is able to
complete restart more quickly than the driver can process the exit of
the previous ublk server. The new ublk server comes up, attempts
recovery of the preexisting ublk devices, and observes them still in
state UBLK_S_DEV_LIVE. While this is possible due to the asynchronous
nature of io_uring cleanup and should therefore be handled properly in
the ublk server, it is still preferable to make ublk server exit
handling faster if possible, as we should strive for it to not be a
limiting factor in how fast a ublk server can restart and provide
service again.
Analysis of the issue showed that the vast majority of the time spent in
handling the ublk server exit was in calls to blk_mq_quiesce_queue,
which is essentially just a (relatively expensive) call to
synchronize_rcu. The ublk server exit path currently issues an
unnecessarily large number of calls to blk_mq_quiesce_queue, for two
reasons:
1. It tries to call blk_mq_quiesce_queue once per ublk_queue. However,
blk_mq_quiesce_queue targets the request_queue of the underlying ublk
device, of which there is only one. So the number of calls is larger
than necessary by a factor of nr_hw_queues.
2. In practice, it calls blk_mq_quiesce_queue _more_ than once per
ublk_queue. This is because of a data race where we read
ubq->canceling without any locking when deciding if we should call
ublk_start_cancel. It is thus possible for two calls to
ublk_uring_cmd_cancel_fn against the same ublk_queue to both call
ublk_start_cancel against the same ublk_queue.
Fix this by making the "canceling" flag a per-device state. This
actually matches the existing code better, as there are several places
where the flag is set or cleared for all queues simultaneously, and
there is the general expectation that cancellation corresponds with ublk
server exit. This per-device canceling flag is then checked under a
(new) lock (addressing the data race (2) above), and the queue is only
quiesced if it is cleared (addressing (1) above). The result is just one
call to blk_mq_quiesce_queue per ublk device.
To minimize the number of cache lines that are accessed in the hot path,
the per-queue canceling flag is kept. The values of the per-device
canceling flag and all per-queue canceling flags should always match.
In our setup, where one ublk server handles I/O for 128 ublk devices,
each having 24 hardware queues of depth 4096, here are the results
before and after this patch, where teardown time is measured from the
first call to io_ring_ctx_wait_and_kill to the return from the last
ublk_ch_release:
before after
number of calls to blk_mq_quiesce_queue: 6469 256
teardown time: 11.14s 2.44s
There are still some potential optimizations here, but this takes care
of a big chunk of the ublk server exit handling delay.
Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20250703-ublk_too_many_quiesce-v2-1-3527b5339eeb@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2025-07-03 23:41:07 -06:00
|
|
|
|
2025-07-03 23:41:08 -06:00
|
|
|
|
2025-05-23 00:35:20 +08:00
|
|
|
|
2025-07-03 23:41:08 -06:00
|
|
|
|
2025-05-23 00:35:20 +08:00
|
|
|
|
2025-07-03 23:41:08 -06:00
|
|
|
|
2025-05-23 00:35:20 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
for-6.3/block-2023-02-16
-----BEGIN PGP SIGNATURE-----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=poRc
-----END PGP SIGNATURE-----
Merge tag 'for-6.3/block-2023-02-16' of git://git.kernel.dk/linux
Pull block updates from Jens Axboe:
- NVMe updates via Christoph:
- Small improvements to the logging functionality (Amit Engel)
- Authentication cleanups (Hannes Reinecke)
- Cleanup and optimize the DMA mapping cod in the PCIe driver
(Keith Busch)
- Work around the command effects for Format NVM (Keith Busch)
- Misc cleanups (Keith Busch, Christoph Hellwig)
- Fix and cleanup freeing single sgl (Keith Busch)
- MD updates via Song:
- Fix a rare crash during the takeover process
- Don't update recovery_cp when curr_resync is ACTIVE
- Free writes_pending in md_stop
- Change active_io to percpu
- Updates to drbd, inching us closer to unifying the out-of-tree driver
with the in-tree one (Andreas, Christoph, Lars, Robert)
- BFQ update adding support for multi-actuator drives (Paolo, Federico,
Davide)
- Make brd compliant with REQ_NOWAIT (me)
- Fix for IOPOLL and queue entering, fixing stalled IO waiting on
timeouts (me)
- Fix for REQ_NOWAIT with multiple bios (me)
- Fix memory leak in blktrace cleanup (Greg)
- Clean up sbitmap and fix a potential hang (Kemeng)
- Clean up some bits in BFQ, and fix a bug in the request injection
(Kemeng)
- Clean up the request allocation and issue code, and fix some bugs
related to that (Kemeng)
- ublk updates and fixes:
- Add support for unprivileged ublk (Ming)
- Improve device deletion handling (Ming)
- Misc (Liu, Ziyang)
- s390 dasd fixes (Alexander, Qiheng)
- Improve utility of request caching and fixes (Anuj, Xiao)
- zoned cleanups (Pankaj)
- More constification for kobjs (Thomas)
- blk-iocost cleanups (Yu)
- Remove bio splitting from drivers that don't need it (Christoph)
- Switch blk-cgroups to use struct gendisk. Some of this is now
incomplete as select late reverts were done. (Christoph)
- Add bvec initialization helpers, and convert callers to use that
rather than open-coding it (Christoph)
- Misc fixes and cleanups (Jinke, Keith, Arnd, Bart, Li, Martin,
Matthew, Ulf, Zhong)
* tag 'for-6.3/block-2023-02-16' of git://git.kernel.dk/linux: (169 commits)
brd: use radix_tree_maybe_preload instead of radix_tree_preload
block: use proper return value from bio_failfast()
block: bio-integrity: Copy flags when bio_integrity_payload is cloned
block: Fix io statistics for cgroup in throttle path
brd: mark as nowait compatible
brd: check for REQ_NOWAIT and set correct page allocation mask
brd: return 0/-error from brd_insert_page()
block: sync mixed merged request's failfast with 1st bio's
Revert "blk-cgroup: pin the gendisk in struct blkcg_gq"
Revert "blk-cgroup: pass a gendisk to blkg_lookup"
Revert "blk-cgroup: delay blk-cgroup initialization until add_disk"
Revert "blk-cgroup: delay calling blkcg_exit_disk until disk_release"
Revert "blk-cgroup: move the cgroup information to struct gendisk"
nvme-pci: remove iod use_sgls
nvme-pci: fix freeing single sgl
block: ublk: check IO buffer based on flag need_get_data
s390/dasd: Fix potential memleak in dasd_eckd_init()
s390/dasd: sort out physical vs virtual pointers usage
block: Remove the ALLOC_CACHE_SLACK constant
block: make kobj_type structures constant
...
2023-02-20 14:27:21 -08:00
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-04-09 21:30:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-04-09 21:30:17 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-04-09 21:30:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:17 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-04-09 21:30:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:16 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:15 +08:00
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:15 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-10 20:41:36 +08:00
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-04-09 21:30:17 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-04-09 21:30:16 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:13 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-10 20:41:36 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
2026-04-10 20:41:36 +08:00
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:17 +08:00
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
2026-04-09 21:30:17 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-10 20:41:36 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:17 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-04-09 21:30:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-04-09 21:30:16 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-04-09 21:30:16 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-04-09 21:30:16 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:17 +08:00
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:17 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-23 11:30:57 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-23 11:30:58 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-23 11:30:57 +08:00
|
|
|
|
2026-04-23 11:30:58 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
|
|
|
|
|
2026-04-23 11:30:58 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-23 11:30:58 +08:00
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
2026-04-23 11:30:57 +08:00
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
|
|
|
|
|
2026-04-23 11:30:58 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
|
|
|
|
|
2026-04-23 11:30:58 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
2026-04-23 11:30:58 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:55 +08:00
|
|
|
|
2026-04-09 21:30:16 +08:00
|
|
|
|
2026-04-23 11:30:58 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:16 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:17 +08:00
|
|
|
|
2026-04-09 21:30:16 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:16 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:17 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-04-23 11:30:57 +08:00
|
|
|
|
2026-04-09 21:30:16 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-04-09 21:30:17 +08:00
|
|
|
|
2026-04-09 21:30:16 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-23 11:30:57 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-04-09 21:30:16 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:53 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
2026-03-31 23:31:53 +08:00
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:14 +08:00
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:53 +08:00
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:14 +08:00
|
|
|
|
2026-03-31 23:31:53 +08:00
|
|
|
|
2026-04-09 21:30:14 +08:00
|
|
|
|
2026-03-31 23:31:53 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-09 21:30:14 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:53 +08:00
|
|
|
|
2026-04-09 21:30:14 +08:00
|
|
|
|
2026-03-31 23:31:53 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-15 19:48:14 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
|
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-06-03 12:06:01 +08:00
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-21 13:59:50 +03:00
|
|
|
|
2025-05-23 00:35:20 +08:00
|
|
|
|
2026-01-13 00:05:01 +02:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-12-01 14:41:44 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
2026-02-18 20:59:30 -08:00
|
|
|
|
|
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
2023-04-18 21:18:10 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2025-12-01 14:41:44 -07:00
|
|
|
|
|
|
|
|
|
2023-01-04 21:32:35 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2026-01-30 10:14:12 -07:00
|
|
|
|
|
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2023-04-18 21:18:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-06-03 12:06:01 +08:00
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
2023-06-03 12:06:01 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-04-18 21:18:10 +08:00
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
2023-04-18 21:18:10 +08:00
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
|
|
|
|
|
2023-04-18 21:18:10 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2026-01-13 00:05:00 +02:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2024-02-23 15:55:39 +08:00
|
|
|
|
|
|
|
|
|
2024-07-24 22:33:11 +08:00
|
|
|
|
2024-02-23 15:55:39 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
|
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
ublk_drv: add SET_PARAMS/GET_PARAMS control command
Add two commands to set/get parameters generically.
One important goal of ublk is to provide generic framework for making
block device by userspace flexibly.
As one generic block device, there are still lots of block parameters,
such as max_sectors, write_cache/fua, discard related limits,
zoned parameters, ...., so this patch starts to add generic mechanism
for set/get device parameters.
Both generic block parameters(all kinds of queue settings) and ublk
feature parameters can be covered with this way, then it becomes quite
easy to extend in future.
Add two parameter types are used so far: basic(covers basic queue setting
and misc settings which can't be grouped easily) and discard, basic type
must be set, and discard type becomes optional now
This way provides mechanism to simulate any kind of generic block device
from userspace easily, from both block queue setting viewpoint or ublk
feature viewpoint.
The style of putting all parameters together is suggested by Christoph.
Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220730092750.1118167-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-30 17:27:49 +08:00
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
2026-01-30 10:14:14 -07:00
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
|
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
2022-09-23 23:39:18 +08:00
|
|
|
|
2025-04-21 13:59:50 +03:00
|
|
|
|
2026-03-05 14:31:46 -05:00
|
|
|
|
2025-04-21 13:59:50 +03:00
|
|
|
|
2025-05-23 00:35:20 +08:00
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
2025-05-23 00:35:20 +08:00
|
|
|
|
2026-01-13 00:05:01 +02:00
|
|
|
|
|
|
|
|
|
2025-05-23 00:35:20 +08:00
|
|
|
|
2026-03-31 23:31:52 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2024-11-19 11:06:46 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2022-07-18 09:54:31 +08:00
|
|
|
|
ublk_drv: add mechanism for supporting unprivileged ublk device
unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.
Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.
In case of UBLK_F_UNPRIVILEGED_DEV:
1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.
2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.
Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.
For applying this mechanism, system administrator needs to take
the following policies:
1) chmod 0666 /dev/ublk-control
2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN
Both can be done via one simple udev rule.
Userspace:
https://github.com/ming1/ubdsrv/tree/unprivileged-ublk
'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.
Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-06 12:17:11 +08:00
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:08 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
2026-01-30 10:14:13 -07:00
|
|
|
|
2025-02-25 14:24:55 -07:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
|
|
|
|
|
2026-01-08 02:19:37 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-05-20 12:54:33 +08:00
|
|
|
|
2023-05-19 14:50:29 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-06-20 20:01:32 +02:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2023-06-20 20:01:32 +02:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-06-20 20:01:32 +02:00
|
|
|
|
2023-01-26 19:53:46 +08:00
|
|
|
|
|
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 21:31:48 -07:00
|
|
|
|
|
|
|
|
|
2023-10-12 10:06:00 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 21:31:48 -07:00
|
|
|
|
|
|
|
|
|
2023-10-12 10:06:00 -05:00
|
|
|
|
2025-02-28 21:31:48 -07:00
|
|
|
|
2023-10-12 10:06:00 -05:00
|
|
|
|
|
|
|
|
|
2025-02-28 21:31:48 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-10-12 10:06:00 -05:00
|
|
|
|
|
|
|
|
|
2025-02-28 21:31:48 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 12:17:10 +08:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|
2024-06-02 17:23:26 -07:00
|
|
|
|
ublk_drv: add io_uring based userspace block driver
This is the driver part of userspace block driver(ublk driver), the other
part is userspace daemon part(ublksrv)[1].
The two parts communicate by io_uring's IORING_OP_URING_CMD with one
shared cmd buffer for storing io command, and the buffer is read only for
ublksrv, each io command is indexed by io request tag directly, and is
written by ublk driver.
For example, when one READ io request is submitted to ublk block driver,
ublk driver stores the io command into cmd buffer first, then completes
one IORING_OP_URING_CMD for notifying ublksrv, and the URING_CMD is issued
to ublk driver beforehand by ublksrv for getting notification of any new
io request, and each URING_CMD is associated with one io request by tag.
After ublksrv gets the io command, it translates and handles the ublk io
request, such as, for the ublk-loop target, ublksrv translates the request
into same request on another file or disk, like the kernel loop block
driver. In ublksrv's implementation, the io is still handled by io_uring,
and share same ring with IORING_OP_URING_CMD command. When the target io
request is done, the same IORING_OP_URING_CMD is issued to ublk driver for
both committing io request result and getting future notification of new
io request.
Another thing done by ublk driver is to copy data between kernel io
request and ublksrv's io buffer:
1) before ubsrv handles WRITE request, copy the request's data into
ublksrv's userspace io buffer, so that ublksrv can handle the write
request
2) after ubsrv handles READ request, copy ublksrv's userspace io buffer
into this READ request, then ublk driver can complete the READ request
Zero copy may be switched if mm is ready to support it.
ublk driver doesn't handle any logic of the specific user space driver,
so it is small/simple enough.
[1] ublksrv
https://github.com/ming1/ubdsrv
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20220713140711.97356-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-07-13 22:07:10 +08:00
|
|
|
|