commit bcb44433bb upstream.
Storage devices which report supporting discard commands like
WRITE_SAME_16 with unmap, but reject discard commands sent to the
storage device. This is a clear storage firmware bug but it doesn't
change the fact that should a program cause discards to be sent to a
multipath device layered on this buggy storage, all paths can end up
failed at the same time from the discards, causing possible I/O loss.
The first discard to a path will fail with Illegal Request, Invalid
field in cdb, e.g.:
kernel: sd 8:0:8:19: [sdfn] tag#0 FAILED Result: hostbyte=DID_OK driverbyte=DRIVER_SENSE
kernel: sd 8:0:8:19: [sdfn] tag#0 Sense Key : Illegal Request [current]
kernel: sd 8:0:8:19: [sdfn] tag#0 Add. Sense: Invalid field in cdb
kernel: sd 8:0:8:19: [sdfn] tag#0 CDB: Write same(16) 93 08 00 00 00 00 00 a0 08 00 00 00 80 00 00 00
kernel: blk_update_request: critical target error, dev sdfn, sector 10487808
The SCSI layer converts this to the BLK_STS_TARGET error number, the sd
device disables its support for discard on this path, and because of the
BLK_STS_TARGET error multipath fails the discard without failing any
path or retrying down a different path. But subsequent discards can
cause path failures. Any discards sent to the path which already failed
a discard ends up failing with EIO from blk_cloned_rq_check_limits with
an "over max size limit" error since the discard limit was set to 0 by
the sd driver for the path. As the error is EIO, this now fails the
path and multipath tries to send the discard down the next path. This
cycle continues as discards are sent until all paths fail.
Fix this by training DM core to disable DISCARD if the underlying
storage already did so.
Also, fix branching in dm_done() and clone_endio() to reflect the
mutually exclussive nature of the IO operations in question.
Reported-by: David Jeffery <djeffery@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
[bwh: Backported to 3.16:
- Keep using op & flag to check operation type
- Keep using bdev_get_queue() to find queue in clone_endio()
- WRITE_ZEROES is not handled
- Use queue_flag_clear() instead of blk_queue_flag_clear()
- Adjust filenames, context
- Declare disable_discard() static as its only user is in the same
source file]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit eb40c0acdc upstream.
Some devices don't use blk_integrity but still want stable pages
because they do their own checksumming. Examples include rbd and iSCSI
when data digests are negotiated. Stacking DM (and thus LVM) on top of
these devices results in sporadic checksum errors.
Set BDI_CAP_STABLE_WRITES if any underlying device has it set.
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
[bwh: Backported to 3.16: request_queue::backing_dev_info is a struct
not a pointer]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit e406f12dde upstream.
mddev->sync_thread can be set to NULL on kzalloc failure downstream.
The patch checks for such a scenario and frees allocated resources.
Committer node:
Added similar fix to raid5.c, as suggested by Guoqing.
Acked-by: Guoqing Jiang <gqjiang@suse.com>
Signed-off-by: Aditya Pakki <pakki001@umn.edu>
Signed-off-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit b761dcf121 upstream.
In reshape_request it already adds len to sector_nr already. It's wrong to add len to
sector_nr again after adding pages to bio. If there is bad block it can't copy one chunk
at a time, it needs to goto read_more. Now the sector_nr is wrong. It can cause data
corruption.
Signed-off-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 58ac323084 upstream.
Stale && dirty keys can be produced in the follow way:
After writeback in write_dirty_finish(), dirty keys k1 will
replace by clean keys k2
==>ret = bch_btree_insert(dc->disk.c, &keys, NULL, &w->key);
==>btree_insert_fn(struct btree_op *b_op, struct btree *b)
==>static int bch_btree_insert_node(struct btree *b,
struct btree_op *op,
struct keylist *insert_keys,
atomic_t *journal_ref,
Then two steps:
A) update k1 to k2 in btree node memory;
bch_btree_insert_keys(b, op, insert_keys, replace_key)
B) Write the bset(contains k2) to cache disk by a 30s delay work
bch_btree_leaf_dirty(b, journal_ref).
But before the 30s delay work write the bset to cache device,
these things happened:
A) GC works, and reclaim the bucket k2 point to;
B) Allocator works, and invalidate the bucket k2 point to,
and increase the gen of the bucket, and place it into free_inc
fifo;
C) Until now, the 30s delay work still does not finish work,
so in the disk, the key still is k1, it is dirty and stale
(its gen is smaller than the gen of the bucket). and then the
machine power off suddenly happens;
D) When the machine power on again, after the btree reconstruction,
the stale dirty key appear.
In bch_extent_bad(), when expensive_debug_checks is off, it would
treat the dirty key as good even it is stale keys, and it would
cause bellow probelms:
A) In read_dirty() it would cause machine crash:
BUG_ON(ptr_stale(dc->disk.c, &w->key, 0));
B) It could be worse when reads hits stale dirty keys, it would
read old incorrect data.
This patch tolerate the existence of these stale && dirty keys,
and treat them as bad key in bch_extent_bad().
(Coly Li: fix indent which was modified by sender's email client)
Signed-off-by: Tang Junhui <tang.junhui.linux@gmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
[bwh: Backported to 3.16: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 4ae280b4ee upstream.
When provisioning a new data block for a virtual block, either because
the block was previously unallocated or because we are breaking sharing,
if the whole block of data is being overwritten the bio that triggered
the provisioning is issued immediately, skipping copying or zeroing of
the data block.
When this bio completes the new mapping is inserted in to the pool's
metadata by process_prepared_mapping(), where the bio completion is
signaled to the upper layers.
This completion is signaled without first committing the metadata. If
the bio in question has the REQ_FUA flag set and the system crashes
right after its completion and before the next metadata commit, then the
write is lost despite the REQ_FUA flag requiring that I/O completion for
this request must only be signaled after the data has been committed to
non-volatile storage.
Fix this by deferring the completion of overwrite bios, with the REQ_FUA
flag set, until after the metadata has been committed.
Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com>
Acked-by: Joe Thornber <ejt@redhat.com>
Acked-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
[bwh: Backported to 3.16:
- bio_endio() takes an error parameter
- Adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 800a7340ab upstream.
In copy_params(), the struct 'dm_ioctl' is first copied from the user
space buffer 'user' to 'param_kernel' and the field 'data_size' is
checked against 'minimum_data_size' (size of 'struct dm_ioctl' payload
up to its 'data' member). If the check fails, an error code EINVAL will be
returned. Otherwise, param_kernel->data_size is used to do a second copy,
which copies from the same user-space buffer to 'dmi'. After the second
copy, only 'dmi->data_size' is checked against 'param_kernel->data_size'.
Given that the buffer 'user' resides in the user space, a malicious
user-space process can race to change the content in the buffer between
the two copies. This way, the attacker can inject inconsistent data
into 'dmi' (versus previously validated 'param_kernel').
Fix redundant copying of 'minimum_data_size' from user-space buffer by
using the first copy stored in 'param_kernel'. Also remove the
'data_size' check after the second copy because it is now unnecessary.
Signed-off-by: Wenwen Wang <wang6495@umn.edu>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 2d6cb6edd2 upstream.
refill->end record the last key of writeback, for example, at the first
time, keys (1,128K) to (1,1024K) are flush to the backend device, but
the end key (1,1024K) is not included, since the bellow code:
if (bkey_cmp(k, refill->end) >= 0) {
ret = MAP_DONE;
goto out;
}
And in the next time when we refill writeback keybuf again, we searched
key start from (1,1024K), and got a key bigger than it, so the key
(1,1024K) missed.
This patch modify the above code, and let the end key to be included to
the writeback key buffer.
Signed-off-by: Tang Junhui <tang.junhui.linux@gmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit c157313791 upstream.
Currently, Cache missed IOs are identified by s->cache_miss, but actually,
there are many situations that missed IOs are not assigned a value for
s->cache_miss in cached_dev_cache_miss(), for example, a bypassed IO
(s->iop.bypass = 1), or the cache_bio allocate failed. In these situations,
it will go to out_put or out_submit, and s->cache_miss is null, which leads
bch_mark_cache_accounting() to treat this IO as a hit IO.
[ML: applied by 3-way merge]
Signed-off-by: tang.junhui <tang.junhui@zte.com.cn>
Reviewed-by: Michael Lyle <mlyle@lyle.org>
Reviewed-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit c7cd55504a upstream.
Commit 7e6358d244 ("dm: fix various targets to dm_register_target
after module __init resources created") inadvertently introduced this
bug when it moved dm_register_target() after the call to KMEM_CACHE().
Fixes: 7e6358d244 ("dm: fix various targets to dm_register_target after module __init resources created")
Signed-off-by: Shenghui Wang <shhuiw@foxmail.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 0f843e65d9 upstream.
After write SSD completed, bcache schedules journal_write work to
system_wq, which is a public workqueue in system, without WQ_MEM_RECLAIM
flag. system_wq is also a bound wq, and there may be no idle kworker on
current processor. Creating a new kworker may unfortunately need to
reclaim memory first, by shrinking cache and slab used by vfs, which
depends on bcache device. That's a deadlock.
This patch create a new workqueue for journal_write with WQ_MEM_RECLAIM
flag. It's rescuer thread will work to avoid the deadlock.
Signed-off-by: Guoju Fang <fangguoju@gmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 16c1fdf4cf upstream.
Fixes an error condition reported by checkpatch.pl which is caused by
assigning a variable in an if condition.
Signed-off-by: Florian Schmaus <flo@geekplace.eu>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
[bwh: Backported to 3.16: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 330a4db89d upstream.
mutex_destroy does nothing most of time, but it's better to call
it to make the code future proof and it also has some meaning
for like mutex debug.
As Coly pointed out in a previous review, bcache_exit() may not be
able to handle all the references properly if userspace registers
cache and backing devices right before bch_debug_init runs and
bch_debug_init failes later. So not exposing userspace interface
until everything is ready to avoid that issue.
Signed-off-by: Liang Chen <liangchen.linux@gmail.com>
Reviewed-by: Michael Lyle <mlyle@lyle.org>
Reviewed-by: Coly Li <colyli@suse.de>
Reviewed-by: Eric Wheeler <bcache@linux.ewheeler.net>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 81baf90af2 upstream.
alloc_workqueue replaces deprecated create_workqueue().
Dedicated workqueues have been used since bcache_wq and moving_gc_wq
are workqueues for writes and are being used on a memory reclaim path.
WQ_MEM_RECLAIM has been set to ensure forward progress under memory
pressure.
Since there are only a fixed number of work items, explicit concurrency
limit is unnecessary here.
Signed-off-by: Bhaktipriya Shridhar <bhaktipriya96@gmail.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@fb.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 77b5a08427 upstream.
This is horribly confusing, it breaks the flow of the code without
it being apparent in the caller.
Signed-off-by: Jens Axboe <axboe@fb.com>
Acked-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 432061b3da upstream.
There's a XFS on dm-crypt deadlock, recursing back to itself due to the
crypto subsystems use of GFP_KERNEL, reported here:
https://bugzilla.kernel.org/show_bug.cgi?id=200835
* dm-crypt calls crypt_convert in xts mode
* init_crypt from xts.c calls kmalloc(GFP_KERNEL)
* kmalloc(GFP_KERNEL) recurses into the XFS filesystem, the filesystem
tries to submit some bios and wait for them, causing a deadlock
Fix this by updating both the DM crypt and integrity targets to no
longer use the CRYPTO_TFM_REQ_MAY_SLEEP flag, which will change the
crypto allocations from GFP_KERNEL to GFP_ATOMIC, therefore they can't
recurse into a filesystem. A GFP_ATOMIC allocation can fail, but
init_crypt() in xts.c handles the allocation failure gracefully - it
will fall back to preallocated buffer if the allocation fails.
The crypto API maintainer says that the crypto API only needs to
allocate memory when dealing with unaligned buffers and therefore
turning CRYPTO_TFM_REQ_MAY_SLEEP off is safe (see this discussion:
https://www.redhat.com/archives/dm-devel/2018-August/msg00195.html )
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
[bwh: Backported to 3.16:
- Drop change to crypt_alloc_req_aead() in dm-crypt
- Drop changes to dm-integrity
- Adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit fd2fa95416 upstream.
policy_hint_size starts as 0 during __write_initial_superblock(). It
isn't until the policy is loaded that policy_hint_size is set in-core
(cmd->policy_hint_size). But it never got recorded in the on-disk
superblock because __commit_transaction() didn't deal with transfering
the in-core cmd->policy_hint_size to the on-disk superblock.
The in-core cmd->policy_hint_size gets initialized by metadata_open()'s
__begin_transaction_flags() which re-reads all superblock fields.
Because the superblock's policy_hint_size was never properly stored, when
the cache was created, hints_array_available() would always return false
when re-activating a previously created cache. This means
__load_mappings() always considered the hints invalid and never made use
of the hints (these hints served to optimize).
Another detremental side-effect of this oversight is the cache_check
utility would fail with: "invalid hint width: 0"
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 41c73a49df upstream.
If the first allocation attempt using GFP_NOWAIT fails, drop the lock
and retry using GFP_NOIO allocation (lock is dropped because the
allocation can take some time).
Note that we won't do GFP_NOIO allocation when we loop for the second
time, because the lock shouldn't be dropped between __wait_for_free_buffer
and __get_unclaimed_buffer.
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 9ea61cac0b upstream.
We've seen in-field reports showing _lots_ (18 in one case, 41 in
another) of tasks all sitting there blocked on:
mutex_lock+0x4c/0x68
dm_bufio_shrink_count+0x38/0x78
shrink_slab.part.54.constprop.65+0x100/0x464
shrink_zone+0xa8/0x198
In the two cases analyzed, we see one task that looks like this:
Workqueue: kverityd verity_prefetch_io
__switch_to+0x9c/0xa8
__schedule+0x440/0x6d8
schedule+0x94/0xb4
schedule_timeout+0x204/0x27c
schedule_timeout_uninterruptible+0x44/0x50
wait_iff_congested+0x9c/0x1f0
shrink_inactive_list+0x3a0/0x4cc
shrink_lruvec+0x418/0x5cc
shrink_zone+0x88/0x198
try_to_free_pages+0x51c/0x588
__alloc_pages_nodemask+0x648/0xa88
__get_free_pages+0x34/0x7c
alloc_buffer+0xa4/0x144
__bufio_new+0x84/0x278
dm_bufio_prefetch+0x9c/0x154
verity_prefetch_io+0xe8/0x10c
process_one_work+0x240/0x424
worker_thread+0x2fc/0x424
kthread+0x10c/0x114
...and that looks to be the one holding the mutex.
The problem has been reproduced on fairly easily:
0. Be running Chrome OS w/ verity enabled on the root filesystem
1. Pick test patch: http://crosreview.com/412360
2. Install launchBalloons.sh and balloon.arm from
http://crbug.com/468342
...that's just a memory stress test app.
3. On a 4GB rk3399 machine, run
nice ./launchBalloons.sh 4 900 100000
...that tries to eat 4 * 900 MB of memory and keep accessing.
4. Login to the Chrome web browser and restore many tabs
With that, I've seen printouts like:
DOUG: long bufio 90758 ms
...and stack trace always show's we're in dm_bufio_prefetch().
The problem is that we try to allocate memory with GFP_NOIO while
we're holding the dm_bufio lock. Instead we should be using
GFP_NOWAIT. Using GFP_NOIO can cause us to sleep while holding the
lock and that causes the above problems.
The current behavior explained by David Rientjes:
It will still try reclaim initially because __GFP_WAIT (or
__GFP_KSWAPD_RECLAIM) is set by GFP_NOIO. This is the cause of
contention on dm_bufio_lock() that the thread holds. You want to
pass GFP_NOWAIT instead of GFP_NOIO to alloc_buffer() when holding a
mutex that can be contended by a concurrent slab shrinker (if
count_objects didn't use a trylock, this pattern would trivially
deadlock).
This change significantly increases responsiveness of the system while
in this state. It makes a real difference because it unblocks kswapd.
In the bug report analyzed, kswapd was hung:
kswapd0 D ffffffc000204fd8 0 72 2 0x00000000
Call trace:
[<ffffffc000204fd8>] __switch_to+0x9c/0xa8
[<ffffffc00090b794>] __schedule+0x440/0x6d8
[<ffffffc00090bac0>] schedule+0x94/0xb4
[<ffffffc00090be44>] schedule_preempt_disabled+0x28/0x44
[<ffffffc00090d900>] __mutex_lock_slowpath+0x120/0x1ac
[<ffffffc00090d9d8>] mutex_lock+0x4c/0x68
[<ffffffc000708e7c>] dm_bufio_shrink_count+0x38/0x78
[<ffffffc00030b268>] shrink_slab.part.54.constprop.65+0x100/0x464
[<ffffffc00030dbd8>] shrink_zone+0xa8/0x198
[<ffffffc00030e578>] balance_pgdat+0x328/0x508
[<ffffffc00030eb7c>] kswapd+0x424/0x51c
[<ffffffc00023f06c>] kthread+0x10c/0x114
[<ffffffc000203dd0>] ret_from_fork+0x10/0x40
By unblocking kswapd memory pressure should be reduced.
Suggested-by: David Rientjes <rientjes@google.com>
Reviewed-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit a685557fbb upstream.
Discards issued to a DM thin device can complete to userspace (via
fstrim) _before_ the metadata changes associated with the discards is
reflected in the thinp superblock (e.g. free blocks). As such, if a
user constructs a test that loops repeatedly over these steps, block
allocation can fail due to discards not having completed yet:
1) fill thin device via filesystem file
2) remove file
3) fstrim
From initial report, here:
https://www.redhat.com/archives/dm-devel/2018-April/msg00022.html
"The root cause of this issue is that dm-thin will first remove
mapping and increase corresponding blocks' reference count to prevent
them from being reused before DISCARD bios get processed by the
underlying layers. However. increasing blocks' reference count could
also increase the nr_allocated_this_transaction in struct sm_disk
which makes smd->old_ll.nr_allocated +
smd->nr_allocated_this_transaction bigger than smd->old_ll.nr_blocks.
In this case, alloc_data_block() will never commit metadata to reset
the begin pointer of struct sm_disk, because sm_disk_get_nr_free()
always return an underflow value."
While there is room for improvement to the space-map accounting that
thinp is making use of: the reality is this test is inherently racey and
will result in the previous iteration's fstrim's discard(s) completing
vs concurrent block allocation, via dd, in the next iteration of the
loop.
No amount of space map accounting improvements will be able to allow
user's to use a block before a discard of that block has completed.
So the best we can really do is allow DM thinp to gracefully handle such
aggressive use of all the pool's data by degrading the pool into
out-of-data-space (OODS) mode. We _should_ get that behaviour already
(if space map accounting didn't falsely cause alloc_data_block() to
believe free space was available).. but short of that we handle the
current reality that dm_pool_alloc_data_block() can return -ENOSPC.
Reported-by: Dennis Yang <dennisyang@qnap.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 86755b7a96 upstream.
This can happen e.g. during disk cloning.
This is an incomplete fix: it does not catch duplicate UUIDs earlier
when things are still unattached. It does not unregister the device.
Further changes to cope better with this are planned but conflict with
Coly's ongoing improvements to handling device errors. In the meantime,
one can manually stop the device after this has happened.
Attempts to attach a duplicate device result in:
[ 136.372404] loop: module loaded
[ 136.424461] bcache: register_bdev() registered backing device loop0
[ 136.424464] bcache: bch_cached_dev_attach() Tried to attach loop0 but duplicate UUID already attached
My test procedure is:
dd if=/dev/sdb1 of=imgfile bs=1024 count=262144
losetup -f imgfile
Signed-off-by: Michael Lyle <mlyle@lyle.org>
Reviewed-by: Tang Junhui <tang.junhui@zte.com.cn>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 01a69cab01 upstream.
In the case of 'recover', an r10bio with R10BIO_WriteError &
R10BIO_IsRecover will be progressed by handle_write_completed().
This function traverses all r10bio->devs[copies].
If devs[m].repl_bio != NULL, it thinks conf->mirrors[dev].replacement
is also not NULL. However, this is not always true.
When there is an rdev of raid10 has replacement, then each r10bio
->devs[m].repl_bio != NULL in conf->r10buf_pool. However, in 'recover',
even if corresponded replacement is NULL, it doesn't clear r10bio
->devs[m].repl_bio, resulting in replacement NULL deference.
This bug was introduced when replacement support for raid10 was
added in Linux 3.3.
As NeilBrown suggested:
Elsewhere the determination of "is this device part of the
resync/recovery" is made by resting bio->bi_end_io.
If this is end_sync_write, then we tried to write here.
If it is NULL, then we didn't try to write.
Fixes: 9ad1aefc8a ("md/raid10: Handle replacement devices during resync.")
Cc: stable (V3.3+)
Suggested-by: NeilBrown <neilb@suse.com>
Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Shaohua Li <sh.li@alibaba-inc.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit bc68d0a435 upstream.
When inserting a new key/value pair into a btree we walk down the spine of
btree nodes performing the following 2 operations:
i) space for a new entry
ii) adjusting the first key entry if the new key is lower than any in the node.
If the _root_ node is full, the function btree_split_beneath() allocates 2 new
nodes, and redistibutes the root nodes entries between them. The root node is
left with 2 entries corresponding to the 2 new nodes.
btree_split_beneath() then adjusts the spine to point to one of the two new
children. This means the first key is never adjusted if the new key was lower,
ie. operation (ii) gets missed out. This can result in the new key being
'lost' for a period; until another low valued key is inserted that will uncover
it.
This is a serious bug, and quite hard to make trigger in normal use. A
reproducing test case ("thin create devices-in-reverse-order") is
available as part of the thin-provision-tools project:
https://github.com/jthornber/thin-provisioning-tools/blob/master/functional-tests/device-mapper/dm-tests.scm#L593
Fix the issue by changing btree_split_beneath() so it no longer adjusts
the spine. Instead it unlocks both the new nodes, and lets the main
loop in btree_insert_raw() relock the appropriate one and make any
neccessary adjustments.
Reported-by: Monty Pavel <monty_pavel@sina.com>
Signed-off-by: Joe Thornber <thornber@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 490ae017f5 upstream.
For btree removal, there is a corner case that a single thread
could takes 6 locks which is more than THIN_MAX_CONCURRENT_LOCKS(5)
and leads to deadlock.
A btree removal might eventually call
rebalance_children()->rebalance3() to rebalance entries of three
neighbor child nodes when shadow_spine has already acquired two
write locks. In rebalance3(), it tries to shadow and acquire the
write locks of all three child nodes. However, shadowing a child
node requires acquiring a read lock of the original child node and
a write lock of the new block. Although the read lock will be
released after block shadowing, shadowing the third child node
in rebalance3() could still take the sixth lock.
(2 write locks for shadow_spine +
2 write locks for the first two child nodes's shadow +
1 write lock for the last child node's shadow +
1 read lock for the last child node)
Signed-off-by: Dennis Yang <dennisyang@qnap.com>
Acked-by: Joe Thornber <thornber@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 7e6358d244 upstream.
A NULL pointer is seen if two concurrent "vgchange -ay -K <vg name>"
processes race to load the dm-thin-pool module:
PID: 25992 TASK: ffff883cd7d23500 CPU: 4 COMMAND: "vgchange"
#0 [ffff883cd743d600] machine_kexec at ffffffff81038fa9
0000001 [ffff883cd743d660] crash_kexec at ffffffff810c5992
0000002 [ffff883cd743d730] oops_end at ffffffff81515c90
0000003 [ffff883cd743d760] no_context at ffffffff81049f1b
0000004 [ffff883cd743d7b0] __bad_area_nosemaphore at ffffffff8104a1a5
0000005 [ffff883cd743d800] bad_area at ffffffff8104a2ce
0000006 [ffff883cd743d830] __do_page_fault at ffffffff8104aa6f
0000007 [ffff883cd743d950] do_page_fault at ffffffff81517bae
0000008 [ffff883cd743d980] page_fault at ffffffff81514f95
[exception RIP: kmem_cache_alloc+108]
RIP: ffffffff8116ef3c RSP: ffff883cd743da38 RFLAGS: 00010046
RAX: 0000000000000004 RBX: ffffffff81121b90 RCX: ffff881bf1e78cc0
RDX: 0000000000000000 RSI: 00000000000000d0 RDI: 0000000000000000
RBP: ffff883cd743da68 R8: ffff881bf1a4eb00 R9: 0000000080042000
R10: 0000000000002000 R11: 0000000000000000 R12: 00000000000000d0
R13: 0000000000000000 R14: 00000000000000d0 R15: 0000000000000246
ORIG_RAX: ffffffffffffffff CS: 0010 SS: 0018
0000009 [ffff883cd743da70] mempool_alloc_slab at ffffffff81121ba5
0000010 [ffff883cd743da80] mempool_create_node at ffffffff81122083
0000011 [ffff883cd743dad0] mempool_create at ffffffff811220f4
0000012 [ffff883cd743dae0] pool_ctr at ffffffffa08de049 [dm_thin_pool]
0000013 [ffff883cd743dbd0] dm_table_add_target at ffffffffa0005f2f [dm_mod]
0000014 [ffff883cd743dc30] table_load at ffffffffa0008ba9 [dm_mod]
0000015 [ffff883cd743dc90] ctl_ioctl at ffffffffa0009dc4 [dm_mod]
The race results in a NULL pointer because:
Process A (vgchange -ay -K):
a. send DM_LIST_VERSIONS_CMD ioctl;
b. pool_target not registered;
c. modprobe dm_thin_pool and wait until end.
Process B (vgchange -ay -K):
a. send DM_LIST_VERSIONS_CMD ioctl;
b. pool_target registered;
c. table_load->dm_table_add_target->pool_ctr;
d. _new_mapping_cache is NULL and panic.
Note:
1. process A and process B are two concurrent processes.
2. pool_target can be detected by process B but
_new_mapping_cache initialization has not ended.
To fix dm-thin-pool, and other targets (cache, multipath, and snapshot)
with the same problem, simply dm_register_target() after all resources
created during module init (as labelled with __init) are finished.
Signed-off-by: monty <monty_pavel@sina.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
[bwh: Backported to 3.16: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit ff658e9c1a upstream.
Currently the cleanup of all error cases are open-coded. Introduce a
common exit path and labels.
Signed-off-by: Johannes Thumshirn <morbidrsa@gmail.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit e393aa2446 upstream.
When we send a read request and hit the clean data in cache device, there
is a situation called cache read race in bcache(see the commit in the tail
of cache_look_up(), the following explaination just copy from there):
The bucket we're reading from might be reused while our bio is in flight,
and we could then end up reading the wrong data. We guard against this
by checking (in bch_cache_read_endio()) if the pointer is stale again;
if so, we treat it as an error (s->iop.error = -EINTR) and reread from
the backing device (but we don't pass that error up anywhere)
It should be noted that cache read race happened under normal
circumstances, not the circumstance when SSD failed, it was counted
and shown in /sys/fs/bcache/XXX/internal/cache_read_races.
Without this patch, when we use writeback mode, we will never reread from
the backing device when cache read race happened, until the whole cache
device is clean, because the condition
(s->recoverable && (dc && !atomic_read(&dc->has_dirty))) is false in
cached_dev_read_error(). In this situation, the s->iop.error(= -EINTR)
will be passed up, at last, user will receive -EINTR when it's bio end,
this is not suitable, and wield to up-application.
In this patch, we use s->read_dirty_data to judge whether the read
request hit dirty data in cache device, it is safe to reread data from
the backing device when the read request hit clean data. This can not
only handle cache read race, but also recover data when failed read
request from cache device.
[edited by mlyle to fix up whitespace, commit log title, comment
spelling]
Fixes: d59b237959 ("bcache: only permit to recovery read error when cache device is clean")
Signed-off-by: Hua Rui <huarui.dev@gmail.com>
Reviewed-by: Michael Lyle <mlyle@lyle.org>
Reviewed-by: Coly Li <colyli@suse.de>
Signed-off-by: Michael Lyle <mlyle@lyle.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 74d4108d9e upstream.
The default max_cache_size_bytes for dm-bufio is meant to be the lesser
of 25% of the size of the vmalloc area and 2% of the size of lowmem.
However, on 32-bit systems the intermediate result in the expression
(VMALLOC_END - VMALLOC_START) * DM_BUFIO_VMALLOC_PERCENT / 100
overflows, causing the wrong result to be computed. For example, on a
32-bit system where the vmalloc area is 520093696 bytes, the result is
1174405 rather than the expected 130023424, which makes the maximum
cache size much too small (far less than 2% of lowmem). This causes
severe performance problems for dm-verity users on affected systems.
Fix this by using mult_frac() to correctly multiply by a percentage. Do
this for all places in dm-bufio that multiply by a percentage. Also
replace (VMALLOC_END - VMALLOC_START) with VMALLOC_TOTAL, which contrary
to the comment is now defined in include/linux/vmalloc.h.
Depends-on: 9993bc635 ("sched/x86: Fix overflow in cyc2ns_offset")
Fixes: 95d402f057 ("dm: add bufio")
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 8a74d29d54 upstream.
A DM device with a mix of discard capabilities (due to some underlying
devices not having discard support) _should_ just return -EOPNOTSUPP for
the region of the device that doesn't support discards (even if only by
way of the underlying driver formally not supporting discards). BUT,
that does ask the underlying driver to handle something that it never
advertised support for. In doing so we're exposing users to the
potential for a underlying disk driver hanging if/when a discard is
issued a the device that is incapable and never claimed to support
discards.
Fix this by requiring that each DM target in a DM table provide discard
support as a prereq for a DM device to advertise support for discards.
This may cause some configurations that were happily supporting discards
(even in the face of a mix of discard support) to stop supporting
discards -- but the risk of users hitting driver hangs, and forced
reboots, outweighs supporting those fringe mixed discard
configurations.
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
[bwh: Backported to 3.16: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit b9a41d21dc upstream.
The following BUG_ON was hit when testing repeat creation and removal of
DM devices:
kernel BUG at drivers/md/dm.c:2919!
CPU: 7 PID: 750 Comm: systemd-udevd Not tainted 4.1.44
Call Trace:
[<ffffffff81649e8b>] dm_get_from_kobject+0x34/0x3a
[<ffffffff81650ef1>] dm_attr_show+0x2b/0x5e
[<ffffffff817b46d1>] ? mutex_lock+0x26/0x44
[<ffffffff811df7f5>] sysfs_kf_seq_show+0x83/0xcf
[<ffffffff811de257>] kernfs_seq_show+0x23/0x25
[<ffffffff81199118>] seq_read+0x16f/0x325
[<ffffffff811de994>] kernfs_fop_read+0x3a/0x13f
[<ffffffff8117b625>] __vfs_read+0x26/0x9d
[<ffffffff8130eb59>] ? security_file_permission+0x3c/0x44
[<ffffffff8117bdb8>] ? rw_verify_area+0x83/0xd9
[<ffffffff8117be9d>] vfs_read+0x8f/0xcf
[<ffffffff81193e34>] ? __fdget_pos+0x12/0x41
[<ffffffff8117c686>] SyS_read+0x4b/0x76
[<ffffffff817b606e>] system_call_fastpath+0x12/0x71
The bug can be easily triggered, if an extra delay (e.g. 10ms) is added
between the test of DMF_FREEING & DMF_DELETING and dm_get() in
dm_get_from_kobject().
To fix it, we need to ensure the test of DMF_FREEING & DMF_DELETING and
dm_get() are done in an atomic way, so _minor_lock is used.
The other callers of dm_get() have also been checked to be OK: some
callers invoke dm_get() under _minor_lock, some callers invoke it under
_hash_lock, and dm_start_request() invoke it after increasing
md->open_count.
Signed-off-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit d59b237959 upstream.
When bcache does read I/Os, for example in writeback or writethrough mode,
if a read request on cache device is failed, bcache will try to recovery
the request by reading from cached device. If the data on cached device is
not synced with cache device, then requester will get a stale data.
For critical storage system like database, providing stale data from
recovery may result an application level data corruption, which is
unacceptible.
With this patch, for a failed read request in writeback or writethrough
mode, recovery a recoverable read request only happens when cache device
is clean. That is to say, all data on cached device is up to update.
For other cache modes in bcache, read request will never hit
cached_dev_read_error(), they don't need this patch.
Please note, because cache mode can be switched arbitrarily in run time, a
writethrough mode might be switched from a writeback mode. Therefore
checking dc->has_data in writethrough mode still makes sense.
Changelog:
V4: Fix parens error pointed by Michael Lyle.
v3: By response from Kent Oversteet, he thinks recovering stale data is a
bug to fix, and option to permit it is unnecessary. So this version
the sysfs file is removed.
v2: rename sysfs entry from allow_stale_data_on_failure to
allow_stale_data_on_failure, and fix the confusing commit log.
v1: initial patch posted.
[small change to patch comment spelling by mlyle]
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Michael Lyle <mlyle@lyle.org>
Reported-by: Arne Wolf <awolf@lenovo.com>
Reviewed-by: Michael Lyle <mlyle@lyle.org>
Cc: Kent Overstreet <kent.overstreet@gmail.com>
Cc: Nix <nix@esperi.org.uk>
Cc: Kai Krakow <hurikhan77@gmail.com>
Cc: Eric Wheeler <bcache@lists.ewheeler.net>
Cc: Junhui Tang <tang.junhui@zte.com.cn>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 91af8300d9 upstream.
In bcache code, sysfs entries are created before all resources get
allocated, e.g. allocation thread of a cache set.
There is posibility for NULL pointer deference if a resource is accessed
but which is not initialized yet. Indeed Jorg Bornschein catches one on
cache set allocation thread and gets a kernel oops.
The reason for this bug is, when bch_bucket_alloc() is called during
cache set registration and attaching, ca->alloc_thread is not properly
allocated and initialized yet, call wake_up_process() on ca->alloc_thread
triggers NULL pointer deference failure. A simple and fast fix is, before
waking up ca->alloc_thread, checking whether it is allocated, and only
wake up ca->alloc_thread when it is not NULL.
Signed-off-by: Coly Li <colyli@suse.de>
Reported-by: Jorg Bornschein <jb@capsec.org>
Cc: Kent Overstreet <kent.overstreet@gmail.com>
Reviewed-by: Michael Lyle <mlyle@lyle.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 175206cf9a upstream.
bcache uses a Proportion-Differentiation Controller algorithm to control
writeback rate to cached devices. In the PD controller algorithm, dirty
stripes of thin flash device should not be counted in, because flash only
volumes never write back dirty data.
Currently dirty stripe counter for thin flash device is not initialized
when the thin flash device starts. Which means the following calculation
in PD controller will reference an undefined dirty stripes number, and
all cached devices attached to the same cache set where the thin flash
device lies on may have an inaccurate writeback rate.
This patch calles bch_sectors_dirty_init() in flash_dev_run(), to
correctly initialize dirty stripe counter when the thin flash device
starts to run. This patch also does following parameter data type change,
-void bch_sectors_dirty_init(struct cached_dev *dc);
+void bch_sectors_dirty_init(struct bcache_device *);
to call this function conveniently in flash_dev_run().
(Commit log is composed by Coly Li)
Signed-off-by: Tang Junhui <tang.junhui@zte.com.cn>
Reviewed-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
[bwh: Backported to 3.16: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 9276717b9e upstream.
Most importantly, solve a crash where %llu was used to format signed
numbers. This would cause a buffer overflow when reading sysfs
writeback_rate_debug, as only 20 bytes were allocated for this and
%llu writes 20 characters plus a null.
Always use the units mechanism rather than having different output
paths for simplicity.
Also, correct problems with display output where 1.10 was a larger
number than 1.09, by multiplying by 10 and then dividing by 1024 instead
of dividing by 100. (Remainders of >= 1000 would print as .10).
Minor changes: Always display the decimal point instead of trying to
omit it based on number of digits shown. Decide what units to use
based on 1000 as a threshold, not 1024 (in other words, always print
at most 3 digits before the decimal point).
Signed-off-by: Michael Lyle <mlyle@lyle.org>
Reported-by: Dmitry Yu Okunev <dyokunev@ut.mephi.ru>
Acked-by: Kent Overstreet <kent.overstreet@gmail.com>
Reviewed-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 77fa100f27 upstream.
If you encounter any errors in bch_cached_dev_attach it will return
a negative error code. The variable 'v' which stores the result is
unsigned, thus user space sees a very large value returned for bytes
written which can cause incorrect user space behavior. Utilize 1
signed variable to use throughout the function to preserve error return
capability.
Signed-off-by: Tony Asleson <tasleson@redhat.com>
Acked-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit a8394090a9 upstream.
__update_write_rate() uses a Proportion-Differentiation Controller
algorithm to control writeback rate. A dirty target number is used in
this PD controller to control writeback rate. A larger target number
will make the writeback rate smaller, on the versus, a smaller target
number will make the writeback rate larger.
bcache uses the following steps to calculate the target number,
1) cache_sectors = all-buckets-of-cache-set * buckets-size
2) cache_dirty_target = cache_sectors * cached-device-writeback_percent
3) target = cache_dirty_target *
(sectors-of-cached-device/sectors-of-all-cached-devices-of-this-cache-set)
The calculation at step 1) for cache_sectors is incorrect, which does
not consider dirty blocks occupied by flash only volume.
A flash only volume can be took as a bcache device without cached
device. All data sectors allocated for it are persistent on cache device
and marked dirty, they are not touched by bcache writeback and garbage
collection code. So data blocks of flash only volume should be ignore
when calculating cache_sectors of cache set.
Current code does not subtract dirty sectors of flash only volume, which
results a larger target number from the above 3 steps. And in sequence
the cache device's writeback rate is smaller then a correct value,
writeback speed is slower on all cached devices.
This patch fixes the incorrect slower writeback rate by subtracting
dirty sectors of flash only volumes in __update_writeback_rate().
(Commit log composed by Coly Li to pass checkpatch.pl checking)
Signed-off-by: Tang Junhui <tang.junhui@zte.com.cn>
Reviewed-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 69daf03ade upstream.
Since bypassed IOs use no bucket, so do not subtract sectors_to_gc to
trigger gc thread.
Signed-off-by: tang.junhui <tang.junhui@zte.com.cn>
Acked-by: Coly Li <colyli@suse.de>
Reviewed-by: Eric Wheeler <bcache@linux.ewheeler.net>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
[bwh: Backported to 3.16: adjust context]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit c81ffa32a2 upstream.
Sequential write IOs were tested with bs=1M by FIO in writeback cache
mode, these IOs were expected to be bypassed, but actually they did not.
We debug the code, and find in check_should_bypass():
if (!congested &&
mode == CACHE_MODE_WRITEBACK &&
op_is_write(bio_op(bio)) &&
(bio-＞bi_opf & REQ_SYNC))
goto rescale
that means, If in writeback mode, a write IO with REQ_SYNC flag will not
be bypassed though it is a sequential large IO, It's not a correct thing
to do actually, so this patch remove these codes.
Signed-off-by: tang.junhui <tang.junhui@zte.com.cn>
Reviewed-by: Kent Overstreet <kent.overstreet@gmail.com>
Reviewed-by: Eric Wheeler <bcache@linux.ewheeler.net>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
[bwh: Backported to 3.16: deleted code is slightly different]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 4b758df21e upstream.
If blkdev_get_by_path() in register_bcache() fails, we try to lookup the
block device using lookup_bdev() to detect which situation we are in to
properly report error. However we never drop the reference returned to
us from lookup_bdev(). Fix that.
Signed-off-by: Jan Kara <jack@suse.cz>
Acked-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit e8a27f836f upstream.
bitmap_resize() does not work for file-backed bitmaps.
The buffer_heads are allocated and initialized when
the bitmap is read from the file, but resize doesn't
read from the file, it loads from the internal bitmap.
When it comes time to write the new bitmap, the bh is
non-existent and we crash.
The common case when growing an array involves making the array larger,
and that normally means making the bitmap larger. Doing
that inside the kernel is possible, but would need more code.
It is probably easier to require people who use file-backed
bitmaps to remove them and re-add after a reshape.
So this patch disables the resizing of arrays which have
file-backed bitmaps. This is better than crashing.
Reported-by: Zhilong Liu <zlliu@suse.com>
Fixes: d60b479d17 ("md/bitmap: add bitmap_resize function to allow bitmap resizing.")
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
mips-gcc-5.3 warns about correct code on linux-3.18 and earlier:
In file included from ../include/linux/blkdev.h:4:0,
from ../drivers/md/dm-bufio.h:12,
from ../drivers/md/dm-bufio.c:9:
../drivers/md/dm-bufio.c: In function 'alloc_buffer':
../include/linux/sched.h:1975:56: warning: 'noio_flag' may be used uninitialized in this function [-Wmaybe-uninitialized]
current->flags = (current->flags & ~PF_MEMALLOC_NOIO) | flags;
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~
../drivers/md/dm-bufio.c:325:11: note: 'noio_flag' was declared here
The warning disappeared on later kernels with this commit: be0c37c985
("MIPS: Rearrange PTE bits into fixed positions.") I assume this only
happened because it changed some inlining decisions.
On 3.18.y, we can shut up the warning by adding an extra initialization.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 604407890e upstream.
Using the same rate limiting state for different kinds of messages
is wrong because this can cause a high frequency message to suppress
a report of a low frequency message. Hence use a unique rate limiting
state per message type.
Fixes: 71a16736a1 ("dm: use local printk ratelimit")
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit 7e96d55963 upstream.
Since thread_group worker and raid5d kthread are not in sync, if
worker writes stripe before raid5d then requests will be waiting
for issue_pendig.
Issue observed when building raid5 with ext4, in some build runs
jbd2 would get hung and requests were waiting in the HW engine
waiting to be issued.
Fix this by adding a call to async_tx_issue_pending_all in the
raid5_do_work.
Signed-off-by: Ofer Heifetz <oferh@marvell.com>
Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit b5d27718f3 upstream.
The raid5 md device is created by the disks which we don't use the total size. For example,
the size of the device is 5G and it just uses 3G of the devices to create one raid5 device.
Then change the chunksize and wait reshape to finish. After reshape finishing stop the raid
and assemble it again. It fails.
mdadm -CR /dev/md0 -l5 -n3 /dev/loop[0-2] --size=3G --chunk=32 --assume-clean
mdadm /dev/md0 --grow --chunk=64
wait reshape to finish
mdadm -S /dev/md0
mdadm -As
The error messages:
[197519.814302] md: loop1 does not have a valid v1.2 superblock, not importing!
[197519.821686] md: md_import_device returned -22
After reshape the data offset is changed. It selects backwards direction in this condition.
In function super_1_load it compares the available space of the underlying device with
sb->data_size. The new data offset gets bigger after reshape. So super_1_load returns -EINVAL.
rdev->sectors is updated in md_finish_reshape. Then sb->data_size is set in super_1_sync based
on rdev->sectors. So add md_finish_reshape in end_reshape.
Signed-off-by: Xiao Ni <xni@redhat.com>
Acked-by: Guoqing Jiang <gqjiang@suse.com>
Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit d67a5f4b59 upstream.
Commit df2cb6daa4 ("block: Avoid deadlocks with bio allocation by
stacking drivers") created a workqueue for every bio set and code
in bio_alloc_bioset() that tries to resolve some low-memory deadlocks
by redirecting bios queued on current->bio_list to the workqueue if the
system is low on memory. However other deadlocks (see below **) may
happen, without any low memory condition, because generic_make_request
is queuing bios to current->bio_list (rather than submitting them).
** the related dm-snapshot deadlock is detailed here:
https://www.redhat.com/archives/dm-devel/2016-July/msg00065.html
Fix this deadlock by redirecting any bios on current->bio_list to the
bio_set's rescue workqueue on every schedule() call. Consequently,
when the process blocks on a mutex, the bios queued on
current->bio_list are dispatched to independent workqueus and they can
complete without waiting for the mutex to be available.
The structure blk_plug contains an entry cb_list and this list can contain
arbitrary callback functions that are called when the process blocks.
To implement this fix DM (ab)uses the onstack plug's cb_list interface
to get its flush_current_bio_list() called at schedule() time.
This fixes the snapshot deadlock - if the map method blocks,
flush_current_bio_list() will be called and it redirects bios waiting
on current->bio_list to appropriate workqueues.
Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1267650
Depends-on: df2cb6daa4 ("block: Avoid deadlocks with bio allocation by stacking drivers")
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
commit f9c79bc05a upstream.
The function flush_signals clears all pending signals for the process. It
may be used by kernel threads when we need to prepare a kernel thread for
responding to signals. However using this function for an userspaces
processes is incorrect - clearing signals without the program expecting it
can cause misbehavior.
The raid1 and raid5 code uses flush_signals in its request routine because
it wants to prepare for an interruptible wait. This patch drops
flush_signals and uses sigprocmask instead to block all signals (including
SIGKILL) around the schedule() call. The signals are not lost, but the
schedule() call won't respond to them.
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Acked-by: NeilBrown <neilb@suse.com>
Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
