2019-05-27 08:55:01 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-10-13 19:01:08 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-05-16 19:58:40 +00:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-01-26 10:58:16 +01:00
|
|
|
|
|
|
|
|
|
2016-06-02 15:05:43 -03:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-04-10 12:05:01 -07:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
2009-02-12 05:03:37 +00:00
|
|
|
|
2011-05-20 12:50:29 -07:00
|
|
|
|
2023-04-19 14:52:53 +02:00
|
|
|
|
net: Always untag vlan-tagged traffic on input.
Currently the functionality to untag traffic on input resides
as part of the vlan module and is build only when VLAN support
is enabled in the kernel. When VLAN is disabled, the function
vlan_untag() turns into a stub and doesn't really untag the
packets. This seems to create an interesting interaction
between VMs supporting checksum offloading and some network drivers.
There are some drivers that do not allow the user to change
tx-vlan-offload feature of the driver. These drivers also seem
to assume that any VLAN-tagged traffic they transmit will
have the vlan information in the vlan_tci and not in the vlan
header already in the skb. When transmitting skbs that already
have tagged data with partial checksum set, the checksum doesn't
appear to be updated correctly by the card thus resulting in a
failure to establish TCP connections.
The following is a packet trace taken on the receiver where a
sender is a VM with a VLAN configued. The host VM is running on
doest not have VLAN support and the outging interface on the
host is tg3:
10:12:43.503055 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27243,
offset 0, flags [DF], proto TCP (6), length 60)
10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x48d9), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294837885 ecr 0,nop,wscale 7], length 0
10:12:44.505556 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27244,
offset 0, flags [DF], proto TCP (6), length 60)
10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x44ee), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294838888 ecr 0,nop,wscale 7], length 0
This connection finally times out.
I've only access to the TG3 hardware in this configuration thus have
only tested this with TG3 driver. There are a lot of other drivers
that do not permit user changes to vlan acceleration features, and
I don't know if they all suffere from a similar issue.
The patch attempt to fix this another way. It moves the vlan header
stipping code out of the vlan module and always builds it into the
kernel network core. This way, even if vlan is not supported on
a virtualizatoin host, the virtual machines running on top of such
host will still work with VLANs enabled.
CC: Patrick McHardy <kaber@trash.net>
CC: Nithin Nayak Sujir <nsujir@broadcom.com>
CC: Michael Chan <mchan@broadcom.com>
CC: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Vladislav Yasevich <vyasevic@redhat.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-08-08 14:42:13 -04:00
|
|
|
|
2019-07-07 15:01:57 +01:00
|
|
|
|
2021-02-18 18:31:24 +01:00
|
|
|
|
2023-09-25 13:03:06 +01:00
|
|
|
|
net: add skb_crc32c()
Add skb_crc32c(), which calculates the CRC32C of a sk_buff. It will
replace __skb_checksum(), which unnecessarily supports arbitrary
checksums. Compared to __skb_checksum(), skb_crc32c():
- Uses the correct type for CRC32C values (u32, not __wsum).
- Does not require the caller to provide a skb_checksum_ops struct.
- Is faster because it does not use indirect calls and does not use
the very slow crc32c_combine().
According to commit 2817a336d4d5 ("net: skb_checksum: allow custom
update/combine for walking skb") which added __skb_checksum(), the
original motivation for the abstraction layer was to avoid code
duplication for CRC32C and other checksums in the future. However:
- No additional checksums showed up after CRC32C. __skb_checksum()
is only used with the "regular" net checksum and CRC32C.
- Indirect calls are expensive. Commit 2544af0344ba ("net: avoid
indirect calls in L4 checksum calculation") worked around this
using the INDIRECT_CALL_1 macro. But that only avoided the indirect
call for the net checksum, and at the cost of an extra branch.
- The checksums use different types (__wsum and u32), causing casts
to be needed.
- It made the checksums of fragments be combined (rather than
chained) for both checksums, despite this being highly
counterproductive for CRC32C due to how slow crc32c_combine() is.
This can clearly be seen in commit 4c2f24549644 ("sctp: linearize
early if it's not GSO") which tried to work around this performance
bug. With a dedicated function for each checksum, we can instead
just use the proper strategy for each checksum.
As shown by the following tables, the new function skb_crc32c() is
faster than __skb_checksum(), with the improvement varying greatly from
5% to 2500% depending on the case. The largest improvements come from
fragmented packets, mainly due to eliminating the inefficient
crc32c_combine(). But linear packets are improved too, especially
shorter ones, mainly due to eliminating indirect calls. These
benchmarks were done on AMD Zen 5. On that CPU, Linux uses IBRS instead
of retpoline; an even greater improvement might be seen with retpoline:
Linear sk_buffs
Length in bytes __skb_checksum cycles skb_crc32c cycles
=============== ===================== =================
64 43 18
256 94 77
1420 204 161
16384 1735 1642
Nonlinear sk_buffs (even split between head and one fragment)
Length in bytes __skb_checksum cycles skb_crc32c cycles
=============== ===================== =================
64 579 22
256 829 77
1420 1506 194
16384 4365 1682
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Link: https://patch.msgid.link/20250519175012.36581-3-ebiggers@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2025-05-19 10:50:04 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-18 19:29:39 +01:00
|
|
|
|
2023-06-08 19:17:37 +00:00
|
|
|
|
2024-03-06 16:00:23 +00:00
|
|
|
|
2014-01-09 10:02:46 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2019-07-07 15:01:54 +01:00
|
|
|
|
2020-01-09 07:59:19 -08:00
|
|
|
|
2021-10-29 11:01:44 +08:00
|
|
|
|
2026-05-12 09:48:58 +00:00
|
|
|
|
2026-02-01 15:33:17 +01:00
|
|
|
|
net: skbuff: don't include <net/page_pool/types.h> to <linux/skbuff.h>
Currently, touching <net/page_pool/types.h> triggers a rebuild of more
than half of the kernel. That's because it's included in
<linux/skbuff.h>. And each new include to page_pool/types.h adds more
[useless] data for the toolchain to process per each source file from
that pile.
In commit 6a5bcd84e886 ("page_pool: Allow drivers to hint on SKB
recycling"), Matteo included it to be able to call a couple of functions
defined there. Then, in commit 57f05bc2ab24 ("page_pool: keep pp info as
long as page pool owns the page") one of the calls was removed, so only
one was left. It's the call to page_pool_return_skb_page() in
napi_frag_unref(). The function is external and doesn't have any
dependencies. Having very niche page_pool_types.h included only for that
looks like an overkill.
As %PP_SIGNATURE is not local to page_pool.c (was only in the
early submissions), nothing holds this function there. Teleport
page_pool_return_skb_page() to skbuff.c, just next to the main consumer,
skb_pp_recycle(), and rename it to napi_pp_put_page(), as it doesn't
work with skbs at all and the former name tells nothing. The #if guards
here are only to not compile and have it in the vmlinux when not needed
-- both call sites are already guarded.
Now, touching page_pool_types.h only triggers rebuilding of the drivers
using it and a couple of core networking files.
Suggested-by: Jakub Kicinski <kuba@kernel.org> # make skbuff.h less heavy
Suggested-by: Alexander Duyck <alexanderduyck@fb.com> # move to skbuff.c
Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230804180529.2483231-3-aleksander.lobakin@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2023-08-04 20:05:25 +02:00
|
|
|
|
2025-09-16 17:09:30 -07:00
|
|
|
|
2023-04-19 14:52:53 +02:00
|
|
|
|
2025-10-31 18:33:28 +08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2016-12-24 11:46:01 -08:00
|
|
|
|
2009-04-14 19:39:12 -04:00
|
|
|
|
2012-04-05 11:35:15 +02:00
|
|
|
|
2015-01-30 13:29:32 -05:00
|
|
|
|
|
|
|
|
|
2019-05-29 17:13:48 +02:00
|
|
|
|
2023-01-25 23:14:17 -08:00
|
|
|
|
2006-10-19 16:08:53 -04:00
|
|
|
|
2022-05-15 21:24:55 -07:00
|
|
|
|
2025-05-08 00:48:22 +00:00
|
|
|
|
2026-04-02 11:41:14 +08:00
|
|
|
|
2024-09-10 17:14:49 +00:00
|
|
|
|
2019-03-25 09:17:23 -07:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-02-06 17:31:03 +00:00
|
|
|
|
2025-02-18 19:29:39 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-02-06 17:31:03 +00:00
|
|
|
|
2026-04-10 11:47:32 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-02-06 17:31:03 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-02-14 14:34:03 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: skb: export skb drop reaons to user by TRACE_DEFINE_ENUM
As Eric reported, the 'reason' field is not presented when trace the
kfree_skb event by perf:
$ perf record -e skb:kfree_skb -a sleep 10
$ perf script
ip_defrag 14605 [021] 221.614303: skb:kfree_skb:
skbaddr=0xffff9d2851242700 protocol=34525 location=0xffffffffa39346b1
reason:
The cause seems to be passing kernel address directly to TP_printk(),
which is not right. As the enum 'skb_drop_reason' is not exported to
user space through TRACE_DEFINE_ENUM(), perf can't get the drop reason
string from the 'reason' field, which is a number.
Therefore, we introduce the macro DEFINE_DROP_REASON(), which is used
to define the trace enum by TRACE_DEFINE_ENUM(). With the help of
DEFINE_DROP_REASON(), now we can remove the auto-generate that we
introduced in the commit ec43908dd556
("net: skb: use auto-generation to convert skb drop reason to string"),
and define the string array 'drop_reasons'.
Hmmmm...now we come back to the situation that have to maintain drop
reasons in both enum skb_drop_reason and DEFINE_DROP_REASON. But they
are both in dropreason.h, which makes it easier.
After this commit, now the format of kfree_skb is like this:
$ cat /tracing/events/skb/kfree_skb/format
name: kfree_skb
ID: 1524
format:
field:unsigned short common_type; offset:0; size:2; signed:0;
field:unsigned char common_flags; offset:2; size:1; signed:0;
field:unsigned char common_preempt_count; offset:3; size:1; signed:0;
field:int common_pid; offset:4; size:4; signed:1;
field:void * skbaddr; offset:8; size:8; signed:0;
field:void * location; offset:16; size:8; signed:0;
field:unsigned short protocol; offset:24; size:2; signed:0;
field:enum skb_drop_reason reason; offset:28; size:4; signed:0;
print fmt: "skbaddr=%p protocol=%u location=%p reason: %s", REC->skbaddr, REC->protocol, REC->location, __print_symbolic(REC->reason, { 1, "NOT_SPECIFIED" }, { 2, "NO_SOCKET" } ......
Fixes: ec43908dd556 ("net: skb: use auto-generation to convert skb drop reason to string")
Link: https://lore.kernel.org/netdev/CANn89i+bx0ybvE55iMYf5GJM48WwV1HNpdm9Q6t-HaEstqpCSA@mail.gmail.com/
Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2022-09-05 11:50:15 +08:00
|
|
|
|
|
|
|
|
|
2023-04-19 14:52:53 +02:00
|
|
|
|
2022-10-29 15:45:16 +00:00
|
|
|
|
net: skb: export skb drop reaons to user by TRACE_DEFINE_ENUM
As Eric reported, the 'reason' field is not presented when trace the
kfree_skb event by perf:
$ perf record -e skb:kfree_skb -a sleep 10
$ perf script
ip_defrag 14605 [021] 221.614303: skb:kfree_skb:
skbaddr=0xffff9d2851242700 protocol=34525 location=0xffffffffa39346b1
reason:
The cause seems to be passing kernel address directly to TP_printk(),
which is not right. As the enum 'skb_drop_reason' is not exported to
user space through TRACE_DEFINE_ENUM(), perf can't get the drop reason
string from the 'reason' field, which is a number.
Therefore, we introduce the macro DEFINE_DROP_REASON(), which is used
to define the trace enum by TRACE_DEFINE_ENUM(). With the help of
DEFINE_DROP_REASON(), now we can remove the auto-generate that we
introduced in the commit ec43908dd556
("net: skb: use auto-generation to convert skb drop reason to string"),
and define the string array 'drop_reasons'.
Hmmmm...now we come back to the situation that have to maintain drop
reasons in both enum skb_drop_reason and DEFINE_DROP_REASON. But they
are both in dropreason.h, which makes it easier.
After this commit, now the format of kfree_skb is like this:
$ cat /tracing/events/skb/kfree_skb/format
name: kfree_skb
ID: 1524
format:
field:unsigned short common_type; offset:0; size:2; signed:0;
field:unsigned char common_flags; offset:2; size:1; signed:0;
field:unsigned char common_preempt_count; offset:3; size:1; signed:0;
field:int common_pid; offset:4; size:4; signed:1;
field:void * skbaddr; offset:8; size:8; signed:0;
field:void * location; offset:16; size:8; signed:0;
field:unsigned short protocol; offset:24; size:2; signed:0;
field:enum skb_drop_reason reason; offset:28; size:4; signed:0;
print fmt: "skbaddr=%p protocol=%u location=%p reason: %s", REC->skbaddr, REC->protocol, REC->location, __print_symbolic(REC->reason, { 1, "NOT_SPECIFIED" }, { 2, "NO_SOCKET" } ......
Fixes: ec43908dd556 ("net: skb: use auto-generation to convert skb drop reason to string")
Link: https://lore.kernel.org/netdev/CANn89i+bx0ybvE55iMYf5GJM48WwV1HNpdm9Q6t-HaEstqpCSA@mail.gmail.com/
Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2022-09-05 11:50:15 +08:00
|
|
|
|
|
|
|
|
|
2023-04-19 14:52:53 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-06 10:24:35 +08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2013-02-11 13:30:38 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-02-13 11:20:27 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2013-02-11 13:30:38 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2013-02-11 13:30:38 +00:00
|
|
|
|
2013-02-13 11:20:27 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2020-04-27 18:37:43 +02:00
|
|
|
|
2013-02-13 11:20:27 +00:00
|
|
|
|
2012-05-16 19:58:40 +00:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-02-11 13:30:38 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2013-02-11 13:30:38 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2013-02-11 13:30:38 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-07-31 16:44:19 -07:00
|
|
|
|
2025-11-18 15:06:43 +08:00
|
|
|
|
2025-11-18 15:06:44 +08:00
|
|
|
|
2025-11-18 15:06:45 +08:00
|
|
|
|
2021-02-13 14:12:13 +00:00
|
|
|
|
|
|
|
|
|
2024-06-20 15:21:54 +02:00
|
|
|
|
2021-02-13 14:12:13 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-06-20 15:21:54 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-02-13 14:12:13 +00:00
|
|
|
|
2021-09-14 11:49:35 +08:00
|
|
|
|
2021-02-13 14:12:13 +00:00
|
|
|
|
|
|
|
|
|
2024-06-20 15:21:54 +02:00
|
|
|
|
2021-02-13 14:12:13 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-06-20 15:21:54 +02:00
|
|
|
|
2024-08-01 17:19:56 -07:00
|
|
|
|
|
|
|
|
|
2024-06-20 15:21:54 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-02-13 14:12:13 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-13 22:57:49 +08:00
|
|
|
|
2021-09-14 11:49:35 +08:00
|
|
|
|
|
|
|
|
|
2024-06-20 15:21:53 +02:00
|
|
|
|
2024-08-01 17:19:56 -07:00
|
|
|
|
|
|
|
|
|
2024-02-28 17:30:08 +08:00
|
|
|
|
2021-02-13 14:12:13 +00:00
|
|
|
|
|
|
|
|
|
2024-06-20 15:21:53 +02:00
|
|
|
|
2021-02-13 14:12:13 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-10-16 18:29:11 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-12 13:15:15 +00:00
|
|
|
|
skbuff: introduce {,__}napi_build_skb() which reuses NAPI cache heads
Instead of just bulk-flushing skbuff_heads queued up through
napi_consume_skb() or __kfree_skb_defer(), try to reuse them
on allocation path.
If the cache is empty on allocation, bulk-allocate the first
16 elements, which is more efficient than per-skb allocation.
If the cache is full on freeing, bulk-wipe the second half of
the cache (32 elements).
This also includes custom KASAN poisoning/unpoisoning to be
double sure there are no use-after-free cases.
To not change current behaviour, introduce a new function,
napi_build_skb(), to optionally use a new approach later
in drivers.
Note on selected bulk size, 16:
- this equals to XDP_BULK_QUEUE_SIZE, DEV_MAP_BULK_SIZE
and especially VETH_XDP_BATCH, which is also used to
bulk-allocate skbuff_heads and was tested on powerful
setups;
- this also showed the best performance in the actual
test series (from the array of {8, 16, 32}).
Suggested-by: Edward Cree <ecree.xilinx@gmail.com> # Divide on two halves
Suggested-by: Eric Dumazet <edumazet@google.com> # KASAN poisoning
Cc: Dmitry Vyukov <dvyukov@google.com> # Help with KASAN
Cc: Paolo Abeni <pabeni@redhat.com> # Reduced batch size
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-02-13 14:12:25 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-06-20 15:21:54 +02:00
|
|
|
|
2022-06-15 11:24:26 +08:00
|
|
|
|
2026-05-28 11:34:32 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-06-20 15:21:54 +02:00
|
|
|
|
|
|
|
|
|
2022-06-15 11:24:26 +08:00
|
|
|
|
2024-06-20 15:21:54 +02:00
|
|
|
|
2022-06-15 11:24:26 +08:00
|
|
|
|
skbuff: introduce {,__}napi_build_skb() which reuses NAPI cache heads
Instead of just bulk-flushing skbuff_heads queued up through
napi_consume_skb() or __kfree_skb_defer(), try to reuse them
on allocation path.
If the cache is empty on allocation, bulk-allocate the first
16 elements, which is more efficient than per-skb allocation.
If the cache is full on freeing, bulk-wipe the second half of
the cache (32 elements).
This also includes custom KASAN poisoning/unpoisoning to be
double sure there are no use-after-free cases.
To not change current behaviour, introduce a new function,
napi_build_skb(), to optionally use a new approach later
in drivers.
Note on selected bulk size, 16:
- this equals to XDP_BULK_QUEUE_SIZE, DEV_MAP_BULK_SIZE
and especially VETH_XDP_BATCH, which is also used to
bulk-allocate skbuff_heads and was tested on powerful
setups;
- this also showed the best performance in the actual
test series (from the array of {8, 16, 32}).
Suggested-by: Edward Cree <ecree.xilinx@gmail.com> # Divide on two halves
Suggested-by: Eric Dumazet <edumazet@google.com> # KASAN poisoning
Cc: Dmitry Vyukov <dvyukov@google.com> # Help with KASAN
Cc: Paolo Abeni <pabeni@redhat.com> # Reduced batch size
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-02-13 14:12:25 +00:00
|
|
|
|
|
|
|
|
|
2025-11-18 15:06:46 +08:00
|
|
|
|
|
|
|
|
|
2024-06-20 15:21:54 +02:00
|
|
|
|
2025-10-16 18:29:11 +00:00
|
|
|
|
skbuff: introduce {,__}napi_build_skb() which reuses NAPI cache heads
Instead of just bulk-flushing skbuff_heads queued up through
napi_consume_skb() or __kfree_skb_defer(), try to reuse them
on allocation path.
If the cache is empty on allocation, bulk-allocate the first
16 elements, which is more efficient than per-skb allocation.
If the cache is full on freeing, bulk-wipe the second half of
the cache (32 elements).
This also includes custom KASAN poisoning/unpoisoning to be
double sure there are no use-after-free cases.
To not change current behaviour, introduce a new function,
napi_build_skb(), to optionally use a new approach later
in drivers.
Note on selected bulk size, 16:
- this equals to XDP_BULK_QUEUE_SIZE, DEV_MAP_BULK_SIZE
and especially VETH_XDP_BATCH, which is also used to
bulk-allocate skbuff_heads and was tested on powerful
setups;
- this also showed the best performance in the actual
test series (from the array of {8, 16, 32}).
Suggested-by: Edward Cree <ecree.xilinx@gmail.com> # Divide on two halves
Suggested-by: Eric Dumazet <edumazet@google.com> # KASAN poisoning
Cc: Dmitry Vyukov <dvyukov@google.com> # Help with KASAN
Cc: Paolo Abeni <pabeni@redhat.com> # Reduced batch size
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-02-13 14:12:25 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-09 20:38:36 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-25 18:17:47 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-05-28 11:34:32 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-25 18:17:47 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-05-28 11:34:32 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-25 18:17:47 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-10-16 18:29:11 +00:00
|
|
|
|
2026-01-09 20:38:36 +00:00
|
|
|
|
2025-02-25 18:17:47 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-12-07 22:02:59 -08:00
|
|
|
|
|
|
|
|
|
2019-04-12 17:07:37 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-02-21 19:21:12 -08:00
|
|
|
|
2019-04-12 17:07:37 +02:00
|
|
|
|
|
|
|
|
|
net: generalize skb freeing deferral to per-cpu lists
Logic added in commit f35f821935d8 ("tcp: defer skb freeing after socket
lock is released") helped bulk TCP flows to move the cost of skbs
frees outside of critical section where socket lock was held.
But for RPC traffic, or hosts with RFS enabled, the solution is far from
being ideal.
For RPC traffic, recvmsg() has to return to user space right after
skb payload has been consumed, meaning that BH handler has no chance
to pick the skb before recvmsg() thread. This issue is more visible
with BIG TCP, as more RPC fit one skb.
For RFS, even if BH handler picks the skbs, they are still picked
from the cpu on which user thread is running.
Ideally, it is better to free the skbs (and associated page frags)
on the cpu that originally allocated them.
This patch removes the per socket anchor (sk->defer_list) and
instead uses a per-cpu list, which will hold more skbs per round.
This new per-cpu list is drained at the end of net_action_rx(),
after incoming packets have been processed, to lower latencies.
In normal conditions, skbs are added to the per-cpu list with
no further action. In the (unlikely) cases where the cpu does not
run net_action_rx() handler fast enough, we use an IPI to raise
NET_RX_SOFTIRQ on the remote cpu.
Also, we do not bother draining the per-cpu list from dev_cpu_dead()
This is because skbs in this list have no requirement on how fast
they should be freed.
Note that we can add in the future a small per-cpu cache
if we see any contention on sd->defer_lock.
Tested on a pair of hosts with 100Gbit NIC, RFS enabled,
and /proc/sys/net/ipv4/tcp_rmem[2] tuned to 16MB to work around
page recycling strategy used by NIC driver (its page pool capacity
being too small compared to number of skbs/pages held in sockets
receive queues)
Note that this tuning was only done to demonstrate worse
conditions for skb freeing for this particular test.
These conditions can happen in more general production workload.
10 runs of one TCP_STREAM flow
Before:
Average throughput: 49685 Mbit.
Kernel profiles on cpu running user thread recvmsg() show high cost for
skb freeing related functions (*)
57.81% [kernel] [k] copy_user_enhanced_fast_string
(*) 12.87% [kernel] [k] skb_release_data
(*) 4.25% [kernel] [k] __free_one_page
(*) 3.57% [kernel] [k] __list_del_entry_valid
1.85% [kernel] [k] __netif_receive_skb_core
1.60% [kernel] [k] __skb_datagram_iter
(*) 1.59% [kernel] [k] free_unref_page_commit
(*) 1.16% [kernel] [k] __slab_free
1.16% [kernel] [k] _copy_to_iter
(*) 1.01% [kernel] [k] kfree
(*) 0.88% [kernel] [k] free_unref_page
0.57% [kernel] [k] ip6_rcv_core
0.55% [kernel] [k] ip6t_do_table
0.54% [kernel] [k] flush_smp_call_function_queue
(*) 0.54% [kernel] [k] free_pcppages_bulk
0.51% [kernel] [k] llist_reverse_order
0.38% [kernel] [k] process_backlog
(*) 0.38% [kernel] [k] free_pcp_prepare
0.37% [kernel] [k] tcp_recvmsg_locked
(*) 0.37% [kernel] [k] __list_add_valid
0.34% [kernel] [k] sock_rfree
0.34% [kernel] [k] _raw_spin_lock_irq
(*) 0.33% [kernel] [k] __page_cache_release
0.33% [kernel] [k] tcp_v6_rcv
(*) 0.33% [kernel] [k] __put_page
(*) 0.29% [kernel] [k] __mod_zone_page_state
0.27% [kernel] [k] _raw_spin_lock
After patch:
Average throughput: 73076 Mbit.
Kernel profiles on cpu running user thread recvmsg() looks better:
81.35% [kernel] [k] copy_user_enhanced_fast_string
1.95% [kernel] [k] _copy_to_iter
1.95% [kernel] [k] __skb_datagram_iter
1.27% [kernel] [k] __netif_receive_skb_core
1.03% [kernel] [k] ip6t_do_table
0.60% [kernel] [k] sock_rfree
0.50% [kernel] [k] tcp_v6_rcv
0.47% [kernel] [k] ip6_rcv_core
0.45% [kernel] [k] read_tsc
0.44% [kernel] [k] _raw_spin_lock_irqsave
0.37% [kernel] [k] _raw_spin_lock
0.37% [kernel] [k] native_irq_return_iret
0.33% [kernel] [k] __inet6_lookup_established
0.31% [kernel] [k] ip6_protocol_deliver_rcu
0.29% [kernel] [k] tcp_rcv_established
0.29% [kernel] [k] llist_reverse_order
v2: kdoc issue (kernel bots)
do not defer if (alloc_cpu == smp_processor_id()) (Paolo)
replace the sk_buff_head with a single-linked list (Jakub)
add a READ_ONCE()/WRITE_ONCE() for the lockless read of sd->defer_list
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Link: https://lore.kernel.org/r/20220422201237.416238-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-04-22 13:12:37 -07:00
|
|
|
|
2019-04-12 17:07:37 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-10-29 17:36:19 +00:00
|
|
|
|
2019-04-12 17:07:37 +02:00
|
|
|
|
|
|
|
|
|
2025-07-02 15:38:11 +02:00
|
|
|
|
2022-12-07 22:02:59 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-08-01 17:19:56 -07:00
|
|
|
|
|
|
|
|
|
2022-12-07 22:02:59 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-09 20:38:36 +00:00
|
|
|
|
2025-07-02 15:38:11 +02:00
|
|
|
|
2022-12-07 22:02:59 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-02 15:38:11 +02:00
|
|
|
|
2022-12-07 22:02:59 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: introduce build_skb()
One of the thing we discussed during netdev 2011 conference was the idea
to change some network drivers to allocate/populate their skb at RX
completion time, right before feeding the skb to network stack.
In old days, we allocated skbs when populating the RX ring.
This means bringing into cpu cache sk_buff and skb_shared_info cache
lines (since we clear/initialize them), then 'queue' skb->data to NIC.
By the time NIC fills a frame in skb->data buffer and host can process
it, cpu probably threw away the cache lines from its caches, because lot
of things happened between the allocation and final use.
So the deal would be to allocate only the data buffer for the NIC to
populate its RX ring buffer. And use build_skb() at RX completion to
attach a data buffer (now filled with an ethernet frame) to a new skb,
initialize the skb_shared_info portion, and give the hot skb to network
stack.
build_skb() is the function to allocate an skb, caller providing the
data buffer that should be attached to it. Drivers are expected to call
skb_reserve() right after build_skb() to adjust skb->data to the
Ethernet frame (usually skipping NET_SKB_PAD and NET_IP_ALIGN, but some
drivers might add a hardware provided alignment)
Data provided to build_skb() MUST have been allocated by a prior
kmalloc() call, with enough room to add SKB_DATA_ALIGN(sizeof(struct
skb_shared_info)) bytes at the end of the data without corrupting
incoming frame.
data = kmalloc(NET_SKB_PAD + NET_IP_ALIGN + 1536 +
SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
GFP_ATOMIC);
...
skb = build_skb(data);
if (!skb) {
recycle_data(data);
} else {
skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
...
}
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Eilon Greenstein <eilong@broadcom.com>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Tom Herbert <therbert@google.com>
CC: Jamal Hadi Salim <hadi@mojatatu.com>
CC: Stephen Hemminger <shemminger@vyatta.com>
CC: Thomas Graf <tgraf@infradead.org>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-11-14 06:03:34 +00:00
|
|
|
|
2015-04-24 16:05:01 -07:00
|
|
|
|
net: introduce build_skb()
One of the thing we discussed during netdev 2011 conference was the idea
to change some network drivers to allocate/populate their skb at RX
completion time, right before feeding the skb to network stack.
In old days, we allocated skbs when populating the RX ring.
This means bringing into cpu cache sk_buff and skb_shared_info cache
lines (since we clear/initialize them), then 'queue' skb->data to NIC.
By the time NIC fills a frame in skb->data buffer and host can process
it, cpu probably threw away the cache lines from its caches, because lot
of things happened between the allocation and final use.
So the deal would be to allocate only the data buffer for the NIC to
populate its RX ring buffer. And use build_skb() at RX completion to
attach a data buffer (now filled with an ethernet frame) to a new skb,
initialize the skb_shared_info portion, and give the hot skb to network
stack.
build_skb() is the function to allocate an skb, caller providing the
data buffer that should be attached to it. Drivers are expected to call
skb_reserve() right after build_skb() to adjust skb->data to the
Ethernet frame (usually skipping NET_SKB_PAD and NET_IP_ALIGN, but some
drivers might add a hardware provided alignment)
Data provided to build_skb() MUST have been allocated by a prior
kmalloc() call, with enough room to add SKB_DATA_ALIGN(sizeof(struct
skb_shared_info)) bytes at the end of the data without corrupting
incoming frame.
data = kmalloc(NET_SKB_PAD + NET_IP_ALIGN + 1536 +
SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
GFP_ATOMIC);
...
skb = build_skb(data);
if (!skb) {
recycle_data(data);
} else {
skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
...
}
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Eilon Greenstein <eilong@broadcom.com>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Tom Herbert <therbert@google.com>
CC: Jamal Hadi Salim <hadi@mojatatu.com>
CC: Stephen Hemminger <shemminger@vyatta.com>
CC: Thomas Graf <tgraf@infradead.org>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-11-14 06:03:34 +00:00
|
|
|
|
2022-12-07 22:02:59 -08:00
|
|
|
|
net: introduce build_skb()
One of the thing we discussed during netdev 2011 conference was the idea
to change some network drivers to allocate/populate their skb at RX
completion time, right before feeding the skb to network stack.
In old days, we allocated skbs when populating the RX ring.
This means bringing into cpu cache sk_buff and skb_shared_info cache
lines (since we clear/initialize them), then 'queue' skb->data to NIC.
By the time NIC fills a frame in skb->data buffer and host can process
it, cpu probably threw away the cache lines from its caches, because lot
of things happened between the allocation and final use.
So the deal would be to allocate only the data buffer for the NIC to
populate its RX ring buffer. And use build_skb() at RX completion to
attach a data buffer (now filled with an ethernet frame) to a new skb,
initialize the skb_shared_info portion, and give the hot skb to network
stack.
build_skb() is the function to allocate an skb, caller providing the
data buffer that should be attached to it. Drivers are expected to call
skb_reserve() right after build_skb() to adjust skb->data to the
Ethernet frame (usually skipping NET_SKB_PAD and NET_IP_ALIGN, but some
drivers might add a hardware provided alignment)
Data provided to build_skb() MUST have been allocated by a prior
kmalloc() call, with enough room to add SKB_DATA_ALIGN(sizeof(struct
skb_shared_info)) bytes at the end of the data without corrupting
incoming frame.
data = kmalloc(NET_SKB_PAD + NET_IP_ALIGN + 1536 +
SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
GFP_ATOMIC);
...
skb = build_skb(data);
if (!skb) {
recycle_data(data);
} else {
skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
...
}
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Eilon Greenstein <eilong@broadcom.com>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Tom Herbert <therbert@google.com>
CC: Jamal Hadi Salim <hadi@mojatatu.com>
CC: Stephen Hemminger <shemminger@vyatta.com>
CC: Thomas Graf <tgraf@infradead.org>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-11-14 06:03:34 +00:00
|
|
|
|
|
|
|
|
|
2022-12-07 22:02:59 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: introduce build_skb()
One of the thing we discussed during netdev 2011 conference was the idea
to change some network drivers to allocate/populate their skb at RX
completion time, right before feeding the skb to network stack.
In old days, we allocated skbs when populating the RX ring.
This means bringing into cpu cache sk_buff and skb_shared_info cache
lines (since we clear/initialize them), then 'queue' skb->data to NIC.
By the time NIC fills a frame in skb->data buffer and host can process
it, cpu probably threw away the cache lines from its caches, because lot
of things happened between the allocation and final use.
So the deal would be to allocate only the data buffer for the NIC to
populate its RX ring buffer. And use build_skb() at RX completion to
attach a data buffer (now filled with an ethernet frame) to a new skb,
initialize the skb_shared_info portion, and give the hot skb to network
stack.
build_skb() is the function to allocate an skb, caller providing the
data buffer that should be attached to it. Drivers are expected to call
skb_reserve() right after build_skb() to adjust skb->data to the
Ethernet frame (usually skipping NET_SKB_PAD and NET_IP_ALIGN, but some
drivers might add a hardware provided alignment)
Data provided to build_skb() MUST have been allocated by a prior
kmalloc() call, with enough room to add SKB_DATA_ALIGN(sizeof(struct
skb_shared_info)) bytes at the end of the data without corrupting
incoming frame.
data = kmalloc(NET_SKB_PAD + NET_IP_ALIGN + 1536 +
SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
GFP_ATOMIC);
...
skb = build_skb(data);
if (!skb) {
recycle_data(data);
} else {
skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
...
}
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Eilon Greenstein <eilong@broadcom.com>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Tom Herbert <therbert@google.com>
CC: Jamal Hadi Salim <hadi@mojatatu.com>
CC: Stephen Hemminger <shemminger@vyatta.com>
CC: Thomas Graf <tgraf@infradead.org>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-11-14 06:03:34 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-04-24 16:05:01 -07:00
|
|
|
|
net: introduce build_skb()
One of the thing we discussed during netdev 2011 conference was the idea
to change some network drivers to allocate/populate their skb at RX
completion time, right before feeding the skb to network stack.
In old days, we allocated skbs when populating the RX ring.
This means bringing into cpu cache sk_buff and skb_shared_info cache
lines (since we clear/initialize them), then 'queue' skb->data to NIC.
By the time NIC fills a frame in skb->data buffer and host can process
it, cpu probably threw away the cache lines from its caches, because lot
of things happened between the allocation and final use.
So the deal would be to allocate only the data buffer for the NIC to
populate its RX ring buffer. And use build_skb() at RX completion to
attach a data buffer (now filled with an ethernet frame) to a new skb,
initialize the skb_shared_info portion, and give the hot skb to network
stack.
build_skb() is the function to allocate an skb, caller providing the
data buffer that should be attached to it. Drivers are expected to call
skb_reserve() right after build_skb() to adjust skb->data to the
Ethernet frame (usually skipping NET_SKB_PAD and NET_IP_ALIGN, but some
drivers might add a hardware provided alignment)
Data provided to build_skb() MUST have been allocated by a prior
kmalloc() call, with enough room to add SKB_DATA_ALIGN(sizeof(struct
skb_shared_info)) bytes at the end of the data without corrupting
incoming frame.
data = kmalloc(NET_SKB_PAD + NET_IP_ALIGN + 1536 +
SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
GFP_ATOMIC);
...
skb = build_skb(data);
if (!skb) {
recycle_data(data);
} else {
skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
...
}
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Eilon Greenstein <eilong@broadcom.com>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Tom Herbert <therbert@google.com>
CC: Jamal Hadi Salim <hadi@mojatatu.com>
CC: Stephen Hemminger <shemminger@vyatta.com>
CC: Thomas Graf <tgraf@infradead.org>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-11-14 06:03:34 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-08-01 17:19:56 -07:00
|
|
|
|
|
|
|
|
|
2019-04-12 17:07:37 +02:00
|
|
|
|
net: introduce build_skb()
One of the thing we discussed during netdev 2011 conference was the idea
to change some network drivers to allocate/populate their skb at RX
completion time, right before feeding the skb to network stack.
In old days, we allocated skbs when populating the RX ring.
This means bringing into cpu cache sk_buff and skb_shared_info cache
lines (since we clear/initialize them), then 'queue' skb->data to NIC.
By the time NIC fills a frame in skb->data buffer and host can process
it, cpu probably threw away the cache lines from its caches, because lot
of things happened between the allocation and final use.
So the deal would be to allocate only the data buffer for the NIC to
populate its RX ring buffer. And use build_skb() at RX completion to
attach a data buffer (now filled with an ethernet frame) to a new skb,
initialize the skb_shared_info portion, and give the hot skb to network
stack.
build_skb() is the function to allocate an skb, caller providing the
data buffer that should be attached to it. Drivers are expected to call
skb_reserve() right after build_skb() to adjust skb->data to the
Ethernet frame (usually skipping NET_SKB_PAD and NET_IP_ALIGN, but some
drivers might add a hardware provided alignment)
Data provided to build_skb() MUST have been allocated by a prior
kmalloc() call, with enough room to add SKB_DATA_ALIGN(sizeof(struct
skb_shared_info)) bytes at the end of the data without corrupting
incoming frame.
data = kmalloc(NET_SKB_PAD + NET_IP_ALIGN + 1536 +
SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
GFP_ATOMIC);
...
skb = build_skb(data);
if (!skb) {
recycle_data(data);
} else {
skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
...
}
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Eilon Greenstein <eilong@broadcom.com>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Tom Herbert <therbert@google.com>
CC: Jamal Hadi Salim <hadi@mojatatu.com>
CC: Stephen Hemminger <shemminger@vyatta.com>
CC: Thomas Graf <tgraf@infradead.org>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-11-14 06:03:34 +00:00
|
|
|
|
|
|
|
|
|
2026-01-09 20:38:36 +00:00
|
|
|
|
2021-02-13 14:11:26 +00:00
|
|
|
|
net: introduce build_skb()
One of the thing we discussed during netdev 2011 conference was the idea
to change some network drivers to allocate/populate their skb at RX
completion time, right before feeding the skb to network stack.
In old days, we allocated skbs when populating the RX ring.
This means bringing into cpu cache sk_buff and skb_shared_info cache
lines (since we clear/initialize them), then 'queue' skb->data to NIC.
By the time NIC fills a frame in skb->data buffer and host can process
it, cpu probably threw away the cache lines from its caches, because lot
of things happened between the allocation and final use.
So the deal would be to allocate only the data buffer for the NIC to
populate its RX ring buffer. And use build_skb() at RX completion to
attach a data buffer (now filled with an ethernet frame) to a new skb,
initialize the skb_shared_info portion, and give the hot skb to network
stack.
build_skb() is the function to allocate an skb, caller providing the
data buffer that should be attached to it. Drivers are expected to call
skb_reserve() right after build_skb() to adjust skb->data to the
Ethernet frame (usually skipping NET_SKB_PAD and NET_IP_ALIGN, but some
drivers might add a hardware provided alignment)
Data provided to build_skb() MUST have been allocated by a prior
kmalloc() call, with enough room to add SKB_DATA_ALIGN(sizeof(struct
skb_shared_info)) bytes at the end of the data without corrupting
incoming frame.
data = kmalloc(NET_SKB_PAD + NET_IP_ALIGN + 1536 +
SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
GFP_ATOMIC);
...
skb = build_skb(data);
if (!skb) {
recycle_data(data);
} else {
skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
...
}
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Eilon Greenstein <eilong@broadcom.com>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Tom Herbert <therbert@google.com>
CC: Jamal Hadi Salim <hadi@mojatatu.com>
CC: Stephen Hemminger <shemminger@vyatta.com>
CC: Thomas Graf <tgraf@infradead.org>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-11-14 06:03:34 +00:00
|
|
|
|
2021-02-13 14:11:26 +00:00
|
|
|
|
net: introduce build_skb()
One of the thing we discussed during netdev 2011 conference was the idea
to change some network drivers to allocate/populate their skb at RX
completion time, right before feeding the skb to network stack.
In old days, we allocated skbs when populating the RX ring.
This means bringing into cpu cache sk_buff and skb_shared_info cache
lines (since we clear/initialize them), then 'queue' skb->data to NIC.
By the time NIC fills a frame in skb->data buffer and host can process
it, cpu probably threw away the cache lines from its caches, because lot
of things happened between the allocation and final use.
So the deal would be to allocate only the data buffer for the NIC to
populate its RX ring buffer. And use build_skb() at RX completion to
attach a data buffer (now filled with an ethernet frame) to a new skb,
initialize the skb_shared_info portion, and give the hot skb to network
stack.
build_skb() is the function to allocate an skb, caller providing the
data buffer that should be attached to it. Drivers are expected to call
skb_reserve() right after build_skb() to adjust skb->data to the
Ethernet frame (usually skipping NET_SKB_PAD and NET_IP_ALIGN, but some
drivers might add a hardware provided alignment)
Data provided to build_skb() MUST have been allocated by a prior
kmalloc() call, with enough room to add SKB_DATA_ALIGN(sizeof(struct
skb_shared_info)) bytes at the end of the data without corrupting
incoming frame.
data = kmalloc(NET_SKB_PAD + NET_IP_ALIGN + 1536 +
SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
GFP_ATOMIC);
...
skb = build_skb(data);
if (!skb) {
recycle_data(data);
} else {
skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
...
}
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Eilon Greenstein <eilong@broadcom.com>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Tom Herbert <therbert@google.com>
CC: Jamal Hadi Salim <hadi@mojatatu.com>
CC: Stephen Hemminger <shemminger@vyatta.com>
CC: Thomas Graf <tgraf@infradead.org>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-11-14 06:03:34 +00:00
|
|
|
|
2015-04-24 16:05:01 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-03-08 15:17:20 +02:00
|
|
|
|
2015-04-24 16:05:01 -07:00
|
|
|
|
2023-03-08 15:17:19 +02:00
|
|
|
|
2015-04-24 16:05:01 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: introduce build_skb()
One of the thing we discussed during netdev 2011 conference was the idea
to change some network drivers to allocate/populate their skb at RX
completion time, right before feeding the skb to network stack.
In old days, we allocated skbs when populating the RX ring.
This means bringing into cpu cache sk_buff and skb_shared_info cache
lines (since we clear/initialize them), then 'queue' skb->data to NIC.
By the time NIC fills a frame in skb->data buffer and host can process
it, cpu probably threw away the cache lines from its caches, because lot
of things happened between the allocation and final use.
So the deal would be to allocate only the data buffer for the NIC to
populate its RX ring buffer. And use build_skb() at RX completion to
attach a data buffer (now filled with an ethernet frame) to a new skb,
initialize the skb_shared_info portion, and give the hot skb to network
stack.
build_skb() is the function to allocate an skb, caller providing the
data buffer that should be attached to it. Drivers are expected to call
skb_reserve() right after build_skb() to adjust skb->data to the
Ethernet frame (usually skipping NET_SKB_PAD and NET_IP_ALIGN, but some
drivers might add a hardware provided alignment)
Data provided to build_skb() MUST have been allocated by a prior
kmalloc() call, with enough room to add SKB_DATA_ALIGN(sizeof(struct
skb_shared_info)) bytes at the end of the data without corrupting
incoming frame.
data = kmalloc(NET_SKB_PAD + NET_IP_ALIGN + 1536 +
SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
GFP_ATOMIC);
...
skb = build_skb(data);
if (!skb) {
recycle_data(data);
} else {
skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
...
}
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Eilon Greenstein <eilong@broadcom.com>
CC: Ben Hutchings <bhutchings@solarflare.com>
CC: Tom Herbert <therbert@google.com>
CC: Jamal Hadi Salim <hadi@mojatatu.com>
CC: Stephen Hemminger <shemminger@vyatta.com>
CC: Thomas Graf <tgraf@infradead.org>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2011-11-14 06:03:34 +00:00
|
|
|
|
|
|
|
|
|
2019-04-12 17:07:37 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 18:29:04 -08:00
|
|
|
|
2019-04-12 17:07:37 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-02-13 14:11:26 +00:00
|
|
|
|
2019-04-12 17:07:37 +02:00
|
|
|
|
2021-02-13 14:11:26 +00:00
|
|
|
|
2019-04-12 17:07:37 +02:00
|
|
|
|
2023-03-08 15:17:19 +02:00
|
|
|
|
2019-04-12 17:07:37 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
skbuff: introduce {,__}napi_build_skb() which reuses NAPI cache heads
Instead of just bulk-flushing skbuff_heads queued up through
napi_consume_skb() or __kfree_skb_defer(), try to reuse them
on allocation path.
If the cache is empty on allocation, bulk-allocate the first
16 elements, which is more efficient than per-skb allocation.
If the cache is full on freeing, bulk-wipe the second half of
the cache (32 elements).
This also includes custom KASAN poisoning/unpoisoning to be
double sure there are no use-after-free cases.
To not change current behaviour, introduce a new function,
napi_build_skb(), to optionally use a new approach later
in drivers.
Note on selected bulk size, 16:
- this equals to XDP_BULK_QUEUE_SIZE, DEV_MAP_BULK_SIZE
and especially VETH_XDP_BATCH, which is also used to
bulk-allocate skbuff_heads and was tested on powerful
setups;
- this also showed the best performance in the actual
test series (from the array of {8, 16, 32}).
Suggested-by: Edward Cree <ecree.xilinx@gmail.com> # Divide on two halves
Suggested-by: Eric Dumazet <edumazet@google.com> # KASAN poisoning
Cc: Dmitry Vyukov <dvyukov@google.com> # Help with KASAN
Cc: Paolo Abeni <pabeni@redhat.com> # Reduced batch size
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-02-13 14:12:25 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 18:29:04 -08:00
|
|
|
|
skbuff: introduce {,__}napi_build_skb() which reuses NAPI cache heads
Instead of just bulk-flushing skbuff_heads queued up through
napi_consume_skb() or __kfree_skb_defer(), try to reuse them
on allocation path.
If the cache is empty on allocation, bulk-allocate the first
16 elements, which is more efficient than per-skb allocation.
If the cache is full on freeing, bulk-wipe the second half of
the cache (32 elements).
This also includes custom KASAN poisoning/unpoisoning to be
double sure there are no use-after-free cases.
To not change current behaviour, introduce a new function,
napi_build_skb(), to optionally use a new approach later
in drivers.
Note on selected bulk size, 16:
- this equals to XDP_BULK_QUEUE_SIZE, DEV_MAP_BULK_SIZE
and especially VETH_XDP_BATCH, which is also used to
bulk-allocate skbuff_heads and was tested on powerful
setups;
- this also showed the best performance in the actual
test series (from the array of {8, 16, 32}).
Suggested-by: Edward Cree <ecree.xilinx@gmail.com> # Divide on two halves
Suggested-by: Eric Dumazet <edumazet@google.com> # KASAN poisoning
Cc: Dmitry Vyukov <dvyukov@google.com> # Help with KASAN
Cc: Paolo Abeni <pabeni@redhat.com> # Reduced batch size
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-02-13 14:12:25 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-16 20:27:15 +00:00
|
|
|
|
skbuff: introduce {,__}napi_build_skb() which reuses NAPI cache heads
Instead of just bulk-flushing skbuff_heads queued up through
napi_consume_skb() or __kfree_skb_defer(), try to reuse them
on allocation path.
If the cache is empty on allocation, bulk-allocate the first
16 elements, which is more efficient than per-skb allocation.
If the cache is full on freeing, bulk-wipe the second half of
the cache (32 elements).
This also includes custom KASAN poisoning/unpoisoning to be
double sure there are no use-after-free cases.
To not change current behaviour, introduce a new function,
napi_build_skb(), to optionally use a new approach later
in drivers.
Note on selected bulk size, 16:
- this equals to XDP_BULK_QUEUE_SIZE, DEV_MAP_BULK_SIZE
and especially VETH_XDP_BATCH, which is also used to
bulk-allocate skbuff_heads and was tested on powerful
setups;
- this also showed the best performance in the actual
test series (from the array of {8, 16, 32}).
Suggested-by: Edward Cree <ecree.xilinx@gmail.com> # Divide on two halves
Suggested-by: Eric Dumazet <edumazet@google.com> # KASAN poisoning
Cc: Dmitry Vyukov <dvyukov@google.com> # Help with KASAN
Cc: Paolo Abeni <pabeni@redhat.com> # Reduced batch size
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-02-13 14:12:25 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-09 20:38:36 +00:00
|
|
|
|
skbuff: introduce {,__}napi_build_skb() which reuses NAPI cache heads
Instead of just bulk-flushing skbuff_heads queued up through
napi_consume_skb() or __kfree_skb_defer(), try to reuse them
on allocation path.
If the cache is empty on allocation, bulk-allocate the first
16 elements, which is more efficient than per-skb allocation.
If the cache is full on freeing, bulk-wipe the second half of
the cache (32 elements).
This also includes custom KASAN poisoning/unpoisoning to be
double sure there are no use-after-free cases.
To not change current behaviour, introduce a new function,
napi_build_skb(), to optionally use a new approach later
in drivers.
Note on selected bulk size, 16:
- this equals to XDP_BULK_QUEUE_SIZE, DEV_MAP_BULK_SIZE
and especially VETH_XDP_BATCH, which is also used to
bulk-allocate skbuff_heads and was tested on powerful
setups;
- this also showed the best performance in the actual
test series (from the array of {8, 16, 32}).
Suggested-by: Edward Cree <ecree.xilinx@gmail.com> # Divide on two halves
Suggested-by: Eric Dumazet <edumazet@google.com> # KASAN poisoning
Cc: Dmitry Vyukov <dvyukov@google.com> # Help with KASAN
Cc: Paolo Abeni <pabeni@redhat.com> # Reduced batch size
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-02-13 14:12:25 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-06 18:29:04 -08:00
|
|
|
|
skbuff: introduce {,__}napi_build_skb() which reuses NAPI cache heads
Instead of just bulk-flushing skbuff_heads queued up through
napi_consume_skb() or __kfree_skb_defer(), try to reuse them
on allocation path.
If the cache is empty on allocation, bulk-allocate the first
16 elements, which is more efficient than per-skb allocation.
If the cache is full on freeing, bulk-wipe the second half of
the cache (32 elements).
This also includes custom KASAN poisoning/unpoisoning to be
double sure there are no use-after-free cases.
To not change current behaviour, introduce a new function,
napi_build_skb(), to optionally use a new approach later
in drivers.
Note on selected bulk size, 16:
- this equals to XDP_BULK_QUEUE_SIZE, DEV_MAP_BULK_SIZE
and especially VETH_XDP_BATCH, which is also used to
bulk-allocate skbuff_heads and was tested on powerful
setups;
- this also showed the best performance in the actual
test series (from the array of {8, 16, 32}).
Suggested-by: Edward Cree <ecree.xilinx@gmail.com> # Divide on two halves
Suggested-by: Eric Dumazet <edumazet@google.com> # KASAN poisoning
Cc: Dmitry Vyukov <dvyukov@google.com> # Help with KASAN
Cc: Paolo Abeni <pabeni@redhat.com> # Reduced batch size
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-02-13 14:12:25 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 04:13:59 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-02-06 17:31:02 +00:00
|
|
|
|
2026-01-16 04:13:59 +00:00
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
2023-08-31 18:37:50 +00:00
|
|
|
|
2023-02-06 17:31:02 +00:00
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
2023-02-06 17:31:02 +00:00
|
|
|
|
2023-02-06 17:31:03 +00:00
|
|
|
|
|
|
|
|
|
2024-03-06 16:00:23 +00:00
|
|
|
|
2023-02-06 17:31:03 +00:00
|
|
|
|
|
|
|
|
|
net: avoid skb end_offset change in __skb_unclone_keeptruesize()
Once initial skb->head has been allocated from skb_small_head_cache,
we need to make sure to use the same strategy whenever skb->head
has to be re-allocated, as found by syzbot [1]
This means kmalloc_reserve() can not fallback from using
skb_small_head_cache to generic (power-of-two) kmem caches.
It seems that we probably want to rework things in the future,
to partially revert following patch, because we no longer use
ksize() for skb allocated in TX path.
2b88cba55883 ("net: preserve skb_end_offset() in skb_unclone_keeptruesize()")
Ideally, TCP stack should never put payload in skb->head,
this effort has to be completed.
In the mean time, add a sanity check.
[1]
BUG: KASAN: invalid-free in slab_free mm/slub.c:3787 [inline]
BUG: KASAN: invalid-free in kmem_cache_free+0xee/0x5c0 mm/slub.c:3809
Free of addr ffff88806cdee800 by task syz-executor239/5189
CPU: 0 PID: 5189 Comm: syz-executor239 Not tainted 6.2.0-rc8-syzkaller-02400-gd1fabc68f8e0 #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/21/2023
Call Trace:
<TASK>
__dump_stack lib/dump_stack.c:88 [inline]
dump_stack_lvl+0xd1/0x138 lib/dump_stack.c:106
print_address_description mm/kasan/report.c:306 [inline]
print_report+0x15e/0x45d mm/kasan/report.c:417
kasan_report_invalid_free+0x9b/0x1b0 mm/kasan/report.c:482
____kasan_slab_free+0x1a5/0x1c0 mm/kasan/common.c:216
kasan_slab_free include/linux/kasan.h:177 [inline]
slab_free_hook mm/slub.c:1781 [inline]
slab_free_freelist_hook+0x8b/0x1c0 mm/slub.c:1807
slab_free mm/slub.c:3787 [inline]
kmem_cache_free+0xee/0x5c0 mm/slub.c:3809
skb_kfree_head net/core/skbuff.c:857 [inline]
skb_kfree_head net/core/skbuff.c:853 [inline]
skb_free_head+0x16f/0x1a0 net/core/skbuff.c:872
skb_release_data+0x57a/0x820 net/core/skbuff.c:901
skb_release_all net/core/skbuff.c:966 [inline]
__kfree_skb+0x4f/0x70 net/core/skbuff.c:980
tcp_wmem_free_skb include/net/tcp.h:302 [inline]
tcp_rtx_queue_purge net/ipv4/tcp.c:3061 [inline]
tcp_write_queue_purge+0x617/0xcf0 net/ipv4/tcp.c:3074
tcp_v4_destroy_sock+0x125/0x810 net/ipv4/tcp_ipv4.c:2302
inet_csk_destroy_sock+0x19a/0x440 net/ipv4/inet_connection_sock.c:1195
__tcp_close+0xb96/0xf50 net/ipv4/tcp.c:3021
tcp_close+0x2d/0xc0 net/ipv4/tcp.c:3033
inet_release+0x132/0x270 net/ipv4/af_inet.c:426
__sock_release+0xcd/0x280 net/socket.c:651
sock_close+0x1c/0x20 net/socket.c:1393
__fput+0x27c/0xa90 fs/file_table.c:320
task_work_run+0x16f/0x270 kernel/task_work.c:179
resume_user_mode_work include/linux/resume_user_mode.h:49 [inline]
exit_to_user_mode_loop kernel/entry/common.c:171 [inline]
exit_to_user_mode_prepare+0x23c/0x250 kernel/entry/common.c:203
__syscall_exit_to_user_mode_work kernel/entry/common.c:285 [inline]
syscall_exit_to_user_mode+0x1d/0x50 kernel/entry/common.c:296
do_syscall_64+0x46/0xb0 arch/x86/entry/common.c:86
entry_SYSCALL_64_after_hwframe+0x63/0xcd
RIP: 0033:0x7f2511f546c3
Code: c7 c2 c0 ff ff ff f7 d8 64 89 02 48 c7 c0 ff ff ff ff eb b7 0f 1f 00 64 8b 04 25 18 00 00 00 85 c0 75 14 b8 03 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 45 c3 0f 1f 40 00 48 83 ec 18 89 7c 24 0c e8
RSP: 002b:00007ffef0103d48 EFLAGS: 00000246 ORIG_RAX: 0000000000000003
RAX: 0000000000000000 RBX: 0000000000000004 RCX: 00007f2511f546c3
RDX: 0000000000000978 RSI: 00000000200000c0 RDI: 0000000000000003
RBP: 0000000000000000 R08: 0000000000000002 R09: 0000000000003434
R10: 0000000000000000 R11: 0000000000000246 R12: 00007ffef0103d6c
R13: 00007ffef0103d80 R14: 00007ffef0103dc0 R15: 0000000000000003
</TASK>
Allocated by task 5189:
kasan_save_stack+0x22/0x40 mm/kasan/common.c:45
kasan_set_track+0x25/0x30 mm/kasan/common.c:52
____kasan_kmalloc mm/kasan/common.c:374 [inline]
____kasan_kmalloc mm/kasan/common.c:333 [inline]
__kasan_kmalloc+0xa5/0xb0 mm/kasan/common.c:383
kasan_kmalloc include/linux/kasan.h:211 [inline]
__do_kmalloc_node mm/slab_common.c:968 [inline]
__kmalloc_node_track_caller+0x5b/0xc0 mm/slab_common.c:988
kmalloc_reserve+0xf1/0x230 net/core/skbuff.c:539
pskb_expand_head+0x237/0x1160 net/core/skbuff.c:1995
__skb_unclone_keeptruesize+0x93/0x220 net/core/skbuff.c:2094
skb_unclone_keeptruesize include/linux/skbuff.h:1910 [inline]
skb_prepare_for_shift net/core/skbuff.c:3804 [inline]
skb_shift+0xef8/0x1e20 net/core/skbuff.c:3877
tcp_skb_shift net/ipv4/tcp_input.c:1538 [inline]
tcp_shift_skb_data net/ipv4/tcp_input.c:1646 [inline]
tcp_sacktag_walk+0x93b/0x18a0 net/ipv4/tcp_input.c:1713
tcp_sacktag_write_queue+0x1599/0x31d0 net/ipv4/tcp_input.c:1974
tcp_ack+0x2e9f/0x5a10 net/ipv4/tcp_input.c:3847
tcp_rcv_established+0x667/0x2230 net/ipv4/tcp_input.c:6006
tcp_v4_do_rcv+0x670/0x9b0 net/ipv4/tcp_ipv4.c:1721
sk_backlog_rcv include/net/sock.h:1113 [inline]
__release_sock+0x133/0x3b0 net/core/sock.c:2921
release_sock+0x58/0x1b0 net/core/sock.c:3488
tcp_sendmsg+0x3a/0x50 net/ipv4/tcp.c:1485
inet_sendmsg+0x9d/0xe0 net/ipv4/af_inet.c:825
sock_sendmsg_nosec net/socket.c:722 [inline]
sock_sendmsg+0xde/0x190 net/socket.c:745
sock_write_iter+0x295/0x3d0 net/socket.c:1136
call_write_iter include/linux/fs.h:2189 [inline]
new_sync_write fs/read_write.c:491 [inline]
vfs_write+0x9ed/0xdd0 fs/read_write.c:584
ksys_write+0x1ec/0x250 fs/read_write.c:637
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x39/0xb0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x63/0xcd
The buggy address belongs to the object at ffff88806cdee800
which belongs to the cache kmalloc-1k of size 1024
The buggy address is located 0 bytes inside of
1024-byte region [ffff88806cdee800, ffff88806cdeec00)
The buggy address belongs to the physical page:
page:ffffea0001b37a00 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x6cde8
head:ffffea0001b37a00 order:3 compound_mapcount:0 subpages_mapcount:0 compound_pincount:0
flags: 0xfff00000010200(slab|head|node=0|zone=1|lastcpupid=0x7ff)
raw: 00fff00000010200 ffff888012441dc0 dead000000000122 0000000000000000
raw: 0000000000000000 0000000000100010 00000001ffffffff 0000000000000000
page dumped because: kasan: bad access detected
page_owner tracks the page as allocated
page last allocated via order 3, migratetype Unmovable, gfp_mask 0x1f2a20(GFP_ATOMIC|__GFP_NOWARN|__GFP_NORETRY|__GFP_COMP|__GFP_NOMEMALLOC|__GFP_MEMALLOC|__GFP_HARDWALL), pid 75, tgid 75 (kworker/u4:4), ts 96369578780, free_ts 26734162530
prep_new_page mm/page_alloc.c:2531 [inline]
get_page_from_freelist+0x119c/0x2ce0 mm/page_alloc.c:4283
__alloc_pages+0x1cb/0x5b0 mm/page_alloc.c:5549
alloc_pages+0x1aa/0x270 mm/mempolicy.c:2287
alloc_slab_page mm/slub.c:1851 [inline]
allocate_slab+0x25f/0x350 mm/slub.c:1998
new_slab mm/slub.c:2051 [inline]
___slab_alloc+0xa91/0x1400 mm/slub.c:3193
__slab_alloc.constprop.0+0x56/0xa0 mm/slub.c:3292
__slab_alloc_node mm/slub.c:3345 [inline]
slab_alloc_node mm/slub.c:3442 [inline]
__kmem_cache_alloc_node+0x1a4/0x430 mm/slub.c:3491
__do_kmalloc_node mm/slab_common.c:967 [inline]
__kmalloc_node_track_caller+0x4b/0xc0 mm/slab_common.c:988
kmalloc_reserve+0xf1/0x230 net/core/skbuff.c:539
__alloc_skb+0x129/0x330 net/core/skbuff.c:608
__netdev_alloc_skb+0x74/0x410 net/core/skbuff.c:672
__netdev_alloc_skb_ip_align include/linux/skbuff.h:3203 [inline]
netdev_alloc_skb_ip_align include/linux/skbuff.h:3213 [inline]
batadv_iv_ogm_aggregate_new+0x106/0x4e0 net/batman-adv/bat_iv_ogm.c:558
batadv_iv_ogm_queue_add net/batman-adv/bat_iv_ogm.c:670 [inline]
batadv_iv_ogm_schedule_buff+0xe6b/0x1450 net/batman-adv/bat_iv_ogm.c:849
batadv_iv_ogm_schedule net/batman-adv/bat_iv_ogm.c:868 [inline]
batadv_iv_ogm_schedule net/batman-adv/bat_iv_ogm.c:861 [inline]
batadv_iv_send_outstanding_bat_ogm_packet+0x744/0x910 net/batman-adv/bat_iv_ogm.c:1712
process_one_work+0x9bf/0x1710 kernel/workqueue.c:2289
worker_thread+0x669/0x1090 kernel/workqueue.c:2436
page last free stack trace:
reset_page_owner include/linux/page_owner.h:24 [inline]
free_pages_prepare mm/page_alloc.c:1446 [inline]
free_pcp_prepare+0x66a/0xc20 mm/page_alloc.c:1496
free_unref_page_prepare mm/page_alloc.c:3369 [inline]
free_unref_page+0x1d/0x490 mm/page_alloc.c:3464
free_contig_range+0xb5/0x180 mm/page_alloc.c:9488
destroy_args+0xa8/0x64c mm/debug_vm_pgtable.c:998
debug_vm_pgtable+0x28de/0x296f mm/debug_vm_pgtable.c:1318
do_one_initcall+0x141/0x790 init/main.c:1306
do_initcall_level init/main.c:1379 [inline]
do_initcalls init/main.c:1395 [inline]
do_basic_setup init/main.c:1414 [inline]
kernel_init_freeable+0x6f9/0x782 init/main.c:1634
kernel_init+0x1e/0x1d0 init/main.c:1522
ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308
Memory state around the buggy address:
ffff88806cdee700: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
ffff88806cdee780: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
>ffff88806cdee800: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
^
ffff88806cdee880: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
Fixes: bf9f1baa279f ("net: add dedicated kmem_cache for typical/small skb->head")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Tested-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2023-02-27 14:17:06 +00:00
|
|
|
|
2026-01-16 04:13:59 +00:00
|
|
|
|
2023-02-06 17:31:03 +00:00
|
|
|
|
net: avoid skb end_offset change in __skb_unclone_keeptruesize()
Once initial skb->head has been allocated from skb_small_head_cache,
we need to make sure to use the same strategy whenever skb->head
has to be re-allocated, as found by syzbot [1]
This means kmalloc_reserve() can not fallback from using
skb_small_head_cache to generic (power-of-two) kmem caches.
It seems that we probably want to rework things in the future,
to partially revert following patch, because we no longer use
ksize() for skb allocated in TX path.
2b88cba55883 ("net: preserve skb_end_offset() in skb_unclone_keeptruesize()")
Ideally, TCP stack should never put payload in skb->head,
this effort has to be completed.
In the mean time, add a sanity check.
[1]
BUG: KASAN: invalid-free in slab_free mm/slub.c:3787 [inline]
BUG: KASAN: invalid-free in kmem_cache_free+0xee/0x5c0 mm/slub.c:3809
Free of addr ffff88806cdee800 by task syz-executor239/5189
CPU: 0 PID: 5189 Comm: syz-executor239 Not tainted 6.2.0-rc8-syzkaller-02400-gd1fabc68f8e0 #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/21/2023
Call Trace:
<TASK>
__dump_stack lib/dump_stack.c:88 [inline]
dump_stack_lvl+0xd1/0x138 lib/dump_stack.c:106
print_address_description mm/kasan/report.c:306 [inline]
print_report+0x15e/0x45d mm/kasan/report.c:417
kasan_report_invalid_free+0x9b/0x1b0 mm/kasan/report.c:482
____kasan_slab_free+0x1a5/0x1c0 mm/kasan/common.c:216
kasan_slab_free include/linux/kasan.h:177 [inline]
slab_free_hook mm/slub.c:1781 [inline]
slab_free_freelist_hook+0x8b/0x1c0 mm/slub.c:1807
slab_free mm/slub.c:3787 [inline]
kmem_cache_free+0xee/0x5c0 mm/slub.c:3809
skb_kfree_head net/core/skbuff.c:857 [inline]
skb_kfree_head net/core/skbuff.c:853 [inline]
skb_free_head+0x16f/0x1a0 net/core/skbuff.c:872
skb_release_data+0x57a/0x820 net/core/skbuff.c:901
skb_release_all net/core/skbuff.c:966 [inline]
__kfree_skb+0x4f/0x70 net/core/skbuff.c:980
tcp_wmem_free_skb include/net/tcp.h:302 [inline]
tcp_rtx_queue_purge net/ipv4/tcp.c:3061 [inline]
tcp_write_queue_purge+0x617/0xcf0 net/ipv4/tcp.c:3074
tcp_v4_destroy_sock+0x125/0x810 net/ipv4/tcp_ipv4.c:2302
inet_csk_destroy_sock+0x19a/0x440 net/ipv4/inet_connection_sock.c:1195
__tcp_close+0xb96/0xf50 net/ipv4/tcp.c:3021
tcp_close+0x2d/0xc0 net/ipv4/tcp.c:3033
inet_release+0x132/0x270 net/ipv4/af_inet.c:426
__sock_release+0xcd/0x280 net/socket.c:651
sock_close+0x1c/0x20 net/socket.c:1393
__fput+0x27c/0xa90 fs/file_table.c:320
task_work_run+0x16f/0x270 kernel/task_work.c:179
resume_user_mode_work include/linux/resume_user_mode.h:49 [inline]
exit_to_user_mode_loop kernel/entry/common.c:171 [inline]
exit_to_user_mode_prepare+0x23c/0x250 kernel/entry/common.c:203
__syscall_exit_to_user_mode_work kernel/entry/common.c:285 [inline]
syscall_exit_to_user_mode+0x1d/0x50 kernel/entry/common.c:296
do_syscall_64+0x46/0xb0 arch/x86/entry/common.c:86
entry_SYSCALL_64_after_hwframe+0x63/0xcd
RIP: 0033:0x7f2511f546c3
Code: c7 c2 c0 ff ff ff f7 d8 64 89 02 48 c7 c0 ff ff ff ff eb b7 0f 1f 00 64 8b 04 25 18 00 00 00 85 c0 75 14 b8 03 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 45 c3 0f 1f 40 00 48 83 ec 18 89 7c 24 0c e8
RSP: 002b:00007ffef0103d48 EFLAGS: 00000246 ORIG_RAX: 0000000000000003
RAX: 0000000000000000 RBX: 0000000000000004 RCX: 00007f2511f546c3
RDX: 0000000000000978 RSI: 00000000200000c0 RDI: 0000000000000003
RBP: 0000000000000000 R08: 0000000000000002 R09: 0000000000003434
R10: 0000000000000000 R11: 0000000000000246 R12: 00007ffef0103d6c
R13: 00007ffef0103d80 R14: 00007ffef0103dc0 R15: 0000000000000003
</TASK>
Allocated by task 5189:
kasan_save_stack+0x22/0x40 mm/kasan/common.c:45
kasan_set_track+0x25/0x30 mm/kasan/common.c:52
____kasan_kmalloc mm/kasan/common.c:374 [inline]
____kasan_kmalloc mm/kasan/common.c:333 [inline]
__kasan_kmalloc+0xa5/0xb0 mm/kasan/common.c:383
kasan_kmalloc include/linux/kasan.h:211 [inline]
__do_kmalloc_node mm/slab_common.c:968 [inline]
__kmalloc_node_track_caller+0x5b/0xc0 mm/slab_common.c:988
kmalloc_reserve+0xf1/0x230 net/core/skbuff.c:539
pskb_expand_head+0x237/0x1160 net/core/skbuff.c:1995
__skb_unclone_keeptruesize+0x93/0x220 net/core/skbuff.c:2094
skb_unclone_keeptruesize include/linux/skbuff.h:1910 [inline]
skb_prepare_for_shift net/core/skbuff.c:3804 [inline]
skb_shift+0xef8/0x1e20 net/core/skbuff.c:3877
tcp_skb_shift net/ipv4/tcp_input.c:1538 [inline]
tcp_shift_skb_data net/ipv4/tcp_input.c:1646 [inline]
tcp_sacktag_walk+0x93b/0x18a0 net/ipv4/tcp_input.c:1713
tcp_sacktag_write_queue+0x1599/0x31d0 net/ipv4/tcp_input.c:1974
tcp_ack+0x2e9f/0x5a10 net/ipv4/tcp_input.c:3847
tcp_rcv_established+0x667/0x2230 net/ipv4/tcp_input.c:6006
tcp_v4_do_rcv+0x670/0x9b0 net/ipv4/tcp_ipv4.c:1721
sk_backlog_rcv include/net/sock.h:1113 [inline]
__release_sock+0x133/0x3b0 net/core/sock.c:2921
release_sock+0x58/0x1b0 net/core/sock.c:3488
tcp_sendmsg+0x3a/0x50 net/ipv4/tcp.c:1485
inet_sendmsg+0x9d/0xe0 net/ipv4/af_inet.c:825
sock_sendmsg_nosec net/socket.c:722 [inline]
sock_sendmsg+0xde/0x190 net/socket.c:745
sock_write_iter+0x295/0x3d0 net/socket.c:1136
call_write_iter include/linux/fs.h:2189 [inline]
new_sync_write fs/read_write.c:491 [inline]
vfs_write+0x9ed/0xdd0 fs/read_write.c:584
ksys_write+0x1ec/0x250 fs/read_write.c:637
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x39/0xb0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x63/0xcd
The buggy address belongs to the object at ffff88806cdee800
which belongs to the cache kmalloc-1k of size 1024
The buggy address is located 0 bytes inside of
1024-byte region [ffff88806cdee800, ffff88806cdeec00)
The buggy address belongs to the physical page:
page:ffffea0001b37a00 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x6cde8
head:ffffea0001b37a00 order:3 compound_mapcount:0 subpages_mapcount:0 compound_pincount:0
flags: 0xfff00000010200(slab|head|node=0|zone=1|lastcpupid=0x7ff)
raw: 00fff00000010200 ffff888012441dc0 dead000000000122 0000000000000000
raw: 0000000000000000 0000000000100010 00000001ffffffff 0000000000000000
page dumped because: kasan: bad access detected
page_owner tracks the page as allocated
page last allocated via order 3, migratetype Unmovable, gfp_mask 0x1f2a20(GFP_ATOMIC|__GFP_NOWARN|__GFP_NORETRY|__GFP_COMP|__GFP_NOMEMALLOC|__GFP_MEMALLOC|__GFP_HARDWALL), pid 75, tgid 75 (kworker/u4:4), ts 96369578780, free_ts 26734162530
prep_new_page mm/page_alloc.c:2531 [inline]
get_page_from_freelist+0x119c/0x2ce0 mm/page_alloc.c:4283
__alloc_pages+0x1cb/0x5b0 mm/page_alloc.c:5549
alloc_pages+0x1aa/0x270 mm/mempolicy.c:2287
alloc_slab_page mm/slub.c:1851 [inline]
allocate_slab+0x25f/0x350 mm/slub.c:1998
new_slab mm/slub.c:2051 [inline]
___slab_alloc+0xa91/0x1400 mm/slub.c:3193
__slab_alloc.constprop.0+0x56/0xa0 mm/slub.c:3292
__slab_alloc_node mm/slub.c:3345 [inline]
slab_alloc_node mm/slub.c:3442 [inline]
__kmem_cache_alloc_node+0x1a4/0x430 mm/slub.c:3491
__do_kmalloc_node mm/slab_common.c:967 [inline]
__kmalloc_node_track_caller+0x4b/0xc0 mm/slab_common.c:988
kmalloc_reserve+0xf1/0x230 net/core/skbuff.c:539
__alloc_skb+0x129/0x330 net/core/skbuff.c:608
__netdev_alloc_skb+0x74/0x410 net/core/skbuff.c:672
__netdev_alloc_skb_ip_align include/linux/skbuff.h:3203 [inline]
netdev_alloc_skb_ip_align include/linux/skbuff.h:3213 [inline]
batadv_iv_ogm_aggregate_new+0x106/0x4e0 net/batman-adv/bat_iv_ogm.c:558
batadv_iv_ogm_queue_add net/batman-adv/bat_iv_ogm.c:670 [inline]
batadv_iv_ogm_schedule_buff+0xe6b/0x1450 net/batman-adv/bat_iv_ogm.c:849
batadv_iv_ogm_schedule net/batman-adv/bat_iv_ogm.c:868 [inline]
batadv_iv_ogm_schedule net/batman-adv/bat_iv_ogm.c:861 [inline]
batadv_iv_send_outstanding_bat_ogm_packet+0x744/0x910 net/batman-adv/bat_iv_ogm.c:1712
process_one_work+0x9bf/0x1710 kernel/workqueue.c:2289
worker_thread+0x669/0x1090 kernel/workqueue.c:2436
page last free stack trace:
reset_page_owner include/linux/page_owner.h:24 [inline]
free_pages_prepare mm/page_alloc.c:1446 [inline]
free_pcp_prepare+0x66a/0xc20 mm/page_alloc.c:1496
free_unref_page_prepare mm/page_alloc.c:3369 [inline]
free_unref_page+0x1d/0x490 mm/page_alloc.c:3464
free_contig_range+0xb5/0x180 mm/page_alloc.c:9488
destroy_args+0xa8/0x64c mm/debug_vm_pgtable.c:998
debug_vm_pgtable+0x28de/0x296f mm/debug_vm_pgtable.c:1318
do_one_initcall+0x141/0x790 init/main.c:1306
do_initcall_level init/main.c:1379 [inline]
do_initcalls init/main.c:1395 [inline]
do_basic_setup init/main.c:1414 [inline]
kernel_init_freeable+0x6f9/0x782 init/main.c:1634
kernel_init+0x1e/0x1d0 init/main.c:1522
ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308
Memory state around the buggy address:
ffff88806cdee700: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
ffff88806cdee780: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
>ffff88806cdee800: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
^
ffff88806cdee880: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
Fixes: bf9f1baa279f ("net: add dedicated kmem_cache for typical/small skb->head")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Tested-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2023-02-27 14:17:06 +00:00
|
|
|
|
2026-01-16 04:13:59 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-02-06 17:31:03 +00:00
|
|
|
|
2023-08-31 18:37:50 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-02-06 17:31:02 +00:00
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
|
|
|
|
|
2026-01-16 04:13:59 +00:00
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 04:13:59 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-16 20:27:16 +00:00
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
2021-09-22 14:17:19 +08:00
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-16 20:27:16 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-16 20:27:15 +00:00
|
|
|
|
2025-11-16 20:27:16 +00:00
|
|
|
|
|
|
|
|
|
2025-11-16 20:27:17 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-16 20:27:16 +00:00
|
|
|
|
2025-11-16 20:27:17 +00:00
|
|
|
|
2025-11-16 20:27:16 +00:00
|
|
|
|
|
|
|
|
|
2021-02-13 14:12:38 +00:00
|
|
|
|
2025-11-16 20:27:16 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 04:13:59 +00:00
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 04:13:59 +00:00
|
|
|
|
2021-02-13 14:11:39 +00:00
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
2022-10-25 15:39:35 -07:00
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-13 13:10:17 +00:00
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-16 16:44:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-02-13 14:11:39 +00:00
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
|
|
|
|
|
2021-02-13 14:11:39 +00:00
|
|
|
|
2021-02-13 14:11:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-08-24 22:56:54 +09:00
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
|
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
2015-05-06 21:11:57 -07:00
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
|
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
2021-01-15 15:04:40 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-18 19:29:39 +01:00
|
|
|
|
2021-01-15 15:04:40 +00:00
|
|
|
|
2015-11-06 16:28:21 -08:00
|
|
|
|
2015-05-13 13:34:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
2023-02-06 17:31:00 +00:00
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-13 22:57:49 +08:00
|
|
|
|
2019-06-07 21:20:35 +02:00
|
|
|
|
|
|
|
|
|
2024-10-28 19:53:39 +08:00
|
|
|
|
2019-06-07 21:20:35 +02:00
|
|
|
|
|
|
|
|
|
2024-06-20 15:21:54 +02:00
|
|
|
|
|
|
|
|
|
2019-06-07 21:20:35 +02:00
|
|
|
|
|
|
|
|
|
2024-10-28 19:53:39 +08:00
|
|
|
|
2024-06-20 15:21:54 +02:00
|
|
|
|
|
|
|
|
|
2019-06-07 21:20:35 +02:00
|
|
|
|
|
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-05-06 21:12:03 -07:00
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
2006-08-07 16:09:04 -07:00
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-05-13 13:34:13 -07:00
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-05-13 13:34:13 -07:00
|
|
|
|
2006-07-31 22:35:23 -07:00
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
2024-03-26 21:02:12 -07:00
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
2015-08-24 22:56:54 +09:00
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-26 21:02:12 -07:00
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
2024-03-26 21:02:12 -07:00
|
|
|
|
2021-01-13 08:18:19 -08:00
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
2022-09-28 10:43:09 +02:00
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
2022-06-08 09:04:37 -07:00
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
2021-01-13 08:18:19 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-18 19:29:40 +01:00
|
|
|
|
2021-01-13 08:18:19 -08:00
|
|
|
|
2015-11-06 16:28:21 -08:00
|
|
|
|
2021-02-13 14:12:49 +00:00
|
|
|
|
|
|
|
|
|
2015-05-13 13:34:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
2025-02-18 19:29:40 +01:00
|
|
|
|
|
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-06-20 15:21:54 +02:00
|
|
|
|
|
|
|
|
|
2022-09-28 10:43:09 +02:00
|
|
|
|
2025-02-18 19:29:40 +01:00
|
|
|
|
|
|
|
|
|
2024-06-20 15:21:54 +02:00
|
|
|
|
2022-09-28 10:43:09 +02:00
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-02-13 14:12:49 +00:00
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
2015-05-06 21:12:03 -07:00
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
|
|
|
|
|
2022-09-28 10:43:09 +02:00
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-05-13 13:34:13 -07:00
|
|
|
|
2015-05-06 21:11:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-05-13 13:34:13 -07:00
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
|
|
|
|
|
2024-03-26 21:02:12 -07:00
|
|
|
|
2014-12-09 19:40:49 -08:00
|
|
|
|
2008-10-07 14:22:33 -07:00
|
|
|
|
2013-11-01 14:07:47 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-09-13 13:48:41 +00:00
|
|
|
|
|
|
|
|
|
2013-11-01 14:07:47 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-07-13 19:26:39 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2013-06-24 06:26:00 -07:00
|
|
|
|
2006-07-13 19:26:39 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2006-07-13 19:26:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-02-12 10:50:57 +01:00
|
|
|
|
|
|
|
|
|
2024-02-12 10:50:56 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-02-17 12:12:14 +01:00
|
|
|
|
2024-02-12 10:50:56 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-02-12 10:50:57 +01:00
|
|
|
|
2024-02-12 10:50:56 +01:00
|
|
|
|
|
|
|
|
|
2024-12-03 18:37:25 +01:00
|
|
|
|
2024-02-12 10:50:56 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: skbuff: don't include <net/page_pool/types.h> to <linux/skbuff.h>
Currently, touching <net/page_pool/types.h> triggers a rebuild of more
than half of the kernel. That's because it's included in
<linux/skbuff.h>. And each new include to page_pool/types.h adds more
[useless] data for the toolchain to process per each source file from
that pile.
In commit 6a5bcd84e886 ("page_pool: Allow drivers to hint on SKB
recycling"), Matteo included it to be able to call a couple of functions
defined there. Then, in commit 57f05bc2ab24 ("page_pool: keep pp info as
long as page pool owns the page") one of the calls was removed, so only
one was left. It's the call to page_pool_return_skb_page() in
napi_frag_unref(). The function is external and doesn't have any
dependencies. Having very niche page_pool_types.h included only for that
looks like an overkill.
As %PP_SIGNATURE is not local to page_pool.c (was only in the
early submissions), nothing holds this function there. Teleport
page_pool_return_skb_page() to skbuff.c, just next to the main consumer,
skb_pp_recycle(), and rename it to napi_pp_put_page(), as it doesn't
work with skbs at all and the former name tells nothing. The #if guards
here are only to not compile and have it in the vmlinux when not needed
-- both call sites are already guarded.
Now, touching page_pool_types.h only triggers rebuilding of the drivers
using it and a couple of core networking files.
Suggested-by: Jakub Kicinski <kuba@kernel.org> # make skbuff.h less heavy
Suggested-by: Alexander Duyck <alexanderduyck@fb.com> # move to skbuff.c
Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230804180529.2483231-3-aleksander.lobakin@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2023-08-04 20:05:25 +02:00
|
|
|
|
2024-06-28 00:32:42 +00:00
|
|
|
|
net: skbuff: don't include <net/page_pool/types.h> to <linux/skbuff.h>
Currently, touching <net/page_pool/types.h> triggers a rebuild of more
than half of the kernel. That's because it's included in
<linux/skbuff.h>. And each new include to page_pool/types.h adds more
[useless] data for the toolchain to process per each source file from
that pile.
In commit 6a5bcd84e886 ("page_pool: Allow drivers to hint on SKB
recycling"), Matteo included it to be able to call a couple of functions
defined there. Then, in commit 57f05bc2ab24 ("page_pool: keep pp info as
long as page pool owns the page") one of the calls was removed, so only
one was left. It's the call to page_pool_return_skb_page() in
napi_frag_unref(). The function is external and doesn't have any
dependencies. Having very niche page_pool_types.h included only for that
looks like an overkill.
As %PP_SIGNATURE is not local to page_pool.c (was only in the
early submissions), nothing holds this function there. Teleport
page_pool_return_skb_page() to skbuff.c, just next to the main consumer,
skb_pp_recycle(), and rename it to napi_pp_put_page(), as it doesn't
work with skbs at all and the former name tells nothing. The #if guards
here are only to not compile and have it in the vmlinux when not needed
-- both call sites are already guarded.
Now, touching page_pool_types.h only triggers rebuilding of the drivers
using it and a couple of core networking files.
Suggested-by: Jakub Kicinski <kuba@kernel.org> # make skbuff.h less heavy
Suggested-by: Alexander Duyck <alexanderduyck@fb.com> # move to skbuff.c
Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230804180529.2483231-3-aleksander.lobakin@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2023-08-04 20:05:25 +02:00
|
|
|
|
2024-09-10 17:14:49 +00:00
|
|
|
|
net: skbuff: don't include <net/page_pool/types.h> to <linux/skbuff.h>
Currently, touching <net/page_pool/types.h> triggers a rebuild of more
than half of the kernel. That's because it's included in
<linux/skbuff.h>. And each new include to page_pool/types.h adds more
[useless] data for the toolchain to process per each source file from
that pile.
In commit 6a5bcd84e886 ("page_pool: Allow drivers to hint on SKB
recycling"), Matteo included it to be able to call a couple of functions
defined there. Then, in commit 57f05bc2ab24 ("page_pool: keep pp info as
long as page pool owns the page") one of the calls was removed, so only
one was left. It's the call to page_pool_return_skb_page() in
napi_frag_unref(). The function is external and doesn't have any
dependencies. Having very niche page_pool_types.h included only for that
looks like an overkill.
As %PP_SIGNATURE is not local to page_pool.c (was only in the
early submissions), nothing holds this function there. Teleport
page_pool_return_skb_page() to skbuff.c, just next to the main consumer,
skb_pp_recycle(), and rename it to napi_pp_put_page(), as it doesn't
work with skbs at all and the former name tells nothing. The #if guards
here are only to not compile and have it in the vmlinux when not needed
-- both call sites are already guarded.
Now, touching page_pool_types.h only triggers rebuilding of the drivers
using it and a couple of core networking files.
Suggested-by: Jakub Kicinski <kuba@kernel.org> # make skbuff.h less heavy
Suggested-by: Alexander Duyck <alexanderduyck@fb.com> # move to skbuff.c
Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230804180529.2483231-3-aleksander.lobakin@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2023-08-04 20:05:25 +02:00
|
|
|
|
2025-04-09 12:41:36 +02:00
|
|
|
|
net: skbuff: don't include <net/page_pool/types.h> to <linux/skbuff.h>
Currently, touching <net/page_pool/types.h> triggers a rebuild of more
than half of the kernel. That's because it's included in
<linux/skbuff.h>. And each new include to page_pool/types.h adds more
[useless] data for the toolchain to process per each source file from
that pile.
In commit 6a5bcd84e886 ("page_pool: Allow drivers to hint on SKB
recycling"), Matteo included it to be able to call a couple of functions
defined there. Then, in commit 57f05bc2ab24 ("page_pool: keep pp info as
long as page pool owns the page") one of the calls was removed, so only
one was left. It's the call to page_pool_return_skb_page() in
napi_frag_unref(). The function is external and doesn't have any
dependencies. Having very niche page_pool_types.h included only for that
looks like an overkill.
As %PP_SIGNATURE is not local to page_pool.c (was only in the
early submissions), nothing holds this function there. Teleport
page_pool_return_skb_page() to skbuff.c, just next to the main consumer,
skb_pp_recycle(), and rename it to napi_pp_put_page(), as it doesn't
work with skbs at all and the former name tells nothing. The #if guards
here are only to not compile and have it in the vmlinux when not needed
-- both call sites are already guarded.
Now, touching page_pool_types.h only triggers rebuilding of the drivers
using it and a couple of core networking files.
Suggested-by: Jakub Kicinski <kuba@kernel.org> # make skbuff.h less heavy
Suggested-by: Alexander Duyck <alexanderduyck@fb.com> # move to skbuff.c
Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230804180529.2483231-3-aleksander.lobakin@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2023-08-04 20:05:25 +02:00
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:49 +00:00
|
|
|
|
net: skbuff: don't include <net/page_pool/types.h> to <linux/skbuff.h>
Currently, touching <net/page_pool/types.h> triggers a rebuild of more
than half of the kernel. That's because it's included in
<linux/skbuff.h>. And each new include to page_pool/types.h adds more
[useless] data for the toolchain to process per each source file from
that pile.
In commit 6a5bcd84e886 ("page_pool: Allow drivers to hint on SKB
recycling"), Matteo included it to be able to call a couple of functions
defined there. Then, in commit 57f05bc2ab24 ("page_pool: keep pp info as
long as page pool owns the page") one of the calls was removed, so only
one was left. It's the call to page_pool_return_skb_page() in
napi_frag_unref(). The function is external and doesn't have any
dependencies. Having very niche page_pool_types.h included only for that
looks like an overkill.
As %PP_SIGNATURE is not local to page_pool.c (was only in the
early submissions), nothing holds this function there. Teleport
page_pool_return_skb_page() to skbuff.c, just next to the main consumer,
skb_pp_recycle(), and rename it to napi_pp_put_page(), as it doesn't
work with skbs at all and the former name tells nothing. The #if guards
here are only to not compile and have it in the vmlinux when not needed
-- both call sites are already guarded.
Now, touching page_pool_types.h only triggers rebuilding of the drivers
using it and a couple of core networking files.
Suggested-by: Jakub Kicinski <kuba@kernel.org> # make skbuff.h less heavy
Suggested-by: Alexander Duyck <alexanderduyck@fb.com> # move to skbuff.c
Signed-off-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Link: https://lore.kernel.org/r/20230804180529.2483231-3-aleksander.lobakin@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2023-08-04 20:05:25 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
2022-10-21 10:58:22 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-06-28 00:32:42 +00:00
|
|
|
|
2022-10-21 10:58:22 +08:00
|
|
|
|
|
|
|
|
|
2024-05-02 10:54:22 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:49 +00:00
|
|
|
|
2024-05-02 10:54:22 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:49 +00:00
|
|
|
|
2025-04-09 12:41:36 +02:00
|
|
|
|
2024-09-10 17:14:49 +00:00
|
|
|
|
2024-05-02 10:54:22 -07:00
|
|
|
|
2024-09-10 17:14:49 +00:00
|
|
|
|
2024-05-02 10:54:22 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-10 11:47:32 +08:00
|
|
|
|
2023-02-06 17:31:03 +00:00
|
|
|
|
2026-04-03 09:45:12 +08:00
|
|
|
|
2023-02-06 17:31:03 +00:00
|
|
|
|
|
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
net: allow skb->head to be a page fragment
skb->head is currently allocated from kmalloc(). This is convenient but
has the drawback the data cannot be converted to a page fragment if
needed.
We have three spots were it hurts :
1) GRO aggregation
When a linear skb must be appended to another skb, GRO uses the
frag_list fallback, very inefficient since we keep all struct sk_buff
around. So drivers enabling GRO but delivering linear skbs to network
stack aren't enabling full GRO power.
2) splice(socket -> pipe).
We must copy the linear part to a page fragment.
This kind of defeats splice() purpose (zero copy claim)
3) TCP coalescing.
Recently introduced, this permits to group several contiguous segments
into a single skb. This shortens queue lengths and save kernel memory,
and greatly reduce probabilities of TCP collapses. This coalescing
doesnt work on linear skbs (or we would need to copy data, this would be
too slow)
Given all these issues, the following patch introduces the possibility
of having skb->head be a fragment in itself. We use a new skb flag,
skb->head_frag to carry this information.
build_skb() is changed to accept a frag_size argument. Drivers willing
to provide a page fragment instead of kmalloc() data will set a non zero
value, set to the fragment size.
Then, on situations we need to convert the skb head to a frag in itself,
we can check if skb->head_frag is set and avoid the copies or various
fallbacks we have.
This means drivers currently using frags could be updated to avoid the
current skb->head allocation and reduce their memory footprint (aka skb
truesize). (thats 512 or 1024 bytes saved per skb). This also makes
bpf/netfilter faster since the 'first frag' will be part of skb linear
part, no need to copy data.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Maciej Żenczykowski <maze@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Cc: Ben Hutchings <bhutchings@solarflare.com>
Cc: Matt Carlson <mcarlson@broadcom.com>
Cc: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-27 00:33:38 +00:00
|
|
|
|
2015-05-06 21:12:03 -07:00
|
|
|
|
|
|
|
|
|
2021-06-07 21:02:38 +02:00
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
2021-06-07 21:02:38 +02:00
|
|
|
|
2015-05-06 21:12:03 -07:00
|
|
|
|
2021-06-07 21:02:38 +02:00
|
|
|
|
2026-04-10 11:47:32 +08:00
|
|
|
|
2021-06-07 21:02:38 +02:00
|
|
|
|
net: allow skb->head to be a page fragment
skb->head is currently allocated from kmalloc(). This is convenient but
has the drawback the data cannot be converted to a page fragment if
needed.
We have three spots were it hurts :
1) GRO aggregation
When a linear skb must be appended to another skb, GRO uses the
frag_list fallback, very inefficient since we keep all struct sk_buff
around. So drivers enabling GRO but delivering linear skbs to network
stack aren't enabling full GRO power.
2) splice(socket -> pipe).
We must copy the linear part to a page fragment.
This kind of defeats splice() purpose (zero copy claim)
3) TCP coalescing.
Recently introduced, this permits to group several contiguous segments
into a single skb. This shortens queue lengths and save kernel memory,
and greatly reduce probabilities of TCP collapses. This coalescing
doesnt work on linear skbs (or we would need to copy data, this would be
too slow)
Given all these issues, the following patch introduces the possibility
of having skb->head be a fragment in itself. We use a new skb flag,
skb->head_frag to carry this information.
build_skb() is changed to accept a frag_size argument. Drivers willing
to provide a page fragment instead of kmalloc() data will set a non zero
value, set to the fragment size.
Then, on situations we need to convert the skb head to a frag in itself,
we can check if skb->head_frag is set and avoid the copies or various
fallbacks we have.
This means drivers currently using frags could be updated to avoid the
current skb->head allocation and reduce their memory footprint (aka skb
truesize). (thats 512 or 1024 bytes saved per skb). This also makes
bpf/netfilter faster since the 'first frag' will be part of skb linear
part, no need to copy data.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Maciej Żenczykowski <maze@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Cc: Ben Hutchings <bhutchings@solarflare.com>
Cc: Matt Carlson <mcarlson@broadcom.com>
Cc: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-27 00:33:38 +00:00
|
|
|
|
|
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-09-23 18:39:30 -07:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2024-03-07 12:34:46 +00:00
|
|
|
|
2021-07-16 10:02:18 +03:00
|
|
|
|
2011-07-06 12:22:12 +00:00
|
|
|
|
2022-07-12 21:52:31 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-01-06 14:18:36 -08:00
|
|
|
|
2014-09-23 18:39:30 -07:00
|
|
|
|
2024-04-08 08:29:58 -07:00
|
|
|
|
2011-07-06 12:22:12 +00:00
|
|
|
|
2022-07-12 21:52:31 +01:00
|
|
|
|
2014-09-23 18:39:30 -07:00
|
|
|
|
2022-10-29 15:45:17 +00:00
|
|
|
|
2014-09-23 18:39:30 -07:00
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
2021-07-16 10:02:18 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-02-09 07:02:42 -08:00
|
|
|
|
2021-07-16 10:02:18 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-11-26 23:11:19 +08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-09-29 13:29:15 -07:00
|
|
|
|
2005-08-17 14:57:30 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-06 16:00:23 +00:00
|
|
|
|
2014-12-03 17:04:39 -08:00
|
|
|
|
2005-08-17 14:57:30 -07:00
|
|
|
|
|
|
|
|
|
2014-09-29 13:29:15 -07:00
|
|
|
|
2014-12-03 17:04:39 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-06-30 13:07:59 +03:00
|
|
|
|
2014-12-03 17:04:39 -08:00
|
|
|
|
2005-08-17 14:57:30 -07:00
|
|
|
|
|
|
|
|
|
2014-12-03 17:04:39 -08:00
|
|
|
|
2014-09-29 13:29:15 -07:00
|
|
|
|
2005-08-17 14:57:30 -07:00
|
|
|
|
2007-04-20 17:09:22 -07:00
|
|
|
|
2017-06-30 13:07:59 +03:00
|
|
|
|
2014-12-03 17:04:39 -08:00
|
|
|
|
|
|
|
|
|
2024-03-06 16:00:23 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2017-06-12 11:23:42 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-06-02 05:19:30 +00:00
|
|
|
|
2005-04-19 22:39:42 -07:00
|
|
|
|
2022-06-08 09:04:36 -07:00
|
|
|
|
2025-10-14 17:19:03 +00:00
|
|
|
|
2025-10-31 18:33:28 +08:00
|
|
|
|
2025-10-14 17:19:03 +00:00
|
|
|
|
2025-10-31 18:33:28 +08:00
|
|
|
|
2025-10-14 17:19:03 +00:00
|
|
|
|
|
|
|
|
|
2025-10-31 18:33:28 +08:00
|
|
|
|
|
|
|
|
|
2025-10-14 17:19:03 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-06 20:29:33 +00:00
|
|
|
|
2025-11-11 15:12:35 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2025-11-06 20:29:33 +00:00
|
|
|
|
|
|
|
|
|
2008-10-01 02:33:12 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
2008-10-01 02:33:12 -07:00
|
|
|
|
|
|
|
|
|
2017-07-23 19:54:47 +02:00
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
2007-11-26 23:11:19 +08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-11-26 23:11:19 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2023-01-13 14:51:59 +01:00
|
|
|
|
2026-05-11 07:23:08 +00:00
|
|
|
|
2024-06-17 11:09:09 -07:00
|
|
|
|
2023-01-13 14:51:59 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-04-19 14:52:53 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-13 14:51:59 +01:00
|
|
|
|
|
|
|
|
|
2023-02-16 15:47:18 +00:00
|
|
|
|
2023-01-13 14:51:59 +01:00
|
|
|
|
2024-06-17 11:09:09 -07:00
|
|
|
|
2023-01-13 14:51:59 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-03-20 21:28:35 -08:00
|
|
|
|
2024-06-17 11:09:09 -07:00
|
|
|
|
|
|
|
|
|
2006-03-20 21:28:35 -08:00
|
|
|
|
2022-01-09 14:36:26 +08:00
|
|
|
|
2006-03-20 21:28:35 -08:00
|
|
|
|
2024-06-17 11:09:09 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-03-20 21:28:35 -08:00
|
|
|
|
2022-08-21 13:18:58 +08:00
|
|
|
|
2026-05-11 07:23:08 +00:00
|
|
|
|
|
|
|
|
|
2006-03-20 21:28:35 -08:00
|
|
|
|
2024-06-17 11:09:09 -07:00
|
|
|
|
2023-01-13 14:51:59 +01:00
|
|
|
|
2006-03-20 21:28:35 -08:00
|
|
|
|
2024-06-17 11:09:09 -07:00
|
|
|
|
2006-03-20 21:28:35 -08:00
|
|
|
|
2023-01-13 14:52:04 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
2023-01-13 14:52:04 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-06 16:00:23 +00:00
|
|
|
|
2023-01-13 14:52:04 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-13 14:51:59 +01:00
|
|
|
|
|
|
|
|
|
2013-06-24 06:26:00 -07:00
|
|
|
|
2023-01-13 14:52:04 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-06-24 06:26:00 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-06-17 11:09:09 -07:00
|
|
|
|
2023-02-03 13:59:29 +01:00
|
|
|
|
2023-01-13 14:52:04 +01:00
|
|
|
|
2023-02-03 13:59:29 +01:00
|
|
|
|
2023-01-13 14:52:04 +01:00
|
|
|
|
2013-06-24 06:26:00 -07:00
|
|
|
|
|
|
|
|
|
2023-01-13 14:52:04 +01:00
|
|
|
|
|
|
|
|
|
2024-03-06 16:00:23 +00:00
|
|
|
|
2013-06-24 06:26:00 -07:00
|
|
|
|
2022-03-04 14:00:41 +08:00
|
|
|
|
2013-06-24 06:26:00 -07:00
|
|
|
|
2019-07-07 05:51:55 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-10-05 17:48:38 +03:00
|
|
|
|
2019-07-07 05:51:55 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-12 17:26:21 +00:00
|
|
|
|
|
|
|
|
|
2019-07-07 05:51:55 -04:00
|
|
|
|
2024-04-07 08:06:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-12 17:26:21 +00:00
|
|
|
|
|
|
|
|
|
2019-07-07 05:51:55 -04:00
|
|
|
|
|
|
|
|
|
2024-04-07 08:06:06 +00:00
|
|
|
|
2019-07-07 05:51:55 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-04-07 08:06:06 +00:00
|
|
|
|
|
|
|
|
|
2019-07-07 05:51:55 -04:00
|
|
|
|
2024-04-07 08:06:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-07 05:51:55 -04:00
|
|
|
|
|
|
|
|
|
2021-12-16 11:28:25 +02:00
|
|
|
|
2019-07-07 05:51:55 -04:00
|
|
|
|
|
|
|
|
|
2019-07-16 11:43:05 -04:00
|
|
|
|
2019-07-07 05:51:55 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:51 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
2019-07-07 05:51:55 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-11-01 09:16:28 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-12 21:52:31 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-11-01 09:16:28 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-08-22 08:23:29 +10:00
|
|
|
|
2009-03-11 09:49:55 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-06-12 11:23:41 +02:00
|
|
|
|
2009-03-11 09:49:55 +00:00
|
|
|
|
2017-06-12 11:23:41 +02:00
|
|
|
|
2023-02-16 15:47:18 +00:00
|
|
|
|
2009-03-11 09:49:55 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-08-22 08:23:29 +10:00
|
|
|
|
2009-03-11 09:49:55 +00:00
|
|
|
|
2017-06-12 11:23:42 +02:00
|
|
|
|
2020-11-16 11:17:58 +01:00
|
|
|
|
2017-06-12 11:23:42 +02:00
|
|
|
|
|
|
|
|
|
2017-09-06 14:44:36 +02:00
|
|
|
|
|
|
|
|
|
2017-06-12 11:23:42 +02:00
|
|
|
|
2017-09-06 14:44:36 +02:00
|
|
|
|
2017-06-12 11:23:42 +02:00
|
|
|
|
2023-02-16 15:47:18 +00:00
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
2017-06-12 11:23:42 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
skbuff: introduce {,__}napi_build_skb() which reuses NAPI cache heads
Instead of just bulk-flushing skbuff_heads queued up through
napi_consume_skb() or __kfree_skb_defer(), try to reuse them
on allocation path.
If the cache is empty on allocation, bulk-allocate the first
16 elements, which is more efficient than per-skb allocation.
If the cache is full on freeing, bulk-wipe the second half of
the cache (32 elements).
This also includes custom KASAN poisoning/unpoisoning to be
double sure there are no use-after-free cases.
To not change current behaviour, introduce a new function,
napi_build_skb(), to optionally use a new approach later
in drivers.
Note on selected bulk size, 16:
- this equals to XDP_BULK_QUEUE_SIZE, DEV_MAP_BULK_SIZE
and especially VETH_XDP_BATCH, which is also used to
bulk-allocate skbuff_heads and was tested on powerful
setups;
- this also showed the best performance in the actual
test series (from the array of {8, 16, 32}).
Suggested-by: Edward Cree <ecree.xilinx@gmail.com> # Divide on two halves
Suggested-by: Eric Dumazet <edumazet@google.com> # KASAN poisoning
Cc: Dmitry Vyukov <dvyukov@google.com> # Help with KASAN
Cc: Paolo Abeni <pabeni@redhat.com> # Reduced batch size
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-02-13 14:12:25 +00:00
|
|
|
|
2016-02-08 13:14:59 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-19 23:29:04 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-06-20 15:21:54 +02:00
|
|
|
|
2016-02-08 13:14:59 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-18 15:06:45 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-19 23:29:04 +01:00
|
|
|
|
2025-10-16 18:29:11 +00:00
|
|
|
|
skbuff: introduce {,__}napi_build_skb() which reuses NAPI cache heads
Instead of just bulk-flushing skbuff_heads queued up through
napi_consume_skb() or __kfree_skb_defer(), try to reuse them
on allocation path.
If the cache is empty on allocation, bulk-allocate the first
16 elements, which is more efficient than per-skb allocation.
If the cache is full on freeing, bulk-wipe the second half of
the cache (32 elements).
This also includes custom KASAN poisoning/unpoisoning to be
double sure there are no use-after-free cases.
To not change current behaviour, introduce a new function,
napi_build_skb(), to optionally use a new approach later
in drivers.
Note on selected bulk size, 16:
- this equals to XDP_BULK_QUEUE_SIZE, DEV_MAP_BULK_SIZE
and especially VETH_XDP_BATCH, which is also used to
bulk-allocate skbuff_heads and was tested on powerful
setups;
- this also showed the best performance in the actual
test series (from the array of {8, 16, 32}).
Suggested-by: Edward Cree <ecree.xilinx@gmail.com> # Divide on two halves
Suggested-by: Eric Dumazet <edumazet@google.com> # KASAN poisoning
Cc: Dmitry Vyukov <dvyukov@google.com> # Help with KASAN
Cc: Paolo Abeni <pabeni@redhat.com> # Reduced batch size
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-02-13 14:12:25 +00:00
|
|
|
|
2025-11-18 15:06:45 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-02-08 13:14:59 +01:00
|
|
|
|
2024-06-20 15:21:54 +02:00
|
|
|
|
2016-02-08 13:14:59 +01:00
|
|
|
|
skbuff: introduce {,__}napi_build_skb() which reuses NAPI cache heads
Instead of just bulk-flushing skbuff_heads queued up through
napi_consume_skb() or __kfree_skb_defer(), try to reuse them
on allocation path.
If the cache is empty on allocation, bulk-allocate the first
16 elements, which is more efficient than per-skb allocation.
If the cache is full on freeing, bulk-wipe the second half of
the cache (32 elements).
This also includes custom KASAN poisoning/unpoisoning to be
double sure there are no use-after-free cases.
To not change current behaviour, introduce a new function,
napi_build_skb(), to optionally use a new approach later
in drivers.
Note on selected bulk size, 16:
- this equals to XDP_BULK_QUEUE_SIZE, DEV_MAP_BULK_SIZE
and especially VETH_XDP_BATCH, which is also used to
bulk-allocate skbuff_heads and was tested on powerful
setups;
- this also showed the best performance in the actual
test series (from the array of {8, 16, 32}).
Suggested-by: Edward Cree <ecree.xilinx@gmail.com> # Divide on two halves
Suggested-by: Eric Dumazet <edumazet@google.com> # KASAN poisoning
Cc: Dmitry Vyukov <dvyukov@google.com> # Help with KASAN
Cc: Paolo Abeni <pabeni@redhat.com> # Reduced batch size
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-02-13 14:12:25 +00:00
|
|
|
|
2023-04-19 19:00:05 -07:00
|
|
|
|
2016-02-08 13:15:04 +01:00
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
2021-02-13 14:13:09 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-07-28 18:24:02 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-07-28 18:24:03 +02:00
|
|
|
|
2021-07-28 18:24:02 +02:00
|
|
|
|
|
|
|
|
|
skbuff: introduce {,__}napi_build_skb() which reuses NAPI cache heads
Instead of just bulk-flushing skbuff_heads queued up through
napi_consume_skb() or __kfree_skb_defer(), try to reuse them
on allocation path.
If the cache is empty on allocation, bulk-allocate the first
16 elements, which is more efficient than per-skb allocation.
If the cache is full on freeing, bulk-wipe the second half of
the cache (32 elements).
This also includes custom KASAN poisoning/unpoisoning to be
double sure there are no use-after-free cases.
To not change current behaviour, introduce a new function,
napi_build_skb(), to optionally use a new approach later
in drivers.
Note on selected bulk size, 16:
- this equals to XDP_BULK_QUEUE_SIZE, DEV_MAP_BULK_SIZE
and especially VETH_XDP_BATCH, which is also used to
bulk-allocate skbuff_heads and was tested on powerful
setups;
- this also showed the best performance in the actual
test series (from the array of {8, 16, 32}).
Suggested-by: Edward Cree <ecree.xilinx@gmail.com> # Divide on two halves
Suggested-by: Eric Dumazet <edumazet@google.com> # KASAN poisoning
Cc: Dmitry Vyukov <dvyukov@google.com> # Help with KASAN
Cc: Paolo Abeni <pabeni@redhat.com> # Reduced batch size
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-02-13 14:12:25 +00:00
|
|
|
|
2016-02-08 13:15:04 +01:00
|
|
|
|
2016-02-08 13:14:59 +01:00
|
|
|
|
2026-01-19 14:41:40 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-02-08 13:14:59 +01:00
|
|
|
|
|
|
|
|
|
2025-11-26 17:43:11 -08:00
|
|
|
|
2016-03-11 09:43:58 +01:00
|
|
|
|
2016-02-08 13:14:59 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-08 09:04:37 -07:00
|
|
|
|
2020-11-24 18:49:29 +08:00
|
|
|
|
2026-04-02 11:41:14 +08:00
|
|
|
|
|
|
|
|
|
2025-11-06 20:29:34 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-06-14 11:48:48 +02:00
|
|
|
|
2016-02-08 13:14:59 +01:00
|
|
|
|
2017-06-14 11:48:48 +02:00
|
|
|
|
2016-02-08 13:14:59 +01:00
|
|
|
|
2023-02-16 15:47:18 +00:00
|
|
|
|
2016-02-08 13:14:59 +01:00
|
|
|
|
|
|
|
|
|
2016-03-20 11:27:47 -07:00
|
|
|
|
2016-02-08 13:14:59 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
skbuff: introduce {,__}napi_build_skb() which reuses NAPI cache heads
Instead of just bulk-flushing skbuff_heads queued up through
napi_consume_skb() or __kfree_skb_defer(), try to reuse them
on allocation path.
If the cache is empty on allocation, bulk-allocate the first
16 elements, which is more efficient than per-skb allocation.
If the cache is full on freeing, bulk-wipe the second half of
the cache (32 elements).
This also includes custom KASAN poisoning/unpoisoning to be
double sure there are no use-after-free cases.
To not change current behaviour, introduce a new function,
napi_build_skb(), to optionally use a new approach later
in drivers.
Note on selected bulk size, 16:
- this equals to XDP_BULK_QUEUE_SIZE, DEV_MAP_BULK_SIZE
and especially VETH_XDP_BATCH, which is also used to
bulk-allocate skbuff_heads and was tested on powerful
setups;
- this also showed the best performance in the actual
test series (from the array of {8, 16, 32}).
Suggested-by: Edward Cree <ecree.xilinx@gmail.com> # Divide on two halves
Suggested-by: Eric Dumazet <edumazet@google.com> # KASAN poisoning
Cc: Dmitry Vyukov <dvyukov@google.com> # Help with KASAN
Cc: Paolo Abeni <pabeni@redhat.com> # Reduced batch size
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
2021-02-13 14:12:25 +00:00
|
|
|
|
2016-02-08 13:14:59 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-11-20 16:31:49 -08:00
|
|
|
|
net: reorganize sk_buff for faster __copy_skb_header()
With proliferation of bit fields in sk_buff, __copy_skb_header() became
quite expensive, showing as the most expensive function in a GSO
workload.
__copy_skb_header() performance is also critical for non GSO TCP
operations, as it is used from skb_clone()
This patch carefully moves all the fields that were not copied in a
separate zone : cloned, nohdr, fclone, peeked, head_frag, xmit_more
Then I moved all other fields and all other copied fields in a section
delimited by headers_start[0]/headers_end[0] section so that we
can use a single memcpy() call, inlined by compiler using long
word load/stores.
I also tried to make all copies in the natural orders of sk_buff,
to help hardware prefetching.
I made sure sk_buff size did not change.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-09-28 22:18:47 -07:00
|
|
|
|
2021-11-20 16:31:49 -08:00
|
|
|
|
|
|
|
|
|
net: reorganize sk_buff for faster __copy_skb_header()
With proliferation of bit fields in sk_buff, __copy_skb_header() became
quite expensive, showing as the most expensive function in a GSO
workload.
__copy_skb_header() performance is also critical for non GSO TCP
operations, as it is used from skb_clone()
This patch carefully moves all the fields that were not copied in a
separate zone : cloned, nohdr, fclone, peeked, head_frag, xmit_more
Then I moved all other fields and all other copied fields in a section
delimited by headers_start[0]/headers_end[0] section so that we
can use a single memcpy() call, inlined by compiler using long
word load/stores.
I also tried to make all copies in the natural orders of sk_buff,
to help hardware prefetching.
I made sure sk_buff size did not change.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-09-28 22:18:47 -07:00
|
|
|
|
2007-10-14 00:37:30 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: reorganize sk_buff for faster __copy_skb_header()
With proliferation of bit fields in sk_buff, __copy_skb_header() became
quite expensive, showing as the most expensive function in a GSO
workload.
__copy_skb_header() performance is also critical for non GSO TCP
operations, as it is used from skb_clone()
This patch carefully moves all the fields that were not copied in a
separate zone : cloned, nohdr, fclone, peeked, head_frag, xmit_more
Then I moved all other fields and all other copied fields in a section
delimited by headers_start[0]/headers_end[0] section so that we
can use a single memcpy() call, inlined by compiler using long
word load/stores.
I also tried to make all copies in the natural orders of sk_buff,
to help hardware prefetching.
I made sure sk_buff size did not change.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-09-28 22:18:47 -07:00
|
|
|
|
2007-10-14 00:37:30 -07:00
|
|
|
|
net: reorganize sk_buff for faster __copy_skb_header()
With proliferation of bit fields in sk_buff, __copy_skb_header() became
quite expensive, showing as the most expensive function in a GSO
workload.
__copy_skb_header() performance is also critical for non GSO TCP
operations, as it is used from skb_clone()
This patch carefully moves all the fields that were not copied in a
separate zone : cloned, nohdr, fclone, peeked, head_frag, xmit_more
Then I moved all other fields and all other copied fields in a section
delimited by headers_start[0]/headers_end[0] section so that we
can use a single memcpy() call, inlined by compiler using long
word load/stores.
I also tried to make all copies in the natural orders of sk_buff,
to help hardware prefetching.
I made sure sk_buff size did not change.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-09-28 22:18:47 -07:00
|
|
|
|
2010-05-11 23:19:48 +00:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
net: reorganize sk_buff for faster __copy_skb_header()
With proliferation of bit fields in sk_buff, __copy_skb_header() became
quite expensive, showing as the most expensive function in a GSO
workload.
__copy_skb_header() performance is also critical for non GSO TCP
operations, as it is used from skb_clone()
This patch carefully moves all the fields that were not copied in a
separate zone : cloned, nohdr, fclone, peeked, head_frag, xmit_more
Then I moved all other fields and all other copied fields in a section
delimited by headers_start[0]/headers_end[0] section so that we
can use a single memcpy() call, inlined by compiler using long
word load/stores.
I also tried to make all copies in the natural orders of sk_buff,
to help hardware prefetching.
I made sure sk_buff size did not change.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-09-28 22:18:47 -07:00
|
|
|
|
2008-07-14 22:49:06 -07:00
|
|
|
|
2021-11-20 16:31:49 -08:00
|
|
|
|
net: reorganize sk_buff for faster __copy_skb_header()
With proliferation of bit fields in sk_buff, __copy_skb_header() became
quite expensive, showing as the most expensive function in a GSO
workload.
__copy_skb_header() performance is also critical for non GSO TCP
operations, as it is used from skb_clone()
This patch carefully moves all the fields that were not copied in a
separate zone : cloned, nohdr, fclone, peeked, head_frag, xmit_more
Then I moved all other fields and all other copied fields in a section
delimited by headers_start[0]/headers_end[0] section so that we
can use a single memcpy() call, inlined by compiler using long
word load/stores.
I also tried to make all copies in the natural orders of sk_buff,
to help hardware prefetching.
I made sure sk_buff size did not change.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-09-28 22:18:47 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-06-10 11:39:50 +03:00
|
|
|
|
2021-11-20 16:31:49 -08:00
|
|
|
|
net: reorganize sk_buff for faster __copy_skb_header()
With proliferation of bit fields in sk_buff, __copy_skb_header() became
quite expensive, showing as the most expensive function in a GSO
workload.
__copy_skb_header() performance is also critical for non GSO TCP
operations, as it is used from skb_clone()
This patch carefully moves all the fields that were not copied in a
separate zone : cloned, nohdr, fclone, peeked, head_frag, xmit_more
Then I moved all other fields and all other copied fields in a section
delimited by headers_start[0]/headers_end[0] section so that we
can use a single memcpy() call, inlined by compiler using long
word load/stores.
I also tried to make all copies in the natural orders of sk_buff,
to help hardware prefetching.
I made sure sk_buff size did not change.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-09-28 22:18:47 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-08-01 11:10:25 +08:00
|
|
|
|
net: reorganize sk_buff for faster __copy_skb_header()
With proliferation of bit fields in sk_buff, __copy_skb_header() became
quite expensive, showing as the most expensive function in a GSO
workload.
__copy_skb_header() performance is also critical for non GSO TCP
operations, as it is used from skb_clone()
This patch carefully moves all the fields that were not copied in a
separate zone : cloned, nohdr, fclone, peeked, head_frag, xmit_more
Then I moved all other fields and all other copied fields in a section
delimited by headers_start[0]/headers_end[0] section so that we
can use a single memcpy() call, inlined by compiler using long
word load/stores.
I also tried to make all copies in the natural orders of sk_buff,
to help hardware prefetching.
I made sure sk_buff size did not change.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-09-28 22:18:47 -07:00
|
|
|
|
2013-06-10 11:39:50 +03:00
|
|
|
|
net: generalize skb freeing deferral to per-cpu lists
Logic added in commit f35f821935d8 ("tcp: defer skb freeing after socket
lock is released") helped bulk TCP flows to move the cost of skbs
frees outside of critical section where socket lock was held.
But for RPC traffic, or hosts with RFS enabled, the solution is far from
being ideal.
For RPC traffic, recvmsg() has to return to user space right after
skb payload has been consumed, meaning that BH handler has no chance
to pick the skb before recvmsg() thread. This issue is more visible
with BIG TCP, as more RPC fit one skb.
For RFS, even if BH handler picks the skbs, they are still picked
from the cpu on which user thread is running.
Ideally, it is better to free the skbs (and associated page frags)
on the cpu that originally allocated them.
This patch removes the per socket anchor (sk->defer_list) and
instead uses a per-cpu list, which will hold more skbs per round.
This new per-cpu list is drained at the end of net_action_rx(),
after incoming packets have been processed, to lower latencies.
In normal conditions, skbs are added to the per-cpu list with
no further action. In the (unlikely) cases where the cpu does not
run net_action_rx() handler fast enough, we use an IPI to raise
NET_RX_SOFTIRQ on the remote cpu.
Also, we do not bother draining the per-cpu list from dev_cpu_dead()
This is because skbs in this list have no requirement on how fast
they should be freed.
Note that we can add in the future a small per-cpu cache
if we see any contention on sd->defer_lock.
Tested on a pair of hosts with 100Gbit NIC, RFS enabled,
and /proc/sys/net/ipv4/tcp_rmem[2] tuned to 16MB to work around
page recycling strategy used by NIC driver (its page pool capacity
being too small compared to number of skbs/pages held in sockets
receive queues)
Note that this tuning was only done to demonstrate worse
conditions for skb freeing for this particular test.
These conditions can happen in more general production workload.
10 runs of one TCP_STREAM flow
Before:
Average throughput: 49685 Mbit.
Kernel profiles on cpu running user thread recvmsg() show high cost for
skb freeing related functions (*)
57.81% [kernel] [k] copy_user_enhanced_fast_string
(*) 12.87% [kernel] [k] skb_release_data
(*) 4.25% [kernel] [k] __free_one_page
(*) 3.57% [kernel] [k] __list_del_entry_valid
1.85% [kernel] [k] __netif_receive_skb_core
1.60% [kernel] [k] __skb_datagram_iter
(*) 1.59% [kernel] [k] free_unref_page_commit
(*) 1.16% [kernel] [k] __slab_free
1.16% [kernel] [k] _copy_to_iter
(*) 1.01% [kernel] [k] kfree
(*) 0.88% [kernel] [k] free_unref_page
0.57% [kernel] [k] ip6_rcv_core
0.55% [kernel] [k] ip6t_do_table
0.54% [kernel] [k] flush_smp_call_function_queue
(*) 0.54% [kernel] [k] free_pcppages_bulk
0.51% [kernel] [k] llist_reverse_order
0.38% [kernel] [k] process_backlog
(*) 0.38% [kernel] [k] free_pcp_prepare
0.37% [kernel] [k] tcp_recvmsg_locked
(*) 0.37% [kernel] [k] __list_add_valid
0.34% [kernel] [k] sock_rfree
0.34% [kernel] [k] _raw_spin_lock_irq
(*) 0.33% [kernel] [k] __page_cache_release
0.33% [kernel] [k] tcp_v6_rcv
(*) 0.33% [kernel] [k] __put_page
(*) 0.29% [kernel] [k] __mod_zone_page_state
0.27% [kernel] [k] _raw_spin_lock
After patch:
Average throughput: 73076 Mbit.
Kernel profiles on cpu running user thread recvmsg() looks better:
81.35% [kernel] [k] copy_user_enhanced_fast_string
1.95% [kernel] [k] _copy_to_iter
1.95% [kernel] [k] __skb_datagram_iter
1.27% [kernel] [k] __netif_receive_skb_core
1.03% [kernel] [k] ip6t_do_table
0.60% [kernel] [k] sock_rfree
0.50% [kernel] [k] tcp_v6_rcv
0.47% [kernel] [k] ip6_rcv_core
0.45% [kernel] [k] read_tsc
0.44% [kernel] [k] _raw_spin_lock_irqsave
0.37% [kernel] [k] _raw_spin_lock
0.37% [kernel] [k] native_irq_return_iret
0.33% [kernel] [k] __inet6_lookup_established
0.31% [kernel] [k] ip6_protocol_deliver_rcu
0.29% [kernel] [k] tcp_rcv_established
0.29% [kernel] [k] llist_reverse_order
v2: kdoc issue (kernel bots)
do not defer if (alloc_cpu == smp_processor_id()) (Paolo)
replace the sk_buff_head with a single-linked list (Jakub)
add a READ_ONCE()/WRITE_ONCE() for the lockless read of sd->defer_list
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Link: https://lore.kernel.org/r/20220422201237.416238-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-04-22 13:12:37 -07:00
|
|
|
|
xps: fix xps for stacked devices
A typical qdisc setup is the following :
bond0 : bonding device, using HTB hierarchy
eth1/eth2 : slaves, multiqueue NIC, using MQ + FQ qdisc
XPS allows to spread packets on specific tx queues, based on the cpu
doing the send.
Problem is that dequeues from bond0 qdisc can happen on random cpus,
due to the fact that qdisc_run() can dequeue a batch of packets.
CPUA -> queue packet P1 on bond0 qdisc, P1->ooo_okay=1
CPUA -> queue packet P2 on bond0 qdisc, P2->ooo_okay=0
CPUB -> dequeue packet P1 from bond0
enqueue packet on eth1/eth2
CPUC -> dequeue packet P2 from bond0
enqueue packet on eth1/eth2 using sk cache (ooo_okay is 0)
get_xps_queue() then might select wrong queue for P1, since current cpu
might be different than CPUA.
P2 might be sent on the old queue (stored in sk->sk_tx_queue_mapping),
if CPUC runs a bit faster (or CPUB spins a bit on qdisc lock)
Effect of this bug is TCP reorders, and more generally not optimal
TX queue placement. (A victim bulk flow can be migrated to the wrong TX
queue for a while)
To fix this, we have to record sender cpu number the first time
dev_queue_xmit() is called for one tx skb.
We can union napi_id (used on receive path) and sender_cpu,
granted we clear sender_cpu in skb_scrub_packet() (credit to Willem for
this union idea)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-02-03 23:48:24 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: reorganize sk_buff for faster __copy_skb_header()
With proliferation of bit fields in sk_buff, __copy_skb_header() became
quite expensive, showing as the most expensive function in a GSO
workload.
__copy_skb_header() performance is also critical for non GSO TCP
operations, as it is used from skb_clone()
This patch carefully moves all the fields that were not copied in a
separate zone : cloned, nohdr, fclone, peeked, head_frag, xmit_more
Then I moved all other fields and all other copied fields in a section
delimited by headers_start[0]/headers_end[0] section so that we
can use a single memcpy() call, inlined by compiler using long
word load/stores.
I also tried to make all copies in the natural orders of sk_buff,
to help hardware prefetching.
I made sure sk_buff size did not change.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-09-28 22:18:47 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-10-14 00:37:30 -07:00
|
|
|
|
|
|
|
|
|
2009-05-22 22:11:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-10-14 00:37:52 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-10-14 00:37:30 -07:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2007-03-16 15:00:46 -07:00
|
|
|
|
[SKBUFF]: Keep track of writable header len of headerless clones
Currently NAT (and others) that want to modify cloned skbs copy them,
even if in the vast majority of cases its not necessary because the
skb is a clone made by TCP and the portion NAT wants to modify is
actually writable because TCP release the header reference before
cloning.
The problem is that there is no clean way for NAT to find out how
long the writable header area is, so this patch introduces skb->hdr_len
to hold this length. When a headerless skb is cloned skb->hdr_len
is set to the current headroom, for regular clones it is copied from
the original. A new function skb_clone_writable(skb, len) returns
whether the skb is writable up to len bytes from skb->data. To avoid
enlarging the skb the mac_len field is reduced to 16 bit and the
new hdr_len field is put in the remaining 16 bit.
I've done a few rough benchmarks of NAT (not with this exact patch,
but a very similar one). As expected it saves huge amounts of system
time in case of sendfile, bringing it down to basically the same
amount as without NAT, with sendmsg it only helps on loopback,
probably because of the large MTU.
Transmit a 1GB file using sendfile/sendmsg over eth0/lo with and
without NAT:
- sendfile eth0, no NAT: sys 0m0.388s
- sendfile eth0, NAT: sys 0m1.835s
- sendfile eth0: NAT + path: sys 0m0.370s (~ -80%)
- sendfile lo, no NAT: sys 0m0.258s
- sendfile lo, NAT: sys 0m2.609s
- sendfile lo, NAT + patch: sys 0m0.260s (~ -90%)
- sendmsg eth0, no NAT: sys 0m2.508s
- sendmsg eth0, NAT: sys 0m2.539s
- sendmsg eth0, NAT + patch: sys 0m2.445s (no change)
- sendmsg lo, no NAT: sys 0m2.151s
- sendmsg lo, NAT: sys 0m3.557s
- sendmsg lo, NAT + patch: sys 0m2.159s (~ -40%)
I expect other users can see a similar performance improvement,
packet mangling iptables targets, ipip and ip_gre come to mind ..
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-06-25 04:35:20 -07:00
|
|
|
|
2008-01-07 21:56:41 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2018-04-07 13:42:39 -07:00
|
|
|
|
2018-07-13 13:21:07 +02:00
|
|
|
|
2021-06-07 21:02:38 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-01-07 21:56:41 -08:00
|
|
|
|
net: allow skb->head to be a page fragment
skb->head is currently allocated from kmalloc(). This is convenient but
has the drawback the data cannot be converted to a page fragment if
needed.
We have three spots were it hurts :
1) GRO aggregation
When a linear skb must be appended to another skb, GRO uses the
frag_list fallback, very inefficient since we keep all struct sk_buff
around. So drivers enabling GRO but delivering linear skbs to network
stack aren't enabling full GRO power.
2) splice(socket -> pipe).
We must copy the linear part to a page fragment.
This kind of defeats splice() purpose (zero copy claim)
3) TCP coalescing.
Recently introduced, this permits to group several contiguous segments
into a single skb. This shortens queue lengths and save kernel memory,
and greatly reduce probabilities of TCP collapses. This coalescing
doesnt work on linear skbs (or we would need to copy data, this would be
too slow)
Given all these issues, the following patch introduces the possibility
of having skb->head be a fragment in itself. We use a new skb flag,
skb->head_frag to carry this information.
build_skb() is changed to accept a frag_size argument. Drivers willing
to provide a page fragment instead of kmalloc() data will set a non zero
value, set to the fragment size.
Then, on situations we need to convert the skb head to a frag in itself,
we can check if skb->head_frag is set and avoid the copies or various
fallbacks we have.
This means drivers currently using frags could be updated to avoid the
current skb->head allocation and reduce their memory footprint (aka skb
truesize). (thats 512 or 1024 bytes saved per skb). This also makes
bpf/netfilter faster since the 'first frag' will be part of skb linear
part, no need to copy data.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Maciej Żenczykowski <maze@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Cc: Ben Hutchings <bhutchings@solarflare.com>
Cc: Matt Carlson <mcarlson@broadcom.com>
Cc: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-27 00:33:38 +00:00
|
|
|
|
2008-01-07 21:56:41 -08:00
|
|
|
|
|
|
|
|
|
2017-06-30 13:07:58 +03:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-10-14 00:37:52 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-03 15:16:58 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-10-14 00:37:52 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
2007-10-14 00:37:52 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-02-15 10:49:32 -08:00
|
|
|
|
2017-08-03 16:29:43 -04:00
|
|
|
|
2023-02-14 10:57:40 -05:00
|
|
|
|
2017-08-03 16:29:43 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-02-14 10:57:40 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:43 -04:00
|
|
|
|
2023-02-14 10:57:40 -05:00
|
|
|
|
2017-08-03 16:29:43 -04:00
|
|
|
|
|
|
|
|
|
2022-11-15 09:10:56 +00:00
|
|
|
|
2017-08-03 16:29:43 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-11-15 09:10:56 +00:00
|
|
|
|
2017-08-03 16:29:43 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-02-15 10:49:32 -08:00
|
|
|
|
2017-08-03 16:29:43 -04:00
|
|
|
|
2018-02-15 10:49:32 -08:00
|
|
|
|
2017-08-03 16:29:43 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-02-15 10:49:32 -08:00
|
|
|
|
2017-08-03 16:29:43 -04:00
|
|
|
|
2025-05-08 00:48:24 +00:00
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
2022-09-23 17:39:04 +01:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:43 -04:00
|
|
|
|
|
|
|
|
|
2025-05-08 00:48:24 +00:00
|
|
|
|
2017-08-03 16:29:43 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
2024-04-19 12:08:39 +01:00
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
2022-09-23 17:39:04 +01:00
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
|
|
|
|
|
2022-09-23 17:39:04 +01:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
|
|
|
|
|
2022-09-23 17:39:04 +01:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-01-06 14:18:37 -08:00
|
|
|
|
2025-05-08 00:48:24 +00:00
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
|
|
|
|
|
2022-09-23 17:39:04 +01:00
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-12 21:52:27 +01:00
|
|
|
|
2024-04-19 12:08:39 +01:00
|
|
|
|
2022-07-12 21:52:27 +01:00
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-23 17:39:04 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-23 17:39:04 +01:00
|
|
|
|
2025-05-08 00:48:24 +00:00
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:43 -04:00
|
|
|
|
2022-09-23 17:39:04 +01:00
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
2019-05-30 18:01:21 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-01-06 14:18:41 -08:00
|
|
|
|
2019-05-30 18:01:21 -04:00
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-05-08 00:48:24 +00:00
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
2021-01-06 14:18:37 -08:00
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-23 17:39:04 +01:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
|
|
|
|
|
2021-06-09 18:41:57 -04:00
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
2017-08-09 19:09:43 -04:00
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-06-09 18:41:57 -04:00
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
|
|
|
|
|
2021-06-09 18:41:57 -04:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
2021-06-27 18:48:21 -04:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-01-06 14:18:31 -08:00
|
|
|
|
2024-04-19 12:08:39 +01:00
|
|
|
|
|
|
|
|
|
2021-01-06 14:18:31 -08:00
|
|
|
|
2022-09-23 17:39:04 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-01-06 14:18:31 -08:00
|
|
|
|
|
|
|
|
|
2022-09-23 17:39:04 +01:00
|
|
|
|
2021-01-06 14:18:31 -08:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
2021-01-06 14:18:37 -08:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
2022-09-23 17:39:04 +01:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
2021-01-06 14:18:35 -08:00
|
|
|
|
2022-09-23 17:39:04 +01:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
2021-01-06 14:18:35 -08:00
|
|
|
|
2024-04-19 12:08:39 +01:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
2021-01-06 14:18:37 -08:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
2024-04-19 12:08:39 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
|
|
|
|
|
2025-05-08 00:48:24 +00:00
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-04-19 12:08:40 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-06-27 13:59:45 +01:00
|
|
|
|
|
|
|
|
|
2024-04-19 12:08:40 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:42 -04:00
|
|
|
|
2025-05-08 00:48:24 +00:00
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
2017-10-19 12:40:39 -04:00
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
2022-04-28 11:57:46 +01:00
|
|
|
|
2017-10-19 12:40:39 -04:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
2017-10-19 12:40:39 -04:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-06-27 13:59:41 +01:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-07-12 21:52:31 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
sock: enable MSG_ZEROCOPY
Prepare the datapath for refcounted ubuf_info. Clone ubuf_info with
skb_zerocopy_clone() wherever needed due to skb split, merge, resize
or clone.
Split skb_orphan_frags into two variants. The split, merge, .. paths
support reference counted zerocopy buffers, so do not do a deep copy.
Add skb_orphan_frags_rx for paths that may loop packets to receive
sockets. That is not allowed, as it may cause unbounded latency.
Deep copy all zerocopy copy buffers, ref-counted or not, in this path.
The exact locations to modify were chosen by exhaustively searching
through all code that might modify skb_frag references and/or the
the SKBTX_DEV_ZEROCOPY tx_flags bit.
The changes err on the safe side, in two ways.
(1) legacy ubuf_info paths virtio and tap are not modified. They keep
a 1:1 ubuf_info to sk_buff relationship. Calls to skb_orphan_frags
still call skb_copy_ubufs and thus copy frags in this case.
(2) not all copies deep in the stack are addressed yet. skb_shift,
skb_split and skb_try_coalesce can be refined to avoid copying.
These are not in the hot path and this patch is hairy enough as
is, so that is left for future refinement.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-03 16:29:41 -04:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-30 15:32:40 -05:00
|
|
|
|
2017-08-03 16:29:39 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-07-10 10:55:09 +00:00
|
|
|
|
|
|
|
|
|
2011-08-31 08:03:29 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-01-06 14:18:38 -08:00
|
|
|
|
2011-08-31 08:03:29 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-07-06 12:22:12 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-04-28 04:32:31 +00:00
|
|
|
|
2017-08-03 16:29:38 -04:00
|
|
|
|
2011-07-06 12:22:12 +00:00
|
|
|
|
2017-08-03 16:29:38 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:52 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-28 12:38:13 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-04-28 04:32:31 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:38 -04:00
|
|
|
|
2023-04-28 04:32:31 +00:00
|
|
|
|
2011-07-06 12:22:12 +00:00
|
|
|
|
|
|
|
|
|
2013-07-12 16:17:23 +09:00
|
|
|
|
2011-07-06 12:22:12 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:38 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
2017-08-03 16:29:38 -04:00
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
2017-08-03 16:29:38 -04:00
|
|
|
|
|
|
|
|
|
2023-04-28 04:32:31 +00:00
|
|
|
|
2017-08-03 16:29:38 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-04-28 04:32:31 +00:00
|
|
|
|
2017-08-03 16:29:38 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
|
|
|
|
|
2011-07-06 12:22:12 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-07-17 02:05:29 +00:00
|
|
|
|
2011-10-19 23:01:49 +00:00
|
|
|
|
2011-07-06 12:22:12 +00:00
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:38 -04:00
|
|
|
|
2024-02-14 14:34:03 -08:00
|
|
|
|
2013-07-12 16:17:23 +09:00
|
|
|
|
2011-07-06 12:22:12 +00:00
|
|
|
|
2024-02-14 14:34:03 -08:00
|
|
|
|
|
|
|
|
|
2017-08-03 16:29:38 -04:00
|
|
|
|
2011-08-31 08:03:29 +00:00
|
|
|
|
2017-12-20 17:37:50 -05:00
|
|
|
|
sock: enable MSG_ZEROCOPY
Prepare the datapath for refcounted ubuf_info. Clone ubuf_info with
skb_zerocopy_clone() wherever needed due to skb split, merge, resize
or clone.
Split skb_orphan_frags into two variants. The split, merge, .. paths
support reference counted zerocopy buffers, so do not do a deep copy.
Add skb_orphan_frags_rx for paths that may loop packets to receive
sockets. That is not allowed, as it may cause unbounded latency.
Deep copy all zerocopy copy buffers, ref-counted or not, in this path.
The exact locations to modify were chosen by exhaustively searching
through all code that might modify skb_frag references and/or the
the SKBTX_DEV_ZEROCOPY tx_flags bit.
The changes err on the safe side, in two ways.
(1) legacy ubuf_info paths virtio and tap are not modified. They keep
a 1:1 ubuf_info to sk_buff relationship. Calls to skb_orphan_frags
still call skb_copy_ubufs and thus copy frags in this case.
(2) not all copies deep in the stack are addressed yet. skb_shift,
skb_split and skb_try_coalesce can be refined to avoid copying.
These are not in the hot path and this patch is hairy enough as
is, so that is left for future refinement.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-03 16:29:41 -04:00
|
|
|
|
2011-07-06 12:22:12 +00:00
|
|
|
|
|
|
|
|
|
2012-07-20 09:23:20 +00:00
|
|
|
|
2011-07-06 12:22:12 +00:00
|
|
|
|
2007-10-14 00:37:52 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-09-29 13:29:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-12-03 17:04:39 -08:00
|
|
|
|
2007-10-14 00:37:52 -07:00
|
|
|
|
2012-07-20 09:23:10 +00:00
|
|
|
|
|
|
|
|
|
2011-07-06 12:22:12 +00:00
|
|
|
|
2007-10-14 00:37:52 -07:00
|
|
|
|
2017-06-30 13:07:59 +03:00
|
|
|
|
2014-12-03 17:04:39 -08:00
|
|
|
|
2017-06-30 13:07:59 +03:00
|
|
|
|
2022-07-07 19:18:46 +00:00
|
|
|
|
2007-10-14 00:37:52 -07:00
|
|
|
|
2012-07-31 16:44:19 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-06 16:00:23 +00:00
|
|
|
|
2007-10-14 00:37:52 -07:00
|
|
|
|
|
|
|
|
|
2008-08-30 12:16:35 +02:00
|
|
|
|
2007-10-14 00:37:52 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2018-08-03 16:58:09 +09:00
|
|
|
|
2013-03-07 13:21:40 +00:00
|
|
|
|
2013-10-19 11:42:54 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-03-07 13:21:40 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-03-07 13:21:46 +00:00
|
|
|
|
2013-03-07 13:21:40 +00:00
|
|
|
|
2018-08-03 16:58:09 +09:00
|
|
|
|
2013-03-07 13:21:40 +00:00
|
|
|
|
2018-04-30 10:16:11 +03:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-10-14 00:37:30 -07:00
|
|
|
|
|
|
|
|
|
2006-06-22 02:40:14 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2018-04-30 10:16:11 +03:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2012-07-31 16:44:19 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-10-07 07:46:04 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2024-04-27 20:24:19 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2010-09-01 05:25:10 +00:00
|
|
|
|
2024-09-10 17:14:52 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-04-27 20:24:19 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-10-09 11:37:59 -04:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2018-04-30 10:16:11 +03:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2014-06-12 01:36:26 +03:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2011-12-03 21:39:53 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-06-12 01:36:26 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-06-12 01:36:26 +03:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2011-12-03 21:39:53 +00:00
|
|
|
|
2014-06-12 01:36:26 +03:00
|
|
|
|
|
|
|
|
|
2010-09-01 05:25:10 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-12-03 21:39:53 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-03-27 18:55:52 -03:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-11-07 14:57:15 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
sock: enable MSG_ZEROCOPY
Prepare the datapath for refcounted ubuf_info. Clone ubuf_info with
skb_zerocopy_clone() wherever needed due to skb split, merge, resize
or clone.
Split skb_orphan_frags into two variants. The split, merge, .. paths
support reference counted zerocopy buffers, so do not do a deep copy.
Add skb_orphan_frags_rx for paths that may loop packets to receive
sockets. That is not allowed, as it may cause unbounded latency.
Deep copy all zerocopy copy buffers, ref-counted or not, in this path.
The exact locations to modify were chosen by exhaustively searching
through all code that might modify skb_frag references and/or the
the SKBTX_DEV_ZEROCOPY tx_flags bit.
The changes err on the safe side, in two ways.
(1) legacy ubuf_info paths virtio and tap are not modified. They keep
a 1:1 ubuf_info to sk_buff relationship. Calls to skb_orphan_frags
still call skb_copy_ubufs and thus copy frags in this case.
(2) not all copies deep in the stack are addressed yet. skb_shift,
skb_split and skb_try_coalesce can be refined to avoid copying.
These are not in the hot path and this patch is hairy enough as
is, so that is left for future refinement.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-03 16:29:41 -04:00
|
|
|
|
|
|
|
|
|
2012-07-20 09:23:10 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-07-06 12:22:12 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2011-08-22 23:44:58 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
net: skbuff: propagate shared-frag marker through frag-transfer helpers
Two frag-transfer helpers (__pskb_copy_fclone() and skb_shift()) fail
to propagate the SKBFL_SHARED_FRAG bit in skb_shinfo()->flags when
moving frags from source to destination. __pskb_copy_fclone() defers
the rest of the shinfo metadata to skb_copy_header() after copying
frag descriptors, but that helper only carries over gso_{size,segs,
type} and never touches skb_shinfo()->flags; skb_shift() moves frag
descriptors directly and leaves flags untouched. As a result, the
destination skb keeps a reference to the same externally-owned or
page-cache-backed pages while reporting skb_has_shared_frag() as
false.
The mismatch is harmful in any in-place writer that uses
skb_has_shared_frag() to decide whether shared pages must be detoured
through skb_cow_data(). ESP input is one such writer (esp4.c,
esp6.c), and a single nft 'dup to <local>' rule -- or any other
nf_dup_ipv4() / xt_TEE caller -- is enough to land a pskb_copy()'d
skb in esp_input() with the marker stripped, letting an unprivileged
user write into the page cache of a root-owned read-only file via
authencesn-ESN stray writes.
Set SKBFL_SHARED_FRAG on the destination whenever frag descriptors
were actually moved from the source. skb_copy() and skb_copy_expand()
share skb_copy_header() too but linearize all paged data into freshly
allocated head storage and emerge with nr_frags == 0, so
skb_has_shared_frag() returns false on its own; they need no change.
The same omission exists in skb_gro_receive() and skb_gro_receive_list().
The former moves the incoming skb's frag descriptors into the
accumulator's last sub-skb via two paths (a direct frag-move loop and
the head_frag + memcpy path); the latter chains the incoming skb whole
onto p's frag_list. Downstream skb_segment() reads only
skb_shinfo(p)->flags, and skb_segment_list() reuses each sub-skb's
shinfo as the nskb -- both p and lp must carry the marker.
The same omission also exists in tcp_clone_payload(), which builds an
MTU probe skb by moving frag descriptors from skbs on sk_write_queue
into a freshly allocated nskb. The helper falls into the same family
and warrants the same fix for consistency; no TCP TX-side in-place
writer is currently known to reach a user page through this gap, but
a future consumer depending on the marker would regress silently.
The same omission exists in skb_segment(): the per-iteration flag
merge takes only head_skb's flag, and the inner switch that rebinds
frag_skb to list_skb on head_skb-frags exhaustion does not fold the
new frag_skb's flag into nskb. Fold frag_skb's flag at both sites
so segments drawing frags from frag_list members carry the marker.
Fixes: cef401de7be8 ("net: fix possible wrong checksum generation")
Fixes: f4c50a4034e6 ("xfrm: esp: avoid in-place decrypt on shared skb frags")
Suggested-by: Sabrina Dubroca <sd@queasysnail.net>
Suggested-by: Sultan Alsawaf <sultan@kerneltoast.com>
Suggested-by: Ben Hutchings <ben@decadent.org.uk>
Suggested-by: Lin Ma <malin89@huawei.com>
Suggested-by: Jingguo Tan <tanjingguo@huawei.com>
Suggested-by: Aaron Esau <aaron1esau@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Hyunwoo Kim <imv4bel@gmail.com>
Tested-by: Rajat Gupta <rajat.gupta@oss.qualcomm.com>
Link: https://patch.msgid.link/ageeJfJHwgzmKXbh@v4bel
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-05-16 07:28:53 +09:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2010-08-23 00:13:46 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-04-30 10:16:11 +03:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-06-12 01:36:26 +03:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-11-07 14:18:26 +01:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-05 21:19:39 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2005-07-08 14:57:47 -07:00
|
|
|
|
2005-10-07 07:46:04 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-10-25 15:39:35 -07:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2017-01-27 07:11:27 -08:00
|
|
|
|
2022-10-25 15:39:35 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-10-01 07:09:38 -07:00
|
|
|
|
|
|
|
|
|
2017-10-09 11:37:59 -04:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-07-12 21:52:31 +01:00
|
|
|
|
|
|
|
|
|
2012-07-31 16:44:19 -07:00
|
|
|
|
|
|
|
|
|
2022-10-25 15:39:35 -07:00
|
|
|
|
2023-02-06 17:31:02 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2022-10-25 15:39:35 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2010-09-01 05:25:10 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2010-07-22 19:09:08 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2012-05-04 14:26:51 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-07-20 09:23:10 +00:00
|
|
|
|
|
|
|
|
|
sock: enable MSG_ZEROCOPY
Prepare the datapath for refcounted ubuf_info. Clone ubuf_info with
skb_zerocopy_clone() wherever needed due to skb split, merge, resize
or clone.
Split skb_orphan_frags into two variants. The split, merge, .. paths
support reference counted zerocopy buffers, so do not do a deep copy.
Add skb_orphan_frags_rx for paths that may loop packets to receive
sockets. That is not allowed, as it may cause unbounded latency.
Deep copy all zerocopy copy buffers, ref-counted or not, in this path.
The exact locations to modify were chosen by exhaustively searching
through all code that might modify skb_frag references and/or the
the SKBTX_DEV_ZEROCOPY tx_flags bit.
The changes err on the safe side, in two ways.
(1) legacy ubuf_info paths virtio and tap are not modified. They keep
a 1:1 ubuf_info to sk_buff relationship. Calls to skb_orphan_frags
still call skb_copy_ubufs and thus copy frags in this case.
(2) not all copies deep in the stack are addressed yet. skb_shift,
skb_split and skb_try_coalesce can be refined to avoid copying.
These are not in the hot path and this patch is hairy enough as
is, so that is left for future refinement.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-03 16:29:41 -04:00
|
|
|
|
2017-08-31 16:48:22 -07:00
|
|
|
|
2010-09-02 23:09:32 +00:00
|
|
|
|
2011-08-22 23:44:58 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2010-09-02 23:09:32 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
2012-05-04 14:26:51 +00:00
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
2010-09-02 23:09:32 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: allow skb->head to be a page fragment
skb->head is currently allocated from kmalloc(). This is convenient but
has the drawback the data cannot be converted to a page fragment if
needed.
We have three spots were it hurts :
1) GRO aggregation
When a linear skb must be appended to another skb, GRO uses the
frag_list fallback, very inefficient since we keep all struct sk_buff
around. So drivers enabling GRO but delivering linear skbs to network
stack aren't enabling full GRO power.
2) splice(socket -> pipe).
We must copy the linear part to a page fragment.
This kind of defeats splice() purpose (zero copy claim)
3) TCP coalescing.
Recently introduced, this permits to group several contiguous segments
into a single skb. This shortens queue lengths and save kernel memory,
and greatly reduce probabilities of TCP collapses. This coalescing
doesnt work on linear skbs (or we would need to copy data, this would be
too slow)
Given all these issues, the following patch introduces the possibility
of having skb->head be a fragment in itself. We use a new skb flag,
skb->head_frag to carry this information.
build_skb() is changed to accept a frag_size argument. Drivers willing
to provide a page fragment instead of kmalloc() data will set a non zero
value, set to the fragment size.
Then, on situations we need to convert the skb head to a frag in itself,
we can check if skb->head_frag is set and avoid the copies or various
fallbacks we have.
This means drivers currently using frags could be updated to avoid the
current skb->head allocation and reduce their memory footprint (aka skb
truesize). (thats 512 or 1024 bytes saved per skb). This also makes
bpf/netfilter faster since the 'first frag' will be part of skb linear
part, no need to copy data.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Maciej Żenczykowski <maze@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Cc: Ben Hutchings <bhutchings@solarflare.com>
Cc: Matt Carlson <mcarlson@broadcom.com>
Cc: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-27 00:33:38 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-02-21 19:21:12 -08:00
|
|
|
|
|
|
|
|
|
2007-04-19 20:43:29 -07:00
|
|
|
|
2007-04-09 11:45:04 -07:00
|
|
|
|
|
|
|
|
|
2007-04-19 20:29:13 -07:00
|
|
|
|
2013-06-06 21:27:21 +08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
[SKBUFF]: Keep track of writable header len of headerless clones
Currently NAT (and others) that want to modify cloned skbs copy them,
even if in the vast majority of cases its not necessary because the
skb is a clone made by TCP and the portion NAT wants to modify is
actually writable because TCP release the header reference before
cloning.
The problem is that there is no clean way for NAT to find out how
long the writable header area is, so this patch introduces skb->hdr_len
to hold this length. When a headerless skb is cloned skb->hdr_len
is set to the current headroom, for regular clones it is copied from
the original. A new function skb_clone_writable(skb, len) returns
whether the skb is writable up to len bytes from skb->data. To avoid
enlarging the skb the mac_len field is reduced to 16 bit and the
new hdr_len field is put in the remaining 16 bit.
I've done a few rough benchmarks of NAT (not with this exact patch,
but a very similar one). As expected it saves huge amounts of system
time in case of sendfile, bringing it down to basically the same
amount as without NAT, with sendmsg it only helps on loopback,
probably because of the large MTU.
Transmit a 1GB file using sendfile/sendmsg over eth0/lo with and
without NAT:
- sendfile eth0, no NAT: sys 0m0.388s
- sendfile eth0, NAT: sys 0m1.835s
- sendfile eth0: NAT + path: sys 0m0.370s (~ -80%)
- sendfile lo, no NAT: sys 0m0.258s
- sendfile lo, NAT: sys 0m2.609s
- sendfile lo, NAT + patch: sys 0m0.260s (~ -90%)
- sendmsg eth0, no NAT: sys 0m2.508s
- sendmsg eth0, NAT: sys 0m2.539s
- sendmsg eth0, NAT + patch: sys 0m2.445s (no change)
- sendmsg lo, no NAT: sys 0m2.151s
- sendmsg lo, NAT: sys 0m3.557s
- sendmsg lo, NAT + patch: sys 0m2.159s (~ -40%)
I expect other users can see a similar performance improvement,
packet mangling iptables targets, ipip and ip_gre come to mind ..
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-06-25 04:35:20 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2017-01-27 07:11:27 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2011-07-06 12:22:12 +00:00
|
|
|
|
2026-04-10 11:47:32 +08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
net: avoid skb end_offset change in __skb_unclone_keeptruesize()
Once initial skb->head has been allocated from skb_small_head_cache,
we need to make sure to use the same strategy whenever skb->head
has to be re-allocated, as found by syzbot [1]
This means kmalloc_reserve() can not fallback from using
skb_small_head_cache to generic (power-of-two) kmem caches.
It seems that we probably want to rework things in the future,
to partially revert following patch, because we no longer use
ksize() for skb allocated in TX path.
2b88cba55883 ("net: preserve skb_end_offset() in skb_unclone_keeptruesize()")
Ideally, TCP stack should never put payload in skb->head,
this effort has to be completed.
In the mean time, add a sanity check.
[1]
BUG: KASAN: invalid-free in slab_free mm/slub.c:3787 [inline]
BUG: KASAN: invalid-free in kmem_cache_free+0xee/0x5c0 mm/slub.c:3809
Free of addr ffff88806cdee800 by task syz-executor239/5189
CPU: 0 PID: 5189 Comm: syz-executor239 Not tainted 6.2.0-rc8-syzkaller-02400-gd1fabc68f8e0 #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/21/2023
Call Trace:
<TASK>
__dump_stack lib/dump_stack.c:88 [inline]
dump_stack_lvl+0xd1/0x138 lib/dump_stack.c:106
print_address_description mm/kasan/report.c:306 [inline]
print_report+0x15e/0x45d mm/kasan/report.c:417
kasan_report_invalid_free+0x9b/0x1b0 mm/kasan/report.c:482
____kasan_slab_free+0x1a5/0x1c0 mm/kasan/common.c:216
kasan_slab_free include/linux/kasan.h:177 [inline]
slab_free_hook mm/slub.c:1781 [inline]
slab_free_freelist_hook+0x8b/0x1c0 mm/slub.c:1807
slab_free mm/slub.c:3787 [inline]
kmem_cache_free+0xee/0x5c0 mm/slub.c:3809
skb_kfree_head net/core/skbuff.c:857 [inline]
skb_kfree_head net/core/skbuff.c:853 [inline]
skb_free_head+0x16f/0x1a0 net/core/skbuff.c:872
skb_release_data+0x57a/0x820 net/core/skbuff.c:901
skb_release_all net/core/skbuff.c:966 [inline]
__kfree_skb+0x4f/0x70 net/core/skbuff.c:980
tcp_wmem_free_skb include/net/tcp.h:302 [inline]
tcp_rtx_queue_purge net/ipv4/tcp.c:3061 [inline]
tcp_write_queue_purge+0x617/0xcf0 net/ipv4/tcp.c:3074
tcp_v4_destroy_sock+0x125/0x810 net/ipv4/tcp_ipv4.c:2302
inet_csk_destroy_sock+0x19a/0x440 net/ipv4/inet_connection_sock.c:1195
__tcp_close+0xb96/0xf50 net/ipv4/tcp.c:3021
tcp_close+0x2d/0xc0 net/ipv4/tcp.c:3033
inet_release+0x132/0x270 net/ipv4/af_inet.c:426
__sock_release+0xcd/0x280 net/socket.c:651
sock_close+0x1c/0x20 net/socket.c:1393
__fput+0x27c/0xa90 fs/file_table.c:320
task_work_run+0x16f/0x270 kernel/task_work.c:179
resume_user_mode_work include/linux/resume_user_mode.h:49 [inline]
exit_to_user_mode_loop kernel/entry/common.c:171 [inline]
exit_to_user_mode_prepare+0x23c/0x250 kernel/entry/common.c:203
__syscall_exit_to_user_mode_work kernel/entry/common.c:285 [inline]
syscall_exit_to_user_mode+0x1d/0x50 kernel/entry/common.c:296
do_syscall_64+0x46/0xb0 arch/x86/entry/common.c:86
entry_SYSCALL_64_after_hwframe+0x63/0xcd
RIP: 0033:0x7f2511f546c3
Code: c7 c2 c0 ff ff ff f7 d8 64 89 02 48 c7 c0 ff ff ff ff eb b7 0f 1f 00 64 8b 04 25 18 00 00 00 85 c0 75 14 b8 03 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 45 c3 0f 1f 40 00 48 83 ec 18 89 7c 24 0c e8
RSP: 002b:00007ffef0103d48 EFLAGS: 00000246 ORIG_RAX: 0000000000000003
RAX: 0000000000000000 RBX: 0000000000000004 RCX: 00007f2511f546c3
RDX: 0000000000000978 RSI: 00000000200000c0 RDI: 0000000000000003
RBP: 0000000000000000 R08: 0000000000000002 R09: 0000000000003434
R10: 0000000000000000 R11: 0000000000000246 R12: 00007ffef0103d6c
R13: 00007ffef0103d80 R14: 00007ffef0103dc0 R15: 0000000000000003
</TASK>
Allocated by task 5189:
kasan_save_stack+0x22/0x40 mm/kasan/common.c:45
kasan_set_track+0x25/0x30 mm/kasan/common.c:52
____kasan_kmalloc mm/kasan/common.c:374 [inline]
____kasan_kmalloc mm/kasan/common.c:333 [inline]
__kasan_kmalloc+0xa5/0xb0 mm/kasan/common.c:383
kasan_kmalloc include/linux/kasan.h:211 [inline]
__do_kmalloc_node mm/slab_common.c:968 [inline]
__kmalloc_node_track_caller+0x5b/0xc0 mm/slab_common.c:988
kmalloc_reserve+0xf1/0x230 net/core/skbuff.c:539
pskb_expand_head+0x237/0x1160 net/core/skbuff.c:1995
__skb_unclone_keeptruesize+0x93/0x220 net/core/skbuff.c:2094
skb_unclone_keeptruesize include/linux/skbuff.h:1910 [inline]
skb_prepare_for_shift net/core/skbuff.c:3804 [inline]
skb_shift+0xef8/0x1e20 net/core/skbuff.c:3877
tcp_skb_shift net/ipv4/tcp_input.c:1538 [inline]
tcp_shift_skb_data net/ipv4/tcp_input.c:1646 [inline]
tcp_sacktag_walk+0x93b/0x18a0 net/ipv4/tcp_input.c:1713
tcp_sacktag_write_queue+0x1599/0x31d0 net/ipv4/tcp_input.c:1974
tcp_ack+0x2e9f/0x5a10 net/ipv4/tcp_input.c:3847
tcp_rcv_established+0x667/0x2230 net/ipv4/tcp_input.c:6006
tcp_v4_do_rcv+0x670/0x9b0 net/ipv4/tcp_ipv4.c:1721
sk_backlog_rcv include/net/sock.h:1113 [inline]
__release_sock+0x133/0x3b0 net/core/sock.c:2921
release_sock+0x58/0x1b0 net/core/sock.c:3488
tcp_sendmsg+0x3a/0x50 net/ipv4/tcp.c:1485
inet_sendmsg+0x9d/0xe0 net/ipv4/af_inet.c:825
sock_sendmsg_nosec net/socket.c:722 [inline]
sock_sendmsg+0xde/0x190 net/socket.c:745
sock_write_iter+0x295/0x3d0 net/socket.c:1136
call_write_iter include/linux/fs.h:2189 [inline]
new_sync_write fs/read_write.c:491 [inline]
vfs_write+0x9ed/0xdd0 fs/read_write.c:584
ksys_write+0x1ec/0x250 fs/read_write.c:637
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x39/0xb0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x63/0xcd
The buggy address belongs to the object at ffff88806cdee800
which belongs to the cache kmalloc-1k of size 1024
The buggy address is located 0 bytes inside of
1024-byte region [ffff88806cdee800, ffff88806cdeec00)
The buggy address belongs to the physical page:
page:ffffea0001b37a00 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x6cde8
head:ffffea0001b37a00 order:3 compound_mapcount:0 subpages_mapcount:0 compound_pincount:0
flags: 0xfff00000010200(slab|head|node=0|zone=1|lastcpupid=0x7ff)
raw: 00fff00000010200 ffff888012441dc0 dead000000000122 0000000000000000
raw: 0000000000000000 0000000000100010 00000001ffffffff 0000000000000000
page dumped because: kasan: bad access detected
page_owner tracks the page as allocated
page last allocated via order 3, migratetype Unmovable, gfp_mask 0x1f2a20(GFP_ATOMIC|__GFP_NOWARN|__GFP_NORETRY|__GFP_COMP|__GFP_NOMEMALLOC|__GFP_MEMALLOC|__GFP_HARDWALL), pid 75, tgid 75 (kworker/u4:4), ts 96369578780, free_ts 26734162530
prep_new_page mm/page_alloc.c:2531 [inline]
get_page_from_freelist+0x119c/0x2ce0 mm/page_alloc.c:4283
__alloc_pages+0x1cb/0x5b0 mm/page_alloc.c:5549
alloc_pages+0x1aa/0x270 mm/mempolicy.c:2287
alloc_slab_page mm/slub.c:1851 [inline]
allocate_slab+0x25f/0x350 mm/slub.c:1998
new_slab mm/slub.c:2051 [inline]
___slab_alloc+0xa91/0x1400 mm/slub.c:3193
__slab_alloc.constprop.0+0x56/0xa0 mm/slub.c:3292
__slab_alloc_node mm/slub.c:3345 [inline]
slab_alloc_node mm/slub.c:3442 [inline]
__kmem_cache_alloc_node+0x1a4/0x430 mm/slub.c:3491
__do_kmalloc_node mm/slab_common.c:967 [inline]
__kmalloc_node_track_caller+0x4b/0xc0 mm/slab_common.c:988
kmalloc_reserve+0xf1/0x230 net/core/skbuff.c:539
__alloc_skb+0x129/0x330 net/core/skbuff.c:608
__netdev_alloc_skb+0x74/0x410 net/core/skbuff.c:672
__netdev_alloc_skb_ip_align include/linux/skbuff.h:3203 [inline]
netdev_alloc_skb_ip_align include/linux/skbuff.h:3213 [inline]
batadv_iv_ogm_aggregate_new+0x106/0x4e0 net/batman-adv/bat_iv_ogm.c:558
batadv_iv_ogm_queue_add net/batman-adv/bat_iv_ogm.c:670 [inline]
batadv_iv_ogm_schedule_buff+0xe6b/0x1450 net/batman-adv/bat_iv_ogm.c:849
batadv_iv_ogm_schedule net/batman-adv/bat_iv_ogm.c:868 [inline]
batadv_iv_ogm_schedule net/batman-adv/bat_iv_ogm.c:861 [inline]
batadv_iv_send_outstanding_bat_ogm_packet+0x744/0x910 net/batman-adv/bat_iv_ogm.c:1712
process_one_work+0x9bf/0x1710 kernel/workqueue.c:2289
worker_thread+0x669/0x1090 kernel/workqueue.c:2436
page last free stack trace:
reset_page_owner include/linux/page_owner.h:24 [inline]
free_pages_prepare mm/page_alloc.c:1446 [inline]
free_pcp_prepare+0x66a/0xc20 mm/page_alloc.c:1496
free_unref_page_prepare mm/page_alloc.c:3369 [inline]
free_unref_page+0x1d/0x490 mm/page_alloc.c:3464
free_contig_range+0xb5/0x180 mm/page_alloc.c:9488
destroy_args+0xa8/0x64c mm/debug_vm_pgtable.c:998
debug_vm_pgtable+0x28de/0x296f mm/debug_vm_pgtable.c:1318
do_one_initcall+0x141/0x790 init/main.c:1306
do_initcall_level init/main.c:1379 [inline]
do_initcalls init/main.c:1395 [inline]
do_basic_setup init/main.c:1414 [inline]
kernel_init_freeable+0x6f9/0x782 init/main.c:1634
kernel_init+0x1e/0x1d0 init/main.c:1522
ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308
Memory state around the buggy address:
ffff88806cdee700: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
ffff88806cdee780: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
>ffff88806cdee800: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
^
ffff88806cdee880: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
Fixes: bf9f1baa279f ("net: add dedicated kmem_cache for typical/small skb->head")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Tested-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2023-02-27 14:17:06 +00:00
|
|
|
|
2022-02-21 19:21:13 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-02 11:52:15 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-22 13:28:37 +03:00
|
|
|
|
|
|
|
|
|
2021-08-02 11:52:15 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-22 13:28:37 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-02 11:52:15 +03:00
|
|
|
|
|
|
|
|
|
2021-10-22 13:28:37 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-02 11:52:15 +03:00
|
|
|
|
|
|
|
|
|
2021-10-22 13:28:37 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-02 11:52:15 +03:00
|
|
|
|
|
|
|
|
|
2021-10-22 13:28:37 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-08-02 11:52:15 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-07-08 14:57:47 -07:00
|
|
|
|
2005-10-07 07:46:04 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-04-27 20:24:19 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2024-09-10 17:14:52 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-04-27 20:24:19 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-10 18:30:09 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-10-09 11:37:59 -04:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2018-04-30 10:16:11 +03:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2013-10-19 11:42:54 -07:00
|
|
|
|
2007-04-10 18:30:09 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2017-08-22 15:12:14 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2017-08-22 15:12:14 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-22 15:12:14 -07:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-02-09 23:24:36 +09:00
|
|
|
|
2017-08-22 15:12:14 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-06-23 02:06:41 -07:00
|
|
|
|
|
|
|
|
|
2007-02-09 23:24:36 +09:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-06-23 02:06:41 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-06-23 02:06:41 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-06-23 02:06:41 -07:00
|
|
|
|
2007-04-19 20:43:29 -07:00
|
|
|
|
2006-06-23 02:06:41 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-22 15:12:14 -07:00
|
|
|
|
|
|
|
|
|
2006-06-23 02:06:41 -07:00
|
|
|
|
2007-02-09 23:24:36 +09:00
|
|
|
|
2017-08-22 15:12:14 -07:00
|
|
|
|
2007-02-09 23:24:36 +09:00
|
|
|
|
2013-11-07 14:18:24 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
networking: make skb_put & friends return void pointers
It seems like a historic accident that these return unsigned char *,
and in many places that means casts are required, more often than not.
Make these functions (skb_put, __skb_put and pskb_put) return void *
and remove all the casts across the tree, adding a (u8 *) cast only
where the unsigned char pointer was used directly, all done with the
following spatch:
@@
expression SKB, LEN;
typedef u8;
identifier fn = { skb_put, __skb_put };
@@
- *(fn(SKB, LEN))
+ *(u8 *)fn(SKB, LEN)
@@
expression E, SKB, LEN;
identifier fn = { skb_put, __skb_put };
type T;
@@
- E = ((T *)(fn(SKB, LEN)))
+ E = fn(SKB, LEN)
which actually doesn't cover pskb_put since there are only three
users overall.
A handful of stragglers were converted manually, notably a macro in
drivers/isdn/i4l/isdn_bsdcomp.c and, oddly enough, one of the many
instances in net/bluetooth/hci_sock.c. In the former file, I also
had to fix one whitespace problem spatch introduced.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-06-16 14:29:21 +02:00
|
|
|
|
2013-11-07 14:18:24 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-03-27 17:43:41 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
networking: make skb_put & friends return void pointers
It seems like a historic accident that these return unsigned char *,
and in many places that means casts are required, more often than not.
Make these functions (skb_put, __skb_put and pskb_put) return void *
and remove all the casts across the tree, adding a (u8 *) cast only
where the unsigned char pointer was used directly, all done with the
following spatch:
@@
expression SKB, LEN;
typedef u8;
identifier fn = { skb_put, __skb_put };
@@
- *(fn(SKB, LEN))
+ *(u8 *)fn(SKB, LEN)
@@
expression E, SKB, LEN;
identifier fn = { skb_put, __skb_put };
type T;
@@
- E = ((T *)(fn(SKB, LEN)))
+ E = fn(SKB, LEN)
which actually doesn't cover pskb_put since there are only three
users overall.
A handful of stragglers were converted manually, notably a macro in
drivers/isdn/i4l/isdn_bsdcomp.c and, oddly enough, one of the many
instances in net/bluetooth/hci_sock.c. In the former file, I also
had to fix one whitespace problem spatch introduced.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-06-16 14:29:21 +02:00
|
|
|
|
2008-03-27 17:43:41 -07:00
|
|
|
|
networking: make skb_put & friends return void pointers
It seems like a historic accident that these return unsigned char *,
and in many places that means casts are required, more often than not.
Make these functions (skb_put, __skb_put and pskb_put) return void *
and remove all the casts across the tree, adding a (u8 *) cast only
where the unsigned char pointer was used directly, all done with the
following spatch:
@@
expression SKB, LEN;
typedef u8;
identifier fn = { skb_put, __skb_put };
@@
- *(fn(SKB, LEN))
+ *(u8 *)fn(SKB, LEN)
@@
expression E, SKB, LEN;
identifier fn = { skb_put, __skb_put };
type T;
@@
- E = ((T *)(fn(SKB, LEN)))
+ E = fn(SKB, LEN)
which actually doesn't cover pskb_put since there are only three
users overall.
A handful of stragglers were converted manually, notably a macro in
drivers/isdn/i4l/isdn_bsdcomp.c and, oddly enough, one of the many
instances in net/bluetooth/hci_sock.c. In the former file, I also
had to fix one whitespace problem spatch introduced.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-06-16 14:29:21 +02:00
|
|
|
|
2008-03-27 17:43:41 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-03-27 17:52:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
networking: make skb_push & __skb_push return void pointers
It seems like a historic accident that these return unsigned char *,
and in many places that means casts are required, more often than not.
Make these functions return void * and remove all the casts across
the tree, adding a (u8 *) cast only where the unsigned char pointer
was used directly, all done with the following spatch:
@@
expression SKB, LEN;
typedef u8;
identifier fn = { skb_push, __skb_push, skb_push_rcsum };
@@
- *(fn(SKB, LEN))
+ *(u8 *)fn(SKB, LEN)
@@
expression E, SKB, LEN;
identifier fn = { skb_push, __skb_push, skb_push_rcsum };
type T;
@@
- E = ((T *)(fn(SKB, LEN)))
+ E = fn(SKB, LEN)
@@
expression SKB, LEN;
identifier fn = { skb_push, __skb_push, skb_push_rcsum };
@@
- fn(SKB, LEN)[0]
+ *(u8 *)fn(SKB, LEN)
Note that the last part there converts from push(...)[0] to the
more idiomatic *(u8 *)push(...).
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-06-16 14:29:23 +02:00
|
|
|
|
2008-03-27 17:52:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-08-02 15:34:52 +05:30
|
|
|
|
2008-03-27 17:52:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-03-27 17:47:24 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
networking: make skb_pull & friends return void pointers
It seems like a historic accident that these return unsigned char *,
and in many places that means casts are required, more often than not.
Make these functions return void * and remove all the casts across
the tree, adding a (u8 *) cast only where the unsigned char pointer
was used directly, all done with the following spatch:
@@
expression SKB, LEN;
typedef u8;
identifier fn = {
skb_pull,
__skb_pull,
skb_pull_inline,
__pskb_pull_tail,
__pskb_pull,
pskb_pull
};
@@
- *(fn(SKB, LEN))
+ *(u8 *)fn(SKB, LEN)
@@
expression E, SKB, LEN;
identifier fn = {
skb_pull,
__skb_pull,
skb_pull_inline,
__pskb_pull_tail,
__pskb_pull,
pskb_pull
};
type T;
@@
- E = ((T *)(fn(SKB, LEN)))
+ E = fn(SKB, LEN)
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-06-16 14:29:22 +02:00
|
|
|
|
2008-03-27 17:47:24 -07:00
|
|
|
|
2010-05-02 02:21:44 -07:00
|
|
|
|
2008-03-27 17:47:24 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-12-01 10:54:52 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-03-27 17:54:01 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-06-09 16:13:38 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2006-06-09 16:13:38 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-07-13 19:26:39 -07:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-07-13 19:26:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-07-30 20:20:28 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
2006-07-13 19:26:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-07-13 19:26:39 -07:00
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
2006-07-13 19:26:39 -07:00
|
|
|
|
2006-07-30 20:20:28 -07:00
|
|
|
|
2006-07-13 19:26:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-08-22 23:44:58 +00:00
|
|
|
|
2006-07-13 19:26:39 -07:00
|
|
|
|
2010-08-23 00:13:46 -07:00
|
|
|
|
2006-07-13 19:26:39 -07:00
|
|
|
|
2006-07-30 20:20:28 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2006-07-13 19:26:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-04-19 02:24:53 +00:00
|
|
|
|
2006-07-13 19:26:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-07-30 20:20:28 -07:00
|
|
|
|
2006-07-13 19:26:39 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-07-13 19:26:39 -07:00
|
|
|
|
|
|
|
|
|
2007-04-19 20:29:13 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-05-22 14:06:40 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2017-04-26 09:07:46 -07:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-05-22 14:06:40 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-04-18 11:43:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-05-22 14:06:40 +02:00
|
|
|
|
2018-04-18 11:43:15 -07:00
|
|
|
|
2026-05-22 14:06:40 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-12-14 22:07:39 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-04-18 11:43:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
networking: make skb_pull & friends return void pointers
It seems like a historic accident that these return unsigned char *,
and in many places that means casts are required, more often than not.
Make these functions return void * and remove all the casts across
the tree, adding a (u8 *) cast only where the unsigned char pointer
was used directly, all done with the following spatch:
@@
expression SKB, LEN;
typedef u8;
identifier fn = {
skb_pull,
__skb_pull,
skb_pull_inline,
__pskb_pull_tail,
__pskb_pull,
pskb_pull
};
@@
- *(fn(SKB, LEN))
+ *(u8 *)fn(SKB, LEN)
@@
expression E, SKB, LEN;
identifier fn = {
skb_pull,
__skb_pull,
skb_pull_inline,
__pskb_pull_tail,
__pskb_pull,
pskb_pull
};
type T;
@@
- E = ((T *)(fn(SKB, LEN)))
+ E = fn(SKB, LEN)
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-06-16 14:29:22 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-19 20:43:29 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2024-09-10 17:14:52 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-10-09 11:37:59 -04:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2010-08-23 00:13:46 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-10-14 18:51:36 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-12-14 23:11:58 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-02-25 00:37:32 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-02-20 07:40:52 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-08-22 23:44:58 +00:00
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
|
|
|
|
|
2017-07-17 17:25:02 +08:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-07-17 17:25:02 +08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
sock: enable MSG_ZEROCOPY
Prepare the datapath for refcounted ubuf_info. Clone ubuf_info with
skb_zerocopy_clone() wherever needed due to skb split, merge, resize
or clone.
Split skb_orphan_frags into two variants. The split, merge, .. paths
support reference counted zerocopy buffers, so do not do a deep copy.
Add skb_orphan_frags_rx for paths that may loop packets to receive
sockets. That is not allowed, as it may cause unbounded latency.
Deep copy all zerocopy copy buffers, ref-counted or not, in this path.
The exact locations to modify were chosen by exhaustively searching
through all code that might modify skb_frag references and/or the
the SKBTX_DEV_ZEROCOPY tx_flags bit.
The changes err on the safe side, in two ways.
(1) legacy ubuf_info paths virtio and tap are not modified. They keep
a 1:1 ubuf_info to sk_buff relationship. Calls to skb_orphan_frags
still call skb_copy_ubufs and thus copy frags in this case.
(2) not all copies deep in the stack are addressed yet. skb_shift,
skb_split and skb_try_coalesce can be refined to avoid copying.
These are not in the hot path and this patch is hairy enough as
is, so that is left for future refinement.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-03 16:29:41 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-19 20:29:13 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2011-07-29 18:37:31 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2007-03-27 18:55:52 -03:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:52 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2012-04-05 11:35:15 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-07-25 21:43:18 -07:00
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2012-04-05 11:35:15 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2011-07-06 12:22:12 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-01-19 17:03:56 -08:00
|
|
|
|
|
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
2012-04-23 23:06:11 -04:00
|
|
|
|
|
|
|
|
|
2013-01-11 14:46:37 +00:00
|
|
|
|
2009-01-19 17:03:56 -08:00
|
|
|
|
net: use a per task frag allocator
We currently use a per socket order-0 page cache for tcp_sendmsg()
operations.
This page is used to build fragments for skbs.
Its done to increase probability of coalescing small write() into
single segments in skbs still in write queue (not yet sent)
But it wastes a lot of memory for applications handling many mostly
idle sockets, since each socket holds one page in sk->sk_sndmsg_page
Its also quite inefficient to build TSO 64KB packets, because we need
about 16 pages per skb on arches where PAGE_SIZE = 4096, so we hit
page allocator more than wanted.
This patch adds a per task frag allocator and uses bigger pages,
if available. An automatic fallback is done in case of memory pressure.
(up to 32768 bytes per frag, thats order-3 pages on x86)
This increases TCP stream performance by 20% on loopback device,
but also benefits on other network devices, since 8x less frags are
mapped on transmit and unmapped on tx completion. Alexander Duyck
mentioned a probable performance win on systems with IOMMU enabled.
Its possible some SG enabled hardware cant cope with bigger fragments,
but their ndo_start_xmit() should already handle this, splitting a
fragment in sub fragments, since some arches have PAGE_SIZE=65536
Successfully tested on various ethernet devices.
(ixgbe, igb, bnx2x, tg3, mellanox mlx4)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Ben Hutchings <bhutchings@solarflare.com>
Cc: Vijay Subramanian <subramanian.vijay@gmail.com>
Cc: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Vijay Subramanian <subramanian.vijay@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-09-23 23:04:42 +00:00
|
|
|
|
2009-01-19 17:03:56 -08:00
|
|
|
|
net: use a per task frag allocator
We currently use a per socket order-0 page cache for tcp_sendmsg()
operations.
This page is used to build fragments for skbs.
Its done to increase probability of coalescing small write() into
single segments in skbs still in write queue (not yet sent)
But it wastes a lot of memory for applications handling many mostly
idle sockets, since each socket holds one page in sk->sk_sndmsg_page
Its also quite inefficient to build TSO 64KB packets, because we need
about 16 pages per skb on arches where PAGE_SIZE = 4096, so we hit
page allocator more than wanted.
This patch adds a per task frag allocator and uses bigger pages,
if available. An automatic fallback is done in case of memory pressure.
(up to 32768 bytes per frag, thats order-3 pages on x86)
This increases TCP stream performance by 20% on loopback device,
but also benefits on other network devices, since 8x less frags are
mapped on transmit and unmapped on tx completion. Alexander Duyck
mentioned a probable performance win on systems with IOMMU enabled.
Its possible some SG enabled hardware cant cope with bigger fragments,
but their ndo_start_xmit() should already handle this, splitting a
fragment in sub fragments, since some arches have PAGE_SIZE=65536
Successfully tested on various ethernet devices.
(ixgbe, igb, bnx2x, tg3, mellanox mlx4)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Ben Hutchings <bhutchings@solarflare.com>
Cc: Vijay Subramanian <subramanian.vijay@gmail.com>
Cc: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Vijay Subramanian <subramanian.vijay@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-09-23 23:04:42 +00:00
|
|
|
|
|
|
|
|
|
2009-02-01 00:41:42 -08:00
|
|
|
|
net: use a per task frag allocator
We currently use a per socket order-0 page cache for tcp_sendmsg()
operations.
This page is used to build fragments for skbs.
Its done to increase probability of coalescing small write() into
single segments in skbs still in write queue (not yet sent)
But it wastes a lot of memory for applications handling many mostly
idle sockets, since each socket holds one page in sk->sk_sndmsg_page
Its also quite inefficient to build TSO 64KB packets, because we need
about 16 pages per skb on arches where PAGE_SIZE = 4096, so we hit
page allocator more than wanted.
This patch adds a per task frag allocator and uses bigger pages,
if available. An automatic fallback is done in case of memory pressure.
(up to 32768 bytes per frag, thats order-3 pages on x86)
This increases TCP stream performance by 20% on loopback device,
but also benefits on other network devices, since 8x less frags are
mapped on transmit and unmapped on tx completion. Alexander Duyck
mentioned a probable performance win on systems with IOMMU enabled.
Its possible some SG enabled hardware cant cope with bigger fragments,
but their ndo_start_xmit() should already handle this, splitting a
fragment in sub fragments, since some arches have PAGE_SIZE=65536
Successfully tested on various ethernet devices.
(ixgbe, igb, bnx2x, tg3, mellanox mlx4)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Ben Hutchings <bhutchings@solarflare.com>
Cc: Vijay Subramanian <subramanian.vijay@gmail.com>
Cc: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Vijay Subramanian <subramanian.vijay@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-09-23 23:04:42 +00:00
|
|
|
|
2009-02-01 00:41:42 -08:00
|
|
|
|
net: use a per task frag allocator
We currently use a per socket order-0 page cache for tcp_sendmsg()
operations.
This page is used to build fragments for skbs.
Its done to increase probability of coalescing small write() into
single segments in skbs still in write queue (not yet sent)
But it wastes a lot of memory for applications handling many mostly
idle sockets, since each socket holds one page in sk->sk_sndmsg_page
Its also quite inefficient to build TSO 64KB packets, because we need
about 16 pages per skb on arches where PAGE_SIZE = 4096, so we hit
page allocator more than wanted.
This patch adds a per task frag allocator and uses bigger pages,
if available. An automatic fallback is done in case of memory pressure.
(up to 32768 bytes per frag, thats order-3 pages on x86)
This increases TCP stream performance by 20% on loopback device,
but also benefits on other network devices, since 8x less frags are
mapped on transmit and unmapped on tx completion. Alexander Duyck
mentioned a probable performance win on systems with IOMMU enabled.
Its possible some SG enabled hardware cant cope with bigger fragments,
but their ndo_start_xmit() should already handle this, splitting a
fragment in sub fragments, since some arches have PAGE_SIZE=65536
Successfully tested on various ethernet devices.
(ixgbe, igb, bnx2x, tg3, mellanox mlx4)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Ben Hutchings <bhutchings@solarflare.com>
Cc: Vijay Subramanian <subramanian.vijay@gmail.com>
Cc: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Vijay Subramanian <subramanian.vijay@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-09-23 23:04:42 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-02-01 00:41:42 -08:00
|
|
|
|
net: use a per task frag allocator
We currently use a per socket order-0 page cache for tcp_sendmsg()
operations.
This page is used to build fragments for skbs.
Its done to increase probability of coalescing small write() into
single segments in skbs still in write queue (not yet sent)
But it wastes a lot of memory for applications handling many mostly
idle sockets, since each socket holds one page in sk->sk_sndmsg_page
Its also quite inefficient to build TSO 64KB packets, because we need
about 16 pages per skb on arches where PAGE_SIZE = 4096, so we hit
page allocator more than wanted.
This patch adds a per task frag allocator and uses bigger pages,
if available. An automatic fallback is done in case of memory pressure.
(up to 32768 bytes per frag, thats order-3 pages on x86)
This increases TCP stream performance by 20% on loopback device,
but also benefits on other network devices, since 8x less frags are
mapped on transmit and unmapped on tx completion. Alexander Duyck
mentioned a probable performance win on systems with IOMMU enabled.
Its possible some SG enabled hardware cant cope with bigger fragments,
but their ndo_start_xmit() should already handle this, splitting a
fragment in sub fragments, since some arches have PAGE_SIZE=65536
Successfully tested on various ethernet devices.
(ixgbe, igb, bnx2x, tg3, mellanox mlx4)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Ben Hutchings <bhutchings@solarflare.com>
Cc: Vijay Subramanian <subramanian.vijay@gmail.com>
Cc: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Vijay Subramanian <subramanian.vijay@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-09-23 23:04:42 +00:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
|
|
|
|
|
2012-04-22 12:26:16 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-02 15:38:07 +02:00
|
|
|
|
|
|
|
|
|
2012-04-23 23:06:11 -04:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
2012-04-22 12:26:16 +00:00
|
|
|
|
2012-04-23 23:06:11 -04:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
2009-01-19 17:03:56 -08:00
|
|
|
|
2013-01-11 14:46:37 +00:00
|
|
|
|
2009-01-19 17:03:56 -08:00
|
|
|
|
2012-04-23 23:06:11 -04:00
|
|
|
|
2012-04-22 12:26:16 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-04-23 23:06:11 -04:00
|
|
|
|
2012-04-22 12:26:16 +00:00
|
|
|
|
|
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
2009-02-01 00:41:42 -08:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
|
|
|
|
|
2009-01-19 17:03:56 -08:00
|
|
|
|
2012-04-23 23:06:11 -04:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
|
|
|
|
|
2012-04-23 23:06:11 -04:00
|
|
|
|
|
|
|
|
|
2013-01-11 14:46:37 +00:00
|
|
|
|
2012-04-23 23:35:04 -04:00
|
|
|
|
2025-07-02 15:38:07 +02:00
|
|
|
|
2008-07-15 00:49:11 -07:00
|
|
|
|
|
|
|
|
|
2012-04-23 23:06:11 -04:00
|
|
|
|
2008-07-15 00:49:11 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-04-23 23:06:11 -04:00
|
|
|
|
2008-06-27 17:27:21 -07:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
2008-07-15 00:49:11 -07:00
|
|
|
|
2013-01-05 21:31:18 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-07-15 00:49:11 -07:00
|
|
|
|
2013-01-11 14:46:37 +00:00
|
|
|
|
|
|
|
|
|
2008-07-15 00:49:11 -07:00
|
|
|
|
2025-07-02 15:38:07 +02:00
|
|
|
|
2013-01-11 14:46:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-19 10:15:51 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-07-15 00:49:11 -07:00
|
|
|
|
2012-04-23 23:06:11 -04:00
|
|
|
|
2008-07-15 00:49:11 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-04-23 23:06:11 -04:00
|
|
|
|
2008-07-15 00:49:11 -07:00
|
|
|
|
|
|
|
|
|
2012-04-23 23:06:11 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-07-15 00:49:11 -07:00
|
|
|
|
2016-03-07 14:11:04 -08:00
|
|
|
|
2025-07-02 15:38:07 +02:00
|
|
|
|
2008-07-15 00:49:11 -07:00
|
|
|
|
2012-04-27 02:10:03 +00:00
|
|
|
|
2012-05-02 18:18:42 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
2008-07-15 00:49:11 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-01-11 14:46:37 +00:00
|
|
|
|
2012-05-03 01:09:42 +00:00
|
|
|
|
2025-07-02 15:38:07 +02:00
|
|
|
|
2012-04-23 23:06:11 -04:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:52 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:51 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-08-22 23:44:58 +00:00
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
2025-07-02 15:38:07 +02:00
|
|
|
|
2012-04-23 23:06:11 -04:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
|
|
|
|
|
2016-03-07 14:11:04 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-04-23 23:06:11 -04:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-03-07 14:11:04 -08:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
2015-05-21 17:00:00 +02:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
2016-09-17 21:02:10 -04:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
2012-04-22 12:26:16 +00:00
|
|
|
|
|
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-06-12 15:24:40 +02:00
|
|
|
|
2014-01-22 19:36:57 +01:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
|
|
|
|
|
2010-05-20 10:43:18 +02:00
|
|
|
|
|
|
|
|
|
2016-03-07 14:11:04 -08:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
2015-05-21 17:00:00 +02:00
|
|
|
|
2016-09-17 21:02:10 -04:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
2010-05-20 10:43:18 +02:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
2015-05-21 17:00:01 +02:00
|
|
|
|
2007-11-06 23:30:13 -08:00
|
|
|
|
2023-06-23 23:54:59 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-03-30 19:32:24 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-06-23 23:54:59 +01:00
|
|
|
|
2021-03-30 19:32:24 -07:00
|
|
|
|
|
|
|
|
|
2023-06-23 23:54:59 +01:00
|
|
|
|
2021-03-30 19:32:24 -07:00
|
|
|
|
2025-04-15 13:17:29 +02:00
|
|
|
|
2017-07-28 16:22:42 -07:00
|
|
|
|
2025-07-08 13:40:53 +08:00
|
|
|
|
2017-07-28 16:22:42 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-15 22:31:34 -07:00
|
|
|
|
2017-07-28 16:22:42 -07:00
|
|
|
|
2025-04-15 13:17:29 +02:00
|
|
|
|
2025-07-08 13:40:53 +08:00
|
|
|
|
|
|
|
|
|
2017-07-28 16:22:42 -07:00
|
|
|
|
2023-06-23 23:54:59 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-07-28 16:22:42 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-22 20:08:26 -07:00
|
|
|
|
2017-07-28 16:22:42 -07:00
|
|
|
|
|
|
|
|
|
2019-07-22 20:08:26 -07:00
|
|
|
|
2017-07-28 16:22:42 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-22 20:08:26 -07:00
|
|
|
|
2017-07-28 16:22:42 -07:00
|
|
|
|
|
|
|
|
|
2023-06-23 23:54:59 +01:00
|
|
|
|
|
|
|
|
|
2025-04-15 13:17:29 +02:00
|
|
|
|
|
|
|
|
|
2023-06-23 23:54:59 +01:00
|
|
|
|
|
|
|
|
|
2025-07-08 13:40:53 +08:00
|
|
|
|
|
|
|
|
|
2023-06-23 23:54:59 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-07-28 16:22:42 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-03-30 19:32:24 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-15 13:17:29 +02:00
|
|
|
|
2021-03-30 19:32:24 -07:00
|
|
|
|
2017-07-28 16:22:42 -07:00
|
|
|
|
|
|
|
|
|
2025-04-15 13:17:29 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-03-30 19:32:24 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-15 13:17:29 +02:00
|
|
|
|
2021-03-30 19:32:24 -07:00
|
|
|
|
|
|
|
|
|
2005-04-19 22:30:14 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-20 16:40:01 -07:00
|
|
|
|
2005-04-19 22:30:14 -07:00
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
|
|
|
|
|
2005-04-19 22:30:14 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2005-04-19 22:30:14 -07:00
|
|
|
|
|
|
|
|
|
2007-03-31 11:55:19 -03:00
|
|
|
|
2005-04-19 22:30:14 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:52 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-19 22:30:14 -07:00
|
|
|
|
|
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2005-04-19 22:30:14 -07:00
|
|
|
|
2008-07-25 21:43:18 -07:00
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
2005-04-19 22:30:14 -07:00
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
|
|
|
|
|
2005-04-19 22:30:14 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-19 22:30:14 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2005-04-19 22:30:14 -07:00
|
|
|
|
|
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
|
|
|
|
|
2005-04-19 22:30:14 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
2005-04-19 22:30:14 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-19 22:30:14 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
2005-04-19 22:30:14 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2025-05-19 10:50:08 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
|
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-05-19 10:50:08 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:52 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2012-04-05 11:35:15 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-07-25 21:43:18 -07:00
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2012-04-05 11:35:15 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
|
|
|
|
|
2006-11-14 21:36:14 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
|
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
|
|
|
|
|
2025-05-19 10:50:08 -07:00
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
2025-05-19 10:50:08 -07:00
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-05-19 10:50:08 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-01-08 22:24:28 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-11-14 21:37:33 -08:00
|
|
|
|
2020-07-10 20:07:10 -04:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
|
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2020-07-10 20:07:10 -04:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-07-11 00:12:07 -04:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:52 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-07-25 21:43:18 -07:00
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-11-14 21:36:34 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
|
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-07-11 00:12:07 -04:00
|
|
|
|
2017-07-31 08:15:47 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-07-10 20:07:10 -04:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-01-08 22:24:28 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
net: add skb_crc32c()
Add skb_crc32c(), which calculates the CRC32C of a sk_buff. It will
replace __skb_checksum(), which unnecessarily supports arbitrary
checksums. Compared to __skb_checksum(), skb_crc32c():
- Uses the correct type for CRC32C values (u32, not __wsum).
- Does not require the caller to provide a skb_checksum_ops struct.
- Is faster because it does not use indirect calls and does not use
the very slow crc32c_combine().
According to commit 2817a336d4d5 ("net: skb_checksum: allow custom
update/combine for walking skb") which added __skb_checksum(), the
original motivation for the abstraction layer was to avoid code
duplication for CRC32C and other checksums in the future. However:
- No additional checksums showed up after CRC32C. __skb_checksum()
is only used with the "regular" net checksum and CRC32C.
- Indirect calls are expensive. Commit 2544af0344ba ("net: avoid
indirect calls in L4 checksum calculation") worked around this
using the INDIRECT_CALL_1 macro. But that only avoided the indirect
call for the net checksum, and at the cost of an extra branch.
- The checksums use different types (__wsum and u32), causing casts
to be needed.
- It made the checksums of fragments be combined (rather than
chained) for both checksums, despite this being highly
counterproductive for CRC32C due to how slow crc32c_combine() is.
This can clearly be seen in commit 4c2f24549644 ("sctp: linearize
early if it's not GSO") which tried to work around this performance
bug. With a dedicated function for each checksum, we can instead
just use the proper strategy for each checksum.
As shown by the following tables, the new function skb_crc32c() is
faster than __skb_checksum(), with the improvement varying greatly from
5% to 2500% depending on the case. The largest improvements come from
fragmented packets, mainly due to eliminating the inefficient
crc32c_combine(). But linear packets are improved too, especially
shorter ones, mainly due to eliminating indirect calls. These
benchmarks were done on AMD Zen 5. On that CPU, Linux uses IBRS instead
of retpoline; an even greater improvement might be seen with retpoline:
Linear sk_buffs
Length in bytes __skb_checksum cycles skb_crc32c cycles
=============== ===================== =================
64 43 18
256 94 77
1420 204 161
16384 1735 1642
Nonlinear sk_buffs (even split between head and one fragment)
Length in bytes __skb_checksum cycles skb_crc32c cycles
=============== ===================== =================
64 579 22
256 829 77
1420 1506 194
16384 4365 1682
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Link: https://patch.msgid.link/20250519175012.36581-3-ebiggers@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2025-05-19 10:50:04 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-08 14:05:42 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-26 09:31:26 -08:00
|
|
|
|
2018-11-08 14:05:42 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-12 14:47:18 -08:00
|
|
|
|
2018-11-08 14:05:42 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-26 09:31:26 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-08 14:05:42 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-26 09:31:26 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-08 14:05:42 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-12 14:47:18 -08:00
|
|
|
|
2018-11-08 14:05:42 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-12-13 15:22:17 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-07-15 16:59:00 -07:00
|
|
|
|
2013-12-13 15:22:17 +01:00
|
|
|
|
2021-07-15 16:59:00 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-12-13 15:22:17 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-01-29 01:05:28 +09:00
|
|
|
|
2013-12-13 15:22:17 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-26 22:37:45 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-12-13 15:22:17 +01:00
|
|
|
|
2014-03-26 22:37:45 +00:00
|
|
|
|
|
|
|
|
|
2013-12-13 15:22:17 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-26 22:37:45 +00:00
|
|
|
|
2013-12-13 15:22:17 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-26 22:37:45 +00:00
|
|
|
|
|
|
|
|
|
2013-12-13 15:22:17 +01:00
|
|
|
|
|
|
|
|
|
2014-03-26 22:37:45 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-12-13 15:22:17 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-02-14 14:34:03 -08:00
|
|
|
|
|
|
|
|
|
2013-12-13 15:22:17 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-22 18:09:03 +02:00
|
|
|
|
2013-12-13 15:22:17 +01:00
|
|
|
|
2014-03-26 22:37:45 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
sock: enable MSG_ZEROCOPY
Prepare the datapath for refcounted ubuf_info. Clone ubuf_info with
skb_zerocopy_clone() wherever needed due to skb split, merge, resize
or clone.
Split skb_orphan_frags into two variants. The split, merge, .. paths
support reference counted zerocopy buffers, so do not do a deep copy.
Add skb_orphan_frags_rx for paths that may loop packets to receive
sockets. That is not allowed, as it may cause unbounded latency.
Deep copy all zerocopy copy buffers, ref-counted or not, in this path.
The exact locations to modify were chosen by exhaustively searching
through all code that might modify skb_frag references and/or the
the SKBTX_DEV_ZEROCOPY tx_flags bit.
The changes err on the safe side, in two ways.
(1) legacy ubuf_info paths virtio and tap are not modified. They keep
a 1:1 ubuf_info to sk_buff relationship. Calls to skb_orphan_frags
still call skb_copy_ubufs and thus copy frags in this case.
(2) not all copies deep in the stack are addressed yet. skb_shift,
skb_split and skb_try_coalesce can be refined to avoid copying.
These are not in the hot path and this patch is hairy enough as
is, so that is left for future refinement.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-03 16:29:41 -04:00
|
|
|
|
2014-03-26 22:37:45 +00:00
|
|
|
|
2013-12-13 15:22:17 +01:00
|
|
|
|
2019-07-22 20:08:26 -07:00
|
|
|
|
|
|
|
|
|
2013-12-13 15:22:17 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-22 20:08:26 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-12-13 15:22:17 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-26 22:37:45 +00:00
|
|
|
|
|
|
|
|
|
2013-12-13 15:22:17 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2006-11-14 21:24:49 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2006-08-29 16:44:56 -07:00
|
|
|
|
2010-12-14 15:24:08 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-01-08 22:24:28 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-03-27 18:55:52 -03:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-07-10 20:07:10 -04:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-08-29 16:44:56 -07:00
|
|
|
|
2006-11-20 18:07:29 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2006-11-14 21:24:49 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2023-08-18 09:40:39 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2023-08-18 09:40:39 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-18 09:40:39 +00:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2023-10-03 18:19:20 +00:00
|
|
|
|
|
|
|
|
|
2023-08-18 09:40:39 +00:00
|
|
|
|
2023-10-03 18:19:20 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2023-08-18 09:40:39 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
tcp: use an RB tree for ooo receive queue
Over the years, TCP BDP has increased by several orders of magnitude,
and some people are considering to reach the 2 Gbytes limit.
Even with current window scale limit of 14, ~1 Gbytes maps to ~740,000
MSS.
In presence of packet losses (or reorders), TCP stores incoming packets
into an out of order queue, and number of skbs sitting there waiting for
the missing packets to be received can be in the 10^5 range.
Most packets are appended to the tail of this queue, and when
packets can finally be transferred to receive queue, we scan the queue
from its head.
However, in presence of heavy losses, we might have to find an arbitrary
point in this queue, involving a linear scan for every incoming packet,
throwing away cpu caches.
This patch converts it to a RB tree, to get bounded latencies.
Yaogong wrote a preliminary patch about 2 years ago.
Eric did the rebase, added ofo_last_skb cache, polishing and tests.
Tested with network dropping between 1 and 10 % packets, with good
success (about 30 % increase of throughput in stress tests)
Next step would be to also use an RB tree for the write queue at sender
side ;)
Signed-off-by: Yaogong Wang <wygivan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Acked-By: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-07 14:49:28 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-08-02 23:34:38 +00:00
|
|
|
|
tcp: use an RB tree for ooo receive queue
Over the years, TCP BDP has increased by several orders of magnitude,
and some people are considering to reach the 2 Gbytes limit.
Even with current window scale limit of 14, ~1 Gbytes maps to ~740,000
MSS.
In presence of packet losses (or reorders), TCP stores incoming packets
into an out of order queue, and number of skbs sitting there waiting for
the missing packets to be received can be in the 10^5 range.
Most packets are appended to the tail of this queue, and when
packets can finally be transferred to receive queue, we scan the queue
from its head.
However, in presence of heavy losses, we might have to find an arbitrary
point in this queue, involving a linear scan for every incoming packet,
throwing away cpu caches.
This patch converts it to a RB tree, to get bounded latencies.
Yaogong wrote a preliminary patch about 2 years ago.
Eric did the rebase, added ofo_last_skb cache, polishing and tests.
Tested with network dropping between 1 and 10 % packets, with good
success (about 30 % increase of throughput in stress tests)
Next step would be to also use an RB tree for the write queue at sender
side ;)
Signed-off-by: Yaogong Wang <wygivan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Acked-By: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-07 14:49:28 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-08-02 23:34:38 +00:00
|
|
|
|
tcp: use an RB tree for ooo receive queue
Over the years, TCP BDP has increased by several orders of magnitude,
and some people are considering to reach the 2 Gbytes limit.
Even with current window scale limit of 14, ~1 Gbytes maps to ~740,000
MSS.
In presence of packet losses (or reorders), TCP stores incoming packets
into an out of order queue, and number of skbs sitting there waiting for
the missing packets to be received can be in the 10^5 range.
Most packets are appended to the tail of this queue, and when
packets can finally be transferred to receive queue, we scan the queue
from its head.
However, in presence of heavy losses, we might have to find an arbitrary
point in this queue, involving a linear scan for every incoming packet,
throwing away cpu caches.
This patch converts it to a RB tree, to get bounded latencies.
Yaogong wrote a preliminary patch about 2 years ago.
Eric did the rebase, added ofo_last_skb cache, polishing and tests.
Tested with network dropping between 1 and 10 % packets, with good
success (about 30 % increase of throughput in stress tests)
Next step would be to also use an RB tree for the write queue at sender
side ;)
Signed-off-by: Yaogong Wang <wygivan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Acked-By: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-07 14:49:28 -07:00
|
|
|
|
2017-09-23 12:39:12 -07:00
|
|
|
|
2018-08-02 23:34:38 +00:00
|
|
|
|
tcp: use an RB tree for ooo receive queue
Over the years, TCP BDP has increased by several orders of magnitude,
and some people are considering to reach the 2 Gbytes limit.
Even with current window scale limit of 14, ~1 Gbytes maps to ~740,000
MSS.
In presence of packet losses (or reorders), TCP stores incoming packets
into an out of order queue, and number of skbs sitting there waiting for
the missing packets to be received can be in the 10^5 range.
Most packets are appended to the tail of this queue, and when
packets can finally be transferred to receive queue, we scan the queue
from its head.
However, in presence of heavy losses, we might have to find an arbitrary
point in this queue, involving a linear scan for every incoming packet,
throwing away cpu caches.
This patch converts it to a RB tree, to get bounded latencies.
Yaogong wrote a preliminary patch about 2 years ago.
Eric did the rebase, added ofo_last_skb cache, polishing and tests.
Tested with network dropping between 1 and 10 % packets, with good
success (about 30 % increase of throughput in stress tests)
Next step would be to also use an RB tree for the write queue at sender
side ;)
Signed-off-by: Yaogong Wang <wygivan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Acked-By: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-07 14:49:28 -07:00
|
|
|
|
2017-09-23 12:39:12 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-08-02 23:34:38 +00:00
|
|
|
|
tcp: use an RB tree for ooo receive queue
Over the years, TCP BDP has increased by several orders of magnitude,
and some people are considering to reach the 2 Gbytes limit.
Even with current window scale limit of 14, ~1 Gbytes maps to ~740,000
MSS.
In presence of packet losses (or reorders), TCP stores incoming packets
into an out of order queue, and number of skbs sitting there waiting for
the missing packets to be received can be in the 10^5 range.
Most packets are appended to the tail of this queue, and when
packets can finally be transferred to receive queue, we scan the queue
from its head.
However, in presence of heavy losses, we might have to find an arbitrary
point in this queue, involving a linear scan for every incoming packet,
throwing away cpu caches.
This patch converts it to a RB tree, to get bounded latencies.
Yaogong wrote a preliminary patch about 2 years ago.
Eric did the rebase, added ofo_last_skb cache, polishing and tests.
Tested with network dropping between 1 and 10 % packets, with good
success (about 30 % increase of throughput in stress tests)
Next step would be to also use an RB tree for the write queue at sender
side ;)
Signed-off-by: Yaogong Wang <wygivan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Acked-By: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-07 14:49:28 -07:00
|
|
|
|
2017-09-23 12:39:12 -07:00
|
|
|
|
2018-08-02 23:34:38 +00:00
|
|
|
|
tcp: use an RB tree for ooo receive queue
Over the years, TCP BDP has increased by several orders of magnitude,
and some people are considering to reach the 2 Gbytes limit.
Even with current window scale limit of 14, ~1 Gbytes maps to ~740,000
MSS.
In presence of packet losses (or reorders), TCP stores incoming packets
into an out of order queue, and number of skbs sitting there waiting for
the missing packets to be received can be in the 10^5 range.
Most packets are appended to the tail of this queue, and when
packets can finally be transferred to receive queue, we scan the queue
from its head.
However, in presence of heavy losses, we might have to find an arbitrary
point in this queue, involving a linear scan for every incoming packet,
throwing away cpu caches.
This patch converts it to a RB tree, to get bounded latencies.
Yaogong wrote a preliminary patch about 2 years ago.
Eric did the rebase, added ofo_last_skb cache, polishing and tests.
Tested with network dropping between 1 and 10 % packets, with good
success (about 30 % increase of throughput in stress tests)
Next step would be to also use an RB tree for the write queue at sender
side ;)
Signed-off-by: Yaogong Wang <wygivan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Acked-By: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-09-07 14:49:28 -07:00
|
|
|
|
|
|
|
|
|
2023-08-18 17:41:45 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-08-09 19:25:21 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-08-09 19:25:21 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2005-08-09 19:25:21 -07:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2005-08-09 19:25:21 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2005-08-09 19:25:21 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2005-08-09 19:25:21 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2005-08-09 19:25:21 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-08-09 19:25:21 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-08-09 19:25:21 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-08-09 19:25:21 -07:00
|
|
|
|
2008-04-14 00:05:09 -07:00
|
|
|
|
2005-08-09 19:25:21 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-03-27 18:55:52 -03:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:52 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-19 20:29:13 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-08-22 23:44:58 +00:00
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:52 +00:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-11-02 22:58:44 -04:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-07-12 21:52:31 +01:00
|
|
|
|
|
|
|
|
|
2021-11-02 22:58:44 -04:00
|
|
|
|
sock: enable MSG_ZEROCOPY
Prepare the datapath for refcounted ubuf_info. Clone ubuf_info with
skb_zerocopy_clone() wherever needed due to skb split, merge, resize
or clone.
Split skb_orphan_frags into two variants. The split, merge, .. paths
support reference counted zerocopy buffers, so do not do a deep copy.
Add skb_orphan_frags_rx for paths that may loop packets to receive
sockets. That is not allowed, as it may cause unbounded latency.
Deep copy all zerocopy copy buffers, ref-counted or not, in this path.
The exact locations to modify were chosen by exhaustively searching
through all code that might modify skb_frag references and/or the
the SKBTX_DEV_ZEROCOPY tx_flags bit.
The changes err on the safe side, in two ways.
(1) legacy ubuf_info paths virtio and tap are not modified. They keep
a 1:1 ubuf_info to sk_buff relationship. Calls to skb_orphan_frags
still call skb_copy_ubufs and thus copy frags in this case.
(2) not all copies deep in the stack are addressed yet. skb_shift,
skb_split and skb_try_coalesce can be refined to avoid copying.
These are not in the hot path and this patch is hairy enough as
is, so that is left for future refinement.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-03 16:29:41 -04:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2008-11-25 13:57:01 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
|
|
|
|
|
2021-11-01 17:45:55 -07:00
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-11-21 01:47:11 +00:00
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-22 20:08:26 -07:00
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
|
|
|
|
|
tcp: enhance tcp_collapse_retrans() with skb_shift()
In commit 2331ccc5b323 ("tcp: enhance tcp collapsing"),
we made a first step allowing copying right skb to left skb head.
Since all skbs in socket write queue are headless (but possibly the very
first one), this strategy often does not work.
This patch extends tcp_collapse_retrans() to perform frag shifting,
thanks to skb_shift() helper.
This helper needs to not BUG on non headless skbs, as callers are ok
with that.
Tested:
Following packetdrill test now passes :
0.000 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
+0 bind(3, ..., ...) = 0
+0 listen(3, 1) = 0
+0 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 8>
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 8>
+.100 < . 1:1(0) ack 1 win 257
+0 accept(3, ..., ...) = 4
+0 setsockopt(4, SOL_TCP, TCP_NODELAY, [1], 4) = 0
+0 write(4, ..., 200) = 200
+0 > P. 1:201(200) ack 1
+.001 write(4, ..., 200) = 200
+0 > P. 201:401(200) ack 1
+.001 write(4, ..., 200) = 200
+0 > P. 401:601(200) ack 1
+.001 write(4, ..., 200) = 200
+0 > P. 601:801(200) ack 1
+.001 write(4, ..., 200) = 200
+0 > P. 801:1001(200) ack 1
+.001 write(4, ..., 100) = 100
+0 > P. 1001:1101(100) ack 1
+.001 write(4, ..., 100) = 100
+0 > P. 1101:1201(100) ack 1
+.001 write(4, ..., 100) = 100
+0 > P. 1201:1301(100) ack 1
+.001 write(4, ..., 100) = 100
+0 > P. 1301:1401(100) ack 1
+.099 < . 1:1(0) ack 201 win 257
+.001 < . 1:1(0) ack 201 win 257 <nop,nop,sack 1001:1401>
+0 > P. 201:1001(800) ack 1
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-11-15 12:51:50 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
sock: enable MSG_ZEROCOPY
Prepare the datapath for refcounted ubuf_info. Clone ubuf_info with
skb_zerocopy_clone() wherever needed due to skb split, merge, resize
or clone.
Split skb_orphan_frags into two variants. The split, merge, .. paths
support reference counted zerocopy buffers, so do not do a deep copy.
Add skb_orphan_frags_rx for paths that may loop packets to receive
sockets. That is not allowed, as it may cause unbounded latency.
Deep copy all zerocopy copy buffers, ref-counted or not, in this path.
The exact locations to modify were chosen by exhaustively searching
through all code that might modify skb_frag references and/or the
the SKBTX_DEV_ZEROCOPY tx_flags bit.
The changes err on the safe side, in two ways.
(1) legacy ubuf_info paths virtio and tap are not modified. They keep
a 1:1 ubuf_info to sk_buff relationship. Calls to skb_orphan_frags
still call skb_copy_ubufs and thus copy frags in this case.
(2) not all copies deep in the stack are addressed yet. skb_shift,
skb_split and skb_try_coalesce can be refined to avoid copying.
These are not in the hot path and this patch is hairy enough as
is, so that is left for future refinement.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-03 16:29:41 -04:00
|
|
|
|
|
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
2024-05-30 16:36:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-08-15 22:09:42 +08:00
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-11-25 13:57:01 -08:00
|
|
|
|
|
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
|
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
|
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-05-02 10:54:22 -07:00
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
|
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
2021-06-07 21:02:38 +02:00
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: skbuff: propagate shared-frag marker through frag-transfer helpers
Two frag-transfer helpers (__pskb_copy_fclone() and skb_shift()) fail
to propagate the SKBFL_SHARED_FRAG bit in skb_shinfo()->flags when
moving frags from source to destination. __pskb_copy_fclone() defers
the rest of the shinfo metadata to skb_copy_header() after copying
frag descriptors, but that helper only carries over gso_{size,segs,
type} and never touches skb_shinfo()->flags; skb_shift() moves frag
descriptors directly and leaves flags untouched. As a result, the
destination skb keeps a reference to the same externally-owned or
page-cache-backed pages while reporting skb_has_shared_frag() as
false.
The mismatch is harmful in any in-place writer that uses
skb_has_shared_frag() to decide whether shared pages must be detoured
through skb_cow_data(). ESP input is one such writer (esp4.c,
esp6.c), and a single nft 'dup to <local>' rule -- or any other
nf_dup_ipv4() / xt_TEE caller -- is enough to land a pskb_copy()'d
skb in esp_input() with the marker stripped, letting an unprivileged
user write into the page cache of a root-owned read-only file via
authencesn-ESN stray writes.
Set SKBFL_SHARED_FRAG on the destination whenever frag descriptors
were actually moved from the source. skb_copy() and skb_copy_expand()
share skb_copy_header() too but linearize all paged data into freshly
allocated head storage and emerge with nr_frags == 0, so
skb_has_shared_frag() returns false on its own; they need no change.
The same omission exists in skb_gro_receive() and skb_gro_receive_list().
The former moves the incoming skb's frag descriptors into the
accumulator's last sub-skb via two paths (a direct frag-move loop and
the head_frag + memcpy path); the latter chains the incoming skb whole
onto p's frag_list. Downstream skb_segment() reads only
skb_shinfo(p)->flags, and skb_segment_list() reuses each sub-skb's
shinfo as the nskb -- both p and lp must carry the marker.
The same omission also exists in tcp_clone_payload(), which builds an
MTU probe skb by moving frag descriptors from skbs on sk_write_queue
into a freshly allocated nskb. The helper falls into the same family
and warrants the same fix for consistency; no TCP TX-side in-place
writer is currently known to reach a user page through this gap, but
a future consumer depending on the marker would regress silently.
The same omission exists in skb_segment(): the per-iteration flag
merge takes only head_skb's flag, and the inner switch that rebinds
frag_skb to list_skb on head_skb-frags exhaustion does not fold the
new frag_skb's flag into nskb. Fold frag_skb's flag at both sites
so segments drawing frags from frag_list members carry the marker.
Fixes: cef401de7be8 ("net: fix possible wrong checksum generation")
Fixes: f4c50a4034e6 ("xfrm: esp: avoid in-place decrypt on shared skb frags")
Suggested-by: Sabrina Dubroca <sd@queasysnail.net>
Suggested-by: Sultan Alsawaf <sultan@kerneltoast.com>
Suggested-by: Ben Hutchings <ben@decadent.org.uk>
Suggested-by: Lin Ma <malin89@huawei.com>
Suggested-by: Jingguo Tan <tanjingguo@huawei.com>
Suggested-by: Aaron Esau <aaron1esau@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Hyunwoo Kim <imv4bel@gmail.com>
Tested-by: Rajat Gupta <rajat.gupta@oss.qualcomm.com>
Link: https://patch.msgid.link/ageeJfJHwgzmKXbh@v4bel
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-05-16 07:28:53 +09:00
|
|
|
|
|
|
|
|
|
2022-06-22 18:09:03 +02:00
|
|
|
|
|
|
|
|
|
tcp: Try to restore large SKBs while SACK processing
During SACK processing, most of the benefits of TSO are eaten by
the SACK blocks that one-by-one fragment SKBs to MSS sized chunks.
Then we're in problems when cleanup work for them has to be done
when a large cumulative ACK comes. Try to return back to pre-split
state already while more and more SACK info gets discovered by
combining newly discovered SACK areas with the previous skb if
that's SACKed as well.
This approach has a number of benefits:
1) The processing overhead is spread more equally over the RTT
2) Write queue has less skbs to process (affect everything
which has to walk in the queue past the sacked areas)
3) Write queue is consistent whole the time, so no other parts
of TCP has to be aware of this (this was not the case with
some other approach that was, well, quite intrusive all
around).
4) Clean_rtx_queue can release most of the pages using single
put_page instead of previous PAGE_SIZE/mss+1 calls
In case a hole is fully filled by the new SACK block, we attempt
to combine the next skb too which allows construction of skbs
that are even larger than what tso split them to and it handles
hole per on every nth patterns that often occur during slow start
overshoot pretty nicely. Though this to be really useful also
a retransmission would have to get lost since cumulative ACKs
advance one hole at a time in the most typical case.
TODO: handle upwards only merging. That should be rather easy
when segment is fully sacked but I'm leaving that as future
work item (it won't make very large difference anyway since
this current approach already covers quite a lot of normal
cases).
I was earlier thinking of some sophisticated way of tracking
timestamps of the first and the last segment but later on
realized that it won't be that necessary at all to store the
timestamp of the last segment. The cases that can occur are
basically either:
1) ambiguous => no sensible measurement can be taken anyway
2) non-ambiguous is due to reordering => having the timestamp
of the last segment there is just skewing things more off
than does some good since the ack got triggered by one of
the holes (besides some substle issues that would make
determining right hole/skb even harder problem). Anyway,
it has nothing to do with this change then.
I choose to route some abnormal looking cases with goto noop,
some could be handled differently (eg., by stopping the
walking at that skb but again). In general, they either
shouldn't happen at all or are rare enough to make no difference
in practice.
In theory this change (as whole) could cause some macroscale
regression (global) because of cache misses that are taken over
the round-trip time but it gets very likely better because of much
less (local) cache misses per other write queue walkers and the
big recovery clearing cumulative ack.
Worth to note that these benefits would be very easy to get also
without TSO/GSO being on as long as the data is in pages so that
we can merge them. Currently I won't let that happen because
DSACK splitting at fragment that would mess up pcounts due to
sk_can_gso in tcp_set_skb_tso_segs. Once DSACKs fragments gets
avoided, we have some conditions that can be made less strict.
TODO: I will probably have to convert the excessive pointer
passing to struct sacktag_state... :-)
My testing revealed that considerable amount of skbs couldn't
be shifted because they were cloned (most likely still awaiting
tx reclaim)...
[The rest is considering future work instead since I got
repeatably EFAULT to tcpdump's recvfrom when I added
pskb_expand_head to deal with clones, so I separated that
into another, later patch]
...To counter that, I gave up on the fifth advantage:
5) When growing previous SACK block, less allocs for new skbs
are done, basically a new alloc is needed only when new hole
is detected and when the previous skb runs out of frags space
...which now only happens of if reclaim is fast enough to dispose
the clone before the SACK block comes in (the window is RTT long),
otherwise we'll have to alloc some.
With clones being handled I got these numbers (will be somewhat
worse without that), taken with fine-grained mibs:
TCPSackShifted 398
TCPSackMerged 877
TCPSackShiftFallback 320
TCPSACKCOLLAPSEFALLBACKGSO 0
TCPSACKCOLLAPSEFALLBACKSKBBITS 0
TCPSACKCOLLAPSEFALLBACKSKBDATA 0
TCPSACKCOLLAPSEFALLBACKBELOW 0
TCPSACKCOLLAPSEFALLBACKFIRST 1
TCPSACKCOLLAPSEFALLBACKPREVBITS 318
TCPSACKCOLLAPSEFALLBACKMSS 1
TCPSACKCOLLAPSEFALLBACKNOHEAD 0
TCPSACKCOLLAPSEFALLBACKSHIFT 0
TCPSACKCOLLAPSENOOPSEQ 0
TCPSACKCOLLAPSENOOPSMALLPCOUNT 0
TCPSACKCOLLAPSENOOPSMALLLEN 0
TCPSACKCOLLAPSEHOLE 12
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-11-24 21:20:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-01-09 17:18:33 -05:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-11-07 14:18:26 +01:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
2013-11-07 14:18:26 +01:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-11-07 14:18:26 +01:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-03-30 22:57:33 -03:00
|
|
|
|
2014-09-04 23:44:36 +09:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
2008-02-13 15:03:25 -08:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-06-23 23:33:48 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
2013-06-23 23:33:48 -07:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
2009-01-29 16:07:52 -08:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
2009-05-18 21:43:27 -07:00
|
|
|
|
2009-01-29 16:07:52 -08:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:52 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-01-09 17:18:33 -05:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
2021-01-09 17:18:33 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
2021-01-09 17:18:33 -05:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
2021-01-09 17:18:33 -05:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-04-05 11:35:15 +02:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-01-09 17:18:33 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
2007-06-23 23:11:52 -07:00
|
|
|
|
2012-04-05 11:35:15 +02:00
|
|
|
|
2007-06-23 23:11:52 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2010-08-23 00:13:46 -07:00
|
|
|
|
2009-01-29 16:12:42 -08:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
2009-01-29 16:12:42 -08:00
|
|
|
|
|
|
|
|
|
2009-01-29 16:07:52 -08:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-04-05 11:35:15 +02:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-06-23 20:59:51 -07:00
|
|
|
|
2024-08-15 13:21:14 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-06-23 21:00:17 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-02-22 11:46:35 +01:00
|
|
|
|
2005-06-23 21:00:17 -07:00
|
|
|
|
2023-10-17 11:39:06 +02:00
|
|
|
|
2015-02-22 11:46:35 +01:00
|
|
|
|
2006-06-26 00:00:57 -07:00
|
|
|
|
|
|
|
|
|
2021-03-01 15:09:44 +00:00
|
|
|
|
|
|
|
|
|
2005-06-23 21:00:17 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-02-22 11:46:35 +01:00
|
|
|
|
2005-06-23 21:00:17 -07:00
|
|
|
|
2015-02-22 11:46:35 +01:00
|
|
|
|
2023-10-17 11:39:06 +02:00
|
|
|
|
2005-06-23 21:00:17 -07:00
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2005-06-23 21:00:17 -07:00
|
|
|
|
2015-05-21 16:59:58 +02:00
|
|
|
|
2023-05-22 13:11:11 +01:00
|
|
|
|
2015-05-21 16:59:58 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-22 13:11:11 +01:00
|
|
|
|
2022-07-12 21:52:31 +01:00
|
|
|
|
2015-05-21 16:59:58 +02:00
|
|
|
|
2022-10-27 04:03:46 +00:00
|
|
|
|
2015-05-21 16:59:58 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-03-20 22:43:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-02-12 22:03:25 -08:00
|
|
|
|
2006-08-29 16:44:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-03-20 22:43:56 -08:00
|
|
|
|
networking: make skb_pull & friends return void pointers
It seems like a historic accident that these return unsigned char *,
and in many places that means casts are required, more often than not.
Make these functions return void * and remove all the casts across
the tree, adding a (u8 *) cast only where the unsigned char pointer
was used directly, all done with the following spatch:
@@
expression SKB, LEN;
typedef u8;
identifier fn = {
skb_pull,
__skb_pull,
skb_pull_inline,
__pskb_pull_tail,
__pskb_pull,
pskb_pull
};
@@
- *(fn(SKB, LEN))
+ *(u8 *)fn(SKB, LEN)
@@
expression E, SKB, LEN;
identifier fn = {
skb_pull,
__skb_pull,
skb_pull_inline,
__pskb_pull_tail,
__pskb_pull,
pskb_pull
};
type T;
@@
- E = ((T *)(fn(SKB, LEN)))
+ E = fn(SKB, LEN)
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-06-16 14:29:22 +02:00
|
|
|
|
2006-03-20 22:43:56 -08:00
|
|
|
|
2015-09-28 17:24:25 -07:00
|
|
|
|
|
|
|
|
|
2006-03-20 22:43:56 -08:00
|
|
|
|
2015-09-28 17:24:25 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-03-20 22:43:56 -08:00
|
|
|
|
2006-03-20 22:47:55 -08:00
|
|
|
|
|
|
|
|
|
2018-03-21 16:31:03 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-11 09:12:12 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-21 16:31:03 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-01-25 11:26:44 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-01-08 11:28:38 +09:00
|
|
|
|
net: fix wrong network header length
When clatd starts with ebpf offloaing, and NETIF_F_GRO_FRAGLIST is enable,
several skbs are gathered in skb_shinfo(skb)->frag_list. The first skb's
ipv6 header will be changed to ipv4 after bpf_skb_proto_6_to_4,
network_header\transport_header\mac_header have been updated as ipv4 acts,
but other skbs in frag_list didnot update anything, just ipv6 packets.
udp_queue_rcv_skb will call skb_segment_list to traverse other skbs in
frag_list and make sure right udp payload is delivered to user space.
Unfortunately, other skbs in frag_list who are still ipv6 packets are
updated like the first skb and will have wrong transport header length.
e.g.before bpf_skb_proto_6_to_4,the first skb and other skbs in frag_list
has the same network_header(24)& transport_header(64), after
bpf_skb_proto_6_to_4, ipv6 protocol has been changed to ipv4, the first
skb's network_header is 44,transport_header is 64, other skbs in frag_list
didnot change.After skb_segment_list, the other skbs in frag_list has
different network_header(24) and transport_header(44), so there will be 20
bytes different from original,that is difference between ipv6 header and
ipv4 header. Just change transport_header to be the same with original.
Actually, there are two solutions to fix it, one is traversing all skbs
and changing every skb header in bpf_skb_proto_6_to_4, the other is
modifying frag_list skb's header in skb_segment_list. Considering
efficiency, adopt the second one--- when the first skb and other skbs in
frag_list has different network_header length, restore them to make sure
right udp payload is delivered to user space.
Signed-off-by: Lina Wang <lina.wang@mediatek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2022-05-05 13:48:49 +08:00
|
|
|
|
2020-01-25 11:26:44 +01:00
|
|
|
|
net: fix memory leak in skb_segment_list for GRO packets
When skb_segment_list() is called during packet forwarding, it handles
packets that were aggregated by the GRO engine.
Historically, the segmentation logic in skb_segment_list assumes that
individual segments are split from a parent SKB and may need to carry
their own socket memory accounting. Accordingly, the code transfers
truesize from the parent to the newly created segments.
Prior to commit ed4cccef64c1 ("gro: fix ownership transfer"), this
truesize subtraction in skb_segment_list() was valid because fragments
still carry a reference to the original socket.
However, commit ed4cccef64c1 ("gro: fix ownership transfer") changed
this behavior by ensuring that fraglist entries are explicitly
orphaned (skb->sk = NULL) to prevent illegal orphaning later in the
stack. This change meant that the entire socket memory charge remained
with the head SKB, but the corresponding accounting logic in
skb_segment_list() was never updated.
As a result, the current code unconditionally adds each fragment's
truesize to delta_truesize and subtracts it from the parent SKB. Since
the fragments are no longer charged to the socket, this subtraction
results in an effective under-count of memory when the head is freed.
This causes sk_wmem_alloc to remain non-zero, preventing socket
destruction and leading to a persistent memory leak.
The leak can be observed via KMEMLEAK when tearing down the networking
environment:
unreferenced object 0xffff8881e6eb9100 (size 2048):
comm "ping", pid 6720, jiffies 4295492526
backtrace:
kmem_cache_alloc_noprof+0x5c6/0x800
sk_prot_alloc+0x5b/0x220
sk_alloc+0x35/0xa00
inet6_create.part.0+0x303/0x10d0
__sock_create+0x248/0x640
__sys_socket+0x11b/0x1d0
Since skb_segment_list() is exclusively used for SKB_GSO_FRAGLIST
packets constructed by GRO, the truesize adjustment is removed.
The call to skb_release_head_state() must be preserved. As documented in
commit cf673ed0e057 ("net: fix fraglist segmentation reference count
leak"), it is still required to correctly drop references to SKB
extensions that may be overwritten during __copy_skb_header().
Fixes: ed4cccef64c1 ("gro: fix ownership transfer")
Signed-off-by: Mohammad Heib <mheib@redhat.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20260104213101.352887-1-mheib@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2026-01-04 23:31:01 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-01-25 11:26:44 +01:00
|
|
|
|
|
|
|
|
|
2023-07-07 10:11:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-01-25 11:26:44 +01:00
|
|
|
|
|
|
|
|
|
2023-01-30 12:51:48 -08:00
|
|
|
|
2020-01-25 11:26:44 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: fix memory leak in skb_segment_list for GRO packets
When skb_segment_list() is called during packet forwarding, it handles
packets that were aggregated by the GRO engine.
Historically, the segmentation logic in skb_segment_list assumes that
individual segments are split from a parent SKB and may need to carry
their own socket memory accounting. Accordingly, the code transfers
truesize from the parent to the newly created segments.
Prior to commit ed4cccef64c1 ("gro: fix ownership transfer"), this
truesize subtraction in skb_segment_list() was valid because fragments
still carry a reference to the original socket.
However, commit ed4cccef64c1 ("gro: fix ownership transfer") changed
this behavior by ensuring that fraglist entries are explicitly
orphaned (skb->sk = NULL) to prevent illegal orphaning later in the
stack. This change meant that the entire socket memory charge remained
with the head SKB, but the corresponding accounting logic in
skb_segment_list() was never updated.
As a result, the current code unconditionally adds each fragment's
truesize to delta_truesize and subtracts it from the parent SKB. Since
the fragments are no longer charged to the socket, this subtraction
results in an effective under-count of memory when the head is freed.
This causes sk_wmem_alloc to remain non-zero, preventing socket
destruction and leading to a persistent memory leak.
The leak can be observed via KMEMLEAK when tearing down the networking
environment:
unreferenced object 0xffff8881e6eb9100 (size 2048):
comm "ping", pid 6720, jiffies 4295492526
backtrace:
kmem_cache_alloc_noprof+0x5c6/0x800
sk_prot_alloc+0x5b/0x220
sk_alloc+0x35/0xa00
inet6_create.part.0+0x303/0x10d0
__sock_create+0x248/0x640
__sys_socket+0x11b/0x1d0
Since skb_segment_list() is exclusively used for SKB_GSO_FRAGLIST
packets constructed by GRO, the truesize adjustment is removed.
The call to skb_release_head_state() must be preserved. As documented in
commit cf673ed0e057 ("net: fix fraglist segmentation reference count
leak"), it is still required to correctly drop references to SKB
extensions that may be overwritten during __copy_skb_header().
Fixes: ed4cccef64c1 ("gro: fix ownership transfer")
Signed-off-by: Mohammad Heib <mheib@redhat.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20260104213101.352887-1-mheib@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2026-01-04 23:31:01 +02:00
|
|
|
|
|
|
|
|
|
2021-01-08 11:28:38 +09:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-01-25 11:26:44 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-01-08 11:28:38 +09:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-01-25 11:26:44 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-03-30 18:51:29 +02:00
|
|
|
|
net: fix wrong network header length
When clatd starts with ebpf offloaing, and NETIF_F_GRO_FRAGLIST is enable,
several skbs are gathered in skb_shinfo(skb)->frag_list. The first skb's
ipv6 header will be changed to ipv4 after bpf_skb_proto_6_to_4,
network_header\transport_header\mac_header have been updated as ipv4 acts,
but other skbs in frag_list didnot update anything, just ipv6 packets.
udp_queue_rcv_skb will call skb_segment_list to traverse other skbs in
frag_list and make sure right udp payload is delivered to user space.
Unfortunately, other skbs in frag_list who are still ipv6 packets are
updated like the first skb and will have wrong transport header length.
e.g.before bpf_skb_proto_6_to_4,the first skb and other skbs in frag_list
has the same network_header(24)& transport_header(64), after
bpf_skb_proto_6_to_4, ipv6 protocol has been changed to ipv4, the first
skb's network_header is 44,transport_header is 64, other skbs in frag_list
didnot change.After skb_segment_list, the other skbs in frag_list has
different network_header(24) and transport_header(44), so there will be 20
bytes different from original,that is difference between ipv6 header and
ipv4 header. Just change transport_header to be the same with original.
Actually, there are two solutions to fix it, one is traversing all skbs
and changing every skb header in bpf_skb_proto_6_to_4, the other is
modifying frag_list skb's header in skb_segment_list. Considering
efficiency, adopt the second one--- when the first skb and other skbs in
frag_list has different network_header length, restore them to make sure
right udp payload is delivered to user space.
Signed-off-by: Lina Wang <lina.wang@mediatek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2022-05-05 13:48:49 +08:00
|
|
|
|
2021-09-02 23:56:23 +01:00
|
|
|
|
2020-01-25 11:26:44 +01:00
|
|
|
|
|
|
|
|
|
net: fix wrong network header length
When clatd starts with ebpf offloaing, and NETIF_F_GRO_FRAGLIST is enable,
several skbs are gathered in skb_shinfo(skb)->frag_list. The first skb's
ipv6 header will be changed to ipv4 after bpf_skb_proto_6_to_4,
network_header\transport_header\mac_header have been updated as ipv4 acts,
but other skbs in frag_list didnot update anything, just ipv6 packets.
udp_queue_rcv_skb will call skb_segment_list to traverse other skbs in
frag_list and make sure right udp payload is delivered to user space.
Unfortunately, other skbs in frag_list who are still ipv6 packets are
updated like the first skb and will have wrong transport header length.
e.g.before bpf_skb_proto_6_to_4,the first skb and other skbs in frag_list
has the same network_header(24)& transport_header(64), after
bpf_skb_proto_6_to_4, ipv6 protocol has been changed to ipv4, the first
skb's network_header is 44,transport_header is 64, other skbs in frag_list
didnot change.After skb_segment_list, the other skbs in frag_list has
different network_header(24) and transport_header(44), so there will be 20
bytes different from original,that is difference between ipv6 header and
ipv4 header. Just change transport_header to be the same with original.
Actually, there are two solutions to fix it, one is traversing all skbs
and changing every skb header in bpf_skb_proto_6_to_4, the other is
modifying frag_list skb's header in skb_segment_list. Considering
efficiency, adopt the second one--- when the first skb and other skbs in
frag_list has different network_header length, restore them to make sure
right udp payload is delivered to user space.
Signed-off-by: Lina Wang <lina.wang@mediatek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2022-05-05 13:48:49 +08:00
|
|
|
|
2020-01-25 11:26:44 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-30 12:51:48 -08:00
|
|
|
|
2020-01-25 11:26:44 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
2014-03-10 18:29:19 +02:00
|
|
|
|
2006-06-27 13:22:38 -07:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
2008-04-13 21:52:48 -07:00
|
|
|
|
|
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
2014-03-10 18:29:19 +02:00
|
|
|
|
|
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-10 19:27:59 +02:00
|
|
|
|
2014-03-10 18:29:19 +02:00
|
|
|
|
|
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
2014-03-10 18:29:19 +02:00
|
|
|
|
2016-04-10 21:45:03 -04:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
2016-04-10 21:45:03 -04:00
|
|
|
|
2023-08-31 02:17:02 -06:00
|
|
|
|
|
|
|
|
|
2013-03-07 09:28:01 +00:00
|
|
|
|
2016-05-02 09:38:18 -07:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
2023-08-31 02:17:02 -06:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
net: gso: fix panic on frag_list with mixed head alloc types
Since commit 3dcbdb134f32 ("net: gso: Fix skb_segment splat when
splitting gso_size mangled skb having linear-headed frag_list"), it is
allowed to change gso_size of a GRO packet. However, that commit assumes
that "checking the first list_skb member suffices; i.e if either of the
list_skb members have non head_frag head, then the first one has too".
It turns out this assumption does not hold. We've seen BUG_ON being hit
in skb_segment when skbs on the frag_list had differing head_frag with
the vmxnet3 driver. This happens because __netdev_alloc_skb and
__napi_alloc_skb can return a skb that is page backed or kmalloced
depending on the requested size. As the result, the last small skb in
the GRO packet can be kmalloced.
There are three different locations where this can be fixed:
(1) We could check head_frag in GRO and not allow GROing skbs with
different head_frag. However, that would lead to performance
regression on normal forward paths with unmodified gso_size, where
!head_frag in the last packet is not a problem.
(2) Set a flag in bpf_skb_net_grow and bpf_skb_net_shrink indicating
that NETIF_F_SG is undesirable. That would need to eat a bit in
sk_buff. Furthermore, that flag can be unset when all skbs on the
frag_list are page backed. To retain good performance,
bpf_skb_net_grow/shrink would have to walk the frag_list.
(3) Walk the frag_list in skb_segment when determining whether
NETIF_F_SG should be cleared. This of course slows things down.
This patch implements (3). To limit the performance impact in
skb_segment, the list is walked only for skbs with SKB_GSO_DODGY set
that have gso_size changed. Normal paths thus will not hit it.
We could check only the last skb but since we need to walk the whole
list anyway, let's stay on the safe side.
Fixes: 3dcbdb134f32 ("net: gso: Fix skb_segment splat when splitting gso_size mangled skb having linear-headed frag_list")
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://lore.kernel.org/r/e04426a6a91baf4d1081e1b478c82b5de25fdf21.1667407944.git.jbenc@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-11-02 17:53:25 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: gso: Fix skb_segment splat when splitting gso_size mangled skb having linear-headed frag_list
Historically, support for frag_list packets entering skb_segment() was
limited to frag_list members terminating on exact same gso_size
boundaries. This is verified with a BUG_ON since commit 89319d3801d1
("net: Add frag_list support to skb_segment"), quote:
As such we require all frag_list members terminate on exact MSS
boundaries. This is checked using BUG_ON.
As there should only be one producer in the kernel of such packets,
namely GRO, this requirement should not be difficult to maintain.
However, since commit 6578171a7ff0 ("bpf: add bpf_skb_change_proto helper"),
the "exact MSS boundaries" assumption no longer holds:
An eBPF program using bpf_skb_change_proto() DOES modify 'gso_size', but
leaves the frag_list members as originally merged by GRO with the
original 'gso_size'. Example of such programs are bpf-based NAT46 or
NAT64.
This lead to a kernel BUG_ON for flows involving:
- GRO generating a frag_list skb
- bpf program performing bpf_skb_change_proto() or bpf_skb_adjust_room()
- skb_segment() of the skb
See example BUG_ON reports in [0].
In commit 13acc94eff12 ("net: permit skb_segment on head_frag frag_list skb"),
skb_segment() was modified to support the "gso_size mangling" case of
a frag_list GRO'ed skb, but *only* for frag_list members having
head_frag==true (having a page-fragment head).
Alas, GRO packets having frag_list members with a linear kmalloced head
(head_frag==false) still hit the BUG_ON.
This commit adds support to skb_segment() for a 'head_skb' packet having
a frag_list whose members are *non* head_frag, with gso_size mangled, by
disabling SG and thus falling-back to copying the data from the given
'head_skb' into the generated segmented skbs - as suggested by Willem de
Bruijn [1].
Since this approach involves the penalty of skb_copy_and_csum_bits()
when building the segments, care was taken in order to enable this
solution only when required:
- untrusted gso_size, by testing SKB_GSO_DODGY is set
(SKB_GSO_DODGY is set by any gso_size mangling functions in
net/core/filter.c)
- the frag_list is non empty, its item is a non head_frag, *and* the
headlen of the given 'head_skb' does not match the gso_size.
[0]
https://lore.kernel.org/netdev/20190826170724.25ff616f@pixies/
https://lore.kernel.org/netdev/9265b93f-253d-6b8c-f2b8-4b54eff1835c@fb.com/
[1]
https://lore.kernel.org/netdev/CA+FuTSfVsgNDi7c=GUU8nMg2hWxF2SjCNLXetHeVPdnxAW5K-w@mail.gmail.com/
Fixes: 6578171a7ff0 ("bpf: add bpf_skb_change_proto helper")
Suggested-by: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Alexander Duyck <alexander.duyck@gmail.com>
Signed-off-by: Shmulik Ladkani <shmulik.ladkani@gmail.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2019-09-06 12:23:50 +03:00
|
|
|
|
|
|
|
|
|
2014-06-08 23:48:54 -07:00
|
|
|
|
2020-08-01 17:36:05 +08:00
|
|
|
|
2013-03-07 09:28:01 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-05-02 09:38:18 -07:00
|
|
|
|
2016-02-05 15:28:26 -08:00
|
|
|
|
2014-06-04 17:20:02 -07:00
|
|
|
|
2016-09-19 12:58:47 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-04-19 21:26:07 +03:00
|
|
|
|
2016-09-19 12:58:47 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-04-19 21:26:07 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-09-19 12:58:47 +02:00
|
|
|
|
2017-04-19 21:26:07 +03:00
|
|
|
|
2016-09-19 12:58:47 +02:00
|
|
|
|
2017-04-19 21:26:07 +03:00
|
|
|
|
|
|
|
|
|
2017-04-08 02:07:08 +03:00
|
|
|
|
2016-09-19 12:58:47 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-04-19 21:26:07 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-09-19 12:58:47 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-12 16:46:21 +00:00
|
|
|
|
2016-09-19 12:58:47 +02:00
|
|
|
|
2023-12-12 16:46:21 +00:00
|
|
|
|
2016-05-02 09:38:12 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-04-10 21:45:03 -04:00
|
|
|
|
|
|
|
|
|
2016-09-19 12:58:47 +02:00
|
|
|
|
2014-03-10 18:29:19 +02:00
|
|
|
|
|
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
2023-08-31 02:17:02 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
2014-03-10 18:29:04 +02:00
|
|
|
|
2006-10-29 15:59:41 -08:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
2016-06-02 15:05:40 -03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
2014-03-10 18:29:19 +02:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
2021-01-15 17:36:38 +08:00
|
|
|
|
2014-03-10 19:27:59 +02:00
|
|
|
|
|
|
|
|
|
2013-11-21 11:10:04 -08:00
|
|
|
|
2023-08-31 02:17:02 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-11-21 11:10:04 -08:00
|
|
|
|
2014-03-10 19:27:59 +02:00
|
|
|
|
|
|
|
|
|
2014-03-10 19:28:08 +02:00
|
|
|
|
2014-03-10 19:27:59 +02:00
|
|
|
|
2013-11-21 11:10:04 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-10 18:29:14 +02:00
|
|
|
|
2013-11-21 11:10:04 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-10 18:29:14 +02:00
|
|
|
|
2013-11-21 11:10:04 -08:00
|
|
|
|
2008-12-15 23:26:06 -08:00
|
|
|
|
2014-03-10 19:27:59 +02:00
|
|
|
|
2008-12-15 23:26:06 -08:00
|
|
|
|
2013-11-21 11:10:04 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-05-04 14:26:56 +00:00
|
|
|
|
2008-12-15 23:26:06 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-05-04 14:26:56 +00:00
|
|
|
|
2008-12-15 23:26:06 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-01-19 17:56:56 +01:00
|
|
|
|
|
|
|
|
|
2021-01-15 17:36:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-07-31 16:44:19 -07:00
|
|
|
|
2014-03-10 18:29:19 +02:00
|
|
|
|
2012-07-31 16:44:19 -07:00
|
|
|
|
2008-12-15 23:26:06 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-10 18:29:19 +02:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
2013-10-19 11:42:54 -07:00
|
|
|
|
2014-07-31 10:33:06 -04:00
|
|
|
|
2013-02-14 14:02:41 +00:00
|
|
|
|
2014-03-10 18:29:19 +02:00
|
|
|
|
2013-02-14 14:02:41 +00:00
|
|
|
|
|
|
|
|
|
2008-12-15 23:26:06 -08:00
|
|
|
|
2013-11-21 11:10:04 -08:00
|
|
|
|
2013-05-19 15:46:49 +00:00
|
|
|
|
2008-12-15 23:26:06 -08:00
|
|
|
|
2016-02-05 15:27:43 -08:00
|
|
|
|
2020-03-17 14:08:38 +05:30
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-07-10 20:07:10 -04:00
|
|
|
|
2020-03-17 14:08:38 +05:30
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-08-22 22:44:11 -07:00
|
|
|
|
|
|
|
|
|
2020-03-17 14:08:38 +05:30
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-10 18:29:04 +02:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
2014-03-10 18:29:19 +02:00
|
|
|
|
2007-03-27 18:55:52 -03:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
net: skbuff: propagate shared-frag marker through frag-transfer helpers
Two frag-transfer helpers (__pskb_copy_fclone() and skb_shift()) fail
to propagate the SKBFL_SHARED_FRAG bit in skb_shinfo()->flags when
moving frags from source to destination. __pskb_copy_fclone() defers
the rest of the shinfo metadata to skb_copy_header() after copying
frag descriptors, but that helper only carries over gso_{size,segs,
type} and never touches skb_shinfo()->flags; skb_shift() moves frag
descriptors directly and leaves flags untouched. As a result, the
destination skb keeps a reference to the same externally-owned or
page-cache-backed pages while reporting skb_has_shared_frag() as
false.
The mismatch is harmful in any in-place writer that uses
skb_has_shared_frag() to decide whether shared pages must be detoured
through skb_cow_data(). ESP input is one such writer (esp4.c,
esp6.c), and a single nft 'dup to <local>' rule -- or any other
nf_dup_ipv4() / xt_TEE caller -- is enough to land a pskb_copy()'d
skb in esp_input() with the marker stripped, letting an unprivileged
user write into the page cache of a root-owned read-only file via
authencesn-ESN stray writes.
Set SKBFL_SHARED_FRAG on the destination whenever frag descriptors
were actually moved from the source. skb_copy() and skb_copy_expand()
share skb_copy_header() too but linearize all paged data into freshly
allocated head storage and emerge with nr_frags == 0, so
skb_has_shared_frag() returns false on its own; they need no change.
The same omission exists in skb_gro_receive() and skb_gro_receive_list().
The former moves the incoming skb's frag descriptors into the
accumulator's last sub-skb via two paths (a direct frag-move loop and
the head_frag + memcpy path); the latter chains the incoming skb whole
onto p's frag_list. Downstream skb_segment() reads only
skb_shinfo(p)->flags, and skb_segment_list() reuses each sub-skb's
shinfo as the nskb -- both p and lp must carry the marker.
The same omission also exists in tcp_clone_payload(), which builds an
MTU probe skb by moving frag descriptors from skbs on sk_write_queue
into a freshly allocated nskb. The helper falls into the same family
and warrants the same fix for consistency; no TCP TX-side in-place
writer is currently known to reach a user page through this gap, but
a future consumer depending on the marker would regress silently.
The same omission exists in skb_segment(): the per-iteration flag
merge takes only head_skb's flag, and the inner switch that rebinds
frag_skb to list_skb on head_skb-frags exhaustion does not fold the
new frag_skb's flag into nskb. Fold frag_skb's flag at both sites
so segments drawing frags from frag_list members carry the marker.
Fixes: cef401de7be8 ("net: fix possible wrong checksum generation")
Fixes: f4c50a4034e6 ("xfrm: esp: avoid in-place decrypt on shared skb frags")
Suggested-by: Sabrina Dubroca <sd@queasysnail.net>
Suggested-by: Sultan Alsawaf <sultan@kerneltoast.com>
Suggested-by: Ben Hutchings <ben@decadent.org.uk>
Suggested-by: Lin Ma <malin89@huawei.com>
Suggested-by: Jingguo Tan <tanjingguo@huawei.com>
Suggested-by: Aaron Esau <aaron1esau@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Hyunwoo Kim <imv4bel@gmail.com>
Tested-by: Rajat Gupta <rajat.gupta@oss.qualcomm.com>
Link: https://patch.msgid.link/ageeJfJHwgzmKXbh@v4bel
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-05-16 07:28:53 +09:00
|
|
|
|
|
|
|
|
|
2021-01-06 14:18:38 -08:00
|
|
|
|
2013-01-25 20:34:37 +00:00
|
|
|
|
2023-08-31 02:17:02 -06:00
|
|
|
|
2017-12-22 19:00:17 -05:00
|
|
|
|
|
|
|
|
|
2013-11-21 11:10:04 -08:00
|
|
|
|
|
|
|
|
|
2023-08-31 02:17:02 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-11-21 11:10:04 -08:00
|
|
|
|
2014-03-10 19:27:59 +02:00
|
|
|
|
|
|
|
|
|
2014-03-10 19:28:08 +02:00
|
|
|
|
net: skbuff: propagate shared-frag marker through frag-transfer helpers
Two frag-transfer helpers (__pskb_copy_fclone() and skb_shift()) fail
to propagate the SKBFL_SHARED_FRAG bit in skb_shinfo()->flags when
moving frags from source to destination. __pskb_copy_fclone() defers
the rest of the shinfo metadata to skb_copy_header() after copying
frag descriptors, but that helper only carries over gso_{size,segs,
type} and never touches skb_shinfo()->flags; skb_shift() moves frag
descriptors directly and leaves flags untouched. As a result, the
destination skb keeps a reference to the same externally-owned or
page-cache-backed pages while reporting skb_has_shared_frag() as
false.
The mismatch is harmful in any in-place writer that uses
skb_has_shared_frag() to decide whether shared pages must be detoured
through skb_cow_data(). ESP input is one such writer (esp4.c,
esp6.c), and a single nft 'dup to <local>' rule -- or any other
nf_dup_ipv4() / xt_TEE caller -- is enough to land a pskb_copy()'d
skb in esp_input() with the marker stripped, letting an unprivileged
user write into the page cache of a root-owned read-only file via
authencesn-ESN stray writes.
Set SKBFL_SHARED_FRAG on the destination whenever frag descriptors
were actually moved from the source. skb_copy() and skb_copy_expand()
share skb_copy_header() too but linearize all paged data into freshly
allocated head storage and emerge with nr_frags == 0, so
skb_has_shared_frag() returns false on its own; they need no change.
The same omission exists in skb_gro_receive() and skb_gro_receive_list().
The former moves the incoming skb's frag descriptors into the
accumulator's last sub-skb via two paths (a direct frag-move loop and
the head_frag + memcpy path); the latter chains the incoming skb whole
onto p's frag_list. Downstream skb_segment() reads only
skb_shinfo(p)->flags, and skb_segment_list() reuses each sub-skb's
shinfo as the nskb -- both p and lp must carry the marker.
The same omission also exists in tcp_clone_payload(), which builds an
MTU probe skb by moving frag descriptors from skbs on sk_write_queue
into a freshly allocated nskb. The helper falls into the same family
and warrants the same fix for consistency; no TCP TX-side in-place
writer is currently known to reach a user page through this gap, but
a future consumer depending on the marker would regress silently.
The same omission exists in skb_segment(): the per-iteration flag
merge takes only head_skb's flag, and the inner switch that rebinds
frag_skb to list_skb on head_skb-frags exhaustion does not fold the
new frag_skb's flag into nskb. Fold frag_skb's flag at both sites
so segments drawing frags from frag_list members carry the marker.
Fixes: cef401de7be8 ("net: fix possible wrong checksum generation")
Fixes: f4c50a4034e6 ("xfrm: esp: avoid in-place decrypt on shared skb frags")
Suggested-by: Sabrina Dubroca <sd@queasysnail.net>
Suggested-by: Sultan Alsawaf <sultan@kerneltoast.com>
Suggested-by: Ben Hutchings <ben@decadent.org.uk>
Suggested-by: Lin Ma <malin89@huawei.com>
Suggested-by: Jingguo Tan <tanjingguo@huawei.com>
Suggested-by: Aaron Esau <aaron1esau@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Hyunwoo Kim <imv4bel@gmail.com>
Tested-by: Rajat Gupta <rajat.gupta@oss.qualcomm.com>
Link: https://patch.msgid.link/ageeJfJHwgzmKXbh@v4bel
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-05-16 07:28:53 +09:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-21 16:31:03 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-11-21 11:10:04 -08:00
|
|
|
|
2018-03-21 16:31:03 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-22 19:00:17 -05:00
|
|
|
|
2014-03-10 19:27:59 +02:00
|
|
|
|
2013-11-21 11:10:04 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-07-19 16:04:38 -07:00
|
|
|
|
2013-11-21 11:10:04 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-21 16:31:03 -07:00
|
|
|
|
2024-05-02 10:54:22 -07:00
|
|
|
|
2014-03-10 18:29:04 +02:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
2014-03-10 18:29:04 +02:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
2008-12-15 23:26:06 -08:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-10 18:29:14 +02:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
2014-03-10 18:29:04 +02:00
|
|
|
|
2008-12-15 23:26:06 -08:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
2014-03-10 18:29:04 +02:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
2008-12-15 23:26:06 -08:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-03-07 09:28:01 +00:00
|
|
|
|
2013-05-19 15:46:49 +00:00
|
|
|
|
2016-02-05 15:27:43 -08:00
|
|
|
|
2018-07-19 16:04:38 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-02-05 15:27:43 -08:00
|
|
|
|
|
|
|
|
|
2016-02-05 15:27:37 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-06-04 17:20:02 -07:00
|
|
|
|
2016-02-05 15:27:37 -08:00
|
|
|
|
2013-03-07 09:28:01 +00:00
|
|
|
|
2014-03-10 18:29:19 +02:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
2014-10-03 20:59:19 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-10-27 10:30:51 -07:00
|
|
|
|
2016-04-10 21:45:03 -04:00
|
|
|
|
2016-09-19 12:58:47 +02:00
|
|
|
|
2016-04-10 21:45:03 -04:00
|
|
|
|
2016-09-19 12:58:47 +02:00
|
|
|
|
2016-04-10 21:45:03 -04:00
|
|
|
|
|
|
|
|
|
2016-09-19 12:58:47 +02:00
|
|
|
|
2016-04-10 21:45:03 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-09-19 12:58:47 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-04-10 21:45:03 -04:00
|
|
|
|
|
|
|
|
|
2014-10-27 10:30:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-12-20 14:29:08 -08:00
|
|
|
|
2006-06-22 03:02:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-18 17:15:27 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: openvswitch: Set OvS recirc_id from tc chain index
Offloaded OvS datapath rules are translated one to one to tc rules,
for example the following simplified OvS rule:
recirc_id(0),in_port(dev1),eth_type(0x0800),ct_state(-trk) actions:ct(),recirc(2)
Will be translated to the following tc rule:
$ tc filter add dev dev1 ingress \
prio 1 chain 0 proto ip \
flower tcp ct_state -trk \
action ct pipe \
action goto chain 2
Received packets will first travel though tc, and if they aren't stolen
by it, like in the above rule, they will continue to OvS datapath.
Since we already did some actions (action ct in this case) which might
modify the packets, and updated action stats, we would like to continue
the proccessing with the correct recirc_id in OvS (here recirc_id(2))
where we left off.
To support this, introduce a new skb extension for tc, which
will be used for translating tc chain to ovs recirc_id to
handle these miss cases. Last tc chain index will be set
by tc goto chain action and read by OvS datapath.
Signed-off-by: Paul Blakey <paulb@mellanox.com>
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2019-09-04 16:56:37 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-01-09 07:59:19 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-29 11:01:44 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-16 17:09:30 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-02-01 15:33:17 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
2026-04-10 19:21:49 +03:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
2023-08-23 11:28:38 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
2026-04-10 19:21:50 +03:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
2026-02-05 15:44:05 +01:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-15 17:17:36 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2024-03-06 16:00:23 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2023-08-15 17:17:36 +02:00
|
|
|
|
|
|
|
|
|
2018-02-07 17:44:38 -08:00
|
|
|
|
|
|
|
|
|
2007-07-20 10:11:58 +09:00
|
|
|
|
2025-10-16 18:29:11 +00:00
|
|
|
|
|
|
|
|
|
2024-03-06 16:00:23 +00:00
|
|
|
|
2014-09-29 13:29:15 -07:00
|
|
|
|
2005-08-17 14:57:30 -07:00
|
|
|
|
2006-08-26 19:25:52 -07:00
|
|
|
|
2007-07-20 10:11:58 +09:00
|
|
|
|
2023-02-08 14:25:08 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-06 16:00:23 +00:00
|
|
|
|
2023-02-06 17:31:03 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-02-08 14:25:08 +00:00
|
|
|
|
|
|
|
|
|
2023-02-06 17:31:03 +00:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2007-10-30 21:29:29 -07:00
|
|
|
|
skbuff: return -EMSGSIZE in skb_to_sgvec to prevent overflow
This is a defense-in-depth measure in response to bugs like
4d6fa57b4dab ("macsec: avoid heap overflow in skb_to_sgvec"). There's
not only a potential overflow of sglist items, but also a stack overflow
potential, so we fix this by limiting the amount of recursion this function
is allowed to do. Not actually providing a bounded base case is a future
disaster that we can easily avoid here.
As a small matter of house keeping, we take this opportunity to move the
documentation comment over the actual function the documentation is for.
While this could be implemented by using an explicit stack of skbuffs,
when implementing this, the function complexity increased considerably,
and I don't think such complexity and bloat is actually worth it. So,
instead I built this and tested it on x86, x86_64, ARM, ARM64, and MIPS,
and measured the stack usage there. I also reverted the recent MIPS
changes that give it a separate IRQ stack, so that I could experience
some worst-case situations. I found that limiting it to 24 layers deep
yielded a good stack usage with room for safety, as well as being much
deeper than any driver actually ever creates.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: David Howells <dhowells@redhat.com>
Cc: Sabrina Dubroca <sd@queasysnail.net>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-06-04 04:16:22 +02:00
|
|
|
|
|
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
|
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
|
|
|
|
|
skbuff: return -EMSGSIZE in skb_to_sgvec to prevent overflow
This is a defense-in-depth measure in response to bugs like
4d6fa57b4dab ("macsec: avoid heap overflow in skb_to_sgvec"). There's
not only a potential overflow of sglist items, but also a stack overflow
potential, so we fix this by limiting the amount of recursion this function
is allowed to do. Not actually providing a bounded base case is a future
disaster that we can easily avoid here.
As a small matter of house keeping, we take this opportunity to move the
documentation comment over the actual function the documentation is for.
While this could be implemented by using an explicit stack of skbuffs,
when implementing this, the function complexity increased considerably,
and I don't think such complexity and bloat is actually worth it. So,
instead I built this and tested it on x86, x86_64, ARM, ARM64, and MIPS,
and measured the stack usage there. I also reverted the recent MIPS
changes that give it a separate IRQ stack, so that I could experience
some worst-case situations. I found that limiting it to 24 layers deep
yielded a good stack usage with room for safety, as well as being much
deeper than any driver actually ever creates.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: David Howells <dhowells@redhat.com>
Cc: Sabrina Dubroca <sd@queasysnail.net>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-06-04 04:16:22 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-10-24 11:20:47 +02:00
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
2008-07-25 21:43:18 -07:00
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2011-10-18 21:00:24 +00:00
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
|
|
|
|
|
skbuff: return -EMSGSIZE in skb_to_sgvec to prevent overflow
This is a defense-in-depth measure in response to bugs like
4d6fa57b4dab ("macsec: avoid heap overflow in skb_to_sgvec"). There's
not only a potential overflow of sglist items, but also a stack overflow
potential, so we fix this by limiting the amount of recursion this function
is allowed to do. Not actually providing a bounded base case is a future
disaster that we can easily avoid here.
As a small matter of house keeping, we take this opportunity to move the
documentation comment over the actual function the documentation is for.
While this could be implemented by using an explicit stack of skbuffs,
when implementing this, the function complexity increased considerably,
and I don't think such complexity and bloat is actually worth it. So,
instead I built this and tested it on x86, x86_64, ARM, ARM64, and MIPS,
and measured the stack usage there. I also reverted the recent MIPS
changes that give it a separate IRQ stack, so that I could experience
some worst-case situations. I found that limiting it to 24 layers deep
yielded a good stack usage with room for safety, as well as being much
deeper than any driver actually ever creates.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: David Howells <dhowells@redhat.com>
Cc: Sabrina Dubroca <sd@queasysnail.net>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-06-04 04:16:22 +02:00
|
|
|
|
|
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-08-22 23:44:58 +00:00
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-27 15:21:23 -07:00
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
|
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
skbuff: return -EMSGSIZE in skb_to_sgvec to prevent overflow
This is a defense-in-depth measure in response to bugs like
4d6fa57b4dab ("macsec: avoid heap overflow in skb_to_sgvec"). There's
not only a potential overflow of sglist items, but also a stack overflow
potential, so we fix this by limiting the amount of recursion this function
is allowed to do. Not actually providing a bounded base case is a future
disaster that we can easily avoid here.
As a small matter of house keeping, we take this opportunity to move the
documentation comment over the actual function the documentation is for.
While this could be implemented by using an explicit stack of skbuffs,
when implementing this, the function complexity increased considerably,
and I don't think such complexity and bloat is actually worth it. So,
instead I built this and tested it on x86, x86_64, ARM, ARM64, and MIPS,
and measured the stack usage there. I also reverted the recent MIPS
changes that give it a separate IRQ stack, so that I could experience
some worst-case situations. I found that limiting it to 24 layers deep
yielded a good stack usage with room for safety, as well as being much
deeper than any driver actually ever creates.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: David Howells <dhowells@redhat.com>
Cc: Sabrina Dubroca <sd@queasysnail.net>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-06-04 04:16:22 +02:00
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
|
|
|
|
|
skbuff: return -EMSGSIZE in skb_to_sgvec to prevent overflow
This is a defense-in-depth measure in response to bugs like
4d6fa57b4dab ("macsec: avoid heap overflow in skb_to_sgvec"). There's
not only a potential overflow of sglist items, but also a stack overflow
potential, so we fix this by limiting the amount of recursion this function
is allowed to do. Not actually providing a bounded base case is a future
disaster that we can easily avoid here.
As a small matter of house keeping, we take this opportunity to move the
documentation comment over the actual function the documentation is for.
While this could be implemented by using an explicit stack of skbuffs,
when implementing this, the function complexity increased considerably,
and I don't think such complexity and bloat is actually worth it. So,
instead I built this and tested it on x86, x86_64, ARM, ARM64, and MIPS,
and measured the stack usage there. I also reverted the recent MIPS
changes that give it a separate IRQ stack, so that I could experience
some worst-case situations. I found that limiting it to 24 layers deep
yielded a good stack usage with room for safety, as well as being much
deeper than any driver actually ever creates.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: David Howells <dhowells@redhat.com>
Cc: Sabrina Dubroca <sd@queasysnail.net>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-06-04 04:16:22 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
|
|
|
|
|
skbuff: return -EMSGSIZE in skb_to_sgvec to prevent overflow
This is a defense-in-depth measure in response to bugs like
4d6fa57b4dab ("macsec: avoid heap overflow in skb_to_sgvec"). There's
not only a potential overflow of sglist items, but also a stack overflow
potential, so we fix this by limiting the amount of recursion this function
is allowed to do. Not actually providing a bounded base case is a future
disaster that we can easily avoid here.
As a small matter of house keeping, we take this opportunity to move the
documentation comment over the actual function the documentation is for.
While this could be implemented by using an explicit stack of skbuffs,
when implementing this, the function complexity increased considerably,
and I don't think such complexity and bloat is actually worth it. So,
instead I built this and tested it on x86, x86_64, ARM, ARM64, and MIPS,
and measured the stack usage there. I also reverted the recent MIPS
changes that give it a separate IRQ stack, so that I could experience
some worst-case situations. I found that limiting it to 24 layers deep
yielded a good stack usage with room for safety, as well as being much
deeper than any driver actually ever creates.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: David Howells <dhowells@redhat.com>
Cc: Sabrina Dubroca <sd@queasysnail.net>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-06-04 04:16:22 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
2009-06-09 00:18:59 -07:00
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
skbuff: return -EMSGSIZE in skb_to_sgvec to prevent overflow
This is a defense-in-depth measure in response to bugs like
4d6fa57b4dab ("macsec: avoid heap overflow in skb_to_sgvec"). There's
not only a potential overflow of sglist items, but also a stack overflow
potential, so we fix this by limiting the amount of recursion this function
is allowed to do. Not actually providing a bounded base case is a future
disaster that we can easily avoid here.
As a small matter of house keeping, we take this opportunity to move the
documentation comment over the actual function the documentation is for.
While this could be implemented by using an explicit stack of skbuffs,
when implementing this, the function complexity increased considerably,
and I don't think such complexity and bloat is actually worth it. So,
instead I built this and tested it on x86, x86_64, ARM, ARM64, and MIPS,
and measured the stack usage there. I also reverted the recent MIPS
changes that give it a separate IRQ stack, so that I could experience
some worst-case situations. I found that limiting it to 24 layers deep
yielded a good stack usage with room for safety, as well as being much
deeper than any driver actually ever creates.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: David Howells <dhowells@redhat.com>
Cc: Sabrina Dubroca <sd@queasysnail.net>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-06-04 04:16:22 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-01-18 09:54:23 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-08-22 13:57:34 +01:00
|
|
|
|
2014-01-18 09:54:23 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
skbuff: return -EMSGSIZE in skb_to_sgvec to prevent overflow
This is a defense-in-depth measure in response to bugs like
4d6fa57b4dab ("macsec: avoid heap overflow in skb_to_sgvec"). There's
not only a potential overflow of sglist items, but also a stack overflow
potential, so we fix this by limiting the amount of recursion this function
is allowed to do. Not actually providing a bounded base case is a future
disaster that we can easily avoid here.
As a small matter of house keeping, we take this opportunity to move the
documentation comment over the actual function the documentation is for.
While this could be implemented by using an explicit stack of skbuffs,
when implementing this, the function complexity increased considerably,
and I don't think such complexity and bloat is actually worth it. So,
instead I built this and tested it on x86, x86_64, ARM, ARM64, and MIPS,
and measured the stack usage there. I also reverted the recent MIPS
changes that give it a separate IRQ stack, so that I could experience
some worst-case situations. I found that limiting it to 24 layers deep
yielded a good stack usage with room for safety, as well as being much
deeper than any driver actually ever creates.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: David Howells <dhowells@redhat.com>
Cc: Sabrina Dubroca <sd@queasysnail.net>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-06-04 04:16:22 +02:00
|
|
|
|
2014-01-18 09:54:23 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-10-30 21:29:29 -07:00
|
|
|
|
|
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-08-01 17:30:23 +08:00
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2010-08-23 00:13:46 -07:00
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2010-08-23 00:13:46 -07:00
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2010-08-23 00:13:46 -07:00
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2007-04-02 20:19:53 -07:00
|
|
|
|
2026-06-06 19:18:19 -07:00
|
|
|
|
2010-05-31 23:44:05 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-03-18 17:02:59 -04:00
|
|
|
|
|
|
|
|
|
2026-06-06 19:18:19 -07:00
|
|
|
|
|
|
|
|
|
2017-03-18 17:02:59 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2010-05-31 23:44:05 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-10-10 20:17:44 -07:00
|
|
|
|
2010-05-31 23:44:05 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-03-18 17:02:59 -04:00
|
|
|
|
2010-05-31 23:44:05 -07:00
|
|
|
|
2011-05-18 02:21:31 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2010-05-31 23:44:05 -07:00
|
|
|
|
|
|
|
|
|
2021-06-27 18:48:21 -04:00
|
|
|
|
2010-05-31 23:44:05 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-11-30 14:01:08 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-08-31 21:30:27 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-11-30 14:01:08 -05:00
|
|
|
|
|
|
|
|
|
2015-02-18 05:47:55 -08:00
|
|
|
|
2014-08-31 21:30:27 -04:00
|
|
|
|
2023-10-05 11:45:04 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-02-18 05:47:55 -08:00
|
|
|
|
2014-08-31 21:30:27 -04:00
|
|
|
|
2017-06-02 12:38:22 -04:00
|
|
|
|
2016-11-30 14:01:08 -05:00
|
|
|
|
2017-06-02 12:38:22 -04:00
|
|
|
|
2020-11-26 10:12:20 -05:00
|
|
|
|
2017-06-02 12:38:22 -04:00
|
|
|
|
2015-02-18 05:47:55 -08:00
|
|
|
|
2014-08-31 21:30:27 -04:00
|
|
|
|
2016-11-30 14:01:08 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-06-27 18:48:21 -04:00
|
|
|
|
2014-08-31 21:30:27 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-09-10 18:05:26 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-09-04 13:31:35 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-06-30 13:08:01 +03:00
|
|
|
|
2014-09-04 13:31:35 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-09-04 13:31:10 -04:00
|
|
|
|
|
|
|
|
|
2017-03-18 17:03:00 -04:00
|
|
|
|
|
|
|
|
|
2009-02-12 05:03:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-03-18 17:03:00 -04:00
|
|
|
|
|
|
|
|
|
2009-02-12 05:03:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-08-04 22:11:48 -04:00
|
|
|
|
2017-03-18 17:03:00 -04:00
|
|
|
|
2017-04-12 19:24:35 -04:00
|
|
|
|
2023-08-31 13:52:11 +00:00
|
|
|
|
2014-08-04 22:11:47 -04:00
|
|
|
|
2021-11-15 11:02:33 -08:00
|
|
|
|
2022-02-17 09:05:02 -08:00
|
|
|
|
net-timestamp: TCP timestamping
TCP timestamping extends SO_TIMESTAMPING to bytestreams.
Bytestreams do not have a 1:1 relationship between send() buffers and
network packets. The feature interprets a send call on a bytestream as
a request for a timestamp for the last byte in that send() buffer.
The choice corresponds to a request for a timestamp when all bytes in
the buffer have been sent. That assumption depends on in-order kernel
transmission. This is the common case. That said, it is possible to
construct a traffic shaping tree that would result in reordering.
The guarantee is strong, then, but not ironclad.
This implementation supports send and sendpages (splice). GSO replaces
one large packet with multiple smaller packets. This patch also copies
the option into the correct smaller packet.
This patch does not yet support timestamping on data in an initial TCP
Fast Open SYN, because that takes a very different data path.
If ID generation in ee_data is enabled, bytestream timestamps return a
byte offset, instead of the packet counter for datagrams.
The implementation supports a single timestamp per packet. It silenty
replaces requests for previous timestamps. To avoid missing tstamps,
flush the tcp queue by disabling Nagle, cork and autocork. Missing
tstamps can be detected by offset when the ee_data ID is enabled.
Implementation details:
- On GSO, the timestamping code can be included in the main loop. I
moved it into its own loop to reduce the impact on the common case
to a single branch.
- To avoid leaking the absolute seqno to userspace, the offset
returned in ee_data must always be relative. It is an offset between
an skb and sk field. The first is always set (also for GSO & ACK).
The second must also never be uninitialized. Only allow the ID
option on sockets in the ESTABLISHED state, for which the seqno
is available. Never reset it to zero (instead, move it to the
current seqno when reenabling the option).
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-08-04 22:11:49 -04:00
|
|
|
|
2010-05-29 00:20:48 -07:00
|
|
|
|
2009-02-12 05:03:37 +00:00
|
|
|
|
2010-05-29 00:20:48 -07:00
|
|
|
|
2009-02-12 05:03:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-09-04 13:31:10 -04:00
|
|
|
|
2015-01-30 13:29:32 -05:00
|
|
|
|
|
|
|
|
|
2026-02-20 19:38:58 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-01-30 13:29:32 -05:00
|
|
|
|
2024-10-06 07:26:09 +09:00
|
|
|
|
2015-01-30 13:29:32 -05:00
|
|
|
|
|
|
|
|
|
2026-02-20 19:38:58 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-01-30 13:29:32 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-09-04 13:31:10 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-01-30 13:29:32 -05:00
|
|
|
|
2017-12-13 14:41:06 -05:00
|
|
|
|
2015-01-30 13:29:32 -05:00
|
|
|
|
2017-03-03 21:01:03 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-06-30 13:08:01 +03:00
|
|
|
|
2017-03-03 21:01:03 -08:00
|
|
|
|
2017-03-18 17:03:00 -04:00
|
|
|
|
2017-03-03 21:01:03 -08:00
|
|
|
|
2017-12-13 14:41:06 -05:00
|
|
|
|
2017-03-03 21:01:03 -08:00
|
|
|
|
2017-12-13 14:41:06 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-09-04 13:31:10 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-20 15:29:33 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-20 15:29:36 +08:00
|
|
|
|
2025-02-20 15:29:33 +08:00
|
|
|
|
|
|
|
|
|
2025-02-20 15:29:37 +08:00
|
|
|
|
2025-03-18 21:06:42 +02:00
|
|
|
|
|
|
|
|
|
2025-02-20 15:29:33 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-20 15:29:34 +08:00
|
|
|
|
2025-02-20 15:29:35 +08:00
|
|
|
|
2025-02-20 15:29:34 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-20 15:29:35 +08:00
|
|
|
|
2025-02-20 15:29:36 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-20 15:29:35 +08:00
|
|
|
|
2025-02-20 15:29:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-20 15:29:34 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-09-04 13:31:10 -04:00
|
|
|
|
2021-01-20 12:41:55 -08:00
|
|
|
|
2014-09-04 13:31:10 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-03-18 17:03:00 -04:00
|
|
|
|
2023-08-31 13:52:11 +00:00
|
|
|
|
2014-09-04 13:31:10 -04:00
|
|
|
|
2015-03-11 15:43:55 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-20 15:29:34 +08:00
|
|
|
|
2025-02-20 15:29:35 +08:00
|
|
|
|
|
|
|
|
|
2025-02-20 15:29:34 +08:00
|
|
|
|
2025-02-20 15:29:33 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-31 13:52:11 +00:00
|
|
|
|
|
|
|
|
|
2017-05-19 17:52:40 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-31 13:52:11 +00:00
|
|
|
|
2015-03-11 15:43:55 -04:00
|
|
|
|
2014-09-04 13:31:10 -04:00
|
|
|
|
|
|
|
|
|
2016-11-27 23:07:18 -08:00
|
|
|
|
|
|
|
|
|
2023-08-31 13:52:11 +00:00
|
|
|
|
2021-11-15 11:02:33 -08:00
|
|
|
|
2021-01-20 12:41:55 -08:00
|
|
|
|
|
|
|
|
|
2017-03-18 17:03:00 -04:00
|
|
|
|
|
|
|
|
|
2016-11-27 23:07:18 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-01-30 13:29:31 -05:00
|
|
|
|
tcp/udp: Fix memleaks of sk and zerocopy skbs with TX timestamp.
syzkaller reported [0] memory leaks of an UDP socket and ZEROCOPY
skbs. We can reproduce the problem with these sequences:
sk = socket(AF_INET, SOCK_DGRAM, 0)
sk.setsockopt(SOL_SOCKET, SO_TIMESTAMPING, SOF_TIMESTAMPING_TX_SOFTWARE)
sk.setsockopt(SOL_SOCKET, SO_ZEROCOPY, 1)
sk.sendto(b'', MSG_ZEROCOPY, ('127.0.0.1', 53))
sk.close()
sendmsg() calls msg_zerocopy_alloc(), which allocates a skb, sets
skb->cb->ubuf.refcnt to 1, and calls sock_hold(). Here, struct
ubuf_info_msgzc indirectly holds a refcnt of the socket. When the
skb is sent, __skb_tstamp_tx() clones it and puts the clone into
the socket's error queue with the TX timestamp.
When the original skb is received locally, skb_copy_ubufs() calls
skb_unclone(), and pskb_expand_head() increments skb->cb->ubuf.refcnt.
This additional count is decremented while freeing the skb, but struct
ubuf_info_msgzc still has a refcnt, so __msg_zerocopy_callback() is
not called.
The last refcnt is not released unless we retrieve the TX timestamped
skb by recvmsg(). Since we clear the error queue in inet_sock_destruct()
after the socket's refcnt reaches 0, there is a circular dependency.
If we close() the socket holding such skbs, we never call sock_put()
and leak the count, sk, and skb.
TCP has the same problem, and commit e0c8bccd40fc ("net: stream:
purge sk_error_queue in sk_stream_kill_queues()") tried to fix it
by calling skb_queue_purge() during close(). However, there is a
small chance that skb queued in a qdisc or device could be put
into the error queue after the skb_queue_purge() call.
In __skb_tstamp_tx(), the cloned skb should not have a reference
to the ubuf to remove the circular dependency, but skb_clone() does
not call skb_copy_ubufs() for zerocopy skb. So, we need to call
skb_orphan_frags_rx() for the cloned skb to call skb_copy_ubufs().
[0]:
BUG: memory leak
unreferenced object 0xffff88800c6d2d00 (size 1152):
comm "syz-executor392", pid 264, jiffies 4294785440 (age 13.044s)
hex dump (first 32 bytes):
00 00 00 00 00 00 00 00 cd af e8 81 00 00 00 00 ................
02 00 07 40 00 00 00 00 00 00 00 00 00 00 00 00 ...@............
backtrace:
[<0000000055636812>] sk_prot_alloc+0x64/0x2a0 net/core/sock.c:2024
[<0000000054d77b7a>] sk_alloc+0x3b/0x800 net/core/sock.c:2083
[<0000000066f3c7e0>] inet_create net/ipv4/af_inet.c:319 [inline]
[<0000000066f3c7e0>] inet_create+0x31e/0xe40 net/ipv4/af_inet.c:245
[<000000009b83af97>] __sock_create+0x2ab/0x550 net/socket.c:1515
[<00000000b9b11231>] sock_create net/socket.c:1566 [inline]
[<00000000b9b11231>] __sys_socket_create net/socket.c:1603 [inline]
[<00000000b9b11231>] __sys_socket_create net/socket.c:1588 [inline]
[<00000000b9b11231>] __sys_socket+0x138/0x250 net/socket.c:1636
[<000000004fb45142>] __do_sys_socket net/socket.c:1649 [inline]
[<000000004fb45142>] __se_sys_socket net/socket.c:1647 [inline]
[<000000004fb45142>] __x64_sys_socket+0x73/0xb0 net/socket.c:1647
[<0000000066999e0e>] do_syscall_x64 arch/x86/entry/common.c:50 [inline]
[<0000000066999e0e>] do_syscall_64+0x38/0x90 arch/x86/entry/common.c:80
[<0000000017f238c1>] entry_SYSCALL_64_after_hwframe+0x63/0xcd
BUG: memory leak
unreferenced object 0xffff888017633a00 (size 240):
comm "syz-executor392", pid 264, jiffies 4294785440 (age 13.044s)
hex dump (first 32 bytes):
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
00 00 00 00 00 00 00 00 00 2d 6d 0c 80 88 ff ff .........-m.....
backtrace:
[<000000002b1c4368>] __alloc_skb+0x229/0x320 net/core/skbuff.c:497
[<00000000143579a6>] alloc_skb include/linux/skbuff.h:1265 [inline]
[<00000000143579a6>] sock_omalloc+0xaa/0x190 net/core/sock.c:2596
[<00000000be626478>] msg_zerocopy_alloc net/core/skbuff.c:1294 [inline]
[<00000000be626478>] msg_zerocopy_realloc+0x1ce/0x7f0 net/core/skbuff.c:1370
[<00000000cbfc9870>] __ip_append_data+0x2adf/0x3b30 net/ipv4/ip_output.c:1037
[<0000000089869146>] ip_make_skb+0x26c/0x2e0 net/ipv4/ip_output.c:1652
[<00000000098015c2>] udp_sendmsg+0x1bac/0x2390 net/ipv4/udp.c:1253
[<0000000045e0e95e>] inet_sendmsg+0x10a/0x150 net/ipv4/af_inet.c:819
[<000000008d31bfde>] sock_sendmsg_nosec net/socket.c:714 [inline]
[<000000008d31bfde>] sock_sendmsg+0x141/0x190 net/socket.c:734
[<0000000021e21aa4>] __sys_sendto+0x243/0x360 net/socket.c:2117
[<00000000ac0af00c>] __do_sys_sendto net/socket.c:2129 [inline]
[<00000000ac0af00c>] __se_sys_sendto net/socket.c:2125 [inline]
[<00000000ac0af00c>] __x64_sys_sendto+0xe1/0x1c0 net/socket.c:2125
[<0000000066999e0e>] do_syscall_x64 arch/x86/entry/common.c:50 [inline]
[<0000000066999e0e>] do_syscall_64+0x38/0x90 arch/x86/entry/common.c:80
[<0000000017f238c1>] entry_SYSCALL_64_after_hwframe+0x63/0xcd
Fixes: f214f915e7db ("tcp: enable MSG_ZEROCOPY")
Fixes: b5947e5d1e71 ("udp: msg_zerocopy")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2023-04-24 15:20:22 -07:00
|
|
|
|
2023-05-22 17:30:20 +02:00
|
|
|
|
|
|
|
|
|
tcp/udp: Fix memleaks of sk and zerocopy skbs with TX timestamp.
syzkaller reported [0] memory leaks of an UDP socket and ZEROCOPY
skbs. We can reproduce the problem with these sequences:
sk = socket(AF_INET, SOCK_DGRAM, 0)
sk.setsockopt(SOL_SOCKET, SO_TIMESTAMPING, SOF_TIMESTAMPING_TX_SOFTWARE)
sk.setsockopt(SOL_SOCKET, SO_ZEROCOPY, 1)
sk.sendto(b'', MSG_ZEROCOPY, ('127.0.0.1', 53))
sk.close()
sendmsg() calls msg_zerocopy_alloc(), which allocates a skb, sets
skb->cb->ubuf.refcnt to 1, and calls sock_hold(). Here, struct
ubuf_info_msgzc indirectly holds a refcnt of the socket. When the
skb is sent, __skb_tstamp_tx() clones it and puts the clone into
the socket's error queue with the TX timestamp.
When the original skb is received locally, skb_copy_ubufs() calls
skb_unclone(), and pskb_expand_head() increments skb->cb->ubuf.refcnt.
This additional count is decremented while freeing the skb, but struct
ubuf_info_msgzc still has a refcnt, so __msg_zerocopy_callback() is
not called.
The last refcnt is not released unless we retrieve the TX timestamped
skb by recvmsg(). Since we clear the error queue in inet_sock_destruct()
after the socket's refcnt reaches 0, there is a circular dependency.
If we close() the socket holding such skbs, we never call sock_put()
and leak the count, sk, and skb.
TCP has the same problem, and commit e0c8bccd40fc ("net: stream:
purge sk_error_queue in sk_stream_kill_queues()") tried to fix it
by calling skb_queue_purge() during close(). However, there is a
small chance that skb queued in a qdisc or device could be put
into the error queue after the skb_queue_purge() call.
In __skb_tstamp_tx(), the cloned skb should not have a reference
to the ubuf to remove the circular dependency, but skb_clone() does
not call skb_copy_ubufs() for zerocopy skb. So, we need to call
skb_orphan_frags_rx() for the cloned skb to call skb_copy_ubufs().
[0]:
BUG: memory leak
unreferenced object 0xffff88800c6d2d00 (size 1152):
comm "syz-executor392", pid 264, jiffies 4294785440 (age 13.044s)
hex dump (first 32 bytes):
00 00 00 00 00 00 00 00 cd af e8 81 00 00 00 00 ................
02 00 07 40 00 00 00 00 00 00 00 00 00 00 00 00 ...@............
backtrace:
[<0000000055636812>] sk_prot_alloc+0x64/0x2a0 net/core/sock.c:2024
[<0000000054d77b7a>] sk_alloc+0x3b/0x800 net/core/sock.c:2083
[<0000000066f3c7e0>] inet_create net/ipv4/af_inet.c:319 [inline]
[<0000000066f3c7e0>] inet_create+0x31e/0xe40 net/ipv4/af_inet.c:245
[<000000009b83af97>] __sock_create+0x2ab/0x550 net/socket.c:1515
[<00000000b9b11231>] sock_create net/socket.c:1566 [inline]
[<00000000b9b11231>] __sys_socket_create net/socket.c:1603 [inline]
[<00000000b9b11231>] __sys_socket_create net/socket.c:1588 [inline]
[<00000000b9b11231>] __sys_socket+0x138/0x250 net/socket.c:1636
[<000000004fb45142>] __do_sys_socket net/socket.c:1649 [inline]
[<000000004fb45142>] __se_sys_socket net/socket.c:1647 [inline]
[<000000004fb45142>] __x64_sys_socket+0x73/0xb0 net/socket.c:1647
[<0000000066999e0e>] do_syscall_x64 arch/x86/entry/common.c:50 [inline]
[<0000000066999e0e>] do_syscall_64+0x38/0x90 arch/x86/entry/common.c:80
[<0000000017f238c1>] entry_SYSCALL_64_after_hwframe+0x63/0xcd
BUG: memory leak
unreferenced object 0xffff888017633a00 (size 240):
comm "syz-executor392", pid 264, jiffies 4294785440 (age 13.044s)
hex dump (first 32 bytes):
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
00 00 00 00 00 00 00 00 00 2d 6d 0c 80 88 ff ff .........-m.....
backtrace:
[<000000002b1c4368>] __alloc_skb+0x229/0x320 net/core/skbuff.c:497
[<00000000143579a6>] alloc_skb include/linux/skbuff.h:1265 [inline]
[<00000000143579a6>] sock_omalloc+0xaa/0x190 net/core/sock.c:2596
[<00000000be626478>] msg_zerocopy_alloc net/core/skbuff.c:1294 [inline]
[<00000000be626478>] msg_zerocopy_realloc+0x1ce/0x7f0 net/core/skbuff.c:1370
[<00000000cbfc9870>] __ip_append_data+0x2adf/0x3b30 net/ipv4/ip_output.c:1037
[<0000000089869146>] ip_make_skb+0x26c/0x2e0 net/ipv4/ip_output.c:1652
[<00000000098015c2>] udp_sendmsg+0x1bac/0x2390 net/ipv4/udp.c:1253
[<0000000045e0e95e>] inet_sendmsg+0x10a/0x150 net/ipv4/af_inet.c:819
[<000000008d31bfde>] sock_sendmsg_nosec net/socket.c:714 [inline]
[<000000008d31bfde>] sock_sendmsg+0x141/0x190 net/socket.c:734
[<0000000021e21aa4>] __sys_sendto+0x243/0x360 net/socket.c:2117
[<00000000ac0af00c>] __do_sys_sendto net/socket.c:2129 [inline]
[<00000000ac0af00c>] __se_sys_sendto net/socket.c:2125 [inline]
[<00000000ac0af00c>] __x64_sys_sendto+0xe1/0x1c0 net/socket.c:2125
[<0000000066999e0e>] do_syscall_x64 arch/x86/entry/common.c:50 [inline]
[<0000000066999e0e>] do_syscall_64+0x38/0x90 arch/x86/entry/common.c:80
[<0000000017f238c1>] entry_SYSCALL_64_after_hwframe+0x63/0xcd
Fixes: f214f915e7db ("tcp: enable MSG_ZEROCOPY")
Fixes: b5947e5d1e71 ("udp: msg_zerocopy")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2023-04-24 15:20:22 -07:00
|
|
|
|
2023-05-22 17:30:20 +02:00
|
|
|
|
2016-11-27 23:07:18 -08:00
|
|
|
|
2014-09-04 13:31:10 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-01-30 13:29:31 -05:00
|
|
|
|
2017-06-08 11:35:03 -04:00
|
|
|
|
|
|
|
|
|
2015-01-30 13:29:31 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-03-02 11:55:44 -08:00
|
|
|
|
2015-01-30 13:29:31 -05:00
|
|
|
|
2017-03-18 17:03:00 -04:00
|
|
|
|
2014-09-04 13:31:10 -04:00
|
|
|
|
2014-08-04 22:11:48 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-01-20 12:41:55 -08:00
|
|
|
|
2014-08-04 22:11:48 -04:00
|
|
|
|
|
|
|
|
|
2009-02-12 05:03:37 +00:00
|
|
|
|
|
|
|
|
|
2023-04-17 08:53:46 -07:00
|
|
|
|
2011-11-09 10:15:42 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-03-03 21:01:02 -08:00
|
|
|
|
2011-11-09 10:15:42 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-03-03 21:01:02 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-06-30 13:08:01 +03:00
|
|
|
|
2017-03-03 21:01:02 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-11-09 10:15:42 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-04-17 08:53:46 -07:00
|
|
|
|
2011-11-09 10:15:42 +01:00
|
|
|
|
2008-02-04 23:49:54 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-10-10 06:59:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-05 17:06:18 +00:00
|
|
|
|
2018-10-10 06:59:35 -07:00
|
|
|
|
|
|
|
|
|
2008-02-04 23:49:54 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-10-10 06:59:35 -07:00
|
|
|
|
2008-02-04 23:49:54 -05:00
|
|
|
|
2023-05-05 17:06:18 +00:00
|
|
|
|
2008-02-04 23:49:54 -05:00
|
|
|
|
|
|
|
|
|
2009-02-10 02:09:24 -08:00
|
|
|
|
2008-02-04 23:49:54 -05:00
|
|
|
|
2014-01-09 10:02:46 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-11 13:56:05 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-02-19 22:23:04 -08:00
|
|
|
|
2014-03-11 13:56:05 +00:00
|
|
|
|
2020-02-19 22:23:04 -08:00
|
|
|
|
2014-03-11 13:56:05 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-01-09 10:02:46 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-11 13:56:05 +00:00
|
|
|
|
2014-01-09 10:02:46 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-11 13:56:05 +00:00
|
|
|
|
2014-01-09 10:02:46 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-08-06 19:57:18 +08:00
|
|
|
|
2014-01-09 10:02:46 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-11 13:56:05 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-01-09 10:02:46 +00:00
|
|
|
|
2014-03-11 13:56:05 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-01-09 10:02:46 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-11 13:56:05 +00:00
|
|
|
|
2014-01-09 10:02:46 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-11 13:56:05 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-01-09 10:02:46 +00:00
|
|
|
|
2014-03-11 13:56:05 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-01-09 10:02:46 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-03-11 13:56:05 +00:00
|
|
|
|
2014-01-09 10:02:46 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-05-02 14:01:07 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-08-13 05:54:07 +02:00
|
|
|
|
|
|
|
|
|
2015-05-02 14:01:07 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-08-13 05:54:07 +02:00
|
|
|
|
2015-05-02 14:01:07 +02:00
|
|
|
|
2015-08-13 05:54:07 +02:00
|
|
|
|
2015-05-02 14:01:07 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-08-13 05:54:07 +02:00
|
|
|
|
|
|
|
|
|
2015-05-02 14:01:07 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-05-05 00:19:35 +02:00
|
|
|
|
2015-05-02 14:01:07 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-08-13 05:54:07 +02:00
|
|
|
|
2015-05-02 14:01:07 +02:00
|
|
|
|
2015-08-13 05:54:07 +02:00
|
|
|
|
|
|
|
|
|
2015-05-02 14:01:07 +02:00
|
|
|
|
2016-02-24 04:21:42 +01:00
|
|
|
|
2015-05-02 14:01:07 +02:00
|
|
|
|
2016-02-24 04:21:42 +01:00
|
|
|
|
2015-05-02 14:01:07 +02:00
|
|
|
|
2015-08-13 05:54:07 +02:00
|
|
|
|
|
|
|
|
|
2015-05-02 14:01:07 +02:00
|
|
|
|
|
|
|
|
|
2015-08-13 05:54:07 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-05-02 14:01:07 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-06-19 16:22:28 -07:00
|
|
|
|
|
|
|
|
|
2012-05-13 21:56:26 +00:00
|
|
|
|
|
|
|
|
|
2008-06-19 16:22:28 -07:00
|
|
|
|
|
|
|
|
|
2012-05-19 03:02:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-10-22 09:03:40 +00:00
|
|
|
|
|
|
|
|
|
2024-03-06 16:00:23 +00:00
|
|
|
|
2012-10-22 09:03:40 +00:00
|
|
|
|
2012-05-19 03:02:02 +00:00
|
|
|
|
2012-10-22 09:03:40 +00:00
|
|
|
|
2012-05-19 03:02:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-06-08 14:01:44 +00:00
|
|
|
|
2012-05-19 03:02:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-10-04 10:48:35 -07:00
|
|
|
|
2012-05-19 03:02:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-04-13 17:03:53 +08:00
|
|
|
|
skbuff: Optimization of SKB coalescing for page pool
In order to address the issues encountered with commit 1effe8ca4e34
("skbuff: fix coalescing for page_pool fragment recycling"), the
combination of the following condition was excluded from skb coalescing:
from->pp_recycle = 1
from->cloned = 1
to->pp_recycle = 1
However, with page pool environments, the aforementioned combination can
be quite common(ex. NetworkMananger may lead to the additional
packet_type being registered, thus the cloning). In scenarios with a
higher number of small packets, it can significantly affect the success
rate of coalescing. For example, considering packets of 256 bytes size,
our comparison of coalescing success rate is as follows:
Without page pool: 70%
With page pool: 13%
Consequently, this has an impact on performance:
Without page pool: 2.57 Gbits/sec
With page pool: 2.26 Gbits/sec
Therefore, it seems worthwhile to optimize this scenario and enable
coalescing of this particular combination. To achieve this, we need to
ensure the correct increment of the "from" SKB page's page pool
reference count (pp_ref_count).
Following this optimization, the success rate of coalescing measured in
our environment has improved as follows:
With page pool: 60%
This success rate is approaching the rate achieved without using page
pool, and the performance has also been improved:
With page pool: 2.52 Gbits/sec
Below is the performance comparison for small packets before and after
this optimization. We observe no impact to packets larger than 4K.
packet size before after improved
(bytes) (Gbits/sec) (Gbits/sec)
128 1.19 1.27 7.13%
256 2.26 2.52 11.75%
512 4.13 4.81 16.50%
1024 6.17 6.73 9.05%
2048 14.54 15.47 6.45%
4096 25.44 27.87 9.52%
Signed-off-by: Liang Chen <liangchen.linux@gmail.com>
Reviewed-by: Yunsheng Lin <linyunsheng@huawei.com>
Suggested-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Mina Almasry <almasrymina@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2023-12-15 11:30:11 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-06-07 21:02:38 +02:00
|
|
|
|
skbuff: Optimization of SKB coalescing for page pool
In order to address the issues encountered with commit 1effe8ca4e34
("skbuff: fix coalescing for page_pool fragment recycling"), the
combination of the following condition was excluded from skb coalescing:
from->pp_recycle = 1
from->cloned = 1
to->pp_recycle = 1
However, with page pool environments, the aforementioned combination can
be quite common(ex. NetworkMananger may lead to the additional
packet_type being registered, thus the cloning). In scenarios with a
higher number of small packets, it can significantly affect the success
rate of coalescing. For example, considering packets of 256 bytes size,
our comparison of coalescing success rate is as follows:
Without page pool: 70%
With page pool: 13%
Consequently, this has an impact on performance:
Without page pool: 2.57 Gbits/sec
With page pool: 2.26 Gbits/sec
Therefore, it seems worthwhile to optimize this scenario and enable
coalescing of this particular combination. To achieve this, we need to
ensure the correct increment of the "from" SKB page's page pool
reference count (pp_ref_count).
Following this optimization, the success rate of coalescing measured in
our environment has improved as follows:
With page pool: 60%
This success rate is approaching the rate achieved without using page
pool, and the performance has also been improved:
With page pool: 2.52 Gbits/sec
Below is the performance comparison for small packets before and after
this optimization. We observe no impact to packets larger than 4K.
packet size before after improved
(bytes) (Gbits/sec) (Gbits/sec)
128 1.19 1.27 7.13%
256 2.26 2.52 11.75%
512 4.13 4.81 16.50%
1024 6.17 6.73 9.05%
2048 14.54 15.47 6.45%
4096 25.44 27.87 9.52%
Signed-off-by: Liang Chen <liangchen.linux@gmail.com>
Reviewed-by: Yunsheng Lin <linyunsheng@huawei.com>
Suggested-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Mina Almasry <almasrymina@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2023-12-15 11:30:11 +08:00
|
|
|
|
2021-06-07 21:02:38 +02:00
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:52 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-09-15 04:19:52 -07:00
|
|
|
|
|
|
|
|
|
2012-05-19 03:02:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-10-04 10:48:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-05-19 03:02:02 +00:00
|
|
|
|
sock: enable MSG_ZEROCOPY
Prepare the datapath for refcounted ubuf_info. Clone ubuf_info with
skb_zerocopy_clone() wherever needed due to skb split, merge, resize
or clone.
Split skb_orphan_frags into two variants. The split, merge, .. paths
support reference counted zerocopy buffers, so do not do a deep copy.
Add skb_orphan_frags_rx for paths that may loop packets to receive
sockets. That is not allowed, as it may cause unbounded latency.
Deep copy all zerocopy copy buffers, ref-counted or not, in this path.
The exact locations to modify were chosen by exhaustively searching
through all code that might modify skb_frag references and/or the
the SKBTX_DEV_ZEROCOPY tx_flags bit.
The changes err on the safe side, in two ways.
(1) legacy ubuf_info paths virtio and tap are not modified. They keep
a 1:1 ubuf_info to sk_buff relationship. Calls to skb_orphan_frags
still call skb_copy_ubufs and thus copy frags in this case.
(2) not all copies deep in the stack are addressed yet. skb_shift,
skb_split and skb_try_coalesce can be refined to avoid copying.
These are not in the hot path and this patch is hairy enough as
is, so that is left for future refinement.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-08-03 16:29:41 -04:00
|
|
|
|
|
|
|
|
|
2012-05-19 03:02:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-10-04 10:48:35 -07:00
|
|
|
|
|
|
|
|
|
2012-05-19 03:02:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-10-04 10:48:35 -07:00
|
|
|
|
2012-05-19 03:02:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-10-04 10:48:35 -07:00
|
|
|
|
|
|
|
|
|
2012-05-19 03:02:02 +00:00
|
|
|
|
|
|
|
|
|
2012-09-28 20:15:30 +00:00
|
|
|
|
2012-05-19 03:02:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-10-04 10:48:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-05-13 04:16:35 +00:00
|
|
|
|
|
|
|
|
|
2012-05-19 03:02:02 +00:00
|
|
|
|
|
|
|
|
|
2017-10-04 10:48:35 -07:00
|
|
|
|
2012-05-19 03:02:02 +00:00
|
|
|
|
2012-09-18 16:53:21 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-05-02 10:54:22 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-05-19 03:02:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-06-26 16:11:27 +02:00
|
|
|
|
|
|
|
|
|
2013-09-02 15:34:56 +02:00
|
|
|
|
2013-06-26 16:11:27 +02:00
|
|
|
|
|
|
|
|
|
2013-09-02 15:34:56 +02:00
|
|
|
|
2013-06-26 16:11:27 +02:00
|
|
|
|
2024-08-22 13:57:34 +01:00
|
|
|
|
2013-09-02 15:34:56 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2013-06-26 16:11:27 +02:00
|
|
|
|
2013-09-02 15:34:56 +02:00
|
|
|
|
2013-06-26 16:11:27 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-05-04 16:39:18 -07:00
|
|
|
|
2013-06-26 16:11:27 +02:00
|
|
|
|
2019-09-26 20:37:05 +02:00
|
|
|
|
2019-09-29 20:54:03 +02:00
|
|
|
|
2013-06-26 16:11:27 +02:00
|
|
|
|
skbuff: Do not scrub skb mark within the same name space
On Wed, Apr 15, 2015 at 05:41:26PM +0200, Nicolas Dichtel wrote:
> Le 15/04/2015 15:57, Herbert Xu a écrit :
> >On Wed, Apr 15, 2015 at 06:22:29PM +0800, Herbert Xu wrote:
> [snip]
> >Subject: skbuff: Do not scrub skb mark within the same name space
> >
> >The commit ea23192e8e577dfc51e0f4fc5ca113af334edff9 ("tunnels:
> Maybe add a Fixes tag?
> Fixes: ea23192e8e57 ("tunnels: harmonize cleanup done on skb on rx path")
>
> >harmonize cleanup done on skb on rx path") broke anyone trying to
> >use netfilter marking across IPv4 tunnels. While most of the
> >fields that are cleared by skb_scrub_packet don't matter, the
> >netfilter mark must be preserved.
> >
> >This patch rearranges skb_scurb_packet to preserve the mark field.
> nit: s/scurb/scrub
>
> Else it's fine for me.
Sure.
PS I used the wrong email for James the first time around. So
let me repeat the question here. Should secmark be preserved
or cleared across tunnels within the same name space? In fact,
do our security models even support name spaces?
---8<---
The commit ea23192e8e577dfc51e0f4fc5ca113af334edff9 ("tunnels:
harmonize cleanup done on skb on rx path") broke anyone trying to
use netfilter marking across IPv4 tunnels. While most of the
fields that are cleared by skb_scrub_packet don't matter, the
netfilter mark must be preserved.
This patch rearranges skb_scrub_packet to preserve the mark field.
Fixes: ea23192e8e57 ("tunnels: harmonize cleanup done on skb on rx path")
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-04-16 09:03:27 +08:00
|
|
|
|
2018-11-19 16:11:07 +00:00
|
|
|
|
|
|
|
|
|
2018-12-04 08:15:10 +00:00
|
|
|
|
2018-11-19 16:11:07 +00:00
|
|
|
|
2025-02-22 11:35:18 +08:00
|
|
|
|
2018-11-19 16:11:07 +00:00
|
|
|
|
skbuff: Do not scrub skb mark within the same name space
On Wed, Apr 15, 2015 at 05:41:26PM +0200, Nicolas Dichtel wrote:
> Le 15/04/2015 15:57, Herbert Xu a écrit :
> >On Wed, Apr 15, 2015 at 06:22:29PM +0800, Herbert Xu wrote:
> [snip]
> >Subject: skbuff: Do not scrub skb mark within the same name space
> >
> >The commit ea23192e8e577dfc51e0f4fc5ca113af334edff9 ("tunnels:
> Maybe add a Fixes tag?
> Fixes: ea23192e8e57 ("tunnels: harmonize cleanup done on skb on rx path")
>
> >harmonize cleanup done on skb on rx path") broke anyone trying to
> >use netfilter marking across IPv4 tunnels. While most of the
> >fields that are cleared by skb_scrub_packet don't matter, the
> >netfilter mark must be preserved.
> >
> >This patch rearranges skb_scurb_packet to preserve the mark field.
> nit: s/scurb/scrub
>
> Else it's fine for me.
Sure.
PS I used the wrong email for James the first time around. So
let me repeat the question here. Should secmark be preserved
or cleared across tunnels within the same name space? In fact,
do our security models even support name spaces?
---8<---
The commit ea23192e8e577dfc51e0f4fc5ca113af334edff9 ("tunnels:
harmonize cleanup done on skb on rx path") broke anyone trying to
use netfilter marking across IPv4 tunnels. While most of the
fields that are cleared by skb_scrub_packet don't matter, the
netfilter mark must be preserved.
This patch rearranges skb_scrub_packet to preserve the mark field.
Fixes: ea23192e8e57 ("tunnels: harmonize cleanup done on skb on rx path")
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-04-16 09:03:27 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-03-02 11:55:31 -08:00
|
|
|
|
2013-06-26 16:11:27 +02:00
|
|
|
|
|
|
|
|
|
2014-01-26 10:58:16 +01:00
|
|
|
|
net: Always untag vlan-tagged traffic on input.
Currently the functionality to untag traffic on input resides
as part of the vlan module and is build only when VLAN support
is enabled in the kernel. When VLAN is disabled, the function
vlan_untag() turns into a stub and doesn't really untag the
packets. This seems to create an interesting interaction
between VMs supporting checksum offloading and some network drivers.
There are some drivers that do not allow the user to change
tx-vlan-offload feature of the driver. These drivers also seem
to assume that any VLAN-tagged traffic they transmit will
have the vlan information in the vlan_tci and not in the vlan
header already in the skb. When transmitting skbs that already
have tagged data with partial checksum set, the checksum doesn't
appear to be updated correctly by the card thus resulting in a
failure to establish TCP connections.
The following is a packet trace taken on the receiver where a
sender is a VM with a VLAN configued. The host VM is running on
doest not have VLAN support and the outging interface on the
host is tg3:
10:12:43.503055 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27243,
offset 0, flags [DF], proto TCP (6), length 60)
10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x48d9), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294837885 ecr 0,nop,wscale 7], length 0
10:12:44.505556 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27244,
offset 0, flags [DF], proto TCP (6), length 60)
10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x44ee), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294838888 ecr 0,nop,wscale 7], length 0
This connection finally times out.
I've only access to the TG3 hardware in this configuration thus have
only tested this with TG3 driver. There are a lot of other drivers
that do not permit user changes to vlan acceleration features, and
I don't know if they all suffere from a similar issue.
The patch attempt to fix this another way. It moves the vlan header
stipping code out of the vlan module and always builds it into the
kernel network core. This way, even if vlan is not supported on
a virtualizatoin host, the virtual machines running on top of such
host will still work with VLANs enabled.
CC: Patrick McHardy <kaber@trash.net>
CC: Nithin Nayak Sujir <nsujir@broadcom.com>
CC: Michael Chan <mchan@broadcom.com>
CC: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Vladislav Yasevich <vyasevic@redhat.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-08-08 14:42:13 -04:00
|
|
|
|
|
|
|
|
|
2019-04-16 10:22:28 +09:00
|
|
|
|
|
|
|
|
|
2018-03-13 14:51:27 +09:00
|
|
|
|
net: Always untag vlan-tagged traffic on input.
Currently the functionality to untag traffic on input resides
as part of the vlan module and is build only when VLAN support
is enabled in the kernel. When VLAN is disabled, the function
vlan_untag() turns into a stub and doesn't really untag the
packets. This seems to create an interesting interaction
between VMs supporting checksum offloading and some network drivers.
There are some drivers that do not allow the user to change
tx-vlan-offload feature of the driver. These drivers also seem
to assume that any VLAN-tagged traffic they transmit will
have the vlan information in the vlan_tci and not in the vlan
header already in the skb. When transmitting skbs that already
have tagged data with partial checksum set, the checksum doesn't
appear to be updated correctly by the card thus resulting in a
failure to establish TCP connections.
The following is a packet trace taken on the receiver where a
sender is a VM with a VLAN configued. The host VM is running on
doest not have VLAN support and the outging interface on the
host is tg3:
10:12:43.503055 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27243,
offset 0, flags [DF], proto TCP (6), length 60)
10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x48d9), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294837885 ecr 0,nop,wscale 7], length 0
10:12:44.505556 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27244,
offset 0, flags [DF], proto TCP (6), length 60)
10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x44ee), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294838888 ecr 0,nop,wscale 7], length 0
This connection finally times out.
I've only access to the TG3 hardware in this configuration thus have
only tested this with TG3 driver. There are a lot of other drivers
that do not permit user changes to vlan acceleration features, and
I don't know if they all suffere from a similar issue.
The patch attempt to fix this another way. It moves the vlan header
stipping code out of the vlan module and always builds it into the
kernel network core. This way, even if vlan is not supported on
a virtualizatoin host, the virtual machines running on top of such
host will still work with VLANs enabled.
CC: Patrick McHardy <kaber@trash.net>
CC: Nithin Nayak Sujir <nsujir@broadcom.com>
CC: Michael Chan <mchan@broadcom.com>
CC: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Vladislav Yasevich <vyasevic@redhat.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-08-08 14:42:13 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-13 14:51:27 +09:00
|
|
|
|
2018-03-29 19:05:29 +09:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-04-16 10:22:28 +09:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: Always untag vlan-tagged traffic on input.
Currently the functionality to untag traffic on input resides
as part of the vlan module and is build only when VLAN support
is enabled in the kernel. When VLAN is disabled, the function
vlan_untag() turns into a stub and doesn't really untag the
packets. This seems to create an interesting interaction
between VMs supporting checksum offloading and some network drivers.
There are some drivers that do not allow the user to change
tx-vlan-offload feature of the driver. These drivers also seem
to assume that any VLAN-tagged traffic they transmit will
have the vlan information in the vlan_tci and not in the vlan
header already in the skb. When transmitting skbs that already
have tagged data with partial checksum set, the checksum doesn't
appear to be updated correctly by the card thus resulting in a
failure to establish TCP connections.
The following is a packet trace taken on the receiver where a
sender is a VM with a VLAN configued. The host VM is running on
doest not have VLAN support and the outging interface on the
host is tg3:
10:12:43.503055 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27243,
offset 0, flags [DF], proto TCP (6), length 60)
10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x48d9), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294837885 ecr 0,nop,wscale 7], length 0
10:12:44.505556 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27244,
offset 0, flags [DF], proto TCP (6), length 60)
10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x44ee), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294838888 ecr 0,nop,wscale 7], length 0
This connection finally times out.
I've only access to the TG3 hardware in this configuration thus have
only tested this with TG3 driver. There are a lot of other drivers
that do not permit user changes to vlan acceleration features, and
I don't know if they all suffere from a similar issue.
The patch attempt to fix this another way. It moves the vlan header
stipping code out of the vlan module and always builds it into the
kernel network core. This way, even if vlan is not supported on
a virtualizatoin host, the virtual machines running on top of such
host will still work with VLANs enabled.
CC: Patrick McHardy <kaber@trash.net>
CC: Nithin Nayak Sujir <nsujir@broadcom.com>
CC: Michael Chan <mchan@broadcom.com>
CC: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Vladislav Yasevich <vyasevic@redhat.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-08-08 14:42:13 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-01-13 17:13:44 +01:00
|
|
|
|
net: Always untag vlan-tagged traffic on input.
Currently the functionality to untag traffic on input resides
as part of the vlan module and is build only when VLAN support
is enabled in the kernel. When VLAN is disabled, the function
vlan_untag() turns into a stub and doesn't really untag the
packets. This seems to create an interesting interaction
between VMs supporting checksum offloading and some network drivers.
There are some drivers that do not allow the user to change
tx-vlan-offload feature of the driver. These drivers also seem
to assume that any VLAN-tagged traffic they transmit will
have the vlan information in the vlan_tci and not in the vlan
header already in the skb. When transmitting skbs that already
have tagged data with partial checksum set, the checksum doesn't
appear to be updated correctly by the card thus resulting in a
failure to establish TCP connections.
The following is a packet trace taken on the receiver where a
sender is a VM with a VLAN configued. The host VM is running on
doest not have VLAN support and the outging interface on the
host is tg3:
10:12:43.503055 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27243,
offset 0, flags [DF], proto TCP (6), length 60)
10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x48d9), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294837885 ecr 0,nop,wscale 7], length 0
10:12:44.505556 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27244,
offset 0, flags [DF], proto TCP (6), length 60)
10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x44ee), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294838888 ecr 0,nop,wscale 7], length 0
This connection finally times out.
I've only access to the TG3 hardware in this configuration thus have
only tested this with TG3 driver. There are a lot of other drivers
that do not permit user changes to vlan acceleration features, and
I don't know if they all suffere from a similar issue.
The patch attempt to fix this another way. It moves the vlan header
stipping code out of the vlan module and always builds it into the
kernel network core. This way, even if vlan is not supported on
a virtualizatoin host, the virtual machines running on top of such
host will still work with VLANs enabled.
CC: Patrick McHardy <kaber@trash.net>
CC: Nithin Nayak Sujir <nsujir@broadcom.com>
CC: Michael Chan <mchan@broadcom.com>
CC: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Vladislav Yasevich <vyasevic@redhat.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-08-08 14:42:13 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-08-15 04:44:31 -04:00
|
|
|
|
|
|
|
|
|
net: Always untag vlan-tagged traffic on input.
Currently the functionality to untag traffic on input resides
as part of the vlan module and is build only when VLAN support
is enabled in the kernel. When VLAN is disabled, the function
vlan_untag() turns into a stub and doesn't really untag the
packets. This seems to create an interesting interaction
between VMs supporting checksum offloading and some network drivers.
There are some drivers that do not allow the user to change
tx-vlan-offload feature of the driver. These drivers also seem
to assume that any VLAN-tagged traffic they transmit will
have the vlan information in the vlan_tci and not in the vlan
header already in the skb. When transmitting skbs that already
have tagged data with partial checksum set, the checksum doesn't
appear to be updated correctly by the card thus resulting in a
failure to establish TCP connections.
The following is a packet trace taken on the receiver where a
sender is a VM with a VLAN configued. The host VM is running on
doest not have VLAN support and the outging interface on the
host is tg3:
10:12:43.503055 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27243,
offset 0, flags [DF], proto TCP (6), length 60)
10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x48d9), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294837885 ecr 0,nop,wscale 7], length 0
10:12:44.505556 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27244,
offset 0, flags [DF], proto TCP (6), length 60)
10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x44ee), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294838888 ecr 0,nop,wscale 7], length 0
This connection finally times out.
I've only access to the TG3 hardware in this configuration thus have
only tested this with TG3 driver. There are a lot of other drivers
that do not permit user changes to vlan acceleration features, and
I don't know if they all suffere from a similar issue.
The patch attempt to fix this another way. It moves the vlan header
stipping code out of the vlan module and always builds it into the
kernel network core. This way, even if vlan is not supported on
a virtualizatoin host, the virtual machines running on top of such
host will still work with VLANs enabled.
CC: Patrick McHardy <kaber@trash.net>
CC: Nithin Nayak Sujir <nsujir@broadcom.com>
CC: Michael Chan <mchan@broadcom.com>
CC: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Vladislav Yasevich <vyasevic@redhat.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-08-08 14:42:13 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-11-09 23:47:23 +00:00
|
|
|
|
|
|
|
|
|
net: Always untag vlan-tagged traffic on input.
Currently the functionality to untag traffic on input resides
as part of the vlan module and is build only when VLAN support
is enabled in the kernel. When VLAN is disabled, the function
vlan_untag() turns into a stub and doesn't really untag the
packets. This seems to create an interesting interaction
between VMs supporting checksum offloading and some network drivers.
There are some drivers that do not allow the user to change
tx-vlan-offload feature of the driver. These drivers also seem
to assume that any VLAN-tagged traffic they transmit will
have the vlan information in the vlan_tci and not in the vlan
header already in the skb. When transmitting skbs that already
have tagged data with partial checksum set, the checksum doesn't
appear to be updated correctly by the card thus resulting in a
failure to establish TCP connections.
The following is a packet trace taken on the receiver where a
sender is a VM with a VLAN configued. The host VM is running on
doest not have VLAN support and the outging interface on the
host is tg3:
10:12:43.503055 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27243,
offset 0, flags [DF], proto TCP (6), length 60)
10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x48d9), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294837885 ecr 0,nop,wscale 7], length 0
10:12:44.505556 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27244,
offset 0, flags [DF], proto TCP (6), length 60)
10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x44ee), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294838888 ecr 0,nop,wscale 7], length 0
This connection finally times out.
I've only access to the TG3 hardware in this configuration thus have
only tested this with TG3 driver. There are a lot of other drivers
that do not permit user changes to vlan acceleration features, and
I don't know if they all suffere from a similar issue.
The patch attempt to fix this another way. It moves the vlan header
stipping code out of the vlan module and always builds it into the
kernel network core. This way, even if vlan is not supported on
a virtualizatoin host, the virtual machines running on top of such
host will still work with VLANs enabled.
CC: Patrick McHardy <kaber@trash.net>
CC: Nithin Nayak Sujir <nsujir@broadcom.com>
CC: Michael Chan <mchan@broadcom.com>
CC: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Vladislav Yasevich <vyasevic@redhat.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-08-08 14:42:13 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-09-17 04:49:49 -07:00
|
|
|
|
2022-04-16 18:58:00 +08:00
|
|
|
|
2014-11-19 14:05:01 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-19 16:53:25 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-09-19 19:11:09 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
2016-09-29 12:10:41 +03:00
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
|
|
|
|
|
2016-09-29 12:10:41 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
|
|
|
|
|
2016-09-29 12:10:41 +03:00
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: dsa: tag_ocelot: call only the relevant portion of __skb_vlan_pop() on TX
ocelot_xmit_get_vlan_info() calls __skb_vlan_pop() as the most
appropriate helper I could find which strips away a VLAN header.
That's all I need it to do, but __skb_vlan_pop() has more logic, which
will become incompatible with the future revert of commit 6d1ccff62780
("net: reset mac header in dev_start_xmit()").
Namely, it performs a sanity check on skb_mac_header(), which will stop
being set after the above revert, so it will return an error instead of
removing the VLAN tag.
ocelot_xmit_get_vlan_info() gets called in 2 circumstances:
(1) the port is under a VLAN-aware bridge and the bridge sends
VLAN-tagged packets
(2) the port is under a VLAN-aware bridge and somebody else (an 8021q
upper) sends VLAN-tagged packets (using a VID that isn't in the
bridge vlan tables)
In case (1), there is actually no bug to defend against, because
br_dev_xmit() calls skb_reset_mac_header() and things continue to work.
However, in case (2), illustrated using the commands below, it can be
seen that our intervention is needed, since __skb_vlan_pop() complains:
$ ip link add br0 type bridge vlan_filtering 1 && ip link set br0 up
$ ip link set $eth master br0 && ip link set $eth up
$ ip link add link $eth name $eth.100 type vlan id 100 && ip link set $eth.100 up
$ ip addr add 192.168.100.1/24 dev $eth.100
I could fend off the checks in __skb_vlan_pop() with some
skb_mac_header_was_set() calls, but seeing how few callers of
__skb_vlan_pop() there are from TX paths, that seems rather
unproductive.
As an alternative solution, extract the bare minimum logic to strip a
VLAN header, and move it to a new helper named vlan_remove_tag(), close
to the definition of vlan_insert_tag(). Document it appropriately and
make ocelot_xmit_get_vlan_info() call this smaller helper instead.
Seeing that it doesn't appear illegal to test skb->protocol in the TX
path, I guess it would be a good for vlan_remove_tag() to also absorb
the vlan_set_encap_proto() function call.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2023-04-21 01:56:01 +03:00
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-09-19 19:11:09 +03:00
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
2016-09-29 12:10:41 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-01-13 17:13:44 +01:00
|
|
|
|
2018-11-09 00:18:02 +01:00
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
2016-09-20 12:48:37 +03:00
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-09-20 12:48:37 +03:00
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-09-29 12:10:41 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
|
|
|
|
|
2015-01-13 17:13:44 +01:00
|
|
|
|
2016-09-29 12:10:41 +03:00
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
|
|
|
|
|
2016-09-29 12:10:41 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
2015-01-13 17:13:44 +01:00
|
|
|
|
2016-09-29 12:10:41 +03:00
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
2016-04-16 02:27:58 +02:00
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
tc: adjust network header after 2nd vlan push
<tldr>
skb network header of the single-tagged vlan packet continues to point the
vlan payload (e.g. IP) after second vlan tag is pushed by tc act_vlan. This
causes problem at the dissector which expects double-tagged packet network
header to point to the inner vlan.
The fix is to adjust network header in tcf_act_vlan.c but requires
refactoring of skb_vlan_push function.
</tldr>
Consider the following shell script snippet configuring TC rules on the
veth interface:
ip link add veth0 type veth peer veth1
ip link set veth0 up
ip link set veth1 up
tc qdisc add dev veth0 clsact
tc filter add dev veth0 ingress pref 10 chain 0 flower \
num_of_vlans 2 cvlan_ethtype 0x800 action goto chain 5
tc filter add dev veth0 ingress pref 20 chain 0 flower \
num_of_vlans 1 action vlan push id 100 \
protocol 0x8100 action goto chain 5
tc filter add dev veth0 ingress pref 30 chain 5 flower \
num_of_vlans 2 cvlan_ethtype 0x800 action simple sdata "success"
Sending double-tagged vlan packet with the IP payload inside:
cat <<ENDS | text2pcap - - | tcpreplay -i veth1 -
0000 00 00 00 00 00 11 00 00 00 00 00 22 81 00 00 64 ..........."...d
0010 81 00 00 14 08 00 45 04 00 26 04 d2 00 00 7f 11 ......E..&......
0020 18 ef 0a 00 00 01 14 00 00 02 00 00 00 00 00 12 ................
0030 e1 c7 00 00 00 00 00 00 00 00 00 00 ............
ENDS
will match rule 10, goto rule 30 in chain 5 and correctly emit "success" to
the dmesg.
OTOH, sending single-tagged vlan packet:
cat <<ENDS | text2pcap - - | tcpreplay -i veth1 -
0000 00 00 00 00 00 11 00 00 00 00 00 22 81 00 00 14 ..........."....
0010 08 00 45 04 00 2a 04 d2 00 00 7f 11 18 eb 0a 00 ..E..*..........
0020 00 01 14 00 00 02 00 00 00 00 00 16 e1 bf 00 00 ................
0030 00 00 00 00 00 00 00 00 00 00 00 00 ............
ENDS
will match rule 20, will push the second vlan tag but will *not* match
rule 30. IOW, the match at rule 30 fails if the second vlan was freshly
pushed by the kernel.
Lets look at __skb_flow_dissect working on the double-tagged vlan packet.
Here is the relevant code from around net/core/flow_dissector.c:1277
copy-pasted here for convenience:
if (dissector_vlan == FLOW_DISSECTOR_KEY_MAX &&
skb && skb_vlan_tag_present(skb)) {
proto = skb->protocol;
} else {
vlan = __skb_header_pointer(skb, nhoff, sizeof(_vlan),
data, hlen, &_vlan);
if (!vlan) {
fdret = FLOW_DISSECT_RET_OUT_BAD;
break;
}
proto = vlan->h_vlan_encapsulated_proto;
nhoff += sizeof(*vlan);
}
The "else" clause above gets the protocol of the encapsulated packet from
the skb data at the network header location. printk debugging has showed
that in the good double-tagged packet case proto is
htons(0x800 == ETH_P_IP) as expected. However in the single-tagged packet
case proto is garbage leading to the failure to match tc filter 30.
proto is being set from the skb header pointed by nhoff parameter which is
defined at the beginning of __skb_flow_dissect
(net/core/flow_dissector.c:1055 in the current version):
nhoff = skb_network_offset(skb);
Therefore the culprit seems to be that the skb network offset is different
between double-tagged packet received from the interface and single-tagged
packet having its vlan tag pushed by TC.
Lets look at the interesting points of the lifetime of the single/double
tagged packets as they traverse our packet flow.
Both of them will start at __netif_receive_skb_core where the first vlan
tag will be stripped:
if (eth_type_vlan(skb->protocol)) {
skb = skb_vlan_untag(skb);
if (unlikely(!skb))
goto out;
}
At this stage in double-tagged case skb->data points to the second vlan tag
while in single-tagged case skb->data points to the network (eg. IP)
header.
Looking at TC vlan push action (net/sched/act_vlan.c) we have the following
code at tcf_vlan_act (interesting points are in square brackets):
if (skb_at_tc_ingress(skb))
[1] skb_push_rcsum(skb, skb->mac_len);
....
case TCA_VLAN_ACT_PUSH:
err = skb_vlan_push(skb, p->tcfv_push_proto, p->tcfv_push_vid |
(p->tcfv_push_prio << VLAN_PRIO_SHIFT),
0);
if (err)
goto drop;
break;
....
out:
if (skb_at_tc_ingress(skb))
[3] skb_pull_rcsum(skb, skb->mac_len);
And skb_vlan_push (net/core/skbuff.c:6204) function does:
err = __vlan_insert_tag(skb, skb->vlan_proto,
skb_vlan_tag_get(skb));
if (err)
return err;
skb->protocol = skb->vlan_proto;
[2] skb->mac_len += VLAN_HLEN;
in the case of pushing the second tag. Lets look at what happens with
skb->data of the single-tagged packet at each of the above points:
1. As a result of the skb_push_rcsum, skb->data is moved back to the start
of the packet.
2. First VLAN tag is moved from the skb into packet buffer, skb->mac_len is
incremented, skb->data still points to the start of the packet.
3. As a result of the skb_pull_rcsum, skb->data is moved forward by the
modified skb->mac_len, thus pointing to the network header again.
Then __skb_flow_dissect will get confused by having double-tagged vlan
packet with the skb->data at the network header.
The solution for the bug is to preserve "skb->data at second vlan header"
semantics in the skb_vlan_push function. We do this by manipulating
skb->network_header rather than skb->mac_len. skb_vlan_push callers are
updated to do skb_reset_mac_len.
Signed-off-by: Boris Sukholitko <boris.sukholitko@broadcom.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2024-08-22 13:35:08 +03:00
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
2016-02-20 00:29:30 +01:00
|
|
|
|
2014-11-19 14:05:02 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-10-03 00:44:28 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-07 15:01:54 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-12-21 08:50:01 +05:30
|
|
|
|
|
|
|
|
|
2019-07-07 15:01:54 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-10-12 13:55:07 +02:00
|
|
|
|
2019-12-21 08:50:01 +05:30
|
|
|
|
|
|
|
|
|
2019-07-07 15:01:54 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-10-12 13:55:07 +02:00
|
|
|
|
net: Fixed updating of ethertype in skb_mpls_push()
The skb_mpls_push was not updating ethertype of an ethernet packet if
the packet was originally received from a non ARPHRD_ETHER device.
In the below OVS data path flow, since the device corresponding to
port 7 is an l3 device (ARPHRD_NONE) the skb_mpls_push function does
not update the ethertype of the packet even though the previous
push_eth action had added an ethernet header to the packet.
recirc_id(0),in_port(7),eth_type(0x0800),ipv4(tos=0/0xfc,ttl=64,frag=no),
actions:push_eth(src=00:00:00:00:00:00,dst=00:00:00:00:00:00),
push_mpls(label=13,tc=0,ttl=64,bos=1,eth_type=0x8847),4
Fixes: 8822e270d697 ("net: core: move push MPLS functionality from OvS to core helper")
Signed-off-by: Martin Varghese <martin.varghese@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2019-12-05 05:57:22 +05:30
|
|
|
|
2019-07-07 15:01:54 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-12-21 08:50:01 +05:30
|
|
|
|
2019-07-07 15:01:54 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-10-12 13:55:07 +02:00
|
|
|
|
2019-07-07 15:01:54 +01:00
|
|
|
|
2019-10-12 13:55:07 +02:00
|
|
|
|
2019-12-21 08:50:01 +05:30
|
|
|
|
2019-07-07 15:01:54 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-10-02 21:53:08 +02:00
|
|
|
|
2019-07-07 15:01:54 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-07 15:01:55 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-10-12 13:55:07 +02:00
|
|
|
|
2019-12-21 08:50:23 +05:30
|
|
|
|
2019-07-07 15:01:55 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Fixed updating of ethertype in function skb_mpls_pop
The skb_mpls_pop was not updating ethertype of an ethernet packet if the
packet was originally received from a non ARPHRD_ETHER device.
In the below OVS data path flow, since the device corresponding to port 7
is an l3 device (ARPHRD_NONE) the skb_mpls_pop function does not update
the ethertype of the packet even though the previous push_eth action had
added an ethernet header to the packet.
recirc_id(0),in_port(7),eth_type(0x8847),
mpls(label=12/0xfffff,tc=0/0,ttl=0/0x0,bos=1/1),
actions:push_eth(src=00:00:00:00:00:00,dst=00:00:00:00:00:00),
pop_mpls(eth_type=0x800),4
Fixes: ed246cee09b9 ("net: core: move pop MPLS functionality from OvS to core helper")
Signed-off-by: Martin Varghese <martin.varghese@nokia.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2019-12-02 10:49:51 +05:30
|
|
|
|
|
|
|
|
|
2019-07-07 15:01:55 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-10-12 13:55:06 +02:00
|
|
|
|
2019-07-07 15:01:55 +01:00
|
|
|
|
2019-10-12 13:55:07 +02:00
|
|
|
|
2019-07-07 15:01:55 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-10-12 13:55:07 +02:00
|
|
|
|
2019-07-07 15:01:55 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-10-12 13:55:07 +02:00
|
|
|
|
2019-07-07 15:01:55 +01:00
|
|
|
|
2020-10-02 21:53:08 +02:00
|
|
|
|
2019-07-07 15:01:55 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-07 15:01:56 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-07 15:01:57 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-12-03 10:58:21 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-07 15:01:57 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-09-17 04:49:49 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-10-09 12:58:08 +09:00
|
|
|
|
|
|
|
|
|
2023-08-01 20:52:51 +00:00
|
|
|
|
2014-10-09 12:58:08 +09:00
|
|
|
|
|
|
|
|
|
2014-09-17 04:49:49 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-01 20:52:51 +00:00
|
|
|
|
2014-09-17 04:49:49 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-01 20:52:51 +00:00
|
|
|
|
2014-09-17 04:49:49 -07:00
|
|
|
|
|
|
|
|
|
2023-08-01 20:52:51 +00:00
|
|
|
|
2014-09-17 04:49:49 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net, skbuff: do not prefer skb allocation fails early
Commit dcda9b04713c ("mm, tree wide: replace __GFP_REPEAT by
__GFP_RETRY_MAYFAIL with more useful semantic") replaced __GFP_REPEAT in
alloc_skb_with_frags() with __GFP_RETRY_MAYFAIL when the allocation may
directly reclaim.
The previous behavior would require reclaim up to 1 << order pages for
skb aligned header_len of order > PAGE_ALLOC_COSTLY_ORDER before failing,
otherwise the allocations in alloc_skb() would loop in the page allocator
looking for memory. __GFP_RETRY_MAYFAIL makes both allocations failable
under memory pressure, including for the HEAD allocation.
This can cause, among many other things, write() to fail with ENOTCONN
during RPC when under memory pressure.
These allocations should succeed as they did previous to dcda9b04713c
even if it requires calling the oom killer and additional looping in the
page allocator to find memory. There is no way to specify the previous
behavior of __GFP_REPEAT, but it's unlikely to be necessary since the
previous behavior only guaranteed that 1 << order pages would be reclaimed
before failing for order > PAGE_ALLOC_COSTLY_ORDER. That reclaim is not
guaranteed to be contiguous memory, so repeating for such large orders is
usually not beneficial.
Removing the setting of __GFP_RETRY_MAYFAIL to restore the previous
behavior, specifically not allowing alloc_skb() to fail for small orders
and oom kill if necessary rather than allowing RPCs to fail.
Fixes: dcda9b04713c ("mm, tree wide: replace __GFP_REPEAT by __GFP_RETRY_MAYFAIL with more useful semantic")
Signed-off-by: David Rientjes <rientjes@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2019-01-02 13:01:43 -08:00
|
|
|
|
2014-09-17 04:49:49 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-08-01 20:52:51 +00:00
|
|
|
|
2025-09-22 15:19:57 -04:00
|
|
|
|
2014-09-17 04:49:49 -07:00
|
|
|
|
2023-08-01 20:52:51 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-09-17 04:49:49 -07:00
|
|
|
|
|
|
|
|
|
2023-08-01 20:52:51 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-09-17 04:49:49 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-10-25 15:39:35 -07:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-10-25 15:39:35 -07:00
|
|
|
|
2023-02-06 17:31:02 +00:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
2022-10-25 15:39:35 -07:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-05-28 19:43:53 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-10 11:47:32 +08:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
2026-05-26 11:12:39 +07:00
|
|
|
|
|
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-02-21 19:21:12 -08:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-02 15:38:12 +02:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-02-20 07:40:52 -08:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-10-25 15:39:35 -07:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-10-25 15:39:35 -07:00
|
|
|
|
2023-02-06 17:31:02 +00:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
2022-10-25 15:39:35 -07:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
2026-05-28 19:43:53 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
2020-08-15 04:48:53 -04:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
2026-04-10 11:47:32 +08:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-30 07:40:33 -07:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-08-15 04:46:41 -04:00
|
|
|
|
2025-07-02 15:38:12 +02:00
|
|
|
|
2020-08-15 04:46:41 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-10 11:47:32 +08:00
|
|
|
|
2020-08-15 04:46:41 -04:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
2026-05-26 11:12:39 +07:00
|
|
|
|
|
|
|
|
|
2024-03-29 17:55:06 +01:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-02-21 19:21:12 -08:00
|
|
|
|
skbuff: Add pskb_extract() helper function
A pattern of skb usage seen in modules such as RDS-TCP is to
extract `to_copy' bytes from the received TCP segment, starting
at some offset `off' into a new skb `clone'. This is done in
the ->data_ready callback, where the clone skb is queued up for rx on
the PF_RDS socket, while the parent TCP segment is returned unchanged
back to the TCP engine.
The existing code uses the sequence
clone = skb_clone(..);
pskb_pull(clone, off, ..);
pskb_trim(clone, to_copy, ..);
with the intention of discarding the first `off' bytes. However,
skb_clone() + pskb_pull() implies pksb_expand_head(), which ends
up doing a redundant memcpy of bytes that will then get discarded
in __pskb_pull_tail().
To avoid this inefficiency, this commit adds pskb_extract() that
creates the clone, and memcpy's only the relevant header/frag/frag_list
to the start of `clone'. pskb_trim() is then invoked to trim clone
down to the requested to_copy bytes.
Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-04-22 18:36:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
udp: under rx pressure, try to condense skbs
Under UDP flood, many softirq producers try to add packets to
UDP receive queue, and one user thread is burning one cpu trying
to dequeue packets as fast as possible.
Two parts of the per packet cost are :
- copying payload from kernel space to user space,
- freeing memory pieces associated with skb.
If socket is under pressure, softirq handler(s) can try to pull in
skb->head the payload of the packet if it fits.
Meaning the softirq handler(s) can free/reuse the page fragment
immediately, instead of letting udp_recvmsg() do this hundreds of usec
later, possibly from another node.
Additional gains :
- We reduce skb->truesize and thus can store more packets per SO_RCVBUF
- We avoid cache line misses at copyout() time and consume_skb() time,
and avoid one put_page() with potential alien freeing on NUMA hosts.
This comes at the cost of a copy, bounded to available tail room, which
is usually small. (We might have to fix GRO_MAX_HEAD which looks bigger
than necessary)
This patch gave me about 5 % increase in throughput in my tests.
skb_condense() helper could probably used in other contexts.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-12-07 09:19:33 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-12-09 08:02:05 -08:00
|
|
|
|
|
|
|
|
|
2024-09-10 17:14:52 +00:00
|
|
|
|
2016-12-09 08:02:05 -08:00
|
|
|
|
udp: under rx pressure, try to condense skbs
Under UDP flood, many softirq producers try to add packets to
UDP receive queue, and one user thread is burning one cpu trying
to dequeue packets as fast as possible.
Two parts of the per packet cost are :
- copying payload from kernel space to user space,
- freeing memory pieces associated with skb.
If socket is under pressure, softirq handler(s) can try to pull in
skb->head the payload of the packet if it fits.
Meaning the softirq handler(s) can free/reuse the page fragment
immediately, instead of letting udp_recvmsg() do this hundreds of usec
later, possibly from another node.
Additional gains :
- We reduce skb->truesize and thus can store more packets per SO_RCVBUF
- We avoid cache line misses at copyout() time and consume_skb() time,
and avoid one put_page() with potential alien freeing on NUMA hosts.
This comes at the cost of a copy, bounded to available tail room, which
is usually small. (We might have to fix GRO_MAX_HEAD which looks bigger
than necessary)
This patch gave me about 5 % increase in throughput in my tests.
skb_condense() helper could probably used in other contexts.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-12-07 09:19:33 -08:00
|
|
|
|
2016-12-09 08:02:05 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
udp: under rx pressure, try to condense skbs
Under UDP flood, many softirq producers try to add packets to
UDP receive queue, and one user thread is burning one cpu trying
to dequeue packets as fast as possible.
Two parts of the per packet cost are :
- copying payload from kernel space to user space,
- freeing memory pieces associated with skb.
If socket is under pressure, softirq handler(s) can try to pull in
skb->head the payload of the packet if it fits.
Meaning the softirq handler(s) can free/reuse the page fragment
immediately, instead of letting udp_recvmsg() do this hundreds of usec
later, possibly from another node.
Additional gains :
- We reduce skb->truesize and thus can store more packets per SO_RCVBUF
- We avoid cache line misses at copyout() time and consume_skb() time,
and avoid one put_page() with potential alien freeing on NUMA hosts.
This comes at the cost of a copy, bounded to available tail room, which
is usually small. (We might have to fix GRO_MAX_HEAD which looks bigger
than necessary)
This patch gave me about 5 % increase in throughput in my tests.
skb_condense() helper could probably used in other contexts.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-12-07 09:19:33 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-05-07 10:06:13 +01:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-01-09 07:59:24 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-16 10:46:23 +02:00
|
|
|
|
|
|
|
|
|
2020-01-09 07:59:24 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-16 10:46:23 +02:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
2020-05-16 10:46:23 +02:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-18 17:15:27 +01:00
|
|
|
|
|
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-18 17:15:27 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-02-19 17:51:54 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-18 17:15:27 +01:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-01-09 07:59:24 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-27 15:54:13 -07:00
|
|
|
|
2020-01-09 07:59:24 -08:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-18 17:15:27 +01:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-21 19:03:15 +01:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
2018-12-21 19:03:13 +01:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-16 10:46:23 +02:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-07-28 18:24:01 +02:00
|
|
|
|
2018-12-21 19:03:15 +01:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-18 17:15:27 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-29 11:01:44 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-18 17:15:27 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-18 17:15:27 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-29 11:01:44 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-18 17:15:27 +01:00
|
|
|
|
sk_buff: add skb extension infrastructure
This adds an optional extension infrastructure, with ispec (xfrm) and
bridge netfilter as first users.
objdiff shows no changes if kernel is built without xfrm and br_netfilter
support.
The third (planned future) user is Multipath TCP which is still
out-of-tree.
MPTCP needs to map logical mptcp sequence numbers to the tcp sequence
numbers used by individual subflows.
This DSS mapping is read/written from tcp option space on receive and
written to tcp option space on transmitted tcp packets that are part of
and MPTCP connection.
Extending skb_shared_info or adding a private data field to skb fclones
doesn't work for incoming skb, so a different DSS propagation method would
be required for the receive side.
mptcp has same requirements as secpath/bridge netfilter:
1. extension memory is released when the sk_buff is free'd.
2. data is shared after cloning an skb (clone inherits extension)
3. adding extension to an skb will COW the extension buffer if needed.
The "MPTCP upstreaming" effort adds SKB_EXT_MPTCP extension to store the
mapping for tx and rx processing.
Two new members are added to sk_buff:
1. 'active_extensions' byte (filling a hole), telling which extensions
are available for this skb.
This has two purposes.
a) avoids the need to initialize the pointer.
b) allows to "delete" an extension by clearing its bit
value in ->active_extensions.
While it would be possible to store the active_extensions byte
in the extension struct instead of sk_buff, there is one problem
with this:
When an extension has to be disabled, we can always clear the
bit in skb->active_extensions. But in case it would be stored in the
extension buffer itself, we might have to COW it first, if
we are dealing with a cloned skb. On kmalloc failure we would
be unable to turn an extension off.
2. extension pointer, located at the end of the sk_buff.
If the active_extensions byte is 0, the pointer is undefined,
it is not initialized on skb allocation.
This adds extra code to skb clone and free paths (to deal with
refcount/free of extension area) but this replaces similar code that
manages skb->nf_bridge and skb->sp structs in the followup patches of
the series.
It is possible to add support for extensions that are not preseved on
clones/copies.
To do this, it would be needed to define a bitmask of all extensions that
need copy/cow semantics, and change __skb_ext_copy() to check
->active_extensions & SKB_EXT_PRESERVE_ON_CLONE, then just set
->active_extensions to 0 on the new clone.
This isn't done here because all extensions that get added here
need the copy/cow semantics.
v2:
Allocate entire extension space using kmem_cache.
Upside is that this allows better tracking of used memory,
downside is that we will allocate more space than strictly needed in
most cases (its unlikely that all extensions are active/needed at same
time for same skb).
The allocated memory (except the small extension header) is not cleared,
so no additonal overhead aside from memory usage.
Avoid atomic_dec_and_test operation on skb_ext_put()
by using similar trick as kfree_skbmem() does with fclone_ref:
If recount is 1, there is no concurrent user and we can free right away.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2018-12-18 17:15:16 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
net: generalize skb freeing deferral to per-cpu lists
Logic added in commit f35f821935d8 ("tcp: defer skb freeing after socket
lock is released") helped bulk TCP flows to move the cost of skbs
frees outside of critical section where socket lock was held.
But for RPC traffic, or hosts with RFS enabled, the solution is far from
being ideal.
For RPC traffic, recvmsg() has to return to user space right after
skb payload has been consumed, meaning that BH handler has no chance
to pick the skb before recvmsg() thread. This issue is more visible
with BIG TCP, as more RPC fit one skb.
For RFS, even if BH handler picks the skbs, they are still picked
from the cpu on which user thread is running.
Ideally, it is better to free the skbs (and associated page frags)
on the cpu that originally allocated them.
This patch removes the per socket anchor (sk->defer_list) and
instead uses a per-cpu list, which will hold more skbs per round.
This new per-cpu list is drained at the end of net_action_rx(),
after incoming packets have been processed, to lower latencies.
In normal conditions, skbs are added to the per-cpu list with
no further action. In the (unlikely) cases where the cpu does not
run net_action_rx() handler fast enough, we use an IPI to raise
NET_RX_SOFTIRQ on the remote cpu.
Also, we do not bother draining the per-cpu list from dev_cpu_dead()
This is because skbs in this list have no requirement on how fast
they should be freed.
Note that we can add in the future a small per-cpu cache
if we see any contention on sd->defer_lock.
Tested on a pair of hosts with 100Gbit NIC, RFS enabled,
and /proc/sys/net/ipv4/tcp_rmem[2] tuned to 16MB to work around
page recycling strategy used by NIC driver (its page pool capacity
being too small compared to number of skbs/pages held in sockets
receive queues)
Note that this tuning was only done to demonstrate worse
conditions for skb freeing for this particular test.
These conditions can happen in more general production workload.
10 runs of one TCP_STREAM flow
Before:
Average throughput: 49685 Mbit.
Kernel profiles on cpu running user thread recvmsg() show high cost for
skb freeing related functions (*)
57.81% [kernel] [k] copy_user_enhanced_fast_string
(*) 12.87% [kernel] [k] skb_release_data
(*) 4.25% [kernel] [k] __free_one_page
(*) 3.57% [kernel] [k] __list_del_entry_valid
1.85% [kernel] [k] __netif_receive_skb_core
1.60% [kernel] [k] __skb_datagram_iter
(*) 1.59% [kernel] [k] free_unref_page_commit
(*) 1.16% [kernel] [k] __slab_free
1.16% [kernel] [k] _copy_to_iter
(*) 1.01% [kernel] [k] kfree
(*) 0.88% [kernel] [k] free_unref_page
0.57% [kernel] [k] ip6_rcv_core
0.55% [kernel] [k] ip6t_do_table
0.54% [kernel] [k] flush_smp_call_function_queue
(*) 0.54% [kernel] [k] free_pcppages_bulk
0.51% [kernel] [k] llist_reverse_order
0.38% [kernel] [k] process_backlog
(*) 0.38% [kernel] [k] free_pcp_prepare
0.37% [kernel] [k] tcp_recvmsg_locked
(*) 0.37% [kernel] [k] __list_add_valid
0.34% [kernel] [k] sock_rfree
0.34% [kernel] [k] _raw_spin_lock_irq
(*) 0.33% [kernel] [k] __page_cache_release
0.33% [kernel] [k] tcp_v6_rcv
(*) 0.33% [kernel] [k] __put_page
(*) 0.29% [kernel] [k] __mod_zone_page_state
0.27% [kernel] [k] _raw_spin_lock
After patch:
Average throughput: 73076 Mbit.
Kernel profiles on cpu running user thread recvmsg() looks better:
81.35% [kernel] [k] copy_user_enhanced_fast_string
1.95% [kernel] [k] _copy_to_iter
1.95% [kernel] [k] __skb_datagram_iter
1.27% [kernel] [k] __netif_receive_skb_core
1.03% [kernel] [k] ip6t_do_table
0.60% [kernel] [k] sock_rfree
0.50% [kernel] [k] tcp_v6_rcv
0.47% [kernel] [k] ip6_rcv_core
0.45% [kernel] [k] read_tsc
0.44% [kernel] [k] _raw_spin_lock_irqsave
0.37% [kernel] [k] _raw_spin_lock
0.37% [kernel] [k] native_irq_return_iret
0.33% [kernel] [k] __inet6_lookup_established
0.31% [kernel] [k] ip6_protocol_deliver_rcu
0.29% [kernel] [k] tcp_rcv_established
0.29% [kernel] [k] llist_reverse_order
v2: kdoc issue (kernel bots)
do not defer if (alloc_cpu == smp_processor_id()) (Paolo)
replace the sk_buff_head with a single-linked list (Jakub)
add a READ_ONCE()/WRITE_ONCE() for the lockless read of sd->defer_list
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Link: https://lore.kernel.org/r/20220422201237.416238-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-04-22 13:12:37 -07:00
|
|
|
|
2024-04-10 02:28:09 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-04-10 02:28:10 +01:00
|
|
|
|
2024-04-10 02:28:09 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-03-11 19:13:40 +00:00
|
|
|
|
|
|
|
|
|
net: generalize skb freeing deferral to per-cpu lists
Logic added in commit f35f821935d8 ("tcp: defer skb freeing after socket
lock is released") helped bulk TCP flows to move the cost of skbs
frees outside of critical section where socket lock was held.
But for RPC traffic, or hosts with RFS enabled, the solution is far from
being ideal.
For RPC traffic, recvmsg() has to return to user space right after
skb payload has been consumed, meaning that BH handler has no chance
to pick the skb before recvmsg() thread. This issue is more visible
with BIG TCP, as more RPC fit one skb.
For RFS, even if BH handler picks the skbs, they are still picked
from the cpu on which user thread is running.
Ideally, it is better to free the skbs (and associated page frags)
on the cpu that originally allocated them.
This patch removes the per socket anchor (sk->defer_list) and
instead uses a per-cpu list, which will hold more skbs per round.
This new per-cpu list is drained at the end of net_action_rx(),
after incoming packets have been processed, to lower latencies.
In normal conditions, skbs are added to the per-cpu list with
no further action. In the (unlikely) cases where the cpu does not
run net_action_rx() handler fast enough, we use an IPI to raise
NET_RX_SOFTIRQ on the remote cpu.
Also, we do not bother draining the per-cpu list from dev_cpu_dead()
This is because skbs in this list have no requirement on how fast
they should be freed.
Note that we can add in the future a small per-cpu cache
if we see any contention on sd->defer_lock.
Tested on a pair of hosts with 100Gbit NIC, RFS enabled,
and /proc/sys/net/ipv4/tcp_rmem[2] tuned to 16MB to work around
page recycling strategy used by NIC driver (its page pool capacity
being too small compared to number of skbs/pages held in sockets
receive queues)
Note that this tuning was only done to demonstrate worse
conditions for skb freeing for this particular test.
These conditions can happen in more general production workload.
10 runs of one TCP_STREAM flow
Before:
Average throughput: 49685 Mbit.
Kernel profiles on cpu running user thread recvmsg() show high cost for
skb freeing related functions (*)
57.81% [kernel] [k] copy_user_enhanced_fast_string
(*) 12.87% [kernel] [k] skb_release_data
(*) 4.25% [kernel] [k] __free_one_page
(*) 3.57% [kernel] [k] __list_del_entry_valid
1.85% [kernel] [k] __netif_receive_skb_core
1.60% [kernel] [k] __skb_datagram_iter
(*) 1.59% [kernel] [k] free_unref_page_commit
(*) 1.16% [kernel] [k] __slab_free
1.16% [kernel] [k] _copy_to_iter
(*) 1.01% [kernel] [k] kfree
(*) 0.88% [kernel] [k] free_unref_page
0.57% [kernel] [k] ip6_rcv_core
0.55% [kernel] [k] ip6t_do_table
0.54% [kernel] [k] flush_smp_call_function_queue
(*) 0.54% [kernel] [k] free_pcppages_bulk
0.51% [kernel] [k] llist_reverse_order
0.38% [kernel] [k] process_backlog
(*) 0.38% [kernel] [k] free_pcp_prepare
0.37% [kernel] [k] tcp_recvmsg_locked
(*) 0.37% [kernel] [k] __list_add_valid
0.34% [kernel] [k] sock_rfree
0.34% [kernel] [k] _raw_spin_lock_irq
(*) 0.33% [kernel] [k] __page_cache_release
0.33% [kernel] [k] tcp_v6_rcv
(*) 0.33% [kernel] [k] __put_page
(*) 0.29% [kernel] [k] __mod_zone_page_state
0.27% [kernel] [k] _raw_spin_lock
After patch:
Average throughput: 73076 Mbit.
Kernel profiles on cpu running user thread recvmsg() looks better:
81.35% [kernel] [k] copy_user_enhanced_fast_string
1.95% [kernel] [k] _copy_to_iter
1.95% [kernel] [k] __skb_datagram_iter
1.27% [kernel] [k] __netif_receive_skb_core
1.03% [kernel] [k] ip6t_do_table
0.60% [kernel] [k] sock_rfree
0.50% [kernel] [k] tcp_v6_rcv
0.47% [kernel] [k] ip6_rcv_core
0.45% [kernel] [k] read_tsc
0.44% [kernel] [k] _raw_spin_lock_irqsave
0.37% [kernel] [k] _raw_spin_lock
0.37% [kernel] [k] native_irq_return_iret
0.33% [kernel] [k] __inet6_lookup_established
0.31% [kernel] [k] ip6_protocol_deliver_rcu
0.29% [kernel] [k] tcp_rcv_established
0.29% [kernel] [k] llist_reverse_order
v2: kdoc issue (kernel bots)
do not defer if (alloc_cpu == smp_processor_id()) (Paolo)
replace the sk_buff_head with a single-linked list (Jakub)
add a READ_ONCE()/WRITE_ONCE() for the lockless read of sd->defer_list
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Link: https://lore.kernel.org/r/20220422201237.416238-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-04-22 13:12:37 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-28 08:49:34 +00:00
|
|
|
|
2025-09-28 08:49:33 +00:00
|
|
|
|
2022-05-15 21:24:55 -07:00
|
|
|
|
net: generalize skb freeing deferral to per-cpu lists
Logic added in commit f35f821935d8 ("tcp: defer skb freeing after socket
lock is released") helped bulk TCP flows to move the cost of skbs
frees outside of critical section where socket lock was held.
But for RPC traffic, or hosts with RFS enabled, the solution is far from
being ideal.
For RPC traffic, recvmsg() has to return to user space right after
skb payload has been consumed, meaning that BH handler has no chance
to pick the skb before recvmsg() thread. This issue is more visible
with BIG TCP, as more RPC fit one skb.
For RFS, even if BH handler picks the skbs, they are still picked
from the cpu on which user thread is running.
Ideally, it is better to free the skbs (and associated page frags)
on the cpu that originally allocated them.
This patch removes the per socket anchor (sk->defer_list) and
instead uses a per-cpu list, which will hold more skbs per round.
This new per-cpu list is drained at the end of net_action_rx(),
after incoming packets have been processed, to lower latencies.
In normal conditions, skbs are added to the per-cpu list with
no further action. In the (unlikely) cases where the cpu does not
run net_action_rx() handler fast enough, we use an IPI to raise
NET_RX_SOFTIRQ on the remote cpu.
Also, we do not bother draining the per-cpu list from dev_cpu_dead()
This is because skbs in this list have no requirement on how fast
they should be freed.
Note that we can add in the future a small per-cpu cache
if we see any contention on sd->defer_lock.
Tested on a pair of hosts with 100Gbit NIC, RFS enabled,
and /proc/sys/net/ipv4/tcp_rmem[2] tuned to 16MB to work around
page recycling strategy used by NIC driver (its page pool capacity
being too small compared to number of skbs/pages held in sockets
receive queues)
Note that this tuning was only done to demonstrate worse
conditions for skb freeing for this particular test.
These conditions can happen in more general production workload.
10 runs of one TCP_STREAM flow
Before:
Average throughput: 49685 Mbit.
Kernel profiles on cpu running user thread recvmsg() show high cost for
skb freeing related functions (*)
57.81% [kernel] [k] copy_user_enhanced_fast_string
(*) 12.87% [kernel] [k] skb_release_data
(*) 4.25% [kernel] [k] __free_one_page
(*) 3.57% [kernel] [k] __list_del_entry_valid
1.85% [kernel] [k] __netif_receive_skb_core
1.60% [kernel] [k] __skb_datagram_iter
(*) 1.59% [kernel] [k] free_unref_page_commit
(*) 1.16% [kernel] [k] __slab_free
1.16% [kernel] [k] _copy_to_iter
(*) 1.01% [kernel] [k] kfree
(*) 0.88% [kernel] [k] free_unref_page
0.57% [kernel] [k] ip6_rcv_core
0.55% [kernel] [k] ip6t_do_table
0.54% [kernel] [k] flush_smp_call_function_queue
(*) 0.54% [kernel] [k] free_pcppages_bulk
0.51% [kernel] [k] llist_reverse_order
0.38% [kernel] [k] process_backlog
(*) 0.38% [kernel] [k] free_pcp_prepare
0.37% [kernel] [k] tcp_recvmsg_locked
(*) 0.37% [kernel] [k] __list_add_valid
0.34% [kernel] [k] sock_rfree
0.34% [kernel] [k] _raw_spin_lock_irq
(*) 0.33% [kernel] [k] __page_cache_release
0.33% [kernel] [k] tcp_v6_rcv
(*) 0.33% [kernel] [k] __put_page
(*) 0.29% [kernel] [k] __mod_zone_page_state
0.27% [kernel] [k] _raw_spin_lock
After patch:
Average throughput: 73076 Mbit.
Kernel profiles on cpu running user thread recvmsg() looks better:
81.35% [kernel] [k] copy_user_enhanced_fast_string
1.95% [kernel] [k] _copy_to_iter
1.95% [kernel] [k] __skb_datagram_iter
1.27% [kernel] [k] __netif_receive_skb_core
1.03% [kernel] [k] ip6t_do_table
0.60% [kernel] [k] sock_rfree
0.50% [kernel] [k] tcp_v6_rcv
0.47% [kernel] [k] ip6_rcv_core
0.45% [kernel] [k] read_tsc
0.44% [kernel] [k] _raw_spin_lock_irqsave
0.37% [kernel] [k] _raw_spin_lock
0.37% [kernel] [k] native_irq_return_iret
0.33% [kernel] [k] __inet6_lookup_established
0.31% [kernel] [k] ip6_protocol_deliver_rcu
0.29% [kernel] [k] tcp_rcv_established
0.29% [kernel] [k] llist_reverse_order
v2: kdoc issue (kernel bots)
do not defer if (alloc_cpu == smp_processor_id()) (Paolo)
replace the sk_buff_head with a single-linked list (Jakub)
add a READ_ONCE()/WRITE_ONCE() for the lockless read of sd->defer_list
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Link: https://lore.kernel.org/r/20220422201237.416238-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-04-22 13:12:37 -07:00
|
|
|
|
2026-02-16 19:36:53 +00:00
|
|
|
|
net: generalize skb freeing deferral to per-cpu lists
Logic added in commit f35f821935d8 ("tcp: defer skb freeing after socket
lock is released") helped bulk TCP flows to move the cost of skbs
frees outside of critical section where socket lock was held.
But for RPC traffic, or hosts with RFS enabled, the solution is far from
being ideal.
For RPC traffic, recvmsg() has to return to user space right after
skb payload has been consumed, meaning that BH handler has no chance
to pick the skb before recvmsg() thread. This issue is more visible
with BIG TCP, as more RPC fit one skb.
For RFS, even if BH handler picks the skbs, they are still picked
from the cpu on which user thread is running.
Ideally, it is better to free the skbs (and associated page frags)
on the cpu that originally allocated them.
This patch removes the per socket anchor (sk->defer_list) and
instead uses a per-cpu list, which will hold more skbs per round.
This new per-cpu list is drained at the end of net_action_rx(),
after incoming packets have been processed, to lower latencies.
In normal conditions, skbs are added to the per-cpu list with
no further action. In the (unlikely) cases where the cpu does not
run net_action_rx() handler fast enough, we use an IPI to raise
NET_RX_SOFTIRQ on the remote cpu.
Also, we do not bother draining the per-cpu list from dev_cpu_dead()
This is because skbs in this list have no requirement on how fast
they should be freed.
Note that we can add in the future a small per-cpu cache
if we see any contention on sd->defer_lock.
Tested on a pair of hosts with 100Gbit NIC, RFS enabled,
and /proc/sys/net/ipv4/tcp_rmem[2] tuned to 16MB to work around
page recycling strategy used by NIC driver (its page pool capacity
being too small compared to number of skbs/pages held in sockets
receive queues)
Note that this tuning was only done to demonstrate worse
conditions for skb freeing for this particular test.
These conditions can happen in more general production workload.
10 runs of one TCP_STREAM flow
Before:
Average throughput: 49685 Mbit.
Kernel profiles on cpu running user thread recvmsg() show high cost for
skb freeing related functions (*)
57.81% [kernel] [k] copy_user_enhanced_fast_string
(*) 12.87% [kernel] [k] skb_release_data
(*) 4.25% [kernel] [k] __free_one_page
(*) 3.57% [kernel] [k] __list_del_entry_valid
1.85% [kernel] [k] __netif_receive_skb_core
1.60% [kernel] [k] __skb_datagram_iter
(*) 1.59% [kernel] [k] free_unref_page_commit
(*) 1.16% [kernel] [k] __slab_free
1.16% [kernel] [k] _copy_to_iter
(*) 1.01% [kernel] [k] kfree
(*) 0.88% [kernel] [k] free_unref_page
0.57% [kernel] [k] ip6_rcv_core
0.55% [kernel] [k] ip6t_do_table
0.54% [kernel] [k] flush_smp_call_function_queue
(*) 0.54% [kernel] [k] free_pcppages_bulk
0.51% [kernel] [k] llist_reverse_order
0.38% [kernel] [k] process_backlog
(*) 0.38% [kernel] [k] free_pcp_prepare
0.37% [kernel] [k] tcp_recvmsg_locked
(*) 0.37% [kernel] [k] __list_add_valid
0.34% [kernel] [k] sock_rfree
0.34% [kernel] [k] _raw_spin_lock_irq
(*) 0.33% [kernel] [k] __page_cache_release
0.33% [kernel] [k] tcp_v6_rcv
(*) 0.33% [kernel] [k] __put_page
(*) 0.29% [kernel] [k] __mod_zone_page_state
0.27% [kernel] [k] _raw_spin_lock
After patch:
Average throughput: 73076 Mbit.
Kernel profiles on cpu running user thread recvmsg() looks better:
81.35% [kernel] [k] copy_user_enhanced_fast_string
1.95% [kernel] [k] _copy_to_iter
1.95% [kernel] [k] __skb_datagram_iter
1.27% [kernel] [k] __netif_receive_skb_core
1.03% [kernel] [k] ip6t_do_table
0.60% [kernel] [k] sock_rfree
0.50% [kernel] [k] tcp_v6_rcv
0.47% [kernel] [k] ip6_rcv_core
0.45% [kernel] [k] read_tsc
0.44% [kernel] [k] _raw_spin_lock_irqsave
0.37% [kernel] [k] _raw_spin_lock
0.37% [kernel] [k] native_irq_return_iret
0.33% [kernel] [k] __inet6_lookup_established
0.31% [kernel] [k] ip6_protocol_deliver_rcu
0.29% [kernel] [k] tcp_rcv_established
0.29% [kernel] [k] llist_reverse_order
v2: kdoc issue (kernel bots)
do not defer if (alloc_cpu == smp_processor_id()) (Paolo)
replace the sk_buff_head with a single-linked list (Jakub)
add a READ_ONCE()/WRITE_ONCE() for the lockless read of sd->defer_list
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Link: https://lore.kernel.org/r/20220422201237.416238-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-04-22 13:12:37 -07:00
|
|
|
|
2026-03-11 19:13:40 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-02-16 19:36:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-04-12 11:07:18 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-04-10 02:28:09 +01:00
|
|
|
|
net: generalize skb freeing deferral to per-cpu lists
Logic added in commit f35f821935d8 ("tcp: defer skb freeing after socket
lock is released") helped bulk TCP flows to move the cost of skbs
frees outside of critical section where socket lock was held.
But for RPC traffic, or hosts with RFS enabled, the solution is far from
being ideal.
For RPC traffic, recvmsg() has to return to user space right after
skb payload has been consumed, meaning that BH handler has no chance
to pick the skb before recvmsg() thread. This issue is more visible
with BIG TCP, as more RPC fit one skb.
For RFS, even if BH handler picks the skbs, they are still picked
from the cpu on which user thread is running.
Ideally, it is better to free the skbs (and associated page frags)
on the cpu that originally allocated them.
This patch removes the per socket anchor (sk->defer_list) and
instead uses a per-cpu list, which will hold more skbs per round.
This new per-cpu list is drained at the end of net_action_rx(),
after incoming packets have been processed, to lower latencies.
In normal conditions, skbs are added to the per-cpu list with
no further action. In the (unlikely) cases where the cpu does not
run net_action_rx() handler fast enough, we use an IPI to raise
NET_RX_SOFTIRQ on the remote cpu.
Also, we do not bother draining the per-cpu list from dev_cpu_dead()
This is because skbs in this list have no requirement on how fast
they should be freed.
Note that we can add in the future a small per-cpu cache
if we see any contention on sd->defer_lock.
Tested on a pair of hosts with 100Gbit NIC, RFS enabled,
and /proc/sys/net/ipv4/tcp_rmem[2] tuned to 16MB to work around
page recycling strategy used by NIC driver (its page pool capacity
being too small compared to number of skbs/pages held in sockets
receive queues)
Note that this tuning was only done to demonstrate worse
conditions for skb freeing for this particular test.
These conditions can happen in more general production workload.
10 runs of one TCP_STREAM flow
Before:
Average throughput: 49685 Mbit.
Kernel profiles on cpu running user thread recvmsg() show high cost for
skb freeing related functions (*)
57.81% [kernel] [k] copy_user_enhanced_fast_string
(*) 12.87% [kernel] [k] skb_release_data
(*) 4.25% [kernel] [k] __free_one_page
(*) 3.57% [kernel] [k] __list_del_entry_valid
1.85% [kernel] [k] __netif_receive_skb_core
1.60% [kernel] [k] __skb_datagram_iter
(*) 1.59% [kernel] [k] free_unref_page_commit
(*) 1.16% [kernel] [k] __slab_free
1.16% [kernel] [k] _copy_to_iter
(*) 1.01% [kernel] [k] kfree
(*) 0.88% [kernel] [k] free_unref_page
0.57% [kernel] [k] ip6_rcv_core
0.55% [kernel] [k] ip6t_do_table
0.54% [kernel] [k] flush_smp_call_function_queue
(*) 0.54% [kernel] [k] free_pcppages_bulk
0.51% [kernel] [k] llist_reverse_order
0.38% [kernel] [k] process_backlog
(*) 0.38% [kernel] [k] free_pcp_prepare
0.37% [kernel] [k] tcp_recvmsg_locked
(*) 0.37% [kernel] [k] __list_add_valid
0.34% [kernel] [k] sock_rfree
0.34% [kernel] [k] _raw_spin_lock_irq
(*) 0.33% [kernel] [k] __page_cache_release
0.33% [kernel] [k] tcp_v6_rcv
(*) 0.33% [kernel] [k] __put_page
(*) 0.29% [kernel] [k] __mod_zone_page_state
0.27% [kernel] [k] _raw_spin_lock
After patch:
Average throughput: 73076 Mbit.
Kernel profiles on cpu running user thread recvmsg() looks better:
81.35% [kernel] [k] copy_user_enhanced_fast_string
1.95% [kernel] [k] _copy_to_iter
1.95% [kernel] [k] __skb_datagram_iter
1.27% [kernel] [k] __netif_receive_skb_core
1.03% [kernel] [k] ip6t_do_table
0.60% [kernel] [k] sock_rfree
0.50% [kernel] [k] tcp_v6_rcv
0.47% [kernel] [k] ip6_rcv_core
0.45% [kernel] [k] read_tsc
0.44% [kernel] [k] _raw_spin_lock_irqsave
0.37% [kernel] [k] _raw_spin_lock
0.37% [kernel] [k] native_irq_return_iret
0.33% [kernel] [k] __inet6_lookup_established
0.31% [kernel] [k] ip6_protocol_deliver_rcu
0.29% [kernel] [k] tcp_rcv_established
0.29% [kernel] [k] llist_reverse_order
v2: kdoc issue (kernel bots)
do not defer if (alloc_cpu == smp_processor_id()) (Paolo)
replace the sk_buff_head with a single-linked list (Jakub)
add a READ_ONCE()/WRITE_ONCE() for the lockless read of sd->defer_list
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Link: https://lore.kernel.org/r/20220422201237.416238-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-04-22 13:12:37 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-04-21 09:43:53 +00:00
|
|
|
|
|
|
|
|
|
2025-10-15 05:27:15 +00:00
|
|
|
|
2023-04-21 09:43:53 +00:00
|
|
|
|
2025-09-28 08:49:34 +00:00
|
|
|
|
|
|
|
|
|
2024-04-29 13:40:22 +00:00
|
|
|
|
2025-09-28 08:49:34 +00:00
|
|
|
|
2025-09-28 08:49:33 +00:00
|
|
|
|
|
|
|
|
|
2022-05-15 21:24:55 -07:00
|
|
|
|
|
|
|
|
|
2025-09-28 08:49:34 +00:00
|
|
|
|
2022-05-18 11:55:22 -07:00
|
|
|
|
2025-09-28 08:49:33 +00:00
|
|
|
|
|
|
|
|
|
net: generalize skb freeing deferral to per-cpu lists
Logic added in commit f35f821935d8 ("tcp: defer skb freeing after socket
lock is released") helped bulk TCP flows to move the cost of skbs
frees outside of critical section where socket lock was held.
But for RPC traffic, or hosts with RFS enabled, the solution is far from
being ideal.
For RPC traffic, recvmsg() has to return to user space right after
skb payload has been consumed, meaning that BH handler has no chance
to pick the skb before recvmsg() thread. This issue is more visible
with BIG TCP, as more RPC fit one skb.
For RFS, even if BH handler picks the skbs, they are still picked
from the cpu on which user thread is running.
Ideally, it is better to free the skbs (and associated page frags)
on the cpu that originally allocated them.
This patch removes the per socket anchor (sk->defer_list) and
instead uses a per-cpu list, which will hold more skbs per round.
This new per-cpu list is drained at the end of net_action_rx(),
after incoming packets have been processed, to lower latencies.
In normal conditions, skbs are added to the per-cpu list with
no further action. In the (unlikely) cases where the cpu does not
run net_action_rx() handler fast enough, we use an IPI to raise
NET_RX_SOFTIRQ on the remote cpu.
Also, we do not bother draining the per-cpu list from dev_cpu_dead()
This is because skbs in this list have no requirement on how fast
they should be freed.
Note that we can add in the future a small per-cpu cache
if we see any contention on sd->defer_lock.
Tested on a pair of hosts with 100Gbit NIC, RFS enabled,
and /proc/sys/net/ipv4/tcp_rmem[2] tuned to 16MB to work around
page recycling strategy used by NIC driver (its page pool capacity
being too small compared to number of skbs/pages held in sockets
receive queues)
Note that this tuning was only done to demonstrate worse
conditions for skb freeing for this particular test.
These conditions can happen in more general production workload.
10 runs of one TCP_STREAM flow
Before:
Average throughput: 49685 Mbit.
Kernel profiles on cpu running user thread recvmsg() show high cost for
skb freeing related functions (*)
57.81% [kernel] [k] copy_user_enhanced_fast_string
(*) 12.87% [kernel] [k] skb_release_data
(*) 4.25% [kernel] [k] __free_one_page
(*) 3.57% [kernel] [k] __list_del_entry_valid
1.85% [kernel] [k] __netif_receive_skb_core
1.60% [kernel] [k] __skb_datagram_iter
(*) 1.59% [kernel] [k] free_unref_page_commit
(*) 1.16% [kernel] [k] __slab_free
1.16% [kernel] [k] _copy_to_iter
(*) 1.01% [kernel] [k] kfree
(*) 0.88% [kernel] [k] free_unref_page
0.57% [kernel] [k] ip6_rcv_core
0.55% [kernel] [k] ip6t_do_table
0.54% [kernel] [k] flush_smp_call_function_queue
(*) 0.54% [kernel] [k] free_pcppages_bulk
0.51% [kernel] [k] llist_reverse_order
0.38% [kernel] [k] process_backlog
(*) 0.38% [kernel] [k] free_pcp_prepare
0.37% [kernel] [k] tcp_recvmsg_locked
(*) 0.37% [kernel] [k] __list_add_valid
0.34% [kernel] [k] sock_rfree
0.34% [kernel] [k] _raw_spin_lock_irq
(*) 0.33% [kernel] [k] __page_cache_release
0.33% [kernel] [k] tcp_v6_rcv
(*) 0.33% [kernel] [k] __put_page
(*) 0.29% [kernel] [k] __mod_zone_page_state
0.27% [kernel] [k] _raw_spin_lock
After patch:
Average throughput: 73076 Mbit.
Kernel profiles on cpu running user thread recvmsg() looks better:
81.35% [kernel] [k] copy_user_enhanced_fast_string
1.95% [kernel] [k] _copy_to_iter
1.95% [kernel] [k] __skb_datagram_iter
1.27% [kernel] [k] __netif_receive_skb_core
1.03% [kernel] [k] ip6t_do_table
0.60% [kernel] [k] sock_rfree
0.50% [kernel] [k] tcp_v6_rcv
0.47% [kernel] [k] ip6_rcv_core
0.45% [kernel] [k] read_tsc
0.44% [kernel] [k] _raw_spin_lock_irqsave
0.37% [kernel] [k] _raw_spin_lock
0.37% [kernel] [k] native_irq_return_iret
0.33% [kernel] [k] __inet6_lookup_established
0.31% [kernel] [k] ip6_protocol_deliver_rcu
0.29% [kernel] [k] tcp_rcv_established
0.29% [kernel] [k] llist_reverse_order
v2: kdoc issue (kernel bots)
do not defer if (alloc_cpu == smp_processor_id()) (Paolo)
replace the sk_buff_head with a single-linked list (Jakub)
add a READ_ONCE()/WRITE_ONCE() for the lockless read of sd->defer_list
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Link: https://lore.kernel.org/r/20220422201237.416238-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-04-22 13:12:37 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-25 08:40:30 +01:00
|
|
|
|
2025-09-28 08:49:34 +00:00
|
|
|
|
net: generalize skb freeing deferral to per-cpu lists
Logic added in commit f35f821935d8 ("tcp: defer skb freeing after socket
lock is released") helped bulk TCP flows to move the cost of skbs
frees outside of critical section where socket lock was held.
But for RPC traffic, or hosts with RFS enabled, the solution is far from
being ideal.
For RPC traffic, recvmsg() has to return to user space right after
skb payload has been consumed, meaning that BH handler has no chance
to pick the skb before recvmsg() thread. This issue is more visible
with BIG TCP, as more RPC fit one skb.
For RFS, even if BH handler picks the skbs, they are still picked
from the cpu on which user thread is running.
Ideally, it is better to free the skbs (and associated page frags)
on the cpu that originally allocated them.
This patch removes the per socket anchor (sk->defer_list) and
instead uses a per-cpu list, which will hold more skbs per round.
This new per-cpu list is drained at the end of net_action_rx(),
after incoming packets have been processed, to lower latencies.
In normal conditions, skbs are added to the per-cpu list with
no further action. In the (unlikely) cases where the cpu does not
run net_action_rx() handler fast enough, we use an IPI to raise
NET_RX_SOFTIRQ on the remote cpu.
Also, we do not bother draining the per-cpu list from dev_cpu_dead()
This is because skbs in this list have no requirement on how fast
they should be freed.
Note that we can add in the future a small per-cpu cache
if we see any contention on sd->defer_lock.
Tested on a pair of hosts with 100Gbit NIC, RFS enabled,
and /proc/sys/net/ipv4/tcp_rmem[2] tuned to 16MB to work around
page recycling strategy used by NIC driver (its page pool capacity
being too small compared to number of skbs/pages held in sockets
receive queues)
Note that this tuning was only done to demonstrate worse
conditions for skb freeing for this particular test.
These conditions can happen in more general production workload.
10 runs of one TCP_STREAM flow
Before:
Average throughput: 49685 Mbit.
Kernel profiles on cpu running user thread recvmsg() show high cost for
skb freeing related functions (*)
57.81% [kernel] [k] copy_user_enhanced_fast_string
(*) 12.87% [kernel] [k] skb_release_data
(*) 4.25% [kernel] [k] __free_one_page
(*) 3.57% [kernel] [k] __list_del_entry_valid
1.85% [kernel] [k] __netif_receive_skb_core
1.60% [kernel] [k] __skb_datagram_iter
(*) 1.59% [kernel] [k] free_unref_page_commit
(*) 1.16% [kernel] [k] __slab_free
1.16% [kernel] [k] _copy_to_iter
(*) 1.01% [kernel] [k] kfree
(*) 0.88% [kernel] [k] free_unref_page
0.57% [kernel] [k] ip6_rcv_core
0.55% [kernel] [k] ip6t_do_table
0.54% [kernel] [k] flush_smp_call_function_queue
(*) 0.54% [kernel] [k] free_pcppages_bulk
0.51% [kernel] [k] llist_reverse_order
0.38% [kernel] [k] process_backlog
(*) 0.38% [kernel] [k] free_pcp_prepare
0.37% [kernel] [k] tcp_recvmsg_locked
(*) 0.37% [kernel] [k] __list_add_valid
0.34% [kernel] [k] sock_rfree
0.34% [kernel] [k] _raw_spin_lock_irq
(*) 0.33% [kernel] [k] __page_cache_release
0.33% [kernel] [k] tcp_v6_rcv
(*) 0.33% [kernel] [k] __put_page
(*) 0.29% [kernel] [k] __mod_zone_page_state
0.27% [kernel] [k] _raw_spin_lock
After patch:
Average throughput: 73076 Mbit.
Kernel profiles on cpu running user thread recvmsg() looks better:
81.35% [kernel] [k] copy_user_enhanced_fast_string
1.95% [kernel] [k] _copy_to_iter
1.95% [kernel] [k] __skb_datagram_iter
1.27% [kernel] [k] __netif_receive_skb_core
1.03% [kernel] [k] ip6t_do_table
0.60% [kernel] [k] sock_rfree
0.50% [kernel] [k] tcp_v6_rcv
0.47% [kernel] [k] ip6_rcv_core
0.45% [kernel] [k] read_tsc
0.44% [kernel] [k] _raw_spin_lock_irqsave
0.37% [kernel] [k] _raw_spin_lock
0.37% [kernel] [k] native_irq_return_iret
0.33% [kernel] [k] __inet6_lookup_established
0.31% [kernel] [k] ip6_protocol_deliver_rcu
0.29% [kernel] [k] tcp_rcv_established
0.29% [kernel] [k] llist_reverse_order
v2: kdoc issue (kernel bots)
do not defer if (alloc_cpu == smp_processor_id()) (Paolo)
replace the sk_buff_head with a single-linked list (Jakub)
add a READ_ONCE()/WRITE_ONCE() for the lockless read of sd->defer_list
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Link: https://lore.kernel.org/r/20220422201237.416238-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2022-04-22 13:12:37 -07:00
|
|
|
|
2023-05-22 13:11:12 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-07-02 15:38:08 +02:00
|
|
|
|
2023-05-22 13:11:12 +01:00
|
|
|
|
2024-04-29 13:40:21 +00:00
|
|
|
|
2023-05-22 13:11:12 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-05-26 16:39:15 +03:00
|
|
|
|
|
|
|
|
|
2023-05-22 13:11:12 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-09-25 13:03:06 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-09-25 13:03:07 +01:00
|
|
|
|
2023-09-25 13:03:06 +01:00
|
|
|
|
2023-09-25 13:03:07 +01:00
|
|
|
|
2023-09-25 13:03:06 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-09-25 13:03:08 +01:00
|
|
|
|
|
|
|
|
|
2023-09-25 13:03:06 +01:00
|
|
|
|
2023-09-25 13:03:08 +01:00
|
|
|
|
|
|
|
|
|
2023-09-25 13:03:06 +01:00
|
|
|
|
2023-09-25 13:03:08 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-09-25 13:03:06 +01:00
|
|
|
|
2023-09-25 13:03:08 +01:00
|
|
|
|
2025-05-08 00:48:22 +00:00
|
|
|
|
2026-01-22 04:57:19 +00:00
|
|
|
|
2025-05-08 00:48:22 +00:00
|
|
|
|
2026-01-22 04:57:19 +00:00
|
|
|
|
2025-05-08 00:48:22 +00:00
|
|
|
|
2026-01-22 04:57:19 +00:00
|
|
|
|
|
|
|
|
|
2025-05-08 00:48:22 +00:00
|
|
|
|
2026-01-22 04:57:19 +00:00
|
|
|
|
2025-05-08 00:48:22 +00:00
|
|
|
|
2026-01-22 04:57:19 +00:00
|
|
|
|
2025-05-08 00:48:22 +00:00
|
|
|
|
2026-01-22 04:57:19 +00:00
|
|
|
|
2025-05-08 00:48:22 +00:00
|
|
|
|
2026-01-22 04:57:19 +00:00
|
|
|
|
|
|
|
|
|
2025-05-08 00:48:22 +00:00
|
|
|
|
2026-01-22 04:57:19 +00:00
|
|
|
|
2026-02-04 05:30:23 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
