2007-10-15 03:31:52 -06:00
|
|
|
/*
|
|
|
|
* inet fragments management
|
|
|
|
*
|
|
|
|
* This program is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of the GNU General Public License
|
|
|
|
* as published by the Free Software Foundation; either version
|
|
|
|
* 2 of the License, or (at your option) any later version.
|
|
|
|
*
|
|
|
|
* Authors: Pavel Emelyanov <xemul@openvz.org>
|
|
|
|
* Started as consolidation of ipv4/ip_fragment.c,
|
|
|
|
* ipv6/reassembly. and ipv6 nf conntrack reassembly
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/list.h>
|
|
|
|
#include <linux/spinlock.h>
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/timer.h>
|
|
|
|
#include <linux/mm.h>
|
2007-10-15 03:38:08 -06:00
|
|
|
#include <linux/random.h>
|
2007-10-15 03:39:14 -06:00
|
|
|
#include <linux/skbuff.h>
|
|
|
|
#include <linux/rtnetlink.h>
|
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h
percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 02:04:11 -06:00
|
|
|
#include <linux/slab.h>
|
2007-10-15 03:31:52 -06:00
|
|
|
|
2013-03-15 05:32:30 -06:00
|
|
|
#include <net/sock.h>
|
2007-10-15 03:31:52 -06:00
|
|
|
#include <net/inet_frag.h>
|
2013-03-22 02:24:37 -06:00
|
|
|
#include <net/inet_ecn.h>
|
|
|
|
|
2014-07-24 08:50:32 -06:00
|
|
|
#define INETFRAGS_EVICT_BUCKETS 128
|
|
|
|
#define INETFRAGS_EVICT_MAX 512
|
|
|
|
|
2014-07-24 08:50:35 -06:00
|
|
|
/* don't rebuild inetfrag table with new secret more often than this */
|
|
|
|
#define INETFRAGS_MIN_REBUILD_INTERVAL (5 * HZ)
|
|
|
|
|
2013-03-22 02:24:37 -06:00
|
|
|
/* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
|
|
|
|
* Value : 0xff if frame should be dropped.
|
|
|
|
* 0 or INET_ECN_CE value, to be ORed in to final iph->tos field
|
|
|
|
*/
|
|
|
|
const u8 ip_frag_ecn_table[16] = {
|
|
|
|
/* at least one fragment had CE, and others ECT_0 or ECT_1 */
|
|
|
|
[IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = INET_ECN_CE,
|
|
|
|
[IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
|
|
|
|
[IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = INET_ECN_CE,
|
|
|
|
|
|
|
|
/* invalid combinations : drop frame */
|
|
|
|
[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff,
|
|
|
|
[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff,
|
|
|
|
[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff,
|
|
|
|
[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
|
|
|
|
[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff,
|
|
|
|
[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff,
|
|
|
|
[IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
|
|
|
|
};
|
|
|
|
EXPORT_SYMBOL(ip_frag_ecn_table);
|
2007-10-15 03:31:52 -06:00
|
|
|
|
2014-07-24 08:50:30 -06:00
|
|
|
static unsigned int
|
|
|
|
inet_frag_hashfn(const struct inet_frags *f, const struct inet_frag_queue *q)
|
|
|
|
{
|
|
|
|
return f->hashfn(q) & (INETFRAGS_HASHSZ - 1);
|
|
|
|
}
|
|
|
|
|
2014-07-24 08:50:35 -06:00
|
|
|
static bool inet_frag_may_rebuild(struct inet_frags *f)
|
|
|
|
{
|
|
|
|
return time_after(jiffies,
|
|
|
|
f->last_rebuild_jiffies + INETFRAGS_MIN_REBUILD_INTERVAL);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void inet_frag_secret_rebuild(struct inet_frags *f)
|
2007-10-15 03:38:08 -06:00
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
2014-07-24 08:50:36 -06:00
|
|
|
write_seqlock_bh(&f->rnd_seqlock);
|
2014-07-24 08:50:35 -06:00
|
|
|
|
|
|
|
if (!inet_frag_may_rebuild(f))
|
|
|
|
goto out;
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
|
2007-10-15 03:38:08 -06:00
|
|
|
get_random_bytes(&f->rnd, sizeof(u32));
|
2014-07-24 08:50:35 -06:00
|
|
|
|
2007-10-15 03:38:08 -06:00
|
|
|
for (i = 0; i < INETFRAGS_HASHSZ; i++) {
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
struct inet_frag_bucket *hb;
|
2007-10-15 03:38:08 -06:00
|
|
|
struct inet_frag_queue *q;
|
hlist: drop the node parameter from iterators
I'm not sure why, but the hlist for each entry iterators were conceived
list_for_each_entry(pos, head, member)
The hlist ones were greedy and wanted an extra parameter:
hlist_for_each_entry(tpos, pos, head, member)
Why did they need an extra pos parameter? I'm not quite sure. Not only
they don't really need it, it also prevents the iterator from looking
exactly like the list iterator, which is unfortunate.
Besides the semantic patch, there was some manual work required:
- Fix up the actual hlist iterators in linux/list.h
- Fix up the declaration of other iterators based on the hlist ones.
- A very small amount of places were using the 'node' parameter, this
was modified to use 'obj->member' instead.
- Coccinelle didn't handle the hlist_for_each_entry_safe iterator
properly, so those had to be fixed up manually.
The semantic patch which is mostly the work of Peter Senna Tschudin is here:
@@
iterator name hlist_for_each_entry, hlist_for_each_entry_continue, hlist_for_each_entry_from, hlist_for_each_entry_rcu, hlist_for_each_entry_rcu_bh, hlist_for_each_entry_continue_rcu_bh, for_each_busy_worker, ax25_uid_for_each, ax25_for_each, inet_bind_bucket_for_each, sctp_for_each_hentry, sk_for_each, sk_for_each_rcu, sk_for_each_from, sk_for_each_safe, sk_for_each_bound, hlist_for_each_entry_safe, hlist_for_each_entry_continue_rcu, nr_neigh_for_each, nr_neigh_for_each_safe, nr_node_for_each, nr_node_for_each_safe, for_each_gfn_indirect_valid_sp, for_each_gfn_sp, for_each_host;
type T;
expression a,c,d,e;
identifier b;
statement S;
@@
-T b;
<+... when != b
(
hlist_for_each_entry(a,
- b,
c, d) S
|
hlist_for_each_entry_continue(a,
- b,
c) S
|
hlist_for_each_entry_from(a,
- b,
c) S
|
hlist_for_each_entry_rcu(a,
- b,
c, d) S
|
hlist_for_each_entry_rcu_bh(a,
- b,
c, d) S
|
hlist_for_each_entry_continue_rcu_bh(a,
- b,
c) S
|
for_each_busy_worker(a, c,
- b,
d) S
|
ax25_uid_for_each(a,
- b,
c) S
|
ax25_for_each(a,
- b,
c) S
|
inet_bind_bucket_for_each(a,
- b,
c) S
|
sctp_for_each_hentry(a,
- b,
c) S
|
sk_for_each(a,
- b,
c) S
|
sk_for_each_rcu(a,
- b,
c) S
|
sk_for_each_from
-(a, b)
+(a)
S
+ sk_for_each_from(a) S
|
sk_for_each_safe(a,
- b,
c, d) S
|
sk_for_each_bound(a,
- b,
c) S
|
hlist_for_each_entry_safe(a,
- b,
c, d, e) S
|
hlist_for_each_entry_continue_rcu(a,
- b,
c) S
|
nr_neigh_for_each(a,
- b,
c) S
|
nr_neigh_for_each_safe(a,
- b,
c, d) S
|
nr_node_for_each(a,
- b,
c) S
|
nr_node_for_each_safe(a,
- b,
c, d) S
|
- for_each_gfn_sp(a, c, d, b) S
+ for_each_gfn_sp(a, c, d) S
|
- for_each_gfn_indirect_valid_sp(a, c, d, b) S
+ for_each_gfn_indirect_valid_sp(a, c, d) S
|
for_each_host(a,
- b,
c) S
|
for_each_host_safe(a,
- b,
c, d) S
|
for_each_mesh_entry(a,
- b,
c, d) S
)
...+>
[akpm@linux-foundation.org: drop bogus change from net/ipv4/raw.c]
[akpm@linux-foundation.org: drop bogus hunk from net/ipv6/raw.c]
[akpm@linux-foundation.org: checkpatch fixes]
[akpm@linux-foundation.org: fix warnings]
[akpm@linux-foudnation.org: redo intrusive kvm changes]
Tested-by: Peter Senna Tschudin <peter.senna@gmail.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-02-27 18:06:00 -07:00
|
|
|
struct hlist_node *n;
|
2007-10-15 03:38:08 -06:00
|
|
|
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
hb = &f->hash[i];
|
2014-07-24 08:50:36 -06:00
|
|
|
spin_lock(&hb->chain_lock);
|
|
|
|
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
hlist_for_each_entry_safe(q, n, &hb->chain, list) {
|
2014-07-24 08:50:30 -06:00
|
|
|
unsigned int hval = inet_frag_hashfn(f, q);
|
2007-10-15 03:38:08 -06:00
|
|
|
|
|
|
|
if (hval != i) {
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
struct inet_frag_bucket *hb_dest;
|
|
|
|
|
2007-10-15 03:38:08 -06:00
|
|
|
hlist_del(&q->list);
|
|
|
|
|
|
|
|
/* Relink to new hash chain. */
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
hb_dest = &f->hash[hval];
|
2014-07-24 08:50:36 -06:00
|
|
|
|
|
|
|
/* This is the only place where we take
|
|
|
|
* another chain_lock while already holding
|
|
|
|
* one. As this will not run concurrently,
|
|
|
|
* we cannot deadlock on hb_dest lock below, if its
|
|
|
|
* already locked it will be released soon since
|
|
|
|
* other caller cannot be waiting for hb lock
|
|
|
|
* that we've taken above.
|
|
|
|
*/
|
|
|
|
spin_lock_nested(&hb_dest->chain_lock,
|
|
|
|
SINGLE_DEPTH_NESTING);
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
hlist_add_head(&q->list, &hb_dest->chain);
|
2014-07-24 08:50:36 -06:00
|
|
|
spin_unlock(&hb_dest->chain_lock);
|
2007-10-15 03:38:08 -06:00
|
|
|
}
|
|
|
|
}
|
2014-07-24 08:50:36 -06:00
|
|
|
spin_unlock(&hb->chain_lock);
|
2007-10-15 03:38:08 -06:00
|
|
|
}
|
|
|
|
|
2014-07-24 08:50:35 -06:00
|
|
|
f->rebuild = false;
|
|
|
|
f->last_rebuild_jiffies = jiffies;
|
|
|
|
out:
|
2014-07-24 08:50:36 -06:00
|
|
|
write_sequnlock_bh(&f->rnd_seqlock);
|
2007-10-15 03:38:08 -06:00
|
|
|
}
|
|
|
|
|
2014-07-24 08:50:32 -06:00
|
|
|
static bool inet_fragq_should_evict(const struct inet_frag_queue *q)
|
|
|
|
{
|
|
|
|
return q->net->low_thresh == 0 ||
|
|
|
|
frag_mem_limit(q->net) >= q->net->low_thresh;
|
|
|
|
}
|
|
|
|
|
|
|
|
static unsigned int
|
|
|
|
inet_evict_bucket(struct inet_frags *f, struct inet_frag_bucket *hb)
|
|
|
|
{
|
|
|
|
struct inet_frag_queue *fq;
|
|
|
|
struct hlist_node *n;
|
|
|
|
unsigned int evicted = 0;
|
|
|
|
HLIST_HEAD(expired);
|
|
|
|
|
|
|
|
spin_lock(&hb->chain_lock);
|
|
|
|
|
|
|
|
hlist_for_each_entry_safe(fq, n, &hb->chain, list) {
|
|
|
|
if (!inet_fragq_should_evict(fq))
|
|
|
|
continue;
|
|
|
|
|
2015-07-23 04:05:39 -06:00
|
|
|
if (!del_timer(&fq->timer))
|
|
|
|
continue;
|
2014-07-24 08:50:32 -06:00
|
|
|
|
2015-07-23 04:05:37 -06:00
|
|
|
hlist_add_head(&fq->list_evictor, &expired);
|
2014-07-24 08:50:32 -06:00
|
|
|
++evicted;
|
|
|
|
}
|
|
|
|
|
|
|
|
spin_unlock(&hb->chain_lock);
|
|
|
|
|
2015-07-23 04:05:37 -06:00
|
|
|
hlist_for_each_entry_safe(fq, n, &expired, list_evictor)
|
2014-07-24 08:50:32 -06:00
|
|
|
f->frag_expire((unsigned long) fq);
|
|
|
|
|
|
|
|
return evicted;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void inet_frag_worker(struct work_struct *work)
|
|
|
|
{
|
|
|
|
unsigned int budget = INETFRAGS_EVICT_BUCKETS;
|
|
|
|
unsigned int i, evicted = 0;
|
|
|
|
struct inet_frags *f;
|
|
|
|
|
|
|
|
f = container_of(work, struct inet_frags, frags_work);
|
|
|
|
|
|
|
|
BUILD_BUG_ON(INETFRAGS_EVICT_BUCKETS >= INETFRAGS_HASHSZ);
|
|
|
|
|
2014-07-24 08:50:36 -06:00
|
|
|
local_bh_disable();
|
2014-07-24 08:50:32 -06:00
|
|
|
|
|
|
|
for (i = ACCESS_ONCE(f->next_bucket); budget; --budget) {
|
|
|
|
evicted += inet_evict_bucket(f, &f->hash[i]);
|
|
|
|
i = (i + 1) & (INETFRAGS_HASHSZ - 1);
|
|
|
|
if (evicted > INETFRAGS_EVICT_MAX)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
f->next_bucket = i;
|
|
|
|
|
2014-07-24 08:50:36 -06:00
|
|
|
local_bh_enable();
|
|
|
|
|
2014-07-24 08:50:35 -06:00
|
|
|
if (f->rebuild && inet_frag_may_rebuild(f))
|
|
|
|
inet_frag_secret_rebuild(f);
|
2014-07-24 08:50:32 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
static void inet_frag_schedule_worker(struct inet_frags *f)
|
|
|
|
{
|
|
|
|
if (unlikely(!work_pending(&f->frags_work)))
|
|
|
|
schedule_work(&f->frags_work);
|
|
|
|
}
|
|
|
|
|
2014-08-01 04:29:48 -06:00
|
|
|
int inet_frags_init(struct inet_frags *f)
|
2007-10-15 03:31:52 -06:00
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
2014-07-24 08:50:32 -06:00
|
|
|
INIT_WORK(&f->frags_work, inet_frag_worker);
|
|
|
|
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
for (i = 0; i < INETFRAGS_HASHSZ; i++) {
|
|
|
|
struct inet_frag_bucket *hb = &f->hash[i];
|
2007-10-15 03:31:52 -06:00
|
|
|
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
spin_lock_init(&hb->chain_lock);
|
|
|
|
INIT_HLIST_HEAD(&hb->chain);
|
|
|
|
}
|
2014-07-24 08:50:36 -06:00
|
|
|
|
|
|
|
seqlock_init(&f->rnd_seqlock);
|
2014-07-24 08:50:35 -06:00
|
|
|
f->last_rebuild_jiffies = 0;
|
2014-08-01 04:29:48 -06:00
|
|
|
f->frags_cachep = kmem_cache_create(f->frags_cache_name, f->qsize, 0, 0,
|
|
|
|
NULL);
|
|
|
|
if (!f->frags_cachep)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
return 0;
|
2007-10-15 03:31:52 -06:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(inet_frags_init);
|
|
|
|
|
|
|
|
void inet_frags_fini(struct inet_frags *f)
|
|
|
|
{
|
2014-07-24 08:50:32 -06:00
|
|
|
cancel_work_sync(&f->frags_work);
|
2014-08-01 04:29:48 -06:00
|
|
|
kmem_cache_destroy(f->frags_cachep);
|
2007-10-15 03:31:52 -06:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(inet_frags_fini);
|
2007-10-15 03:37:18 -06:00
|
|
|
|
2008-01-22 07:12:39 -07:00
|
|
|
void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f)
|
|
|
|
{
|
2014-07-24 08:50:36 -06:00
|
|
|
unsigned int seq;
|
2014-07-24 08:50:32 -06:00
|
|
|
int i;
|
|
|
|
|
2008-01-22 07:12:39 -07:00
|
|
|
nf->low_thresh = 0;
|
2008-03-28 18:30:18 -06:00
|
|
|
|
2014-07-24 08:50:36 -06:00
|
|
|
evict_again:
|
2015-07-23 04:05:39 -06:00
|
|
|
local_bh_disable();
|
2014-07-24 08:50:36 -06:00
|
|
|
seq = read_seqbegin(&f->rnd_seqlock);
|
2014-07-24 08:50:32 -06:00
|
|
|
|
|
|
|
for (i = 0; i < INETFRAGS_HASHSZ ; i++)
|
|
|
|
inet_evict_bucket(f, &f->hash[i]);
|
|
|
|
|
2014-07-24 08:50:36 -06:00
|
|
|
local_bh_enable();
|
2015-07-23 04:05:39 -06:00
|
|
|
cond_resched();
|
|
|
|
|
|
|
|
if (read_seqretry(&f->rnd_seqlock, seq) ||
|
|
|
|
percpu_counter_sum(&nf->mem))
|
|
|
|
goto evict_again;
|
net: use lib/percpu_counter API for fragmentation mem accounting
Replace the per network namespace shared atomic "mem" accounting
variable, in the fragmentation code, with a lib/percpu_counter.
Getting percpu_counter to scale to the fragmentation code usage
requires some tweaks.
At first view, percpu_counter looks superfast, but it does not
scale on multi-CPU/NUMA machines, because the default batch size
is too small, for frag code usage. Thus, I have adjusted the
batch size by using __percpu_counter_add() directly, instead of
percpu_counter_sub() and percpu_counter_add().
The batch size is increased to 130.000, based on the largest 64K
fragment memory usage. This does introduce some imprecise
memory accounting, but its does not need to be strict for this
use-case.
It is also essential, that the percpu_counter, does not
share cacheline with other writers, to make this scale.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-01-28 16:45:33 -07:00
|
|
|
|
|
|
|
percpu_counter_destroy(&nf->mem);
|
2008-01-22 07:12:39 -07:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(inet_frags_exit_net);
|
|
|
|
|
2014-07-24 08:50:36 -06:00
|
|
|
static struct inet_frag_bucket *
|
|
|
|
get_frag_bucket_locked(struct inet_frag_queue *fq, struct inet_frags *f)
|
|
|
|
__acquires(hb->chain_lock)
|
2007-10-15 03:37:18 -06:00
|
|
|
{
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
struct inet_frag_bucket *hb;
|
2014-07-24 08:50:36 -06:00
|
|
|
unsigned int seq, hash;
|
|
|
|
|
|
|
|
restart:
|
|
|
|
seq = read_seqbegin(&f->rnd_seqlock);
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
|
2014-07-24 08:50:30 -06:00
|
|
|
hash = inet_frag_hashfn(f, fq);
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
hb = &f->hash[hash];
|
|
|
|
|
|
|
|
spin_lock(&hb->chain_lock);
|
2014-07-24 08:50:36 -06:00
|
|
|
if (read_seqretry(&f->rnd_seqlock, seq)) {
|
|
|
|
spin_unlock(&hb->chain_lock);
|
|
|
|
goto restart;
|
|
|
|
}
|
|
|
|
|
|
|
|
return hb;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f)
|
|
|
|
{
|
|
|
|
struct inet_frag_bucket *hb;
|
|
|
|
|
|
|
|
hb = get_frag_bucket_locked(fq, f);
|
2015-07-23 04:05:37 -06:00
|
|
|
hlist_del(&fq->list);
|
2015-07-23 04:05:39 -06:00
|
|
|
fq->flags |= INET_FRAG_COMPLETE;
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
spin_unlock(&hb->chain_lock);
|
2007-10-15 03:37:18 -06:00
|
|
|
}
|
|
|
|
|
|
|
|
void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f)
|
|
|
|
{
|
|
|
|
if (del_timer(&fq->timer))
|
|
|
|
atomic_dec(&fq->refcnt);
|
|
|
|
|
2014-08-01 04:29:44 -06:00
|
|
|
if (!(fq->flags & INET_FRAG_COMPLETE)) {
|
2007-10-15 03:37:18 -06:00
|
|
|
fq_unlink(fq, f);
|
|
|
|
atomic_dec(&fq->refcnt);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(inet_frag_kill);
|
2007-10-15 03:39:14 -06:00
|
|
|
|
2014-07-24 08:50:34 -06:00
|
|
|
void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f)
|
2007-10-15 03:39:14 -06:00
|
|
|
{
|
|
|
|
struct sk_buff *fp;
|
2008-01-22 07:07:25 -07:00
|
|
|
struct netns_frags *nf;
|
2013-01-28 16:45:12 -07:00
|
|
|
unsigned int sum, sum_truesize = 0;
|
2007-10-15 03:39:14 -06:00
|
|
|
|
2014-08-01 04:29:44 -06:00
|
|
|
WARN_ON(!(q->flags & INET_FRAG_COMPLETE));
|
2008-07-25 22:43:18 -06:00
|
|
|
WARN_ON(del_timer(&q->timer) != 0);
|
2007-10-15 03:39:14 -06:00
|
|
|
|
|
|
|
/* Release all fragment data. */
|
|
|
|
fp = q->fragments;
|
2008-01-22 07:07:25 -07:00
|
|
|
nf = q->net;
|
2007-10-15 03:39:14 -06:00
|
|
|
while (fp) {
|
|
|
|
struct sk_buff *xp = fp->next;
|
|
|
|
|
2013-01-28 16:45:12 -07:00
|
|
|
sum_truesize += fp->truesize;
|
2016-01-05 14:17:55 -07:00
|
|
|
kfree_skb(fp);
|
2007-10-15 03:39:14 -06:00
|
|
|
fp = xp;
|
|
|
|
}
|
2013-01-28 16:45:12 -07:00
|
|
|
sum = sum_truesize + f->qsize;
|
2007-10-15 03:39:14 -06:00
|
|
|
|
2007-10-17 20:48:26 -06:00
|
|
|
if (f->destructor)
|
|
|
|
f->destructor(q);
|
2014-08-01 04:29:48 -06:00
|
|
|
kmem_cache_free(f->frags_cachep, q);
|
2015-07-23 04:05:39 -06:00
|
|
|
|
|
|
|
sub_frag_mem_limit(nf, sum);
|
2007-10-15 03:39:14 -06:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(inet_frag_destroy);
|
2007-10-15 03:40:06 -06:00
|
|
|
|
2008-01-22 07:02:14 -07:00
|
|
|
static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf,
|
2014-08-01 04:29:46 -06:00
|
|
|
struct inet_frag_queue *qp_in,
|
|
|
|
struct inet_frags *f,
|
|
|
|
void *arg)
|
2007-10-17 20:44:34 -06:00
|
|
|
{
|
2014-07-24 08:50:36 -06:00
|
|
|
struct inet_frag_bucket *hb = get_frag_bucket_locked(qp_in, f);
|
2007-10-17 20:44:34 -06:00
|
|
|
struct inet_frag_queue *qp;
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
|
2007-10-17 20:44:34 -06:00
|
|
|
#ifdef CONFIG_SMP
|
|
|
|
/* With SMP race we have to recheck hash table, because
|
2014-07-24 08:50:36 -06:00
|
|
|
* such entry could have been created on other cpu before
|
|
|
|
* we acquired hash bucket lock.
|
2007-10-17 20:44:34 -06:00
|
|
|
*/
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
hlist_for_each_entry(qp, &hb->chain, list) {
|
2008-01-22 07:02:14 -07:00
|
|
|
if (qp->net == nf && f->match(qp, arg)) {
|
2007-10-17 20:44:34 -06:00
|
|
|
atomic_inc(&qp->refcnt);
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
spin_unlock(&hb->chain_lock);
|
2014-08-01 04:29:44 -06:00
|
|
|
qp_in->flags |= INET_FRAG_COMPLETE;
|
2007-10-17 20:44:34 -06:00
|
|
|
inet_frag_put(qp_in, f);
|
|
|
|
return qp;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
qp = qp_in;
|
2008-01-22 07:09:37 -07:00
|
|
|
if (!mod_timer(&qp->timer, jiffies + nf->timeout))
|
2007-10-17 20:44:34 -06:00
|
|
|
atomic_inc(&qp->refcnt);
|
|
|
|
|
|
|
|
atomic_inc(&qp->refcnt);
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
hlist_add_head(&qp->list, &hb->chain);
|
2014-07-24 08:50:34 -06:00
|
|
|
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
spin_unlock(&hb->chain_lock);
|
net: fix for a race condition in the inet frag code
I stumbled upon this very serious bug while hunting for another one,
it's a very subtle race condition between inet_frag_evictor,
inet_frag_intern and the IPv4/6 frag_queue and expire functions
(basically the users of inet_frag_kill/inet_frag_put).
What happens is that after a fragment has been added to the hash chain
but before it's been added to the lru_list (inet_frag_lru_add) in
inet_frag_intern, it may get deleted (either by an expired timer if
the system load is high or the timer sufficiently low, or by the
fraq_queue function for different reasons) before it's added to the
lru_list, then after it gets added it's a matter of time for the
evictor to get to a piece of memory which has been freed leading to a
number of different bugs depending on what's left there.
I've been able to trigger this on both IPv4 and IPv6 (which is normal
as the frag code is the same), but it's been much more difficult to
trigger on IPv4 due to the protocol differences about how fragments
are treated.
The setup I used to reproduce this is: 2 machines with 4 x 10G bonded
in a RR bond, so the same flow can be seen on multiple cards at the
same time. Then I used multiple instances of ping/ping6 to generate
fragmented packets and flood the machines with them while running
other processes to load the attacked machine.
*It is very important to have the _same flow_ coming in on multiple CPUs
concurrently. Usually the attacked machine would die in less than 30
minutes, if configured properly to have many evictor calls and timeouts
it could happen in 10 minutes or so.
An important point to make is that any caller (frag_queue or timer) of
inet_frag_kill will remove both the timer refcount and the
original/guarding refcount thus removing everything that's keeping the
frag from being freed at the next inet_frag_put. All of this could
happen before the frag was ever added to the LRU list, then it gets
added and the evictor uses a freed fragment.
An example for IPv6 would be if a fragment is being added and is at
the stage of being inserted in the hash after the hash lock is
released, but before inet_frag_lru_add executes (or is able to obtain
the lru lock) another overlapping fragment for the same flow arrives
at a different CPU which finds it in the hash, but since it's
overlapping it drops it invoking inet_frag_kill and thus removing all
guarding refcounts, and afterwards freeing it by invoking
inet_frag_put which removes the last refcount added previously by
inet_frag_find, then inet_frag_lru_add gets executed by
inet_frag_intern and we have a freed fragment in the lru_list.
The fix is simple, just move the lru_add under the hash chain locked
region so when a removing function is called it'll have to wait for
the fragment to be added to the lru_list, and then it'll remove it (it
works because the hash chain removal is done before the lru_list one
and there's no window between the two list adds when the frag can get
dropped). With this fix applied I couldn't kill the same machine in 24
hours with the same setup.
Fixes: 3ef0eb0db4bf ("net: frag, move LRU list maintenance outside of
rwlock")
CC: Florian Westphal <fw@strlen.de>
CC: Jesper Dangaard Brouer <brouer@redhat.com>
CC: David S. Miller <davem@davemloft.net>
Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-03-03 15:19:18 -07:00
|
|
|
|
2007-10-17 20:44:34 -06:00
|
|
|
return qp;
|
|
|
|
}
|
2007-10-17 20:45:23 -06:00
|
|
|
|
2008-01-22 07:02:14 -07:00
|
|
|
static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
|
2014-08-01 04:29:46 -06:00
|
|
|
struct inet_frags *f,
|
|
|
|
void *arg)
|
2007-10-17 20:45:23 -06:00
|
|
|
{
|
|
|
|
struct inet_frag_queue *q;
|
|
|
|
|
2014-07-24 08:50:32 -06:00
|
|
|
if (frag_mem_limit(nf) > nf->high_thresh) {
|
|
|
|
inet_frag_schedule_worker(f);
|
2014-07-24 08:50:31 -06:00
|
|
|
return NULL;
|
2014-07-24 08:50:32 -06:00
|
|
|
}
|
2014-07-24 08:50:31 -06:00
|
|
|
|
2014-08-01 04:29:48 -06:00
|
|
|
q = kmem_cache_zalloc(f->frags_cachep, GFP_ATOMIC);
|
2015-04-03 02:17:26 -06:00
|
|
|
if (!q)
|
2007-10-17 20:45:23 -06:00
|
|
|
return NULL;
|
|
|
|
|
2012-06-07 19:21:40 -06:00
|
|
|
q->net = nf;
|
2007-10-17 20:46:47 -06:00
|
|
|
f->constructor(q, arg);
|
2015-07-23 04:05:38 -06:00
|
|
|
add_frag_mem_limit(nf, f->qsize);
|
2013-01-28 16:45:12 -07:00
|
|
|
|
2007-10-17 20:45:23 -06:00
|
|
|
setup_timer(&q->timer, f->frag_expire, (unsigned long)q);
|
|
|
|
spin_lock_init(&q->lock);
|
|
|
|
atomic_set(&q->refcnt, 1);
|
|
|
|
|
|
|
|
return q;
|
|
|
|
}
|
2007-10-17 20:46:47 -06:00
|
|
|
|
2008-01-22 07:02:14 -07:00
|
|
|
static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
|
2014-08-01 04:29:46 -06:00
|
|
|
struct inet_frags *f,
|
|
|
|
void *arg)
|
2007-10-17 20:46:47 -06:00
|
|
|
{
|
|
|
|
struct inet_frag_queue *q;
|
|
|
|
|
2008-01-22 07:02:14 -07:00
|
|
|
q = inet_frag_alloc(nf, f, arg);
|
2015-04-03 02:17:26 -06:00
|
|
|
if (!q)
|
2007-10-17 20:46:47 -06:00
|
|
|
return NULL;
|
|
|
|
|
2008-06-27 21:06:08 -06:00
|
|
|
return inet_frag_intern(nf, q, f, arg);
|
2007-10-17 20:46:47 -06:00
|
|
|
}
|
2007-10-17 20:47:21 -06:00
|
|
|
|
2008-01-22 07:02:14 -07:00
|
|
|
struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
|
2014-08-01 04:29:46 -06:00
|
|
|
struct inet_frags *f, void *key,
|
|
|
|
unsigned int hash)
|
2007-10-17 20:47:21 -06:00
|
|
|
{
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
struct inet_frag_bucket *hb;
|
2007-10-17 20:47:21 -06:00
|
|
|
struct inet_frag_queue *q;
|
2013-03-15 05:32:30 -06:00
|
|
|
int depth = 0;
|
2007-10-17 20:47:21 -06:00
|
|
|
|
2014-07-24 08:50:32 -06:00
|
|
|
if (frag_mem_limit(nf) > nf->low_thresh)
|
|
|
|
inet_frag_schedule_worker(f);
|
2014-07-24 08:50:31 -06:00
|
|
|
|
2014-07-24 08:50:30 -06:00
|
|
|
hash &= (INETFRAGS_HASHSZ - 1);
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
hb = &f->hash[hash];
|
|
|
|
|
|
|
|
spin_lock(&hb->chain_lock);
|
|
|
|
hlist_for_each_entry(q, &hb->chain, list) {
|
2008-01-22 07:02:14 -07:00
|
|
|
if (q->net == nf && f->match(q, key)) {
|
2007-10-17 20:47:21 -06:00
|
|
|
atomic_inc(&q->refcnt);
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
spin_unlock(&hb->chain_lock);
|
2007-10-17 20:47:21 -06:00
|
|
|
return q;
|
|
|
|
}
|
2013-03-15 05:32:30 -06:00
|
|
|
depth++;
|
2007-10-17 20:47:21 -06:00
|
|
|
}
|
net: frag queue per hash bucket locking
This patch implements per hash bucket locking for the frag queue
hash. This removes two write locks, and the only remaining write
lock is for protecting hash rebuild. This essentially reduce the
readers-writer lock to a rebuild lock.
This patch is part of "net: frag performance followup"
http://thread.gmane.org/gmane.linux.network/263644
of which two patches have already been accepted:
Same test setup as previous:
(http://thread.gmane.org/gmane.linux.network/257155)
Two 10G interfaces, on seperate NUMA nodes, are under-test, and uses
Ethernet flow-control. A third interface is used for generating the
DoS attack (with trafgen).
Notice, I have changed the frag DoS generator script to be more
efficient/deadly. Before it would only hit one RX queue, now its
sending packets causing multi-queue RX, due to "better" RX hashing.
Test types summary (netperf UDP_STREAM):
Test-20G64K == 2x10G with 65K fragments
Test-20G3F == 2x10G with 3x fragments (3*1472 bytes)
Test-20G64K+DoS == Same as 20G64K with frag DoS
Test-20G3F+DoS == Same as 20G3F with frag DoS
Test-20G64K+MQ == Same as 20G64K with Multi-Queue frag DoS
Test-20G3F+MQ == Same as 20G3F with Multi-Queue frag DoS
When I rebased this-patch(03) (on top of net-next commit a210576c) and
removed the _bh spinlock, I saw a performance regression. BUT this
was caused by some unrelated change in-between. See tests below.
Test (A) is what I reported before for patch-02, accepted in commit 1b5ab0de.
Test (B) verifying-retest of commit 1b5ab0de corrospond to patch-02.
Test (C) is what I reported before for this-patch
Test (D) is net-next master HEAD (commit a210576c), which reveals some
(unknown) performance regression (compared against test (B)).
Test (D) function as a new base-test.
Performance table summary (in Mbit/s):
(#) Test-type: 20G64K 20G3F 20G64K+DoS 20G3F+DoS 20G64K+MQ 20G3F+MQ
---------- ------- ------- ---------- --------- -------- -------
(A) Patch-02 : 18848.7 13230.1 4103.04 5310.36 130.0 440.2
(B) 1b5ab0de : 18841.5 13156.8 4101.08 5314.57 129.0 424.2
(C) Patch-03v1: 18838.0 13490.5 4405.11 6814.72 196.6 461.6
(D) a210576c : 18321.5 11250.4 3635.34 5160.13 119.1 405.2
(E) with _bh : 17247.3 11492.6 3994.74 6405.29 166.7 413.6
(F) without bh: 17471.3 11298.7 3818.05 6102.11 165.7 406.3
Test (E) and (F) is this-patch(03), with(V1) and without(V2) the _bh spinlocks.
I cannot explain the slow down for 20G64K (but its an artificial
"lab-test" so I'm not worried). But the other results does show
improvements. And test (E) "with _bh" version is slightly better.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
----
V2:
- By analysis from Hannes Frederic Sowa and Eric Dumazet, we don't
need the spinlock _bh versions, as Netfilter currently does a
local_bh_disable() before entering inet_fragment.
- Fold-in desc from cover-mail
V3:
- Drop the chain_len counter per hash bucket.
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-04-03 17:38:16 -06:00
|
|
|
spin_unlock(&hb->chain_lock);
|
2007-10-17 20:47:21 -06:00
|
|
|
|
2013-03-15 05:32:30 -06:00
|
|
|
if (depth <= INETFRAGS_MAXDEPTH)
|
|
|
|
return inet_frag_create(nf, f, key);
|
2014-07-24 08:50:35 -06:00
|
|
|
|
|
|
|
if (inet_frag_may_rebuild(f)) {
|
2014-07-24 08:50:36 -06:00
|
|
|
if (!f->rebuild)
|
|
|
|
f->rebuild = true;
|
2014-07-24 08:50:35 -06:00
|
|
|
inet_frag_schedule_worker(f);
|
|
|
|
}
|
|
|
|
|
|
|
|
return ERR_PTR(-ENOBUFS);
|
2007-10-17 20:47:21 -06:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(inet_frag_find);
|
2013-03-15 05:32:30 -06:00
|
|
|
|
|
|
|
void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
|
|
|
|
const char *prefix)
|
|
|
|
{
|
|
|
|
static const char msg[] = "inet_frag_find: Fragment hash bucket"
|
|
|
|
" list length grew over limit " __stringify(INETFRAGS_MAXDEPTH)
|
|
|
|
". Dropping fragment.\n";
|
|
|
|
|
|
|
|
if (PTR_ERR(q) == -ENOBUFS)
|
2014-11-11 11:59:17 -07:00
|
|
|
net_dbg_ratelimited("%s%s", prefix, msg);
|
2013-03-15 05:32:30 -06:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(inet_frag_maybe_warn_overflow);
|