Embarrassingly, nft_pipapo_insert() checked for interval validity in
the first field only.
The start_p and end_p pointers were reset to key data from the first
field at every iteration of the loop which was supposed to go over
the set fields.
Fixes: 3c4287f620 ("nf_tables: Add set type for arbitrary concatenation of ranges")
Reported-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
New elements that reside in the clone are not released in case that the
transaction is aborted.
[16302.231754] ------------[ cut here ]------------
[16302.231756] WARNING: CPU: 0 PID: 100509 at net/netfilter/nf_tables_api.c:1864 nf_tables_chain_destroy+0x26/0x127 [nf_tables]
[...]
[16302.231882] CPU: 0 PID: 100509 Comm: nft Tainted: G W 5.19.0-rc3+ #155
[...]
[16302.231887] RIP: 0010:nf_tables_chain_destroy+0x26/0x127 [nf_tables]
[16302.231899] Code: f3 fe ff ff 41 55 41 54 55 53 48 8b 6f 10 48 89 fb 48 c7 c7 82 96 d9 a0 8b 55 50 48 8b 75 58 e8 de f5 92 e0 83 7d 50 00 74 09 <0f> 0b 5b 5d 41 5c 41 5d c3 4c 8b 65 00 48 8b 7d 08 49 39 fc 74 05
[...]
[16302.231917] Call Trace:
[16302.231919] <TASK>
[16302.231921] __nf_tables_abort.cold+0x23/0x28 [nf_tables]
[16302.231934] nf_tables_abort+0x30/0x50 [nf_tables]
[16302.231946] nfnetlink_rcv_batch+0x41a/0x840 [nfnetlink]
[16302.231952] ? __nla_validate_parse+0x48/0x190
[16302.231959] nfnetlink_rcv+0x110/0x129 [nfnetlink]
[16302.231963] netlink_unicast+0x211/0x340
[16302.231969] netlink_sendmsg+0x21e/0x460
Add nft_set_pipapo_match_destroy() helper function to release the
elements in the lookup tables.
Stefano Brivio says: "We additionally look for elements pointers in the
cloned matching data if priv->dirty is set, because that means that
cloned data might point to additional elements we did not commit to the
working copy yet (such as the abort path case, but perhaps not limited
to it)."
Fixes: 3c4287f620 ("nf_tables: Add set type for arbitrary concatenation of ranges")
Reviewed-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This is needed in case a new transaction is made that doesn't insert any
new elements into an already existing set.
Else, after second 'nft -f ruleset.txt', lookups in such a set will fail
because ->lookup() encounters raw_cpu_ptr(m->scratch) == NULL.
For the initial rule load, insertion of elements takes care of the
allocation, but for rule reloads this isn't guaranteed: we might not
have additions to the set.
Fixes: 3c4287f620 ("nf_tables: Add set type for arbitrary concatenation of ranges")
Reported-by: etkaar <lists.netfilter.org@prvy.eu>
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This patch extends the set infrastructure to add a special catch-all set
element. If the lookup fails to find an element (or range) in the set,
then the catch-all element is selected. Users can specify a mapping,
expression(s) and timeout to be attached to the catch-all element.
This patch adds a catchall list to the set, this list might contain more
than one single catch-all element (e.g. in case that the catch-all
element is removed and a new one is added in the same transaction).
However, most of the time, there will be either one element or no
elements at all in this list.
The catch-all element is identified via NFT_SET_ELEM_CATCHALL flag and
such special element has no NFTA_SET_ELEM_KEY attribute. There is a new
nft_set_elem_catchall object that stores a reference to the dummy
catch-all element (catchall->elem) whose layout is the same of the set
element type to reuse the existing set element codebase.
The set size does not apply to the catch-all element, users can define a
catch-all element even if the set is full.
The check for valid set element flags hava been updates to report
EOPNOTSUPP in case userspace requests flags that are not supported when
using new userspace nftables and old kernel.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Rationale:
Reduces attack surface on kernel devs opening the links for MITM
as HTTPS traffic is much harder to manipulate.
Deterministic algorithm:
For each file:
If not .svg:
For each line:
If doesn't contain `\bxmlns\b`:
For each link, `\bhttp://[^# \t\r\n]*(?:\w|/)`:
If neither `\bgnu\.org/license`, nor `\bmozilla\.org/MPL\b`:
If both the HTTP and HTTPS versions
return 200 OK and serve the same content:
Replace HTTP with HTTPS.
Signed-off-by: Alexander A. Klimov <grandmaster@al2klimov.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Simple fixes which require no deep knowledge of the code.
Cc: Pablo Neira Ayuso <pablo@netfilter.org>
Cc: Jozsef Kadlecsik <kadlec@netfilter.org>
Cc: Florian Westphal <fw@strlen.de>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
In nft_pipapo_insert(), we need to reallocate scratch maps that will
be used for matching by lookup functions, if they have never been
allocated or if the bucket size changes as a result of the insertion.
As pipapo_realloc_scratch() provides a pair of fresh, zeroed out
maps, there's no need to select a particular one after reallocation.
Other than being useless, the existing assignment was also troubled
by the fact that the index was set only on the CPU performing the
actual insertion, as spotted by Florian.
Simply drop the assignment.
Reported-by: Florian Westphal <fw@strlen.de>
Fixes: 3c4287f620 ("nf_tables: Add set type for arbitrary concatenation of ranges")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Now that the kernel specifies binutils 2.23 as the minimum version, we
can remove ifdefs for AVX2 and ADX throughout.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Acked-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Overlapping header include additions in macsec.c
A bug fix in 'net' overlapping with the removal of 'version'
string in ena_netdev.c
Overlapping test additions in selftests Makefile
Overlapping PCI ID table adjustments in iwlwifi driver.
Signed-off-by: David S. Miller <davem@davemloft.net>
...and return -ENOTEMPTY to the front-end on collision, -EEXIST if
an identical element already exists. Together with the previous patch,
element collision will now be returned to the user as -EEXIST.
Reported-by: Phil Sutter <phil@nwl.cc>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
If the AVX2 set is available, we can exploit the repetitive
characteristic of this algorithm to provide a fast, vectorised
version by using 256-bit wide AVX2 operations for bucket loads and
bitwise intersections.
In most cases, this implementation consistently outperforms rbtree
set instances despite the fact they are configured to use a given,
single, ranged data type out of the ones used for performance
measurements by the nft_concat_range.sh kselftest.
That script, injecting packets directly on the ingoing device path
with pktgen, reports, averaged over five runs on a single AMD Epyc
7402 thread (3.35GHz, 768 KiB L1D$, 12 MiB L2$), the figures below.
CONFIG_RETPOLINE was not set here.
Note that this is not a fair comparison over hash and rbtree set
types: non-ranged entries (used to have a reference for hash types)
would be matched faster than this, and matching on a single field
only (which is the case for rbtree) is also significantly faster.
However, it's not possible at the moment to choose this set type
for non-ranged entries, and the current implementation also needs
a few minor adjustments in order to match on less than two fields.
---------------.-----------------------------------.------------.
AMD Epyc 7402 | baselines, Mpps | this patch |
1 thread |___________________________________|____________|
3.35GHz | | | | | |
768KiB L1D$ | netdev | hash | rbtree | | |
---------------| hook | no | single | | pipapo |
type entries | drop | ranges | field | pipapo | AVX2 |
---------------|--------|--------|--------|--------|------------|
net,port | | | | | |
1000 | 19.0 | 10.4 | 3.8 | 4.0 | 7.5 +87% |
---------------|--------|--------|--------|--------|------------|
port,net | | | | | |
100 | 18.8 | 10.3 | 5.8 | 6.3 | 8.1 +29% |
---------------|--------|--------|--------|--------|------------|
net6,port | | | | | |
1000 | 16.4 | 7.6 | 1.8 | 2.1 | 4.8 +128% |
---------------|--------|--------|--------|--------|------------|
port,proto | | | | | |
30000 | 19.6 | 11.6 | 3.9 | 0.5 | 2.6 +420% |
---------------|--------|--------|--------|--------|------------|
net6,port,mac | | | | | |
10 | 16.5 | 5.4 | 4.3 | 3.4 | 4.7 +38% |
---------------|--------|--------|--------|--------|------------|
net6,port,mac, | | | | | |
proto 1000 | 16.5 | 5.7 | 1.9 | 1.4 | 3.6 +26% |
---------------|--------|--------|--------|--------|------------|
net,mac | | | | | |
1000 | 19.0 | 8.4 | 3.9 | 2.5 | 6.4 +156% |
---------------'--------'--------'--------'--------'------------'
A similar strategy could be easily reused to implement specialised
versions for other SIMD sets, and I plan to post at least a NEON
version at a later time.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Move most macros and helpers to a header file, so that they can be
conveniently used by related implementations.
No functional changes are intended here.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
SIMD vector extension sets require stricter alignment than native
instruction sets to operate efficiently (AVX, NEON) or for some
instructions to work at all (AltiVec).
Provide facilities to define arbitrary alignment for lookup tables
and scratch maps. By defining byte alignment with NFT_PIPAPO_ALIGN,
lt_aligned and scratch_aligned pointers become available.
Additional headroom is allocated, and pointers to the possibly
unaligned, originally allocated areas are kept so that they can
be freed.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
While grouping matching bits in groups of four saves memory compared
to the more natural choice of 8-bit words (lookup table size is one
eighth), it comes at a performance cost, as the number of lookup
comparisons is doubled, and those also needs bitshifts and masking.
Introduce support for 8-bit lookup groups, together with a mapping
mechanism to dynamically switch, based on defined per-table size
thresholds and hysteresis, between 8-bit and 4-bit groups, as tables
grow and shrink. Empty sets start with 8-bit groups, and per-field
tables are converted to 4-bit groups if they get too big.
An alternative approach would have been to swap per-set lookup
operation functions as needed, but this doesn't allow for different
group sizes in the same set, which looks desirable if some fields
need significantly more matching data compared to others due to
heavier impact of ranges (e.g. a big number of subnets with
relatively simple port specifications).
Allowing different group sizes for the same lookup functions implies
the need for further conditional clauses, whose cost, however,
appears to be negligible in tests.
The matching rate figures below were obtained for x86_64 running
the nft_concat_range.sh "performance" cases, averaged over five
runs, on a single thread of an AMD Epyc 7402 CPU, and for aarch64
on a single thread of a BCM2711 (Raspberry Pi 4 Model B 4GB),
clocked at a stable 2147MHz frequency:
---------------.-----------------------------------.------------.
AMD Epyc 7402 | baselines, Mpps | this patch |
1 thread |___________________________________|____________|
3.35GHz | | | | | |
768KiB L1D$ | netdev | hash | rbtree | | |
---------------| hook | no | single | pipapo | pipapo |
type entries | drop | ranges | field | 4 bits | bit switch |
---------------|--------|--------|--------|--------|------------|
net,port | | | | | |
1000 | 19.0 | 10.4 | 3.8 | 2.8 | 4.0 +43% |
---------------|--------|--------|--------|--------|------------|
port,net | | | | | |
100 | 18.8 | 10.3 | 5.8 | 5.5 | 6.3 +14% |
---------------|--------|--------|--------|--------|------------|
net6,port | | | | | |
1000 | 16.4 | 7.6 | 1.8 | 1.3 | 2.1 +61% |
---------------|--------|--------|--------|--------|------------|
port,proto | | | | | [1] |
30000 | 19.6 | 11.6 | 3.9 | 0.3 | 0.5 +66% |
---------------|--------|--------|--------|--------|------------|
net6,port,mac | | | | | |
10 | 16.5 | 5.4 | 4.3 | 2.6 | 3.4 +31% |
---------------|--------|--------|--------|--------|------------|
net6,port,mac, | | | | | |
proto 1000 | 16.5 | 5.7 | 1.9 | 1.0 | 1.4 +40% |
---------------|--------|--------|--------|--------|------------|
net,mac | | | | | |
1000 | 19.0 | 8.4 | 3.9 | 1.7 | 2.5 +47% |
---------------'--------'--------'--------'--------'------------'
[1] Causes switch of lookup table buckets for 'port', not 'proto',
to 4-bit groups
---------------.-----------------------------------.------------.
BCM2711 | baselines, Mpps | this patch |
1 thread |___________________________________|____________|
2147MHz | | | | | |
32KiB L1D$ | netdev | hash | rbtree | | |
---------------| hook | no | single | pipapo | pipapo |
type entries | drop | ranges | field | 4 bits | bit switch |
---------------|--------|--------|--------|--------|------------|
net,port | | | | | |
1000 | 1.63 | 1.37 | 0.87 | 0.61 | 0.70 +17% |
---------------|--------|--------|--------|--------|------------|
port,net | | | | | |
100 | 1.64 | 1.36 | 1.02 | 0.78 | 0.81 +4% |
---------------|--------|--------|--------|--------|------------|
net6,port | | | | | |
1000 | 1.56 | 1.27 | 0.65 | 0.34 | 0.50 +47% |
---------------|--------|--------|--------|--------|------------|
port,proto [2] | | | | | |
10000 | 1.68 | 1.43 | 0.84 | 0.30 | 0.40 +13% |
---------------|--------|--------|--------|--------|------------|
net6,port,mac | | | | | |
10 | 1.56 | 1.14 | 1.02 | 0.62 | 0.66 +6% |
---------------|--------|--------|--------|--------|------------|
net6,port,mac, | | | | | |
proto 1000 | 1.56 | 1.12 | 0.64 | 0.27 | 0.40 +48% |
---------------|--------|--------|--------|--------|------------|
net,mac | | | | | |
1000 | 1.63 | 1.26 | 0.87 | 0.41 | 0.53 +29% |
---------------'--------'--------'--------'--------'------------'
[2] Using 10000 entries instead of 30000 as it would take way too
long for the test script to generate all of them
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Get rid of all hardcoded assumptions that buckets in lookup tables
correspond to four-bit groups, and replace them with appropriate
calculations based on a variable group size, now stored in struct
field.
The group size could now be in principle any divisor of eight. Note,
though, that lookup and get functions need an implementation
intimately depending on the group size, and the only supported size
there, currently, is four bits, which is also the initial and only
used size at the moment.
While at it, drop 'groups' from struct nft_pipapo: it was never used.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The current codebase makes use of the zero-length array language
extension to the C90 standard, but the preferred mechanism to declare
variable-length types such as these ones is a flexible array member[1][2],
introduced in C99:
struct foo {
int stuff;
struct boo array[];
};
By making use of the mechanism above, we will get a compiler warning
in case the flexible array does not occur last in the structure, which
will help us prevent some kind of undefined behavior bugs from being
inadvertently introduced[3] to the codebase from now on.
Also, notice that, dynamic memory allocations won't be affected by
this change:
"Flexible array members have incomplete type, and so the sizeof operator
may not be applied. As a quirk of the original implementation of
zero-length arrays, sizeof evaluates to zero."[1]
Lastly, fix checkpatch.pl warning
WARNING: __aligned(size) is preferred over __attribute__((aligned(size)))
in net/bridge/netfilter/ebtables.c
This issue was found with the help of Coccinelle.
[1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
[2] https://github.com/KSPP/linux/issues/21
[3] commit 7649773293 ("cxgb3/l2t: Fix undefined behaviour")
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Fix the following sparse warning:
net/netfilter/nft_set_pipapo.c:739:6: warning: symbol 'nft_pipapo_get' was not declared. Should it be static?
Fixes: 3c4287f620 ("nf_tables: Add set type for arbitrary concatenation of ranges")
Signed-off-by: Chen Wandun <chenwandun@huawei.com>
Acked-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
They do not need to be writeable anymore.
v2: remove left-over __read_mostly annotation in set_pipapo.c (Stefano)
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Phil reports that adding elements, flushing and re-adding them
right away:
nft add table t '{ set s { type ipv4_addr . inet_service; flags interval; }; }'
nft add element t s '{ 10.0.0.1 . 22-25, 10.0.0.1 . 10-20 }'
nft flush set t s
nft add element t s '{ 10.0.0.1 . 10-20, 10.0.0.1 . 22-25 }'
triggers, almost reliably, a crash like this one:
[ 71.319848] general protection fault, probably for non-canonical address 0x6f6b6e696c2e756e: 0000 [#1] PREEMPT SMP PTI
[ 71.321540] CPU: 3 PID: 1201 Comm: kworker/3:2 Not tainted 5.6.0-rc1-00377-g2bb07f4e1d861 #192
[ 71.322746] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ?-20190711_202441-buildvm-armv7-10.arm.fedoraproject.org-2.fc31 04/01/2014
[ 71.324430] Workqueue: events nf_tables_trans_destroy_work [nf_tables]
[ 71.325387] RIP: 0010:nft_set_elem_destroy+0xa5/0x110 [nf_tables]
[ 71.326164] Code: 89 d4 84 c0 74 0e 8b 77 44 0f b6 f8 48 01 df e8 41 ff ff ff 45 84 e4 74 36 44 0f b6 63 08 45 84 e4 74 2c 49 01 dc 49 8b 04 24 <48> 8b 40 38 48 85 c0 74 4f 48 89 e7 4c 8b
[ 71.328423] RSP: 0018:ffffc9000226fd90 EFLAGS: 00010282
[ 71.329225] RAX: 6f6b6e696c2e756e RBX: ffff88813ab79f60 RCX: ffff88813931b5a0
[ 71.330365] RDX: 0000000000000001 RSI: 0000000000000000 RDI: ffff88813ab79f9a
[ 71.331473] RBP: ffff88813ab79f60 R08: 0000000000000008 R09: 0000000000000000
[ 71.332627] R10: 000000000000021c R11: 0000000000000000 R12: ffff88813ab79fc2
[ 71.333615] R13: ffff88813b3adf50 R14: dead000000000100 R15: ffff88813931b8a0
[ 71.334596] FS: 0000000000000000(0000) GS:ffff88813bd80000(0000) knlGS:0000000000000000
[ 71.335780] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 71.336577] CR2: 000055ac683710f0 CR3: 000000013a222003 CR4: 0000000000360ee0
[ 71.337533] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 71.338557] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 71.339718] Call Trace:
[ 71.340093] nft_pipapo_destroy+0x7a/0x170 [nf_tables_set]
[ 71.340973] nft_set_destroy+0x20/0x50 [nf_tables]
[ 71.341879] nf_tables_trans_destroy_work+0x246/0x260 [nf_tables]
[ 71.342916] process_one_work+0x1d5/0x3c0
[ 71.343601] worker_thread+0x4a/0x3c0
[ 71.344229] kthread+0xfb/0x130
[ 71.344780] ? process_one_work+0x3c0/0x3c0
[ 71.345477] ? kthread_park+0x90/0x90
[ 71.346129] ret_from_fork+0x35/0x40
[ 71.346748] Modules linked in: nf_tables_set nf_tables nfnetlink 8021q [last unloaded: nfnetlink]
[ 71.348153] ---[ end trace 2eaa8149ca759bcc ]---
[ 71.349066] RIP: 0010:nft_set_elem_destroy+0xa5/0x110 [nf_tables]
[ 71.350016] Code: 89 d4 84 c0 74 0e 8b 77 44 0f b6 f8 48 01 df e8 41 ff ff ff 45 84 e4 74 36 44 0f b6 63 08 45 84 e4 74 2c 49 01 dc 49 8b 04 24 <48> 8b 40 38 48 85 c0 74 4f 48 89 e7 4c 8b
[ 71.350017] RSP: 0018:ffffc9000226fd90 EFLAGS: 00010282
[ 71.350019] RAX: 6f6b6e696c2e756e RBX: ffff88813ab79f60 RCX: ffff88813931b5a0
[ 71.350019] RDX: 0000000000000001 RSI: 0000000000000000 RDI: ffff88813ab79f9a
[ 71.350020] RBP: ffff88813ab79f60 R08: 0000000000000008 R09: 0000000000000000
[ 71.350021] R10: 000000000000021c R11: 0000000000000000 R12: ffff88813ab79fc2
[ 71.350022] R13: ffff88813b3adf50 R14: dead000000000100 R15: ffff88813931b8a0
[ 71.350025] FS: 0000000000000000(0000) GS:ffff88813bd80000(0000) knlGS:0000000000000000
[ 71.350026] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 71.350027] CR2: 000055ac683710f0 CR3: 000000013a222003 CR4: 0000000000360ee0
[ 71.350028] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 71.350028] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 71.350030] Kernel panic - not syncing: Fatal exception
[ 71.350412] Kernel Offset: disabled
[ 71.365922] ---[ end Kernel panic - not syncing: Fatal exception ]---
which is caused by dangling elements that have been deactivated, but
never removed.
On a flush operation, nft_pipapo_walk() walks through all the elements
in the mapping table, which are then deactivated by nft_flush_set(),
one by one, and added to the commit list for removal. Element data is
then freed.
On transaction commit, nft_pipapo_remove() is called, and failed to
remove these elements, leading to the stale references in the mapping.
The first symptom of this, revealed by KASan, is a one-byte
use-after-free in subsequent calls to nft_pipapo_walk(), which is
usually not enough to trigger a panic. When stale elements are used
more heavily, though, such as double-free via nft_pipapo_destroy()
as in Phil's case, the problem becomes more noticeable.
The issue comes from that fact that, on a flush operation,
nft_pipapo_remove() won't get the actual key data via elem->key,
elements to be deleted upon commit won't be found by the lookup via
pipapo_get(), and removal will be skipped. Key data should be fetched
via nft_set_ext_key(), instead.
Reported-by: Phil Sutter <phil@nwl.cc>
Fixes: 3c4287f620 ("nf_tables: Add set type for arbitrary concatenation of ranges")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
I originally used unlikely() in the if (match_only) clause, which
we hit on the mapping table for the last field in a set, to ensure
we avoid branching to the rest of for loop body, which is executed
more frequently.
However, Pablo reports, this is confusing as it gives the impression
that this is not a common case, and it's actually not the intended
usage of unlikely().
I couldn't observe any statistical difference in matching rates on
x864_64 and aarch64 without it, so just drop it.
Reported-by: Pablo Neira Ayuso <pablo@netfilter.org>
Fixes: 3c4287f620 ("nf_tables: Add set type for arbitrary concatenation of ranges")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
In both insertion and lookup examples, the two element pointers
of rule mapping tables were swapped. Fix that.
Reported-by: Pablo Neira Ayuso <pablo@netfilter.org>
Fixes: 3c4287f620 ("nf_tables: Add set type for arbitrary concatenation of ranges")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This new set type allows for intervals in concatenated fields,
which are expressed in the usual way, that is, simple byte
concatenation with padding to 32 bits for single fields, and
given as ranges by specifying start and end elements containing,
each, the full concatenation of start and end values for the
single fields.
Ranges are expanded to composing netmasks, for each field: these
are inserted as rules in per-field lookup tables. Bits to be
classified are divided in 4-bit groups, and for each group, the
lookup table contains 4^2 buckets, representing all the possible
values of a bit group. This approach was inspired by the Grouper
algorithm:
http://www.cse.usf.edu/~ligatti/projects/grouper/
Matching is performed by a sequence of AND operations between
bucket values, with buckets selected according to the value of
packet bits, for each group. The result of this sequence tells
us which rules matched for a given field.
In order to concatenate several ranged fields, per-field rules
are mapped using mapping arrays, one per field, that specify
which rules should be considered while matching the next field.
The mapping array for the last field contains a reference to
the element originally inserted.
The notes in nft_set_pipapo.c cover the algorithm in deeper
detail.
A pure hash-based approach is of no use here, as ranges need
to be classified. An implementation based on "proxying" the
existing red-black tree set type, creating a tree for each
field, was considered, but deemed impractical due to the fact
that elements would need to be shared between trees, at least
as long as we want to keep UAPI changes to a minimum.
A stand-alone implementation of this algorithm is available at:
https://pipapo.lameexcu.se
together with notes about possible future optimisations
(in pipapo.c).
This algorithm was designed with data locality in mind, and can
be highly optimised for SIMD instruction sets, as the bulk of
the matching work is done with repetitive, simple bitwise
operations.
At this point, without further optimisations, nft_concat_range.sh
reports, for one AMD Epyc 7351 thread (2.9GHz, 512 KiB L1D$, 8 MiB
L2$):
TEST: performance
net,port [ OK ]
baseline (drop from netdev hook): 10190076pps
baseline hash (non-ranged entries): 6179564pps
baseline rbtree (match on first field only): 2950341pps
set with 1000 full, ranged entries: 2304165pps
port,net [ OK ]
baseline (drop from netdev hook): 10143615pps
baseline hash (non-ranged entries): 6135776pps
baseline rbtree (match on first field only): 4311934pps
set with 100 full, ranged entries: 4131471pps
net6,port [ OK ]
baseline (drop from netdev hook): 9730404pps
baseline hash (non-ranged entries): 4809557pps
baseline rbtree (match on first field only): 1501699pps
set with 1000 full, ranged entries: 1092557pps
port,proto [ OK ]
baseline (drop from netdev hook): 10812426pps
baseline hash (non-ranged entries): 6929353pps
baseline rbtree (match on first field only): 3027105pps
set with 30000 full, ranged entries: 284147pps
net6,port,mac [ OK ]
baseline (drop from netdev hook): 9660114pps
baseline hash (non-ranged entries): 3778877pps
baseline rbtree (match on first field only): 3179379pps
set with 10 full, ranged entries: 2082880pps
net6,port,mac,proto [ OK ]
baseline (drop from netdev hook): 9718324pps
baseline hash (non-ranged entries): 3799021pps
baseline rbtree (match on first field only): 1506689pps
set with 1000 full, ranged entries: 783810pps
net,mac [ OK ]
baseline (drop from netdev hook): 10190029pps
baseline hash (non-ranged entries): 5172218pps
baseline rbtree (match on first field only): 2946863pps
set with 1000 full, ranged entries: 1279122pps
v4:
- fix build for 32-bit architectures: 64-bit division needs
div_u64() (kbuild test robot <lkp@intel.com>)
v3:
- rework interface for field length specification,
NFT_SET_SUBKEY disappears and information is stored in
description
- remove scratch area to store closing element of ranges,
as elements now come with an actual attribute to specify
the upper range limit (Pablo Neira Ayuso)
- also remove pointer to 'start' element from mapping table,
closing key is now accessible via extension data
- use bytes right away instead of bits for field lengths,
this way we can also double the inner loop of the lookup
function to take care of upper and lower bits in a single
iteration (minor performance improvement)
- make it clearer that set operations are actually atomic
API-wise, but we can't e.g. implement flush() as one-shot
action
- fix type for 'dup' in nft_pipapo_insert(), check for
duplicates only in the next generation, and in general take
care of differentiating generation mask cases depending on
the operation (Pablo Neira Ayuso)
- report C implementation matching rate in commit message, so
that AVX2 implementation can be compared (Pablo Neira Ayuso)
v2:
- protect access to scratch maps in nft_pipapo_lookup() with
local_bh_disable/enable() (Florian Westphal)
- drop rcu_read_lock/unlock() from nft_pipapo_lookup(), it's
already implied (Florian Westphal)
- explain why partial allocation failures don't need handling
in pipapo_realloc_scratch(), rename 'm' to clone and update
related kerneldoc to make it clear we're not operating on
the live copy (Florian Westphal)
- add expicit check for priv->start_elem in
nft_pipapo_insert() to avoid ending up in nft_pipapo_walk()
with a NULL start element, and also zero it out in every
operation that might make it invalid, so that insertion
doesn't proceed with an invalid element (Florian Westphal)
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>