We want to re-organize the struct sock layout. The sk_peek_off
field location is problematic, as most protocols want it in the
RX read area, while UDP wants it on a cacheline different from
sk_receive_queue.
Create a local (inside udp_sock) copy of the 'peek offset is enabled'
flag and place it inside the same cacheline of reader_queue.
Check such flag before reading sk_peek_off. This will save potential
false sharing and cache misses in the fast-path.
Tested under UDP flood with small packets. The struct sock layout
update causes a 4% performance drop, and this patch restores completely
the original tput.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://lore.kernel.org/r/67ab679c15fbf49fa05b3ffe05d91c47ab84f147.1708426665.git.pabeni@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Use the new KMEM_CACHE() macro instead of direct kmem_cache_create
to simplify the creation of SLAB caches.
And change cache name from 'ip_dst_cache' to 'rtable'.
Signed-off-by: Kunwu Chan <chentao@kylinos.cn>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use the new KMEM_CACHE() macro instead of direct kmem_cache_create
to simplify the creation of SLAB caches.
And change cache name from 'ip_mrt_cache' to 'mfc_cache'.
Signed-off-by: Kunwu Chan <chentao@kylinos.cn>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The variable len being initialized with a value that is never read, an
if statement is initializing it in both paths of the if statement.
The initialization is redundant and can be removed.
Cleans up clang scan build warning:
net/ipv4/tcp_ao.c:512:11: warning: Value stored to 'len' during its
initialization is never read [deadcode.DeadStores]
Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Reviewed-by: Dmitry Safonov <0x7f454c46@gmail.com>
Link: https://lore.kernel.org/r/20240216125443.2107244-1-colin.i.king@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
When SOF_TIMESTAMPING_OPT_ID is used to ambiguate timestamped datagrams,
the sk_tskey can become unpredictable in case of any error happened
during sendmsg(). Move increment later in the code and make decrement of
sk_tskey in error path. This solution is still racy in case of multiple
threads doing snedmsg() over the very same socket in parallel, but still
makes error path much more predictable.
Fixes: 09c2d251b7 ("net-timestamp: add key to disambiguate concurrent datagrams")
Reported-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Vadim Fedorenko <vadfed@meta.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://lore.kernel.org/r/20240213110428.1681540-1-vadfed@meta.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
t->parms.link is read locklessly, annotate these reads
and opposite writes accordingly.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Set scope automatically in ip_route_output_ports() (using the socket
SOCK_LOCALROUTE flag). This way, callers don't have to overload the
tos with the RTO_ONLINK flag, like RT_CONN_FLAGS() does.
For callers that don't pass a struct sock, this doesn't change anything
as the scope is still set to RT_SCOPE_UNIVERSE when sk is NULL.
Callers that passed a struct sock and used RT_CONN_FLAGS(sk) or
RT_CONN_FLAGS_TOS(sk, tos) for the tos are modified to use
ip_sock_tos(sk) and RT_TOS(tos) respectively, as overloading tos with
the RTO_ONLINK flag now becomes unnecessary.
In drivers/net/amt.c, all ip_route_output_ports() calls use a 0 tos
parameter, ignoring the SOCK_LOCALROUTE flag of the socket. But the sk
parameter is a kernel socket, which doesn't have any configuration path
for setting SOCK_LOCALROUTE anyway. Therefore, ip_route_output_ports()
will continue to initialise scope with RT_SCOPE_UNIVERSE and amt.c
doesn't need to be modified.
Also, remove RT_CONN_FLAGS() and RT_CONN_FLAGS_TOS() from route.h as
these macros are now unused.
The objective is to eventually remove RTO_ONLINK entirely to allow
converting ->flowi4_tos to dscp_t. This will ensure proper isolation
between the DSCP and ECN bits, thus minimising the risk of introducing
bugs where TOS values interfere with ECN.
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/dacfd2ab40685e20959ab7b53c427595ba229e7d.1707496938.git.gnault@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tnode_free() should use synchronize_net()
instead of syncronize_rcu() to release RTNL sooner.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Cross-merge networking fixes after downstream PR.
No conflicts.
Adjacent changes:
drivers/net/ethernet/stmicro/stmmac/common.h
38cc3c6dcc ("net: stmmac: protect updates of 64-bit statistics counters")
fd5a6a7131 ("net: stmmac: est: Per Tx-queue error count for HLBF")
c5c3e1bfc9 ("net: stmmac: Offload queueMaxSDU from tc-taprio")
drivers/net/wireless/microchip/wilc1000/netdev.c
c901388028 ("wifi: fill in MODULE_DESCRIPTION()s for wilc1000")
328efda22a ("wifi: wilc1000: do not realloc workqueue everytime an interface is added")
net/unix/garbage.c
11498715f2 ("af_unix: Remove io_uring code for GC.")
1279f9d9de ("af_unix: Call kfree_skb() for dead unix_(sk)->oob_skb in GC.")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
inet_recv_error() is called without holding the socket lock.
IPv6 socket could mutate to IPv4 with IPV6_ADDRFORM
socket option and trigger a KCSAN warning.
Fixes: f4713a3dfa ("net-timestamp: make tcp_recvmsg call ipv6_recv_error for AF_INET6 socks")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the ICMPv6 error is built from a non-linear skb we get the following
splat,
BUG: KASAN: slab-out-of-bounds in do_csum+0x220/0x240
Read of size 4 at addr ffff88811d402c80 by task netperf/820
CPU: 0 PID: 820 Comm: netperf Not tainted 6.8.0-rc1+ #543
...
kasan_report+0xd8/0x110
do_csum+0x220/0x240
csum_partial+0xc/0x20
skb_tunnel_check_pmtu+0xeb9/0x3280
vxlan_xmit_one+0x14c2/0x4080
vxlan_xmit+0xf61/0x5c00
dev_hard_start_xmit+0xfb/0x510
__dev_queue_xmit+0x7cd/0x32a0
br_dev_queue_push_xmit+0x39d/0x6a0
Use skb_checksum instead of csum_partial who cannot deal with non-linear
SKBs.
Fixes: 4cb47a8644 ("tunnels: PMTU discovery support for directly bridged IP packets")
Signed-off-by: Antoine Tenart <atenart@kernel.org>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Florian Westphal says:
====================
nf-next pr 2024-01-29
This batch contains updates for your *next* tree.
First three changes, from Phil Sutter, allow userspace to define
a table that is exclusively owned by a daemon (via netlink socket
aliveness) without auto-removing this table when the userspace program
exits. Such table gets marked as orphaned and a restarting management
daemon may re-attach/reassume ownership.
Next patch, from Pablo, passes already-validated flags variable around
rather than having called code re-fetch it from netlnik message.
Patches 5 and 6 update ipvs and nf_conncount to use the recently
introduced KMEM_CACHE() macro.
Last three patches, from myself, tweak kconfig logic a little to
permit a kernel configuration that can run iptables-over-nftables
but not classic (setsockopt) iptables.
Such builds lack the builtin-filter/mangle/raw/nat/security tables,
the set/getsockopt interface and the "old blob format"
interpreter/traverser. For now, this is 'oldconfig friendly', users
need to manually deselect existing config options for this.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
For now, the packet with an old ack is not accepted if we are in
FIN_WAIT1 state, which can cause retransmission. Taking the following
case as an example:
Client Server
| |
FIN_WAIT1(Send FIN, seq=10) FIN_WAIT1(Send FIN, seq=20, ack=10)
| |
| Send ACK(seq=21, ack=11)
Recv ACK(seq=21, ack=11)
|
Recv FIN(seq=20, ack=10)
In the case above, simultaneous close is happening, and the FIN and ACK
packet that send from the server is out of order. Then, the FIN will be
dropped by the client, as it has an old ack. Then, the server has to
retransmit the FIN, which can cause delay if the server has set the
SO_LINGER on the socket.
Old ack is accepted in the ESTABLISHED and TIME_WAIT state, and I think
it should be better to keep the same logic.
In this commit, we accept old ack in FIN_WAIT1/FIN_WAIT2/CLOSING/LAST_ACK
states. Maybe we should limit it to FIN_WAIT1 for now?
Signed-off-by: Menglong Dong <menglong8.dong@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20240126040519.1846345-1-menglong8.dong@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add hidden IP(6)_NF_IPTABLES_LEGACY symbol.
When any of the "old" builtin tables are enabled the "old" iptables
interface will be supported.
To disable the old set/getsockopt interface the existing options
for the builtin tables need to be turned off:
CONFIG_IP_NF_IPTABLES=m
CONFIG_IP_NF_FILTER is not set
CONFIG_IP_NF_NAT is not set
CONFIG_IP_NF_MANGLE is not set
CONFIG_IP_NF_RAW is not set
CONFIG_IP_NF_SECURITY is not set
Same for CONFIG_IP6_NF_ variants.
This allows to build a kernel that only supports ip(6)tables-nft
(iptables-over-nftables api).
In the future the _LEGACY symbol will become visible and the select
statements will be turned into 'depends on', but for now be on safe side
so "make oldconfig" won't break things.
Signed-off-by: Florian Westphal <fw@strlen.de>
Allows to build kernel that supports the arptables mangle target
via nftables' compat infra but without the arptables get/setsockopt
interface or the old arptables filter interpreter.
IOW, setting IP_NF_ARPFILTER=n will break arptables-legacy, but
arptables-nft will continue to work as long as nftables compat
support is enabled.
Signed-off-by: Florian Westphal <fw@strlen.de>
Reviewed-by: Phil Sutter <phil@nwl.cc>
TCP rx zerocopy intent is to map pages initially allocated
from NIC drivers, not pages owned by a fs.
This patch adds to can_map_frag() these additional checks:
- Page must not be a compound one.
- page->mapping must be NULL.
This fixes the panic reported by ZhangPeng.
syzbot was able to loopback packets built with sendfile(),
mapping pages owned by an ext4 file to TCP rx zerocopy.
r3 = socket$inet_tcp(0x2, 0x1, 0x0)
mmap(&(0x7f0000ff9000/0x4000)=nil, 0x4000, 0x0, 0x12, r3, 0x0)
r4 = socket$inet_tcp(0x2, 0x1, 0x0)
bind$inet(r4, &(0x7f0000000000)={0x2, 0x4e24, @multicast1}, 0x10)
connect$inet(r4, &(0x7f00000006c0)={0x2, 0x4e24, @empty}, 0x10)
r5 = openat$dir(0xffffffffffffff9c, &(0x7f00000000c0)='./file0\x00',
0x181e42, 0x0)
fallocate(r5, 0x0, 0x0, 0x85b8)
sendfile(r4, r5, 0x0, 0x8ba0)
getsockopt$inet_tcp_TCP_ZEROCOPY_RECEIVE(r4, 0x6, 0x23,
&(0x7f00000001c0)={&(0x7f0000ffb000/0x3000)=nil, 0x3000, 0x0, 0x0, 0x0,
0x0, 0x0, 0x0, 0x0}, &(0x7f0000000440)=0x40)
r6 = openat$dir(0xffffffffffffff9c, &(0x7f00000000c0)='./file0\x00',
0x181e42, 0x0)
Fixes: 93ab6cc691 ("tcp: implement mmap() for zero copy receive")
Link: https://lore.kernel.org/netdev/5106a58e-04da-372a-b836-9d3d0bd2507b@huawei.com/T/
Reported-and-bisected-by: ZhangPeng <zhangpeng362@huawei.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Arjun Roy <arjunroy@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: linux-mm@vger.kernel.org
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Daniel Borkmann says:
====================
pull-request: bpf-next 2024-01-26
We've added 107 non-merge commits during the last 4 day(s) which contain
a total of 101 files changed, 6009 insertions(+), 1260 deletions(-).
The main changes are:
1) Add BPF token support to delegate a subset of BPF subsystem
functionality from privileged system-wide daemons such as systemd
through special mount options for userns-bound BPF fs to a trusted
& unprivileged application. With addressed changes from Christian
and Linus' reviews, from Andrii Nakryiko.
2) Support registration of struct_ops types from modules which helps
projects like fuse-bpf that seeks to implement a new struct_ops type,
from Kui-Feng Lee.
3) Add support for retrieval of cookies for perf/kprobe multi links,
from Jiri Olsa.
4) Bigger batch of prep-work for the BPF verifier to eventually support
preserving boundaries and tracking scalars on narrowing fills,
from Maxim Mikityanskiy.
5) Extend the tc BPF flavor to support arbitrary TCP SYN cookies to help
with the scenario of SYN floods, from Kuniyuki Iwashima.
6) Add code generation to inline the bpf_kptr_xchg() helper which
improves performance when stashing/popping the allocated BPF objects,
from Hou Tao.
7) Extend BPF verifier to track aligned ST stores as imprecise spilled
registers, from Yonghong Song.
8) Several fixes to BPF selftests around inline asm constraints and
unsupported VLA code generation, from Jose E. Marchesi.
9) Various updates to the BPF IETF instruction set draft document such
as the introduction of conformance groups for instructions,
from Dave Thaler.
10) Fix BPF verifier to make infinite loop detection in is_state_visited()
exact to catch some too lax spill/fill corner cases,
from Eduard Zingerman.
11) Refactor the BPF verifier pointer ALU check to allow ALU explicitly
instead of implicitly for various register types, from Hao Sun.
12) Fix the flaky tc_redirect_dtime BPF selftest due to slowness
in neighbor advertisement at setup time, from Martin KaFai Lau.
13) Change BPF selftests to skip callback tests for the case when the
JIT is disabled, from Tiezhu Yang.
14) Add a small extension to libbpf which allows to auto create
a map-in-map's inner map, from Andrey Grafin.
* tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next: (107 commits)
selftests/bpf: Add missing line break in test_verifier
bpf, docs: Clarify definitions of various instructions
bpf: Fix error checks against bpf_get_btf_vmlinux().
bpf: One more maintainer for libbpf and BPF selftests
selftests/bpf: Incorporate LSM policy to token-based tests
selftests/bpf: Add tests for LIBBPF_BPF_TOKEN_PATH envvar
libbpf: Support BPF token path setting through LIBBPF_BPF_TOKEN_PATH envvar
selftests/bpf: Add tests for BPF object load with implicit token
selftests/bpf: Add BPF object loading tests with explicit token passing
libbpf: Wire up BPF token support at BPF object level
libbpf: Wire up token_fd into feature probing logic
libbpf: Move feature detection code into its own file
libbpf: Further decouple feature checking logic from bpf_object
libbpf: Split feature detectors definitions from cached results
selftests/bpf: Utilize string values for delegate_xxx mount options
bpf: Support symbolic BPF FS delegation mount options
bpf: Fail BPF_TOKEN_CREATE if no delegation option was set on BPF FS
bpf,selinux: Allocate bpf_security_struct per BPF token
selftests/bpf: Add BPF token-enabled tests
libbpf: Add BPF token support to bpf_prog_load() API
...
====================
Link: https://lore.kernel.org/r/20240126215710.19855-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Instead of performing unconditional system-wide bpf_capable() and
perfmon_capable() calls inside bpf_base_func_proto() function (and other
similar ones) to determine eligibility of a given BPF helper for a given
program, use previously recorded BPF token during BPF_PROG_LOAD command
handling to inform the decision.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20240124022127.2379740-8-andrii@kernel.org
We will support arbitrary SYN Cookie with BPF in the following
patch.
If BPF prog validates ACK and kfunc allocates a reqsk, it will
be carried to cookie_[46]_check() as skb->sk. If skb->sk is not
NULL, we call cookie_bpf_check().
Then, we clear skb->sk and skb->destructor, which are needed not
to hold refcnt for reqsk and the listener. See the following patch
for details.
After that, we finish initialisation for the remaining fields with
cookie_tcp_reqsk_init().
Note that the server side WScale is set only for non-BPF SYN Cookie.
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20240115205514.68364-5-kuniyu@amazon.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
We will support arbitrary SYN Cookie with BPF.
When BPF prog validates ACK and kfunc allocates a reqsk, we need
to call tcp_ns_to_ts() to calculate an offset of TSval for later
use:
time
t0 : Send SYN+ACK
-> tsval = Initial TSval (Random Number)
t1 : Recv ACK of 3WHS
-> tsoff = TSecr - tcp_ns_to_ts(usec_ts_ok, tcp_clock_ns())
= Initial TSval - t1
t2 : Send ACK
-> tsval = t2 + tsoff
= Initial TSval + (t2 - t1)
= Initial TSval + Time Delta (x)
(x) Note that the time delta does not include the initial RTT
from t0 to t1.
Let's move tcp_ns_to_ts() to tcp.h.
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20240115205514.68364-2-kuniyu@amazon.com
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
After the removal of inet_diag_table_mutex, sock_diag_table_mutex
and sock_diag_mutex, I was able so see spinlock contention from
inet_diag_dump_icsk() when running 100 parallel invocations.
It is time to skip over empty buckets.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
TCPDIAG_GETSOCK and DCCPDIAG_GETSOCK diag are serialized
on sock_diag_table_mutex.
This is to make sure inet_diag module is not unloaded
while diag was ongoing.
It is time to get rid of this mutex and use RCU protection,
allowing full parallelism.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Following patch is going to use RCU instead of
sock_diag_table_mutex acquisition.
This patch is a preparation, no change of behavior yet.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
inet_diag_lock_handler() current implementation uses a mutex
to protect inet_diag_table[] array against concurrent changes.
This makes inet_diag dump serialized, thus less scalable
than legacy /proc files.
It is time to switch to full RCU protection.
As a bonus, if a target is statically linked instead of being
modular, inet_diag_lock_handler() & inet_diag_unlock_handler()
reduce to reads only.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Following patch is going to use RCU instead of
inet_diag_table_mutex acquisition.
This patch is a preparation, no change of behavior yet.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Guillaume Nault <gnault@redhat.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
On CPUs with weak memory models, reads and updates performed by tcp_push
to the sk variables can get reordered leaving the socket throttled when
it should not. The tasklet running tcp_wfree() may also not observe the
memory updates in time and will skip flushing any packets throttled by
tcp_push(), delaying the sending. This can pathologically cause 40ms
extra latency due to bad interactions with delayed acks.
Adding a memory barrier in tcp_push removes the bug, similarly to the
previous commit bf06200e73 ("tcp: tsq: fix nonagle handling").
smp_mb__after_atomic() is used to not incur in unnecessary overhead
on x86 since not affected.
Patch has been tested using an AWS c7g.2xlarge instance with Ubuntu
22.04 and Apache Tomcat 9.0.83 running the basic servlet below:
import java.io.IOException;
import java.io.OutputStreamWriter;
import java.io.PrintWriter;
import javax.servlet.ServletException;
import javax.servlet.http.HttpServlet;
import javax.servlet.http.HttpServletRequest;
import javax.servlet.http.HttpServletResponse;
public class HelloWorldServlet extends HttpServlet {
@Override
protected void doGet(HttpServletRequest request, HttpServletResponse response)
throws ServletException, IOException {
response.setContentType("text/html;charset=utf-8");
OutputStreamWriter osw = new OutputStreamWriter(response.getOutputStream(),"UTF-8");
String s = "a".repeat(3096);
osw.write(s,0,s.length());
osw.flush();
}
}
Load was applied using wrk2 (https://github.com/kinvolk/wrk2) from an AWS
c6i.8xlarge instance. Before the patch an additional 40ms latency from P99.99+
values is observed while, with the patch, the extra latency disappears.
No patch and tcp_autocorking=1
./wrk -t32 -c128 -d40s --latency -R10000 http://172.31.60.173:8080/hello/hello
...
50.000% 0.91ms
75.000% 1.13ms
90.000% 1.46ms
99.000% 1.74ms
99.900% 1.89ms
99.990% 41.95ms <<< 40+ ms extra latency
99.999% 48.32ms
100.000% 48.96ms
With patch and tcp_autocorking=1
./wrk -t32 -c128 -d40s --latency -R10000 http://172.31.60.173:8080/hello/hello
...
50.000% 0.90ms
75.000% 1.13ms
90.000% 1.45ms
99.000% 1.72ms
99.900% 1.83ms
99.990% 2.11ms <<< no 40+ ms extra latency
99.999% 2.53ms
100.000% 2.62ms
Patch has been also tested on x86 (m7i.2xlarge instance) which it is not
affected by this issue and the patch doesn't introduce any additional
delay.
Fixes: 7aa5470c2c ("tcp: tsq: move tsq_flags close to sk_wmem_alloc")
Signed-off-by: Salvatore Dipietro <dipiets@amazon.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20240119190133.43698-1-dipiets@amazon.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Pablo Neira Ayuso says:
====================
Netfilter fixes for net
The following batch contains Netfilter fixes for net. Slightly larger
than usual because this batch includes several patches to tighten the
nf_tables control plane to reject inconsistent configuration:
1) Restrict NFTA_SET_POLICY to NFT_SET_POL_PERFORMANCE and
NFT_SET_POL_MEMORY.
2) Bail out if a nf_tables expression registers more than 16 netlink
attributes which is what struct nft_expr_info allows.
3) Bail out if NFT_EXPR_STATEFUL provides no .clone interface, remove
existing fallback to memcpy() when cloning which might accidentally
duplicate memory reference to the same object.
4) Fix br_netfilter interaction with neighbour layer. This requires
three preparation patches:
- Use nf_bridge_get_physinif() in nfnetlink_log
- Use nf_bridge_info_exists() to check in br_netfilter context
is available in nf_queue.
- Pass net to nf_bridge_get_physindev()
And finally, the fix which replaces physindev with physinif
in nf_bridge_info.
Patches from Pavel Tikhomirov.
5) Catch-all deactivation happens in the transaction, hence this
oneliner to check for the next generation. This bug uncovered after
the removal of the _BUSY bit, which happened in set elements back in
summer 2023.
6) Ensure set (total) key length size and concat field length description
is consistent, otherwise bail out.
7) Skip set element with the _DEAD flag on from the netlink dump path.
A tests occasionally shows that dump is mismatching because GC might
lose race to get rid of this element while a netlink dump is in
progress.
8) Reject NFT_SET_CONCAT for field_count < 1.
9) Use IP6_INC_STATS in ipvs to fix preemption BUG splat, patch
from Fedor Pchelkin.
* tag 'nf-24-01-18' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf:
ipvs: avoid stat macros calls from preemptible context
netfilter: nf_tables: reject NFT_SET_CONCAT with not field length description
netfilter: nf_tables: skip dead set elements in netlink dump
netfilter: nf_tables: do not allow mismatch field size and set key length
netfilter: nf_tables: check if catch-all set element is active in next generation
netfilter: bridge: replace physindev with physinif in nf_bridge_info
netfilter: propagate net to nf_bridge_get_physindev
netfilter: nf_queue: remove excess nf_bridge variable
netfilter: nfnetlink_log: use proper helper for fetching physinif
netfilter: nft_limit: do not ignore unsupported flags
netfilter: nf_tables: bail out if stateful expression provides no .clone
netfilter: nf_tables: validate .maxattr at expression registration
netfilter: nf_tables: reject invalid set policy
====================
Link: https://lore.kernel.org/r/20240118161726.14838-1-pablo@netfilter.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
An skb can be added to a neigh->arp_queue while waiting for an arp
reply. Where original skb's skb->dev can be different to neigh's
neigh->dev. For instance in case of bridging dnated skb from one veth to
another, the skb would be added to a neigh->arp_queue of the bridge.
As skb->dev can be reset back to nf_bridge->physindev and used, and as
there is no explicit mechanism that prevents this physindev from been
freed under us (for instance neigh_flush_dev doesn't cleanup skbs from
different device's neigh queue) we can crash on e.g. this stack:
arp_process
neigh_update
skb = __skb_dequeue(&neigh->arp_queue)
neigh_resolve_output(..., skb)
...
br_nf_dev_xmit
br_nf_pre_routing_finish_bridge_slow
skb->dev = nf_bridge->physindev
br_handle_frame_finish
Let's use plain ifindex instead of net_device link. To peek into the
original net_device we will use dev_get_by_index_rcu(). Thus either we
get device and are safe to use it or we don't get it and drop skb.
Fixes: c4e70a87d9 ("netfilter: bridge: rename br_netfilter.c to br_netfilter_hooks.c")
Suggested-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This is a preparation patch for replacing physindev with physinif on
nf_bridge_info structure. We will use dev_get_by_index_rcu to resolve
device, when needed, and it requires net to be available.
Signed-off-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
There is a bug in the bpf_iter_udp_batch() function that stops
the userspace from making forward progress.
The case that triggers the bug is the userspace passed in
a very small read buffer. When the bpf prog does bpf_seq_printf,
the userspace read buffer is not enough to capture the whole bucket.
When the read buffer is not large enough, the kernel will remember
the offset of the bucket in iter->offset such that the next userspace
read() can continue from where it left off.
The kernel will skip the number (== "iter->offset") of sockets in
the next read(). However, the code directly decrements the
"--iter->offset". This is incorrect because the next read() may
not consume the whole bucket either and then the next-next read()
will start from offset 0. The net effect is the userspace will
keep reading from the beginning of a bucket and the process will
never finish. "iter->offset" must always go forward until the
whole bucket is consumed.
This patch fixes it by using a local variable "resume_offset"
and "resume_bucket". "iter->offset" is always reset to 0 before
it may be used. "iter->offset" will be advanced to the
"resume_offset" when it continues from the "resume_bucket" (i.e.
"state->bucket == resume_bucket"). This brings it closer to
the bpf_iter_tcp's offset handling which does not suffer
the same bug.
Cc: Aditi Ghag <aditi.ghag@isovalent.com>
Fixes: c96dac8d36 ("bpf: udp: Implement batching for sockets iterator")
Acked-by: Yonghong Song <yonghong.song@linux.dev>
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Reviewed-by: Aditi Ghag <aditi.ghag@isovalent.com>
Link: https://lore.kernel.org/r/20240112190530.3751661-3-martin.lau@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The current logic is to use a default size 16 to batch the whole bucket.
If it is too small, it will retry with a larger batch size.
The current code accidentally does a state->bucket-- before retrying.
This goes back to retry with the previous bucket which has already
been done. This patch fixed it.
It is hard to create a selftest. I added a WARN_ON(state->bucket < 0),
forced a particular port to be hashed to the first bucket,
created >16 sockets, and observed the for-loop went back
to the "-1" bucket.
Cc: Aditi Ghag <aditi.ghag@isovalent.com>
Fixes: c96dac8d36 ("bpf: udp: Implement batching for sockets iterator")
Acked-by: Yonghong Song <yonghong.song@linux.dev>
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Reviewed-by: Aditi Ghag <aditi.ghag@isovalent.com>
Link: https://lore.kernel.org/r/20240112190530.3751661-2-martin.lau@linux.dev
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
