When skb->len==0, the recv_actor() returns 0 too, but we also use 0
for error conditions. This patch amends this by propagating the errors
to tcp_read_skb() so that we can distinguish skb->len==0 case from
error cases.
Fixes: 04919bed94 ("tcp: Introduce tcp_read_skb()")
Reported-by: Eric Dumazet <edumazet@google.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Cong Wang <cong.wang@bytedance.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
As tcp_read_skb() only reads one skb at a time, the while loop is
unnecessary, we can turn it into an if. This also simplifies the
code logic.
Cc: Eric Dumazet <edumazet@google.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Cong Wang <cong.wang@bytedance.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tcp_cleanup_rbuf() retrieves the skb from sk_receive_queue, it
assumes the skb is not yet dequeued. This is no longer true for
tcp_read_skb() case where we dequeue the skb first.
Fix this by introducing a helper __tcp_cleanup_rbuf() which does
not require any skb and calling it in tcp_read_skb().
Fixes: 04919bed94 ("tcp: Introduce tcp_read_skb()")
Cc: Eric Dumazet <edumazet@google.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Cong Wang <cong.wang@bytedance.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Before commit 965b57b469 ("net: Introduce a new proto_ops
->read_skb()"), skb was not dequeued from receive queue hence
when we close TCP socket skb can be just flushed synchronously.
After this commit, we have to uncharge skb immediately after being
dequeued, otherwise it is still charged in the original sock. And we
still need to retain skb->sk, as eBPF programs may extract sock
information from skb->sk. Therefore, we have to call
skb_set_owner_sk_safe() here.
Fixes: 965b57b469 ("net: Introduce a new proto_ops ->read_skb()")
Reported-and-tested-by: syzbot+a0e6f8738b58f7654417@syzkaller.appspotmail.com
Tested-by: Stanislav Fomichev <sdf@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: Cong Wang <cong.wang@bytedance.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
After the blamed commit, IPv4 SYN packets handled
by a dual stack IPv6 socket are dropped, even if
perfectly valid.
$ nstat | grep MD5
TcpExtTCPMD5Failure 5 0.0
For a dual stack listener, an incoming IPv4 SYN packet
would call tcp_inbound_md5_hash() with @family == AF_INET,
while tp->af_specific is pointing to tcp_sock_ipv6_specific.
Only later when an IPv4-mapped child is created, tp->af_specific
is changed to tcp_sock_ipv6_mapped_specific.
Fixes: 7bbb765b73 ("net/tcp: Merge TCP-MD5 inbound callbacks")
Reported-by: Brian Vazquez <brianvv@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Dmitry Safonov <dima@arista.com>
Tested-by: Leonard Crestez <cdleonard@gmail.com>
Link: https://lore.kernel.org/r/20220726115743.2759832-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Expose TCP rx queue accessor and cleanup, so that TLS can
decrypt directly from the TCP queue. The expectation
is that the caller can access the skb returned from
tcp_recv_skb() and up to inq bytes worth of data (some
of which may be in ->next skbs) and then call
tcp_read_done() when data has been consumed.
The socket lock must be held continuously across
those two operations.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
While reading these sysctl variables, they can be changed concurrently.
Thus, we need to add READ_ONCE() to their readers.
- .sysctl_rmem
- .sysctl_rwmem
- .sysctl_rmem_offset
- .sysctl_wmem_offset
- sysctl_tcp_rmem[1, 2]
- sysctl_tcp_wmem[1, 2]
- sysctl_decnet_rmem[1]
- sysctl_decnet_wmem[1]
- sysctl_tipc_rmem[1]
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
While reading sysctl_tcp_autocorking, it can be changed concurrently.
Thus, we need to add READ_ONCE() to its reader.
Fixes: f54b311142 ("tcp: auto corking")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pavel Begunkov says:
====================
io_uring zerocopy send
The patchset implements io_uring zerocopy send. It works with both registered
and normal buffers, mixing is allowed but not recommended. Apart from usual
request completions, just as with MSG_ZEROCOPY, io_uring separately notifies
the userspace when buffers are freed and can be reused (see API design below),
which is delivered into io_uring's Completion Queue. Those "buffer-free"
notifications are not necessarily per request, but the userspace has control
over it and should explicitly attaching a number of requests to a single
notification. The series also adds some internal optimisations when used with
registered buffers like removing page referencing.
From the kernel networking perspective there are two main changes. The first
one is passing ubuf_info into the network layer from io_uring (inside of an
in kernel struct msghdr). This allows extra optimisations, e.g. ubuf_info
caching on the io_uring side, but also helps to avoid cross-referencing
and synchronisation problems. The second part is an optional optimisation
removing page referencing for requests with registered buffers.
Benchmarking UDP with an optimised version of the selftest (see [1]), which
sends a bunch of requests, waits for completions and repeats. "+ flush" column
posts one additional "buffer-free" notification per request, and just "zc"
doesn't post buffer notifications at all.
NIC (requests / second):
IO size | non-zc | zc | zc + flush
4000 | 495134 | 606420 (+22%) | 558971 (+12%)
1500 | 551808 | 577116 (+4.5%) | 565803 (+2.5%)
1000 | 584677 | 592088 (+1.2%) | 560885 (-4%)
600 | 596292 | 598550 (+0.4%) | 555366 (-6.7%)
dummy (requests / second):
IO size | non-zc | zc | zc + flush
8000 | 1299916 | 2396600 (+84%) | 2224219 (+71%)
4000 | 1869230 | 2344146 (+25%) | 2170069 (+16%)
1200 | 2071617 | 2361960 (+14%) | 2203052 (+6%)
600 | 2106794 | 2381527 (+13%) | 2195295 (+4%)
Previously it also brought a massive performance speedup compared to the
msg_zerocopy tool (see [3]), which is probably not super interesting. There
is also an additional bunch of refcounting optimisations that was omitted from
the series for simplicity and as they don't change the picture drastically,
they will be sent as follow up, as well as flushing optimisations closing the
performance gap b/w two last columns.
For TCP on localhost (with hacks enabling localhost zerocopy) and including
additional overhead for receive:
IO size | non-zc | zc
1200 | 4174 | 4148
4096 | 7597 | 11228
Using a real NIC 1200 bytes, zc is worse than non-zc ~5-10%, maybe the
omitted optimisations will somewhat help, should look better for 4000,
but couldn't test properly because of setup problems.
Links:
liburing (benchmark + tests):
[1] https://github.com/isilence/liburing/tree/zc_v4
kernel repo:
[2] https://github.com/isilence/linux/tree/zc_v4
RFC v1:
[3] https://lore.kernel.org/io-uring/cover.1638282789.git.asml.silence@gmail.com/
RFC v2:
https://lore.kernel.org/io-uring/cover.1640029579.git.asml.silence@gmail.com/
Net patches based:
git@github.com:isilence/linux.git zc_v4-net-base
or
https://github.com/isilence/linux/tree/zc_v4-net-base
API design overview:
The series introduces an io_uring concept of notifactors. From the userspace
perspective it's an entity to which it can bind one or more requests and then
requesting to flush it. Flushing a notifier makes it impossible to attach new
requests to it, and instructs the notifier to post a completion once all
requests attached to it are completed and the kernel doesn't need the buffers
anymore.
Notifications are stored in notification slots, which should be registered as
an array in io_uring. Each slot stores only one notifier at any particular
moment. Flushing removes it from the slot and the slot automatically replaces
it with a new notifier. All operations with notifiers are done by specifying
an index of a slot it's currently in.
When registering a notification the userspace specifies a u64 tag for each
slot, which will be copied in notification completion entries as
cqe::user_data. cqe::res is 0 and cqe::flags is equal to wrap around u32
sequence number counting notifiers of a slot.
====================
Link: https://lore.kernel.org/r/cover.1657643355.git.asml.silence@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Teach tcp how to use external ubuf_info provided in msghdr and
also prepare it for managed frags by sprinkling
skb_zcopy_downgrade_managed() when it could mix managed and not managed
frags.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
While reading sysctl_tcp_fastopen, it can be changed concurrently.
Thus, we need to add READ_ONCE() to its readers.
Fixes: 2100c8d2d9 ("net-tcp: Fast Open base")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
While reading these sysctl knobs, they can be changed concurrently.
Thus, we need to add READ_ONCE() to their readers.
- tcp_retries1
- tcp_retries2
- tcp_orphan_retries
- tcp_fin_timeout
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
While reading sysctl_tcp_reordering, it can be changed concurrently.
Thus, we need to add READ_ONCE() to its readers.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
While reading sysctl_tcp_syn(ack)?_retries, they can be changed
concurrently. Thus, we need to add READ_ONCE() to their readers.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Daniel Borkmann says:
====================
pull-request: bpf-next 2022-07-09
We've added 94 non-merge commits during the last 19 day(s) which contain
a total of 125 files changed, 5141 insertions(+), 6701 deletions(-).
The main changes are:
1) Add new way for performing BTF type queries to BPF, from Daniel Müller.
2) Add inlining of calls to bpf_loop() helper when its function callback is
statically known, from Eduard Zingerman.
3) Implement BPF TCP CC framework usability improvements, from Jörn-Thorben Hinz.
4) Add LSM flavor for attaching per-cgroup BPF programs to existing LSM
hooks, from Stanislav Fomichev.
5) Remove all deprecated libbpf APIs in prep for 1.0 release, from Andrii Nakryiko.
6) Add benchmarks around local_storage to BPF selftests, from Dave Marchevsky.
7) AF_XDP sample removal (given move to libxdp) and various improvements around AF_XDP
selftests, from Magnus Karlsson & Maciej Fijalkowski.
8) Add bpftool improvements for memcg probing and bash completion, from Quentin Monnet.
9) Add arm64 JIT support for BPF-2-BPF coupled with tail calls, from Jakub Sitnicki.
10) Sockmap optimizations around throughput of UDP transmissions which have been
improved by 61%, from Cong Wang.
11) Rework perf's BPF prologue code to remove deprecated functions, from Jiri Olsa.
12) Fix sockmap teardown path to avoid sleepable sk_psock_stop, from John Fastabend.
13) Fix libbpf's cleanup around legacy kprobe/uprobe on error case, from Chuang Wang.
14) Fix libbpf's bpf_helpers.h to work with gcc for the case of its sec/pragma
macro, from James Hilliard.
15) Fix libbpf's pt_regs macros for riscv to use a0 for RC register, from Yixun Lan.
16) Fix bpftool to show the name of type BPF_OBJ_LINK, from Yafang Shao.
* https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next: (94 commits)
selftests/bpf: Fix xdp_synproxy build failure if CONFIG_NF_CONNTRACK=m/n
bpf: Correctly propagate errors up from bpf_core_composites_match
libbpf: Disable SEC pragma macro on GCC
bpf: Check attach_func_proto more carefully in check_return_code
selftests/bpf: Add test involving restrict type qualifier
bpftool: Add support for KIND_RESTRICT to gen min_core_btf command
MAINTAINERS: Add entry for AF_XDP selftests files
selftests, xsk: Rename AF_XDP testing app
bpf, docs: Remove deprecated xsk libbpf APIs description
selftests/bpf: Add benchmark for local_storage RCU Tasks Trace usage
libbpf, riscv: Use a0 for RC register
libbpf: Remove unnecessary usdt_rel_ip assignments
selftests/bpf: Fix few more compiler warnings
selftests/bpf: Fix bogus uninitialized variable warning
bpftool: Remove zlib feature test from Makefile
libbpf: Cleanup the legacy uprobe_event on failed add/attach_event()
libbpf: Fix wrong variable used in perf_event_uprobe_open_legacy()
libbpf: Cleanup the legacy kprobe_event on failed add/attach_event()
selftests/bpf: Add type match test against kernel's task_struct
selftests/bpf: Add nested type to type based tests
...
====================
Link: https://lore.kernel.org/r/20220708233145.32365-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
While reading sysctl_tcp_max_orphans, it can be changed concurrently.
So, we need to add READ_ONCE() to avoid a data-race.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently both splice() and sockmap use ->read_sock() to
read skb from receive queue, but for sockmap we only read
one entire skb at a time, so ->read_sock() is too conservative
to use. Introduce a new proto_ops ->read_skb() which supports
this sematic, with this we can finally pass the ownership of
skb to recv actors.
For non-TCP protocols, all ->read_sock() can be simply
converted to ->read_skb().
Signed-off-by: Cong Wang <cong.wang@bytedance.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20220615162014.89193-3-xiyou.wangcong@gmail.com
This patch inroduces tcp_read_skb() based on tcp_read_sock(),
a preparation for the next patch which actually introduces
a new sock ops.
TCP is special here, because it has tcp_read_sock() which is
mainly used by splice(). tcp_read_sock() supports partial read
and arbitrary offset, neither of them is needed for sockmap.
Signed-off-by: Cong Wang <cong.wang@bytedance.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/20220615162014.89193-2-xiyou.wangcong@gmail.com
Blamed commit only dealt with applications issuing small writes.
Issue here is that we allow to force memory schedule for the sk_buff
allocation, but we have no guarantee that sendmsg() is able to
copy some payload in it.
In this patch, I make sure the socket can use up to tcp_wmem[0] bytes.
For example, if we consider tcp_wmem[0] = 4096 (default on x86),
and initial skb->truesize being 1280, tcp_sendmsg() is able to
copy up to 2816 bytes under memory pressure.
Before this patch a sendmsg() sending more than 2816 bytes
would either block forever (if persistent memory pressure),
or return -EAGAIN.
For bigger MTU networks, it is advised to increase tcp_wmem[0]
to avoid sending too small packets.
v2: deal with zero copy paths.
Fixes: 8e4d980ac2 ("tcp: fix behavior for epoll edge trigger")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: Wei Wang <weiwan@google.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Blamed commit only dealt with applications issuing small writes.
Issue here is that we allow to force memory schedule for the sk_buff
allocation, but we have no guarantee that sendmsg() is able to
copy some payload in it.
In this patch, I make sure the socket can use up to tcp_wmem[0] bytes.
For example, if we consider tcp_wmem[0] = 4096 (default on x86),
and initial skb->truesize being 1280, tcp_sendmsg() is able to
copy up to 2816 bytes under memory pressure.
Before this patch a sendmsg() sending more than 2816 bytes
would either block forever (if persistent memory pressure),
or return -EAGAIN.
For bigger MTU networks, it is advised to increase tcp_wmem[0]
to avoid sending too small packets.
v2: deal with zero copy paths.
Fixes: 8e4d980ac2 ("tcp: fix behavior for epoll edge trigger")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: Wei Wang <weiwan@google.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, tcp_memory_allocated can hit tcp_mem[] limits quite fast.
Each TCP socket can forward allocate up to 2 MB of memory, even after
flow became less active.
10,000 sockets can have reserved 20 GB of memory,
and we have no shrinker in place to reclaim that.
Instead of trying to reclaim the extra allocations in some places,
just keep sk->sk_forward_alloc values as small as possible.
This should not impact performance too much now we have per-cpu
reserves: Changes to tcp_memory_allocated should not be too frequent.
For sockets not using SO_RESERVE_MEM:
- idle sockets (no packets in tx/rx queues) have zero forward alloc.
- non idle sockets have a forward alloc smaller than one page.
Note:
- Removal of SK_RECLAIM_CHUNK and SK_RECLAIM_THRESHOLD
is left to MPTCP maintainers as a follow up.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Each protocol having a ->memory_allocated pointer gets a corresponding
per-cpu reserve, that following patches will use.
Instead of having reserved bytes per socket,
we want to have per-cpu reserves.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Due to memcg interface, SK_MEM_QUANTUM is effectively PAGE_SIZE.
This might change in the future, but it seems better to avoid the
confusion.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We currently have one tcp bind table (bhash) which hashes by port
number only. In the socket bind path, we check for bind conflicts by
traversing the specified port's inet_bind2_bucket while holding the
bucket's spinlock (see inet_csk_get_port() and inet_csk_bind_conflict()).
In instances where there are tons of sockets hashed to the same port
at different addresses, checking for a bind conflict is time-intensive
and can cause softirq cpu lockups, as well as stops new tcp connections
since __inet_inherit_port() also contests for the spinlock.
This patch proposes adding a second bind table, bhash2, that hashes by
port and ip address. Searching the bhash2 table leads to significantly
faster conflict resolution and less time holding the spinlock.
Signed-off-by: Joanne Koong <joannelkoong@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The listen sk is currently stored in two hash tables,
listening_hash (hashed by port) and lhash2 (hashed by port and address).
After commit 0ee58dad5b ("net: tcp6: prefer listeners bound to an address")
and commit d9fbc7f643 ("net: tcp: prefer listeners bound to an address"),
the TCP-SYN lookup fast path does not use listening_hash.
The commit 05c0b35709 ("tcp: seq_file: Replace listening_hash with lhash2")
also moved the seq_file (/proc/net/tcp) iteration usage from
listening_hash to lhash2.
There are still a few listening_hash usages left.
One of them is inet_reuseport_add_sock() which uses the listening_hash
to search a listen sk during the listen() system call. This turns
out to be very slow on use cases that listen on many different
VIPs at a popular port (e.g. 443). [ On top of the slowness in
adding to the tail in the IPv6 case ]. The latter patch has a
selftest to demonstrate this case.
This patch takes this chance to move all remaining listening_hash
usages to lhash2 and then retire listening_hash.
Since most changes need to be done together, it is hard to cut
the listening_hash to lhash2 switch into small patches. The
changes in this patch is highlighted here for the review
purpose.
1. Because of the listening_hash removal, lhash2 can use the
sk->sk_nulls_node instead of the icsk->icsk_listen_portaddr_node.
This will also keep the sk_unhashed() check to work as is
after stop adding sk to listening_hash.
The union is removed from inet_listen_hashbucket because
only nulls_head is needed.
2. icsk->icsk_listen_portaddr_node and its helpers are removed.
3. The current lhash2 users needs to iterate with sk_nulls_node
instead of icsk_listen_portaddr_node.
One case is in the inet[6]_lhash2_lookup().
Another case is the seq_file iterator in tcp_ipv4.c.
One thing to note is sk_nulls_next() is needed
because the old inet_lhash2_for_each_icsk_continue()
does a "next" first before iterating.
4. Move the remaining listening_hash usage to lhash2
inet_reuseport_add_sock() which this series is
trying to improve.
inet_diag.c and mptcp_diag.c are the final two
remaining use cases and is moved to lhash2 now also.
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This is currently done for CMSG_INQ, add an ability to do so via struct
msghdr as well and have CMSG_INQ use that too. If the caller sets
msghdr->msg_get_inq, then we'll pass back the hint in msghdr->msg_inq.
Rearrange struct msghdr a bit so we can add this member while shrinking
it at the same time. On a 64-bit build, it was 96 bytes before this
change and 88 bytes afterwards.
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/650c22ca-cffc-0255-9a05-2413a1e20826@kernel.dk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Logic added in commit f35f821935 ("tcp: defer skb freeing after socket
lock is released") helped bulk TCP flows to move the cost of skbs
frees outside of critical section where socket lock was held.
But for RPC traffic, or hosts with RFS enabled, the solution is far from
being ideal.
For RPC traffic, recvmsg() has to return to user space right after
skb payload has been consumed, meaning that BH handler has no chance
to pick the skb before recvmsg() thread. This issue is more visible
with BIG TCP, as more RPC fit one skb.
For RFS, even if BH handler picks the skbs, they are still picked
from the cpu on which user thread is running.
Ideally, it is better to free the skbs (and associated page frags)
on the cpu that originally allocated them.
This patch removes the per socket anchor (sk->defer_list) and
instead uses a per-cpu list, which will hold more skbs per round.
This new per-cpu list is drained at the end of net_action_rx(),
after incoming packets have been processed, to lower latencies.
In normal conditions, skbs are added to the per-cpu list with
no further action. In the (unlikely) cases where the cpu does not
run net_action_rx() handler fast enough, we use an IPI to raise
NET_RX_SOFTIRQ on the remote cpu.
Also, we do not bother draining the per-cpu list from dev_cpu_dead()
This is because skbs in this list have no requirement on how fast
they should be freed.
Note that we can add in the future a small per-cpu cache
if we see any contention on sd->defer_lock.
Tested on a pair of hosts with 100Gbit NIC, RFS enabled,
and /proc/sys/net/ipv4/tcp_rmem[2] tuned to 16MB to work around
page recycling strategy used by NIC driver (its page pool capacity
being too small compared to number of skbs/pages held in sockets
receive queues)
Note that this tuning was only done to demonstrate worse
conditions for skb freeing for this particular test.
These conditions can happen in more general production workload.
10 runs of one TCP_STREAM flow
Before:
Average throughput: 49685 Mbit.
Kernel profiles on cpu running user thread recvmsg() show high cost for
skb freeing related functions (*)
57.81% [kernel] [k] copy_user_enhanced_fast_string
(*) 12.87% [kernel] [k] skb_release_data
(*) 4.25% [kernel] [k] __free_one_page
(*) 3.57% [kernel] [k] __list_del_entry_valid
1.85% [kernel] [k] __netif_receive_skb_core
1.60% [kernel] [k] __skb_datagram_iter
(*) 1.59% [kernel] [k] free_unref_page_commit
(*) 1.16% [kernel] [k] __slab_free
1.16% [kernel] [k] _copy_to_iter
(*) 1.01% [kernel] [k] kfree
(*) 0.88% [kernel] [k] free_unref_page
0.57% [kernel] [k] ip6_rcv_core
0.55% [kernel] [k] ip6t_do_table
0.54% [kernel] [k] flush_smp_call_function_queue
(*) 0.54% [kernel] [k] free_pcppages_bulk
0.51% [kernel] [k] llist_reverse_order
0.38% [kernel] [k] process_backlog
(*) 0.38% [kernel] [k] free_pcp_prepare
0.37% [kernel] [k] tcp_recvmsg_locked
(*) 0.37% [kernel] [k] __list_add_valid
0.34% [kernel] [k] sock_rfree
0.34% [kernel] [k] _raw_spin_lock_irq
(*) 0.33% [kernel] [k] __page_cache_release
0.33% [kernel] [k] tcp_v6_rcv
(*) 0.33% [kernel] [k] __put_page
(*) 0.29% [kernel] [k] __mod_zone_page_state
0.27% [kernel] [k] _raw_spin_lock
After patch:
Average throughput: 73076 Mbit.
Kernel profiles on cpu running user thread recvmsg() looks better:
81.35% [kernel] [k] copy_user_enhanced_fast_string
1.95% [kernel] [k] _copy_to_iter
1.95% [kernel] [k] __skb_datagram_iter
1.27% [kernel] [k] __netif_receive_skb_core
1.03% [kernel] [k] ip6t_do_table
0.60% [kernel] [k] sock_rfree
0.50% [kernel] [k] tcp_v6_rcv
0.47% [kernel] [k] ip6_rcv_core
0.45% [kernel] [k] read_tsc
0.44% [kernel] [k] _raw_spin_lock_irqsave
0.37% [kernel] [k] _raw_spin_lock
0.37% [kernel] [k] native_irq_return_iret
0.33% [kernel] [k] __inet6_lookup_established
0.31% [kernel] [k] ip6_protocol_deliver_rcu
0.29% [kernel] [k] tcp_rcv_established
0.29% [kernel] [k] llist_reverse_order
v2: kdoc issue (kernel bots)
do not defer if (alloc_cpu == smp_processor_id()) (Paolo)
replace the sk_buff_head with a single-linked list (Jakub)
add a READ_ONCE()/WRITE_ONCE() for the lockless read of sd->defer_list
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Link: https://lore.kernel.org/r/20220422201237.416238-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The internal recvmsg() functions have two parameters 'flags' and 'noblock'
that were merged inside skb_recv_datagram(). As a follow up patch to commit
f4b41f062c ("net: remove noblock parameter from skb_recv_datagram()")
this patch removes the separate 'noblock' parameter for recvmsg().
Analogue to the referenced patch for skb_recv_datagram() the 'flags' and
'noblock' parameters are unnecessarily split up with e.g.
err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
flags & ~MSG_DONTWAIT, &addr_len);
or in
err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg,
sk, msg, size, flags & MSG_DONTWAIT,
flags & ~MSG_DONTWAIT, &addr_len);
instead of simply using only flags all the time and check for MSG_DONTWAIT
where needed (to preserve for the formerly separated no(n)block condition).
Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Link: https://lore.kernel.org/r/20220411124955.154876-1-socketcan@hartkopp.net
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
We had various bugs over the years with code
breaking the assumption that tp->snd_cwnd is greater
than zero.
Lately, syzbot reported the WARN_ON_ONCE(!tp->prior_cwnd) added
in commit 8b8a321ff7 ("tcp: fix zero cwnd in tcp_cwnd_reduction")
can trigger, and without a repro we would have to spend
considerable time finding the bug.
Instead of complaining too late, we want to catch where
and when tp->snd_cwnd is set to an illegal value.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Suggested-by: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Link: https://lore.kernel.org/r/20220405233538.947344-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tcp_should_autocork() is evaluating if it makes senses
to not immediately send current skb, hoping that
user space will add more payload on it by the
time TCP stack reacts to upcoming TX completions.
If current skb got MSG_EOR mark, then we know
that no further data will be added, it is therefore
futile to wait.
SOF_TIMESTAMPING_TX_ACK will become a bit more accurate,
if prior packets are still in qdisc/device queues.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Martin KaFai Lau <kafai@fb.com>
Cc: Willem de Bruijn <willemb@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Link: https://lore.kernel.org/r/20220309054706.2857266-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We have a number of cases where function returns drop/no drop
decision as a boolean. Now that we want to report the reason
code as well we have to pass extra output arguments.
We can make the reason code evaluate correctly as bool.
I believe we're good to reorder the reasons as they are
reported to user space as strings.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
If recv_actor() returns an incorrect value, tcp_read_sock()
might loop forever.
Instead, issue a one time warning and make sure to make progress.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Jakub Sitnicki <jakub@cloudflare.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/r/20220302161723.3910001-2-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The functions do essentially the same work to verify TCP-MD5 sign.
Code can be merged into one family-independent function in order to
reduce copy'n'paste and generated code.
Later with TCP-AO option added, this will allow to create one function
that's responsible for segment verification, that will have all the
different checks for MD5/AO/non-signed packets, which in turn will help
to see checks for all corner-cases in one function, rather than spread
around different families and functions.
Cc: Eric Dumazet <edumazet@google.com>
Cc: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Signed-off-by: Dmitry Safonov <dima@arista.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20220223175740.452397-1-dima@arista.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
syzbot found that mixing sendpage() and sendmsg(MSG_ZEROCOPY)
calls over the same TCP socket would again trigger the
infamous warning in inet_sock_destruct()
WARN_ON(sk_forward_alloc_get(sk));
While Talal took into account a mix of regular copied data
and MSG_ZEROCOPY one in the same skb, the sendpage() path
has been forgotten.
We want the charging to happen for sendpage(), because
pages could be coming from a pipe. What is missing is the
downgrading of pure zerocopy status to make sure
sk_forward_alloc will stay synced.
Add tcp_downgrade_zcopy_pure() helper so that we can
use it from the two callers.
Fixes: 9b65b17db7 ("net: avoid double accounting for pure zerocopy skbs")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: Talal Ahmad <talalahmad@google.com>
Cc: Arjun Roy <arjunroy@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Link: https://lore.kernel.org/r/20220203225547.665114-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We got reports of following warning in inet_sock_destruct()
WARN_ON(sk_forward_alloc_get(sk));
Whenever we add a non zero-copy fragment to a pure zerocopy skb,
we have to anticipate that whole skb->truesize will be uncharged
when skb is finally freed.
skb->data_len is the payload length. But the memory truesize
estimated by __zerocopy_sg_from_iter() is page aligned.
Fixes: 9b65b17db7 ("net: avoid double accounting for pure zerocopy skbs")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Talal Ahmad <talalahmad@google.com>
Cc: Arjun Roy <arjunroy@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Link: https://lore.kernel.org/r/20220201065254.680532-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
sk_gso_max_size is set based on the dst dev. Both users of it
adjust the value by the same offset - (MAX_TCP_HEADER + 1). Rather
than compute the same adjusted value on each call do the adjustment
once when set.
Signed-off-by: David Ahern <dsahern@kernel.org>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20220125024511.27480-1-dsahern@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Without it, splice users can hit the warning
added in commit 79074a72d3 ("net: Flush deferred skb free on socket destroy")
Fixes: f35f821935 ("tcp: defer skb freeing after socket lock is released")
Fixes: 79074a72d3 ("net: Flush deferred skb free on socket destroy")
Suggested-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Gal Pressman <gal@nvidia.com>
Link: https://lore.kernel.org/r/20220120124530.925607-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
