Kazuho Oku reported that setsockopt(SO_INCOMING_CPU) does not work
with setsockopt(SO_REUSEPORT) since v4.6.
With the combination of SO_REUSEPORT and SO_INCOMING_CPU, we could
build a highly efficient server application.
setsockopt(SO_INCOMING_CPU) associates a CPU with a TCP listener
or UDP socket, and then incoming packets processed on the CPU will
likely be distributed to the socket. Technically, a socket could
even receive packets handled on another CPU if no sockets in the
reuseport group have the same CPU receiving the flow.
The logic exists in compute_score() so that a socket will get a higher
score if it has the same CPU with the flow. However, the score gets
ignored after the blamed two commits, which introduced a faster socket
selection algorithm for SO_REUSEPORT.
This patch introduces a counter of sockets with SO_INCOMING_CPU in
a reuseport group to check if we should iterate all sockets to find
a proper one. We increment the counter when
* calling listen() if the socket has SO_INCOMING_CPU and SO_REUSEPORT
* enabling SO_INCOMING_CPU if the socket is in a reuseport group
Also, we decrement it when
* detaching a socket out of the group to apply SO_INCOMING_CPU to
migrated TCP requests
* disabling SO_INCOMING_CPU if the socket is in a reuseport group
When the counter reaches 0, we can get back to the O(1) selection
algorithm.
The overall changes are negligible for the non-SO_INCOMING_CPU case,
and the only notable thing is that we have to update sk_incomnig_cpu
under reuseport_lock. Otherwise, the race prevents transitioning to
the O(n) algorithm and results in the wrong socket selection.
cpu1 (setsockopt) cpu2 (listen)
+-----------------+ +-------------+
lock_sock(sk1) lock_sock(sk2)
reuseport_update_incoming_cpu(sk1, val)
.
| /* set CPU as 0 */
|- WRITE_ONCE(sk1->incoming_cpu, val)
|
| spin_lock_bh(&reuseport_lock)
| reuseport_grow(sk2, reuse)
| .
| |- more_socks_size = reuse->max_socks * 2U;
| |- if (more_socks_size > U16_MAX &&
| | reuse->num_closed_socks)
| | .
| | |- RCU_INIT_POINTER(sk1->sk_reuseport_cb, NULL);
| | `- __reuseport_detach_closed_sock(sk1, reuse)
| | .
| | `- reuseport_put_incoming_cpu(sk1, reuse)
| | .
| | | /* Read shutdown()ed sk1's sk_incoming_cpu
| | | * without lock_sock().
| | | */
| | `- if (sk1->sk_incoming_cpu >= 0)
| | .
| | | /* decrement not-yet-incremented
| | | * count, which is never incremented.
| | | */
| | `- __reuseport_put_incoming_cpu(reuse);
| |
| `- spin_lock_bh(&reuseport_lock)
|
|- spin_lock_bh(&reuseport_lock)
|
|- reuse = rcu_dereference_protected(sk1->sk_reuseport_cb, ...)
|- if (!reuse)
| .
| | /* Cannot increment reuse->incoming_cpu. */
| `- goto out;
|
`- spin_unlock_bh(&reuseport_lock)
Fixes: e32ea7e747 ("soreuseport: fast reuseport UDP socket selection")
Fixes: c125e80b88 ("soreuseport: fast reuseport TCP socket selection")
Reported-by: Kazuho Oku <kazuhooku@gmail.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The parameter 'msg' has never been used by __sock_cmsg_send, so we can remove it
safely.
Reported-by: Zeal Robot <zealci@zte.com.cn>
Signed-off-by: xu xin <xu.xin16@zte.com.cn>
Reviewed-by: Zhang Yunkai <zhang.yunkai@zte.com.cn>
Acked-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit ffa84b5ffb ("net: add netns refcount tracker to struct sock")
added a tracker to sockets, but did not track kernel sockets.
We still have syzbot reports hinting about netns being destroyed
while some kernel TCP sockets had not been dismantled.
This patch tracks kernel sockets, and adds a ref_tracker_dir_print()
call to net_free() right before the netns is freed.
Normally, each layer is responsible for properly releasing its
kernel sockets before last call to net_free().
This debugging facility is enabled with CONFIG_NET_NS_REFCNT_TRACKER=y
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Tested-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 086d49058c ("ipv6: annotate some data-races around sk->sk_prot")
fixed some data-races around sk->sk_prot but it was not enough.
Some functions in inet6_(stream|dgram)_ops still access sk->sk_prot
without lock_sock() or rtnl_lock(), so they need READ_ONCE() to avoid
load tearing.
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Daniel Borkmann says:
====================
pull-request: bpf-next 2022-09-05
The following pull-request contains BPF updates for your *net-next* tree.
We've added 106 non-merge commits during the last 18 day(s) which contain
a total of 159 files changed, 5225 insertions(+), 1358 deletions(-).
There are two small merge conflicts, resolve them as follows:
1) tools/testing/selftests/bpf/DENYLIST.s390x
Commit 27e23836ce ("selftests/bpf: Add lru_bug to s390x deny list") in
bpf tree was needed to get BPF CI green on s390x, but it conflicted with
newly added tests on bpf-next. Resolve by adding both hunks, result:
[...]
lru_bug # prog 'printk': failed to auto-attach: -524
setget_sockopt # attach unexpected error: -524 (trampoline)
cb_refs # expected error message unexpected error: -524 (trampoline)
cgroup_hierarchical_stats # JIT does not support calling kernel function (kfunc)
htab_update # failed to attach: ERROR: strerror_r(-524)=22 (trampoline)
[...]
2) net/core/filter.c
Commit 1227c1771d ("net: Fix data-races around sysctl_[rw]mem_(max|default).")
from net tree conflicts with commit 29003875bd ("bpf: Change bpf_setsockopt(SOL_SOCKET)
to reuse sk_setsockopt()") from bpf-next tree. Take the code as it is from
bpf-next tree, result:
[...]
if (getopt) {
if (optname == SO_BINDTODEVICE)
return -EINVAL;
return sk_getsockopt(sk, SOL_SOCKET, optname,
KERNEL_SOCKPTR(optval),
KERNEL_SOCKPTR(optlen));
}
return sk_setsockopt(sk, SOL_SOCKET, optname,
KERNEL_SOCKPTR(optval), *optlen);
[...]
The main changes are:
1) Add any-context BPF specific memory allocator which is useful in particular for BPF
tracing with bonus of performance equal to full prealloc, from Alexei Starovoitov.
2) Big batch to remove duplicated code from bpf_{get,set}sockopt() helpers as an effort
to reuse the existing core socket code as much as possible, from Martin KaFai Lau.
3) Extend BPF flow dissector for BPF programs to just augment the in-kernel dissector
with custom logic. In other words, allow for partial replacement, from Shmulik Ladkani.
4) Add a new cgroup iterator to BPF with different traversal options, from Hao Luo.
5) Support for BPF to collect hierarchical cgroup statistics efficiently through BPF
integration with the rstat framework, from Yosry Ahmed.
6) Support bpf_{g,s}et_retval() under more BPF cgroup hooks, from Stanislav Fomichev.
7) BPF hash table and local storages fixes under fully preemptible kernel, from Hou Tao.
8) Add various improvements to BPF selftests and libbpf for compilation with gcc BPF
backend, from James Hilliard.
9) Fix verifier helper permissions and reference state management for synchronous
callbacks, from Kumar Kartikeya Dwivedi.
10) Add support for BPF selftest's xskxceiver to also be used against real devices that
support MAC loopback, from Maciej Fijalkowski.
11) Various fixes to the bpf-helpers(7) man page generation script, from Quentin Monnet.
12) Document BPF verifier's tnum_in(tnum_range(), ...) gotchas, from Shung-Hsi Yu.
13) Various minor misc improvements all over the place.
* https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next: (106 commits)
bpf: Optimize rcu_barrier usage between hash map and bpf_mem_alloc.
bpf: Remove usage of kmem_cache from bpf_mem_cache.
bpf: Remove prealloc-only restriction for sleepable bpf programs.
bpf: Prepare bpf_mem_alloc to be used by sleepable bpf programs.
bpf: Remove tracing program restriction on map types
bpf: Convert percpu hash map to per-cpu bpf_mem_alloc.
bpf: Add percpu allocation support to bpf_mem_alloc.
bpf: Batch call_rcu callbacks instead of SLAB_TYPESAFE_BY_RCU.
bpf: Adjust low/high watermarks in bpf_mem_cache
bpf: Optimize call_rcu in non-preallocated hash map.
bpf: Optimize element count in non-preallocated hash map.
bpf: Relax the requirement to use preallocated hash maps in tracing progs.
samples/bpf: Reduce syscall overhead in map_perf_test.
selftests/bpf: Improve test coverage of test_maps
bpf: Convert hash map to bpf_mem_alloc.
bpf: Introduce any context BPF specific memory allocator.
selftest/bpf: Add test for bpf_getsockopt()
bpf: Change bpf_getsockopt(SOL_IPV6) to reuse do_ipv6_getsockopt()
bpf: Change bpf_getsockopt(SOL_IP) to reuse do_ip_getsockopt()
bpf: Change bpf_getsockopt(SOL_TCP) to reuse do_tcp_getsockopt()
...
====================
Link: https://lore.kernel.org/r/20220905161136.9150-1-daniel@iogearbox.net
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This patch changes bpf_getsockopt(SOL_SOCKET) to reuse
sk_getsockopt(). It removes all duplicated code from
bpf_getsockopt(SOL_SOCKET).
Before this patch, there were some optnames available to
bpf_setsockopt(SOL_SOCKET) but missing in bpf_getsockopt(SOL_SOCKET).
It surprises users from time to time. For example, SO_REUSEADDR,
SO_KEEPALIVE, SO_RCVLOWAT, and SO_MAX_PACING_RATE. This patch
automatically closes this gap without duplicating more code.
The only exception is SO_BINDTODEVICE because it needs to acquire a
blocking lock. Thus, SO_BINDTODEVICE is not supported.
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Link: https://lore.kernel.org/r/20220902002912.2894040-1-kafai@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This patch changes sk_getsockopt() to take the sockptr_t argument
such that it can be used by bpf_getsockopt(SOL_SOCKET) in a
latter patch.
security_socket_getpeersec_stream() is not changed. It stays
with the __user ptr (optval.user and optlen.user) to avoid changes
to other security hooks. bpf_getsockopt(SOL_SOCKET) also does not
support SO_PEERSEC.
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Link: https://lore.kernel.org/r/20220902002802.2888419-1-kafai@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
A latter patch refactors bpf_getsockopt(SOL_SOCKET) with the
sock_getsockopt() to avoid code duplication and code
drift between the two duplicates.
The current sock_getsockopt() takes sock ptr as the argument.
The very first thing of this function is to get back the sk ptr
by 'sk = sock->sk'.
bpf_getsockopt() could be called when the sk does not have
the sock ptr created. Meaning sk->sk_socket is NULL. For example,
when a passive tcp connection has just been established but has yet
been accept()-ed. Thus, it cannot use the sock_getsockopt(sk->sk_socket)
or else it will pass a NULL ptr.
This patch moves all sock_getsockopt implementation to the newly
added sk_getsockopt(). The new sk_getsockopt() takes a sk ptr
and immediately gets the sock ptr by 'sock = sk->sk_socket'
The existing sock_getsockopt(sock) is changed to call
sk_getsockopt(sock->sk). All existing callers have both sock->sk
and sk->sk_socket pointer.
The latter patch will make bpf_getsockopt(SOL_SOCKET) call
sk_getsockopt(sk) directly. The bpf_getsockopt(SOL_SOCKET) does
not use the optnames that require sk->sk_socket, so it will
be safe.
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Link: https://lore.kernel.org/r/20220902002756.2887884-1-kafai@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
While reading sysctl_net_busy_read, it can be changed concurrently.
Thus, we need to add READ_ONCE() to its reader.
Fixes: 2d48d67fa8 ("net: poll/select low latency socket support")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
While reading sysctl_optmem_max, 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_[rw]mem_(max|default), they 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>
After the prep work in the previous patches,
this patch removes most of the dup code from bpf_setsockopt(SOL_SOCKET)
and reuses them from sk_setsockopt().
The sock ptr test is added to the SO_RCVLOWAT because
the sk->sk_socket could be NULL in some of the bpf hooks.
The existing optname white-list is refactored into a new
function sol_socket_setsockopt().
Reviewed-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/r/20220817061804.4178920-1-kafai@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When bpf program calling bpf_setsockopt(SOL_SOCKET),
it could be run in softirq and doesn't make sense to do the capable
check. There was a similar situation in bpf_setsockopt(TCP_CONGESTION).
In commit 8d650cdeda ("tcp: fix tcp_set_congestion_control() use from bpf hook"),
tcp_set_congestion_control(..., cap_net_admin) was added to skip
the cap check for bpf prog.
This patch adds sockopt_ns_capable() and sockopt_capable() for
the sk_setsockopt() to use. They will consider the
has_current_bpf_ctx() before doing the ns_capable() and capable() test.
They are in EXPORT_SYMBOL for the ipv6 module to use in a latter patch.
Suggested-by: Stanislav Fomichev <sdf@google.com>
Reviewed-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/r/20220817061723.4175820-1-kafai@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Most of the code in bpf_setsockopt(SOL_SOCKET) are duplicated from
the sk_setsockopt(). The number of supported optnames are
increasing ever and so as the duplicated code.
One issue in reusing sk_setsockopt() is that the bpf prog
has already acquired the sk lock. This patch adds a
has_current_bpf_ctx() to tell if the sk_setsockopt() is called from
a bpf prog. The bpf prog calling bpf_setsockopt() is either running
in_task() or in_serving_softirq(). Both cases have the current->bpf_ctx
initialized. Thus, the has_current_bpf_ctx() only needs to
test !!current->bpf_ctx.
This patch also adds sockopt_{lock,release}_sock() helpers
for sk_setsockopt() to use. These helpers will test
has_current_bpf_ctx() before acquiring/releasing the lock. They are
in EXPORT_SYMBOL for the ipv6 module to use in a latter patch.
Note on the change in sock_setbindtodevice(). sockopt_lock_sock()
is done in sock_setbindtodevice() instead of doing the lock_sock
in sock_bindtoindex(..., lock_sk = true).
Reviewed-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/r/20220817061717.4175589-1-kafai@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
A latter patch refactors bpf_setsockopt(SOL_SOCKET) with the
sock_setsockopt() to avoid code duplication and code
drift between the two duplicates.
The current sock_setsockopt() takes sock ptr as the argument.
The very first thing of this function is to get back the sk ptr
by 'sk = sock->sk'.
bpf_setsockopt() could be called when the sk does not have
the sock ptr created. Meaning sk->sk_socket is NULL. For example,
when a passive tcp connection has just been established but has yet
been accept()-ed. Thus, it cannot use the sock_setsockopt(sk->sk_socket)
or else it will pass a NULL ptr.
This patch moves all sock_setsockopt implementation to the newly
added sk_setsockopt(). The new sk_setsockopt() takes a sk ptr
and immediately gets the sock ptr by 'sock = sk->sk_socket'
The existing sock_setsockopt(sock) is changed to call
sk_setsockopt(sock->sk). All existing callers have both sock->sk
and sk->sk_socket pointer.
The latter patch will make bpf_setsockopt(SOL_SOCKET) call
sk_setsockopt(sk) directly. The bpf_setsockopt(SOL_SOCKET) does
not use the optnames that require sk->sk_socket, so it will
be safe.
Reviewed-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/r/20220817061711.4175048-1-kafai@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
We continuously hold the socket lock during large reads and writes.
This may inflate RTT and negatively impact TCP performance.
Flush the backlog periodically. I tried to pick a flush period (128kB)
which gives significant benefit but the max Bps rate is not yet visibly
impacted.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
sk_memory_allocated_add() has three callers, and returns
to them @memory_allocated.
sk_forced_mem_schedule() is one of them, and ignores
the returned value.
Change sk_memory_allocated_add() to return void.
Change sock_reserve_memory() and __sk_mem_raise_allocated()
to call sk_memory_allocated().
This removes one cache line miss [1] for RPC workloads,
as first skbs in TCP write queue and receive queue go through
sk_forced_mem_schedule().
[1] Cache line holding tcp_memory_allocated.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
These two helpers are only used from core networking.
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>
Check against skb dst in socket backlog has never triggered
in past years.
Keep the check omly for CONFIG_DEBUG_NET=y builds.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
sock_bindtoindex_locked() needs to use WRITE_ONCE(sk->sk_bound_dev_if, val),
because other cpus/threads might locklessly read this field.
sock_getbindtodevice(), sock_getsockopt() need READ_ONCE()
because they run without socket lock held.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The code for gso_max_size was added originally to allow for debugging and
workaround of buggy devices that couldn't support TSO with blocks 64K in
size. The original reason for limiting it to 64K was because that was the
existing limits of IPv4 and non-jumbogram IPv6 length fields.
With the addition of Big TCP we can remove this limit and allow the value
to potentially go up to UINT_MAX and instead be limited by the tso_max_size
value.
So in order to support this we need to go through and clean up the
remaining users of the gso_max_size value so that the values will cap at
64K for non-TCPv6 flows. In addition we can clean up the GSO_MAX_SIZE value
so that 64K becomes GSO_LEGACY_MAX_SIZE and UINT_MAX will now be the upper
limit for GSO_MAX_SIZE.
v6: (edumazet) fixed a compile error if CONFIG_IPV6=n,
in a new sk_trim_gso_size() helper.
netif_set_tso_max_size() caps the requested TSO size
with GSO_MAX_SIZE.
Signed-off-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The commit referenced in the "Fixes" tag added the SO_RCVMARK socket
option for receiving the skb mark in the ancillary data.
Since this is a new capability, and exposes admin configured details
regarding the underlying network setup to sockets, let's align the
needed capabilities with those of SO_MARK.
Fixes: 6fd1d51cfa ("net: SO_RCVMARK socket option for SO_MARK with recvmsg()")
Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
Link: https://lore.kernel.org/r/20220504095459.2663513-1-eyal.birger@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Now we have a separate path for sock_def_write_space() and can go one
step further. When it's called from sock_wfree() we know that there is a
preceding atomic for putting down ->sk_wmem_alloc. We can use it to
replace to replace smb_mb() with a less expensive
smp_mb__after_atomic(). It also removes an extra RCU read lock/unlock as
a small bonus.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For non SOCK_USE_WRITE_QUEUE sockets, sock_wfree() (atomically) puts
->sk_wmem_alloc twice. It's needed to keep the socket alive while
calling ->sk_write_space() after the first put.
However, some sockets, such as UDP, are freed by RCU
(i.e. SOCK_RCU_FREE) and use already RCU-safe sock_def_write_space().
Carve a fast path for such sockets, put down all refs in one go before
calling sock_def_write_space() but guard the socket from being freed
by an RCU read section.
note: because TCP sockets are marked with SOCK_USE_WRITE_QUEUE it
doesn't add extra checks in its path.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Except for minor rounding differences the first ->sk_wmem_alloc test in
sock_def_write_space() is a hand coded version of sock_writeable().
Replace it with the helper, and also kill the following if duplicating
the check.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sock_alloc_send_skb() is simple and just proxying to another function,
so we can inline it and cut associated overhead.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Adding a new socket option, SO_RCVMARK, to indicate that SO_MARK
should be included in the ancillary data returned by recvmsg().
Renamed the sock_recv_ts_and_drops() function to sock_recv_cmsgs().
Signed-off-by: Erin MacNeil <lnx.erin@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Acked-by: Marc Kleine-Budde <mkl@pengutronix.de>
Link: https://lore.kernel.org/r/20220427200259.2564-1-lnx.erin@gmail.com
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>
In order to report the reasons of skb drops in 'sock_queue_rcv_skb()',
introduce the function 'sock_queue_rcv_skb_reason()'.
As the return value of 'sock_queue_rcv_skb()' is used as the error code,
we can't make it as drop reason and have to pass extra output argument.
'sock_queue_rcv_skb()' is used in many places, so we can't change it
directly.
Introduce the new function 'sock_queue_rcv_skb_reason()' and make
'sock_queue_rcv_skb()' an inline call to it.
Reviewed-by: Hao Peng <flyingpeng@tencent.com>
Reviewed-by: Jiang Biao <benbjiang@tencent.com>
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
There's a number of functions and static variables used
under net/core/ but not from the outside. We currently
dump most of them into netdevice.h. That bad for many
reasons:
- netdevice.h is very cluttered, hard to figure out
what the APIs are;
- netdevice.h is very long;
- we have to touch netdevice.h more which causes expensive
incremental builds.
Create a header under net/core/ and move some declarations.
The new header is also a bit of a catch-all but that's
fine, if we create more specific headers people will
likely over-think where their declaration fit best.
And end up putting them in netdevice.h, again.
More work should be done on splitting netdevice.h into more
targeted headers, but that'd be more time consuming so small
steps.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
ENOTSUPP is documented as "should never be seen by user programs",
and thus not exposed in <errno.h>, and thus applications cannot safely
check against it (they get "Unknown error 524" as strerror). We should
rather return the well-known -EOPNOTSUPP.
This is similar to 2230a7ef51 ("drop_monitor: Use correct error
code") and 4a5cdc604b ("net/tls: Fix return values to avoid
ENOTSUPP"), which did not seem to cause problems.
Signed-off-by: Samuel Thibault <samuel.thibault@labri.fr>
Acked-by: Willem de Bruijn <willemb@google.com>
Link: https://lore.kernel.org/r/20220307223126.djzvg44v2o2jkjsx@begin
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
UDP sendmsg() can be lockless, this is causing all kinds
of data races.
This patch converts sk->sk_tskey to remove one of these races.
BUG: KCSAN: data-race in __ip_append_data / __ip_append_data
read to 0xffff8881035d4b6c of 4 bytes by task 8877 on cpu 1:
__ip_append_data+0x1c1/0x1de0 net/ipv4/ip_output.c:994
ip_make_skb+0x13f/0x2d0 net/ipv4/ip_output.c:1636
udp_sendmsg+0x12bd/0x14c0 net/ipv4/udp.c:1249
inet_sendmsg+0x5f/0x80 net/ipv4/af_inet.c:819
sock_sendmsg_nosec net/socket.c:705 [inline]
sock_sendmsg net/socket.c:725 [inline]
____sys_sendmsg+0x39a/0x510 net/socket.c:2413
___sys_sendmsg net/socket.c:2467 [inline]
__sys_sendmmsg+0x267/0x4c0 net/socket.c:2553
__do_sys_sendmmsg net/socket.c:2582 [inline]
__se_sys_sendmmsg net/socket.c:2579 [inline]
__x64_sys_sendmmsg+0x53/0x60 net/socket.c:2579
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x44/0xd0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x44/0xae
write to 0xffff8881035d4b6c of 4 bytes by task 8880 on cpu 0:
__ip_append_data+0x1d8/0x1de0 net/ipv4/ip_output.c:994
ip_make_skb+0x13f/0x2d0 net/ipv4/ip_output.c:1636
udp_sendmsg+0x12bd/0x14c0 net/ipv4/udp.c:1249
inet_sendmsg+0x5f/0x80 net/ipv4/af_inet.c:819
sock_sendmsg_nosec net/socket.c:705 [inline]
sock_sendmsg net/socket.c:725 [inline]
____sys_sendmsg+0x39a/0x510 net/socket.c:2413
___sys_sendmsg net/socket.c:2467 [inline]
__sys_sendmmsg+0x267/0x4c0 net/socket.c:2553
__do_sys_sendmmsg net/socket.c:2582 [inline]
__se_sys_sendmmsg net/socket.c:2579 [inline]
__x64_sys_sendmmsg+0x53/0x60 net/socket.c:2579
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x44/0xd0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x44/0xae
value changed: 0x0000054d -> 0x0000054e
Reported by Kernel Concurrency Sanitizer on:
CPU: 0 PID: 8880 Comm: syz-executor.5 Not tainted 5.17.0-rc2-syzkaller-00167-gdcb85f85fa6f-dirty #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Fixes: 09c2d251b7 ("net-timestamp: add key to disambiguate concurrent datagrams")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
prot->memory_allocated should only be set if prot->sysctl_mem
is also set.
This is a followup of commit 2520611151 ("crypto: af_alg - get
rid of alg_memory_allocated").
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20220216171801.3604366-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
There's not reason SO_MARK would be allowed via setsockopt()
and not via cmsg, let's keep the two consistent. See
commit 079925cce1 ("net: allow SO_MARK with CAP_NET_RAW")
for justification why NET_RAW -> SO_MARK is safe.
Reviewed-by: Maciej Żenczykowski <maze@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Link: https://lore.kernel.org/r/20220131233357.52964-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Disabling rehash behavior did not affect SYN ACK retransmits because hash
was forcefully changed bypassing the sk_rethink_hash function. This patch
adds a condition which checks for rehash mode before resetting hash.
Signed-off-by: Akhmat Karakotov <hmukos@yandex-team.ru>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add the SO_TXREHASH socket option to control hash rethink behavior per socket.
When default mode is set, sockets disable rehash at initialization and use
sysctl option when entering listen state. setsockopt() overrides default
behavior.
Signed-off-by: Akhmat Karakotov <hmukos@yandex-team.ru>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
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>
The cited Fixes patch moved to a deferred skb approach where the skbs
are not freed immediately under the socket lock. Add a WARN_ON_ONCE()
to verify the deferred list is empty on socket destroy, and empty it to
prevent potential memory leaks.
Fixes: f35f821935 ("tcp: defer skb freeing after socket lock is released")
Signed-off-by: Gal Pressman <gal@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Don't forget to release the device in sock_timestamping_bind_phc() after
it was used to get the vclock indices.
Fixes: d463126e23 ("net: sock: extend SO_TIMESTAMPING for PHC binding")
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com>
Cc: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch moves sock_release_ownership() down in include/net/sock.h and
replaces some sk_lock.owned tests with sock_owned_by_user_nocheck().
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.co.jp>
Link: https://lore.kernel.org/r/20211208062158.54132-1-kuniyu@amazon.co.jp
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
A CAP_NET_RAW capable process can already spoof (on transmit) anything
it desires via raw packet sockets... There is no good reason to not
allow it to also be able to play routing tricks on packets from its
own normal sockets.
There is a desire to be able to use SO_MARK for routing table selection
(via ip rule fwmark) from within a user process without having to run
it as root. Granting it CAP_NET_RAW is much less dangerous than
CAP_NET_ADMIN (CAP_NET_RAW doesn't permit persistent state change,
while CAP_NET_ADMIN does - by for example allowing the reconfiguration
of the routing tables and/or bringing up/down devices).
Let's keep CAP_NET_ADMIN for persistent state changes,
while using CAP_NET_RAW for non-configuration related stuff.
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Link: https://lore.kernel.org/r/20211123203715.193413-1-zenczykowski@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
CAP_NET_ADMIN is and should continue to be about configuring the
system as a whole, not about configuring per-socket or per-packet
parameters.
Sending and receiving raw packets is what CAP_NET_RAW is all about.
It can already send packets with any VLAN tag, and any IPv4 TOS
mark, and any IPv6 TCLASS mark, simply by virtue of building
such a raw packet. Not to mention using any protocol and source/
/destination ip address/port tuple.
These are the fields that networking gear uses to prioritize packets.
Hence, a CAP_NET_RAW process is already capable of affecting traffic
prioritization after it hits the wire. This change makes it capable
of affecting traffic prioritization even in the host at the nic and
before that in the queueing disciplines (provided skb->priority is
actually being used for prioritization, and not the TOS/TCLASS field)
Hence it makes sense to allow a CAP_NET_RAW process to set the
priority of sockets and thus packets it sends.
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Link: https://lore.kernel.org/r/20211123203702.193221-1-zenczykowski@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
dev->gso_max_segs is written under RTNL protection, or when the device is
not yet visible, but is read locklessly.
Add netif_set_gso_max_segs() helper.
Add the READ_ONCE()/WRITE_ONCE() pairs, and use netif_set_gso_max_segs()
where we can to better document what is going on.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
dev->gso_max_size is written under RTNL protection, or when the device is
not yet visible, but is read locklessly.
Add the READ_ONCE()/WRITE_ONCE() pairs, and use netif_set_gso_max_size()
where we can to better document what is going on.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
