linux.git/net/core/skmsg.c, branch v6.6

bpf, sockmap: Fix skb refcnt race after locking changes

2023-09-04T07:53:35+00:00

There is a race where skb's from the sk_psock_backlog can be referenced
after userspace side has already skb_consumed() the sk_buff and its refcnt
dropped to zer0 causing use after free.

The flow is the following:

  while ((skb = skb_peek(&psock->ingress_skb))
    sk_psock_handle_Skb(psock, skb, ..., ingress)
    if (!ingress) ...
    sk_psock_skb_ingress
       sk_psock_skb_ingress_enqueue(skb)
          msg->skb = skb
          sk_psock_queue_msg(psock, msg)
    skb_dequeue(&psock->ingress_skb)

The sk_psock_queue_msg() puts the msg on the ingress_msg queue. This is
what the application reads when recvmsg() is called. An application can
read this anytime after the msg is placed on the queue. The recvmsg hook
will also read msg->skb and then after user space reads the msg will call
consume_skb(skb) on it effectively free'ing it.

But, the race is in above where backlog queue still has a reference to
the skb and calls skb_dequeue(). If the skb_dequeue happens after the
user reads and free's the skb we have a use after free.

The !ingress case does not suffer from this problem because it uses
sendmsg_*(sk, msg) which does not pass the sk_buff further down the
stack.

The following splat was observed with 'test_progs -t sockmap_listen':

  [ 1022.710250][ T2556] general protection fault, ...
  [...]
  [ 1022.712830][ T2556] Workqueue: events sk_psock_backlog
  [ 1022.713262][ T2556] RIP: 0010:skb_dequeue+0x4c/0x80
  [ 1022.713653][ T2556] Code: ...
  [...]
  [ 1022.720699][ T2556] Call Trace:
  [ 1022.720984][ T2556]  
  [ 1022.721254][ T2556]  ? die_addr+0x32/0x80^M
  [ 1022.721589][ T2556]  ? exc_general_protection+0x25a/0x4b0
  [ 1022.722026][ T2556]  ? asm_exc_general_protection+0x22/0x30
  [ 1022.722489][ T2556]  ? skb_dequeue+0x4c/0x80
  [ 1022.722854][ T2556]  sk_psock_backlog+0x27a/0x300
  [ 1022.723243][ T2556]  process_one_work+0x2a7/0x5b0
  [ 1022.723633][ T2556]  worker_thread+0x4f/0x3a0
  [ 1022.723998][ T2556]  ? __pfx_worker_thread+0x10/0x10
  [ 1022.724386][ T2556]  kthread+0xfd/0x130
  [ 1022.724709][ T2556]  ? __pfx_kthread+0x10/0x10
  [ 1022.725066][ T2556]  ret_from_fork+0x2d/0x50
  [ 1022.725409][ T2556]  ? __pfx_kthread+0x10/0x10
  [ 1022.725799][ T2556]  ret_from_fork_asm+0x1b/0x30
  [ 1022.726201][ T2556]  

To fix we add an skb_get() before passing the skb to be enqueued in the
engress queue. This bumps the skb->users refcnt so that consume_skb()
and kfree_skb will not immediately free the sk_buff. With this we can
be sure the skb is still around when we do the dequeue. Then we just
need to decrement the refcnt or free the skb in the backlog case which
we do by calling kfree_skb() on the ingress case as well as the sendmsg
case.

Before locking change from fixes tag we had the sock locked so we
couldn't race with user and there was no issue here.

Fixes: 799aa7f98d53e ("skmsg: Avoid lock_sock() in sk_psock_backlog()")
Reported-by: Jiri Olsa  
Signed-off-by: John Fastabend 
Signed-off-by: Daniel Borkmann 
Tested-by: Xu Kuohai 
Tested-by: Jiri Olsa 
Link: https://lore.kernel.org/bpf/20230901202137.214666-1-john.fastabend@gmail.com

Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net

2023-08-10T21:10:53+00:00

Cross-merge networking fixes after downstream PR.

No conflicts.

Adjacent changes:

drivers/net/ethernet/intel/igc/igc_main.c
  06b412589eef ("igc: Add lock to safeguard global Qbv variables")
  d3750076d464 ("igc: Add TransmissionOverrun counter")

drivers/net/ethernet/microsoft/mana/mana_en.c
  a7dfeda6fdec ("net: mana: Fix MANA VF unload when hardware is unresponsive")
  a9ca9f9ceff3 ("page_pool: split types and declarations from page_pool.h")
  92272ec4107e ("eth: add missing xdp.h includes in drivers")

net/mptcp/protocol.h
  511b90e39250 ("mptcp: fix disconnect vs accept race")
  b8dc6d6ce931 ("mptcp: fix rcv buffer auto-tuning")

tools/testing/selftests/net/mptcp/mptcp_join.sh
  c8c101ae390a ("selftests: mptcp: join: fix 'implicit EP' test")
  03668c65d153 ("selftests: mptcp: join: rework detailed report")

Signed-off-by: Jakub Kicinski

bpf, sockmap: Fix bug that strp_done cannot be called

2023-08-10T03:29:02+00:00

strp_done is only called when psock->progs.stream_parser is not NULL,
but stream_parser was set to NULL by sk_psock_stop_strp(), called
by sk_psock_drop() earlier. So, strp_done can never be called.

Introduce SK_PSOCK_RX_ENABLED to mark whether there is strp on psock.
Change the condition for calling strp_done from judging whether
stream_parser is set to judging whether this flag is set. This flag is
only set once when strp_init() succeeds, and will never be cleared later.

Fixes: c0d95d3380ee ("bpf, sockmap: Re-evaluate proto ops when psock is removed from sockmap")
Signed-off-by: Xu Kuohai 
Reviewed-by: John Fastabend 
Link: https://lore.kernel.org/r/20230804073740.194770-3-xukuohai@huaweicloud.com
Signed-off-by: Martin KaFai Lau

net: annotate data-races around sock->ops

2023-08-09T22:32:43+00:00

IPV6_ADDRFORM socket option is evil, because it can change sock->ops
while other threads might read it. Same issue for sk->sk_family
being set to AF_INET.

Adding READ_ONCE() over sock->ops reads is needed for sockets
that might be impacted by IPV6_ADDRFORM.

Note that mptcp_is_tcpsk() can also overwrite sock->ops.

Adding annotations for all sk->sk_family reads will require
more patches :/

BUG: KCSAN: data-race in ____sys_sendmsg / do_ipv6_setsockopt

write to 0xffff888109f24ca0 of 8 bytes by task 4470 on cpu 0:
do_ipv6_setsockopt+0x2c5e/0x2ce0 net/ipv6/ipv6_sockglue.c:491
ipv6_setsockopt+0x57/0x130 net/ipv6/ipv6_sockglue.c:1012
udpv6_setsockopt+0x95/0xa0 net/ipv6/udp.c:1690
sock_common_setsockopt+0x61/0x70 net/core/sock.c:3663
__sys_setsockopt+0x1c3/0x230 net/socket.c:2273
__do_sys_setsockopt net/socket.c:2284 [inline]
__se_sys_setsockopt net/socket.c:2281 [inline]
__x64_sys_setsockopt+0x66/0x80 net/socket.c:2281
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x41/0xc0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x63/0xcd

read to 0xffff888109f24ca0 of 8 bytes by task 4469 on cpu 1:
sock_sendmsg_nosec net/socket.c:724 [inline]
sock_sendmsg net/socket.c:747 [inline]
____sys_sendmsg+0x349/0x4c0 net/socket.c:2503
___sys_sendmsg net/socket.c:2557 [inline]
__sys_sendmmsg+0x263/0x500 net/socket.c:2643
__do_sys_sendmmsg net/socket.c:2672 [inline]
__se_sys_sendmmsg net/socket.c:2669 [inline]
__x64_sys_sendmmsg+0x57/0x60 net/socket.c:2669
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x41/0xc0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x63/0xcd

value changed: 0xffffffff850e32b8 -> 0xffffffff850da890

Reported by Kernel Concurrency Sanitizer on:
CPU: 1 PID: 4469 Comm: syz-executor.1 Not tainted 6.4.0-rc5-syzkaller-00313-g4c605260bc60 #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 05/25/2023

Reported-by: syzbot 
Signed-off-by: Eric Dumazet 
Reviewed-by: Kuniyuki Iwashima 
Link: https://lore.kernel.org/r/20230808135809.2300241-1-edumazet@google.com
Signed-off-by: Jakub Kicinski

bpf, sockmap: Avoid potential NULL dereference in sk_psock_verdict_data_ready()

2023-06-01T12:44:53+00:00

syzbot found sk_psock(sk) could return NULL when called
from sk_psock_verdict_data_ready().

Just make sure to handle this case.

[1]
general protection fault, probably for non-canonical address 0xdffffc000000005c: 0000 [#1] PREEMPT SMP KASAN
KASAN: null-ptr-deref in range [0x00000000000002e0-0x00000000000002e7]
CPU: 0 PID: 15 Comm: ksoftirqd/0 Not tainted 6.4.0-rc3-syzkaller-00588-g4781e965e655 #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 05/16/2023
RIP: 0010:sk_psock_verdict_data_ready+0x19f/0x3c0 net/core/skmsg.c:1213
Code: 4c 89 e6 e8 63 70 5e f9 4d 85 e4 75 75 e8 19 74 5e f9 48 8d bb e0 02 00 00 48 b8 00 00 00 00 00 fc ff df 48 89 fa 48 c1 ea 03 <80> 3c 02 00 0f 85 07 02 00 00 48 89 ef ff 93 e0 02 00 00 e8 29 fd
RSP: 0018:ffffc90000147688 EFLAGS: 00010206
RAX: dffffc0000000000 RBX: 0000000000000000 RCX: 0000000000000100
RDX: 000000000000005c RSI: ffffffff8825ceb7 RDI: 00000000000002e0
RBP: ffff888076518c40 R08: 0000000000000007 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000001 R12: 0000000000000000
R13: 0000000000000000 R14: 0000000000008000 R15: ffff888076518c40
FS: 0000000000000000(0000) GS:ffff8880b9800000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007f901375bab0 CR3: 000000004bf26000 CR4: 00000000003506f0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:

tcp_data_ready+0x10a/0x520 net/ipv4/tcp_input.c:5006
tcp_data_queue+0x25d3/0x4c50 net/ipv4/tcp_input.c:5080
tcp_rcv_established+0x829/0x1f90 net/ipv4/tcp_input.c:6019
tcp_v4_do_rcv+0x65a/0x9c0 net/ipv4/tcp_ipv4.c:1726
tcp_v4_rcv+0x2cbf/0x3340 net/ipv4/tcp_ipv4.c:2148
ip_protocol_deliver_rcu+0x9f/0x480 net/ipv4/ip_input.c:205
ip_local_deliver_finish+0x2ec/0x520 net/ipv4/ip_input.c:233
NF_HOOK include/linux/netfilter.h:303 [inline]
NF_HOOK include/linux/netfilter.h:297 [inline]
ip_local_deliver+0x1ae/0x200 net/ipv4/ip_input.c:254
dst_input include/net/dst.h:468 [inline]
ip_rcv_finish+0x1cf/0x2f0 net/ipv4/ip_input.c:449
NF_HOOK include/linux/netfilter.h:303 [inline]
NF_HOOK include/linux/netfilter.h:297 [inline]
ip_rcv+0xae/0xd0 net/ipv4/ip_input.c:569
__netif_receive_skb_one_core+0x114/0x180 net/core/dev.c:5491
__netif_receive_skb+0x1f/0x1c0 net/core/dev.c:5605
process_backlog+0x101/0x670 net/core/dev.c:5933
__napi_poll+0xb7/0x6f0 net/core/dev.c:6499
napi_poll net/core/dev.c:6566 [inline]
net_rx_action+0x8a9/0xcb0 net/core/dev.c:6699
__do_softirq+0x1d4/0x905 kernel/softirq.c:571
run_ksoftirqd kernel/softirq.c:939 [inline]
run_ksoftirqd+0x31/0x60 kernel/softirq.c:931
smpboot_thread_fn+0x659/0x9e0 kernel/smpboot.c:164
kthread+0x344/0x440 kernel/kthread.c:379
ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:308


Fixes: 6df7f764cd3c ("bpf, sockmap: Wake up polling after data copy")
Reported-by: syzbot 
Signed-off-by: Eric Dumazet 
Signed-off-by: Daniel Borkmann 
Reviewed-by: John Fastabend 
Link: https://lore.kernel.org/bpf/20230530195149.68145-1-edumazet@google.com

bpf, sockmap: Incorrectly handling copied_seq

2023-05-23T14:10:42+00:00

The read_skb() logic is incrementing the tcp->copied_seq which is used for
among other things calculating how many outstanding bytes can be read by
the application. This results in application errors, if the application
does an ioctl(FIONREAD) we return zero because this is calculated from
the copied_seq value.

To fix this we move tcp->copied_seq accounting into the recv handler so
that we update these when the recvmsg() hook is called and data is in
fact copied into user buffers. This gives an accurate FIONREAD value
as expected and improves ACK handling. Before we were calling the
tcp_rcv_space_adjust() which would update 'number of bytes copied to
user in last RTT' which is wrong for programs returning SK_PASS. The
bytes are only copied to the user when recvmsg is handled.

Doing the fix for recvmsg is straightforward, but fixing redirect and
SK_DROP pkts is a bit tricker. Build a tcp_psock_eat() helper and then
call this from skmsg handlers. This fixes another issue where a broken
socket with a BPF program doing a resubmit could hang the receiver. This
happened because although read_skb() consumed the skb through sock_drop()
it did not update the copied_seq. Now if a single reccv socket is
redirecting to many sockets (for example for lb) the receiver sk will be
hung even though we might expect it to continue. The hang comes from
not updating the copied_seq numbers and memory pressure resulting from
that.

We have a slight layer problem of calling tcp_eat_skb even if its not
a TCP socket. To fix we could refactor and create per type receiver
handlers. I decided this is more work than we want in the fix and we
already have some small tweaks depending on caller that use the
helper skb_bpf_strparser(). So we extend that a bit and always set
the strparser bit when it is in use and then we can gate the
seq_copied updates on this.

Fixes: 04919bed948dc ("tcp: Introduce tcp_read_skb()")
Signed-off-by: John Fastabend 
Signed-off-by: Daniel Borkmann 
Reviewed-by: Jakub Sitnicki 
Link: https://lore.kernel.org/bpf/20230523025618.113937-9-john.fastabend@gmail.com

bpf, sockmap: Wake up polling after data copy

2023-05-23T14:10:35+00:00

When TCP stack has data ready to read sk_data_ready() is called. Sockmap
overwrites this with its own handler to call into BPF verdict program.
But, the original TCP socket had sock_def_readable that would additionally
wake up any user space waiters with sk_wake_async().

Sockmap saved the callback when the socket was created so call the saved
data ready callback and then we can wake up any epoll() logic waiting
on the read.

Note we call on 'copied >= 0' to account for returning 0 when a FIN is
received because we need to wake up user for this as well so they
can do the recvmsg() -> 0 and detect the shutdown.

Fixes: 04919bed948dc ("tcp: Introduce tcp_read_skb()")
Signed-off-by: John Fastabend 
Signed-off-by: Daniel Borkmann 
Reviewed-by: Jakub Sitnicki 
Link: https://lore.kernel.org/bpf/20230523025618.113937-8-john.fastabend@gmail.com

bpf, sockmap: Improved check for empty queue

2023-05-23T14:10:11+00:00

We noticed some rare sk_buffs were stepping past the queue when system was
under memory pressure. The general theory is to skip enqueueing
sk_buffs when its not necessary which is the normal case with a system
that is properly provisioned for the task, no memory pressure and enough
cpu assigned.

But, if we can't allocate memory due to an ENOMEM error when enqueueing
the sk_buff into the sockmap receive queue we push it onto a delayed
workqueue to retry later. When a new sk_buff is received we then check
if that queue is empty. However, there is a problem with simply checking
the queue length. When a sk_buff is being processed from the ingress queue
but not yet on the sockmap msg receive queue its possible to also recv
a sk_buff through normal path. It will check the ingress queue which is
zero and then skip ahead of the pkt being processed.

Previously we used sock lock from both contexts which made the problem
harder to hit, but not impossible.

To fix instead of popping the skb from the queue entirely we peek the
skb from the queue and do the copy there. This ensures checks to the
queue length are non-zero while skb is being processed. Then finally
when the entire skb has been copied to user space queue or another
socket we pop it off the queue. This way the queue length check allows
bypassing the queue only after the list has been completely processed.

To reproduce issue we run NGINX compliance test with sockmap running and
observe some flakes in our testing that we attributed to this issue.

Fixes: 04919bed948dc ("tcp: Introduce tcp_read_skb()")
Suggested-by: Jakub Sitnicki 
Signed-off-by: John Fastabend 
Signed-off-by: Daniel Borkmann 
Tested-by: William Findlay 
Reviewed-by: Jakub Sitnicki 
Link: https://lore.kernel.org/bpf/20230523025618.113937-5-john.fastabend@gmail.com

bpf, sockmap: Reschedule is now done through backlog

2023-05-23T14:10:04+00:00

Now that the backlog manages the reschedule() logic correctly we can drop
the partial fix to reschedule from recvmsg hook.

Rescheduling on recvmsg hook was added to address a corner case where we
still had data in the backlog state but had nothing to kick it and
reschedule the backlog worker to run and finish copying data out of the
state. This had a couple limitations, first it required user space to
kick it introducing an unnecessary EBUSY and retry. Second it only
handled the ingress case and egress redirects would still be hung.

With the correct fix, pushing the reschedule logic down to where the
enomem error occurs we can drop this fix.

Fixes: bec217197b412 ("skmsg: Schedule psock work if the cached skb exists on the psock")
Signed-off-by: John Fastabend 
Signed-off-by: Daniel Borkmann 
Reviewed-by: Jakub Sitnicki 
Link: https://lore.kernel.org/bpf/20230523025618.113937-4-john.fastabend@gmail.com

bpf, sockmap: Convert schedule_work into delayed_work

2023-05-23T14:09:56+00:00

Sk_buffs are fed into sockmap verdict programs either from a strparser
(when the user might want to decide how framing of skb is done by attaching
another parser program) or directly through tcp_read_sock. The
tcp_read_sock is the preferred method for performance when the BPF logic is
a stream parser.

The flow for Cilium's common use case with a stream parser is,

 tcp_read_sock()
  sk_psock_verdict_recv
    ret = bpf_prog_run_pin_on_cpu()
    sk_psock_verdict_apply(sock, skb, ret)
     // if system is under memory pressure or app is slow we may
     // need to queue skb. Do this queuing through ingress_skb and
     // then kick timer to wake up handler
     skb_queue_tail(ingress_skb, skb)
     schedule_work(work);

The work queue is wired up to sk_psock_backlog(). This will then walk the
ingress_skb skb list that holds our sk_buffs that could not be handled,
but should be OK to run at some later point. However, its possible that
the workqueue doing this work still hits an error when sending the skb.
When this happens the skbuff is requeued on a temporary 'state' struct
kept with the workqueue. This is necessary because its possible to
partially send an skbuff before hitting an error and we need to know how
and where to restart when the workqueue runs next.

Now for the trouble, we don't rekick the workqueue. This can cause a
stall where the skbuff we just cached on the state variable might never
be sent. This happens when its the last packet in a flow and no further
packets come along that would cause the system to kick the workqueue from
that side.

To fix we could do simple schedule_work(), but while under memory pressure
it makes sense to back off some instead of continue to retry repeatedly. So
instead to fix convert schedule_work to schedule_delayed_work and add
backoff logic to reschedule from backlog queue on errors. Its not obvious
though what a good backoff is so use '1'.

To test we observed some flakes whil running NGINX compliance test with
sockmap we attributed these failed test to this bug and subsequent issue.

>From on list discussion. This commit

 bec217197b41("skmsg: Schedule psock work if the cached skb exists on the psock")

was intended to address similar race, but had a couple cases it missed.
Most obvious it only accounted for receiving traffic on the local socket
so if redirecting into another socket we could still get an sk_buff stuck
here. Next it missed the case where copied=0 in the recv() handler and
then we wouldn't kick the scheduler. Also its sub-optimal to require
userspace to kick the internal mechanisms of sockmap to wake it up and
copy data to user. It results in an extra syscall and requires the app
to actual handle the EAGAIN correctly.

Fixes: 04919bed948dc ("tcp: Introduce tcp_read_skb()")
Signed-off-by: John Fastabend 
Signed-off-by: Daniel Borkmann 
Tested-by: William Findlay 
Reviewed-by: Jakub Sitnicki 
Link: https://lore.kernel.org/bpf/20230523025618.113937-3-john.fastabend@gmail.com