]> git.baikalelectronics.ru Git - kernel.git/commitdiff
tcp: prohibit TCP_REPAIR_OPTIONS if data was already sent
authorLu Wei <luwei32@huawei.com>
Fri, 4 Nov 2022 02:27:23 +0000 (10:27 +0800)
committerDavid S. Miller <davem@davemloft.net>
Mon, 7 Nov 2022 11:29:38 +0000 (11:29 +0000)
If setsockopt with option name of TCP_REPAIR_OPTIONS and opt_code
of TCPOPT_SACK_PERM is called to enable sack after data is sent
and dupacks are received , it will trigger a warning in function
tcp_verify_left_out() as follows:

============================================
WARNING: CPU: 8 PID: 0 at net/ipv4/tcp_input.c:2132
tcp_timeout_mark_lost+0x154/0x160
tcp_enter_loss+0x2b/0x290
tcp_retransmit_timer+0x50b/0x640
tcp_write_timer_handler+0x1c8/0x340
tcp_write_timer+0xe5/0x140
call_timer_fn+0x3a/0x1b0
__run_timers.part.0+0x1bf/0x2d0
run_timer_softirq+0x43/0xb0
__do_softirq+0xfd/0x373
__irq_exit_rcu+0xf6/0x140

The warning is caused in the following steps:
1. a socket named socketA is created
2. socketA enters repair mode without build a connection
3. socketA calls connect() and its state is changed to TCP_ESTABLISHED
   directly
4. socketA leaves repair mode
5. socketA calls sendmsg() to send data, packets_out and sack_outs(dup
   ack receives) increase
6. socketA enters repair mode again
7. socketA calls setsockopt with TCPOPT_SACK_PERM to enable sack
8. retransmit timer expires, it calls tcp_timeout_mark_lost(), lost_out
   increases
9. sack_outs + lost_out > packets_out triggers since lost_out and
   sack_outs increase repeatly

In function tcp_timeout_mark_lost(), tp->sacked_out will be cleared if
Step7 not happen and the warning will not be triggered. As suggested by
Denis and Eric, TCP_REPAIR_OPTIONS should be prohibited if data was
already sent.

socket-tcp tests in CRIU has been tested as follows:
$ sudo ./test/zdtm.py run -t zdtm/static/socket-tcp*  --keep-going \
       --ignore-taint

socket-tcp* represent all socket-tcp tests in test/zdtm/static/.

Fixes: 113136c2cce8 ("tcp: Repair connection-time negotiated parameters")
Signed-off-by: Lu Wei <luwei32@huawei.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/tcp.c

index ef14efa1fb70eaaecfc7884de32b74c49bb25a39..54836a6b81d6fa2059f312c753f62ee658bb6b18 100644 (file)
@@ -3647,7 +3647,7 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname,
        case TCP_REPAIR_OPTIONS:
                if (!tp->repair)
                        err = -EINVAL;
-               else if (sk->sk_state == TCP_ESTABLISHED)
+               else if (sk->sk_state == TCP_ESTABLISHED && !tp->bytes_sent)
                        err = tcp_repair_options_est(sk, optval, optlen);
                else
                        err = -EPERM;