On 8/14/26 03:45, Hyunwoo Kim wrote:
> On Thu, Aug 13, 2026 at 11:42:17AM +0200, Michal Luczaj wrote:
>> On 8/12/26 22:13, Hyunwoo Kim wrote:
>>> commit 002541ef650b ("vsock: Ignore signal/timeout on connect() if
>>> already established") stopped connect() from resetting an established
>>> socket. The check only looks at whether sk_state is TCP_ESTABLISHED at
>>> that moment, and the state can change while connect() sleeps.
>>
>> I guess this makes my fix incomplete. "Fixes: 002541ef650b"?
>
> It is incomplete, yes. But this has been triggerable since d021c344051a, so
> I'd keep Fixes: d021c344051a.
OK, I get it.
>>> A peer RST moves the socket to TCP_CLOSING, and it is not removed from
>>> vsock_connected_table on that path. connect() then wakes up, fails the
>>> check, and resets a socket that had actually connected to TCP_CLOSE and
>>> SS_UNCONNECTED.
>>
>> Thanks for the details. Do I get it right: connect() misses the fact that
>> socket might have already transitioned TCP_ESTABLISHED -> TCP_CLOSING
>> during schedule_timeout()?
>
> Yes, that's it.
>
>>
>> How about:
>>
>> diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
>> index 622dbd046799..39c42ef016c3 100644
>> --- a/net/vmw_vsock/af_vsock.c
>> +++ b/net/vmw_vsock/af_vsock.c
>> @@ -1807,15 +1807,18 @@ static int vsock_connect(struct socket *sock, struct
>> sockaddr_unsized *addr,
>> timeout = schedule_timeout(timeout);
>> lock_sock(sk);
>>
>> - /* Connection established. Whatever happens to socket once we
>> + /* Connection (has been) established. Whatever happens to
>> socket once we
>> * release it, that's not connect()'s concern. No need to go
>> * into signal and timeout handling. Call it a day.
>> *
>> * Note that allowing to "reset" an already established socket
>> * here is racy and insecure.
>> */
>> - if (sk->sk_state == TCP_ESTABLISHED)
>> - break;
>> + if (sk->sk_state == TCP_ESTABLISHED ||
>> + sk->sk_state == TCP_CLOSING) {
>> + err = -sk->sk_err;
>> + goto out_wait;
>> + }
>>
>> /* If connection was _not_ established and a signal/timeout came
>> * to be, we want the socket's state reset. User space may want
>>
>> ?
>
> Yes, I like it better than mine. I confirmed it fixes the issue.
Great, thanks.
> If you don't mind, would you take the patch from here?
Sure, no problem.
Stefano, does this look good to you? And should any sk_err be consumed
here, too? (`err = sock_error(sk)` instead of `err = -sk->sk_err`)
thanks,
Michal
