> What really happens is:
>
> app calls connect()
> socket is set to SS_CONNECTING
> -> signal
> ICMP happens in between
> connect is restarted
> sees sk->err returns and clears error but does _not_ set SS_UNCONNECTED
> application retries connect
> inet_wait_for_connect (because the socket is still SYN_SENT)
> -> sleeps forever because no wake up happens and no error
> is set because it is already cleared.
>
> Here is the better fix which always clears SS_CONNECTING, I'm still
> waiting for Jason's feedback on this one. Seems the "fix" to move
> the tcp_set_state(TCP_CLOSE) out of the icmp error handler was worse
> medicine than the original problem.
Ik, it seems that the machine was running fine for about 4 days and now
I'm looking at it today and their are 54 stuck connections :<
I don't know if your patch helped at all or if some network/user condition
mearly shifted to hide the problem for 4 days.
It is odd, according to my mrtg graphs I'd say that most of these appeared
in the last 20 hours, outbound traffic levels steadally decreased as the
ftp user limit filled with inactive ftpds
Sorry,
Jason
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/