I think I may have narrowed down the cause of sockets stuck in
2.0.30/31/probably 32 too. Basically, very rarely, sockets appear stuck in
state CLOSE.
Currently, we have one in such a state; the timer endlessly cycles from
10s down to 0s. The port number in question appears in the kernel logs
like this:
ferret kernel: Warning: possible SYN flood from x.x.x.x on
163.1.138.204:7186. Sending cookies.
ferret kernel: validated proble(x.x.x.x:20, 163.1.138.204:7186, -<big
number>)
>From netstat after this event:
tcp 0 0 ferret.lmh.ox.ac.uk:7186 *.* CLOSE on (8.12/0)
I'm guessing that narrowing the problem down to SYN protection code thus,
will enable the bug to be fixed easily? Any more info wanted, just ask...
Cheers
Chris