> The RST problem is being caused by our changes. So, looks like
> you guys did the same splitting (of half-open and established but
> unaccepted connections) in 2.3 and thats why you said that you
> saw the problem in 2.3.
Yes, exaclty.
Actually, in 2.3 the problem is more deep.
In 2.3 packet may be queued not only on backlog,
it can spin on listening socket spinlock on another CPU,
so that even if we searched in established connection list,
we would fall to race window during accept().
> Did the splitup bringup other races in the code ?
In 2.2 it should not, if I remember correctly.
But probably, I forgot something. The whole year passed since
that time yet.
> (adding the code to do a established hash queue lookup
> in backlog processing fixes the problem !)
Yes, it does. 2.3/4 does exactly this.
Alexey
-
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/