Kevin> What causes a "Socket destroy delayed" message? I mean,
Kevin> specifically...
Well if we did know really "specifically" the problem should have been
eliminated already ;-)
That message is caused when a socket gets closed and some of the internal
buffers it allocated (sk_buffs) where not freed...
now where can the missing skb be ? guesses:
- in an ARP queue
- really lost when going in and out the device queues
- not freed by the device driver
- etc...
mainly i would check for what happens when dev->start_xmit returns 1
... possible race conditions with sockets closing while the skb is out
of the queue and get into the device queue again, etc.
regards,
Pedro.