Re: oops in tcp_xmit_retransmit_queue() w/ v2.6.32.15

[Eric Dumazet <eric.dumazet@gmail.com>]

Le dimanche 11 juillet 2010 à 19:06 +0200, Eric Dumazet a écrit :
> Le dimanche 11 juillet 2010 à 19:09 +0300, Ilpo Järvinen a écrit :
> > On Thu, 8 Jul 2010, Tejun Heo wrote:
> > 
> > > We've been seeing oops in tcp_xmit_retransmit_queue() w/ 2.6.32.15.
> > > Please see the attached photoshoot.  This is happening on a HPC
> > > cluster and very interestingly caused by one particular job.  How long
> > > it takes isn't clear yet (at least more than a day) but when it
> > > happens it happens on a lot of machines in relatively short time.
> > > 
> > > With a bit of disassemblying, I've found that the oops is happening
> > > during tcp_for_write_queue_from() because the skb->next points to
> > > NULL.
> > > 
> > >  void tcp_xmit_retransmit_queue(struct sock *sk)
> > >  {
> > >  ...
> > > 	if (tp->retransmit_skb_hint) {
> > > 		skb = tp->retransmit_skb_hint;
> > > 		last_lost = TCP_SKB_CB(skb)->end_seq;
> > > 		if (after(last_lost, tp->retransmit_high))
> > > 			last_lost = tp->retransmit_high;
> > > 	} else {
> > > 		skb = tcp_write_queue_head(sk);
> > > 		last_lost = tp->snd_una;
> > > 	}
> > > 
> > >  =>	tcp_for_write_queue_from(skb, sk) {
> > > 		 __u8 sacked = TCP_SKB_CB(skb)->sacked;
> > > 
> > > 		 if (skb == tcp_send_head(sk))
> > > 			 break;
> > > 		 /* we could do better than to assign each time */
> > > 		 if (hole == NULL)
> > > 
> > > This can happen for one of the following reasons,
> > > 
> > > 1. tp->retransmit_skb_hint is NULL and tcp_write_queue_head() is NULL
> > >    too.  ie. tcp_xmit_retransmit_queue() is called on an empty write
> > >    queue for some reason.
> > > 
> > > 2. tp->retransmit_skb_hint is pointing to a skb which is not on the
> > >    write_queue.  ie. somebody forgot to update hint while removing the
> > >    skb from the write queue.
> > 
> > Once again I've read the unlinkers through, and only thing that could 
> > cause this is tcp_send_synack (others do deal with the hints) but I think 
> > Eric already proposed a patch to that but we never got anywhere due to 
> > some counterargument why it wouldn't take place (too far away for me to 
> > remember, see archives about the discussions). ...But if you want be dead 
> > sure some WARN_ON there might not hurt. Also the purging of the whole 
> > queue was a similar suspect I then came across (but that would only 
> > materialize with sk reuse happening e.g., with nfs which the other guys 
> > weren't using).
> > 
> 
> Hmm.
> 
> This sounds familiar to me, but I cannot remember the discussion you
> mention or the patch.
> 
> Or maybe it was the TCP transaction thing ? (including data in SYN or
> SYN-ACK packet)

Hmm, I cannot find where we reset restransmit_skb_hint in
tcp_mtu_probe(), if we call tcp_unlink_write_queue().

if (skb->len <= copy) {
	/* We've eaten all the data from this skb.
	 * Throw it away. */
	TCP_SKB_CB(nskb)->flags |= TCP_SKB_CB(skb)->flags;
<<>>	tcp_unlink_write_queue(skb, sk);
	sk_wmem_free_skb(sk, skb);
} else {


Sorry if this was already discussed. We might add a comment here in anycase ;)