From: Tomas Szepe <szepe@pinerecords.com>
To: "David S. Miller" <davem@redhat.com>
Cc: linux-kernel@vger.kernel.org, netdev@oss.sgi.com, grof@dragon.cz,
volf@dragon.cz
Subject: Re: possible bug in tcp_input.c
Date: Sun, 26 Oct 2003 07:55:19 +0100 [thread overview]
Message-ID: <20031026065519.GC28035@louise.pinerecords.com> (raw)
In-Reply-To: <20031024193034.30f1caed.davem@redhat.com>
On Oct-24 2003, Fri, 19:30 -0700
David S. Miller <davem@redhat.com> wrote:
> > The passed NULL (and yes, this is where we are getting one) is dereferenced
> > immediately in:
> >
> > /* tcp_input.c, line 1133 */
> > static inline int tcp_skb_timedout(struct tcp_opt *tp, struct sk_buff *skb)
> > {
> > return (tcp_time_stamp - TCP_SKB_CB(skb)->when > tp->rto);
> > }
>
> If tp->packets_out is non-zero (which by definition it is
> in your case else the right hand side of the "&&" would not be
> evaluated) then we _MUST_ have some packets in sk->write_queue.
>
> Something is being fiercely corrupted. Probably some piece of
> netfilter is freeing up an SKB one too many times thus corrupting
> the TCP write queue list pointers.
Dave, we've been thinking about this some more and have concluded
that since the problem is a relatively non-fatal one, the kernel
should just print out an "assertion failed" error similar to the
one in tcp_input.c, line 1323 [BUG_TRAP(cnt <= tp->packets_out);]
and maybe fix things up a little rather than oops on a NULL pointer
dereference; The state in question, although invalid, is possible
and should IMHO be checked for as in all the other "if (skb != NULL)
..." places).
What do you think? We keep on trying to locate which code is causing
the corruption, meanwhile the affected system has been running crash-lessly
with the attached patch.
Thanks,
--
Tomas Szepe <szepe@pinerecords.com>
diff -urN a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
--- a/net/ipv4/tcp_input.c 2003-06-13 16:51:39 +0200
+++ b/net/ipv4/tcp_input.c 2003-10-26 07:29:11 +0100
@@ -1138,7 +1138,19 @@
static inline int tcp_head_timedout(struct sock *sk, struct tcp_opt *tp)
{
- return tp->packets_out && tcp_skb_timedout(tp, skb_peek(&sk->write_queue));
+ struct sk_buff *skb = skb_peek(&sk->write_queue);
+
+ if (skb == NULL) {
+ if (tp->packets_out) {
+ printk("KERNEL: assertion (%s) failed at %s(%d)\n",
+ "skb == NULL && tp->packets_out",
+ __FILE__, __LINE__);
+ tp->packets_out = 0;
+ }
+ return 0;
+ } else {
+ return tp->packets_out && tcp_skb_timedout(tp, skb);
+ }
}
/* Linux NewReno/SACK/FACK/ECN state machine.
next prev parent reply other threads:[~2003-10-26 6:55 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2003-10-24 16:29 possible bug in tcp_input.c Tomas Szepe
2003-10-25 2:30 ` David S. Miller
2003-10-25 9:12 ` Tomas Szepe
2003-10-26 6:55 ` Tomas Szepe [this message]
2003-10-27 6:33 ` David S. Miller
[not found] <20031024162959.GB11154@louise.pinerecords.com.suse.lists.linux.kernel>
2003-10-24 17:57 ` Andi Kleen
2003-11-18 13:58 ` Tomas Szepe
2003-11-18 14:01 ` Andi Kleen
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20031026065519.GC28035@louise.pinerecords.com \
--to=szepe@pinerecords.com \
--cc=davem@redhat.com \
--cc=grof@dragon.cz \
--cc=linux-kernel@vger.kernel.org \
--cc=netdev@oss.sgi.com \
--cc=volf@dragon.cz \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).