Re: Help: major pppoe regression since 2.6.35 (panic on first ppp conection)?

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

Le mercredi 22 décembre 2010 à 17:00 +0100, Joel Soete a écrit :
> Hello Jarek,
> 
> Nice to read you :<)
> 
> On 12/22/2010 12:00 PM, Jarek Poplawski wrote:
> > On 2010-12-22 09:22, Andrew Morton wrote:
> >> (cc netdev)
> >>
> >> The bug is still present in 2.6.37-rc6.
> >>
> >> On Sat, 18 Dec 2010 11:33:14 +0000 Joel Soete<soete.joel@scarlet.be>  wrote:
> >>
> >>> Hello all,
> >
> > Hi,
> > Could you reproduce this bug with a vanilla kernel (without nvidia
> > patch)? If so, please include dmesg and .config to the next report.
> >
> Yes (it was already a vanilla kernel but 2.6.35 with my config, even thought same issue occurs some other distro stock 
> kernel 2.6.35), but here are some more dmesg with vanilla 2.6.37-rc6 and rc7 (I just added your debugging patch
> I found here, just because if I don't do it kernel is panicing immediately without letting any chance to capture dmesg (and 
> unfortunately I don't have any more chance to grab panic messages from serial console: no more rs232 on latest office laptop :<)
> 
> So you will find here attached personal config files of 2 kernels and respective dmesg.
> 
> If ever you need more details, don't hesitate to ask me.
> 
> Thanks a lot,
> 	J.

Something overwrites nr_frags in skb_shinfo(skb)

As skb_shinfo follows head portion of an skb, something overflows skb
head

Please try adding some room like in following patch ?

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index e6ba898..adf2834 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -187,6 +187,7 @@ enum {
  * the end of the header data, ie. at skb->end.
  */
 struct skb_shared_info {
+	char		filler[64];
 	unsigned short	nr_frags;
 	unsigned short	gso_size;
 	/* Warning: this field is not always filled in (UFO)! */