Re: [PANIC] lro + iscsi or lro + skb text search causes panic

[David Miller <davem@davemloft.net>]

From: Herbert Xu <herbert@gondor.apana.org.au>
Date: Tue, 27 Jan 2009 09:30:22 +1100

> net: Fix frag_list handling in skb_seq_read
> 
> The frag_list handling was broken in skb_seq_read:
> 
> 1) We didn't add the stepped offset when looking at the head
> are of fragments other than the first.
> 
> 2) We didn't take the stepped offset away when setting the data
> pointer in the head area.
> 
> 3) The frag index wasn't reset.
> 
> This patch fixes both issues.
> 
> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

I see, the code is only clearing the fragment index when it's
advancing from one SKB to the next while already in the middle of a
->frag_list, not when transitioning past the root skb in such a list.

I bet some weird cases happen when "consumed" it's advanced by
the caller the entire length of data returned by the previous
skb_seq_read().

It all seems to be designed to work for that case, however.

Anyways, Herbert's patch looks definitely correct but until
we've gotten these crashes and hangs solved I don't want
to apply it.