2007-06-19 07:56:53

by Olaf Kirch

Subject: [PATCH] net: Make skb_seq_read unmap the last fragment

From: Olaf Kirch <[email protected]>

Make skb_seq_read unmap the last fragment

Having walked through the entire skbuff, skb_seq_read would leave the
last fragment mapped. As a consequence, the unwary caller would leak
kmaps, and proceed with preempt_count off by one. The only (kind of
non-intuitive) workaround is to use skb_seq_read_abort.

This patch makes sure skb_seq_read always unmaps frag_data after having
cycled through the skb's paged part.

Signed-off-by: [email protected]
---
net/core/skbuff.c | 5 +++++
1 file changed, 5 insertions(+)

Index: build-2.6/net/core/skbuff.c
===================================================================
--- build-2.6.orig/net/core/skbuff.c
+++ build-2.6/net/core/skbuff.c
@@ -1706,6 +1706,11 @@ next_skb:
st->stepped_offset += frag->size;
}

+ if (st->frag_data) {
+ kunmap_skb_frag(st->frag_data);
+ st->frag_data = NULL;
+ }
+
if (st->cur_skb->next) {
st->cur_skb = st->cur_skb->next;
st->frag_idx = 0;

--
Olaf Kirch | --- o --- Nous sommes du soleil we love when we play
[email protected] | / | \ sol.dhoop.naytheet.ah kin.ir.samse.qurax

2007-06-24 06:11:59

by David Miller

Subject: Re: [PATCH] net: Make skb_seq_read unmap the last fragment

From: Olaf Kirch <[email protected]>
Date: Tue, 19 Jun 2007 09:56:24 +0200

> From: Olaf Kirch <[email protected]>
>
> Make skb_seq_read unmap the last fragment
>
> Having walked through the entire skbuff, skb_seq_read would leave the
> last fragment mapped. As a consequence, the unwary caller would leak
> kmaps, and proceed with preempt_count off by one. The only (kind of
> non-intuitive) workaround is to use skb_seq_read_abort.
>
> This patch makes sure skb_seq_read always unmaps frag_data after having
> cycled through the skb's paged part.
>
> Signed-off-by: [email protected]

Thanks for finding and fixing this bug.

Patch applied, thanks again Olaf.