lkml.org 
[lkml]   [2011]   [May]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [Stable-review] [patch 35/38] fs/partitions/ldm.c: fix oops caused by corrupted partition table
On Thu, 2011-05-05 at 17:11 -0700, Greg KH wrote:
> 2.6.38-stable review patch. If anyone has any objections, please let us know.
>
> ------------------
>
> From: Timo Warns <Warns@pre-sense.de>
>
> commit c340b1d640001c8c9ecff74f68fd90422ae2448a upstream.
>
> The kernel automatically evaluates partition tables of storage devices.
> The code for evaluating LDM partitions (in fs/partitions/ldm.c) contains
> a bug that causes a kernel oops on certain corrupted LDM partitions.
> A kernel subsystem seems to crash, because, after the oops, the kernel no
> longer recognizes newly connected storage devices.
>
> The patch validates the value of vblk_size.

I don't think this actually fixes the vulnerability, and I don't think
this code works at all.

> [akpm@linux-foundation.org: coding-style fixes]
> Signed-off-by: Timo Warns <warns@pre-sense.de>
> Cc: Eugene Teo <eugeneteo@kernel.sg>
> Cc: Harvey Harrison <harvey.harrison@gmail.com>
> Cc: Richard Russon <rich@flatcap.org>
> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
>
> ---
> fs/partitions/ldm.c | 16 ++++++++++++----
> 1 file changed, 12 insertions(+), 4 deletions(-)
>
> --- a/fs/partitions/ldm.c
> +++ b/fs/partitions/ldm.c
> @@ -1299,6 +1299,11 @@ static bool ldm_frag_add (const u8 *data
>
> BUG_ON (!data || !frags);
>
> + if (size < 2 * VBLK_SIZE_HEAD) {
> + ldm_error("Value of size is to small.");
> + return false;
> + }
> +
> group = get_unaligned_be32(data + 0x08);
> rec = get_unaligned_be16(data + 0x0C);
> num = get_unaligned_be16(data + 0x0E);
> @@ -1306,6 +1311,10 @@ static bool ldm_frag_add (const u8 *data
> ldm_error ("A VBLK claims to have %d parts.", num);
> return false;
> }
> + if (rec >= num) {
> + ldm_error("REC value (%d) exceeds NUM value (%d)", rec, num);
> + return false;
> + }

This is fine for the first fragment we find, when we allocate memory
based on 'num'. However, when we add another fragment, we need to
compare with f->num. So there still seems to be the possibility of a
buffer overflow.

> list_for_each (item, frags) {
> f = list_entry (item, struct frag, list);
> @@ -1334,10 +1343,9 @@ found:
>
> f->map |= (1 << rec);
>
> - if (num > 0) {
> - data += VBLK_SIZE_HEAD;
> - size -= VBLK_SIZE_HEAD;
> - }
> + data += VBLK_SIZE_HEAD;
> + size -= VBLK_SIZE_HEAD;
> +
>
> memcpy (f->data+rec*(size-VBLK_SIZE_HEAD)+VBLK_SIZE_HEAD, data, size);
>
> return true;

The offset used for the destination means that the first VBLK_SIZE_HEAD
bytes of f->data are never initialised!

I suspect (without any knowledge of LDM) that the intent was to use the
header from the first fragment and drop it from the subsequent
fragments, like this:

if (rec == 0)
memcpy(f->data, data, VBLK_SIZE_HEAD);
data += VBLK_SIZE_HEAD;
size -= VBLK_SIZE_HEAD;
memcpy(f->data + VBLK_SIZE_HEAD + rec * size, data, size);
Ben.

--
Ben Hutchings
Once a job is fouled up, anything done to improve it makes it worse.
[unhandled content-type:application/pgp-signature]
\
 
 \ /
  Last update: 2011-05-07 04:27    [from the cache]
©2003-2014 Jasper Spaans. Advertise on this site