On 02/24/2016 12:08 AM, P J P wrote:
> From: Prasad J Pandit <p...@fedoraproject.org>
>
> While computing IP checksum, 'net_checksum_calculate' reads
> payload length from the packet. It could exceed the given 'data'
> buffer size. Add a check to avoid it.
>
> Reported-by: Liu Ling <liuling...@360.cn>
> Signed-off-by: Prasad J Pandit <p...@fedoraproject.org>
> ---
> net/checksum.c | 10 ++++++++--
> 1 file changed, 8 insertions(+), 2 deletions(-)
>
> Update as per review:
> -> https://lists.gnu.org/archive/html/qemu-devel/2016-02/msg04062.html
>
> diff --git a/net/checksum.c b/net/checksum.c
> index 14c0855..bd89083 100644
> --- a/net/checksum.c
> +++ b/net/checksum.c
> @@ -59,6 +59,11 @@ void net_checksum_calculate(uint8_t *data, int length)
> int hlen, plen, proto, csum_offset;
> uint16_t csum;
>
I must say this is still far from perfect, since it has too assumptions
. But I agree we can fix OOB first.
> + /* Ensure data has complete L2 & L3 headers. */
> + if (length < 14 + 20) {
> + return;
> + }
> +
> if ((data[14] & 0xf0) != 0x40)
> return; /* not IPv4 */
> hlen = (data[14] & 0x0f) * 4;
> @@ -76,8 +81,9 @@ void net_checksum_calculate(uint8_t *data, int length)
> return;
> }
>
> - if (plen < csum_offset+2)
> - return;
> + if (plen < csum_offset + 2 || plen + hlen >= length) {
> + return;
> + }
Should we count mac header here? Did "plen + hlen >= length" imply "14 +
hlen + csum_offset + 1" < length?
Looks not. Consider a TCP packet can report evil plen (e.g 20) but just
have 10 bytes payload in fact. In this case:
hlen = 20, plen = 20, csum_offset = 16, length = 44 which can pass all
the above tests, but 14 + hlen + csum_offset = 50 which is greater than
length.
>
> data[14+hlen+csum_offset] = 0;
> data[14+hlen+csum_offset+1] = 0;
