From: Andrey Shvetsov <andrey.shvetsov@xxxxxx> commit 4d1356ac12f4d5180d0df345d85ff0ee42b89c72 upstream. If the length of the socket buffer is 0xFFFFFFFF (max size for an unsigned int), then payload_len becomes 0xFFFFFFF1 after subtracting 14 (ETH_HLEN). Then, mdp_len is set to payload_len + 16 (MDP_HDR_LEN) which overflows and results in a value of 2. These values for payload_len and mdp_len will pass current buffer size checks. This patch checks if derived from skb->len sum may overflow. The check is based on the following idea: For any `unsigned V1, V2` and derived `unsigned SUM = V1 + V2`, `V1 + V2` overflows iif `SUM < V1`. Reported-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> Signed-off-by: Andrey Shvetsov <andrey.shvetsov@xxxxxx> Cc: stable <stable@xxxxxxxxxxxxxxx> Link: https://lore.kernel.org/r/20200116172238.6046-1-andrey.shvetsov@xxxxxxxxxxxxx Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> --- drivers/staging/most/aim-network/networking.c | 10 ++++++++++ 1 file changed, 10 insertions(+) --- a/drivers/staging/most/aim-network/networking.c +++ b/drivers/staging/most/aim-network/networking.c @@ -87,6 +87,11 @@ static int skb_to_mamac(const struct sk_ unsigned int payload_len = skb->len - ETH_HLEN; unsigned int mdp_len = payload_len + MDP_HDR_LEN; + if (mdp_len < skb->len) { + pr_err("drop: too large packet! (%u)\n", skb->len); + return -EINVAL; + } + if (mbo->buffer_length < mdp_len) { pr_err("drop: too small buffer! (%d for %d)\n", mbo->buffer_length, mdp_len); @@ -134,6 +139,11 @@ static int skb_to_mep(const struct sk_bu u8 *buff = mbo->virt_address; unsigned int mep_len = skb->len + MEP_HDR_LEN; + if (mep_len < skb->len) { + pr_err("drop: too large packet! (%u)\n", skb->len); + return -EINVAL; + } + if (mbo->buffer_length < mep_len) { pr_err("drop: too small buffer! (%d for %d)\n", mbo->buffer_length, mep_len);
