net: qdisc_pkt_len_init() should be more robust
authorEric Dumazet <edumazet@google.com>
Fri, 19 Jan 2018 03:59:19 +0000 (19:59 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 31 Jan 2018 13:03:45 +0000 (14:03 +0100)
commit450449fff22687db76d7611acb48f94b685bc762
treed2a88c8c9d9171e6e557366bdfda9d463be92bc8
parentd9bee33e397150b43be9ac2271b813552a94345f
net: qdisc_pkt_len_init() should be more robust

[ Upstream commit 7c68d1a6b4db9012790af7ac0f0fdc0d2083422a ]

Without proper validation of DODGY packets, we might very well
feed qdisc_pkt_len_init() with invalid GSO packets.

tcp_hdrlen() might access out-of-bound data, so let's use
skb_header_pointer() and proper checks.

Whole story is described in commit d0c081b49137 ("flow_dissector:
properly cap thoff field")

We have the goal of validating DODGY packets earlier in the stack,
so we might very well revert this fix in the future.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Cc: Jason Wang <jasowang@redhat.com>
Reported-by: syzbot+9da69ebac7dddd804552@syzkaller.appspotmail.com
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/core/dev.c