gso_segment: Reset skb->mac_len after modifying network header
authorToke Høiland-Jørgensen <toke@toke.dk>
Thu, 13 Sep 2018 14:43:07 +0000 (16:43 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sat, 29 Sep 2018 10:06:00 +0000 (03:06 -0700)
commit13a47054f0b2f00c0f916387a1d4cf5402848add
treee1139230097009131a7b9b61ad056e4c6508cb8d
parent2cc4d365363b1fb681b8231adcf4a8f80082506c
gso_segment: Reset skb->mac_len after modifying network header

[ Upstream commit c56cae23c6b167acc68043c683c4573b80cbcc2c ]

When splitting a GSO segment that consists of encapsulated packets, the
skb->mac_len of the segments can end up being set wrong, causing packet
drops in particular when using act_mirred and ifb interfaces in
combination with a qdisc that splits GSO packets.

This happens because at the time skb_segment() is called, network_header
will point to the inner header, throwing off the calculation in
skb_reset_mac_len(). The network_header is subsequently adjust by the
outer IP gso_segment handlers, but they don't set the mac_len.

Fix this by adding skb_reset_mac_len() calls to both the IPv4 and IPv6
gso_segment handlers, after they modify the network_header.

Many thanks to Eric Dumazet for his help in identifying the cause of
the bug.

Acked-by: Dave Taht <dave.taht@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/ipv4/af_inet.c
net/ipv6/ip6_offload.c