net/xfrm_input: fix possible NULL deref of tunnel.ip6->parms.i_key
authorAlexey Kodanev <alexey.kodanev@oracle.com>
Wed, 10 Aug 2016 10:54:57 +0000 (13:54 +0300)
committerSteffen Klassert <steffen.klassert@secunet.com>
Thu, 11 Aug 2016 11:15:57 +0000 (13:15 +0200)
commit1625f4529957738be7d87cf157e107b8fb9d23b9
treee65c8b228135216df2f5b6e77c6e54a2976e3ec9
parent6916fb3b10b3cbe3b1f9f5b680675f53e4e299eb
net/xfrm_input: fix possible NULL deref of tunnel.ip6->parms.i_key

Running LTP 'icmp-uni-basic.sh -6 -p ipcomp -m tunnel' test over
openvswitch + veth can trigger kernel panic:

  BUG: unable to handle kernel NULL pointer dereference
  at 00000000000000e0 IP: [<ffffffff8169d1d2>] xfrm_input+0x82/0x750
  ...
  [<ffffffff816d472e>] xfrm6_rcv_spi+0x1e/0x20
  [<ffffffffa082c3c2>] xfrm6_tunnel_rcv+0x42/0x50 [xfrm6_tunnel]
  [<ffffffffa082727e>] tunnel6_rcv+0x3e/0x8c [tunnel6]
  [<ffffffff8169f365>] ip6_input_finish+0xd5/0x430
  [<ffffffff8169fc53>] ip6_input+0x33/0x90
  [<ffffffff8169f1d5>] ip6_rcv_finish+0xa5/0xb0
  ...

It seems that tunnel.ip6 can have garbage values and also dereferenced
without a proper check, only tunnel.ip4 is being verified. Fix it by
adding one more if block for AF_INET6 and initialize tunnel.ip6 with NULL
inside xfrm6_rcv_spi() (which is similar to xfrm4_rcv_spi()).

Fixes: 049f8e2 ("xfrm: Override skb->mark with tunnel->parm.i_key in xfrm_input")

Signed-off-by: Alexey Kodanev <alexey.kodanev@oracle.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
net/ipv6/xfrm6_input.c
net/xfrm/xfrm_input.c