tcp: Don't access TCP_SKB_CB before initializing it
authorChristoph Paasch <cpaasch@apple.com>
Mon, 11 Mar 2019 18:41:05 +0000 (11:41 -0700)
committerDavid S. Miller <davem@davemloft.net>
Mon, 11 Mar 2019 22:36:49 +0000 (15:36 -0700)
commitf2feaefdabb0a6253aa020f65e7388f07a9ed47c
treed79b2b98513771ba6e4460a58cf8d25ae93fcded
parent0b31d98d90f09868dce71319615e19cd1f146fb6
tcp: Don't access TCP_SKB_CB before initializing it

Since commit eeea10b83a13 ("tcp: add
tcp_v4_fill_cb()/tcp_v4_restore_cb()"), tcp_vX_fill_cb is only called
after tcp_filter(). That means, TCP_SKB_CB(skb)->end_seq still points to
the IP-part of the cb.

We thus should not mock with it, as this can trigger bugs (thanks
syzkaller):
[   12.349396] ==================================================================
[   12.350188] BUG: KASAN: slab-out-of-bounds in ip6_datagram_recv_specific_ctl+0x19b3/0x1a20
[   12.351035] Read of size 1 at addr ffff88006adbc208 by task test_ip6_datagr/1799

Setting end_seq is actually no more necessary in tcp_filter as it gets
initialized later on in tcp_vX_fill_cb.

Cc: Eric Dumazet <edumazet@google.com>
Fixes: eeea10b83a13 ("tcp: add tcp_v4_fill_cb()/tcp_v4_restore_cb()")
Signed-off-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/tcp_ipv4.c