tcp: take care of overlaps in tcp_try_coalesce()
authorEric Dumazet <edumazet@google.com>
Wed, 23 May 2012 17:51:37 +0000 (17:51 +0000)
committerDavid S. Miller <davem@davemloft.net>
Thu, 24 May 2012 04:28:21 +0000 (00:28 -0400)
commit1ca7ee30630e1022dbcf1b51be20580815ffab73
tree5aab9c5bd4e31617afce262672c833db26381a23
parente49cc0da7283088c5e03d475ffe2fdcb24a6d5b1
tcp: take care of overlaps in tcp_try_coalesce()

Sergio Correia reported following warning :

WARNING: at net/ipv4/tcp.c:1301 tcp_cleanup_rbuf+0x4f/0x110()

WARN(skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq),
     "cleanup rbuf bug: copied %X seq %X rcvnxt %X\n",
     tp->copied_seq, TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt);

It appears TCP coalescing, and more specifically commit b081f85c297
(net: implement tcp coalescing in tcp_queue_rcv()) should take care of
possible segment overlaps in receive queue. This was properly done in
the case of out_or_order_queue by the caller.

For example, segment at tail of queue have sequence 1000-2000, and we
add a segment with sequence 1500-2500.
This can happen in case of retransmits.

In this case, just don't do the coalescing.

Reported-by: Sergio Correia <lists@uece.net>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Tested-by: Sergio Correia <lists@uece.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/tcp_input.c