tcp: use sequence to break TS ties for RACK loss detection
authorYuchung Cheng <ycheng@google.com>
Fri, 13 Jan 2017 06:11:34 +0000 (22:11 -0800)
committerDavid S. Miller <davem@davemloft.net>
Sat, 14 Jan 2017 03:37:16 +0000 (22:37 -0500)
commit1d0833df594390876647c54c2c88069d29059665
treeabd620968abad8f56e132a108b5bf354eb0d3b54
parent57dde7f70de34d4251f291c9eac7ad920aaf56b2
tcp: use sequence to break TS ties for RACK loss detection

The packets inside a jumbo skb (e.g., TSO) share the same skb
timestamp, even though they are sent sequentially on the wire. Since
RACK is based on time, it can not detect some packets inside the
same skb are lost.  However, we can leverage the packet sequence
numbers as extended timestamps to detect losses. Therefore, when
RACK timestamp is identical to skb's timestamp (i.e., one of the
packets of the skb is acked or sacked), we use the sequence numbers
of the acked and unacked packets to break ties.

We can use the same sequence logic to advance RACK xmit time as
well to detect more losses and avoid timeout.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/linux/tcp.h
include/net/tcp.h
net/ipv4/tcp_input.c
net/ipv4/tcp_recovery.c