tcp: always set retrans_stamp on recovery
authorYuchung Cheng <ycheng@google.com>
Wed, 16 Jan 2019 23:05:30 +0000 (15:05 -0800)
committerDavid S. Miller <davem@davemloft.net>
Thu, 17 Jan 2019 23:12:26 +0000 (15:12 -0800)
commit7ae189759cc48cf8b54beebff566e9fd2d4e7d7c
tree352db05710fad943494d48107a6fb4d7555d0a70
parent7f12422c4873e9b274bc151ea59cb0cdf9415cf1
tcp: always set retrans_stamp on recovery

Previously TCP socket's retrans_stamp is not set if the
retransmission has failed to send. As a result if a socket is
experiencing local issues to retransmit packets, determining when
to abort a socket is complicated w/o knowning the starting time of
the recovery since retrans_stamp may remain zero.

This complication causes sub-optimal behavior that TCP may use the
latest, instead of the first, retransmission time to compute the
elapsed time of a stalling connection due to local issues. Then TCP
may disrecard TCP retries settings and keep retrying until it finally
succeed: not a good idea when the local host is already strained.

The simple fix is to always timestamp the start of a recovery.
It's worth noting that retrans_stamp is also used to compare echo
timestamp values to detect spurious recovery. This patch does
not break that because retrans_stamp is still later than when the
original packet was sent.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/tcp_output.c
net/ipv4/tcp_timer.c