tcp: do tcp_mstamp_refresh before retransmits on TSQ handler
authorKoichiro Den <den@klaipeden.com>
Sun, 22 Oct 2017 04:13:16 +0000 (13:13 +0900)
committerDavid S. Miller <davem@davemloft.net>
Mon, 23 Oct 2017 04:45:24 +0000 (05:45 +0100)
commit3a91d29f20276fa7cd4d0c9c7f3e78b30708159d
tree944c30bb3c71ccf6dee939462dc1fb66b86683d3
parentb5ac3beb5a9f0ef0ea64cd85faf94c0dc4de0e42
tcp: do tcp_mstamp_refresh before retransmits on TSQ handler

When retransmission on TSQ handler was introduced in the commit
f9616c35a0d7 ("tcp: implement TSQ for retransmits"), the retransmitted
skbs' timestamps were updated on the actual transmission. In the later
commit 385e20706fac ("tcp: use tp->tcp_mstamp in output path"), it stops
being done so. In the commit, the comment says "We try to refresh
tp->tcp_mstamp only when necessary", and at present tcp_tsq_handler and
tcp_v4_mtu_reduced applies to this. About the latter, it's okay since
it's rare enough.

About the former, even though possible retransmissions on the tasklet
comes just after the destructor run in NET_RX softirq handling, the time
between them could be nonnegligibly large to the extent that
tcp_rack_advance or rto rearming be affected if other (remaining) RX,
BLOCK and (preceding) TASKLET sofirq handlings are unexpectedly heavy.

So in the same way as tcp_write_timer_handler does, doing tcp_mstamp_refresh
ensures the accuracy of algorithms relying on it.

Fixes: 385e20706fac ("tcp: use tp->tcp_mstamp in output path")
Signed-off-by: Koichiro Den <den@klaipeden.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/tcp_output.c