tcp_bbr: cut pacing rate only if filled pipe
authorNeal Cardwell <ncardwell@google.com>
Fri, 14 Jul 2017 21:49:21 +0000 (17:49 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 11 Aug 2017 15:49:31 +0000 (08:49 -0700)
commit73520d3814b6fd37483dc79e578906301122bbb3
tree9ffebab26f83f8a8a5a52cf9cac7ba3acc6a4a65
parent12d17d78e3f74b5022f61eee7d6de082e472a401
tcp_bbr: cut pacing rate only if filled pipe

[ Upstream commit 4aea287e90dd61a48268ff2994b56f9799441b62 ]

In bbr_set_pacing_rate(), which decides whether to cut the pacing
rate, there was some code that considered exiting STARTUP to be
equivalent to the notion of filling the pipe (i.e.,
bbr_full_bw_reached()). Specifically, as the code was structured,
exiting STARTUP and going into PROBE_RTT could cause us to cut the
pacing rate down to something silly and low, based on whatever
bandwidth samples we've had so far, when it's possible that all of
them have been small app-limited bandwidth samples that are not
representative of the bandwidth available in the path. (The code was
correct at the time it was written, but the state machine changed
without this spot being adjusted correspondingly.)

Fixes: 0f8782ea1497 ("tcp_bbr: add BBR congestion control")
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/ipv4/tcp_bbr.c