tcp: only postpone PROBE_RTT if RTT is < current min_rtt estimate

During loss recovery, retransmitted packets are forced to use TCP
timestamps to calculate the RTT samples, which have a millisecond
granularity. BBR is designed using a microsecond granularity. As a
result, multiple RTT samples could be truncated to the same RTT value
during loss recovery. This is problematic, as BBR will not enter
PROBE_RTT if the RTT sample is <= the current min_rtt sample, meaning
that if there are persistent losses, PROBE_RTT will constantly be
pushed off and potentially never re-entered. This patch makes sure
that BBR enters PROBE_RTT by checking if RTT sample is < the current
min_rtt sample, rather than <=.

The Netflix transport/TCP team discovered this bug in the Linux TCP
BBR code during lab tests.

Fixes: 0f8782ea1497 ("tcp_bbr: add BBR congestion control")
Signed-off-by: Ryan Sharpelletti <sharpelletti@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Link: https://lore.kernel.org/r/20201116174412.1433277-1-sharpelletti.kdev@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

diff --git a/net/ipv4/tcp_bbr.c b/net/ipv4/tcp_bbr.c
index 6c4d79b..6ea3dc2 100644 (file)
--- a/net/ipv4/tcp_bbr.c
+++ b/net/ipv4/tcp_bbr.c
@@ -945,7 +945,7 @@ static void bbr_update_min_rtt(struct sock *sk, const struct rate_sample *rs)
        filter_expired = after(tcp_jiffies32,
                               bbr->min_rtt_stamp + bbr_min_rtt_win_sec * HZ);
        if (rs->rtt_us >= 0 &&
-           (rs->rtt_us <= bbr->min_rtt_us ||
+           (rs->rtt_us < bbr->min_rtt_us ||
             (filter_expired && !rs->is_ack_delayed))) {
                bbr->min_rtt_us = rs->rtt_us;
                bbr->min_rtt_stamp = tcp_jiffies32;