aboutsummaryrefslogtreecommitdiffstats
path: root/src/vnet/tcp
diff options
context:
space:
mode:
authorFlorin Coras <fcoras@cisco.com>2022-01-06 11:58:24 -0800
committerFlorin Coras <fcoras@cisco.com>2022-01-06 16:38:24 -0800
commitc8144356338995a277593bc08315d3b1ddcc8cb3 (patch)
tree4eed6c04bf1e924e9d7312efe20af34db429c88b /src/vnet/tcp
parented5be47009aca45008ba9d44e76ac51ec1f7f156 (diff)
tcp: update snd_congestion only during congestion
If running without sacks, if snd_una does not cover snd_congestion fast recovery can be missed but the two heuristics from RFC6582 should avoid that. Also snd_congestion was used as a means of inferring if the connection recently exited congestion while setting the persist timer but that does not always work correctly if not congested. Type: fix Signed-off-by: Florin Coras <fcoras@cisco.com> Change-Id: I94d4ac738cdd4f7f23f62e97dd63059de1cd4af9
Diffstat (limited to 'src/vnet/tcp')
-rw-r--r--src/vnet/tcp/tcp.c4
-rw-r--r--src/vnet/tcp/tcp_input.c9
-rw-r--r--src/vnet/tcp/tcp_timer.h32
3 files changed, 14 insertions, 31 deletions
diff --git a/src/vnet/tcp/tcp.c b/src/vnet/tcp/tcp.c
index 3d3cc886024..ffe5c895cc9 100644
--- a/src/vnet/tcp/tcp.c
+++ b/src/vnet/tcp/tcp.c
@@ -71,6 +71,10 @@ tcp_add_del_adjacency (tcp_connection_t * tc, u8 is_add)
static void
tcp_cc_init (tcp_connection_t * tc)
{
+ /* As per RFC 6582 initialize "recover" to iss */
+ if (tcp_opts_sack_permitted (&tc->rcv_opts))
+ tc->snd_congestion = tc->iss;
+
tc->cc_algo->init (tc);
}
diff --git a/src/vnet/tcp/tcp_input.c b/src/vnet/tcp/tcp_input.c
index 11491a68257..64b9334cbea 100644
--- a/src/vnet/tcp/tcp_input.c
+++ b/src/vnet/tcp/tcp_input.c
@@ -802,15 +802,6 @@ tcp_cc_update (tcp_connection_t * tc, tcp_rate_sample_t * rs)
/* If a cumulative ack, make sure dupacks is 0 */
tc->rcv_dupacks = 0;
-
- /* When dupacks hits the threshold we only enter fast retransmit if
- * cumulative ack covers more than snd_congestion. Should snd_una
- * wrap this test may fail under otherwise valid circumstances.
- * Therefore, proactively update snd_congestion when wrap detected. */
- if (PREDICT_FALSE
- (seq_leq (tc->snd_congestion, tc->snd_una - tc->bytes_acked)
- && seq_gt (tc->snd_congestion, tc->snd_una)))
- tc->snd_congestion = tc->snd_una - 1;
}
/**
diff --git a/src/vnet/tcp/tcp_timer.h b/src/vnet/tcp/tcp_timer.h
index 5f14a71ee7d..7f7dbf193eb 100644
--- a/src/vnet/tcp/tcp_timer.h
+++ b/src/vnet/tcp/tcp_timer.h
@@ -51,6 +51,13 @@ tcp_timer_update (tcp_timer_wheel_t * tw, tcp_connection_t * tc, u8 timer_id,
timer_id, interval);
}
+always_inline u8
+tcp_timer_is_active (tcp_connection_t *tc, tcp_timers_e timer)
+{
+ return tc->timers[timer] != TCP_TIMER_HANDLE_INVALID ||
+ (tc->pending_timers & (1 << timer));
+}
+
always_inline void
tcp_retransmit_timer_set (tcp_timer_wheel_t * tw, tcp_connection_t * tc)
{
@@ -74,19 +81,6 @@ tcp_persist_timer_set (tcp_timer_wheel_t * tw, tcp_connection_t * tc)
}
always_inline void
-tcp_persist_timer_update (tcp_timer_wheel_t * tw, tcp_connection_t * tc)
-{
- u32 interval;
-
- if (seq_leq (tc->snd_una, tc->snd_congestion + tc->burst_acked))
- interval = 1;
- else
- interval = clib_max ((u32) tc->rto * TCP_TO_TIMER_TICK, 1);
-
- tcp_timer_update (tw, tc, TCP_TIMER_PERSIST, interval);
-}
-
-always_inline void
tcp_persist_timer_reset (tcp_timer_wheel_t * tw, tcp_connection_t * tc)
{
tcp_timer_reset (tw, tc, TCP_TIMER_PERSIST);
@@ -98,21 +92,15 @@ tcp_retransmit_timer_update (tcp_timer_wheel_t * tw, tcp_connection_t * tc)
if (tc->snd_una == tc->snd_nxt)
{
tcp_retransmit_timer_reset (tw, tc);
- if (tc->snd_wnd < tc->snd_mss)
- tcp_persist_timer_update (tw, tc);
+ if (tc->snd_wnd < tc->snd_mss &&
+ !tcp_timer_is_active (tc, TCP_TIMER_PERSIST))
+ tcp_persist_timer_set (tw, tc);
}
else
tcp_timer_update (tw, tc, TCP_TIMER_RETRANSMIT,
clib_max ((u32) tc->rto * TCP_TO_TIMER_TICK, 1));
}
-always_inline u8
-tcp_timer_is_active (tcp_connection_t * tc, tcp_timers_e timer)
-{
- return tc->timers[timer] != TCP_TIMER_HANDLE_INVALID
- || (tc->pending_timers & (1 << timer));
-}
-
always_inline void
tcp_timer_expire_timers (tcp_timer_wheel_t * tw, f64 now)
{