[PATCH net-next 0/2] more accurate DSACK processing for RACK-TLP

This patch series includes two minor improvements to tighten up the accuracy of
the processing of incoming DSACK information, so that RACK-TLP behavior is
faster and more precise: first, to ensure we detect packet loss in some extra
corner cases; and second, to avoid growing the RACK reordering window (and
delaying fast recovery) in cases where it seems clear we don't need to.

Neal Cardwell (1):
  tcp: more accurately check DSACKs to grow RACK reordering window

Yuchung Cheng (1):
  tcp: more accurately detect spurious TLP probes

 net/ipv4/tcp_input.c    | 14 ++++++++++++--
 net/ipv4/tcp_recovery.c |  3 ++-
 2 files changed, 14 insertions(+), 3 deletions(-)

-- 
2.32.0.432.gabb21c7263-goog

From: Yuchung Cheng <redacted>

Previously TLP is considered spurious if the sender receives any
DSACK during a TLP episode. This patch further checks the DSACK
sequences match the TLP's to improve accuracy.

Signed-off-by: Yuchung Cheng <redacted>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Priyaranjan Jha <redacted>
---
 net/ipv4/tcp_input.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 501d8d4d4ba4..98408d520c32 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c

@@ -100,6 +100,7 @@ int sysctl_tcp_max_orphans __read_mostly = NR_FILE;
 #define FLAG_UPDATE_TS_RECENT	0x4000 /* tcp_replace_ts_recent() */
 #define FLAG_NO_CHALLENGE_ACK	0x8000 /* do not call tcp_send_challenge_ack()	*/
 #define FLAG_ACK_MAYBE_DELAYED	0x10000 /* Likely a delayed ACK */
+#define FLAG_DSACK_TLP		0x20000 /* DSACK for tail loss probe */
 
 #define FLAG_ACKED		(FLAG_DATA_ACKED|FLAG_SYN_ACKED)
 #define FLAG_NOT_DUP		(FLAG_DATA|FLAG_WIN_UPDATE|FLAG_ACKED)

@@ -991,6 +992,8 @@ static u32 tcp_dsack_seen(struct tcp_sock *tp, u32 start_seq,
 		return 0;
 	if (seq_len > tp->mss_cache)
 		dup_segs = DIV_ROUND_UP(seq_len, tp->mss_cache);
+	else if (tp->tlp_high_seq && tp->tlp_high_seq == end_seq)
+		state->flag |= FLAG_DSACK_TLP;
 
 	tp->dsack_dups += dup_segs;
 	/* Skip the DSACK if dup segs weren't retransmitted by sender */

@@ -3650,7 +3653,7 @@ static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag)
 	if (!tp->tlp_retrans) {
 		/* TLP of new data has been acknowledged */
 		tp->tlp_high_seq = 0;
-	} else if (flag & FLAG_DSACKING_ACK) {
+	} else if (flag & FLAG_DSACK_TLP) {
 		/* This DSACK means original and TLP probe arrived; no loss */
 		tp->tlp_high_seq = 0;
 	} else if (after(ack, tp->tlp_high_seq)) {

-- 
2.32.0.432.gabb21c7263-goog

Previously, a DSACK could expand the RACK reordering window when no
reordering has been seen, and/or when the DSACK was due to an
unnecessary TLP retransmit (rather than a spurious fast recovery due
to reordering). This could result in unnecessarily growing the RACK
reordering window and thus unnecessarily delaying RACK-based fast
recovery episodes.

To avoid these issues, this commit tightens the conditions under which
a DSACK triggers the RACK reordering window to grow, so that a
connection only expands its RACK reordering window if:

(a) reordering has been seen in the connection
(b) a DSACKed range does not match the most recent TLP retransmit

Signed-off-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <redacted>
Acked-by: Priyaranjan Jha <redacted>
---
 net/ipv4/tcp_input.c    | 9 ++++++++-
 net/ipv4/tcp_recovery.c | 3 ++-
 2 files changed, 10 insertions(+), 2 deletions(-)

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 98408d520c32..3f7bd7ae7d7a 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c

@@ -1001,7 +1001,14 @@ static u32 tcp_dsack_seen(struct tcp_sock *tp, u32 start_seq,
 		return 0;
 
 	tp->rx_opt.sack_ok |= TCP_DSACK_SEEN;
-	tp->rack.dsack_seen = 1;
+	/* We increase the RACK ordering window in rounds where we receive
+	 * DSACKs that may have been due to reordering causing RACK to trigger
+	 * a spurious fast recovery. Thus RACK ignores DSACKs that happen
+	 * without having seen reordering, or that match TLP probes (TLP
+	 * is timer-driven, not triggered by RACK).
+	 */
+	if (tp->reord_seen && !(state->flag & FLAG_DSACK_TLP))
+		tp->rack.dsack_seen = 1;
 
 	state->flag |= FLAG_DSACKING_ACK;
 	/* A spurious retransmission is delivered */

diff --git a/net/ipv4/tcp_recovery.c b/net/ipv4/tcp_recovery.c
index 6f1b4ac7fe99..fd113f6226ef 100644
--- a/net/ipv4/tcp_recovery.c
+++ b/net/ipv4/tcp_recovery.c

@@ -172,7 +172,8 @@ void tcp_rack_reo_timeout(struct sock *sk)
 
 /* Updates the RACK's reo_wnd based on DSACK and no. of recoveries.
  *
- * If DSACK is received, increment reo_wnd by min_rtt/4 (upper bounded
+ * If a DSACK is received that seems like it may have been due to reordering
+ * triggering fast recovery, increment reo_wnd by min_rtt/4 (upper bounded
  * by srtt), since there is possibility that spurious retransmission was
  * due to reordering delay longer than reo_wnd.
  *

-- 
2.32.0.432.gabb21c7263-goog

Hello:

This series was applied to netdev/net-next.git (refs/heads/master):

On Tue, 27 Jul 2021 10:42:56 -0400 you wrote:

This patch series includes two minor improvements to tighten up the accuracy of
the processing of incoming DSACK information, so that RACK-TLP behavior is
faster and more precise: first, to ensure we detect packet loss in some extra
corner cases; and second, to avoid growing the RACK reordering window (and
delaying fast recovery) in cases where it seems clear we don't need to.

Neal Cardwell (1):
  tcp: more accurately check DSACKs to grow RACK reordering window

[...]

Here is the summary with links:
  - [net-next,1/2] tcp: more accurately detect spurious TLP probes
    https://git.kernel.org/netdev/net-next/c/63f367d9de77
  - [net-next,2/2] tcp: more accurately check DSACKs to grow RACK reordering window
    https://git.kernel.org/netdev/net-next/c/a657db0350bb

You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html