From: Eric Dumazet edumazet@google.com
[ Upstream commit d59fc95be9d0fd05ed3ccc11b4a2f832bdf2ee03 ]
For TCP flows using ms RFC 7323 timestamp granularity tcp_rcv_rtt_update() can be fed with 1 ms samples, breaking TCP autotuning for data center flows with sub ms RTT.
Instead, rely on the window based samples, fed by tcp_rcv_rtt_measure()
tcp_rcvbuf_grow() for a 10 second TCP_STREAM sesssion now looks saner. We can see rcvbuf is kept at a reasonable value.
222.234976: tcp:tcp_rcvbuf_grow: time=348 rtt_us=330 copied=110592 inq=0 space=40960 ooo=0 scaling_ratio=230 rcvbuf=131072 ... 222.235276: tcp:tcp_rcvbuf_grow: time=300 rtt_us=288 copied=126976 inq=0 space=110592 ooo=0 scaling_ratio=230 rcvbuf=246187 ... 222.235569: tcp:tcp_rcvbuf_grow: time=294 rtt_us=288 copied=184320 inq=0 space=126976 ooo=0 scaling_ratio=230 rcvbuf=282659 ... 222.235833: tcp:tcp_rcvbuf_grow: time=264 rtt_us=244 copied=373760 inq=0 space=184320 ooo=0 scaling_ratio=230 rcvbuf=410312 ... 222.236142: tcp:tcp_rcvbuf_grow: time=308 rtt_us=219 copied=424960 inq=20480 space=373760 ooo=0 scaling_ratio=230 rcvbuf=832022 ... 222.236378: tcp:tcp_rcvbuf_grow: time=236 rtt_us=219 copied=692224 inq=49152 space=404480 ooo=0 scaling_ratio=230 rcvbuf=900407 ... 222.236602: tcp:tcp_rcvbuf_grow: time=225 rtt_us=219 copied=730112 inq=49152 space=643072 ooo=0 scaling_ratio=230 rcvbuf=1431534 ... 222.237050: tcp:tcp_rcvbuf_grow: time=229 rtt_us=219 copied=1160192 inq=49152 space=680960 ooo=0 scaling_ratio=230 rcvbuf=1515876 ... 222.237618: tcp:tcp_rcvbuf_grow: time=305 rtt_us=218 copied=2228224 inq=49152 space=1111040 ooo=0 scaling_ratio=230 rcvbuf=2473271 ... 222.238591: tcp:tcp_rcvbuf_grow: time=224 rtt_us=218 copied=3063808 inq=360448 space=2179072 ooo=0 scaling_ratio=230 rcvbuf=4850803 ... 222.240647: tcp:tcp_rcvbuf_grow: time=260 rtt_us=218 copied=2752512 inq=0 space=2703360 ooo=0 scaling_ratio=230 rcvbuf=6017914 ... 222.243535: tcp:tcp_rcvbuf_grow: time=224 rtt_us=218 copied=2834432 inq=49152 space=2752512 ooo=0 scaling_ratio=230 rcvbuf=6127331 ... 222.245108: tcp:tcp_rcvbuf_grow: time=240 rtt_us=218 copied=2883584 inq=49152 space=2785280 ooo=0 scaling_ratio=230 rcvbuf=6200275 ... 222.245333: tcp:tcp_rcvbuf_grow: time=224 rtt_us=218 copied=2859008 inq=0 space=2834432 ooo=0 scaling_ratio=230 rcvbuf=6309692 ... 222.301021: tcp:tcp_rcvbuf_grow: time=222 rtt_us=218 copied=2883584 inq=0 space=2859008 ooo=0 scaling_ratio=230 rcvbuf=6364400 ... 222.989242: tcp:tcp_rcvbuf_grow: time=225 rtt_us=218 copied=2899968 inq=0 space=2883584 ooo=0 scaling_ratio=230 rcvbuf=6419108 ... 224.139553: tcp:tcp_rcvbuf_grow: time=224 rtt_us=218 copied=3014656 inq=65536 space=2899968 ooo=0 scaling_ratio=230 rcvbuf=6455580 ... 224.584608: tcp:tcp_rcvbuf_grow: time=232 rtt_us=218 copied=3014656 inq=49152 space=2949120 ooo=0 scaling_ratio=230 rcvbuf=6564997 ... 230.145560: tcp:tcp_rcvbuf_grow: time=223 rtt_us=218 copied=2981888 inq=0 space=2965504 ooo=0 scaling_ratio=230 rcvbuf=6601469 ...
Signed-off-by: Eric Dumazet edumazet@google.com Reviewed-by: Wei Wang weiwan@google.com Link: https://patch.msgid.link/20250513193919.1089692-6-edumazet@google.com Signed-off-by: Jakub Kicinski kuba@kernel.org Signed-off-by: Sasha Levin sashal@kernel.org ---
**YES** This commit should be backported to stable kernel trees based on my analysis of the code changes and their impact. ## Extensive Analysis ### 1. **Bug Impact Assessment - Fixes User-Affecting Issues** The commit addresses a significant performance regression in data center environments. The evidence from the commit message shows real-world impact through detailed tracing data demonstrating improved TCP receive buffer autotuning behavior. Before the fix, zero RTT samples from ms- granularity TCP timestamps would corrupt autotuning for sub-ms RTT flows, causing: - Suboptimal receive buffer sizing - Poor performance in high-bandwidth, low-latency scenarios - Incorrect RTT estimates feeding into congestion control algorithms ### 2. **Code Changes Are Small and Contained** The changes are minimal and surgical: ```c // Modified tcp_rtt_tsopt_us() to take min_delta parameter -static s32 tcp_rtt_tsopt_us(const struct tcp_sock *tp) +static s32 tcp_rtt_tsopt_us(const struct tcp_sock *tp, u32 min_delta) { // When delta is 0, use min_delta instead of hardcoded 1 if (!delta) - delta = 1; + delta = min_delta; } // In tcp_rcv_rtt_measure_ts(): filter out zero samples for autotuning -s32 delta = tcp_rtt_tsopt_us(tp); +s32 delta = tcp_rtt_tsopt_us(tp, 0); -if (delta >= 0) +if (delta > 0) tcp_rcv_rtt_update(tp, delta, 0); // In tcp_ack_update_rtt(): preserve minimum 1µs for congestion control -seq_rtt_us = ca_rtt_us = tcp_rtt_tsopt_us(tp); +seq_rtt_us = ca_rtt_us = tcp_rtt_tsopt_us(tp, 1); ``` ### 3. **Follows Stable Tree Patterns from Similar Commits** Looking at the historical references provided, this commit follows the exact same pattern as previous TCP RTT measurement fixes that were successfully backported: - **Similar Commit #2** (YES): Fixed zero RTT samples from TCP timestamps - nearly identical issue - **Similar Commit #3** (YES): Fixed potential underestimation on rcv_rtt - same subsystem - **Similar Commit #4** (YES): Fixed timestamp refresh in tcp_rcv_space_adjust() - same autotuning code - **Similar Commit #5** (YES): Fixed rtt_min calculation - same RTT measurement subsystem ### 4. **No Architectural Changes or Risky Side Effects** The changes: - Don't introduce new features - Don't modify core TCP state machine logic - Don't change external APIs or user-visible behavior - Only affect internal RTT measurement accuracy - Have clear behavioral improvements with no regression risk ### 5. **Critical Subsystem But Safe Changes** While this touches the critical TCP input path, the modifications are: - Isolated to RTT measurement functions - Backward compatible (no protocol changes) - Defensive (filtering bad samples rather than adding new logic) - Following established patterns from prior successful fixes ### 6. **Data Center Performance Is Critical** Modern kernels are heavily used in data center environments where: - Sub-millisecond RTTs are common - TCP autotuning performance directly affects application throughput - The specific scenario (ms timestamp granularity breaking autotuning) is widespread - Performance regressions in this area affect major workloads ### 7. **Clear Technical Merit** The fix addresses a fundamental measurement accuracy problem where: - 1ms timestamp granularity creates zero RTT deltas for fast networks - Zero RTT samples poison autotuning algorithms designed for accurate measurements - The solution elegantly separates concerns: filter zeros for autotuning, preserve minimums for congestion control - The trace data in the commit message demonstrates measurable improvement ### 8. **Low Risk Profile** The changes have minimal regression risk because: - They only affect edge cases (zero timestamp deltas) - The fallback behavior (using window-based samples) was already present - No changes to normal case processing - Clear separation between autotuning and congestion control requirements This commit represents exactly the type of fix that stable trees are designed for: a small, contained fix for a user-affecting performance issue in a critical subsystem, with clear evidence of improvement and minimal regression risk.
net/ipv4/tcp_input.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index cf4fef18a9cad..61ada4682094f 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -707,7 +707,7 @@ static inline void tcp_rcv_rtt_measure(struct tcp_sock *tp) tp->rcv_rtt_est.time = tp->tcp_mstamp; }
-static s32 tcp_rtt_tsopt_us(const struct tcp_sock *tp) +static s32 tcp_rtt_tsopt_us(const struct tcp_sock *tp, u32 min_delta) { u32 delta, delta_us;
@@ -717,7 +717,7 @@ static s32 tcp_rtt_tsopt_us(const struct tcp_sock *tp)
if (likely(delta < INT_MAX / (USEC_PER_SEC / TCP_TS_HZ))) { if (!delta) - delta = 1; + delta = min_delta; delta_us = delta * (USEC_PER_SEC / TCP_TS_HZ); return delta_us; } @@ -735,9 +735,9 @@ static inline void tcp_rcv_rtt_measure_ts(struct sock *sk,
if (TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq >= inet_csk(sk)->icsk_ack.rcv_mss) { - s32 delta = tcp_rtt_tsopt_us(tp); + s32 delta = tcp_rtt_tsopt_us(tp, 0);
- if (delta >= 0) + if (delta > 0) tcp_rcv_rtt_update(tp, delta, 0); } } @@ -3220,7 +3220,7 @@ static bool tcp_ack_update_rtt(struct sock *sk, const int flag, */ if (seq_rtt_us < 0 && tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr && flag & FLAG_ACKED) - seq_rtt_us = ca_rtt_us = tcp_rtt_tsopt_us(tp); + seq_rtt_us = ca_rtt_us = tcp_rtt_tsopt_us(tp, 1);
rs->rtt_us = ca_rtt_us; /* RTT of last (S)ACKed packet (or -1) */ if (seq_rtt_us < 0)