Skip to contentSkip to conversation

tcp: do not change rcv_ssthresh in tcp_measure_rcv_mss()

PATCH Applied Nathan Gao 5 participants 7 messages Jul 25 – Aug 3

Subsystems (3)
lore
Share & tools
lore

Nathan Gao

author
#0 lore raw

PATCH 1/1 tcp: do not change rcv_ssthresh in tcp_measure_rcv_mss()

Commit f5da7c45188e ("tcp: adjust rcvq_space after updating scaling ratio") replaced the direct window_clamp update in tcp_measure_rcv_mss() with a call to tcp_set_window_clamp(), a helper that implements the TCP_WINDOW_CLAMP setsockopt. As a side effect, the helper also shrinks rcv_ssthresh via __tcp_adjust_rcv_ssthresh().

As a result, each scaling_ratio decrease detected by tcp_measure_rcv_mss() also cuts rcv_ssthresh. Elsewhere in TCP, rcv_ssthresh is usually cut under memory pressure and grows via tcp_grow_window().

Flows whose segment sizes vary keep scaling_ratio oscillating, which leads to an unstable rcv_ssthresh: a dip of rcv_ssthresh only recovers via tcp_grow_window(), keeping the advertised window at a relatively low level even after the ratio itself has recovered, and can even stall the sender.

Observed on a customer's proxy gateway after upgrading from kernel 6.1 to 6.12: in the worst case, rcv_ssthresh was cut in half by a scaling_ratio dip. P99 latency jumped from <10ms on 6.1 to ~100ms on 6.12, and almost returned to the 6.1 level with this patch applied.

Restore the plain WRITE_ONCE() update of window_clamp, as introduced in commit a2cbb1603943 ("tcp: Update window clamping condition"), and keep the rcvq_space.space adjustment. Now rcv_ssthresh is decoupled from scaling_ratio changes in tcp_measure_rcv_mss().

Fixes f5da7c45188e ("tcp: adjust rcvq_space after updating scaling ratio")
---
 net/ipv4/tcp_input.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/ipv4/tcp_input.c +1 -1
index daff93d513428..5b6378b94701e 100644
@@ -252,7 +252,7 @@ static void tcp_measure_rcv_mss(struct sock *sk, const struct sk_buff *skb)
 				struct tcp_sock *tp = tcp_sk(sk);
 
 				val = tcp_win_from_space(sk, sk->sk_rcvbuf);
-				tcp_set_window_clamp(sk, val);
+				WRITE_ONCE(tp->window_clamp, val);
 
 				if (tp->window_clamp < tp->rcvq_space.space)
 					tp->rcvq_space.space = tp->window_clamp;

base-commit: 78f75d632f74b8de0f081a128588f7c37d0d1164

Signature · 1 line
2.47.3
maintainer
#1 lore raw
Quoted 22 lines · On Fri, 24 Jul 2026 20:08:06 -0700 Nathan Gao wrote:

Commit f5da7c45188e ("tcp: adjust rcvq_space after updating scaling ratio") replaced the direct window_clamp update in tcp_measure_rcv_mss() with a call to tcp_set_window_clamp(), a helper that implements the TCP_WINDOW_CLAMP setsockopt. As a side effect, the helper also shrinks rcv_ssthresh via __tcp_adjust_rcv_ssthresh().

As a result, each scaling_ratio decrease detected by tcp_measure_rcv_mss() also cuts rcv_ssthresh. Elsewhere in TCP, rcv_ssthresh is usually cut under memory pressure and grows via tcp_grow_window().

Flows whose segment sizes vary keep scaling_ratio oscillating, which leads to an unstable rcv_ssthresh: a dip of rcv_ssthresh only recovers via tcp_grow_window(), keeping the advertised window at a relatively low level even after the ratio itself has recovered, and can even stall the sender.

Observed on a customer's proxy gateway after upgrading from kernel 6.1 to 6.12: in the worst case, rcv_ssthresh was cut in half by a scaling_ratio dip. P99 latency jumped from <10ms on 6.1 to ~100ms on 6.12, and almost returned to the 6.1 level with this patch applied.

Quoted 7 lines

Restore the plain WRITE_ONCE() update of window_clamp, as introduced in commit a2cbb1603943 ("tcp: Update window clamping condition"), and keep the rcvq_space.space adjustment. Now rcv_ssthresh is decoupled from scaling_ratio changes in tcp_measure_rcv_mss().

Fixes f5da7c45188e ("tcp: adjust rcvq_space after updating scaling ratio")

Not sure, I mean regression is a regression, but also the previous behavior seems to have just been lucky rather than correct in principle?

Looks like Eric and Neal are AFK, Kuniyuki, Paolo, any opinion on this patch?

Quoted 15 lines
net/ipv4/tcp_input.c +1 -1
index daff93d513428..5b6378b94701e 100644
@@ -252,7 +252,7 @@ static void tcp_measure_rcv_mss(struct sock *sk, const struct sk_buff *skb)
 				struct tcp_sock *tp = tcp_sk(sk);
 
 				val = tcp_win_from_space(sk, sk->sk_rcvbuf);
-				tcp_set_window_clamp(sk, val);
+				WRITE_ONCE(tp->window_clamp, val);
 
 				if (tp->window_clamp < tp->rcvq_space.space)
 					tp->rcvq_space.space = tp->window_clamp;

base-commit: 78f75d632f74b8de0f081a128588f7c37d0d1164

maintainer
#2 lore raw
Quoted 28 lines · Jakub Kicinski
Quoted 27 lines · On Fri, 24 Jul 2026 20:08:06 -0700 Nathan Gao wrote:

Commit f5da7c45188e ("tcp: adjust rcvq_space after updating scaling ratio") replaced the direct window_clamp update in tcp_measure_rcv_mss() with a call to tcp_set_window_clamp(), a helper that implements the TCP_WINDOW_CLAMP setsockopt. As a side effect, the helper also shrinks rcv_ssthresh via __tcp_adjust_rcv_ssthresh().

As a result, each scaling_ratio decrease detected by tcp_measure_rcv_mss() also cuts rcv_ssthresh. Elsewhere in TCP, rcv_ssthresh is usually cut under memory pressure and grows via tcp_grow_window().

Flows whose segment sizes vary keep scaling_ratio oscillating, which leads to an unstable rcv_ssthresh: a dip of rcv_ssthresh only recovers via tcp_grow_window(), keeping the advertised window at a relatively low level even after the ratio itself has recovered, and can even stall the sender.

Observed on a customer's proxy gateway after upgrading from kernel 6.1 to 6.12: in the worst case, rcv_ssthresh was cut in half by a scaling_ratio dip. P99 latency jumped from <10ms on 6.1 to ~100ms on 6.12, and almost returned to the 6.1 level with this patch applied.

Restore the plain WRITE_ONCE() update of window_clamp, as introduced in commit a2cbb1603943 ("tcp: Update window clamping condition"), and keep the rcvq_space.space adjustment. Now rcv_ssthresh is decoupled from scaling_ratio changes in tcp_measure_rcv_mss().

Quoted 8 lines
Quoted 2 lines
Fixes f5da7c45188e ("tcp: adjust rcvq_space after updating scaling ratio")

Not sure, I mean regression is a regression, but also the previous behavior seems to have just been lucky rather than correct in principle?

Looks like Eric and Neal are AFK, Kuniyuki, Paolo, any opinion on this patch?

A quick grep confirm that except for f5da7c45188e, only the control path calls tcp_set_window_clamp(), which IMHO supports this patch rationale. My understanding is also that this patch should not re-introduce the issue addressed by the blamed commit.

It would be great to have a pktdrill tests for at least one of the 2 relevant scenarios (the one described here and the one relevant for f5da7c45188e). My totally uneducated impression is that writing a packet drill for the case described here should be slightly less difficult than the other option, as there is no MTU dependency.

TL;DR: I *think* this patch make sense, pktdrill would be helpful but not a blocker.

/P

maintainer
#3 lore raw
Quoted 42 lines · Paolo Abeni
Quoted 28 lines · Jakub Kicinski
Quoted 27 lines · On Fri, 24 Jul 2026 20:08:06 -0700 Nathan Gao wrote:

Commit f5da7c45188e ("tcp: adjust rcvq_space after updating scaling ratio") replaced the direct window_clamp update in tcp_measure_rcv_mss() with a call to tcp_set_window_clamp(), a helper that implements the TCP_WINDOW_CLAMP setsockopt. As a side effect, the helper also shrinks rcv_ssthresh via __tcp_adjust_rcv_ssthresh().

As a result, each scaling_ratio decrease detected by tcp_measure_rcv_mss() also cuts rcv_ssthresh. Elsewhere in TCP, rcv_ssthresh is usually cut under memory pressure and grows via tcp_grow_window().

Flows whose segment sizes vary keep scaling_ratio oscillating, which leads to an unstable rcv_ssthresh: a dip of rcv_ssthresh only recovers via tcp_grow_window(), keeping the advertised window at a relatively low level even after the ratio itself has recovered, and can even stall the sender.

Observed on a customer's proxy gateway after upgrading from kernel 6.1 to 6.12: in the worst case, rcv_ssthresh was cut in half by a scaling_ratio dip. P99 latency jumped from <10ms on 6.1 to ~100ms on 6.12, and almost returned to the 6.1 level with this patch applied.

Restore the plain WRITE_ONCE() update of window_clamp, as introduced in commit a2cbb1603943 ("tcp: Update window clamping condition"), and keep the rcvq_space.space adjustment. Now rcv_ssthresh is decoupled from scaling_ratio changes in tcp_measure_rcv_mss().

Quoted 8 lines
Quoted 2 lines
Fixes f5da7c45188e ("tcp: adjust rcvq_space after updating scaling ratio")

Not sure, I mean regression is a regression, but also the previous behavior seems to have just been lucky rather than correct in principle?

Looks like Eric and Neal are AFK, Kuniyuki, Paolo, any opinion on this patch?

A quick grep confirm that except for f5da7c45188e, only the control path calls tcp_set_window_clamp(), which IMHO supports this patch rationale. My understanding is also that this patch should not re-introduce the issue addressed by the blamed commit.

Quoted 8 lines

It would be great to have a pktdrill tests for at least one of the 2 relevant scenarios (the one described here and the one relevant for f5da7c45188e). My totally uneducated impression is that writing a packet drill for the case described here should be slightly less difficult than the other option, as there is no MTU dependency.

TL;DR: I *think* this patch make sense, pktdrill would be helpful but not a blocker.

Uhm... Above I did not take in account how far we are in the current release cycle. The issue has been unnoticed for a considerable amount of time, and the chances the fix would introduce some other regressions are not 0, so I think this patch would deserve at least another positive review to be merged now.

Thanks,

Paolo

#4 lore raw
In reply to Paolo Abeni
Quoted 36 lines · Paolo Abeni
Quoted 34 lines · Jakub Kicinski
Quoted 22 lines · On Fri, 24 Jul 2026 20:08:06 -0700 Nathan Gao wrote:

Commit f5da7c45188e ("tcp: adjust rcvq_space after updating scaling ratio") replaced the direct window_clamp update in tcp_measure_rcv_mss() with a call to tcp_set_window_clamp(), a helper that implements the TCP_WINDOW_CLAMP setsockopt. As a side effect, the helper also shrinks rcv_ssthresh via __tcp_adjust_rcv_ssthresh().

As a result, each scaling_ratio decrease detected by tcp_measure_rcv_mss() also cuts rcv_ssthresh. Elsewhere in TCP, rcv_ssthresh is usually cut under memory pressure and grows via tcp_grow_window().

Flows whose segment sizes vary keep scaling_ratio oscillating, which leads to an unstable rcv_ssthresh: a dip of rcv_ssthresh only recovers via tcp_grow_window(), keeping the advertised window at a relatively low level even after the ratio itself has recovered, and can even stall the sender.

Observed on a customer's proxy gateway after upgrading from kernel 6.1 to 6.12: in the worst case, rcv_ssthresh was cut in half by a scaling_ratio dip. P99 latency jumped from <10ms on 6.1 to ~100ms on 6.12, and almost returned to the 6.1 level with this patch applied.

Quoted 7 lines

Restore the plain WRITE_ONCE() update of window_clamp, as introduced in commit a2cbb1603943 ("tcp: Update window clamping condition"), and keep the rcvq_space.space adjustment. Now rcv_ssthresh is decoupled from scaling_ratio changes in tcp_measure_rcv_mss().

Fixes f5da7c45188e ("tcp: adjust rcvq_space after updating scaling ratio")

Not sure, I mean regression is a regression, but also the previous behavior seems to have just been lucky rather than correct in principle?

Quoted 6 lines
Quoted 2 lines

Looks like Eric and Neal are AFK, Kuniyuki, Paolo, any opinion on this patch?

A quick grep confirm that except for f5da7c45188e, only the control path calls tcp_set_window_clamp(), which IMHO supports this patch rationale. My understanding is also that this patch should not re-introduce the issue addressed by the blamed commit.

+1, and

Quoted 2 lines

It would be great to have a pktdrill tests for at least one of the 2

+1. I also told Nathan offlist that it would be nicer to write a packetdrill test.

Quoted 10 lines

relevant scenarios (the one described here and the one relevant for f5da7c45188e). My totally uneducated impression is that writing a packet drill for the case described here should be slightly less difficult than the other option, as there is no MTU dependency.

TL;DR: I *think* this patch make sense, pktdrill would be helpful but not a blocker.

/P

#5 lore raw
In reply to Nathan Gao
This patch was applied to netdev/net.git (main) by Jakub Kicinski <[email protected]>:

Hello:

This patch was applied to netdev/net.git (main)
by Jakub Kicinski <[email protected]>:

Quoted 6 lines · On Fri, 24 Jul 2026 20:08:06 -0700 you wrote:

Commit f5da7c45188e ("tcp: adjust rcvq_space after updating scaling ratio") replaced the direct window_clamp update in tcp_measure_rcv_mss() with a call to tcp_set_window_clamp(), a helper that implements the TCP_WINDOW_CLAMP setsockopt. As a side effect, the helper also shrinks rcv_ssthresh via __tcp_adjust_rcv_ssthresh().

Quoted 6 lines

As a result, each scaling_ratio decrease detected by tcp_measure_rcv_mss() also cuts rcv_ssthresh. Elsewhere in TCP, rcv_ssthresh is usually cut under memory pressure and grows via tcp_grow_window().

[...]

Here is the summary with links:
- [net] tcp: do not change rcv_ssthresh in tcp_measure_rcv_mss()

    https://git.kernel.org/netdev/net/c/0e125ecfe20c

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html