Re: [PATCH v11 net-next 10/15] tcp: accecn: unset ECT if receive or send ACE=0 in AccECN negotiaion
From: Jakub Kicinski <kuba@kernel.org>
Date: 2026-01-27 20:09:59
Also in:
bpf, linux-doc, linux-kselftest
On Fri, 23 Jan 2026 11:07:16 +0100 chia-yu.chang@nokia-bell-labs.com wrote:
Based on specification: https://tools.ietf.org/id/draft-ietf-tcpm-accurate-ecn-28.txt Based on Section 3.1.5 of AccECN spec (RFC9768), a TCP Server in AccECN mode MUST NOT set ECT on any packet for the rest of the connection, if it has received or sent at least one valid SYN or Acceptable SYN/ACK with (AE,CWR,ECE) = (0,0,0) during the handshake. In addition, a host in AccECN mode that is feeding back the IP-ECN field on a SYN or SYN/ACK MUST feed back the IP-ECN field on the latest valid SYN or acceptable SYN/ACK to arrive.
FWIW this series is marked as Changes Requested in patchwork. I don't think I did that, IDK why it's in that state :S I see these warnings when I try to apply it: Applying: selftests/net: packetdrill: add TCP Accurate ECN cases .git/rebase-apply/patch:325: new blank line at EOF. + .git/rebase-apply/patch:1550: new blank line at EOF. + .git/rebase-apply/patch:1641: new blank line at EOF. + warning: 3 lines add whitespace errors. Other than that the AI code review made the following comment, I haven't investigated the expected behavior. Could you TAL?
quoted hunk ↗ jump to hunk
@@ -1103,6 +1104,8 @@ static void reqsk_timer_handler(struct timer_list *t) (!resend || !tcp_rtx_synack(sk_listener, req) || inet_rsk(req)->acked)) { + if (req->num_retrans > 1 && tcp_rsk(req)->accecn_ok) + tcp_rsk(req)->accecn_fail_mode |= TCP_ACCECN_ACE_FAIL_SEND;
Note: This correctly sets the flag on tcp_rsk(req)->accecn_fail_mode, which will be propagated to the child socket via tcp_ecn_openreq_child().
quoted hunk ↗ jump to hunk
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
[ ... ]
quoted hunk ↗ jump to hunk
@@ -481,6 +481,8 @@ static void tcp_ecn_openreq_child(struct sock *sk, tp->syn_ect_snt = treq->syn_ect_snt; tcp_accecn_third_ack(sk, skb, treq->syn_ect_snt); tp->saw_accecn_opt = treq->saw_accecn_opt; + if (treq->accecn_fail_mode & TCP_ACCECN_ACE_FAIL_SEND) + tcp_accecn_fail_mode_set(tp, TCP_ACCECN_ACE_FAIL_SEND);
This propagates TCP_ACCECN_ACE_FAIL_SEND from the request socket to the child, but TCP_ACCECN_ACE_FAIL_RECV is not propagated here.
quoted hunk ↗ jump to hunk
@@ -749,16 +751,32 @@ struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, */ if (!tcp_oow_rate_limited(sock_net(sk), skb, LINUX_MIB_TCPACKSKIPPEDSYNRECV, - &tcp_rsk(req)->last_oow_ack_time) && - - !tcp_rtx_synack(sk, req)) { - unsigned long expires = jiffies; - - expires += tcp_reqsk_timeout(req); - if (!fastopen) - mod_timer_pending(&req->rsk_timer, expires); - else - req->rsk_timer.expires = expires; + &tcp_rsk(req)->last_oow_ack_time)) { + if (tcp_rsk(req)->accecn_ok) { + u8 ect_rcv = TCP_SKB_CB(skb)->ip_dsfield & + INET_ECN_MASK; + + tcp_rsk(req)->syn_ect_rcv = ect_rcv; + if (tcp_accecn_ace(tcp_hdr(skb)) == 0x0) { + u8 fail_mode = TCP_ACCECN_ACE_FAIL_RECV; + + tcp_accecn_fail_mode_set(tcp_sk(sk), + fail_mode);
Should this be setting the flag on the request socket instead of the
listener socket? Here tcp_sk(sk) is the listener, but reqsk_timer_handler()
uses tcp_rsk(req)->accecn_fail_mode for the same purpose. The flag set
on the listener socket will not be propagated to the child socket, so
the child may incorrectly set ECT on outgoing packets even after receiving
a SYN with ACE=0.
For consistency with reqsk_timer_handler() and to ensure the flag reaches
the child socket, this might need to be:
tcp_rsk(req)->accecn_fail_mode |= fail_mode;
+ }
+ }
+ if (!tcp_rtx_synack(sk, req)) {
+ unsigned long expires = jiffies;
+
+ tcp_accecn_ace_fail_send_set_retrans(req,
+ tcp_sk(sk));Similar question here: tcp_accecn_ace_fail_send_set_retrans() sets the flag on tcp_sk(sk) (the listener), while reqsk_timer_handler() sets it on tcp_rsk(req). Is this intentional?