[PATCH v1 net] tcp: Disable usec TS for SYN Cookie.

[PATCH v1 net] tcp: Disable usec TS for SYN Cookie.

[Kuniyuki Iwashima]

cookie_tcp_reqsk_alloc() sets tcp_rsk(req)->req_usec_ts to false
unconditionally.

If want_cookie is true in tcp_conn_request(), we should not set
tcp_rsk(req)->req_usec_ts.

Let's not call dst_tcp_usec_ts() for SYN Cookie.

Fixes: 614e8316aa4c ("tcp: add support for usec resolution in TCP TS values")
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
---
 net/ipv4/syncookies.c | 3 ---
 net/ipv4/tcp_input.c  | 3 ++-
 2 files changed, 2 insertions(+), 4 deletions(-)

diff --git a/net/ipv4/syncookies.c b/net/ipv4/syncookies.c
index b5f0a65c6786..f5cd9e325d01 100644
--- a/net/ipv4/syncookies.c
+++ b/net/ipv4/syncookies.c
@@ -76,12 +76,9 @@ u64 cookie_init_timestamp(struct request_sock *req, u64 now)
 	if (ts > ts_now)
 		ts -= (1UL << TSBITS);
 
-	if (tcp_rsk(req)->req_usec_ts)
-		return ts * NSEC_PER_USEC;
 	return ts * NSEC_PER_MSEC;
 }
 
-
 static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
 				   __be16 dport, __u32 sseq, __u32 data)
 {
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index cba89733d121..8bf202b95c68 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -7720,7 +7720,8 @@ int tcp_conn_request(struct request_sock_ops *rsk_ops,
 		st = af_ops->init_seq_and_ts_off(net, skb);
 
 	if (tmp_opt.tstamp_ok) {
-		tcp_rsk(req)->req_usec_ts = dst_tcp_usec_ts(dst);
+		if (!want_cookie)
+			tcp_rsk(req)->req_usec_ts = dst_tcp_usec_ts(dst);
 		tcp_rsk(req)->ts_off = st.ts_off;
 	}
 	if (!want_cookie && !isn) {
-- 
2.54.0.rc1.513.gad8abe7a5a-goog

Re: [PATCH v1 net] tcp: Disable usec TS for SYN Cookie.

[Eric Dumazet]

On Fri, Apr 17, 2026 at 7:50 PM Kuniyuki Iwashima <kuniyu@google.com> wrote:
>
> cookie_tcp_reqsk_alloc() sets tcp_rsk(req)->req_usec_ts to false
> unconditionally.
>
> If want_cookie is true in tcp_conn_request(), we should not set
> tcp_rsk(req)->req_usec_ts.
>
> Let's not call dst_tcp_usec_ts() for SYN Cookie.

May I ask why ?

TCP usec TS are based on routing. this feature is not part of SYN
and/or SYNACK options.

Both side must have:

ip route ... feature tcp_usec_ts

syncookies are orthogonal to this constraint, so TCP flows can use
usec TS just fine.

pw-bot: cr

Re: [PATCH v1 net] tcp: Disable usec TS for SYN Cookie.

[Kuniyuki Iwashima]

On Fri, Apr 17, 2026 at 9:59 PM Eric Dumazet <edumazet@google.com> wrote:
>
> On Fri, Apr 17, 2026 at 7:50 PM Kuniyuki Iwashima <kuniyu@google.com> wrote:
> >
> > cookie_tcp_reqsk_alloc() sets tcp_rsk(req)->req_usec_ts to false
> > unconditionally.
> >
> > If want_cookie is true in tcp_conn_request(), we should not set
> > tcp_rsk(req)->req_usec_ts.
> >
> > Let's not call dst_tcp_usec_ts() for SYN Cookie.
>
> May I ask why ?

Sorry, I missed tcp_skb_timestamp_ts() properly restores the
cookie TS generated by cookie_init_timestamp() to ms unit.

Still we don't need to call dst_tcp_usec_ts() for SYN cookie,
but this was more like a cleanup patch.


>
> TCP usec TS are based on routing. this feature is not part of SYN
> and/or SYNACK options.
>
> Both side must have:
>
> ip route ... feature tcp_usec_ts
>
> syncookies are orthogonal to this constraint, so TCP flows can use
> usec TS just fine.
>
> pw-bot: cr

Re: [PATCH v1 net] tcp: Disable usec TS for SYN Cookie.

[Eric Dumazet]

On Fri, Apr 17, 2026 at 10:33 PM Kuniyuki Iwashima <kuniyu@google.com> wrote:
>
> On Fri, Apr 17, 2026 at 9:59 PM Eric Dumazet <edumazet@google.com> wrote:
> >
> > On Fri, Apr 17, 2026 at 7:50 PM Kuniyuki Iwashima <kuniyu@google.com> wrote:
> > >
> > > cookie_tcp_reqsk_alloc() sets tcp_rsk(req)->req_usec_ts to false
> > > unconditionally.
> > >
> > > If want_cookie is true in tcp_conn_request(), we should not set
> > > tcp_rsk(req)->req_usec_ts.
> > >
> > > Let's not call dst_tcp_usec_ts() for SYN Cookie.
> >
> > May I ask why ?
>
> Sorry, I missed tcp_skb_timestamp_ts() properly restores the
> cookie TS generated by cookie_init_timestamp() to ms unit.
>
> Still we don't need to call dst_tcp_usec_ts() for SYN cookie,
> but this was more like a cleanup patch.

Okay, but consider the standard path (or fast path) has to call it.
dst_tcp_usec_ts() is a mere dst_feature(dst,
RTAX_FEATURE_TCP_USEC_TS), which is pretty fast.
Adding a conditional branch won't help.