Re: kern/58859 (KASSERT in wg_task_retry_handshake)

[riastradh%NetBSD.org@localhost]

Synopsis: KASSERT in wg_task_retry_handshake

Responsible-Changed-From-To: kern-bug-people->riastradh
Responsible-Changed-By: riastradh%NetBSD.org@localhost
Responsible-Changed-When: Thu, 28 Nov 2024 13:10:42 +0000
Responsible-Changed-Why:
Interesting, I haven't seen this.  Can you reproduce it with
sysctl -w net.wg.debug=3
and share dmesg output when it happens again?

It looks like the condition is harmless and maybe just requires the
following patch to avoid tripping the assertion, but I'd like to see
the debug output if possible to confirm (since I haven't seen this in
a much longer time than a few hours of regular activity):

diff -r 902158b0ae9d sys/net/if_wg.c
--- a/sys/net/if_wg.c	Sun Nov 24 14:53:53 2024 +0000
+++ b/sys/net/if_wg.c	Thu Nov 28 13:10:12 2024 +0000
@@ -3386,12 +3386,13 @@ wg_task_retry_handshake(struct wg_softc 
 	WG_TRACE("WGP_TASK_RETRY_HANDSHAKE");
 
 	KASSERT(mutex_owned(wgp->wgp_lock));
-	KASSERT(wgp->wgp_handshake_start_time != 0);
 
 	wgs = wgp->wgp_session_unstable;
 	if (wgs->wgs_state != WGS_STATE_INIT_ACTIVE)
 		return;
 
+	KASSERT(wgp->wgp_handshake_start_time != 0);
+
 	/*
 	 * XXX no real need to assign a new index here, but we do need
 	 * to transition to UNKNOWN temporarily


State-Changed-From-To: open->feedback
State-Changed-By: riastradh%NetBSD.org@localhost
State-Changed-When: Thu, 28 Nov 2024 13:10:42 +0000
State-Changed-Why:
mine