From: Alexandr Nedvedicky <sashan@fastmail.net>
Subject: Re: link mbufs/inpcbs to pf_states, not pf_state_keys
To: David Gwynne <david@gwynne.id.au>
Cc: tech@openbsd.org
Date: Fri, 30 Jan 2026 00:36:55 +0100

Hello,
On Thu, Jan 29, 2026 at 01:50:19AM +0100, Alexandr Nedvedicky wrote:
> Hello,
> 
> On Thu, Jan 29, 2026 at 06:40:01AM +1000, David Gwynne wrote:
> </snip>
> > 
> > this chunk makes more sense if you can see more context.
> > 
> >   1296	int
> >   1297	pf_state_insert(struct pfi_kif *kif, struct pf_state_key **skwp,
> >   1298	    struct pf_state_key **sksp, struct pf_state *st)
> >   1299	{
> >   1300		struct pf_state_key *skw = *skwp;
> >   1301		struct pf_state_key *sks = *sksp;
> >   1302		int same = (skw == sks);
> >   1303	
> >   1304		PF_ASSERT_LOCKED();
> >   1305	
> >   1306		st->kif = kif;
> >   1307		PF_STATE_ENTER_WRITE();
> >   1308	
> >   1309		skw = pf_state_key_attach(skw, st, PF_SK_WIRE);
> >   1310		if (skw == NULL) {
> >   1311			pf_state_key_unref(sks);
> >   1312			PF_STATE_EXIT_WRITE();
> >   1313			return (-1);
> >   1314		}
> >   1315	
> >   1316		if (same) {
> >   1317			/* pf_state_key_attach might have swapped skw */
> >   1318			if (skw != sks) {
> >   1319				pf_state_key_unref(sks);
> >   1320				sks = pf_state_key_ref(skw);

    may be we should do
	     *sksp = sks;
    here if when running through this branch, so things are consistent
    for caller.

thanks and
regards
sashan