From: Alexander Bluhm <bluhm@openbsd.org>
Subject: Re: link mbufs/inpcbs to pf_states, not pf_state_keys
To: David Gwynne <david@gwynne.id.au>
Cc: Alexandr Nedvedicky <sashan@fastmail.net>, tech@openbsd.org
Date: Wed, 21 May 2025 18:36:27 +0900

On Sat, May 10, 2025 at 10:28:56AM +1000, David Gwynne wrote:
> tl;dr: less pointer chasing, and more precise links between things
> representing connections in the kernel.

regress/sys/net/pf_forward and regress/sys/net/pflog fail.

I am not 100% sure that it is related to this diff.  Due to other
bugs, my last successful regress run was at May 9th.  Give me another
day, then I can say if these tests are broken in current already.

bluhm

> Index: kern/uipc_mbuf.c
> ===================================================================
> RCS file: /cvs/src/sys/kern/uipc_mbuf.c,v
> diff -u -p -r1.296 uipc_mbuf.c
> --- kern/uipc_mbuf.c	1 Jan 2025 13:44:22 -0000	1.296
> +++ kern/uipc_mbuf.c	9 May 2025 02:48:26 -0000
> @@ -301,7 +301,7 @@ m_clearhdr(struct mbuf *m)
>  	/* delete all mbuf tags to reset the state */
>  	m_tag_delete_chain(m);
>  #if NPF > 0
> -	pf_mbuf_unlink_state_key(m);
> +	pf_mbuf_unlink_state(m);
>  	pf_mbuf_unlink_inpcb(m);
>  #endif	/* NPF > 0 */
>  
> @@ -429,7 +429,7 @@ m_free(struct mbuf *m)
>  	if (m->m_flags & M_PKTHDR) {
>  		m_tag_delete_chain(m);
>  #if NPF > 0
> -		pf_mbuf_unlink_state_key(m);
> +		pf_mbuf_unlink_state(m);
>  		pf_mbuf_unlink_inpcb(m);
>  #endif	/* NPF > 0 */
>  	}
> @@ -1387,8 +1387,8 @@ m_dup_pkthdr(struct mbuf *to, struct mbu
>  	to->m_pkthdr = from->m_pkthdr;
>  
>  #if NPF > 0
> -	to->m_pkthdr.pf.statekey = NULL;
> -	pf_mbuf_link_state_key(to, from->m_pkthdr.pf.statekey);
> +	to->m_pkthdr.pf.st = NULL;
> +	pf_mbuf_link_state(to, from->m_pkthdr.pf.st);
>  	to->m_pkthdr.pf.inp = NULL;
>  	pf_mbuf_link_inpcb(to, from->m_pkthdr.pf.inp);
>  #endif	/* NPF > 0 */
> @@ -1517,8 +1517,8 @@ m_print(void *v,
>  		    m->m_pkthdr.csum_flags, MCS_BITS);
>  		(*pr)("m_pkthdr.ether_vtag: %u\tm_ptkhdr.ph_rtableid: %u\n",
>  		    m->m_pkthdr.ether_vtag, m->m_pkthdr.ph_rtableid);
> -		(*pr)("m_pkthdr.pf.statekey: %p\tm_pkthdr.pf.inp %p\n",
> -		    m->m_pkthdr.pf.statekey, m->m_pkthdr.pf.inp);
> +		(*pr)("m_pkthdr.pf.st: %p\tm_pkthdr.pf.inp %p\n",
> +		    m->m_pkthdr.pf.st, m->m_pkthdr.pf.inp);
>  		(*pr)("m_pkthdr.pf.qid: %u\tm_pkthdr.pf.tag: %u\n",
>  		    m->m_pkthdr.pf.qid, m->m_pkthdr.pf.tag);
>  		(*pr)("m_pkthdr.pf.flags: %b\n",
> Index: net/if_mpw.c
> ===================================================================
> RCS file: /cvs/src/sys/net/if_mpw.c,v
> diff -u -p -r1.67 if_mpw.c
> --- net/if_mpw.c	2 Mar 2025 21:28:32 -0000	1.67
> +++ net/if_mpw.c	9 May 2025 02:48:26 -0000
> @@ -612,7 +612,7 @@ mpw_input(struct mpw_softc *sc, struct m
>  	m->m_pkthdr.ph_rtableid = ifp->if_rdomain;
>  
>  	/* packet has not been processed by PF yet. */
> -	KASSERT(m->m_pkthdr.pf.statekey == NULL);
> +	KASSERT(m->m_pkthdr.pf.st == NULL);
>  
>  	if_vinput(ifp, m, NULL);
>  	return;
> Index: net/if_tpmr.c
> ===================================================================
> RCS file: /cvs/src/sys/net/if_tpmr.c,v
> diff -u -p -r1.36 if_tpmr.c
> --- net/if_tpmr.c	2 Mar 2025 21:28:32 -0000	1.36
> +++ net/if_tpmr.c	9 May 2025 02:48:26 -0000
> @@ -304,7 +304,7 @@ tpmr_pf(struct ifnet *ifp0, int dir, str
>  		return (NULL);
>  
>  	if (dir == PF_IN && ISSET(m->m_pkthdr.pf.flags, PF_TAG_DIVERTED)) {
> -		pf_mbuf_unlink_state_key(m);
> +		pf_mbuf_unlink_state(m);
>  		pf_mbuf_unlink_inpcb(m);
>  		(*fam->ip_input)(ifp0, m, ns);
>  		return (NULL);
> Index: net/if_veb.c
> ===================================================================
> RCS file: /cvs/src/sys/net/if_veb.c,v
> diff -u -p -r1.37 if_veb.c
> --- net/if_veb.c	2 Mar 2025 21:28:32 -0000	1.37
> +++ net/if_veb.c	9 May 2025 02:48:26 -0000
> @@ -654,7 +654,7 @@ veb_pf(struct ifnet *ifp0, int dir, stru
>  		return (NULL);
>  
>  	if (dir == PF_IN && ISSET(m->m_pkthdr.pf.flags, PF_TAG_DIVERTED)) {
> -		pf_mbuf_unlink_state_key(m);
> +		pf_mbuf_unlink_state(m);
>  		pf_mbuf_unlink_inpcb(m);
>  		(*fam->ip_input)(ifp0, m, ns);
>  		return (NULL);
> Index: net/pf.c
> ===================================================================
> RCS file: /cvs/src/sys/net/pf.c,v
> diff -u -p -r1.1210 pf.c
> --- net/pf.c	1 May 2025 01:10:08 -0000	1.1210
> +++ net/pf.c	9 May 2025 02:48:26 -0000
> @@ -247,15 +247,16 @@ int			 pf_state_insert(struct pfi_kif *,
>  			    struct pf_state_key **, struct pf_state_key **,
>  			    struct pf_state *);
>  
> +int			 pf_state_isvalid(struct pf_state *);
>  int			 pf_state_key_isvalid(struct pf_state_key *);
>  struct pf_state_key	*pf_state_key_ref(struct pf_state_key *);
>  void			 pf_state_key_unref(struct pf_state_key *);
> -void			 pf_state_key_link_reverse(struct pf_state_key *,
> -			    struct pf_state_key *);
> -void			 pf_state_key_unlink_reverse(struct pf_state_key *);
> -void			 pf_state_key_link_inpcb(struct pf_state_key *,
> +void			 pf_state_link_reverse(struct pf_state *,
> +			    struct pf_state *);
> +void			 pf_state_unlink_reverse(struct pf_state *);
> +void			 pf_state_link_inpcb(struct pf_state *,
>  			    struct inpcb *);
> -void			 pf_state_key_unlink_inpcb(struct pf_state_key *);
> +void			 pf_state_unlink_inpcb(struct pf_state *);
>  void			 pf_pktenqueue_delayed(void *);
>  int32_t			 pf_state_expires(const struct pf_state *, uint8_t);
>  
> @@ -860,8 +861,6 @@ pf_state_key_detach(struct pf_state *st,
>  	if (TAILQ_EMPTY(&sk->sk_states)) {
>  		RBT_REMOVE(pf_state_tree, &pf_statetbl, sk);
>  		sk->sk_removed = 1;
> -		pf_state_key_unlink_reverse(sk);
> -		pf_state_key_unlink_inpcb(sk);
>  		pf_state_key_unref(sk);
>  	}
>  
> @@ -1123,13 +1122,40 @@ pf_compare_state_keys(struct pf_state_ke
>  	}
>  }
>  
> +static inline struct pf_state *
> +pf_find_state_lookup(struct pf_pdesc *pd, const struct pf_state_key_cmp *key)
> +{
> +	struct pf_state_key	*sk;
> +	struct pf_state_item	*si;
> +	struct pf_state		*st;
> +	uint8_t			 dir = pd->dir;
> +
> +	sk = RBT_FIND(pf_state_tree, &pf_statetbl, (struct pf_state_key *)key);
> +	if (sk == NULL)
> +		return (NULL);
> +
> +	/* list is sorted, if-bound states before floating ones */
> +	TAILQ_FOREACH(si, &sk->sk_states, si_entry) {
> +		st = si->si_st;
> +		if (st->timeout == PFTM_PURGE)
> +			continue;
> +		if (st->kif != pfi_all && st->kif != pd->kif)
> +			continue;
> +
> +		if (st->key[dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK] == sk)
> +			return (st);
> +	}
> +
> +	return (NULL);
> +}
> +
>  int
>  pf_find_state(struct pf_pdesc *pd, struct pf_state_key_cmp *key,
>      struct pf_state **stp)
>  {
> -	struct pf_state_key	*sk, *pkt_sk;
> -	struct pf_state_item	*si;
>  	struct pf_state		*st = NULL;
> +	struct pf_state		*strev = NULL;
> +	struct inpcb		*inp = NULL;
>  
>  	pf_status.fcounters[FCNT_STATE_SEARCH]++;
>  	if (pf_status.debug >= LOG_DEBUG) {
> @@ -1139,82 +1165,60 @@ pf_find_state(struct pf_pdesc *pd, struc
>  		addlog("\n");
>  	}
>  
> -	pkt_sk = NULL;
> -	sk = NULL;
>  	if (pd->dir == PF_OUT) {
> -		/* first if block deals with outbound forwarded packet */
> -		pkt_sk = pd->m->m_pkthdr.pf.statekey;
> +		strev = pd->m->m_pkthdr.pf.st;
> +		inp = pd->m->m_pkthdr.pf.inp;
>  
> -		if (!pf_state_key_isvalid(pkt_sk)) {
> -			pf_mbuf_unlink_state_key(pd->m);
> -			pkt_sk = NULL;
> -		}
> +		/* first if block deals with outbound forwarded packet */
> +		if (strev != NULL) {
> +			pd->m->m_pkthdr.pf.st = NULL;
> +			KASSERT(inp == NULL);
>  
> -		if (pkt_sk && pf_state_key_isvalid(pkt_sk->sk_reverse))
> -			sk = pkt_sk->sk_reverse;
> +			if (pf_state_isvalid(strev)) {
> +				st = strev->reverse;
> +				if (st != NULL && pf_state_isvalid(st))
> +					goto match;
> +			}
>  
> -		if (pkt_sk == NULL) {
> -			struct inpcb *inp = pd->m->m_pkthdr.pf.inp;
> +			/* this handles st not being valid too */
> +			pf_state_unlink_reverse(strev);
>  
> +		} else if (inp != NULL && READ_ONCE(inp->inp_pf_st) != NULL) {
>  			/* here we deal with local outbound packet */
> -			if (inp != NULL) {
> -				struct pf_state_key	*inp_sk;
> -
> -				mtx_enter(&pf_inp_mtx);
> -				inp_sk = inp->inp_pf_sk;
> -				if (pf_state_key_isvalid(inp_sk)) {
> -					sk = inp_sk;
> +			mtx_enter(&pf_inp_mtx);
> +			st = inp->inp_pf_st;
> +			if (st != NULL) {
> +				if (pf_state_isvalid(st)) {
>  					mtx_leave(&pf_inp_mtx);
> -				} else if (inp_sk != NULL) {
> -					KASSERT(inp_sk->sk_inp == inp);
> -					inp_sk->sk_inp = NULL;
> -					inp->inp_pf_sk = NULL;
> +					goto match;
> +				} else {
> +					KASSERT(st->inp == inp);
> +					st->inp = NULL;
> +					inp->inp_pf_st = NULL;
>  					mtx_leave(&pf_inp_mtx);
>  
> -					pf_state_key_unref(inp_sk);
> +					pf_state_unref(st);
>  					in_pcbunref(inp);
> -				} else
> -					mtx_leave(&pf_inp_mtx);
> -			}
> -		}
> -	}
> -
> -	if (sk == NULL) {
> -		if ((sk = RBT_FIND(pf_state_tree, &pf_statetbl,
> -		    (struct pf_state_key *)key)) == NULL)
> -			return (PF_DROP);
> -		if (pd->dir == PF_OUT && pkt_sk &&
> -		    pf_compare_state_keys(pkt_sk, sk, pd->kif, pd->dir) == 0)
> -			pf_state_key_link_reverse(sk, pkt_sk);
> -		else if (pd->dir == PF_OUT)
> -			pf_state_key_link_inpcb(sk, pd->m->m_pkthdr.pf.inp);
> -	}
> -
> -	/* remove firewall data from outbound packet */
> -	if (pd->dir == PF_OUT)
> -		pf_pkt_addr_changed(pd->m);
> -
> -	/* list is sorted, if-bound states before floating ones */
> -	TAILQ_FOREACH(si, &sk->sk_states, si_entry) {
> -		struct pf_state *sist = si->si_st;
> -		if (sist->timeout != PFTM_PURGE &&
> -		    (sist->kif == pfi_all || sist->kif == pd->kif) &&
> -		    ((sist->key[PF_SK_WIRE]->af == sist->key[PF_SK_STACK]->af &&
> -		      sk == (pd->dir == PF_IN ? sist->key[PF_SK_WIRE] :
> -		    sist->key[PF_SK_STACK])) ||
> -		    (sist->key[PF_SK_WIRE]->af != sist->key[PF_SK_STACK]->af
> -		    && pd->dir == PF_IN && (sk == sist->key[PF_SK_STACK] ||
> -		    sk == sist->key[PF_SK_WIRE])))) {
> -			st = sist;
> -			break;
> +				}
> +			} else
> +				mtx_leave(&pf_inp_mtx);
>  		}
>  	}
>  
> +	st = pf_find_state_lookup(pd, key);
>  	if (st == NULL)
>  		return (PF_DROP);
>  	if (ISSET(st->state_flags, PFSTATE_INP_UNLINKED))
>  		return (PF_DROP);
>  
> +	if (pd->dir == PF_OUT) {
> +		if (strev != NULL)
> +			pf_state_link_reverse(st, strev);
> +		else if (inp != NULL)
> +			pf_state_link_inpcb(st, inp);
> +	}
> +
> +match:
>  	if (st->rule.ptr->pktrate.limit && pd->dir == st->direction) {
>  		pf_add_threshold(&st->rule.ptr->pktrate);
>  		if (pf_check_threshold(&st->rule.ptr->pktrate))
> @@ -1782,6 +1786,9 @@ pf_remove_state(struct pf_state *st)
>  	st->timeout = PFTM_UNLINKED;
>  	mtx_leave(&st->mtx);
>  
> +	pf_state_unlink_reverse(st);
> +	pf_state_unlink_inpcb(st);
> +
>  	/* handle load balancing related tasks */
>  	pf_postprocess_addr(st);
>  
> @@ -1813,52 +1820,47 @@ pf_remove_state(struct pf_state *st)
>  void
>  pf_remove_divert_state(struct inpcb *inp)
>  {
> -	struct pf_state_key	*sk;
> -	struct pf_state_item	*si;
> +	struct pf_state		*st;
>  
>  	PF_ASSERT_UNLOCKED();
>  
> -	if (READ_ONCE(inp->inp_pf_sk) == NULL)
> +	if (READ_ONCE(inp->inp_pf_st) == NULL)
>  		return;
>  
>  	mtx_enter(&pf_inp_mtx);
> -	sk = pf_state_key_ref(inp->inp_pf_sk);
> +	st = pf_state_ref(inp->inp_pf_st);
>  	mtx_leave(&pf_inp_mtx);
> -	if (sk == NULL)
> +	if (st == NULL)
>  		return;
>  
>  	PF_LOCK();
>  	PF_STATE_ENTER_WRITE();
> -	TAILQ_FOREACH(si, &sk->sk_states, si_entry) {
> -		struct pf_state *sist = si->si_st;
> -		if (sk == sist->key[PF_SK_STACK] && sist->rule.ptr &&
> -		    (sist->rule.ptr->divert.type == PF_DIVERT_TO ||
> -		     sist->rule.ptr->divert.type == PF_DIVERT_REPLY)) {
> -			if (sist->key[PF_SK_STACK]->proto == IPPROTO_TCP &&
> -			    sist->key[PF_SK_WIRE] != sist->key[PF_SK_STACK]) {
> -				/*
> -				 * If the local address is translated, keep
> -				 * the state for "tcp.closed" seconds to
> -				 * prevent its source port from being reused.
> -				 */
> -				if (sist->src.state < TCPS_FIN_WAIT_2 ||
> -				    sist->dst.state < TCPS_FIN_WAIT_2) {
> -					pf_set_protostate(sist, PF_PEER_BOTH,
> -					    TCPS_TIME_WAIT);
> -					pf_update_state_timeout(sist,
> -					    PFTM_TCP_CLOSED);
> -					sist->expire = getuptime();
> -				}
> -				sist->state_flags |= PFSTATE_INP_UNLINKED;
> -			} else
> -				pf_remove_state(sist);
> -			break;
> -		}
> +	if (st->rule.ptr &&
> +	    (st->rule.ptr->divert.type == PF_DIVERT_TO ||
> +	     st->rule.ptr->divert.type == PF_DIVERT_REPLY)) {
> +		if (st->key[PF_SK_STACK]->proto == IPPROTO_TCP &&
> +		    st->key[PF_SK_WIRE] != st->key[PF_SK_STACK]) {
> +			/*
> +			 * If the local address is translated, keep
> +			 * the state for "tcp.closed" seconds to
> +			 * prevent its source port from being reused.
> +			 */
> +			if (st->src.state < TCPS_FIN_WAIT_2 ||
> +			    st->dst.state < TCPS_FIN_WAIT_2) {
> +				pf_set_protostate(st, PF_PEER_BOTH,
> +				    TCPS_TIME_WAIT);
> +				pf_update_state_timeout(st,
> +				    PFTM_TCP_CLOSED);
> +				st->expire = getuptime();
> +			}
> +			st->state_flags |= PFSTATE_INP_UNLINKED;
> +		} else
> +			pf_remove_state(st);
>  	}
>  	PF_STATE_EXIT_WRITE();
>  	PF_UNLOCK();
>  
> -	pf_state_key_unref(sk);
> +	pf_state_unref(st);
>  }
>  
>  void
> @@ -7916,9 +7918,9 @@ done:
>  
>  		if (pd.dir == PF_IN) {
>  			KASSERT(inp == NULL);
> -			pf_mbuf_link_state_key(m, st->key[PF_SK_STACK]);
> +			pf_mbuf_link_state(m, st);
>  		} else if (pd.dir == PF_OUT)
> -			pf_state_key_link_inpcb(st->key[PF_SK_STACK], inp);
> +			pf_state_link_inpcb(st, inp);
>  
>  		if (!ISSET(m->m_pkthdr.csum_flags, M_FLOWID)) {
>  			m->m_pkthdr.ph_flowid = st->key[PF_SK_WIRE]->hash;
> @@ -8100,14 +8102,14 @@ done:
>  int
>  pf_ouraddr(struct mbuf *m)
>  {
> -	struct pf_state_key	*sk;
> +	struct pf_state		*st;
>  
>  	if (m->m_pkthdr.pf.flags & PF_TAG_DIVERTED)
>  		return (1);
>  
> -	sk = m->m_pkthdr.pf.statekey;
> -	if (sk != NULL) {
> -		if (READ_ONCE(sk->sk_inp) != NULL)
> +	st = m->m_pkthdr.pf.st;
> +	if (st != NULL) {
> +		if (READ_ONCE(st->inp) != NULL)
>  			return (1);
>  	}
>  
> @@ -8121,7 +8123,7 @@ pf_ouraddr(struct mbuf *m)
>  void
>  pf_pkt_addr_changed(struct mbuf *m)
>  {
> -	pf_mbuf_unlink_state_key(m);
> +	pf_mbuf_unlink_state(m);
>  	pf_mbuf_unlink_inpcb(m);
>  }
>  
> @@ -8129,85 +8131,97 @@ struct inpcb *
>  pf_inp_lookup(struct mbuf *m)
>  {
>  	struct inpcb *inp = NULL;
> -	struct pf_state_key *sk = m->m_pkthdr.pf.statekey;
> +	struct pf_state *st;
> +
> +	st = m->m_pkthdr.pf.st;
> +	if (st == NULL)
> +		return (NULL);
> +	if (!pf_state_isvalid(st)) {
> +		pf_mbuf_unlink_state(m);
> +		return (NULL);
> +	}
>  
> -	if (!pf_state_key_isvalid(sk))
> -		pf_mbuf_unlink_state_key(m);
> -	else if (READ_ONCE(sk->sk_inp) != NULL) {
> +	if (READ_ONCE(st->inp) != NULL) {
>  		mtx_enter(&pf_inp_mtx);
> -		inp = in_pcbref(sk->sk_inp);
> +		inp = in_pcbref(st->inp);
>  		mtx_leave(&pf_inp_mtx);
>  	}
>  
>  	return (inp);
>  }
>  
> +/*
> + * This is called from the IP stack after it's found an inpcb for
> + * an mbuf so it can link the pf_state to that pcb.
> + */
>  void
>  pf_inp_link(struct mbuf *m, struct inpcb *inp)
>  {
> -	struct pf_state_key *sk = m->m_pkthdr.pf.statekey;
> +	struct pf_state *st;
>  
> -	if (!pf_state_key_isvalid(sk)) {
> -		pf_mbuf_unlink_state_key(m);
> +	st = m->m_pkthdr.pf.st;
> +	if (st == NULL)
>  		return;
> -	}
>  
>  	/*
>  	 * we don't need to grab PF-lock here. At worst case we link inp to
>  	 * state, which might be just being marked as deleted by another
>  	 * thread.
>  	 */
> -	pf_state_key_link_inpcb(sk, inp);
> +	if (pf_state_isvalid(st)) {
> +		if (READ_ONCE(st->inp) == NULL)
> +			pf_state_link_inpcb(st, inp);
> +	}
>  
>  	/* The statekey has finished finding the inp, it is no longer needed. */
> -	pf_mbuf_unlink_state_key(m);
> +	pf_mbuf_unlink_state(m);
>  }
>  
>  void
>  pf_inp_unlink(struct inpcb *inp)
>  {
> -	struct pf_state_key *sk;
> +	struct pf_state *st;
>  
> -	if (READ_ONCE(inp->inp_pf_sk) == NULL)
> +	if (READ_ONCE(inp->inp_pf_st) == NULL)
>  		return;
>  
>  	mtx_enter(&pf_inp_mtx);
> -	sk = inp->inp_pf_sk;
> -	if (sk == NULL) {
> +	st = inp->inp_pf_st;
> +	if (st == NULL) {
>  		mtx_leave(&pf_inp_mtx);
>  		return;
>  	}
> -	KASSERT(sk->sk_inp == inp);
> -	sk->sk_inp = NULL;
> -	inp->inp_pf_sk = NULL;
> +	KASSERT(st->inp == inp);
> +	st->inp = NULL;
> +	inp->inp_pf_st = NULL;
>  	mtx_leave(&pf_inp_mtx);
>  
> -	pf_state_key_unref(sk);
> +	pf_state_unref(st);
>  	in_pcbunref(inp);
>  }
>  
>  void
> -pf_state_key_link_reverse(struct pf_state_key *sk, struct pf_state_key *skrev)
> +pf_state_link_reverse(struct pf_state *st, struct pf_state *strev)
>  {
> -	struct pf_state_key *old_reverse;
> +	struct pf_state *ost;
>  
> -	old_reverse = atomic_cas_ptr(&sk->sk_reverse, NULL, skrev);
> -	if (old_reverse != NULL)
> -		KASSERT(old_reverse == skrev);
> +	ost = atomic_cas_ptr(&st->reverse, NULL, strev);
> +	if (ost != NULL)
> +		KASSERT(ost == strev);
>  	else {
> -		pf_state_key_ref(skrev);
> +		pf_state_ref(strev);
>  
>  		/*
> -		 * NOTE: if sk == skrev, then KASSERT() below holds true, we
> +		 * NOTE: if st == strev, then KASSERT() below holds true, we
>  		 * still want to grab a reference in such case, because
> -		 * pf_state_key_unlink_reverse() does not check whether keys
> +		 * pf_state_unlink_reverse() does not check whether states
>  		 * are identical or not.
>  		 */
> -		old_reverse = atomic_cas_ptr(&skrev->sk_reverse, NULL, sk);
> -		if (old_reverse != NULL)
> -			KASSERT(old_reverse == sk);
> +		ost = atomic_cas_ptr(&strev->reverse, NULL, st);
> +		if (ost != NULL)
> +			KASSERT(ost == st);
>  
> -		pf_state_key_ref(sk);
> +		pf_state_ref(st);
>  	}
>  }
>  
> @@ -8243,10 +8257,6 @@ pf_state_key_unref(struct pf_state_key *
>  	if (PF_REF_RELE(sk->sk_refcnt)) {
>  		/* state key must be removed from tree */
>  		KASSERT(!pf_state_key_isvalid(sk));
> -		/* state key must be unlinked from reverse key */
> -		KASSERT(sk->sk_reverse == NULL);
> -		/* state key must be unlinked from socket */
> -		KASSERT(sk->sk_inp == NULL);
>  		pool_put(&pf_state_key_pl, sk);
>  	}
>  }
> @@ -8257,21 +8267,28 @@ pf_state_key_isvalid(struct pf_state_key
>  	return ((sk != NULL) && (sk->sk_removed == 0));
>  }
>  
> +int
> +pf_state_isvalid(struct pf_state *st)
> +{
> +	return (st->timeout < PFTM_MAX);
> +}
> +
>  void
> -pf_mbuf_link_state_key(struct mbuf *m, struct pf_state_key *sk)
> +pf_mbuf_link_state(struct mbuf *m, struct pf_state *st)
>  {
> -	KASSERT(m->m_pkthdr.pf.statekey == NULL);
> -	m->m_pkthdr.pf.statekey = pf_state_key_ref(sk);
> +	KASSERT(m->m_pkthdr.pf.st == NULL);
> +	m->m_pkthdr.pf.st = pf_state_ref(st);
>  }
>  
>  void
> -pf_mbuf_unlink_state_key(struct mbuf *m)
> +pf_mbuf_unlink_state(struct mbuf *m)
>  {
> -	struct pf_state_key *sk = m->m_pkthdr.pf.statekey;
> +	struct pf_state *st;
>  
> -	if (sk != NULL) {
> -		m->m_pkthdr.pf.statekey = NULL;
> -		pf_state_key_unref(sk);
> +	st = m->m_pkthdr.pf.st;
> +	if (st != NULL) {
> +		m->m_pkthdr.pf.st = NULL;
> +		pf_state_unref(st);
>  	}
>  }
>  
> @@ -8294,57 +8311,58 @@ pf_mbuf_unlink_inpcb(struct mbuf *m)
>  }
>  
>  void
> -pf_state_key_link_inpcb(struct pf_state_key *sk, struct inpcb *inp)
> +pf_state_link_inpcb(struct pf_state *st, struct inpcb *inp)
>  {
> -	if (inp == NULL || READ_ONCE(sk->sk_inp) != NULL)
> +	if (inp == NULL || READ_ONCE(st->inp) != NULL)
>  		return;
>  
>  	mtx_enter(&pf_inp_mtx);
> -	if (inp->inp_pf_sk != NULL || sk->sk_inp != NULL) {
> +	if (inp->inp_pf_st != NULL || st->inp != NULL) {
>  		mtx_leave(&pf_inp_mtx);
>  		return;
>  	}
> -	sk->sk_inp = in_pcbref(inp);
> -	inp->inp_pf_sk = pf_state_key_ref(sk);
> +	st->inp = in_pcbref(inp);
> +	inp->inp_pf_st = pf_state_ref(st);
>  	mtx_leave(&pf_inp_mtx);
>  }
>  
>  void
> -pf_state_key_unlink_inpcb(struct pf_state_key *sk)
> +pf_state_unlink_inpcb(struct pf_state *st)
>  {
>  	struct inpcb *inp;
>  
> -	if (READ_ONCE(sk->sk_inp) == NULL)
> +	if (READ_ONCE(st->inp) == NULL)
>  		return;
>  
>  	mtx_enter(&pf_inp_mtx);
> -	inp = sk->sk_inp;
> +	inp = st->inp;
>  	if (inp == NULL) {
>  		mtx_leave(&pf_inp_mtx);
>  		return;
>  	}
> -	KASSERT(inp->inp_pf_sk == sk);
> -	sk->sk_inp = NULL;
> -	inp->inp_pf_sk = NULL;
> +	KASSERT(inp->inp_pf_st == st);
> +	st->inp = NULL;
> +	inp->inp_pf_st = NULL;
>  	mtx_leave(&pf_inp_mtx);
>  
> -	pf_state_key_unref(sk);
> +	pf_state_unref(st);
>  	in_pcbunref(inp);
>  }
>  
>  void
> -pf_state_key_unlink_reverse(struct pf_state_key *sk)
> +pf_state_unlink_reverse(struct pf_state *st)
>  {
> -	struct pf_state_key *skrev = sk->sk_reverse;
> -
> -	/* Note that sk and skrev may be equal, then we unref twice. */
> -	if (skrev != NULL) {
> -		KASSERT(skrev->sk_reverse == sk);
> -		sk->sk_reverse = NULL;
> -		skrev->sk_reverse = NULL;
> -		pf_state_key_unref(skrev);
> -		pf_state_key_unref(sk);
> -	}
> +	struct pf_state *strev;
> + 
> +	/* Note that st and strev may be equal, then we unref twice. */
> +	strev = st->reverse;
> +	if (strev != NULL) {
> +		KASSERT(strev->reverse == st);
> +		st->reverse = NULL;
> +		strev->reverse = NULL;
> +		pf_state_unref(strev);
> +		pf_state_unref(st);
> +       }
>  }
>  
>  struct pf_state *
> @@ -8370,6 +8388,11 @@ pf_state_unref(struct pf_state *st)
>  
>  		pf_state_key_unref(st->key[PF_SK_WIRE]);
>  		pf_state_key_unref(st->key[PF_SK_STACK]);
> +
> +		/* state must be unlinked from reverse */
> +		KASSERT(st->reverse == NULL);
> +		/* state must be unlinked from socket */
> +		KASSERT(st->inp == NULL);
>  
>  		pool_put(&pf_state_pl, st);
>  	}
> Index: net/pfvar.h
> ===================================================================
> RCS file: /cvs/src/sys/net/pfvar.h,v
> diff -u -p -r1.543 pfvar.h
> --- net/pfvar.h	14 Apr 2025 20:02:34 -0000	1.543
> +++ net/pfvar.h	9 May 2025 02:48:26 -0000
> @@ -1849,9 +1849,8 @@ int			 pf_map_addr(sa_family_t, struct p
>  			    struct pf_pool *, enum pf_sn_types);
>  int			 pf_postprocess_addr(struct pf_state *);
>  
> -void			 pf_mbuf_link_state_key(struct mbuf *,
> -			    struct pf_state_key *);
> -void			 pf_mbuf_unlink_state_key(struct mbuf *);
> +void			 pf_mbuf_link_state(struct mbuf *, struct pf_state *);
> +void			 pf_mbuf_unlink_state(struct mbuf *);
>  void			 pf_mbuf_link_inpcb(struct mbuf *, struct inpcb *);
>  void			 pf_mbuf_unlink_inpcb(struct mbuf *);
>  
> Index: net/pfvar_priv.h
> ===================================================================
> RCS file: /cvs/src/sys/net/pfvar_priv.h,v
> diff -u -p -r1.38 pfvar_priv.h
> --- net/pfvar_priv.h	7 Sep 2024 22:41:55 -0000	1.38
> +++ net/pfvar_priv.h	9 May 2025 02:48:26 -0000
> @@ -41,11 +41,6 @@
>  #include <sys/mutex.h>
>  #include <sys/percpu.h>
>  
> -/*
> - * Locks used to protect struct members in this file:
> - *	L	pf_inp_mtx		link pf to inp mutex
> - */
> -
>  struct pfsync_deferral;
>  
>  /*
> @@ -74,8 +69,6 @@ struct pf_state_key {
>  
>  	RBT_ENTRY(pf_state_key)	 sk_entry;
>  	struct pf_statelisthead	 sk_states;
> -	struct pf_state_key	*sk_reverse;
> -	struct inpcb		*sk_inp;	/* [L] */
>  	pf_refcnt_t		 sk_refcnt;
>  	u_int8_t		 sk_removed;
>  };
> @@ -95,7 +88,8 @@ RBT_PROTOTYPE(pf_state_tree, pf_state_ke
>   *	M	pf_state mtx
>   *	P	PF_STATE_LOCK
>   *	S	pfsync
> - *	L	pf_state_list
> + *	L	pf_inp_mtx		link pf to inp mutex
> + *	G	pf_state_list
>   *	g	pf_purge gc
>   */
>  
> @@ -107,7 +101,7 @@ struct pf_state {
>  
>  	TAILQ_ENTRY(pf_state)	 sync_list;	/* [S] */
>  	struct pfsync_deferral	*sync_defer;	/* [S] */
> -	TAILQ_ENTRY(pf_state)	 entry_list;	/* [L] */
> +	TAILQ_ENTRY(pf_state)	 entry_list;	/* [G] */
>  	SLIST_ENTRY(pf_state)	 gc_list;	/* [g] */
>  	RBT_ENTRY(pf_state)	 entry_id;	/* [P] */
>  	struct pf_state_peer	 src;
> @@ -120,6 +114,8 @@ struct pf_state {
>  	struct pf_sn_head	 src_nodes;	/* [I] */
>  	struct pf_state_key	*key[2];	/* [I] stack and wire */
>  	struct pfi_kif		*kif;		/* [I] */
> +	struct pf_state		*reverse;
> +	struct inpcb		*inp;		/* [L] */
>  	struct mutex		 mtx;
>  	pf_refcnt_t		 refcnt;
>  	u_int64_t		 packets[2];
> Index: netinet/in_pcb.h
> ===================================================================
> RCS file: /cvs/src/sys/netinet/in_pcb.h,v
> diff -u -p -r1.167 in_pcb.h
> --- netinet/in_pcb.h	4 May 2025 23:05:17 -0000	1.167
> +++ netinet/in_pcb.h	9 May 2025 02:48:26 -0000
> @@ -110,7 +110,7 @@
>   * Protocol input only reads inp_[lf]addr/port during lookup and is safe.
>   */
>  
> -struct pf_state_key;
> +struct pf_state;
>  
>  union inpaddru {
>  	struct in_addr iau_addr;
> @@ -168,7 +168,7 @@ struct inpcb {
>  
>  	int	inp_cksum6;
>  	struct	icmp6_filter *inp_icmp6filt;
> -	struct	pf_state_key *inp_pf_sk; /* [L] */
> +	struct	pf_state *inp_pf_st;	/* [L] */
>  	struct	mbuf *(*inp_upcall)(void *, struct mbuf *,
>  	    struct ip *, struct ip6_hdr *, void *, int, struct netstack *);
>  	void	*inp_upcall_arg;
> Index: sys/mbuf.h
> ===================================================================
> RCS file: /cvs/src/sys/sys/mbuf.h,v
> diff -u -p -r1.265 mbuf.h
> --- sys/mbuf.h	5 Nov 2024 13:15:13 -0000	1.265
> +++ sys/mbuf.h	9 May 2025 02:48:26 -0000
> @@ -90,11 +90,11 @@ struct m_hdr {
>  };
>  
>  /* pf stuff */
> -struct pf_state_key;
> +struct pf_state;
>  struct inpcb;
>  
>  struct pkthdr_pf {
> -	struct pf_state_key *statekey;	/* pf stackside statekey */
> +	struct pf_state	*st;		/* pf state */
>  	struct inpcb	*inp;		/* connected pcb for outgoing packet */
>  	u_int32_t	 qid;		/* queue id */
>  	u_int16_t	 tag;		/* tag id */
> @@ -325,7 +325,7 @@ u_int mextfree_register(void (*)(caddr_t
>  	(to)->m_pkthdr = (from)->m_pkthdr;				\
>  	(from)->m_flags &= ~M_PKTHDR;					\
>  	SLIST_INIT(&(from)->m_pkthdr.ph_tags);				\
> -	(from)->m_pkthdr.pf.statekey = NULL;				\
> +	(from)->m_pkthdr.pf.st = NULL;					\
>  } while (/* CONSTCOND */ 0)
>  
>  /*