xref: /freebsd/sys/netpfil/pf/pf.c (revision aa790af10a9a2ce36ca0b2bc445a01431b8bc0c4)

/*-
 * SPDX-License-Identifier: BSD-2-Clause
 *
 * Copyright (c) 2001 Daniel Hartmeier
 * Copyright (c) 2002 - 2008 Henning Brauer
 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 *    - Redistributions of source code must retain the above copyright
 *      notice, this list of conditions and the following disclaimer.
 *    - Redistributions in binary form must reproduce the above
 *      copyright notice, this list of conditions and the following
 *      disclaimer in the documentation and/or other materials provided
 *      with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 *
 * Effort sponsored in part by the Defense Advanced Research Projects
 * Agency (DARPA) and Air Force Research Laboratory, Air Force
 * Materiel Command, USAF, under agreement number F30602-01-2-0537.
 *
 *	$OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
 */

#include <sys/cdefs.h>
#include "opt_bpf.h"
#include "opt_inet.h"
#include "opt_inet6.h"
#include "opt_pf.h"
#include "opt_sctp.h"

#include <sys/param.h>
#include <sys/bus.h>
#include <sys/endian.h>
#include <sys/gsb_crc32.h>
#include <sys/hash.h>
#include <sys/interrupt.h>
#include <sys/kernel.h>
#include <sys/kthread.h>
#include <sys/limits.h>
#include <sys/mbuf.h>
#include <sys/random.h>
#include <sys/refcount.h>
#include <sys/sdt.h>
#include <sys/socket.h>
#include <sys/sysctl.h>
#include <sys/taskqueue.h>
#include <sys/ucred.h>

#include <crypto/sha2/sha512.h>

#include <net/if.h>
#include <net/if_var.h>
#include <net/if_private.h>
#include <net/if_types.h>
#include <net/if_vlan_var.h>
#include <net/route.h>
#include <net/route/nhop.h>
#include <net/vnet.h>

#include <net/pfil.h>
#include <net/pfvar.h>
#include <net/if_pflog.h>
#include <net/if_pfsync.h>

#include <netinet/in_pcb.h>
#include <netinet/in_var.h>
#include <netinet/in_fib.h>
#include <netinet/ip.h>
#include <netinet/ip_fw.h>
#include <netinet/ip_icmp.h>
#include <netinet/icmp_var.h>
#include <netinet/ip_var.h>
#include <netinet/tcp.h>
#include <netinet/tcp_fsm.h>
#include <netinet/tcp_seq.h>
#include <netinet/tcp_timer.h>
#include <netinet/tcp_var.h>
#include <netinet/udp.h>
#include <netinet/udp_var.h>

/* dummynet */
#include <netinet/ip_dummynet.h>
#include <netinet/ip_fw.h>
#include <netpfil/ipfw/dn_heap.h>
#include <netpfil/ipfw/ip_fw_private.h>
#include <netpfil/ipfw/ip_dn_private.h>

#ifdef INET6
#include <netinet/ip6.h>
#include <netinet/icmp6.h>
#include <netinet6/nd6.h>
#include <netinet6/ip6_var.h>
#include <netinet6/in6_pcb.h>
#include <netinet6/in6_fib.h>
#include <netinet6/scope6_var.h>
#endif /* INET6 */

#include <netinet/sctp_header.h>
#include <netinet/sctp_crc32.h>

#include <netipsec/ah.h>

#include <machine/in_cksum.h>
#include <security/mac/mac_framework.h>

SDT_PROVIDER_DEFINE(pf);
SDT_PROBE_DEFINE2(pf, , test, reason_set, "int", "int");
SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *",
    "struct pf_kstate *");
SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *",
    "struct pf_state_key_cmp *", "int", "struct pf_pdesc *",
    "struct pf_kstate *");
SDT_PROBE_DEFINE2(pf, ip, , bound_iface, "struct pf_kstate *",
    "struct pfi_kkif *");
SDT_PROBE_DEFINE4(pf, ip, route_to, entry, "struct mbuf *",
    "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *");
SDT_PROBE_DEFINE1(pf, ip, route_to, drop, "int");
SDT_PROBE_DEFINE2(pf, ip, route_to, output, "struct ifnet *", "int");
SDT_PROBE_DEFINE4(pf, ip6, route_to, entry, "struct mbuf *",
    "struct pf_pdesc *", "struct pf_kstate *", "struct ifnet *");
SDT_PROBE_DEFINE1(pf, ip6, route_to, drop, "int");
SDT_PROBE_DEFINE2(pf, ip6, route_to, output, "struct ifnet *", "int");
SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *",
    "struct pf_krule *", "struct mbuf *", "int");
SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t",
    "struct pf_sctp_source *");
SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t",
    "struct pf_kstate *", "struct pf_sctp_source *");
SDT_PROBE_DEFINE4(pf, sctp, multihome_scan, entry, "int",
    "int", "struct pf_pdesc *", "int");
SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, param, "uint16_t", "uint16_t");
SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv4, "struct in_addr *",
    "int");
SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv6, "struct in_addr6 *",
    "int");

SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *",
    "struct mbuf *");
SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *");
SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch,
    "int", "struct pf_keth_rule *", "char *");
SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *");
SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match,
    "int", "struct pf_keth_rule *");
SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t");
SDT_PROBE_DEFINE2(pf, , log, log, "int", "const char *");

/*
 * Global variables
 */

/* state tables */
VNET_DEFINE(struct pf_altqqueue,	 pf_altqs[4]);
VNET_DEFINE(struct pf_kpalist,		 pf_pabuf[3]);
VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_active);
VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_active);
VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_inactive);
VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_inactive);
VNET_DEFINE(struct pf_kstatus,		 pf_status);

VNET_DEFINE(u_int32_t,			 ticket_altqs_active);
VNET_DEFINE(u_int32_t,			 ticket_altqs_inactive);
VNET_DEFINE(int,			 altqs_inactive_open);
VNET_DEFINE(u_int32_t,			 ticket_pabuf);

static const int			 PF_HDR_LIMIT = 20;	/* arbitrary limit */

VNET_DEFINE(SHA512_CTX,			 pf_tcp_secret_ctx);
#define	V_pf_tcp_secret_ctx		 VNET(pf_tcp_secret_ctx)
VNET_DEFINE(u_char,			 pf_tcp_secret[16]);
#define	V_pf_tcp_secret			 VNET(pf_tcp_secret)
VNET_DEFINE(int,			 pf_tcp_secret_init);
#define	V_pf_tcp_secret_init		 VNET(pf_tcp_secret_init)
VNET_DEFINE(int,			 pf_tcp_iss_off);
#define	V_pf_tcp_iss_off		 VNET(pf_tcp_iss_off)
VNET_DECLARE(int,			 pf_vnet_active);
#define	V_pf_vnet_active		 VNET(pf_vnet_active)

VNET_DEFINE_STATIC(uint32_t, pf_purge_idx);
#define V_pf_purge_idx	VNET(pf_purge_idx)

#ifdef PF_WANT_32_TO_64_COUNTER
VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter);
#define	V_pf_counter_periodic_iter	VNET(pf_counter_periodic_iter)

VNET_DEFINE(struct allrulelist_head, pf_allrulelist);
VNET_DEFINE(size_t, pf_allrulecount);
VNET_DEFINE(struct pf_krule *, pf_rulemarker);
#endif

#define PF_SCTP_MAX_ENDPOINTS		8

struct pf_sctp_endpoint;
RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint);
struct pf_sctp_source {
	sa_family_t			af;
	struct pf_addr			addr;
	TAILQ_ENTRY(pf_sctp_source)	entry;
};
TAILQ_HEAD(pf_sctp_sources, pf_sctp_source);
struct pf_sctp_endpoint
{
	uint32_t		 v_tag;
	struct pf_sctp_sources	 sources;
	RB_ENTRY(pf_sctp_endpoint)	entry;
};
static int
pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b)
{
	return (a->v_tag - b->v_tag);
}
RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare);
RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare);
VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints);
#define V_pf_sctp_endpoints	VNET(pf_sctp_endpoints)
static struct mtx_padalign pf_sctp_endpoints_mtx;
MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF);
#define	PF_SCTP_ENDPOINTS_LOCK()	mtx_lock(&pf_sctp_endpoints_mtx)
#define	PF_SCTP_ENDPOINTS_UNLOCK()	mtx_unlock(&pf_sctp_endpoints_mtx)

/*
 * Queue for pf_intr() sends.
 */
static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
struct pf_send_entry {
	STAILQ_ENTRY(pf_send_entry)	pfse_next;
	struct mbuf			*pfse_m;
	enum {
		PFSE_IP,
		PFSE_IP6,
		PFSE_ICMP,
		PFSE_ICMP6,
	}				pfse_type;
	struct {
		int		type;
		int		code;
		int		mtu;
	} icmpopts;
};

STAILQ_HEAD(pf_send_head, pf_send_entry);
VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue);
#define	V_pf_sendqueue	VNET(pf_sendqueue)

static struct mtx_padalign pf_sendqueue_mtx;
MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
#define	PF_SENDQ_LOCK()		mtx_lock(&pf_sendqueue_mtx)
#define	PF_SENDQ_UNLOCK()	mtx_unlock(&pf_sendqueue_mtx)

/*
 * Queue for pf_overload_task() tasks.
 */
struct pf_overload_entry {
	SLIST_ENTRY(pf_overload_entry)	next;
	struct pf_addr  		addr;
	sa_family_t			af;
	uint8_t				dir;
	struct pf_krule  		*rule;
};

SLIST_HEAD(pf_overload_head, pf_overload_entry);
VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue);
#define V_pf_overloadqueue	VNET(pf_overloadqueue)
VNET_DEFINE_STATIC(struct task, pf_overloadtask);
#define	V_pf_overloadtask	VNET(pf_overloadtask)

static struct mtx_padalign pf_overloadqueue_mtx;
MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
    "pf overload/flush queue", MTX_DEF);
#define	PF_OVERLOADQ_LOCK()	mtx_lock(&pf_overloadqueue_mtx)
#define	PF_OVERLOADQ_UNLOCK()	mtx_unlock(&pf_overloadqueue_mtx)

VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules);
struct mtx_padalign pf_unlnkdrules_mtx;
MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
    MTX_DEF);

struct sx pf_config_lock;
SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config");

struct mtx_padalign pf_table_stats_lock;
MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats",
    MTX_DEF);

VNET_DEFINE_STATIC(uma_zone_t,	pf_sources_z);
#define	V_pf_sources_z	VNET(pf_sources_z)
uma_zone_t		pf_mtag_z;
VNET_DEFINE(uma_zone_t,	 pf_state_z);
VNET_DEFINE(uma_zone_t,	 pf_state_key_z);
VNET_DEFINE(uma_zone_t,	 pf_udp_mapping_z);

VNET_DEFINE(struct unrhdr64, pf_stateid);

static void		 pf_src_tree_remove_state(struct pf_kstate *);
static int		 pf_check_threshold(struct pf_kthreshold *);

static void		 pf_change_ap(struct pf_pdesc *, struct pf_addr *, u_int16_t *,
			    struct pf_addr *, u_int16_t);
static int		 pf_modulate_sack(struct pf_pdesc *,
			    struct tcphdr *, struct pf_state_peer *);
int			 pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *,
			    u_int16_t *, u_int16_t *);
static void		 pf_change_icmp(struct pf_addr *, u_int16_t *,
			    struct pf_addr *, struct pf_addr *, u_int16_t,
			    u_int16_t *, u_int16_t *, u_int16_t *,
			    u_int16_t *, u_int8_t, sa_family_t);
int			 pf_change_icmp_af(struct mbuf *, int,
			    struct pf_pdesc *, struct pf_pdesc *,
			    struct pf_addr *, struct pf_addr *, sa_family_t,
			    sa_family_t);
int			 pf_translate_icmp_af(int, void *);
static void		 pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
			    int, sa_family_t, struct pf_krule *, int);
static void		 pf_detach_state(struct pf_kstate *);
static int		 pf_state_key_attach(struct pf_state_key *,
			    struct pf_state_key *, struct pf_kstate *);
static void		 pf_state_key_detach(struct pf_kstate *, int);
static int		 pf_state_key_ctor(void *, int, void *, int);
static u_int32_t	 pf_tcp_iss(struct pf_pdesc *);
static __inline void	 pf_dummynet_flag_remove(struct mbuf *m,
			    struct pf_mtag *pf_mtag);
static int		 pf_dummynet(struct pf_pdesc *, struct pf_kstate *,
			    struct pf_krule *, struct mbuf **);
static int		 pf_dummynet_route(struct pf_pdesc *,
			    struct pf_kstate *, struct pf_krule *,
			    struct ifnet *, const struct sockaddr *, struct mbuf **);
static int		 pf_test_eth_rule(int, struct pfi_kkif *,
			    struct mbuf **);
static int		 pf_test_rule(struct pf_krule **, struct pf_kstate **,
			    struct pf_pdesc *, struct pf_krule **,
			    struct pf_kruleset **, u_short *, struct inpcb *);
static int		 pf_create_state(struct pf_krule *,
			    struct pf_test_ctx *,
			    struct pf_kstate **, u_int16_t, u_int16_t);
static int		 pf_state_key_addr_setup(struct pf_pdesc *,
			    struct pf_state_key_cmp *, int);
static int		 pf_tcp_track_full(struct pf_kstate *,
			    struct pf_pdesc *, u_short *, int *,
			    struct pf_state_peer *, struct pf_state_peer *,
			    u_int8_t, u_int8_t);
static int		 pf_tcp_track_sloppy(struct pf_kstate *,
			    struct pf_pdesc *, u_short *,
			    struct pf_state_peer *, struct pf_state_peer *,
			    u_int8_t, u_int8_t);
static int		 pf_test_state(struct pf_kstate **, struct pf_pdesc *,
			    u_short *);
int			 pf_icmp_state_lookup(struct pf_state_key_cmp *,
			    struct pf_pdesc *, struct pf_kstate **,
			    u_int16_t, u_int16_t, int, int *, int, int);
static int		 pf_test_state_icmp(struct pf_kstate **,
			    struct pf_pdesc *, u_short *);
static int		 pf_sctp_track(struct pf_kstate *, struct pf_pdesc *,
			    u_short *);
static void		 pf_sctp_multihome_detach_addr(const struct pf_kstate *);
static void		 pf_sctp_multihome_delayed(struct pf_pdesc *,
			    struct pfi_kkif *, struct pf_kstate *, int);
static u_int16_t	 pf_calc_mss(struct pf_addr *, sa_family_t,
				int, u_int16_t);
static int		 pf_check_proto_cksum(struct mbuf *, int, int,
			    u_int8_t, sa_family_t);
static int		 pf_walk_option(struct pf_pdesc *, struct ip *,
			    int, int, u_short *);
static int		 pf_walk_header(struct pf_pdesc *, struct ip *, u_short *);
#ifdef INET6
static int		 pf_walk_option6(struct pf_pdesc *, struct ip6_hdr *,
			    int, int, u_short *);
static int		 pf_walk_header6(struct pf_pdesc *, struct ip6_hdr *,
			    u_short *);
#endif
static void		 pf_print_state_parts(struct pf_kstate *,
			    struct pf_state_key *, struct pf_state_key *);
static int		 pf_patch_8(struct pf_pdesc *, u_int8_t *, u_int8_t,
			    bool);
static int		 pf_find_state(struct pf_pdesc *,
			    const struct pf_state_key_cmp *, struct pf_kstate **);
static bool		 pf_src_connlimit(struct pf_kstate *);
static int		 pf_match_rcvif(struct mbuf *, struct pf_krule *);
static void		 pf_counters_inc(int, struct pf_pdesc *,
			    struct pf_kstate *, struct pf_krule *,
			    struct pf_krule *);
static void		 pf_log_matches(struct pf_pdesc *, struct pf_krule *,
			    struct pf_krule *, struct pf_kruleset *,
			    struct pf_krule_slist *);
static void		 pf_overload_task(void *v, int pending);
static u_short		 pf_insert_src_node(struct pf_ksrc_node *[PF_SN_MAX],
			    struct pf_srchash *[PF_SN_MAX], struct pf_krule *,
			    struct pf_addr *, sa_family_t, struct pf_addr *,
			    struct pfi_kkif *, sa_family_t, pf_sn_types_t);
static u_int		 pf_purge_expired_states(u_int, int);
static void		 pf_purge_unlinked_rules(void);
static int		 pf_mtag_uminit(void *, int, int);
static void		 pf_mtag_free(struct m_tag *);
static void		 pf_packet_rework_nat(struct pf_pdesc *, int,
			    struct pf_state_key *);
#ifdef INET
static int		 pf_route(struct pf_krule *,
			    struct ifnet *, struct pf_kstate *,
			    struct pf_pdesc *, struct inpcb *);
#endif /* INET */
#ifdef INET6
static void		 pf_change_a6(struct pf_addr *, u_int16_t *,
			    struct pf_addr *, u_int8_t);
static int		 pf_route6(struct pf_krule *,
			    struct ifnet *, struct pf_kstate *,
			    struct pf_pdesc *, struct inpcb *);
#endif /* INET6 */
static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t);

int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);

extern int pf_end_threads;
extern struct proc *pf_purge_proc;

VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);

#define	PACKET_UNDO_NAT(_pd, _off, _s)					\
	do {								\
		struct pf_state_key *nk;				\
		if ((pd->dir) == PF_OUT)				\
			nk = (_s)->key[PF_SK_STACK];			\
		else							\
			nk = (_s)->key[PF_SK_WIRE];			\
		pf_packet_rework_nat(_pd, _off, nk);		\
	} while (0)

#define	PACKET_LOOPED(pd)	((pd)->pf_mtag &&			\
				 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED)

static struct pfi_kkif *
BOUND_IFACE(struct pf_kstate *st, struct pf_pdesc *pd)
{
	struct pfi_kkif *k = pd->kif;

	SDT_PROBE2(pf, ip, , bound_iface, st, k);

	/* Floating unless otherwise specified. */
	if (! (st->rule->rule_flag & PFRULE_IFBOUND))
		return (V_pfi_all);

	/*
	 * Initially set to all, because we don't know what interface we'll be
	 * sending this out when we create the state.
	 */
	if (st->rule->rt == PF_REPLYTO || (pd->af != pd->naf && st->direction == PF_IN))
		return (V_pfi_all);

	/*
	 * If this state is created based on another state (e.g. SCTP
	 * multihome) always set it floating initially. We can't know for sure
	 * what interface the actual traffic for this state will come in on.
	 */
	if (pd->related_rule)
		return (V_pfi_all);

	/* Don't overrule the interface for states created on incoming packets. */
	if (st->direction == PF_IN)
		return (k);

	/* No route-to, so don't overrule. */
	if (st->act.rt != PF_ROUTETO)
		return (k);

	/* Bind to the route-to interface. */
	return (st->act.rt_kif);
}

#define	STATE_INC_COUNTERS(s)						\
	do {								\
		struct pf_krule_item *mrm;				\
		counter_u64_add(s->rule->states_cur, 1);		\
		counter_u64_add(s->rule->states_tot, 1);		\
		if (s->anchor != NULL) {				\
			counter_u64_add(s->anchor->states_cur, 1);	\
			counter_u64_add(s->anchor->states_tot, 1);	\
		}							\
		if (s->nat_rule != NULL) {				\
			counter_u64_add(s->nat_rule->states_cur, 1);\
			counter_u64_add(s->nat_rule->states_tot, 1);\
		}							\
		SLIST_FOREACH(mrm, &s->match_rules, entry) {		\
			counter_u64_add(mrm->r->states_cur, 1);		\
			counter_u64_add(mrm->r->states_tot, 1);		\
		}							\
	} while (0)

#define	STATE_DEC_COUNTERS(s)						\
	do {								\
		struct pf_krule_item *mrm;				\
		if (s->nat_rule != NULL)				\
			counter_u64_add(s->nat_rule->states_cur, -1);\
		if (s->anchor != NULL)				\
			counter_u64_add(s->anchor->states_cur, -1);	\
		counter_u64_add(s->rule->states_cur, -1);		\
		SLIST_FOREACH(mrm, &s->match_rules, entry)		\
			counter_u64_add(mrm->r->states_cur, -1);	\
	} while (0)

MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items");
VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
VNET_DEFINE(struct pf_idhash *, pf_idhash);
VNET_DEFINE(struct pf_srchash *, pf_srchash);
VNET_DEFINE(struct pf_udpendpointhash *, pf_udpendpointhash);
VNET_DEFINE(struct pf_udpendpointmapping *, pf_udpendpointmapping);

SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
    "pf(4)");

VNET_DEFINE(u_long, pf_hashmask);
VNET_DEFINE(u_long, pf_srchashmask);
VNET_DEFINE(u_long, pf_udpendpointhashmask);
VNET_DEFINE_STATIC(u_long, pf_hashsize);
#define V_pf_hashsize	VNET(pf_hashsize)
VNET_DEFINE_STATIC(u_long, pf_srchashsize);
#define V_pf_srchashsize	VNET(pf_srchashsize)
VNET_DEFINE_STATIC(u_long, pf_udpendpointhashsize);
#define V_pf_udpendpointhashsize	VNET(pf_udpendpointhashsize)
u_long	pf_ioctl_maxcount = 65535;

SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN,
    &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable");
SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN,
    &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable");
SYSCTL_ULONG(_net_pf, OID_AUTO, udpendpoint_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN,
    &VNET_NAME(pf_udpendpointhashsize), 0, "Size of pf(4) endpoint hashtable");
SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN,
    &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call");

VNET_DEFINE(void *, pf_swi_cookie);
VNET_DEFINE(struct intr_event *, pf_swi_ie);

VNET_DEFINE(uint32_t, pf_hashseed);
#define	V_pf_hashseed	VNET(pf_hashseed)

static void
pf_sctp_checksum(struct mbuf *m, int off)
{
	uint32_t sum = 0;

	/* Zero out the checksum, to enable recalculation. */
	m_copyback(m, off + offsetof(struct sctphdr, checksum),
	    sizeof(sum), (caddr_t)&sum);

	sum = sctp_calculate_cksum(m, off);

	m_copyback(m, off + offsetof(struct sctphdr, checksum),
	    sizeof(sum), (caddr_t)&sum);
}

int
pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
{

	switch (af) {
#ifdef INET
	case AF_INET:
		if (a->addr32[0] > b->addr32[0])
			return (1);
		if (a->addr32[0] < b->addr32[0])
			return (-1);
		break;
#endif /* INET */
#ifdef INET6
	case AF_INET6:
		if (a->addr32[3] > b->addr32[3])
			return (1);
		if (a->addr32[3] < b->addr32[3])
			return (-1);
		if (a->addr32[2] > b->addr32[2])
			return (1);
		if (a->addr32[2] < b->addr32[2])
			return (-1);
		if (a->addr32[1] > b->addr32[1])
			return (1);
		if (a->addr32[1] < b->addr32[1])
			return (-1);
		if (a->addr32[0] > b->addr32[0])
			return (1);
		if (a->addr32[0] < b->addr32[0])
			return (-1);
		break;
#endif /* INET6 */
	default:
		unhandled_af(af);
	}
	return (0);
}

static bool
pf_is_loopback(sa_family_t af, struct pf_addr *addr)
{
	switch (af) {
#ifdef INET
	case AF_INET:
		return IN_LOOPBACK(ntohl(addr->v4.s_addr));
#endif /* INET */
	case AF_INET6:
		return IN6_IS_ADDR_LOOPBACK(&addr->v6);
	default:
		unhandled_af(af);
	}
}

static void
pf_packet_rework_nat(struct pf_pdesc *pd, int off, struct pf_state_key *nk)
{

	switch (pd->virtual_proto) {
	case IPPROTO_TCP: {
		struct tcphdr *th = &pd->hdr.tcp;

		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af))
			pf_change_ap(pd, pd->src, &th->th_sport,
			    &nk->addr[pd->sidx], nk->port[pd->sidx]);
		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af))
			pf_change_ap(pd, pd->dst, &th->th_dport,
			    &nk->addr[pd->didx], nk->port[pd->didx]);
		m_copyback(pd->m, off, sizeof(*th), (caddr_t)th);
		break;
	}
	case IPPROTO_UDP: {
		struct udphdr *uh = &pd->hdr.udp;

		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af))
			pf_change_ap(pd, pd->src, &uh->uh_sport,
			    &nk->addr[pd->sidx], nk->port[pd->sidx]);
		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af))
			pf_change_ap(pd, pd->dst, &uh->uh_dport,
			    &nk->addr[pd->didx], nk->port[pd->didx]);
		m_copyback(pd->m, off, sizeof(*uh), (caddr_t)uh);
		break;
	}
	case IPPROTO_SCTP: {
		struct sctphdr *sh = &pd->hdr.sctp;

		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) {
			pf_change_ap(pd, pd->src, &sh->src_port,
			    &nk->addr[pd->sidx], nk->port[pd->sidx]);
		}
		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) {
			pf_change_ap(pd, pd->dst, &sh->dest_port,
			    &nk->addr[pd->didx], nk->port[pd->didx]);
		}

		break;
	}
	case IPPROTO_ICMP: {
		struct icmp *ih = &pd->hdr.icmp;

		if (nk->port[pd->sidx] != ih->icmp_id) {
			pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
			    ih->icmp_cksum, ih->icmp_id,
			    nk->port[pd->sidx], 0);
			ih->icmp_id = nk->port[pd->sidx];
			pd->sport = &ih->icmp_id;

			m_copyback(pd->m, off, ICMP_MINLEN, (caddr_t)ih);
		}
		/* FALLTHROUGH */
	}
	default:
		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) {
			switch (pd->af) {
			case AF_INET:
				pf_change_a(&pd->src->v4.s_addr,
				    pd->ip_sum, nk->addr[pd->sidx].v4.s_addr,
				    0);
				break;
			case AF_INET6:
				pf_addrcpy(pd->src, &nk->addr[pd->sidx],
				    pd->af);
				break;
			default:
				unhandled_af(pd->af);
			}
		}
		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) {
			switch (pd->af) {
			case AF_INET:
				pf_change_a(&pd->dst->v4.s_addr,
				    pd->ip_sum, nk->addr[pd->didx].v4.s_addr,
				    0);
				break;
			case AF_INET6:
				pf_addrcpy(pd->dst, &nk->addr[pd->didx],
				    pd->af);
				break;
			default:
				unhandled_af(pd->af);
			}
		}
		break;
	}
}

static __inline uint32_t
pf_hashkey(const struct pf_state_key *sk)
{
	uint32_t h;

	h = murmur3_32_hash32((const uint32_t *)sk,
	    sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
	    V_pf_hashseed);

	return (h & V_pf_hashmask);
}

__inline uint32_t
pf_hashsrc(struct pf_addr *addr, sa_family_t af)
{
	uint32_t h;

	switch (af) {
	case AF_INET:
		h = murmur3_32_hash32((uint32_t *)&addr->v4,
		    sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
		break;
	case AF_INET6:
		h = murmur3_32_hash32((uint32_t *)&addr->v6,
		    sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
		break;
	default:
		unhandled_af(af);
	}

	return (h & V_pf_srchashmask);
}

static inline uint32_t
pf_hashudpendpoint(struct pf_udp_endpoint *endpoint)
{
	uint32_t h;

	h = murmur3_32_hash32((uint32_t *)endpoint,
	    sizeof(struct pf_udp_endpoint_cmp)/sizeof(uint32_t),
	    V_pf_hashseed);
	return (h & V_pf_udpendpointhashmask);
}

#ifdef ALTQ
static int
pf_state_hash(struct pf_kstate *s)
{
	u_int32_t hv = (intptr_t)s / sizeof(*s);

	hv ^= crc32(&s->src, sizeof(s->src));
	hv ^= crc32(&s->dst, sizeof(s->dst));
	if (hv == 0)
		hv = 1;
	return (hv);
}
#endif /* ALTQ */

static __inline void
pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate)
{
	if (which == PF_PEER_DST || which == PF_PEER_BOTH)
		s->dst.state = newstate;
	if (which == PF_PEER_DST)
		return;
	if (s->src.state == newstate)
		return;
	if (s->creatorid == V_pf_status.hostid &&
	    s->key[PF_SK_STACK] != NULL &&
	    s->key[PF_SK_STACK]->proto == IPPROTO_TCP &&
	    !(TCPS_HAVEESTABLISHED(s->src.state) ||
	    s->src.state == TCPS_CLOSED) &&
	    (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED))
		atomic_add_32(&V_pf_status.states_halfopen, -1);

	s->src.state = newstate;
}

bool
pf_init_threshold(struct pf_kthreshold *threshold,
    u_int32_t limit, u_int32_t seconds)
{
	threshold->limit = limit;
	threshold->seconds = seconds;
	threshold->cr = counter_rate_alloc(M_NOWAIT, seconds);

	return (threshold->cr != NULL);
}

static int
pf_check_threshold(struct pf_kthreshold *threshold)
{
	return (counter_ratecheck(threshold->cr, threshold->limit) < 0);
}

static bool
pf_src_connlimit(struct pf_kstate *state)
{
	struct pf_overload_entry	*pfoe;
	struct pf_ksrc_node		*src_node = state->sns[PF_SN_LIMIT];
	bool				 limited = false;

	PF_STATE_LOCK_ASSERT(state);
	PF_SRC_NODE_LOCK(src_node);

	src_node->conn++;
	state->src.tcp_est = 1;

	if (state->rule->max_src_conn &&
	    state->rule->max_src_conn <
	    src_node->conn) {
		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
		limited = true;
	}

	if (state->rule->max_src_conn_rate.limit &&
	    pf_check_threshold(&src_node->conn_rate)) {
		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
		limited = true;
	}

	if (!limited)
		goto done;

	/* Kill this state. */
	state->timeout = PFTM_PURGE;
	pf_set_protostate(state, PF_PEER_BOTH, TCPS_CLOSED);

	if (state->rule->overload_tbl == NULL)
		goto done;

	/* Schedule overloading and flushing task. */
	pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
	if (pfoe == NULL)
		goto done;  /* too bad :( */

	bcopy(&src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
	pfoe->af = state->key[PF_SK_WIRE]->af;
	pfoe->rule = state->rule;
	pfoe->dir = state->direction;
	PF_OVERLOADQ_LOCK();
	SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
	PF_OVERLOADQ_UNLOCK();
	taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);

done:
	PF_SRC_NODE_UNLOCK(src_node);
	return (limited);
}

static void
pf_overload_task(void *v, int pending)
{
	struct pf_overload_head queue;
	struct pfr_addr p;
	struct pf_overload_entry *pfoe, *pfoe1;
	uint32_t killed = 0;

	CURVNET_SET((struct vnet *)v);

	PF_OVERLOADQ_LOCK();
	queue = V_pf_overloadqueue;
	SLIST_INIT(&V_pf_overloadqueue);
	PF_OVERLOADQ_UNLOCK();

	bzero(&p, sizeof(p));
	SLIST_FOREACH(pfoe, &queue, next) {
		counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
		if (V_pf_status.debug >= PF_DEBUG_MISC) {
			printf("%s: blocking address ", __func__);
			pf_print_host(&pfoe->addr, 0, pfoe->af);
			printf("\n");
		}

		p.pfra_af = pfoe->af;
		switch (pfoe->af) {
#ifdef INET
		case AF_INET:
			p.pfra_net = 32;
			p.pfra_ip4addr = pfoe->addr.v4;
			break;
#endif /* INET */
#ifdef INET6
		case AF_INET6:
			p.pfra_net = 128;
			p.pfra_ip6addr = pfoe->addr.v6;
			break;
#endif /* INET6 */
		default:
			unhandled_af(pfoe->af);
		}

		PF_RULES_WLOCK();
		pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
		PF_RULES_WUNLOCK();
	}

	/*
	 * Remove those entries, that don't need flushing.
	 */
	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
		if (pfoe->rule->flush == 0) {
			SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
			free(pfoe, M_PFTEMP);
		} else
			counter_u64_add(
			    V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);

	/* If nothing to flush, return. */
	if (SLIST_EMPTY(&queue)) {
		CURVNET_RESTORE();
		return;
	}

	for (int i = 0; i <= V_pf_hashmask; i++) {
		struct pf_idhash *ih = &V_pf_idhash[i];
		struct pf_state_key *sk;
		struct pf_kstate *s;

		PF_HASHROW_LOCK(ih);
		LIST_FOREACH(s, &ih->states, entry) {
		    sk = s->key[PF_SK_WIRE];
		    SLIST_FOREACH(pfoe, &queue, next)
			if (sk->af == pfoe->af &&
			    ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
			    pfoe->rule == s->rule) &&
			    ((pfoe->dir == PF_OUT &&
			    PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
			    (pfoe->dir == PF_IN &&
			    PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
				s->timeout = PFTM_PURGE;
				pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED);
				killed++;
			}
		}
		PF_HASHROW_UNLOCK(ih);
	}
	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
		free(pfoe, M_PFTEMP);
	if (V_pf_status.debug >= PF_DEBUG_MISC)
		printf("%s: %u states killed", __func__, killed);

	CURVNET_RESTORE();
}

/*
 * On node found always returns locked. On not found its configurable.
 */
struct pf_ksrc_node *
pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af,
    struct pf_srchash **sh, pf_sn_types_t sn_type, bool returnlocked)
{
	struct pf_ksrc_node *n;

	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);

	*sh = &V_pf_srchash[pf_hashsrc(src, af)];
	PF_HASHROW_LOCK(*sh);
	LIST_FOREACH(n, &(*sh)->nodes, entry)
		if (n->rule == rule && n->af == af && n->type == sn_type &&
		    ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
		    (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
			break;

	if (n == NULL && !returnlocked)
		PF_HASHROW_UNLOCK(*sh);

	return (n);
}

bool
pf_src_node_exists(struct pf_ksrc_node **sn, struct pf_srchash *sh)
{
	struct pf_ksrc_node	*cur;

	if ((*sn) == NULL)
		return (false);

	KASSERT(sh != NULL, ("%s: sh is NULL", __func__));

	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
	PF_HASHROW_LOCK(sh);
	LIST_FOREACH(cur, &(sh->nodes), entry) {
		if (cur == (*sn) &&
		    cur->expire != 1) /* Ignore nodes being killed */
			return (true);
	}
	PF_HASHROW_UNLOCK(sh);
	(*sn) = NULL;
	return (false);
}

static void
pf_free_src_node(struct pf_ksrc_node *sn)
{

	for (int i = 0; i < 2; i++) {
		counter_u64_free(sn->bytes[i]);
		counter_u64_free(sn->packets[i]);
	}
	counter_rate_free(sn->conn_rate.cr);
	uma_zfree(V_pf_sources_z, sn);
}

static u_short
pf_insert_src_node(struct pf_ksrc_node *sns[PF_SN_MAX],
    struct pf_srchash *snhs[PF_SN_MAX], struct pf_krule *rule,
    struct pf_addr *src, sa_family_t af, struct pf_addr *raddr,
    struct pfi_kkif *rkif, sa_family_t raf, pf_sn_types_t sn_type)
{
	u_short			 reason = 0;
	struct pf_krule		*r_track = rule;
	struct pf_ksrc_node	**sn = &(sns[sn_type]);
	struct pf_srchash	**sh = &(snhs[sn_type]);

	KASSERT(sn_type != PF_SN_LIMIT || (raddr == NULL && rkif == NULL),
	    ("%s: raddr and rkif must be NULL for PF_SN_LIMIT", __func__));

	KASSERT(sn_type != PF_SN_LIMIT || (rule->rule_flag & PFRULE_SRCTRACK),
	    ("%s: PF_SN_LIMIT only valid for rules with PFRULE_SRCTRACK", __func__));

	/*
	 * XXX: There could be a KASSERT for
	 * sn_type == PF_SN_LIMIT || (pool->opts & PF_POOL_STICKYADDR)
	 * but we'd need to pass pool *only* for this KASSERT.
	 */

	if ( (rule->rule_flag & PFRULE_SRCTRACK) &&
	    !(rule->rule_flag & PFRULE_RULESRCTRACK))
		r_track = &V_pf_default_rule;

	/*
	 * Request the sh to always be locked, as we might insert a new sn.
	 */
	if (*sn == NULL)
		*sn = pf_find_src_node(src, r_track, af, sh, sn_type, true);

	if (*sn == NULL) {
		PF_HASHROW_ASSERT(*sh);

		if (sn_type == PF_SN_LIMIT && rule->max_src_nodes &&
		    counter_u64_fetch(r_track->src_nodes[sn_type]) >= rule->max_src_nodes) {
			counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1);
			reason = PFRES_SRCLIMIT;
			goto done;
		}

		(*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
		if ((*sn) == NULL) {
			reason = PFRES_MEMORY;
			goto done;
		}

		for (int i = 0; i < 2; i++) {
			(*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT);
			(*sn)->packets[i] = counter_u64_alloc(M_NOWAIT);

			if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) {
				pf_free_src_node(*sn);
				reason = PFRES_MEMORY;
				goto done;
			}
		}

		if (sn_type == PF_SN_LIMIT)
			if (! pf_init_threshold(&(*sn)->conn_rate,
			    rule->max_src_conn_rate.limit,
			    rule->max_src_conn_rate.seconds)) {
				pf_free_src_node(*sn);
				reason = PFRES_MEMORY;
				goto done;
			}

		MPASS((*sn)->lock == NULL);
		(*sn)->lock = &(*sh)->lock;

		(*sn)->af = af;
		(*sn)->rule = r_track;
		pf_addrcpy(&(*sn)->addr, src, af);
		if (raddr != NULL)
			pf_addrcpy(&(*sn)->raddr, raddr, raf);
		(*sn)->rkif = rkif;
		(*sn)->raf = raf;
		LIST_INSERT_HEAD(&(*sh)->nodes, *sn, entry);
		(*sn)->creation = time_uptime;
		(*sn)->ruletype = rule->action;
		(*sn)->type = sn_type;
		counter_u64_add(r_track->src_nodes[sn_type], 1);
		counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
	} else {
		if (sn_type == PF_SN_LIMIT && rule->max_src_states &&
		    (*sn)->states >= rule->max_src_states) {
			counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
			    1);
			reason = PFRES_SRCLIMIT;
			goto done;
		}
	}
done:
	if (reason == 0)
		(*sn)->states++;
	else
		(*sn) = NULL;

	PF_HASHROW_UNLOCK(*sh);
	return (reason);
}

void
pf_unlink_src_node(struct pf_ksrc_node *src)
{
	PF_SRC_NODE_LOCK_ASSERT(src);

	LIST_REMOVE(src, entry);
	if (src->rule)
		counter_u64_add(src->rule->src_nodes[src->type], -1);
}

u_int
pf_free_src_nodes(struct pf_ksrc_node_list *head)
{
	struct pf_ksrc_node *sn, *tmp;
	u_int count = 0;

	LIST_FOREACH_SAFE(sn, head, entry, tmp) {
		pf_free_src_node(sn);
		count++;
	}

	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);

	return (count);
}

void
pf_mtag_initialize(void)
{

	pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
	    sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
	    UMA_ALIGN_PTR, 0);
}

/* Per-vnet data storage structures initialization. */
void
pf_initialize(void)
{
	struct pf_keyhash	*kh;
	struct pf_idhash	*ih;
	struct pf_srchash	*sh;
	struct pf_udpendpointhash	*uh;
	u_int i;

	if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize))
		V_pf_hashsize = PF_HASHSIZ;
	if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize))
		V_pf_srchashsize = PF_SRCHASHSIZ;
	if (V_pf_udpendpointhashsize == 0 || !powerof2(V_pf_udpendpointhashsize))
		V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ;

	V_pf_hashseed = arc4random();

	/* States and state keys storage. */
	V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate),
	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
	V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
	uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
	uma_zone_set_warning(V_pf_state_z, "PF states limit reached");

	V_pf_state_key_z = uma_zcreate("pf state keys",
	    sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
	    UMA_ALIGN_PTR, 0);

	V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash),
	    M_PFHASH, M_NOWAIT | M_ZERO);
	V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash),
	    M_PFHASH, M_NOWAIT | M_ZERO);
	if (V_pf_keyhash == NULL || V_pf_idhash == NULL) {
		printf("pf: Unable to allocate memory for "
		    "state_hashsize %lu.\n", V_pf_hashsize);

		free(V_pf_keyhash, M_PFHASH);
		free(V_pf_idhash, M_PFHASH);

		V_pf_hashsize = PF_HASHSIZ;
		V_pf_keyhash = mallocarray(V_pf_hashsize,
		    sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO);
		V_pf_idhash = mallocarray(V_pf_hashsize,
		    sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO);
	}

	V_pf_hashmask = V_pf_hashsize - 1;
	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
	    i++, kh++, ih++) {
		mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
		mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
	}

	/* Source nodes. */
	V_pf_sources_z = uma_zcreate("pf source nodes",
	    sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
	    0);
	V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
	uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
	uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");

	V_pf_srchash = mallocarray(V_pf_srchashsize,
	    sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO);
	if (V_pf_srchash == NULL) {
		printf("pf: Unable to allocate memory for "
		    "source_hashsize %lu.\n", V_pf_srchashsize);

		V_pf_srchashsize = PF_SRCHASHSIZ;
		V_pf_srchash = mallocarray(V_pf_srchashsize,
		    sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO);
	}

	V_pf_srchashmask = V_pf_srchashsize - 1;
	for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++)
		mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);


	/* UDP endpoint mappings. */
	V_pf_udp_mapping_z = uma_zcreate("pf UDP mappings",
	    sizeof(struct pf_udp_mapping), NULL, NULL, NULL, NULL,
	    UMA_ALIGN_PTR, 0);
	V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize,
	    sizeof(struct pf_udpendpointhash), M_PFHASH, M_NOWAIT | M_ZERO);
	if (V_pf_udpendpointhash == NULL) {
		printf("pf: Unable to allocate memory for "
		    "udpendpoint_hashsize %lu.\n", V_pf_udpendpointhashsize);

		V_pf_udpendpointhashsize = PF_UDPENDHASHSIZ;
		V_pf_udpendpointhash = mallocarray(V_pf_udpendpointhashsize,
		    sizeof(struct pf_udpendpointhash), M_PFHASH, M_WAITOK | M_ZERO);
	}

	V_pf_udpendpointhashmask = V_pf_udpendpointhashsize - 1;
	for (i = 0, uh = V_pf_udpendpointhash;
	    i <= V_pf_udpendpointhashmask;
	    i++, uh++) {
		mtx_init(&uh->lock, "pf_udpendpointhash", NULL,
		    MTX_DEF | MTX_DUPOK);
	}

	/* Anchors */
	V_pf_anchor_z = uma_zcreate("pf anchors",
	    sizeof(struct pf_kanchor), NULL, NULL, NULL, NULL,
	    UMA_ALIGN_PTR, 0);
	V_pf_limits[PF_LIMIT_ANCHORS].zone = V_pf_anchor_z;
	uma_zone_set_max(V_pf_anchor_z, PF_ANCHOR_HIWAT);
	uma_zone_set_warning(V_pf_anchor_z, "PF anchor limit reached");

	V_pf_eth_anchor_z = uma_zcreate("pf Ethernet anchors",
	    sizeof(struct pf_keth_anchor), NULL, NULL, NULL, NULL,
	    UMA_ALIGN_PTR, 0);
	V_pf_limits[PF_LIMIT_ETH_ANCHORS].zone = V_pf_eth_anchor_z;
	uma_zone_set_max(V_pf_eth_anchor_z, PF_ANCHOR_HIWAT);
	uma_zone_set_warning(V_pf_eth_anchor_z, "PF Ethernet anchor limit reached");

	/* ALTQ */
	TAILQ_INIT(&V_pf_altqs[0]);
	TAILQ_INIT(&V_pf_altqs[1]);
	TAILQ_INIT(&V_pf_altqs[2]);
	TAILQ_INIT(&V_pf_altqs[3]);
	TAILQ_INIT(&V_pf_pabuf[0]);
	TAILQ_INIT(&V_pf_pabuf[1]);
	TAILQ_INIT(&V_pf_pabuf[2]);
	V_pf_altqs_active = &V_pf_altqs[0];
	V_pf_altq_ifs_active = &V_pf_altqs[1];
	V_pf_altqs_inactive = &V_pf_altqs[2];
	V_pf_altq_ifs_inactive = &V_pf_altqs[3];

	/* Send & overload+flush queues. */
	STAILQ_INIT(&V_pf_sendqueue);
	SLIST_INIT(&V_pf_overloadqueue);
	TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);

	/* Unlinked, but may be referenced rules. */
	TAILQ_INIT(&V_pf_unlinked_rules);
}

void
pf_mtag_cleanup(void)
{

	uma_zdestroy(pf_mtag_z);
}

void
pf_cleanup(void)
{
	struct pf_keyhash	*kh;
	struct pf_idhash	*ih;
	struct pf_srchash	*sh;
	struct pf_udpendpointhash	*uh;
	struct pf_send_entry	*pfse, *next;
	u_int i;

	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash;
	    i <= V_pf_hashmask;
	    i++, kh++, ih++) {
		KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
		    __func__));
		KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
		    __func__));
		mtx_destroy(&kh->lock);
		mtx_destroy(&ih->lock);
	}
	free(V_pf_keyhash, M_PFHASH);
	free(V_pf_idhash, M_PFHASH);

	for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
		KASSERT(LIST_EMPTY(&sh->nodes),
		    ("%s: source node hash not empty", __func__));
		mtx_destroy(&sh->lock);
	}
	free(V_pf_srchash, M_PFHASH);

	for (i = 0, uh = V_pf_udpendpointhash;
	    i <= V_pf_udpendpointhashmask;
	    i++, uh++) {
		KASSERT(LIST_EMPTY(&uh->endpoints),
		    ("%s: udp endpoint hash not empty", __func__));
		mtx_destroy(&uh->lock);
	}
	free(V_pf_udpendpointhash, M_PFHASH);

	STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
		m_freem(pfse->pfse_m);
		free(pfse, M_PFTEMP);
	}
	MPASS(RB_EMPTY(&V_pf_sctp_endpoints));

	uma_zdestroy(V_pf_sources_z);
	uma_zdestroy(V_pf_state_z);
	uma_zdestroy(V_pf_state_key_z);
	uma_zdestroy(V_pf_udp_mapping_z);
	uma_zdestroy(V_pf_anchor_z);
	uma_zdestroy(V_pf_eth_anchor_z);
}

static int
pf_mtag_uminit(void *mem, int size, int how)
{
	struct m_tag *t;

	t = (struct m_tag *)mem;
	t->m_tag_cookie = MTAG_ABI_COMPAT;
	t->m_tag_id = PACKET_TAG_PF;
	t->m_tag_len = sizeof(struct pf_mtag);
	t->m_tag_free = pf_mtag_free;

	return (0);
}

static void
pf_mtag_free(struct m_tag *t)
{

	uma_zfree(pf_mtag_z, t);
}

struct pf_mtag *
pf_get_mtag(struct mbuf *m)
{
	struct m_tag *mtag;

	if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
		return ((struct pf_mtag *)(mtag + 1));

	mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
	if (mtag == NULL)
		return (NULL);
	bzero(mtag + 1, sizeof(struct pf_mtag));
	m_tag_prepend(m, mtag);

	return ((struct pf_mtag *)(mtag + 1));
}

static int
pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
    struct pf_kstate *s)
{
	struct pf_keyhash	*khs, *khw, *kh;
	struct pf_state_key	*sk, *cur;
	struct pf_kstate	*si, *olds = NULL;
	int idx;

	NET_EPOCH_ASSERT();
	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
	KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
	KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));

	/*
	 * We need to lock hash slots of both keys. To avoid deadlock
	 * we always lock the slot with lower address first. Unlock order
	 * isn't important.
	 *
	 * We also need to lock ID hash slot before dropping key
	 * locks. On success we return with ID hash slot locked.
	 */

	if (skw == sks) {
		khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
		PF_HASHROW_LOCK(khs);
	} else {
		khs = &V_pf_keyhash[pf_hashkey(sks)];
		khw = &V_pf_keyhash[pf_hashkey(skw)];
		if (khs == khw) {
			PF_HASHROW_LOCK(khs);
		} else if (khs < khw) {
			PF_HASHROW_LOCK(khs);
			PF_HASHROW_LOCK(khw);
		} else {
			PF_HASHROW_LOCK(khw);
			PF_HASHROW_LOCK(khs);
		}
	}

#define	KEYS_UNLOCK()	do {			\
	if (khs != khw) {			\
		PF_HASHROW_UNLOCK(khs);		\
		PF_HASHROW_UNLOCK(khw);		\
	} else					\
		PF_HASHROW_UNLOCK(khs);		\
} while (0)

	/*
	 * First run: start with wire key.
	 */
	sk = skw;
	kh = khw;
	idx = PF_SK_WIRE;

	MPASS(s->lock == NULL);
	s->lock = &V_pf_idhash[PF_IDHASH(s)].lock;

keyattach:
	LIST_FOREACH(cur, &kh->keys, entry)
		if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
			break;

	if (cur != NULL) {
		/* Key exists. Check for same kif, if none, add to key. */
		TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
			struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];

			PF_HASHROW_LOCK(ih);
			if (si->kif == s->kif &&
			    ((si->key[PF_SK_WIRE]->af == sk->af &&
			    si->direction == s->direction) ||
			    (si->key[PF_SK_WIRE]->af !=
			    si->key[PF_SK_STACK]->af &&
			    sk->af == si->key[PF_SK_STACK]->af &&
			    si->direction != s->direction))) {
				bool reuse = false;

				if (sk->proto == IPPROTO_TCP &&
				    si->src.state >= TCPS_FIN_WAIT_2 &&
				    si->dst.state >= TCPS_FIN_WAIT_2)
					reuse = true;

				if (V_pf_status.debug >= PF_DEBUG_MISC) {
					printf("pf: %s key attach "
					    "%s on %s: ",
					    (idx == PF_SK_WIRE) ?
					    "wire" : "stack",
					    reuse ? "reuse" : "failed",
					    s->kif->pfik_name);
					pf_print_state_parts(s,
					    (idx == PF_SK_WIRE) ?
					    sk : NULL,
					    (idx == PF_SK_STACK) ?
					    sk : NULL);
					printf(", existing: ");
					pf_print_state_parts(si,
					    (idx == PF_SK_WIRE) ?
					    sk : NULL,
					    (idx == PF_SK_STACK) ?
					    sk : NULL);
					printf("\n");
				}

				if (reuse) {
					/*
					 * New state matches an old >FIN_WAIT_2
					 * state. We can't drop key hash locks,
					 * thus we can't unlink it properly.
					 *
					 * As a workaround we drop it into
					 * TCPS_CLOSED state, schedule purge
					 * ASAP and push it into the very end
					 * of the slot TAILQ, so that it won't
					 * conflict with our new state.
					 */
					pf_set_protostate(si, PF_PEER_BOTH,
					    TCPS_CLOSED);
					si->timeout = PFTM_PURGE;
					olds = si;
				} else {
					s->timeout = PFTM_UNLINKED;
					if (idx == PF_SK_STACK)
						/*
						 * Remove the wire key from
						 * the hash. Other threads
						 * can't be referencing it
						 * because we still hold the
						 * hash lock.
						 */
						pf_state_key_detach(s,
						    PF_SK_WIRE);
					PF_HASHROW_UNLOCK(ih);
					KEYS_UNLOCK();
					if (idx == PF_SK_WIRE)
						/*
						 * We've not inserted either key.
						 * Free both.
						 */
						uma_zfree(V_pf_state_key_z, skw);
					if (skw != sks)
						uma_zfree(
						    V_pf_state_key_z,
						    sks);
					return (EEXIST); /* collision! */
				}
			}
			PF_HASHROW_UNLOCK(ih);
		}
		uma_zfree(V_pf_state_key_z, sk);
		s->key[idx] = cur;
	} else {
		LIST_INSERT_HEAD(&kh->keys, sk, entry);
		s->key[idx] = sk;
	}

stateattach:
	/* List is sorted, if-bound states before floating. */
	if (s->kif == V_pfi_all)
		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
	else
		TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);

	if (olds) {
		TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
		    key_list[idx]);
		olds = NULL;
	}

	/*
	 * Attach done. See how should we (or should not?)
	 * attach a second key.
	 */
	if (sks == skw) {
		s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
		idx = PF_SK_STACK;
		sks = NULL;
		goto stateattach;
	} else if (sks != NULL) {
		/*
		 * Continue attaching with stack key.
		 */
		sk = sks;
		kh = khs;
		idx = PF_SK_STACK;
		sks = NULL;
		goto keyattach;
	}

	PF_STATE_LOCK(s);
	KEYS_UNLOCK();

	KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
	    ("%s failure", __func__));

	return (0);
#undef	KEYS_UNLOCK
}

static void
pf_detach_state(struct pf_kstate *s)
{
	struct pf_state_key *sks = s->key[PF_SK_STACK];
	struct pf_keyhash *kh;

	NET_EPOCH_ASSERT();
	MPASS(s->timeout >= PFTM_MAX);

	pf_sctp_multihome_detach_addr(s);

	if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr)
		V_pflow_export_state_ptr(s);

	if (sks != NULL) {
		kh = &V_pf_keyhash[pf_hashkey(sks)];
		PF_HASHROW_LOCK(kh);
		if (s->key[PF_SK_STACK] != NULL)
			pf_state_key_detach(s, PF_SK_STACK);
		/*
		 * If both point to same key, then we are done.
		 */
		if (sks == s->key[PF_SK_WIRE]) {
			pf_state_key_detach(s, PF_SK_WIRE);
			PF_HASHROW_UNLOCK(kh);
			return;
		}
		PF_HASHROW_UNLOCK(kh);
	}

	if (s->key[PF_SK_WIRE] != NULL) {
		kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
		PF_HASHROW_LOCK(kh);
		if (s->key[PF_SK_WIRE] != NULL)
			pf_state_key_detach(s, PF_SK_WIRE);
		PF_HASHROW_UNLOCK(kh);
	}
}

static void
pf_state_key_detach(struct pf_kstate *s, int idx)
{
	struct pf_state_key *sk = s->key[idx];
#ifdef INVARIANTS
	struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];

	PF_HASHROW_ASSERT(kh);
#endif /* INVARIANTS */
	TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
	s->key[idx] = NULL;

	if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
		LIST_REMOVE(sk, entry);
		uma_zfree(V_pf_state_key_z, sk);
	}
}

static int
pf_state_key_ctor(void *mem, int size, void *arg, int flags)
{
	struct pf_state_key *sk = mem;

	bzero(sk, sizeof(struct pf_state_key_cmp));
	TAILQ_INIT(&sk->states[PF_SK_WIRE]);
	TAILQ_INIT(&sk->states[PF_SK_STACK]);

	return (0);
}

static int
pf_state_key_addr_setup(struct pf_pdesc *pd,
    struct pf_state_key_cmp *key, int multi)
{
	struct pf_addr *saddr = pd->src;
	struct pf_addr *daddr = pd->dst;
#ifdef INET6
	struct nd_neighbor_solicit nd;
	struct pf_addr *target;

	if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6)
		goto copy;

	switch (pd->hdr.icmp6.icmp6_type) {
	case ND_NEIGHBOR_SOLICIT:
		if (multi)
			return (-1);
		if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), NULL,
		    NULL, pd->af))
			return (-1);
		target = (struct pf_addr *)&nd.nd_ns_target;
		daddr = target;
		break;
	case ND_NEIGHBOR_ADVERT:
		if (multi)
			return (-1);
		if (!pf_pull_hdr(pd->m, pd->off, &nd, sizeof(nd), NULL,
		    NULL, pd->af))
			return (-1);
		target = (struct pf_addr *)&nd.nd_ns_target;
		saddr = target;
		if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) {
			key->addr[pd->didx].addr32[0] = 0;
			key->addr[pd->didx].addr32[1] = 0;
			key->addr[pd->didx].addr32[2] = 0;
			key->addr[pd->didx].addr32[3] = 0;
			daddr = NULL; /* overwritten */
		}
		break;
	default:
		if (multi) {
			key->addr[pd->sidx].addr32[0] = IPV6_ADDR_INT32_MLL;
			key->addr[pd->sidx].addr32[1] = 0;
			key->addr[pd->sidx].addr32[2] = 0;
			key->addr[pd->sidx].addr32[3] = IPV6_ADDR_INT32_ONE;
			saddr = NULL; /* overwritten */
		}
	}
copy:
#endif /* INET6 */
	if (saddr)
		pf_addrcpy(&key->addr[pd->sidx], saddr, pd->af);
	if (daddr)
		pf_addrcpy(&key->addr[pd->didx], daddr, pd->af);

	return (0);
}

int
pf_state_key_setup(struct pf_pdesc *pd, u_int16_t sport, u_int16_t dport,
    struct pf_state_key **sk, struct pf_state_key **nk)
{
	*sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
	if (*sk == NULL)
		return (ENOMEM);

	if (pf_state_key_addr_setup(pd, (struct pf_state_key_cmp *)*sk,
	    0)) {
		uma_zfree(V_pf_state_key_z, *sk);
		*sk = NULL;
		return (ENOMEM);
	}

	(*sk)->port[pd->sidx] = sport;
	(*sk)->port[pd->didx] = dport;
	(*sk)->proto = pd->proto;
	(*sk)->af = pd->af;

	*nk = pf_state_key_clone(*sk);
	if (*nk == NULL) {
		uma_zfree(V_pf_state_key_z, *sk);
		*sk = NULL;
		return (ENOMEM);
	}

	if (pd->af != pd->naf) {
		(*sk)->port[pd->sidx] = pd->osport;
		(*sk)->port[pd->didx] = pd->odport;

		(*nk)->af = pd->naf;

		/*
		 * We're overwriting an address here, so potentially there's bits of an IPv6
		 * address left in here. Clear that out first.
		 */
		bzero(&(*nk)->addr[0], sizeof((*nk)->addr[0]));
		bzero(&(*nk)->addr[1], sizeof((*nk)->addr[1]));
		if (pd->dir == PF_IN) {
			pf_addrcpy(&(*nk)->addr[pd->didx], &pd->nsaddr,
			    pd->naf);
			pf_addrcpy(&(*nk)->addr[pd->sidx], &pd->ndaddr,
			    pd->naf);
			(*nk)->port[pd->didx] = pd->nsport;
			(*nk)->port[pd->sidx] = pd->ndport;
		} else {
			pf_addrcpy(&(*nk)->addr[pd->sidx], &pd->nsaddr,
			    pd->naf);
			pf_addrcpy(&(*nk)->addr[pd->didx], &pd->ndaddr,
			    pd->naf);
			(*nk)->port[pd->sidx] = pd->nsport;
			(*nk)->port[pd->didx] = pd->ndport;
		}

		switch (pd->proto) {
		case IPPROTO_ICMP:
			(*nk)->proto = IPPROTO_ICMPV6;
			break;
		case IPPROTO_ICMPV6:
			(*nk)->proto = IPPROTO_ICMP;
			break;
		default:
			(*nk)->proto = pd->proto;
		}
	}

	return (0);
}

struct pf_state_key *
pf_state_key_clone(const struct pf_state_key *orig)
{
	struct pf_state_key *sk;

	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
	if (sk == NULL)
		return (NULL);

	bcopy(orig, sk, sizeof(struct pf_state_key_cmp));

	return (sk);
}

int
pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif,
    struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s)
{
	struct pf_idhash *ih;
	struct pf_kstate *cur;
	int error;

	NET_EPOCH_ASSERT();

	KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
	    ("%s: sks not pristine", __func__));
	KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
	    ("%s: skw not pristine", __func__));
	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));

	s->kif = kif;
	s->orig_kif = orig_kif;

	if (s->id == 0 && s->creatorid == 0) {
		s->id = alloc_unr64(&V_pf_stateid);
		s->id = htobe64(s->id);
		s->creatorid = V_pf_status.hostid;
	}

	/* Returns with ID locked on success. */
	if ((error = pf_state_key_attach(skw, sks, s)) != 0)
		return (error);
	skw = sks = NULL;

	ih = &V_pf_idhash[PF_IDHASH(s)];
	PF_HASHROW_ASSERT(ih);
	LIST_FOREACH(cur, &ih->states, entry)
		if (cur->id == s->id && cur->creatorid == s->creatorid)
			break;

	if (cur != NULL) {
		s->timeout = PFTM_UNLINKED;
		PF_HASHROW_UNLOCK(ih);
		if (V_pf_status.debug >= PF_DEBUG_MISC) {
			printf("pf: state ID collision: "
			    "id: %016llx creatorid: %08x\n",
			    (unsigned long long)be64toh(s->id),
			    ntohl(s->creatorid));
		}
		pf_detach_state(s);
		return (EEXIST);
	}
	LIST_INSERT_HEAD(&ih->states, s, entry);
	/* One for keys, one for ID hash. */
	refcount_init(&s->refs, 2);

	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
	if (V_pfsync_insert_state_ptr != NULL)
		V_pfsync_insert_state_ptr(s);

	/* Returns locked. */
	return (0);
}

/*
 * Find state by ID: returns with locked row on success.
 */
struct pf_kstate *
pf_find_state_byid(uint64_t id, uint32_t creatorid)
{
	struct pf_idhash *ih;
	struct pf_kstate *s;

	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);

	ih = &V_pf_idhash[PF_IDHASHID(id)];

	PF_HASHROW_LOCK(ih);
	LIST_FOREACH(s, &ih->states, entry)
		if (s->id == id && s->creatorid == creatorid)
			break;

	if (s == NULL)
		PF_HASHROW_UNLOCK(ih);

	return (s);
}

/*
 * Find state by key.
 * Returns with ID hash slot locked on success.
 */
static int
pf_find_state(struct pf_pdesc *pd, const struct pf_state_key_cmp *key,
    struct pf_kstate **state)
{
	struct pf_keyhash	*kh;
	struct pf_state_key	*sk;
	struct pf_kstate	*s;
	int idx;

	*state = NULL;

	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);

	kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)];

	PF_HASHROW_LOCK(kh);
	LIST_FOREACH(sk, &kh->keys, entry)
		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
			break;
	if (sk == NULL) {
		PF_HASHROW_UNLOCK(kh);
		return (PF_DROP);
	}

	idx = (pd->dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);

	/* List is sorted, if-bound states before floating ones. */
	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
		if (s->kif == V_pfi_all || s->kif == pd->kif ||
		    s->orig_kif == pd->kif) {
			PF_STATE_LOCK(s);
			PF_HASHROW_UNLOCK(kh);
			if (__predict_false(s->timeout >= PFTM_MAX)) {
				/*
				 * State is either being processed by
				 * pf_remove_state() in an other thread, or
				 * is scheduled for immediate expiry.
				 */
				PF_STATE_UNLOCK(s);
				SDT_PROBE5(pf, ip, state, lookup, pd->kif,
				    key, (pd->dir), pd, *state);
				return (PF_DROP);
			}
			goto out;
		}

	/* Look through the other list, in case of AF-TO */
	idx = idx == PF_SK_WIRE ? PF_SK_STACK : PF_SK_WIRE;
	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
		if (s->key[PF_SK_WIRE]->af == s->key[PF_SK_STACK]->af)
			continue;
		if (s->kif == V_pfi_all || s->kif == pd->kif ||
		    s->orig_kif == pd->kif) {
			PF_STATE_LOCK(s);
			PF_HASHROW_UNLOCK(kh);
			if (__predict_false(s->timeout >= PFTM_MAX)) {
				/*
				 * State is either being processed by
				 * pf_remove_state() in an other thread, or
				 * is scheduled for immediate expiry.
				 */
				PF_STATE_UNLOCK(s);
				SDT_PROBE5(pf, ip, state, lookup, pd->kif,
				    key, (pd->dir), pd, NULL);
				return (PF_DROP);
			}
			goto out;
		}
	}

	PF_HASHROW_UNLOCK(kh);

out:
	SDT_PROBE5(pf, ip, state, lookup, pd->kif, key, (pd->dir), pd, *state);

	if (s == NULL || s->timeout == PFTM_PURGE) {
		if (s)
			PF_STATE_UNLOCK(s);
		return (PF_DROP);
	}

	if ((s)->rule->pktrate.limit && pd->dir == (s)->direction) {
		if (pf_check_threshold(&(s)->rule->pktrate)) {
			PF_STATE_UNLOCK(s);
			return (PF_DROP);
		}
	}
	if (PACKET_LOOPED(pd)) {
		PF_STATE_UNLOCK(s);
		return (PF_PASS);
	}

	*state = s;

	return (PF_MATCH);
}

/*
 * Returns with ID hash slot locked on success.
 */
struct pf_kstate *
pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more)
{
	struct pf_keyhash	*kh;
	struct pf_state_key	*sk;
	struct pf_kstate	*s, *ret = NULL;
	int			 idx, inout = 0;

	if (more != NULL)
		*more = 0;

	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);

	kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)];

	PF_HASHROW_LOCK(kh);
	LIST_FOREACH(sk, &kh->keys, entry)
		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
			break;
	if (sk == NULL) {
		PF_HASHROW_UNLOCK(kh);
		return (NULL);
	}
	switch (dir) {
	case PF_IN:
		idx = PF_SK_WIRE;
		break;
	case PF_OUT:
		idx = PF_SK_STACK;
		break;
	case PF_INOUT:
		idx = PF_SK_WIRE;
		inout = 1;
		break;
	default:
		panic("%s: dir %u", __func__, dir);
	}
second_run:
	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
		if (more == NULL) {
			PF_STATE_LOCK(s);
			PF_HASHROW_UNLOCK(kh);
			return (s);
		}

		if (ret)
			(*more)++;
		else {
			ret = s;
			PF_STATE_LOCK(s);
		}
	}
	if (inout == 1) {
		inout = 0;
		idx = PF_SK_STACK;
		goto second_run;
	}
	PF_HASHROW_UNLOCK(kh);

	return (ret);
}

/*
 * FIXME
 * This routine is inefficient -- locks the state only to unlock immediately on
 * return.
 * It is racy -- after the state is unlocked nothing stops other threads from
 * removing it.
 */
bool
pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir)
{
	struct pf_kstate *s;

	s = pf_find_state_all(key, dir, NULL);
	if (s != NULL) {
		PF_STATE_UNLOCK(s);
		return (true);
	}
	return (false);
}

void
pf_state_peer_hton(const struct pf_state_peer *s, struct pf_state_peer_export *d)
{
	d->seqlo = htonl(s->seqlo);
	d->seqhi = htonl(s->seqhi);
	d->seqdiff = htonl(s->seqdiff);
	d->max_win = htons(s->max_win);
	d->mss = htons(s->mss);
	d->state = s->state;
	d->wscale = s->wscale;
	if (s->scrub) {
		d->scrub.pfss_flags = htons(
		    s->scrub->pfss_flags & PFSS_TIMESTAMP);
		d->scrub.pfss_ttl = (s)->scrub->pfss_ttl;
		d->scrub.pfss_ts_mod = htonl((s)->scrub->pfss_ts_mod);
		d->scrub.scrub_flag = PF_SCRUB_FLAG_VALID;
	}
}

void
pf_state_peer_ntoh(const struct pf_state_peer_export *s, struct pf_state_peer *d)
{
	d->seqlo = ntohl(s->seqlo);
	d->seqhi = ntohl(s->seqhi);
	d->seqdiff = ntohl(s->seqdiff);
	d->max_win = ntohs(s->max_win);
	d->mss = ntohs(s->mss);
	d->state = s->state;
	d->wscale = s->wscale;
	if (s->scrub.scrub_flag == PF_SCRUB_FLAG_VALID &&
	    d->scrub != NULL) {
		d->scrub->pfss_flags = ntohs(s->scrub.pfss_flags) &
		    PFSS_TIMESTAMP;
		d->scrub->pfss_ttl = s->scrub.pfss_ttl;
		d->scrub->pfss_ts_mod = ntohl(s->scrub.pfss_ts_mod);
	}
}

struct pf_udp_mapping *
pf_udp_mapping_create(sa_family_t af, struct pf_addr *src_addr, uint16_t src_port,
    struct pf_addr *nat_addr, uint16_t nat_port)
{
	struct pf_udp_mapping *mapping;

	mapping = uma_zalloc(V_pf_udp_mapping_z, M_NOWAIT | M_ZERO);
	if (mapping == NULL)
		return (NULL);
	pf_addrcpy(&mapping->endpoints[0].addr, src_addr, af);
	mapping->endpoints[0].port = src_port;
	mapping->endpoints[0].af = af;
	mapping->endpoints[0].mapping = mapping;
	pf_addrcpy(&mapping->endpoints[1].addr, nat_addr, af);
	mapping->endpoints[1].port = nat_port;
	mapping->endpoints[1].af = af;
	mapping->endpoints[1].mapping = mapping;
	refcount_init(&mapping->refs, 1);
	return (mapping);
}

int
pf_udp_mapping_insert(struct pf_udp_mapping *mapping)
{
	struct pf_udpendpointhash *h0, *h1;
	struct pf_udp_endpoint *endpoint;
	int ret = EEXIST;

	h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])];
	h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])];
	if (h0 == h1) {
		PF_HASHROW_LOCK(h0);
	} else if (h0 < h1) {
		PF_HASHROW_LOCK(h0);
		PF_HASHROW_LOCK(h1);
	} else {
		PF_HASHROW_LOCK(h1);
		PF_HASHROW_LOCK(h0);
	}

	LIST_FOREACH(endpoint, &h0->endpoints, entry) {
		if (bcmp(endpoint, &mapping->endpoints[0],
		    sizeof(struct pf_udp_endpoint_cmp)) == 0)
			break;
	}
	if (endpoint != NULL)
		goto cleanup;
	LIST_FOREACH(endpoint, &h1->endpoints, entry) {
		if (bcmp(endpoint, &mapping->endpoints[1],
		    sizeof(struct pf_udp_endpoint_cmp)) == 0)
			break;
	}
	if (endpoint != NULL)
		goto cleanup;
	LIST_INSERT_HEAD(&h0->endpoints, &mapping->endpoints[0], entry);
	LIST_INSERT_HEAD(&h1->endpoints, &mapping->endpoints[1], entry);
	ret = 0;

cleanup:
	if (h0 != h1) {
		PF_HASHROW_UNLOCK(h0);
		PF_HASHROW_UNLOCK(h1);
	} else {
		PF_HASHROW_UNLOCK(h0);
	}
	return (ret);
}

void
pf_udp_mapping_release(struct pf_udp_mapping *mapping)
{
	/* refcount is synchronized on the source endpoint's row lock */
	struct pf_udpendpointhash *h0, *h1;

	if (mapping == NULL)
		return;

	h0 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[0])];
	PF_HASHROW_LOCK(h0);
	if (refcount_release(&mapping->refs)) {
		LIST_REMOVE(&mapping->endpoints[0], entry);
		PF_HASHROW_UNLOCK(h0);
		h1 = &V_pf_udpendpointhash[pf_hashudpendpoint(&mapping->endpoints[1])];
		PF_HASHROW_LOCK(h1);
		LIST_REMOVE(&mapping->endpoints[1], entry);
		PF_HASHROW_UNLOCK(h1);

		uma_zfree(V_pf_udp_mapping_z, mapping);
	} else {
			PF_HASHROW_UNLOCK(h0);
	}
}


struct pf_udp_mapping *
pf_udp_mapping_find(struct pf_udp_endpoint_cmp *key)
{
	struct pf_udpendpointhash *uh;
	struct pf_udp_endpoint *endpoint;

	uh = &V_pf_udpendpointhash[pf_hashudpendpoint((struct pf_udp_endpoint*)key)];

	PF_HASHROW_LOCK(uh);
	LIST_FOREACH(endpoint, &uh->endpoints, entry) {
		if (bcmp(endpoint, key, sizeof(struct pf_udp_endpoint_cmp)) == 0 &&
			bcmp(endpoint, &endpoint->mapping->endpoints[0],
			    sizeof(struct pf_udp_endpoint_cmp)) == 0)
			break;
	}
	if (endpoint == NULL) {
		PF_HASHROW_UNLOCK(uh);
		return (NULL);
	}
	refcount_acquire(&endpoint->mapping->refs);
	PF_HASHROW_UNLOCK(uh);
	return (endpoint->mapping);
}
/* END state table stuff */

static void
pf_send(struct pf_send_entry *pfse)
{

	PF_SENDQ_LOCK();
	STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
	PF_SENDQ_UNLOCK();
	swi_sched(V_pf_swi_cookie, 0);
}

static bool
pf_isforlocal(struct mbuf *m, int af)
{
	switch (af) {
#ifdef INET
	case AF_INET: {
		struct ip *ip = mtod(m, struct ip *);

		return (in_localip(ip->ip_dst));
	}
#endif /* INET */
#ifdef INET6
	case AF_INET6: {
		struct ip6_hdr *ip6;
		struct in6_ifaddr *ia;
		ip6 = mtod(m, struct ip6_hdr *);
		ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
		if (ia == NULL)
			return (false);
		return (! (ia->ia6_flags & IN6_IFF_NOTREADY));
	}
#endif /* INET6 */
	default:
		unhandled_af(af);
	}

	return (false);
}

int
pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type,
    int *icmp_dir, u_int16_t *virtual_id, u_int16_t *virtual_type)
{
	/*
	 * ICMP types marked with PF_OUT are typically responses to
	 * PF_IN, and will match states in the opposite direction.
	 * PF_IN ICMP types need to match a state with that type.
	 */
	*icmp_dir = PF_OUT;

	/* Queries (and responses) */
	switch (pd->af) {
#ifdef INET
	case AF_INET:
		switch (type) {
		case ICMP_ECHO:
			*icmp_dir = PF_IN;
			/* FALLTHROUGH */
		case ICMP_ECHOREPLY:
			*virtual_type = ICMP_ECHO;
			*virtual_id = pd->hdr.icmp.icmp_id;
			break;

		case ICMP_TSTAMP:
			*icmp_dir = PF_IN;
			/* FALLTHROUGH */
		case ICMP_TSTAMPREPLY:
			*virtual_type = ICMP_TSTAMP;
			*virtual_id = pd->hdr.icmp.icmp_id;
			break;

		case ICMP_IREQ:
			*icmp_dir = PF_IN;
			/* FALLTHROUGH */
		case ICMP_IREQREPLY:
			*virtual_type = ICMP_IREQ;
			*virtual_id = pd->hdr.icmp.icmp_id;
			break;

		case ICMP_MASKREQ:
			*icmp_dir = PF_IN;
			/* FALLTHROUGH */
		case ICMP_MASKREPLY:
			*virtual_type = ICMP_MASKREQ;
			*virtual_id = pd->hdr.icmp.icmp_id;
			break;

		case ICMP_IPV6_WHEREAREYOU:
			*icmp_dir = PF_IN;
			/* FALLTHROUGH */
		case ICMP_IPV6_IAMHERE:
			*virtual_type = ICMP_IPV6_WHEREAREYOU;
			*virtual_id = 0; /* Nothing sane to match on! */
			break;

		case ICMP_MOBILE_REGREQUEST:
			*icmp_dir = PF_IN;
			/* FALLTHROUGH */
		case ICMP_MOBILE_REGREPLY:
			*virtual_type = ICMP_MOBILE_REGREQUEST;
			*virtual_id = 0; /* Nothing sane to match on! */
			break;

		case ICMP_ROUTERSOLICIT:
			*icmp_dir = PF_IN;
			/* FALLTHROUGH */
		case ICMP_ROUTERADVERT:
			*virtual_type = ICMP_ROUTERSOLICIT;
			*virtual_id = 0; /* Nothing sane to match on! */
			break;

		/* These ICMP types map to other connections */
		case ICMP_UNREACH:
		case ICMP_SOURCEQUENCH:
		case ICMP_REDIRECT:
		case ICMP_TIMXCEED:
		case ICMP_PARAMPROB:
			/* These will not be used, but set them anyway */
			*icmp_dir = PF_IN;
			*virtual_type = type;
			*virtual_id = 0;
			*virtual_type = htons(*virtual_type);
			return (1);  /* These types match to another state */

		/*
		 * All remaining ICMP types get their own states,
		 * and will only match in one direction.
		 */
		default:
			*icmp_dir = PF_IN;
			*virtual_type = type;
			*virtual_id = 0;
			break;
		}
		break;
#endif /* INET */
#ifdef INET6
	case AF_INET6:
		switch (type) {
		case ICMP6_ECHO_REQUEST:
			*icmp_dir = PF_IN;
			/* FALLTHROUGH */
		case ICMP6_ECHO_REPLY:
			*virtual_type = ICMP6_ECHO_REQUEST;
			*virtual_id = pd->hdr.icmp6.icmp6_id;
			break;

		case MLD_LISTENER_QUERY:
		case MLD_LISTENER_REPORT: {
			/*
			 * Listener Report can be sent by clients
			 * without an associated Listener Query.
			 * In addition to that, when Report is sent as a
			 * reply to a Query its source and destination
			 * address are different.
			 */
			*icmp_dir = PF_IN;
			*virtual_type = MLD_LISTENER_QUERY;
			*virtual_id = 0;
			break;
		}
		case MLD_MTRACE:
			*icmp_dir = PF_IN;
			/* FALLTHROUGH */
		case MLD_MTRACE_RESP:
			*virtual_type = MLD_MTRACE;
			*virtual_id = 0; /* Nothing sane to match on! */
			break;

		case ND_NEIGHBOR_SOLICIT:
			*icmp_dir = PF_IN;
			/* FALLTHROUGH */
		case ND_NEIGHBOR_ADVERT: {
			*virtual_type = ND_NEIGHBOR_SOLICIT;
			*virtual_id = 0;
			break;
		}

		/*
		 * These ICMP types map to other connections.
		 * ND_REDIRECT can't be in this list because the triggering
		 * packet header is optional.
		 */
		case ICMP6_DST_UNREACH:
		case ICMP6_PACKET_TOO_BIG:
		case ICMP6_TIME_EXCEEDED:
		case ICMP6_PARAM_PROB:
			/* These will not be used, but set them anyway */
			*icmp_dir = PF_IN;
			*virtual_type = type;
			*virtual_id = 0;
			*virtual_type = htons(*virtual_type);
			return (1);  /* These types match to another state */
		/*
		 * All remaining ICMP6 types get their own states,
		 * and will only match in one direction.
		 */
		default:
			*icmp_dir = PF_IN;
			*virtual_type = type;
			*virtual_id = 0;
			break;
		}
		break;
#endif /* INET6 */
	default:
		unhandled_af(pd->af);
	}
	*virtual_type = htons(*virtual_type);
	return (0);  /* These types match to their own state */
}

void
pf_intr(void *v)
{
	struct epoch_tracker et;
	struct pf_send_head queue;
	struct pf_send_entry *pfse, *next;

	CURVNET_SET((struct vnet *)v);

	PF_SENDQ_LOCK();
	queue = V_pf_sendqueue;
	STAILQ_INIT(&V_pf_sendqueue);
	PF_SENDQ_UNLOCK();

	NET_EPOCH_ENTER(et);

	STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
		switch (pfse->pfse_type) {
#ifdef INET
		case PFSE_IP: {
			if (pf_isforlocal(pfse->pfse_m, AF_INET)) {
				KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif,
				    ("%s: rcvif != loif", __func__));

				pfse->pfse_m->m_flags |= M_SKIP_FIREWALL;
				pfse->pfse_m->m_pkthdr.csum_flags |=
				    CSUM_IP_VALID | CSUM_IP_CHECKED |
				    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
				pfse->pfse_m->m_pkthdr.csum_data = 0xffff;
				ip_input(pfse->pfse_m);
			} else {
				ip_output(pfse->pfse_m, NULL, NULL, 0, NULL,
				    NULL);
			}
			break;
		}
		case PFSE_ICMP:
			icmp_error(pfse->pfse_m, pfse->icmpopts.type,
			    pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
			break;
#endif /* INET */
#ifdef INET6
		case PFSE_IP6:
			if (pf_isforlocal(pfse->pfse_m, AF_INET6)) {
				KASSERT(pfse->pfse_m->m_pkthdr.rcvif == V_loif,
				    ("%s: rcvif != loif", __func__));

				pfse->pfse_m->m_flags |= M_SKIP_FIREWALL |
				    M_LOOP;
				pfse->pfse_m->m_pkthdr.csum_flags |=
				    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
				pfse->pfse_m->m_pkthdr.csum_data = 0xffff;
				ip6_input(pfse->pfse_m);
			} else {
				ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL,
				    NULL, NULL);
			}
			break;
		case PFSE_ICMP6:
			icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
			    pfse->icmpopts.code, pfse->icmpopts.mtu);
			break;
#endif /* INET6 */
		default:
			panic("%s: unknown type", __func__);
		}
		free(pfse, M_PFTEMP);
	}
	NET_EPOCH_EXIT(et);
	CURVNET_RESTORE();
}

#define	pf_purge_thread_period	(hz / 10)

#ifdef PF_WANT_32_TO_64_COUNTER
static void
pf_status_counter_u64_periodic(void)
{

	PF_RULES_RASSERT();

	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) {
		return;
	}

	for (int i = 0; i < FCNT_MAX; i++) {
		pf_counter_u64_periodic(&V_pf_status.fcounters[i]);
	}
}

static void
pf_kif_counter_u64_periodic(void)
{
	struct pfi_kkif *kif;
	size_t r, run;

	PF_RULES_RASSERT();

	if (__predict_false(V_pf_allkifcount == 0)) {
		return;
	}

	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) {
		return;
	}

	run = V_pf_allkifcount / 10;
	if (run < 5)
		run = 5;

	for (r = 0; r < run; r++) {
		kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist);
		if (kif == NULL) {
			LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist);
			LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist);
			break;
		}

		LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist);
		LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist);

		for (int i = 0; i < 2; i++) {
			for (int j = 0; j < 2; j++) {
				for (int k = 0; k < 2; k++) {
					pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]);
					pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]);
				}
			}
		}
	}
}

static void
pf_rule_counter_u64_periodic(void)
{
	struct pf_krule *rule;
	size_t r, run;

	PF_RULES_RASSERT();

	if (__predict_false(V_pf_allrulecount == 0)) {
		return;
	}

	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) {
		return;
	}

	run = V_pf_allrulecount / 10;
	if (run < 5)
		run = 5;

	for (r = 0; r < run; r++) {
		rule = LIST_NEXT(V_pf_rulemarker, allrulelist);
		if (rule == NULL) {
			LIST_REMOVE(V_pf_rulemarker, allrulelist);
			LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist);
			break;
		}

		LIST_REMOVE(V_pf_rulemarker, allrulelist);
		LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist);

		pf_counter_u64_periodic(&rule->evaluations);
		for (int i = 0; i < 2; i++) {
			pf_counter_u64_periodic(&rule->packets[i]);
			pf_counter_u64_periodic(&rule->bytes[i]);
		}
	}
}

static void
pf_counter_u64_periodic_main(void)
{
	PF_RULES_RLOCK_TRACKER;

	V_pf_counter_periodic_iter++;

	PF_RULES_RLOCK();
	pf_counter_u64_critical_enter();
	pf_status_counter_u64_periodic();
	pf_kif_counter_u64_periodic();
	pf_rule_counter_u64_periodic();
	pf_counter_u64_critical_exit();
	PF_RULES_RUNLOCK();
}
#else
#define	pf_counter_u64_periodic_main()	do { } while (0)
#endif

void
pf_purge_thread(void *unused __unused)
{
	struct epoch_tracker	 et;

	VNET_ITERATOR_DECL(vnet_iter);

	sx_xlock(&pf_end_lock);
	while (pf_end_threads == 0) {
		sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period);

		VNET_LIST_RLOCK();
		NET_EPOCH_ENTER(et);
		VNET_FOREACH(vnet_iter) {
			CURVNET_SET(vnet_iter);

			/* Wait until V_pf_default_rule is initialized. */
			if (V_pf_vnet_active == 0) {
				CURVNET_RESTORE();
				continue;
			}

			pf_counter_u64_periodic_main();

			/*
			 *  Process 1/interval fraction of the state
			 * table every run.
			 */
			V_pf_purge_idx =
			    pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask /
			    (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));

			/*
			 * Purge other expired types every
			 * PFTM_INTERVAL seconds.
			 */
			if (V_pf_purge_idx == 0) {
				/*
				 * Order is important:
				 * - states and src nodes reference rules
				 * - states and rules reference kifs
				 */
				pf_purge_expired_fragments();
				pf_purge_expired_src_nodes();
				pf_purge_unlinked_rules();
				pfi_kkif_purge();
			}
			CURVNET_RESTORE();
		}
		NET_EPOCH_EXIT(et);
		VNET_LIST_RUNLOCK();
	}

	pf_end_threads++;
	sx_xunlock(&pf_end_lock);
	kproc_exit(0);
}

void
pf_unload_vnet_purge(void)
{

	/*
	 * To cleanse up all kifs and rules we need
	 * two runs: first one clears reference flags,
	 * then pf_purge_expired_states() doesn't
	 * raise them, and then second run frees.
	 */
	pf_purge_unlinked_rules();
	pfi_kkif_purge();

	/*
	 * Now purge everything.
	 */
	pf_purge_expired_states(0, V_pf_hashmask);
	pf_purge_fragments(UINT_MAX);
	pf_purge_expired_src_nodes();

	/*
	 * Now all kifs & rules should be unreferenced,
	 * thus should be successfully freed.
	 */
	pf_purge_unlinked_rules();
	pfi_kkif_purge();
}

u_int32_t
pf_state_expires(const struct pf_kstate *state)
{
	u_int32_t	timeout;
	u_int32_t	start;
	u_int32_t	end;
	u_int32_t	states;

	/* handle all PFTM_* > PFTM_MAX here */
	if (state->timeout == PFTM_PURGE)
		return (time_uptime);
	KASSERT(state->timeout != PFTM_UNLINKED,
	    ("pf_state_expires: timeout == PFTM_UNLINKED"));
	KASSERT((state->timeout < PFTM_MAX),
	    ("pf_state_expires: timeout > PFTM_MAX"));
	timeout = state->rule->timeout[state->timeout];
	if (!timeout)
		timeout = V_pf_default_rule.timeout[state->timeout];
	start = state->rule->timeout[PFTM_ADAPTIVE_START];
	if (start && state->rule != &V_pf_default_rule) {
		end = state->rule->timeout[PFTM_ADAPTIVE_END];
		states = counter_u64_fetch(state->rule->states_cur);
	} else {
		start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
		end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
		states = V_pf_status.states;
	}
	if (end && states > start && start < end) {
		if (states < end) {
			timeout = (u_int64_t)timeout * (end - states) /
			    (end - start);
			return ((state->expire / 1000) + timeout);
		}
		else
			return (time_uptime);
	}
	return ((state->expire / 1000) + timeout);
}

void
pf_purge_expired_src_nodes(void)
{
	struct pf_ksrc_node_list	 freelist;
	struct pf_srchash	*sh;
	struct pf_ksrc_node	*cur, *next;
	int i;

	LIST_INIT(&freelist);
	for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
	    PF_HASHROW_LOCK(sh);
	    LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
		if (cur->states == 0 && cur->expire <= time_uptime) {
			pf_unlink_src_node(cur);
			LIST_INSERT_HEAD(&freelist, cur, entry);
		} else if (cur->rule != NULL)
			cur->rule->rule_ref |= PFRULE_REFS;
	    PF_HASHROW_UNLOCK(sh);
	}

	pf_free_src_nodes(&freelist);

	V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
}

static void
pf_src_tree_remove_state(struct pf_kstate *s)
{
	uint32_t timeout;

	timeout = s->rule->timeout[PFTM_SRC_NODE] ?
	    s->rule->timeout[PFTM_SRC_NODE] :
	    V_pf_default_rule.timeout[PFTM_SRC_NODE];

	for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) {
		if (s->sns[sn_type] == NULL)
			continue;
		PF_SRC_NODE_LOCK(s->sns[sn_type]);
		if (sn_type == PF_SN_LIMIT && s->src.tcp_est)
			--(s->sns[sn_type]->conn);
		if (--(s->sns[sn_type]->states) == 0)
			s->sns[sn_type]->expire = time_uptime + timeout;
		PF_SRC_NODE_UNLOCK(s->sns[sn_type]);
		s->sns[sn_type] = NULL;
	}

}

/*
 * Unlink and potentilly free a state. Function may be
 * called with ID hash row locked, but always returns
 * unlocked, since it needs to go through key hash locking.
 */
int
pf_remove_state(struct pf_kstate *s)
{
	struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];

	NET_EPOCH_ASSERT();
	PF_HASHROW_ASSERT(ih);

	if (s->timeout == PFTM_UNLINKED) {
		/*
		 * State is being processed
		 * by pf_remove_state() in
		 * an other thread.
		 */
		PF_HASHROW_UNLOCK(ih);
		return (0);	/* XXXGL: undefined actually */
	}

	if (s->src.state == PF_TCPS_PROXY_DST) {
		/* XXX wire key the right one? */
		pf_send_tcp(s->rule, s->key[PF_SK_WIRE]->af,
		    &s->key[PF_SK_WIRE]->addr[1],
		    &s->key[PF_SK_WIRE]->addr[0],
		    s->key[PF_SK_WIRE]->port[1],
		    s->key[PF_SK_WIRE]->port[0],
		    s->src.seqhi, s->src.seqlo + 1,
		    TH_RST|TH_ACK, 0, 0, 0, M_SKIP_FIREWALL, s->tag, 0,
		    s->act.rtableid);
	}

	LIST_REMOVE(s, entry);
	pf_src_tree_remove_state(s);

	if (V_pfsync_delete_state_ptr != NULL)
		V_pfsync_delete_state_ptr(s);

	STATE_DEC_COUNTERS(s);

	s->timeout = PFTM_UNLINKED;

	/* Ensure we remove it from the list of halfopen states, if needed. */
	if (s->key[PF_SK_STACK] != NULL &&
	    s->key[PF_SK_STACK]->proto == IPPROTO_TCP)
		pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED);

	PF_HASHROW_UNLOCK(ih);

	pf_detach_state(s);

	pf_udp_mapping_release(s->udp_mapping);

	/* pf_state_insert() initialises refs to 2 */
	return (pf_release_staten(s, 2));
}

struct pf_kstate *
pf_alloc_state(int flags)
{

	return (uma_zalloc(V_pf_state_z, flags | M_ZERO));
}

void
pf_free_state(struct pf_kstate *cur)
{
	struct pf_krule_item *ri;

	KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
	KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
	    cur->timeout));

	while ((ri = SLIST_FIRST(&cur->match_rules))) {
		SLIST_REMOVE_HEAD(&cur->match_rules, entry);
		free(ri, M_PF_RULE_ITEM);
	}

	pf_normalize_tcp_cleanup(cur);
	uma_zfree(V_pf_state_z, cur);
	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
}

/*
 * Called only from pf_purge_thread(), thus serialized.
 */
static u_int
pf_purge_expired_states(u_int i, int maxcheck)
{
	struct pf_idhash *ih;
	struct pf_kstate *s;
	struct pf_krule_item *mrm;
	size_t count __unused;

	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);

	/*
	 * Go through hash and unlink states that expire now.
	 */
	while (maxcheck > 0) {
		count = 0;
		ih = &V_pf_idhash[i];

		/* only take the lock if we expect to do work */
		if (!LIST_EMPTY(&ih->states)) {
relock:
			PF_HASHROW_LOCK(ih);
			LIST_FOREACH(s, &ih->states, entry) {
				if (pf_state_expires(s) <= time_uptime) {
					V_pf_status.states -=
					    pf_remove_state(s);
					goto relock;
				}
				s->rule->rule_ref |= PFRULE_REFS;
				if (s->nat_rule != NULL)
					s->nat_rule->rule_ref |= PFRULE_REFS;
				if (s->anchor != NULL)
					s->anchor->rule_ref |= PFRULE_REFS;
				s->kif->pfik_flags |= PFI_IFLAG_REFS;
				SLIST_FOREACH(mrm, &s->match_rules, entry)
					mrm->r->rule_ref |= PFRULE_REFS;
				if (s->act.rt_kif)
					s->act.rt_kif->pfik_flags |= PFI_IFLAG_REFS;
				count++;
			}
			PF_HASHROW_UNLOCK(ih);
		}

		SDT_PROBE2(pf, purge, state, rowcount, i, count);

		/* Return when we hit end of hash. */
		if (++i > V_pf_hashmask) {
			V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
			return (0);
		}

		maxcheck--;
	}

	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);

	return (i);
}

static void
pf_purge_unlinked_rules(void)
{
	struct pf_krulequeue tmpq;
	struct pf_krule *r, *r1;

	/*
	 * If we have overloading task pending, then we'd
	 * better skip purging this time. There is a tiny
	 * probability that overloading task references
	 * an already unlinked rule.
	 */
	PF_OVERLOADQ_LOCK();
	if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
		PF_OVERLOADQ_UNLOCK();
		return;
	}
	PF_OVERLOADQ_UNLOCK();

	/*
	 * Do naive mark-and-sweep garbage collecting of old rules.
	 * Reference flag is raised by pf_purge_expired_states()
	 * and pf_purge_expired_src_nodes().
	 *
	 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
	 * use a temporary queue.
	 */
	TAILQ_INIT(&tmpq);
	PF_UNLNKDRULES_LOCK();
	TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
		if (!(r->rule_ref & PFRULE_REFS)) {
			TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
			TAILQ_INSERT_TAIL(&tmpq, r, entries);
		} else
			r->rule_ref &= ~PFRULE_REFS;
	}
	PF_UNLNKDRULES_UNLOCK();

	if (!TAILQ_EMPTY(&tmpq)) {
		PF_CONFIG_LOCK();
		PF_RULES_WLOCK();
		TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
			TAILQ_REMOVE(&tmpq, r, entries);
			pf_free_rule(r);
		}
		PF_RULES_WUNLOCK();
		PF_CONFIG_UNLOCK();
	}
}

void
pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
{
	switch (af) {
#ifdef INET
	case AF_INET: {
		u_int32_t a = ntohl(addr->addr32[0]);
		printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
		    (a>>8)&255, a&255);
		if (p) {
			p = ntohs(p);
			printf(":%u", p);
		}
		break;
	}
#endif /* INET */
#ifdef INET6
	case AF_INET6: {
		u_int16_t b;
		u_int8_t i, curstart, curend, maxstart, maxend;
		curstart = curend = maxstart = maxend = 255;
		for (i = 0; i < 8; i++) {
			if (!addr->addr16[i]) {
				if (curstart == 255)
					curstart = i;
				curend = i;
			} else {
				if ((curend - curstart) >
				    (maxend - maxstart)) {
					maxstart = curstart;
					maxend = curend;
				}
				curstart = curend = 255;
			}
		}
		if ((curend - curstart) >
		    (maxend - maxstart)) {
			maxstart = curstart;
			maxend = curend;
		}
		for (i = 0; i < 8; i++) {
			if (i >= maxstart && i <= maxend) {
				if (i == 0)
					printf(":");
				if (i == maxend)
					printf(":");
			} else {
				b = ntohs(addr->addr16[i]);
				printf("%x", b);
				if (i < 7)
					printf(":");
			}
		}
		if (p) {
			p = ntohs(p);
			printf("[%u]", p);
		}
		break;
	}
#endif /* INET6 */
	default:
		unhandled_af(af);
	}
}

void
pf_print_state(struct pf_kstate *s)
{
	pf_print_state_parts(s, NULL, NULL);
}

static void
pf_print_state_parts(struct pf_kstate *s,
    struct pf_state_key *skwp, struct pf_state_key *sksp)
{
	struct pf_state_key *skw, *sks;
	u_int8_t proto, dir;

	/* Do our best to fill these, but they're skipped if NULL */
	skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
	sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
	proto = skw ? skw->proto : (sks ? sks->proto : 0);
	dir = s ? s->direction : 0;

	switch (proto) {
	case IPPROTO_IPV4:
		printf("IPv4");
		break;
	case IPPROTO_IPV6:
		printf("IPv6");
		break;
	case IPPROTO_TCP:
		printf("TCP");
		break;
	case IPPROTO_UDP:
		printf("UDP");
		break;
	case IPPROTO_ICMP:
		printf("ICMP");
		break;
	case IPPROTO_ICMPV6:
		printf("ICMPv6");
		break;
	default:
		printf("%u", proto);
		break;
	}
	switch (dir) {
	case PF_IN:
		printf(" in");
		break;
	case PF_OUT:
		printf(" out");
		break;
	}
	if (skw) {
		printf(" wire: ");
		pf_print_host(&skw->addr[0], skw->port[0], skw->af);
		printf(" ");
		pf_print_host(&skw->addr[1], skw->port[1], skw->af);
	}
	if (sks) {
		printf(" stack: ");
		if (sks != skw) {
			pf_print_host(&sks->addr[0], sks->port[0], sks->af);
			printf(" ");
			pf_print_host(&sks->addr[1], sks->port[1], sks->af);
		} else
			printf("-");
	}
	if (s) {
		if (proto == IPPROTO_TCP) {
			printf(" [lo=%u high=%u win=%u modulator=%u",
			    s->src.seqlo, s->src.seqhi,
			    s->src.max_win, s->src.seqdiff);
			if (s->src.wscale && s->dst.wscale)
				printf(" wscale=%u",
				    s->src.wscale & PF_WSCALE_MASK);
			printf("]");
			printf(" [lo=%u high=%u win=%u modulator=%u",
			    s->dst.seqlo, s->dst.seqhi,
			    s->dst.max_win, s->dst.seqdiff);
			if (s->src.wscale && s->dst.wscale)
				printf(" wscale=%u",
				s->dst.wscale & PF_WSCALE_MASK);
			printf("]");
		}
		printf(" %u:%u", s->src.state, s->dst.state);
		if (s->rule)
			printf(" @%d", s->rule->nr);
	}
}

void
pf_print_flags(uint16_t f)
{
	if (f)
		printf(" ");
	if (f & TH_FIN)
		printf("F");
	if (f & TH_SYN)
		printf("S");
	if (f & TH_RST)
		printf("R");
	if (f & TH_PUSH)
		printf("P");
	if (f & TH_ACK)
		printf("A");
	if (f & TH_URG)
		printf("U");
	if (f & TH_ECE)
		printf("E");
	if (f & TH_CWR)
		printf("W");
	if (f & TH_AE)
		printf("e");
}

#define	PF_SET_SKIP_STEPS(i)					\
	do {							\
		while (head[i] != cur) {			\
			head[i]->skip[i] = cur;			\
			head[i] = TAILQ_NEXT(head[i], entries);	\
		}						\
	} while (0)

void
pf_calc_skip_steps(struct pf_krulequeue *rules)
{
	struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT];
	int i;

	cur = TAILQ_FIRST(rules);
	prev = cur;
	for (i = 0; i < PF_SKIP_COUNT; ++i)
		head[i] = cur;
	while (cur != NULL) {
		if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
			PF_SET_SKIP_STEPS(PF_SKIP_IFP);
		if (cur->direction != prev->direction)
			PF_SET_SKIP_STEPS(PF_SKIP_DIR);
		if (cur->af != prev->af)
			PF_SET_SKIP_STEPS(PF_SKIP_AF);
		if (cur->proto != prev->proto)
			PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
		if (cur->src.neg != prev->src.neg ||
		    pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
			PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
		if (cur->dst.neg != prev->dst.neg ||
		    pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
			PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
		if (cur->src.port[0] != prev->src.port[0] ||
		    cur->src.port[1] != prev->src.port[1] ||
		    cur->src.port_op != prev->src.port_op)
			PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
		if (cur->dst.port[0] != prev->dst.port[0] ||
		    cur->dst.port[1] != prev->dst.port[1] ||
		    cur->dst.port_op != prev->dst.port_op)
			PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);

		prev = cur;
		cur = TAILQ_NEXT(cur, entries);
	}
	for (i = 0; i < PF_SKIP_COUNT; ++i)
		PF_SET_SKIP_STEPS(i);
}

int
pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
{
	if (aw1->type != aw2->type)
		return (1);
	switch (aw1->type) {
	case PF_ADDR_ADDRMASK:
	case PF_ADDR_RANGE:
		if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
			return (1);
		if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
			return (1);
		return (0);
	case PF_ADDR_DYNIFTL:
		return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
	case PF_ADDR_NONE:
	case PF_ADDR_NOROUTE:
	case PF_ADDR_URPFFAILED:
		return (0);
	case PF_ADDR_TABLE:
		return (aw1->p.tbl != aw2->p.tbl);
	default:
		printf("invalid address type: %d\n", aw1->type);
		return (1);
	}
}

/**
 * Checksum updates are a little complicated because the checksum in the TCP/UDP
 * header isn't always a full checksum. In some cases (i.e. output) it's a
 * pseudo-header checksum, which is a partial checksum over src/dst IP
 * addresses, protocol number and length.
 *
 * That means we have the following cases:
 *  * Input or forwarding: we don't have TSO, the checksum fields are full
 *  	checksums, we need to update the checksum whenever we change anything.
 *  * Output (i.e. the checksum is a pseudo-header checksum):
 *  	x The field being updated is src/dst address or affects the length of
 *  	the packet. We need to update the pseudo-header checksum (note that this
 *  	checksum is not ones' complement).
 *  	x Some other field is being modified (e.g. src/dst port numbers): We
 *  	don't have to update anything.
 **/
u_int16_t
pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
{
	u_int32_t x;

	x = cksum + old - new;
	x = (x + (x >> 16)) & 0xffff;

	/* optimise: eliminate a branch when not udp */
	if (udp && cksum == 0x0000)
		return cksum;
	if (udp && x == 0x0000)
		x = 0xffff;

	return (u_int16_t)(x);
}

static int
pf_patch_8(struct pf_pdesc *pd, u_int8_t *f, u_int8_t v, bool hi)
{
	int	 rewrite = 0;

	if (*f != v) {
		uint16_t old = htons(hi ? (*f << 8) : *f);
		uint16_t new = htons(hi ? ( v << 8) :  v);

		*f = v;

		if (! (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
		    CSUM_DELAY_DATA_IPV6)))
			*pd->pcksum = pf_cksum_fixup(*pd->pcksum, old, new,
			    pd->proto == IPPROTO_UDP);

		rewrite = 1;
	}

	return (rewrite);
}

int
pf_patch_16(struct pf_pdesc *pd, void *f, u_int16_t v, bool hi)
{
	int rewrite = 0;
	u_int8_t *fb = (u_int8_t *)f;
	u_int8_t *vb = (u_int8_t *)&v;

	rewrite += pf_patch_8(pd, fb++, *vb++, hi);
	rewrite += pf_patch_8(pd, fb++, *vb++, !hi);

	return (rewrite);
}

int
pf_patch_32(struct pf_pdesc *pd, void *f, u_int32_t v, bool hi)
{
	int rewrite = 0;
	u_int8_t *fb = (u_int8_t *)f;
	u_int8_t *vb = (u_int8_t *)&v;

	rewrite += pf_patch_8(pd, fb++, *vb++, hi);
	rewrite += pf_patch_8(pd, fb++, *vb++, !hi);
	rewrite += pf_patch_8(pd, fb++, *vb++, hi);
	rewrite += pf_patch_8(pd, fb++, *vb++, !hi);

	return (rewrite);
}

u_int16_t
pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
        u_int16_t new, u_int8_t udp)
{
	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
		return (cksum);

	return (pf_cksum_fixup(cksum, old, new, udp));
}

static void
pf_change_ap(struct pf_pdesc *pd, struct pf_addr *a, u_int16_t *p,
        struct pf_addr *an, u_int16_t pn)
{
	struct pf_addr	ao;
	u_int16_t	po;
	uint8_t		u = pd->virtual_proto == IPPROTO_UDP;

	MPASS(pd->pcksum);
	if (pd->af == AF_INET) {
		MPASS(pd->ip_sum);
	}

	pf_addrcpy(&ao, a, pd->af);
	if (pd->af == pd->naf)
		pf_addrcpy(a, an, pd->af);

	if (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
		*pd->pcksum = ~*pd->pcksum;

	if (p == NULL)  /* no port -> done. no cksum to worry about. */
		return;
	po = *p;
	*p = pn;

	switch (pd->af) {
#ifdef INET
	case AF_INET:
		switch (pd->naf) {
		case AF_INET:
			*pd->ip_sum = pf_cksum_fixup(pf_cksum_fixup(*pd->ip_sum,
			    ao.addr16[0], an->addr16[0], 0),
			    ao.addr16[1], an->addr16[1], 0);
			*p = pn;

			*pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum,
			    ao.addr16[0], an->addr16[0], u),
			    ao.addr16[1], an->addr16[1], u);

			*pd->pcksum = pf_proto_cksum_fixup(pd->m, *pd->pcksum, po, pn, u);
			break;
#ifdef INET6
		case AF_INET6:
			*pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
			   pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
			    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum,
			    ao.addr16[0], an->addr16[0], u),
			    ao.addr16[1], an->addr16[1], u),
			    0,            an->addr16[2], u),
			    0,            an->addr16[3], u),
			    0,            an->addr16[4], u),
			    0,            an->addr16[5], u),
			    0,            an->addr16[6], u),
			    0,            an->addr16[7], u),
			    po, pn, u);
			break;
#endif /* INET6 */
		default:
			unhandled_af(pd->naf);
		}
		break;
#endif /* INET */
#ifdef INET6
	case AF_INET6:
		switch (pd->naf) {
#ifdef INET
		case AF_INET:
			*pd->pcksum = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
			    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
			    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum,
			    ao.addr16[0], an->addr16[0], u),
			    ao.addr16[1], an->addr16[1], u),
			    ao.addr16[2], 0,             u),
			    ao.addr16[3], 0,             u),
			    ao.addr16[4], 0,             u),
			    ao.addr16[5], 0,             u),
			    ao.addr16[6], 0,             u),
			    ao.addr16[7], 0,             u),
			    po, pn, u);
			break;
#endif /* INET */
		case AF_INET6:
			*pd->pcksum  = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
			    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
			    pf_cksum_fixup(pf_cksum_fixup(*pd->pcksum,
			    ao.addr16[0], an->addr16[0], u),
			    ao.addr16[1], an->addr16[1], u),
			    ao.addr16[2], an->addr16[2], u),
			    ao.addr16[3], an->addr16[3], u),
			    ao.addr16[4], an->addr16[4], u),
			    ao.addr16[5], an->addr16[5], u),
			    ao.addr16[6], an->addr16[6], u),
			    ao.addr16[7], an->addr16[7], u);

			*pd->pcksum = pf_proto_cksum_fixup(pd->m, *pd->pcksum, po, pn, u);
			break;
		default:
			unhandled_af(pd->naf);
		}
		break;
#endif /* INET6 */
	default:
		unhandled_af(pd->af);
	}

	if (pd->m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
	    CSUM_DELAY_DATA_IPV6)) {
		*pd->pcksum = ~*pd->pcksum;
		if (! *pd->pcksum)
			*pd->pcksum = 0xffff;
	}
}

/* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
void
pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
{
	u_int32_t	ao;

	memcpy(&ao, a, sizeof(ao));
	memcpy(a, &an, sizeof(u_int32_t));
	*c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
	    ao % 65536, an % 65536, u);
}

void
pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
{
	u_int32_t	ao;

	memcpy(&ao, a, sizeof(ao));
	memcpy(a, &an, sizeof(u_int32_t));

	*c = pf_proto_cksum_fixup(m,
	    pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
	    ao % 65536, an % 65536, udp);
}

#ifdef INET6
static void
pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
{
	struct pf_addr	ao;

	pf_addrcpy(&ao, a, AF_INET6);
	pf_addrcpy(a, an, AF_INET6);

	*c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
	    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
	    pf_cksum_fixup(pf_cksum_fixup(*c,
	    ao.addr16[0], an->addr16[0], u),
	    ao.addr16[1], an->addr16[1], u),
	    ao.addr16[2], an->addr16[2], u),
	    ao.addr16[3], an->addr16[3], u),
	    ao.addr16[4], an->addr16[4], u),
	    ao.addr16[5], an->addr16[5], u),
	    ao.addr16[6], an->addr16[6], u),
	    ao.addr16[7], an->addr16[7], u);
}
#endif /* INET6 */

static void
pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
    struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
    u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
{
	struct pf_addr	oia, ooa;

	pf_addrcpy(&oia, ia, af);
	if (oa)
		pf_addrcpy(&ooa, oa, af);

	/* Change inner protocol port, fix inner protocol checksum. */
	if (ip != NULL) {
		u_int16_t	oip = *ip;
		u_int32_t	opc;

		if (pc != NULL)
			opc = *pc;
		*ip = np;
		if (pc != NULL)
			*pc = pf_cksum_fixup(*pc, oip, *ip, u);
		*ic = pf_cksum_fixup(*ic, oip, *ip, 0);
		if (pc != NULL)
			*ic = pf_cksum_fixup(*ic, opc, *pc, 0);
	}
	/* Change inner ip address, fix inner ip and icmp checksums. */
	pf_addrcpy(ia, na, af);
	switch (af) {
#ifdef INET
	case AF_INET: {
		u_int32_t	 oh2c = *h2c;

		*h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
		    oia.addr16[0], ia->addr16[0], 0),
		    oia.addr16[1], ia->addr16[1], 0);
		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
		    oia.addr16[0], ia->addr16[0], 0),
		    oia.addr16[1], ia->addr16[1], 0);
		*ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
		break;
	}
#endif /* INET */
#ifdef INET6
	case AF_INET6:
		*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
		    pf_cksum_fixup(pf_cksum_fixup(*ic,
		    oia.addr16[0], ia->addr16[0], u),
		    oia.addr16[1], ia->addr16[1], u),
		    oia.addr16[2], ia->addr16[2], u),
		    oia.addr16[3], ia->addr16[3], u),
		    oia.addr16[4], ia->addr16[4], u),
		    oia.addr16[5], ia->addr16[5], u),
		    oia.addr16[6], ia->addr16[6], u),
		    oia.addr16[7], ia->addr16[7], u);
		break;
#endif /* INET6 */
	}
	/* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
	if (oa) {
		pf_addrcpy(oa, na, af);
		switch (af) {
#ifdef INET
		case AF_INET:
			*hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
			    ooa.addr16[0], oa->addr16[0], 0),
			    ooa.addr16[1], oa->addr16[1], 0);
			break;
#endif /* INET */
#ifdef INET6
		case AF_INET6:
			*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
			    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
			    pf_cksum_fixup(pf_cksum_fixup(*ic,
			    ooa.addr16[0], oa->addr16[0], u),
			    ooa.addr16[1], oa->addr16[1], u),
			    ooa.addr16[2], oa->addr16[2], u),
			    ooa.addr16[3], oa->addr16[3], u),
			    ooa.addr16[4], oa->addr16[4], u),
			    ooa.addr16[5], oa->addr16[5], u),
			    ooa.addr16[6], oa->addr16[6], u),
			    ooa.addr16[7], oa->addr16[7], u);
			break;
#endif /* INET6 */
		}
	}
}

int
pf_translate_af(struct pf_pdesc *pd)
{
#if defined(INET) && defined(INET6)
	struct mbuf		*mp;
	struct ip		*ip4;
	struct ip6_hdr		*ip6;
	struct icmp6_hdr	*icmp;
	struct m_tag		*mtag;
	struct pf_fragment_tag	*ftag;
	int			 hlen;

	hlen = pd->naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6);

	/* trim the old header */
	m_adj(pd->m, pd->off);

	/* prepend a new one */
	M_PREPEND(pd->m, hlen, M_NOWAIT);
	if (pd->m == NULL)
		return (-1);

	switch (pd->naf) {
	case AF_INET:
		ip4 = mtod(pd->m, struct ip *);
		bzero(ip4, hlen);
		ip4->ip_v = IPVERSION;
		ip4->ip_hl = hlen >> 2;
		ip4->ip_tos = pd->tos;
		ip4->ip_len = htons(hlen + (pd->tot_len - pd->off));
		ip_fillid(ip4, V_ip_random_id);
		ip4->ip_ttl = pd->ttl;
		ip4->ip_p = pd->proto;
		ip4->ip_src = pd->nsaddr.v4;
		ip4->ip_dst = pd->ndaddr.v4;
		pd->src = (struct pf_addr *)&ip4->ip_src;
		pd->dst = (struct pf_addr *)&ip4->ip_dst;
		pd->off = sizeof(struct ip);
		break;
	case AF_INET6:
		ip6 = mtod(pd->m, struct ip6_hdr *);
		bzero(ip6, hlen);
		ip6->ip6_vfc = IPV6_VERSION;
		ip6->ip6_flow |= htonl((u_int32_t)pd->tos << 20);
		ip6->ip6_plen = htons(pd->tot_len - pd->off);
		ip6->ip6_nxt = pd->proto;
		if (!pd->ttl || pd->ttl > IPV6_DEFHLIM)
			ip6->ip6_hlim = IPV6_DEFHLIM;
		else
			ip6->ip6_hlim = pd->ttl;
		ip6->ip6_src = pd->nsaddr.v6;
		ip6->ip6_dst = pd->ndaddr.v6;
		pd->src = (struct pf_addr *)&ip6->ip6_src;
		pd->dst = (struct pf_addr *)&ip6->ip6_dst;
		pd->off = sizeof(struct ip6_hdr);

		/*
		 * If we're dealing with a reassembled packet we need to adjust
		 * the header length from the IPv4 header size to IPv6 header
		 * size.
		 */
		mtag = m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL);
		if (mtag) {
			ftag = (struct pf_fragment_tag *)(mtag + 1);
			ftag->ft_hdrlen = sizeof(*ip6);
			ftag->ft_maxlen -= sizeof(struct ip6_hdr) -
			    sizeof(struct ip) + sizeof(struct ip6_frag);
		}
		break;
	default:
		return (-1);
	}

	/* recalculate icmp/icmp6 checksums */
	if (pd->proto == IPPROTO_ICMP || pd->proto == IPPROTO_ICMPV6) {
		int off;
		if ((mp = m_pulldown(pd->m, hlen, sizeof(*icmp), &off)) ==
		    NULL) {
			pd->m = NULL;
			return (-1);
		}
		icmp = (struct icmp6_hdr *)(mp->m_data + off);
		icmp->icmp6_cksum = 0;
		icmp->icmp6_cksum = pd->naf == AF_INET ?
		    in4_cksum(pd->m, 0, hlen, ntohs(ip4->ip_len) - hlen) :
		    in6_cksum(pd->m, IPPROTO_ICMPV6, hlen,
			ntohs(ip6->ip6_plen));
	}
#endif /* INET && INET6 */

	return (0);
}

int
pf_change_icmp_af(struct mbuf *m, int off, struct pf_pdesc *pd,
    struct pf_pdesc *pd2, struct pf_addr *src, struct pf_addr *dst,
    sa_family_t af, sa_family_t naf)
{
#if defined(INET) && defined(INET6)
	struct mbuf	*n = NULL;
	struct ip	*ip4;
	struct ip6_hdr	*ip6;
	int		 hlen, olen, mlen;

	if (af == naf || (af != AF_INET && af != AF_INET6) ||
	    (naf != AF_INET && naf != AF_INET6))
		return (-1);

	/* split the mbuf chain on the inner ip/ip6 header boundary */
	if ((n = m_split(m, off, M_NOWAIT)) == NULL)
		return (-1);

	/* old header */
	olen = pd2->off - off;
	/* new header */
	hlen = naf == AF_INET ? sizeof(*ip4) : sizeof(*ip6);

	/* trim old header */
	m_adj(n, olen);

	/* prepend a new one */
	M_PREPEND(n, hlen, M_NOWAIT);
	if (n == NULL)
		return (-1);

	/* translate inner ip/ip6 header */
	switch (naf) {
	case AF_INET:
		ip4 = mtod(n, struct ip *);
		bzero(ip4, sizeof(*ip4));
		ip4->ip_v = IPVERSION;
		ip4->ip_hl = sizeof(*ip4) >> 2;
		ip4->ip_len = htons(sizeof(*ip4) + pd2->tot_len - olen);
		ip_fillid(ip4, V_ip_random_id);
		ip4->ip_off = htons(IP_DF);
		ip4->ip_ttl = pd2->ttl;
		if (pd2->proto == IPPROTO_ICMPV6)
			ip4->ip_p = IPPROTO_ICMP;
		else
			ip4->ip_p = pd2->proto;
		ip4->ip_src = src->v4;
		ip4->ip_dst = dst->v4;
		ip4->ip_sum = in_cksum(n, ip4->ip_hl << 2);
		break;
	case AF_INET6:
		ip6 = mtod(n, struct ip6_hdr *);
		bzero(ip6, sizeof(*ip6));
		ip6->ip6_vfc = IPV6_VERSION;
		ip6->ip6_plen = htons(pd2->tot_len - olen);
		if (pd2->proto == IPPROTO_ICMP)
			ip6->ip6_nxt = IPPROTO_ICMPV6;
		else
			ip6->ip6_nxt = pd2->proto;
		if (!pd2->ttl || pd2->ttl > IPV6_DEFHLIM)
			ip6->ip6_hlim = IPV6_DEFHLIM;
		else
			ip6->ip6_hlim = pd2->ttl;
		ip6->ip6_src = src->v6;
		ip6->ip6_dst = dst->v6;
		break;
	default:
		unhandled_af(naf);
	}

	/* adjust payload offset and total packet length */
	pd2->off += hlen - olen;
	pd->tot_len += hlen - olen;

	/* merge modified inner packet with the original header */
	mlen = n->m_pkthdr.len;
	m_cat(m, n);
	m->m_pkthdr.len += mlen;
#endif /* INET && INET6 */

	return (0);
}

#define PTR_IP(field)	(offsetof(struct ip, field))
#define PTR_IP6(field)	(offsetof(struct ip6_hdr, field))

int
pf_translate_icmp_af(int af, void *arg)
{
#if defined(INET) && defined(INET6)
	struct icmp		*icmp4;
	struct icmp6_hdr	*icmp6;
	u_int32_t		 mtu;
	int32_t			 ptr = -1;
	u_int8_t		 type;
	u_int8_t		 code;

	switch (af) {
	case AF_INET:
		icmp6 = arg;
		type = icmp6->icmp6_type;
		code = icmp6->icmp6_code;
		mtu = ntohl(icmp6->icmp6_mtu);

		switch (type) {
		case ICMP6_ECHO_REQUEST:
			type = ICMP_ECHO;
			break;
		case ICMP6_ECHO_REPLY:
			type = ICMP_ECHOREPLY;
			break;
		case ICMP6_DST_UNREACH:
			type = ICMP_UNREACH;
			switch (code) {
			case ICMP6_DST_UNREACH_NOROUTE:
			case ICMP6_DST_UNREACH_BEYONDSCOPE:
			case ICMP6_DST_UNREACH_ADDR:
				code = ICMP_UNREACH_HOST;
				break;
			case ICMP6_DST_UNREACH_ADMIN:
				code = ICMP_UNREACH_HOST_PROHIB;
				break;
			case ICMP6_DST_UNREACH_NOPORT:
				code = ICMP_UNREACH_PORT;
				break;
			default:
				return (-1);
			}
			break;
		case ICMP6_PACKET_TOO_BIG:
			type = ICMP_UNREACH;
			code = ICMP_UNREACH_NEEDFRAG;
			mtu -= 20;
			break;
		case ICMP6_TIME_EXCEEDED:
			type = ICMP_TIMXCEED;
			break;
		case ICMP6_PARAM_PROB:
			switch (code) {
			case ICMP6_PARAMPROB_HEADER:
				type = ICMP_PARAMPROB;
				code = ICMP_PARAMPROB_ERRATPTR;
				ptr = ntohl(icmp6->icmp6_pptr);

				if (ptr == PTR_IP6(ip6_vfc))
					; /* preserve */
				else if (ptr == PTR_IP6(ip6_vfc) + 1)
					ptr = PTR_IP(ip_tos);
				else if (ptr == PTR_IP6(ip6_plen) ||
				    ptr == PTR_IP6(ip6_plen) + 1)
					ptr = PTR_IP(ip_len);
				else if (ptr == PTR_IP6(ip6_nxt))
					ptr = PTR_IP(ip_p);
				else if (ptr == PTR_IP6(ip6_hlim))
					ptr = PTR_IP(ip_ttl);
				else if (ptr >= PTR_IP6(ip6_src) &&
				    ptr < PTR_IP6(ip6_dst))
					ptr = PTR_IP(ip_src);
				else if (ptr >= PTR_IP6(ip6_dst) &&
				    ptr < sizeof(struct ip6_hdr))
					ptr = PTR_IP(ip_dst);
				else {
					return (-1);
				}
				break;
			case ICMP6_PARAMPROB_NEXTHEADER:
				type = ICMP_UNREACH;
				code = ICMP_UNREACH_PROTOCOL;
				break;
			default:
				return (-1);
			}
			break;
		default:
			return (-1);
		}
		if (icmp6->icmp6_type != type) {
			icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum,
			    icmp6->icmp6_type, type, 0);
			icmp6->icmp6_type = type;
		}
		if (icmp6->icmp6_code != code) {
			icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum,
			    icmp6->icmp6_code, code, 0);
			icmp6->icmp6_code = code;
		}
		if (icmp6->icmp6_mtu != htonl(mtu)) {
			icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum,
			    htons(ntohl(icmp6->icmp6_mtu)), htons(mtu), 0);
			/* aligns well with a icmpv4 nextmtu */
			icmp6->icmp6_mtu = htonl(mtu);
		}
		if (ptr >= 0 && icmp6->icmp6_pptr != htonl(ptr)) {
			icmp6->icmp6_cksum = pf_cksum_fixup(icmp6->icmp6_cksum,
			    htons(ntohl(icmp6->icmp6_pptr)), htons(ptr), 0);
			/* icmpv4 pptr is a one most significant byte */
			icmp6->icmp6_pptr = htonl(ptr << 24);
		}
		break;
	case AF_INET6:
		icmp4 = arg;
		type = icmp4->icmp_type;
		code = icmp4->icmp_code;
		mtu = ntohs(icmp4->icmp_nextmtu);

		switch (type) {
		case ICMP_ECHO:
			type = ICMP6_ECHO_REQUEST;
			break;
		case ICMP_ECHOREPLY:
			type = ICMP6_ECHO_REPLY;
			break;
		case ICMP_UNREACH:
			type = ICMP6_DST_UNREACH;
			switch (code) {
			case ICMP_UNREACH_NET:
			case ICMP_UNREACH_HOST:
			case ICMP_UNREACH_NET_UNKNOWN:
			case ICMP_UNREACH_HOST_UNKNOWN:
			case ICMP_UNREACH_ISOLATED:
			case ICMP_UNREACH_TOSNET:
			case ICMP_UNREACH_TOSHOST:
				code = ICMP6_DST_UNREACH_NOROUTE;
				break;
			case ICMP_UNREACH_PORT:
				code = ICMP6_DST_UNREACH_NOPORT;
				break;
			case ICMP_UNREACH_NET_PROHIB:
			case ICMP_UNREACH_HOST_PROHIB:
			case ICMP_UNREACH_FILTER_PROHIB:
			case ICMP_UNREACH_PRECEDENCE_CUTOFF:
				code = ICMP6_DST_UNREACH_ADMIN;
				break;
			case ICMP_UNREACH_PROTOCOL:
				type = ICMP6_PARAM_PROB;
				code = ICMP6_PARAMPROB_NEXTHEADER;
				ptr = offsetof(struct ip6_hdr, ip6_nxt);
				break;
			case ICMP_UNREACH_NEEDFRAG:
				type = ICMP6_PACKET_TOO_BIG;
				code = 0;
				mtu += 20;
				break;
			default:
				return (-1);
			}
			break;
		case ICMP_TIMXCEED:
			type = ICMP6_TIME_EXCEEDED;
			break;
		case ICMP_PARAMPROB:
			type = ICMP6_PARAM_PROB;
			switch (code) {
			case ICMP_PARAMPROB_ERRATPTR:
				code = ICMP6_PARAMPROB_HEADER;
				break;
			case ICMP_PARAMPROB_LENGTH:
				code = ICMP6_PARAMPROB_HEADER;
				break;
			default:
				return (-1);
			}

			ptr = icmp4->icmp_pptr;
			if (ptr == 0 || ptr == PTR_IP(ip_tos))
				; /* preserve */
			else if (ptr == PTR_IP(ip_len) ||
			    ptr == PTR_IP(ip_len) + 1)
				ptr = PTR_IP6(ip6_plen);
			else if (ptr == PTR_IP(ip_ttl))
				ptr = PTR_IP6(ip6_hlim);
			else if (ptr == PTR_IP(ip_p))
				ptr = PTR_IP6(ip6_nxt);
			else if (ptr >= PTR_IP(ip_src) && ptr < PTR_IP(ip_dst))
				ptr = PTR_IP6(ip6_src);
			else if (ptr >= PTR_IP(ip_dst) &&
			    ptr < sizeof(struct ip))
				ptr = PTR_IP6(ip6_dst);
			else {
				return (-1);
			}
			break;
		default:
			return (-1);
		}
		if (icmp4->icmp_type != type) {
			icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum,
			    icmp4->icmp_type, type, 0);
			icmp4->icmp_type = type;
		}
		if (icmp4->icmp_code != code) {
			icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum,
			    icmp4->icmp_code, code, 0);
			icmp4->icmp_code = code;
		}
		if (icmp4->icmp_nextmtu != htons(mtu)) {
			icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum,
			    icmp4->icmp_nextmtu, htons(mtu), 0);
			icmp4->icmp_nextmtu = htons(mtu);
		}
		if (ptr >= 0 && icmp4->icmp_void != ptr) {
			icmp4->icmp_cksum = pf_cksum_fixup(icmp4->icmp_cksum,
			    htons(icmp4->icmp_pptr), htons(ptr), 0);
			icmp4->icmp_void = htonl(ptr);
		}
		break;
	default:
		unhandled_af(af);
	}
#endif /* INET && INET6 */

	return (0);
}

/*
 * Need to modulate the sequence numbers in the TCP SACK option
 * (credits to Krzysztof Pfaff for report and patch)
 */
static int
pf_modulate_sack(struct pf_pdesc *pd, struct tcphdr *th,
    struct pf_state_peer *dst)
{
	struct sackblk	 sack;
	int		 copyback = 0, i;
	int		 olen, optsoff;
	uint8_t		 opts[MAX_TCPOPTLEN], *opt, *eoh;

	olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr);
	optsoff = pd->off + sizeof(struct tcphdr);
#define	TCPOLEN_MINSACK	(TCPOLEN_SACK + 2)
	if (olen < TCPOLEN_MINSACK ||
	    !pf_pull_hdr(pd->m, optsoff, opts, olen, NULL, NULL, pd->af))
		return (0);

	eoh = opts + olen;
	opt = opts;
	while ((opt = pf_find_tcpopt(opt, opts, olen,
	    TCPOPT_SACK, TCPOLEN_MINSACK)) != NULL)
	{
		size_t safelen = MIN(opt[1], (eoh - opt));
		for (i = 2; i + TCPOLEN_SACK <= safelen; i += TCPOLEN_SACK) {
			size_t startoff = (opt + i) - opts;
			memcpy(&sack, &opt[i], sizeof(sack));
			pf_patch_32(pd, &sack.start,
			    htonl(ntohl(sack.start) - dst->seqdiff),
			    PF_ALGNMNT(startoff));
			pf_patch_32(pd, &sack.end,
			    htonl(ntohl(sack.end) - dst->seqdiff),
			    PF_ALGNMNT(startoff + sizeof(sack.start)));
			memcpy(&opt[i], &sack, sizeof(sack));
		}
		copyback = 1;
		opt += opt[1];
	}

	if (copyback)
		m_copyback(pd->m, optsoff, olen, (caddr_t)opts);

	return (copyback);
}

struct mbuf *
pf_build_tcp(const struct pf_krule *r, sa_family_t af,
    const struct pf_addr *saddr, const struct pf_addr *daddr,
    u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
    u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl,
    int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, u_int sack,
    int rtableid)
{
	struct mbuf	*m;
	int		 len, tlen;
#ifdef INET
	struct ip	*h = NULL;
#endif /* INET */
#ifdef INET6
	struct ip6_hdr	*h6 = NULL;
#endif /* INET6 */
	struct tcphdr	*th;
	char		*opt;
	struct pf_mtag  *pf_mtag;

	len = 0;
	th = NULL;

	/* maximum segment size tcp option */
	tlen = sizeof(struct tcphdr);
	if (mss)
		tlen += 4;
	if (sack)
		tlen += 2;

	switch (af) {
#ifdef INET
	case AF_INET:
		len = sizeof(struct ip) + tlen;
		break;
#endif /* INET */
#ifdef INET6
	case AF_INET6:
		len = sizeof(struct ip6_hdr) + tlen;
		break;
#endif /* INET6 */
	default:
		unhandled_af(af);
	}

	m = m_gethdr(M_NOWAIT, MT_DATA);
	if (m == NULL)
		return (NULL);

#ifdef MAC
	mac_netinet_firewall_send(m);
#endif
	if ((pf_mtag = pf_get_mtag(m)) == NULL) {
		m_freem(m);
		return (NULL);
	}
	m->m_flags |= mbuf_flags;
	pf_mtag->tag = mtag_tag;
	pf_mtag->flags = mtag_flags;

	if (rtableid >= 0)
		M_SETFIB(m, rtableid);

#ifdef ALTQ
	if (r != NULL && r->qid) {
		pf_mtag->qid = r->qid;

		/* add hints for ecn */
		pf_mtag->hdr = mtod(m, struct ip *);
	}
#endif /* ALTQ */
	m->m_data += max_linkhdr;
	m->m_pkthdr.len = m->m_len = len;
	/* The rest of the stack assumes a rcvif, so provide one.
	 * This is a locally generated packet, so .. close enough. */
	m->m_pkthdr.rcvif = V_loif;
	bzero(m->m_data, len);
	switch (af) {
#ifdef INET
	case AF_INET:
		m->m_pkthdr.csum_flags |= CSUM_TCP;
		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);

		h = mtod(m, struct ip *);

		h->ip_p = IPPROTO_TCP;
		h->ip_len = htons(tlen);
		h->ip_v = 4;
		h->ip_hl = sizeof(*h) >> 2;
		h->ip_tos = IPTOS_LOWDELAY;
		h->ip_len = htons(len);
		h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
		h->ip_ttl = ttl ? ttl : V_ip_defttl;
		h->ip_sum = 0;
		h->ip_src.s_addr = saddr->v4.s_addr;
		h->ip_dst.s_addr = daddr->v4.s_addr;

		th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
		th->th_sum = in_pseudo(h->ip_src.s_addr, h->ip_dst.s_addr,
		    htons(len - sizeof(struct ip) + IPPROTO_TCP));
		break;
#endif /* INET */
#ifdef INET6
	case AF_INET6:
		m->m_pkthdr.csum_flags |= CSUM_TCP_IPV6;
		m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);

		h6 = mtod(m, struct ip6_hdr *);

		/* IP header fields included in the TCP checksum */
		h6->ip6_nxt = IPPROTO_TCP;
		h6->ip6_plen = htons(tlen);
		h6->ip6_vfc |= IPV6_VERSION;
		h6->ip6_hlim = V_ip6_defhlim;
		memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
		memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));

		th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
		th->th_sum = in6_cksum_pseudo(h6, len - sizeof(struct ip6_hdr),
		    IPPROTO_TCP, 0);
		break;
#endif /* INET6 */
	}

	/* TCP header */
	th->th_sport = sport;
	th->th_dport = dport;
	th->th_seq = htonl(seq);
	th->th_ack = htonl(ack);
	th->th_off = tlen >> 2;
	tcp_set_flags(th, tcp_flags);
	th->th_win = htons(win);

	opt = (char *)(th + 1);
	if (mss) {
		opt = (char *)(th + 1);
		opt[0] = TCPOPT_MAXSEG;
		opt[1] = 4;
		mss = htons(mss);
		memcpy((opt + 2), &mss, 2);
		opt += 4;
	}
	if (sack) {
		opt[0] = TCPOPT_SACK_PERMITTED;
		opt[1] = 2;
		opt += 2;
	}

	return (m);
}

static void
pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd,
    uint8_t ttl, int rtableid)
{
	struct mbuf		*m;
#ifdef INET
	struct ip		*h = NULL;
#endif /* INET */
#ifdef INET6
	struct ip6_hdr		*h6 = NULL;
#endif /* INET6 */
	struct sctphdr		*hdr;
	struct sctp_chunkhdr	*chunk;
	struct pf_send_entry	*pfse;
	int			 off = 0;

	MPASS(af == pd->af);

	m = m_gethdr(M_NOWAIT, MT_DATA);
	if (m == NULL)
		return;

	m->m_data += max_linkhdr;
	m->m_flags |= M_SKIP_FIREWALL;
	/* The rest of the stack assumes a rcvif, so provide one.
	 * This is a locally generated packet, so .. close enough. */
	m->m_pkthdr.rcvif = V_loif;

	/* IPv4|6 header */
	switch (af) {
#ifdef INET
	case AF_INET:
		bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk));

		h = mtod(m, struct ip *);

		/* IP header fields included in the TCP checksum */

		h->ip_p = IPPROTO_SCTP;
		h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk));
		h->ip_ttl = ttl ? ttl : V_ip_defttl;
		h->ip_src = pd->dst->v4;
		h->ip_dst = pd->src->v4;

		off += sizeof(struct ip);
		break;
#endif /* INET */
#ifdef INET6
	case AF_INET6:
		bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk));

		h6 = mtod(m, struct ip6_hdr *);

		/* IP header fields included in the TCP checksum */
		h6->ip6_vfc |= IPV6_VERSION;
		h6->ip6_nxt = IPPROTO_SCTP;
		h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk));
		h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim;
		memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr));
		memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr));

		off += sizeof(struct ip6_hdr);
		break;
#endif /* INET6 */
	default:
		unhandled_af(af);
	}

	/* SCTP header */
	hdr = mtodo(m, off);

	hdr->src_port = pd->hdr.sctp.dest_port;
	hdr->dest_port = pd->hdr.sctp.src_port;
	hdr->v_tag = pd->sctp_initiate_tag;
	hdr->checksum = 0;

	/* Abort chunk. */
	off += sizeof(struct sctphdr);
	chunk = mtodo(m, off);

	chunk->chunk_type = SCTP_ABORT_ASSOCIATION;
	chunk->chunk_length = htons(sizeof(*chunk));

	/* SCTP checksum */
	off += sizeof(*chunk);
	m->m_pkthdr.len = m->m_len = off;

	pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk));

	if (rtableid >= 0)
		M_SETFIB(m, rtableid);

	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
	if (pfse == NULL) {
		m_freem(m);
		return;
	}

	switch (af) {
#ifdef INET
	case AF_INET:
		pfse->pfse_type = PFSE_IP;
		break;
#endif /* INET */
#ifdef INET6
	case AF_INET6:
		pfse->pfse_type = PFSE_IP6;
		break;
#endif /* INET6 */
	}

	pfse->pfse_m = m;
	pf_send(pfse);
}

void
pf_send_tcp(const struct pf_krule *r, sa_family_t af,
    const struct pf_addr *saddr, const struct pf_addr *daddr,
    u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
    u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl,
    int mbuf_flags, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid)
{
	struct pf_send_entry *pfse;
	struct mbuf	*m;

	m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags,
	    win, mss, ttl, mbuf_flags, mtag_tag, mtag_flags, 0, rtableid);
	if (m == NULL)
		return;

	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
	if (pfse == NULL) {
		m_freem(m);
		return;
	}

	switch (af) {
#ifdef INET
	case AF_INET:
		pfse->pfse_type = PFSE_IP;
		break;
#endif /* INET */
#ifdef INET6
	case AF_INET6:
		pfse->pfse_type = PFSE_IP6;
		break;
#endif /* INET6 */
	default:
		unhandled_af(af);
	}

	pfse->pfse_m = m;
	pf_send(pfse);
}

static void
pf_undo_nat(struct pf_krule *nr, struct pf_pdesc *pd, uint16_t bip_sum)
{
	/* undo NAT changes, if they have taken place */
	if (nr != NULL) {
		pf_addrcpy(pd->src, &pd->osrc, pd->af);
		pf_addrcpy(pd->dst, &pd->odst, pd->af);
		if (pd->sport)
			*pd->sport = pd->osport;
		if (pd->dport)
			*pd->dport = pd->odport;
		if (pd->ip_sum)
			*pd->ip_sum = bip_sum;
		m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any);
	}
}

static void
pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd,
    struct tcphdr *th, u_int16_t bproto_sum, u_int16_t bip_sum,
    u_short *reason, int rtableid)
{
	pf_undo_nat(nr, pd, bip_sum);

	if (pd->proto == IPPROTO_TCP &&
	    ((r->rule_flag & PFRULE_RETURNRST) ||
	    (r->rule_flag & PFRULE_RETURN)) &&
	    !(tcp_get_flags(th) & TH_RST)) {
		u_int32_t	 ack = ntohl(th->th_seq) + pd->p_len;

		if (pf_check_proto_cksum(pd->m, pd->off, pd->tot_len - pd->off,
		    IPPROTO_TCP, pd->af))
			REASON_SET(reason, PFRES_PROTCKSUM);
		else {
			if (tcp_get_flags(th) & TH_SYN)
				ack++;
			if (tcp_get_flags(th) & TH_FIN)
				ack++;
			pf_send_tcp(r, pd->af, pd->dst,
				pd->src, th->th_dport, th->th_sport,
				ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
				r->return_ttl, M_SKIP_FIREWALL, 0, 0, rtableid);
		}
	} else if (pd->proto == IPPROTO_SCTP &&
	    (r->rule_flag & PFRULE_RETURN)) {
		pf_send_sctp_abort(pd->af, pd, r->return_ttl, rtableid);
	} else if (pd->proto != IPPROTO_ICMP && pd->af == AF_INET &&
		r->return_icmp)
		pf_send_icmp(pd->m, r->return_icmp >> 8,
			r->return_icmp & 255, 0, pd->af, r, rtableid);
	else if (pd->proto != IPPROTO_ICMPV6 && pd->af == AF_INET6 &&
		r->return_icmp6)
		pf_send_icmp(pd->m, r->return_icmp6 >> 8,
			r->return_icmp6 & 255, 0, pd->af, r, rtableid);
}

static int
pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
{
	struct m_tag *mtag;
	u_int8_t mpcp;

	mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
	if (mtag == NULL)
		return (0);

	if (prio == PF_PRIO_ZERO)
		prio = 0;

	mpcp = *(uint8_t *)(mtag + 1);

	return (mpcp == prio);
}

static int
pf_icmp_to_bandlim(uint8_t type)
{
	switch (type) {
		case ICMP_ECHO:
		case ICMP_ECHOREPLY:
			return (BANDLIM_ICMP_ECHO);
		case ICMP_TSTAMP:
		case ICMP_TSTAMPREPLY:
			return (BANDLIM_ICMP_TSTAMP);
		case ICMP_UNREACH:
		default:
			return (BANDLIM_ICMP_UNREACH);
	}
}

static void
pf_send_challenge_ack(struct pf_pdesc *pd, struct pf_kstate *s,
    struct pf_state_peer *src, struct pf_state_peer *dst)
{
	/*
	 * We are sending challenge ACK as a response to SYN packet, which
	 * matches existing state (modulo TCP window check). Therefore packet
	 * must be sent on behalf of destination.
	 *
	 * We expect sender to remain either silent, or send RST packet
	 * so both, firewall and remote peer, can purge dead state from
	 * memory.
	 */
	pf_send_tcp(s->rule, pd->af, pd->dst, pd->src,
	    pd->hdr.tcp.th_dport, pd->hdr.tcp.th_sport, dst->seqlo,
	    src->seqlo, TH_ACK, 0, 0, s->rule->return_ttl, 0, 0, 0,
	    s->rule->rtableid);
}

static void
pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, int mtu,
    sa_family_t af, struct pf_krule *r, int rtableid)
{
	struct pf_send_entry *pfse;
	struct mbuf *m0;
	struct pf_mtag *pf_mtag;

	/* ICMP packet rate limitation. */
	switch (af) {
#ifdef INET6
	case AF_INET6:
		if (icmp6_ratelimit(NULL, type, code))
			return;
		break;
#endif /* INET6 */
#ifdef INET
	case AF_INET:
		if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0)
			return;
		break;
#endif /* INET */
	}

	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
	if (pfse == NULL)
		return;

	if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
		free(pfse, M_PFTEMP);
		return;
	}

	if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
		free(pfse, M_PFTEMP);
		return;
	}
	/* XXX: revisit */
	m0->m_flags |= M_SKIP_FIREWALL;

	if (rtableid >= 0)
		M_SETFIB(m0, rtableid);

#ifdef ALTQ
	if (r->qid) {
		pf_mtag->qid = r->qid;
		/* add hints for ecn */
		pf_mtag->hdr = mtod(m0, struct ip *);
	}
#endif /* ALTQ */

	switch (af) {
#ifdef INET
	case AF_INET:
		pfse->pfse_type = PFSE_ICMP;
		break;
#endif /* INET */
#ifdef INET6
	case AF_INET6:
		pfse->pfse_type = PFSE_ICMP6;
		break;
#endif /* INET6 */
	}
	pfse->pfse_m = m0;
	pfse->icmpopts.type = type;
	pfse->icmpopts.code = code;
	pfse->icmpopts.mtu = mtu;
	pf_send(pfse);
}

/*
 * Return ((n = 0) == (a = b [with mask m]))
 * Note: n != 0 => returns (a != b [with mask m])
 */
int
pf_match_addr(u_int8_t n, const struct pf_addr *a, const struct pf_addr *m,
    const struct pf_addr *b, sa_family_t af)
{
	switch (af) {
#ifdef INET
	case AF_INET:
		if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4))
			return (n == 0);
		break;
#endif /* INET */
#ifdef INET6
	case AF_INET6:
		if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6))
			return (n == 0);
		break;
#endif /* INET6 */
	}

	return (n != 0);
}

/*
 * Return 1 if b <= a <= e, otherwise return 0.
 */
int
pf_match_addr_range(const struct pf_addr *b, const struct pf_addr *e,
    const struct pf_addr *a, sa_family_t af)
{
	switch (af) {
#ifdef INET
	case AF_INET:
		if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
		    (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
			return (0);
		break;
#endif /* INET */
#ifdef INET6
	case AF_INET6: {
		int	i;

		/* check a >= b */
		for (i = 0; i < 4; ++i)
			if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
				break;
			else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
				return (0);
		/* check a <= e */
		for (i = 0; i < 4; ++i)
			if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
				break;
			else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
				return (0);
		break;
	}
#endif /* INET6 */
	}
	return (1);
}

static int
pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
{
	switch (op) {
	case PF_OP_IRG:
		return ((p > a1) && (p < a2));
	case PF_OP_XRG:
		return ((p < a1) || (p > a2));
	case PF_OP_RRG:
		return ((p >= a1) && (p <= a2));
	case PF_OP_EQ:
		return (p == a1);
	case PF_OP_NE:
		return (p != a1);
	case PF_OP_LT:
		return (p < a1);
	case PF_OP_LE:
		return (p <= a1);
	case PF_OP_GT:
		return (p > a1);
	case PF_OP_GE:
		return (p >= a1);
	}
	return (0); /* never reached */
}

int
pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
{
	return (pf_match(op, ntohs(a1), ntohs(a2), ntohs(p)));
}

static int
pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
{
	if (u == -1 && op != PF_OP_EQ && op != PF_OP_NE)
		return (0);
	return (pf_match(op, a1, a2, u));
}

static int
pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
{
	if (g == -1 && op != PF_OP_EQ && op != PF_OP_NE)
		return (0);
	return (pf_match(op, a1, a2, g));
}

int
pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag)
{
	if (*tag == -1)
		*tag = mtag;

	return ((!r->match_tag_not && r->match_tag == *tag) ||
	    (r->match_tag_not && r->match_tag != *tag));
}

static int
pf_match_rcvif(struct mbuf *m, struct pf_krule *r)
{
	struct ifnet *ifp = m->m_pkthdr.rcvif;
	struct pfi_kkif *kif;

	if (ifp == NULL)
		return (0);

	kif = (struct pfi_kkif *)ifp->if_pf_kif;

	if (kif == NULL) {
		DPFPRINTF(PF_DEBUG_URGENT,
		    "%s: kif == NULL, @%d via %s", __func__, r->nr,
			r->rcv_ifname);
		return (0);
	}

	return (pfi_kkif_match(r->rcv_kif, kif));
}

int
pf_tag_packet(struct pf_pdesc *pd, int tag)
{

	KASSERT(tag > 0, ("%s: tag %d", __func__, tag));

	if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL))
		return (ENOMEM);

	pd->pf_mtag->tag = tag;

	return (0);
}

/*
 * XXX: We rely on malloc(9) returning pointer aligned addresses.
 */
#define	PF_ANCHORSTACK_MATCH	0x00000001
#define	PF_ANCHORSTACK_MASK	(PF_ANCHORSTACK_MATCH)

#define	PF_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
#define	PF_ANCHOR_RULE(f)	(struct pf_krule *)			\
				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
#define	PF_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 			\
				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
} while (0)

enum pf_test_status
pf_step_into_anchor(struct pf_test_ctx *ctx, struct pf_krule *r)
{
	enum pf_test_status	rv;

	PF_RULES_RASSERT();

	if (ctx->depth >= PF_ANCHOR_STACK_MAX) {
		printf("%s: anchor stack overflow on %s\n",
		    __func__, r->anchor->name);
		return (PF_TEST_FAIL);
	}

	ctx->depth++;

	if (r->anchor_wildcard) {
		struct pf_kanchor *child;
		rv = PF_TEST_OK;
		RB_FOREACH(child, pf_kanchor_node, &r->anchor->children) {
			rv = pf_match_rule(ctx, &child->ruleset);
			if ((rv == PF_TEST_QUICK) || (rv == PF_TEST_FAIL)) {
				/*
				 * we either hit a rule with quick action
				 * (more likely), or hit some runtime
				 * error (e.g. pool_get() failure).
				 */
				break;
			}
		}
	} else {
		rv = pf_match_rule(ctx, &r->anchor->ruleset);
		/*
		 * Unless errors occured, stop iff any rule matched
		 * within quick anchors.
		 */
		if (rv != PF_TEST_FAIL && r->quick == PF_TEST_QUICK &&
		    *ctx->am == r)
			rv = PF_TEST_QUICK;
	}

	ctx->depth--;

	return (rv);
}

struct pf_keth_anchor_stackframe {
	struct pf_keth_ruleset	*rs;
	struct pf_keth_rule	*r;	/* XXX: + match bit */
	struct pf_keth_anchor	*child;
};

#define	PF_ETH_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
#define	PF_ETH_ANCHOR_RULE(f)	(struct pf_keth_rule *)			\
				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
#define	PF_ETH_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 		\
				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
} while (0)

void
pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth,
    struct pf_keth_ruleset **rs, struct pf_keth_rule **r,
    struct pf_keth_rule **a, int *match)
{
	struct pf_keth_anchor_stackframe	*f;

	NET_EPOCH_ASSERT();

	if (match)
		*match = 0;
	if (*depth >= PF_ANCHOR_STACK_MAX) {
		printf("%s: anchor stack overflow on %s\n",
		    __func__, (*r)->anchor->name);
		*r = TAILQ_NEXT(*r, entries);
		return;
	} else if (*depth == 0 && a != NULL)
		*a = *r;
	f = stack + (*depth)++;
	f->rs = *rs;
	f->r = *r;
	if ((*r)->anchor_wildcard) {
		struct pf_keth_anchor_node *parent = &(*r)->anchor->children;

		if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) {
			*r = NULL;
			return;
		}
		*rs = &f->child->ruleset;
	} else {
		f->child = NULL;
		*rs = &(*r)->anchor->ruleset;
	}
	*r = TAILQ_FIRST((*rs)->active.rules);
}

int
pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth,
    struct pf_keth_ruleset **rs, struct pf_keth_rule **r,
    struct pf_keth_rule **a, int *match)
{
	struct pf_keth_anchor_stackframe	*f;
	struct pf_keth_rule *fr;
	int quick = 0;

	NET_EPOCH_ASSERT();

	do {
		if (*depth <= 0)
			break;
		f = stack + *depth - 1;
		fr = PF_ETH_ANCHOR_RULE(f);
		if (f->child != NULL) {
			/*
			 * This block traverses through
			 * a wildcard anchor.
			 */
			if (match != NULL && *match) {
				/*
				 * If any of "*" matched, then
				 * "foo/ *" matched, mark frame
				 * appropriately.
				 */
				PF_ETH_ANCHOR_SET_MATCH(f);
				*match = 0;
			}
			f->child = RB_NEXT(pf_keth_anchor_node,
			    &fr->anchor->children, f->child);
			if (f->child != NULL) {
				*rs = &f->child->ruleset;
				*r = TAILQ_FIRST((*rs)->active.rules);
				if (*r == NULL)
					continue;
				else
					break;
			}
		}
		(*depth)--;
		if (*depth == 0 && a != NULL)
			*a = NULL;
		*rs = f->rs;
		if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match))
			quick = fr->quick;
		*r = TAILQ_NEXT(fr, entries);
	} while (*r == NULL);

	return (quick);
}

void
pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
    struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
{
	switch (af) {
#ifdef INET
	case AF_INET:
		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
		break;
#endif /* INET */
#ifdef INET6
	case AF_INET6:
		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
		naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
		((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
		naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
		((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
		naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
		((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
		break;
#endif /* INET6 */
	}
}

void
pf_addr_inc(struct pf_addr *addr, sa_family_t af)
{
	switch (af) {
#ifdef INET
	case AF_INET:
		addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
		break;
#endif /* INET */
#ifdef INET6
	case AF_INET6:
		if (addr->addr32[3] == 0xffffffff) {
			addr->addr32[3] = 0;
			if (addr->addr32[2] == 0xffffffff) {
				addr->addr32[2] = 0;
				if (addr->addr32[1] == 0xffffffff) {
					addr->addr32[1] = 0;
					addr->addr32[0] =
					    htonl(ntohl(addr->addr32[0]) + 1);
				} else
					addr->addr32[1] =
					    htonl(ntohl(addr->addr32[1]) + 1);
			} else
				addr->addr32[2] =
				    htonl(ntohl(addr->addr32[2]) + 1);
		} else
			addr->addr32[3] =
			    htonl(ntohl(addr->addr32[3]) + 1);
		break;
#endif /* INET6 */
	}
}

void
pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a)
{
	/*
	 * Modern rules use the same flags in rules as they do in states.
	 */
	a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID|
	    PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO));

	/*
	 * Old-style scrub rules have different flags which need to be translated.
	 */
	if (r->rule_flag & PFRULE_RANDOMID)
		a->flags |= PFSTATE_RANDOMID;
	if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) {
		a->flags |= PFSTATE_SETTOS;
		a->set_tos = r->set_tos;
	}

	if (r->qid)
		a->qid = r->qid;
	if (r->pqid)
		a->pqid = r->pqid;
	if (r->rtableid >= 0)
		a->rtableid = r->rtableid;
	a->log |= r->log;
	if (r->min_ttl)
		a->min_ttl = r->min_ttl;
	if (r->max_mss)
		a->max_mss = r->max_mss;
	if (r->dnpipe)
		a->dnpipe = r->dnpipe;
	if (r->dnrpipe)
		a->dnrpipe = r->dnrpipe;
	if (r->dnpipe || r->dnrpipe) {
		if (r->free_flags & PFRULE_DN_IS_PIPE)
			a->flags |= PFSTATE_DN_IS_PIPE;
		else
			a->flags &= ~PFSTATE_DN_IS_PIPE;
	}
	if (r->scrub_flags & PFSTATE_SETPRIO) {
		a->set_prio[0] = r->set_prio[0];
		a->set_prio[1] = r->set_prio[1];
	}
	if (r->allow_opts)
		a->allow_opts = r->allow_opts;
	if (r->max_pkt_size)
		a->max_pkt_size = r->max_pkt_size;
}

int
pf_socket_lookup(struct pf_pdesc *pd)
{
	struct pf_addr		*saddr, *daddr;
	u_int16_t		 sport, dport;
	struct inpcbinfo	*pi;
	struct inpcb		*inp;

	pd->lookup.uid = -1;
	pd->lookup.gid = -1;

	switch (pd->proto) {
	case IPPROTO_TCP:
		sport = pd->hdr.tcp.th_sport;
		dport = pd->hdr.tcp.th_dport;
		pi = &V_tcbinfo;
		break;
	case IPPROTO_UDP:
		sport = pd->hdr.udp.uh_sport;
		dport = pd->hdr.udp.uh_dport;
		pi = &V_udbinfo;
		break;
	default:
		return (-1);
	}
	if (pd->dir == PF_IN) {
		saddr = pd->src;
		daddr = pd->dst;
	} else {
		u_int16_t	p;

		p = sport;
		sport = dport;
		dport = p;
		saddr = pd->dst;
		daddr = pd->src;
	}
	switch (pd->af) {
#ifdef INET
	case AF_INET:
		inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
		    dport, INPLOOKUP_RLOCKPCB, NULL, pd->m);
		if (inp == NULL) {
			inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
			   daddr->v4, dport, INPLOOKUP_WILDCARD |
			   INPLOOKUP_RLOCKPCB, NULL, pd->m);
			if (inp == NULL)
				return (-1);
		}
		break;
#endif /* INET */
#ifdef INET6
	case AF_INET6:
		inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
		    dport, INPLOOKUP_RLOCKPCB, NULL, pd->m);
		if (inp == NULL) {
			inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
			    &daddr->v6, dport, INPLOOKUP_WILDCARD |
			    INPLOOKUP_RLOCKPCB, NULL, pd->m);
			if (inp == NULL)
				return (-1);
		}
		break;
#endif /* INET6 */
	default:
		unhandled_af(pd->af);
	}
	INP_RLOCK_ASSERT(inp);
	pd->lookup.uid = inp->inp_cred->cr_uid;
	pd->lookup.gid = inp->inp_cred->cr_gid;
	INP_RUNLOCK(inp);

	return (1);
}

/* post: r  => (r[0] == type /\ r[1] >= min_typelen >= 2  "validity"
 *                      /\ (eoh - r) >= min_typelen >= 2  "safety"  )
 *
 * warning: r + r[1] may exceed opts bounds for r[1] > min_typelen
 */
uint8_t*
pf_find_tcpopt(u_int8_t *opt, u_int8_t *opts, size_t hlen, u_int8_t type,
    u_int8_t min_typelen)
{
	uint8_t	*eoh = opts + hlen;

	if (min_typelen < 2)
		return (NULL);

	while ((eoh - opt) >= min_typelen) {
		switch (*opt) {
		case TCPOPT_EOL:
			/* FALLTHROUGH - Workaround the failure of some
			 systems to NOP-pad their bzero'd option buffers,
			 producing spurious EOLs */
		case TCPOPT_NOP:
			opt++;
			continue;
		default:
		if (opt[0] == type &&
			    opt[1] >= min_typelen)
			return (opt);
		}

		opt += MAX(opt[1], 2); /* evade infinite loops */
	}

	return (NULL);
}

u_int8_t
pf_get_wscale(struct pf_pdesc *pd)
{
	int	 olen;
	uint8_t	 opts[MAX_TCPOPTLEN], *opt;
	uint8_t	 wscale = 0;

	olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr);
	if (olen < TCPOLEN_WINDOW || !pf_pull_hdr(pd->m,
	    pd->off + sizeof(struct tcphdr), opts, olen, NULL, NULL, pd->af))
		return (0);

	opt = opts;
	while ((opt = pf_find_tcpopt(opt, opts, olen,
		    TCPOPT_WINDOW, TCPOLEN_WINDOW)) != NULL) {
		wscale = opt[2];
		wscale = MIN(wscale, TCP_MAX_WINSHIFT);
		wscale |= PF_WSCALE_FLAG;

		opt += opt[1];
	}

	return (wscale);
}

u_int16_t
pf_get_mss(struct pf_pdesc *pd)
{
	int		 olen;
	uint8_t		 opts[MAX_TCPOPTLEN], *opt;
	u_int16_t	 mss = V_tcp_mssdflt;

	olen = (pd->hdr.tcp.th_off << 2) - sizeof(struct tcphdr);
	if (olen < TCPOLEN_MAXSEG || !pf_pull_hdr(pd->m,
	    pd->off + sizeof(struct tcphdr), opts, olen, NULL, NULL, pd->af))
		return (0);

	opt = opts;
	while ((opt = pf_find_tcpopt(opt, opts, olen,
	    TCPOPT_MAXSEG, TCPOLEN_MAXSEG)) != NULL) {
		memcpy(&mss, (opt + 2), 2);
		mss = ntohs(mss);
		opt += opt[1];
	}

	return (mss);
}

static u_int16_t
pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
{
	struct nhop_object *nh;
#ifdef INET6
	struct in6_addr		dst6;
	uint32_t		scopeid;
#endif /* INET6 */
	int			 hlen = 0;
	uint16_t		 mss = 0;

	NET_EPOCH_ASSERT();

	switch (af) {
#ifdef INET
	case AF_INET:
		hlen = sizeof(struct ip);
		nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0);
		if (nh != NULL)
			mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
		break;
#endif /* INET */
#ifdef INET6
	case AF_INET6:
		hlen = sizeof(struct ip6_hdr);
		in6_splitscope(&addr->v6, &dst6, &scopeid);
		nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0);
		if (nh != NULL)
			mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
		break;
#endif /* INET6 */
	}

	mss = max(V_tcp_mssdflt, mss);
	mss = min(mss, offer);
	mss = max(mss, 64);		/* sanity - at least max opt space */
	return (mss);
}

static u_int32_t
pf_tcp_iss(struct pf_pdesc *pd)
{
	SHA512_CTX ctx;
	union {
		uint8_t bytes[SHA512_DIGEST_LENGTH];
		uint32_t words[1];
	} digest;

	if (V_pf_tcp_secret_init == 0) {
		arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
		SHA512_Init(&V_pf_tcp_secret_ctx);
		SHA512_Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
		    sizeof(V_pf_tcp_secret));
		V_pf_tcp_secret_init = 1;
	}

	ctx = V_pf_tcp_secret_ctx;

	SHA512_Update(&ctx, &pd->hdr.tcp.th_sport, sizeof(u_short));
	SHA512_Update(&ctx, &pd->hdr.tcp.th_dport, sizeof(u_short));
	switch (pd->af) {
	case AF_INET6:
		SHA512_Update(&ctx, &pd->src->v6, sizeof(struct in6_addr));
		SHA512_Update(&ctx, &pd->dst->v6, sizeof(struct in6_addr));
		break;
	case AF_INET:
		SHA512_Update(&ctx, &pd->src->v4, sizeof(struct in_addr));
		SHA512_Update(&ctx, &pd->dst->v4, sizeof(struct in_addr));
		break;
	}
	SHA512_Final(digest.bytes, &ctx);
	V_pf_tcp_iss_off += 4096;
#define	ISN_RANDOM_INCREMENT (4096 - 1)
	return (digest.words[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
	    V_pf_tcp_iss_off);
#undef	ISN_RANDOM_INCREMENT
}

static bool
pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r)
{
	bool match = true;

	/* Always matches if not set */
	if (! r->isset)
		return (!r->neg);

	for (int i = 0; i < ETHER_ADDR_LEN; i++) {
		if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) {
			match = false;
			break;
		}
	}

	return (match ^ r->neg);
}

static int
pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag)
{
	if (*tag == -1)
		*tag = mtag;

	return ((!r->match_tag_not && r->match_tag == *tag) ||
	    (r->match_tag_not && r->match_tag != *tag));
}

static void
pf_bridge_to(struct ifnet *ifp, struct mbuf *m)
{
	/* If we don't have the interface drop the packet. */
	if (ifp == NULL) {
		m_freem(m);
		return;
	}

	switch (ifp->if_type) {
	case IFT_ETHER:
	case IFT_XETHER:
	case IFT_L2VLAN:
	case IFT_BRIDGE:
	case IFT_IEEE8023ADLAG:
		break;
	default:
		m_freem(m);
		return;
	}

	ifp->if_transmit(ifp, m);
}

static int
pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0)
{
#ifdef INET
	struct ip ip;
#endif /* INET */
#ifdef INET6
	struct ip6_hdr ip6;
#endif /* INET6 */
	struct mbuf *m = *m0;
	struct ether_header *e;
	struct pf_keth_rule *r, *rm, *a = NULL;
	struct pf_keth_ruleset *ruleset = NULL;
	struct pf_mtag *mtag;
	struct pf_keth_ruleq *rules;
	struct pf_addr *src = NULL, *dst = NULL;
	struct pfi_kkif *bridge_to;
	sa_family_t af = 0;
	uint16_t proto;
	int asd = 0, match = 0;
	int tag = -1;
	uint8_t action;
	struct pf_keth_anchor_stackframe	anchor_stack[PF_ANCHOR_STACK_MAX];

	MPASS(kif->pfik_ifp->if_vnet == curvnet);
	NET_EPOCH_ASSERT();

	PF_RULES_RLOCK_TRACKER;

	SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m);

	mtag = pf_find_mtag(m);
	if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
		/* Dummynet re-injects packets after they've
		 * completed their delay. We've already
		 * processed them, so pass unconditionally. */

		/* But only once. We may see the packet multiple times (e.g.
		 * PFIL_IN/PFIL_OUT). */
		pf_dummynet_flag_remove(m, mtag);

		return (PF_PASS);
	}

	if (__predict_false(m->m_len < sizeof(struct ether_header)) &&
	    (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) {
		DPFPRINTF(PF_DEBUG_URGENT,
		    "%s: m_len < sizeof(struct ether_header)"
		     ", pullup failed", __func__);
		return (PF_DROP);
	}
	e = mtod(m, struct ether_header *);
	proto = ntohs(e->ether_type);

	switch (proto) {
#ifdef INET
	case ETHERTYPE_IP: {
		if (m_length(m, NULL) < (sizeof(struct ether_header) +
		    sizeof(ip)))
			return (PF_DROP);

		af = AF_INET;
		m_copydata(m, sizeof(struct ether_header), sizeof(ip),
		    (caddr_t)&ip);
		src = (struct pf_addr *)&ip.ip_src;
		dst = (struct pf_addr *)&ip.ip_dst;
		break;
	}
#endif /* INET */
#ifdef INET6
	case ETHERTYPE_IPV6: {
		if (m_length(m, NULL) < (sizeof(struct ether_header) +
		    sizeof(ip6)))
			return (PF_DROP);

		af = AF_INET6;
		m_copydata(m, sizeof(struct ether_header), sizeof(ip6),
		    (caddr_t)&ip6);
		src = (struct pf_addr *)&ip6.ip6_src;
		dst = (struct pf_addr *)&ip6.ip6_dst;
		break;
	}
#endif /* INET6 */
	}

	PF_RULES_RLOCK();

	ruleset = V_pf_keth;
	rules = atomic_load_ptr(&ruleset->active.rules);
	for (r = TAILQ_FIRST(rules), rm = NULL; r != NULL;) {
		counter_u64_add(r->evaluations, 1);
		SDT_PROBE2(pf, eth, test_rule, test, r->nr, r);

		if (pfi_kkif_match(r->kif, kif) == r->ifnot) {
			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
			    "kif");
			r = r->skip[PFE_SKIP_IFP].ptr;
		}
		else if (r->direction && r->direction != dir) {
			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
			    "dir");
			r = r->skip[PFE_SKIP_DIR].ptr;
		}
		else if (r->proto && r->proto != proto) {
			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
			    "proto");
			r = r->skip[PFE_SKIP_PROTO].ptr;
		}
		else if (! pf_match_eth_addr(e->ether_shost, &r->src)) {
			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
			    "src");
			r = r->skip[PFE_SKIP_SRC_ADDR].ptr;
		}
		else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) {
			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
			    "dst");
			r = r->skip[PFE_SKIP_DST_ADDR].ptr;
		}
		else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af,
		    r->ipsrc.neg, kif, M_GETFIB(m))) {
			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
			    "ip_src");
			r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr;
		}
		else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af,
		    r->ipdst.neg, kif, M_GETFIB(m))) {
			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
			    "ip_dst");
			r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr;
		}
		else if (r->match_tag && !pf_match_eth_tag(m, r, &tag,
		    mtag ? mtag->tag : 0)) {
			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
			    "match_tag");
			r = TAILQ_NEXT(r, entries);
		}
		else {
			if (r->tag)
				tag = r->tag;
			if (r->anchor == NULL) {
				/* Rule matches */
				rm = r;

				SDT_PROBE2(pf, eth, test_rule, match, r->nr, r);

				if (r->quick)
					break;

				r = TAILQ_NEXT(r, entries);
			} else {
				pf_step_into_keth_anchor(anchor_stack, &asd,
				    &ruleset, &r, &a, &match);
			}
		}
		if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd,
		    &ruleset, &r, &a, &match))
			break;
	}

	r = rm;

	SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r);

	/* Default to pass. */
	if (r == NULL) {
		PF_RULES_RUNLOCK();
		return (PF_PASS);
	}

	/* Execute action. */
	counter_u64_add(r->packets[dir == PF_OUT], 1);
	counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL));
	pf_update_timestamp(r);

	/* Shortcut. Don't tag if we're just going to drop anyway. */
	if (r->action == PF_DROP) {
		PF_RULES_RUNLOCK();
		return (PF_DROP);
	}

	if (tag > 0) {
		if (mtag == NULL)
			mtag = pf_get_mtag(m);
		if (mtag == NULL) {
			PF_RULES_RUNLOCK();
			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
			return (PF_DROP);
		}
		mtag->tag = tag;
	}

	if (r->qid != 0) {
		if (mtag == NULL)
			mtag = pf_get_mtag(m);
		if (mtag == NULL) {
			PF_RULES_RUNLOCK();
			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
			return (PF_DROP);
		}
		mtag->qid = r->qid;
	}

	action = r->action;
	bridge_to = r->bridge_to;

	/* Dummynet */
	if (r->dnpipe) {
		struct ip_fw_args dnflow;

		/* Drop packet if dummynet is not loaded. */
		if (ip_dn_io_ptr == NULL) {
			PF_RULES_RUNLOCK();
			m_freem(m);
			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
			return (PF_DROP);
		}
		if (mtag == NULL)
			mtag = pf_get_mtag(m);
		if (mtag == NULL) {
			PF_RULES_RUNLOCK();
			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
			return (PF_DROP);
		}

		bzero(&dnflow, sizeof(dnflow));

		/* We don't have port numbers here, so we set 0.  That means
		 * that we'll be somewhat limited in distinguishing flows (i.e.
		 * only based on IP addresses, not based on port numbers), but
		 * it's better than nothing. */
		dnflow.f_id.dst_port = 0;
		dnflow.f_id.src_port = 0;
		dnflow.f_id.proto = 0;

		dnflow.rule.info = r->dnpipe;
		dnflow.rule.info |= IPFW_IS_DUMMYNET;
		if (r->dnflags & PFRULE_DN_IS_PIPE)
			dnflow.rule.info |= IPFW_IS_PIPE;

		dnflow.f_id.extra = dnflow.rule.info;

		dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT;
		dnflow.flags |= IPFW_ARGS_ETHER;
		dnflow.ifp = kif->pfik_ifp;

		switch (af) {
		case AF_INET:
			dnflow.f_id.addr_type = 4;
			dnflow.f_id.src_ip = src->v4.s_addr;
			dnflow.f_id.dst_ip = dst->v4.s_addr;
			break;
		case AF_INET6:
			dnflow.flags |= IPFW_ARGS_IP6;
			dnflow.f_id.addr_type = 6;
			dnflow.f_id.src_ip6 = src->v6;
			dnflow.f_id.dst_ip6 = dst->v6;
			break;
		}

		PF_RULES_RUNLOCK();

		mtag->flags |= PF_MTAG_FLAG_DUMMYNET;
		ip_dn_io_ptr(m0, &dnflow);
		if (*m0 != NULL)
			pf_dummynet_flag_remove(m, mtag);
	} else {
		PF_RULES_RUNLOCK();
	}

	if (action == PF_PASS && bridge_to) {
		pf_bridge_to(bridge_to->pfik_ifp, *m0);
		*m0 = NULL; /* We've eaten the packet. */
	}

	return (action);
}

#define PF_TEST_ATTRIB(t, a)		\
	if (t) {			\
		r = a;			\
		continue;		\
	} else do {			\
	} while (0)

static __inline u_short
pf_rule_apply_nat(struct pf_test_ctx *ctx, struct pf_krule *r)
{
	struct pf_pdesc	*pd = ctx->pd;
	u_short		 transerror;
	u_int8_t	 nat_action;

	if (r->rule_flag & PFRULE_AFTO) {
		/* Don't translate if there was an old style NAT rule */
		if (ctx->nr != NULL)
			return (PFRES_TRANSLATE);

		/* pass af-to rules, unsupported on match rules */
		KASSERT(r->action != PF_MATCH, ("%s: af-to on match rule", __func__));
		/* XXX I can imagine scenarios where we have both NAT and RDR source tracking */
		ctx->nat_pool = &(r->nat);
		ctx->nr = r;
		pd->naf = r->naf;
		if (pf_get_transaddr_af(ctx->nr, pd) == -1) {
			return (PFRES_TRANSLATE);
		}
		return (PFRES_MATCH);
	} else if (r->rdr.cur || r->nat.cur) {
		/* Don't translate if there was an old style NAT rule */
		if (ctx->nr != NULL)
			return (PFRES_TRANSLATE);

		/* match/pass nat-to/rdr-to rules */
		ctx->nr = r;
		if (r->nat.cur) {
			nat_action = PF_NAT;
			ctx->nat_pool = &(r->nat);
		} else {
			nat_action = PF_RDR;
			ctx->nat_pool = &(r->rdr);
		}

		transerror = pf_get_transaddr(ctx, ctx->nr,
		    nat_action, ctx->nat_pool);
		if (transerror == PFRES_MATCH) {
			ctx->rewrite += pf_translate_compat(ctx);
			return(PFRES_MATCH);
		}
		return (transerror);
	}

	return (PFRES_MAX);
}

enum pf_test_status
pf_match_rule(struct pf_test_ctx *ctx, struct pf_kruleset *ruleset)
{
	struct pf_krule_item	*ri;
	struct pf_krule		*r;
	struct pf_krule		*save_a;
	struct pf_kruleset	*save_aruleset;
	struct pf_pdesc		*pd = ctx->pd;
	u_short			 transerror;

	r = TAILQ_FIRST(ruleset->rules[PF_RULESET_FILTER].active.ptr);
	while (r != NULL) {
		if (ctx->pd->related_rule) {
			*ctx->rm = ctx->pd->related_rule;
			break;
		}
		pf_counter_u64_add(&r->evaluations, 1);
		PF_TEST_ATTRIB(pfi_kkif_match(r->kif, pd->kif) == r->ifnot,
			r->skip[PF_SKIP_IFP]);
		PF_TEST_ATTRIB(r->direction && r->direction != pd->dir,
			r->skip[PF_SKIP_DIR]);
		PF_TEST_ATTRIB(r->af && r->af != pd->af,
			r->skip[PF_SKIP_AF]);
		PF_TEST_ATTRIB(r->proto && r->proto != pd->proto,
			r->skip[PF_SKIP_PROTO]);
		PF_TEST_ATTRIB(PF_MISMATCHAW(&r->src.addr, &pd->nsaddr, pd->naf,
		    r->src.neg, pd->kif, M_GETFIB(pd->m)),
			r->skip[PF_SKIP_SRC_ADDR]);
		PF_TEST_ATTRIB(PF_MISMATCHAW(&r->dst.addr, &pd->ndaddr, pd->af,
		    r->dst.neg, NULL, M_GETFIB(pd->m)),
			r->skip[PF_SKIP_DST_ADDR]);
		switch (pd->virtual_proto) {
		case PF_VPROTO_FRAGMENT:
			/* tcp/udp only. port_op always 0 in other cases */
			PF_TEST_ATTRIB((r->src.port_op || r->dst.port_op),
				TAILQ_NEXT(r, entries));
			PF_TEST_ATTRIB((pd->proto == IPPROTO_TCP && r->flagset),
				TAILQ_NEXT(r, entries));
			/* icmp only. type/code always 0 in other cases */
			PF_TEST_ATTRIB((r->type || r->code),
				TAILQ_NEXT(r, entries));
			/* tcp/udp only. {uid|gid}.op always 0 in other cases */
			PF_TEST_ATTRIB((r->gid.op || r->uid.op),
				TAILQ_NEXT(r, entries));
			break;

		case IPPROTO_TCP:
			PF_TEST_ATTRIB((r->flagset & tcp_get_flags(ctx->th))
			    != r->flags,
				TAILQ_NEXT(r, entries));
			/* FALLTHROUGH */
		case IPPROTO_SCTP:
		case IPPROTO_UDP:
			/* tcp/udp only. port_op always 0 in other cases */
			PF_TEST_ATTRIB(r->src.port_op && !pf_match_port(r->src.port_op,
			    r->src.port[0], r->src.port[1], pd->nsport),
				r->skip[PF_SKIP_SRC_PORT]);
			/* tcp/udp only. port_op always 0 in other cases */
			PF_TEST_ATTRIB(r->dst.port_op && !pf_match_port(r->dst.port_op,
			    r->dst.port[0], r->dst.port[1], pd->ndport),
				r->skip[PF_SKIP_DST_PORT]);
			/* tcp/udp only. uid.op always 0 in other cases */
			PF_TEST_ATTRIB(r->uid.op && (pd->lookup.done || (pd->lookup.done =
			    pf_socket_lookup(pd), 1)) &&
			    !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
			    pd->lookup.uid),
				TAILQ_NEXT(r, entries));
			/* tcp/udp only. gid.op always 0 in other cases */
			PF_TEST_ATTRIB(r->gid.op && (pd->lookup.done || (pd->lookup.done =
			    pf_socket_lookup(pd), 1)) &&
			    !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
			    pd->lookup.gid),
				TAILQ_NEXT(r, entries));
			break;

		case IPPROTO_ICMP:
		case IPPROTO_ICMPV6:
			/* icmp only. type always 0 in other cases */
			PF_TEST_ATTRIB(r->type && r->type != ctx->icmptype + 1,
				TAILQ_NEXT(r, entries));
			/* icmp only. type always 0 in other cases */
			PF_TEST_ATTRIB(r->code && r->code != ctx->icmpcode + 1,
				TAILQ_NEXT(r, entries));
			break;

		default:
			break;
		}
		PF_TEST_ATTRIB(r->tos && !(r->tos == pd->tos),
			TAILQ_NEXT(r, entries));
		PF_TEST_ATTRIB(r->prio &&
		    !pf_match_ieee8021q_pcp(r->prio, pd->m),
			TAILQ_NEXT(r, entries));
		PF_TEST_ATTRIB(r->prob &&
		    r->prob <= arc4random(),
			TAILQ_NEXT(r, entries));
		PF_TEST_ATTRIB(r->match_tag && !pf_match_tag(pd->m, r,
		    &ctx->tag, pd->pf_mtag ? pd->pf_mtag->tag : 0),
			TAILQ_NEXT(r, entries));
		PF_TEST_ATTRIB((r->rcv_kif && pf_match_rcvif(pd->m, r) ==
		   r->rcvifnot),
			TAILQ_NEXT(r, entries));
		PF_TEST_ATTRIB((r->rule_flag & PFRULE_FRAGMENT &&
		    pd->virtual_proto != PF_VPROTO_FRAGMENT),
			TAILQ_NEXT(r, entries));
		PF_TEST_ATTRIB(r->os_fingerprint != PF_OSFP_ANY &&
		    (pd->virtual_proto != IPPROTO_TCP || !pf_osfp_match(
		    pf_osfp_fingerprint(pd, ctx->th),
		    r->os_fingerprint)),
			TAILQ_NEXT(r, entries));
		/* must be last! */
		if (r->pktrate.limit) {
			PF_TEST_ATTRIB((pf_check_threshold(&r->pktrate)),
			    TAILQ_NEXT(r, entries));
		}
		/* FALLTHROUGH */
		if (r->tag)
			ctx->tag = r->tag;
		if (r->anchor == NULL) {
			if (r->action == PF_MATCH) {
				/*
				 * Apply translations before increasing counters,
				 * in case it fails.
				 */
				transerror = pf_rule_apply_nat(ctx, r);
				switch (transerror) {
				case PFRES_MATCH:
					/* Translation action found in rule and applied successfully */
				case PFRES_MAX:
					/* No translation action found in rule */
					break;
				default:
					/* Translation action found in rule but failed to apply */
					REASON_SET(&ctx->reason, transerror);
					return (PF_TEST_FAIL);
				}
				ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO);
				if (ri == NULL) {
					REASON_SET(&ctx->reason, PFRES_MEMORY);
					return (PF_TEST_FAIL);
				}
				ri->r = r;
				SLIST_INSERT_HEAD(&ctx->rules, ri, entry);
				pf_counter_u64_critical_enter();
				pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
				pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
				pf_counter_u64_critical_exit();
				pf_rule_to_actions(r, &pd->act);
				if (r->log)
					PFLOG_PACKET(r->action, PFRES_MATCH, r,
					    ctx->a, ruleset, pd, 1, NULL);
			} else {
				/*
				 * found matching r
				 */
				*ctx->rm = r;
				/*
				 * anchor, with ruleset, where r belongs to
				 */
				*ctx->am = ctx->a;
				/*
				 * ruleset where r belongs to
				 */
				*ctx->rsm = ruleset;
				/*
				 * ruleset, where anchor belongs to.
				 */
				ctx->arsm = ctx->aruleset;
			}
			if (pd->act.log & PF_LOG_MATCHES)
				pf_log_matches(pd, r, ctx->a, ruleset, &ctx->rules);
			if (r->quick) {
				ctx->test_status = PF_TEST_QUICK;
				break;
			}
		} else {
			save_a = ctx->a;
			save_aruleset = ctx->aruleset;

			ctx->a = r;			/* remember anchor */
			ctx->aruleset = ruleset;	/* and its ruleset */
			if (ctx->a->quick)
				ctx->test_status = PF_TEST_QUICK;
			/*
			 * Note: we don't need to restore if we are not going
			 * to continue with ruleset evaluation.
			 */
			if (pf_step_into_anchor(ctx, r) != PF_TEST_OK) {
				break;
			}
			ctx->a = save_a;
			ctx->aruleset = save_aruleset;
		}
		r = TAILQ_NEXT(r, entries);
	}

	return (ctx->test_status);
}

static int
pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm,
    struct pf_pdesc *pd, struct pf_krule **am,
    struct pf_kruleset **rsm, u_short *reason, struct inpcb *inp)
{
	struct pf_krule		*r = NULL;
	struct pf_kruleset	*ruleset = NULL;
	struct pf_krule_item	*ri;
	struct pf_test_ctx	 ctx;
	u_short			 transerror;
	int			 action = PF_PASS;
	u_int16_t		 bproto_sum = 0, bip_sum = 0;
	enum pf_test_status	 rv;

	PF_RULES_RASSERT();

	bzero(&ctx, sizeof(ctx));
	ctx.tag = -1;
	ctx.pd = pd;
	ctx.rm = rm;
	ctx.am = am;
	ctx.rsm = rsm;
	ctx.th = &pd->hdr.tcp;
	ctx.reason = *reason;
	SLIST_INIT(&ctx.rules);

	pf_addrcpy(&pd->nsaddr, pd->src, pd->af);
	pf_addrcpy(&pd->ndaddr, pd->dst, pd->af);

	if (inp != NULL) {
		INP_LOCK_ASSERT(inp);
		pd->lookup.uid = inp->inp_cred->cr_uid;
		pd->lookup.gid = inp->inp_cred->cr_gid;
		pd->lookup.done = 1;
	}

	if (pd->ip_sum)
		bip_sum = *pd->ip_sum;

	switch (pd->virtual_proto) {
	case IPPROTO_TCP:
		bproto_sum = ctx.th->th_sum;
		pd->nsport = ctx.th->th_sport;
		pd->ndport = ctx.th->th_dport;
		break;
	case IPPROTO_UDP:
		bproto_sum = pd->hdr.udp.uh_sum;
		pd->nsport = pd->hdr.udp.uh_sport;
		pd->ndport = pd->hdr.udp.uh_dport;
		break;
	case IPPROTO_SCTP:
		pd->nsport = pd->hdr.sctp.src_port;
		pd->ndport = pd->hdr.sctp.dest_port;
		break;
#ifdef INET
	case IPPROTO_ICMP:
		MPASS(pd->af == AF_INET);
		ctx.icmptype = pd->hdr.icmp.icmp_type;
		ctx.icmpcode = pd->hdr.icmp.icmp_code;
		ctx.state_icmp = pf_icmp_mapping(pd, ctx.icmptype,
		    &ctx.icmp_dir, &ctx.virtual_id, &ctx.virtual_type);
		if (ctx.icmp_dir == PF_IN) {
			pd->nsport = ctx.virtual_id;
			pd->ndport = ctx.virtual_type;
		} else {
			pd->nsport = ctx.virtual_type;
			pd->ndport = ctx.virtual_id;
		}
		break;
#endif /* INET */
#ifdef INET6
	case IPPROTO_ICMPV6:
		MPASS(pd->af == AF_INET6);
		ctx.icmptype = pd->hdr.icmp6.icmp6_type;
		ctx.icmpcode = pd->hdr.icmp6.icmp6_code;
		ctx.state_icmp = pf_icmp_mapping(pd, ctx.icmptype,
		    &ctx.icmp_dir, &ctx.virtual_id, &ctx.virtual_type);
		if (ctx.icmp_dir == PF_IN) {
			pd->nsport = ctx.virtual_id;
			pd->ndport = ctx.virtual_type;
		} else {
			pd->nsport = ctx.virtual_type;
			pd->ndport = ctx.virtual_id;
		}

		break;
#endif /* INET6 */
	default:
		pd->nsport = pd->ndport = 0;
		break;
	}
	pd->osport = pd->nsport;
	pd->odport = pd->ndport;

	/* check packet for BINAT/NAT/RDR */
	transerror = pf_get_translation(&ctx);
	switch (transerror) {
	default:
		/* A translation error occurred. */
		REASON_SET(&ctx.reason, transerror);
		goto cleanup;
	case PFRES_MAX:
		/* No match. */
		break;
	case PFRES_MATCH:
		KASSERT(ctx.sk != NULL, ("%s: null sk", __func__));
		KASSERT(ctx.nk != NULL, ("%s: null nk", __func__));
		if (ctx.nr->log) {
			PFLOG_PACKET(ctx.nr->action, PFRES_MATCH, ctx.nr, ctx.a,
			    ruleset, pd, 1, NULL);
		}

		ctx.rewrite += pf_translate_compat(&ctx);
		ctx.nat_pool = &(ctx.nr->rdr);
	}

	ruleset = &pf_main_ruleset;
	rv = pf_match_rule(&ctx, ruleset);
	if (rv == PF_TEST_FAIL) {
		/*
		 * Reason has been set in pf_match_rule() already.
		 */
		goto cleanup;
	}

	r = *ctx.rm;			/* matching rule */
	ctx.a = *ctx.am;		/* rule that defines an anchor containing 'r' */
	ruleset = *ctx.rsm;		/* ruleset of the anchor defined by the rule 'a' */
	ctx.aruleset = ctx.arsm;	/* ruleset of the 'a' rule itself */

	REASON_SET(&ctx.reason, PFRES_MATCH);

	/* apply actions for last matching pass/block rule */
	pf_rule_to_actions(r, &pd->act);
	transerror = pf_rule_apply_nat(&ctx, r);
	switch (transerror) {
	case PFRES_MATCH:
		/* Translation action found in rule and applied successfully */
	case PFRES_MAX:
		/* No translation action found in rule */
		break;
	default:
		/* Translation action found in rule but failed to apply */
		REASON_SET(&ctx.reason, transerror);
		goto cleanup;
	}

	if (r->log) {
		if (ctx.rewrite)
			m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any);
		PFLOG_PACKET(r->action, ctx.reason, r, ctx.a, ruleset, pd, 1, NULL);
	}
	if (pd->act.log & PF_LOG_MATCHES)
		pf_log_matches(pd, r, ctx.a, ruleset, &ctx.rules);
	if (pd->virtual_proto != PF_VPROTO_FRAGMENT &&
	   (r->action == PF_DROP) &&
	    ((r->rule_flag & PFRULE_RETURNRST) ||
	    (r->rule_flag & PFRULE_RETURNICMP) ||
	    (r->rule_flag & PFRULE_RETURN))) {
		pf_return(r, ctx.nr, pd, ctx.th, bproto_sum,
		    bip_sum, &ctx.reason, r->rtableid);
	}

	if (r->action == PF_DROP)
		goto cleanup;

	if (ctx.tag > 0 && pf_tag_packet(pd, ctx.tag)) {
		REASON_SET(&ctx.reason, PFRES_MEMORY);
		goto cleanup;
	}
	if (pd->act.rtableid >= 0)
		M_SETFIB(pd->m, pd->act.rtableid);

	if (r->rt) {
		/*
		 * Set act.rt here instead of in pf_rule_to_actions() because
		 * it is applied only from the last pass rule. For rules
		 * with the prefer-ipv6-nexthop option act.rt_af is a hint
		 * about AF of the forwarded packet and might be changed.
		 */
		pd->act.rt = r->rt;
		if (r->rt == PF_REPLYTO)
			pd->act.rt_af = pd->af;
		else
			pd->act.rt_af = pd->naf;
		if ((transerror = pf_map_addr_sn(pd->af, r, pd->src,
		    &pd->act.rt_addr, &pd->act.rt_af, &pd->act.rt_kif, NULL,
		    &(r->route), PF_SN_ROUTE)) != PFRES_MATCH) {
			REASON_SET(&ctx.reason, transerror);
			goto cleanup;
		}
	}

	if (pd->virtual_proto != PF_VPROTO_FRAGMENT &&
	   (!ctx.state_icmp && (r->keep_state || ctx.nr != NULL ||
	    (pd->flags & PFDESC_TCP_NORM)))) {
		bool nat64;

		action = pf_create_state(r, &ctx, sm, bproto_sum, bip_sum);
		ctx.sk = ctx.nk = NULL;
		if (action != PF_PASS) {
			pf_udp_mapping_release(ctx.udp_mapping);
			if (r->log || (ctx.nr != NULL && ctx.nr->log) ||
			    ctx.reason == PFRES_MEMORY)
				pd->act.log |= PF_LOG_FORCE;
			if (action == PF_DROP &&
			    (r->rule_flag & PFRULE_RETURN))
				pf_return(r, ctx.nr, pd, ctx.th,
				    bproto_sum, bip_sum, &ctx.reason,
				    pd->act.rtableid);
			*reason = ctx.reason;
			return (action);
		}

		nat64 = pd->af != pd->naf;
		if (nat64) {
			int			 ret;

			if (ctx.sk == NULL)
				ctx.sk = (*sm)->key[pd->dir == PF_IN ? PF_SK_STACK : PF_SK_WIRE];
			if (ctx.nk == NULL)
				ctx.nk = (*sm)->key[pd->dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK];

			if (pd->dir == PF_IN) {
				ret = pf_translate(pd, &ctx.sk->addr[pd->didx],
				    ctx.sk->port[pd->didx], &ctx.sk->addr[pd->sidx],
				    ctx.sk->port[pd->sidx], ctx.virtual_type,
				    ctx.icmp_dir);
			} else {
				ret = pf_translate(pd, &ctx.sk->addr[pd->sidx],
				    ctx.sk->port[pd->sidx], &ctx.sk->addr[pd->didx],
				    ctx.sk->port[pd->didx], ctx.virtual_type,
				    ctx.icmp_dir);
			}

			if (ret < 0)
				goto cleanup;

			ctx.rewrite += ret;

			if (ctx.rewrite && ctx.sk->af != ctx.nk->af)
				action = PF_AFRT;
		}
	} else {
		while ((ri = SLIST_FIRST(&ctx.rules))) {
			SLIST_REMOVE_HEAD(&ctx.rules, entry);
			free(ri, M_PF_RULE_ITEM);
		}

		uma_zfree(V_pf_state_key_z, ctx.sk);
		uma_zfree(V_pf_state_key_z, ctx.nk);
		ctx.sk = ctx.nk = NULL;
		pf_udp_mapping_release(ctx.udp_mapping);
	}

	/* copy back packet headers if we performed NAT operations */
	if (ctx.rewrite)
		m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any);

	if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
	    pd->dir == PF_OUT &&
	    V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, pd->m)) {
		/*
		 * We want the state created, but we dont
		 * want to send this in case a partner
		 * firewall has to know about it to allow
		 * replies through it.
		 */
		*reason = ctx.reason;
		return (PF_DEFER);
	}

	*reason = ctx.reason;
	return (action);

cleanup:
	while ((ri = SLIST_FIRST(&ctx.rules))) {
		SLIST_REMOVE_HEAD(&ctx.rules, entry);
		free(ri, M_PF_RULE_ITEM);
	}

	uma_zfree(V_pf_state_key_z, ctx.sk);
	uma_zfree(V_pf_state_key_z, ctx.nk);
	pf_udp_mapping_release(ctx.udp_mapping);
	*reason = ctx.reason;

	return (PF_DROP);
}

static int
pf_create_state(struct pf_krule *r, struct pf_test_ctx *ctx,
    struct pf_kstate **sm, u_int16_t bproto_sum, u_int16_t bip_sum)
{
	struct pf_pdesc		*pd = ctx->pd;
	struct pf_kstate	*s = NULL;
	struct pf_ksrc_node	*sns[PF_SN_MAX] = { NULL };
	/*
	 * XXXKS: The hash for PF_SN_LIMIT and PF_SN_ROUTE should be the same
	 *        but for PF_SN_NAT it is different. Don't try optimizing it,
	 *        just store all 3 hashes.
	 */
	struct pf_srchash	*snhs[PF_SN_MAX] = { NULL };
	struct tcphdr		*th = &pd->hdr.tcp;
	u_int16_t		 mss = V_tcp_mssdflt;
	u_short			 sn_reason;
	struct pf_krule_item	*ri;

	/* check maximums */
	if (r->max_states &&
	    (counter_u64_fetch(r->states_cur) >= r->max_states)) {
		counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
		REASON_SET(&ctx->reason, PFRES_MAXSTATES);
		goto csfailed;
	}
	/* src node for limits */
	if ((r->rule_flag & PFRULE_SRCTRACK) &&
	    (sn_reason = pf_insert_src_node(sns, snhs, r, pd->src, pd->af,
	    NULL, NULL, pd->af, PF_SN_LIMIT)) != 0) {
		REASON_SET(&ctx->reason, sn_reason);
		goto csfailed;
	}
	/* src node for route-to rule */
	if (r->rt) {
		if ((r->route.opts & PF_POOL_STICKYADDR) &&
		    (sn_reason = pf_insert_src_node(sns, snhs, r, pd->src,
		    pd->af, &pd->act.rt_addr, pd->act.rt_kif, pd->act.rt_af,
		    PF_SN_ROUTE)) != 0) {
			REASON_SET(&ctx->reason, sn_reason);
			goto csfailed;
		}
	}
	/* src node for translation rule */
	if (ctx->nr != NULL) {
		KASSERT(ctx->nat_pool != NULL, ("%s: nat_pool is NULL", __func__));
		/*
		 * The NAT addresses are chosen during ruleset parsing.
		 * The new afto code stores post-nat addresses in nsaddr.
		 * The old nat code (also used for new nat-to rules) creates
		 * state keys and stores addresses in them.
		 */
		if ((ctx->nat_pool->opts & PF_POOL_STICKYADDR) &&
		    (sn_reason = pf_insert_src_node(sns, snhs, ctx->nr,
		    ctx->sk ? &(ctx->sk->addr[pd->sidx]) : pd->src, pd->af,
		    ctx->nk ? &(ctx->nk->addr[1]) : &(pd->nsaddr), NULL,
		    pd->naf, PF_SN_NAT)) != 0 ) {
			REASON_SET(&ctx->reason, sn_reason);
			goto csfailed;
		}
	}
	s = pf_alloc_state(M_NOWAIT);
	if (s == NULL) {
		REASON_SET(&ctx->reason, PFRES_MEMORY);
		goto csfailed;
	}
	s->rule = r;
	s->nat_rule = ctx->nr;
	s->anchor = ctx->a;
	memcpy(&s->match_rules, &ctx->rules, sizeof(s->match_rules));
	memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions));

	if (pd->act.allow_opts)
		s->state_flags |= PFSTATE_ALLOWOPTS;
	if (r->rule_flag & PFRULE_STATESLOPPY)
		s->state_flags |= PFSTATE_SLOPPY;
	if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */
		s->state_flags |= PFSTATE_SCRUB_TCP;
	if ((r->rule_flag & PFRULE_PFLOW) ||
	    (ctx->nr != NULL && ctx->nr->rule_flag & PFRULE_PFLOW))
		s->state_flags |= PFSTATE_PFLOW;

	s->act.log = pd->act.log & PF_LOG_ALL;
	s->sync_state = PFSYNC_S_NONE;
	s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */

	if (ctx->nr != NULL)
		s->act.log |= ctx->nr->log & PF_LOG_ALL;
	switch (pd->proto) {
	case IPPROTO_TCP:
		s->src.seqlo = ntohl(th->th_seq);
		s->src.seqhi = s->src.seqlo + pd->p_len + 1;
		if ((tcp_get_flags(th) & (TH_SYN|TH_ACK)) == TH_SYN &&
		    r->keep_state == PF_STATE_MODULATE) {
			/* Generate sequence number modulator */
			if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
			    0)
				s->src.seqdiff = 1;
			pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum,
			    htonl(s->src.seqlo + s->src.seqdiff), 0);
			ctx->rewrite = 1;
		} else
			s->src.seqdiff = 0;
		if (tcp_get_flags(th) & TH_SYN) {
			s->src.seqhi++;
			s->src.wscale = pf_get_wscale(pd);
		}
		s->src.max_win = MAX(ntohs(th->th_win), 1);
		if (s->src.wscale & PF_WSCALE_MASK) {
			/* Remove scale factor from initial window */
			int win = s->src.max_win;
			win += 1 << (s->src.wscale & PF_WSCALE_MASK);
			s->src.max_win = (win - 1) >>
			    (s->src.wscale & PF_WSCALE_MASK);
		}
		if (tcp_get_flags(th) & TH_FIN)
			s->src.seqhi++;
		s->dst.seqhi = 1;
		s->dst.max_win = 1;
		pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT);
		pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED);
		s->timeout = PFTM_TCP_FIRST_PACKET;
		atomic_add_32(&V_pf_status.states_halfopen, 1);
		break;
	case IPPROTO_UDP:
		pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE);
		pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC);
		s->timeout = PFTM_UDP_FIRST_PACKET;
		break;
	case IPPROTO_SCTP:
		pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT);
		pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED);
		s->timeout = PFTM_SCTP_FIRST_PACKET;
		break;
	case IPPROTO_ICMP:
#ifdef INET6
	case IPPROTO_ICMPV6:
#endif /* INET6 */
		s->timeout = PFTM_ICMP_FIRST_PACKET;
		break;
	default:
		pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE);
		pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC);
		s->timeout = PFTM_OTHER_FIRST_PACKET;
	}

	s->creation = s->expire = pf_get_uptime();

	if (pd->proto == IPPROTO_TCP) {
		if (s->state_flags & PFSTATE_SCRUB_TCP &&
		    pf_normalize_tcp_init(pd, th, &s->src)) {
			REASON_SET(&ctx->reason, PFRES_MEMORY);
			goto csfailed;
		}
		if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub &&
		    pf_normalize_tcp_stateful(pd, &ctx->reason, th, s,
		    &s->src, &s->dst, &ctx->rewrite)) {
			/* This really shouldn't happen!!! */
			DPFPRINTF(PF_DEBUG_URGENT,
			    "%s: tcp normalize failed on first "
			     "pkt", __func__);
			goto csfailed;
		}
	} else if (pd->proto == IPPROTO_SCTP) {
		if (pf_normalize_sctp_init(pd, &s->src, &s->dst))
			goto csfailed;
		if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP)))
			goto csfailed;
	}
	s->direction = pd->dir;

	/*
	 * sk/nk could already been setup by pf_get_translation().
	 */
	if (ctx->sk == NULL && ctx->nk == NULL) {
		MPASS(pd->sport == NULL || (pd->osport == *pd->sport));
		MPASS(pd->dport == NULL || (pd->odport == *pd->dport));
		if (pf_state_key_setup(pd, pd->nsport, pd->ndport,
		    &ctx->sk, &ctx->nk)) {
			goto csfailed;
		}
	} else
		KASSERT((ctx->sk != NULL && ctx->nk != NULL), ("%s: nr %p sk %p, nk %p",
		    __func__, ctx->nr, ctx->sk, ctx->nk));

	/* Swap sk/nk for PF_OUT. */
	if (pf_state_insert(BOUND_IFACE(s, pd), pd->kif,
	    (pd->dir == PF_IN) ? ctx->sk : ctx->nk,
	    (pd->dir == PF_IN) ? ctx->nk : ctx->sk, s)) {
		REASON_SET(&ctx->reason, PFRES_STATEINS);
		goto drop;
	} else
		*sm = s;
	ctx->sk = ctx->nk = NULL;

	STATE_INC_COUNTERS(s);

	/*
	 * Lock order is important: first state, then source node.
	 */
	for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) {
		if (pf_src_node_exists(&sns[sn_type], snhs[sn_type])) {
			s->sns[sn_type] = sns[sn_type];
			PF_HASHROW_UNLOCK(snhs[sn_type]);
		}
	}

	if (ctx->tag > 0)
		s->tag = ctx->tag;
	if (pd->proto == IPPROTO_TCP && (tcp_get_flags(th) & (TH_SYN|TH_ACK)) ==
	    TH_SYN && r->keep_state == PF_STATE_SYNPROXY && pd->dir == PF_IN) {
		pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC);
		pf_undo_nat(ctx->nr, pd, bip_sum);
		s->src.seqhi = arc4random();
		/* Find mss option */
		int rtid = M_GETFIB(pd->m);
		mss = pf_get_mss(pd);
		mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
		mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
		s->src.mss = mss;
		pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport,
		    th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
		    TH_SYN|TH_ACK, 0, s->src.mss, 0, M_SKIP_FIREWALL, 0, 0,
		    pd->act.rtableid);
		REASON_SET(&ctx->reason, PFRES_SYNPROXY);
		return (PF_SYNPROXY_DROP);
	}

	s->udp_mapping = ctx->udp_mapping;

	return (PF_PASS);

csfailed:
	while ((ri = SLIST_FIRST(&ctx->rules))) {
		SLIST_REMOVE_HEAD(&ctx->rules, entry);
		free(ri, M_PF_RULE_ITEM);
	}

	uma_zfree(V_pf_state_key_z, ctx->sk);
	uma_zfree(V_pf_state_key_z, ctx->nk);

	for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) {
		if (pf_src_node_exists(&sns[sn_type], snhs[sn_type])) {
			if (--sns[sn_type]->states == 0 &&
			    sns[sn_type]->expire == 0) {
				pf_unlink_src_node(sns[sn_type]);
				pf_free_src_node(sns[sn_type]);
				counter_u64_add(
				    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
			}
			PF_HASHROW_UNLOCK(snhs[sn_type]);
		}
	}

drop:
	if (s != NULL) {
		pf_src_tree_remove_state(s);
		s->timeout = PFTM_UNLINKED;
		pf_free_state(s);
	}

	return (PF_DROP);
}

int
pf_translate(struct pf_pdesc *pd, struct pf_addr *saddr, u_int16_t sport,
    struct pf_addr *daddr, u_int16_t dport, u_int16_t virtual_type,
    int icmp_dir)
{
	/*
	 * pf_translate() implements OpenBSD's "new" NAT approach.
	 * We don't follow it, because it involves a breaking syntax change
	 * (removing nat/rdr rules, moving it into regular pf rules.)
	 * It also moves NAT processing to be done after normal rules evaluation
	 * whereas in FreeBSD that's done before rules processing.
	 *
	 * We adopt the function only for nat64, and keep other NAT processing
	 * before rules processing.
	 */
	int	rewrite = 0;
	int	afto = pd->af != pd->naf;

	MPASS(afto);

	switch (pd->proto) {
	case IPPROTO_TCP:
	case IPPROTO_UDP:
	case IPPROTO_SCTP:
		if (afto || *pd->sport != sport) {
			pf_change_ap(pd, pd->src, pd->sport,
			    saddr, sport);
			rewrite = 1;
		}
		if (afto || *pd->dport != dport) {
			pf_change_ap(pd, pd->dst, pd->dport,
			    daddr, dport);
			rewrite = 1;
		}
		break;

#ifdef INET
	case IPPROTO_ICMP:
		/* pf_translate() is also used when logging invalid packets */
		if (pd->af != AF_INET)
			return (0);

		if (afto) {
			if (pf_translate_icmp_af(AF_INET6, &pd->hdr.icmp))
				return (-1);
			pd->proto = IPPROTO_ICMPV6;
			rewrite = 1;
		}
		if (virtual_type == htons(ICMP_ECHO)) {
			u_int16_t icmpid = (icmp_dir == PF_IN) ? sport : dport;

			if (icmpid != pd->hdr.icmp.icmp_id) {
				pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
				    pd->hdr.icmp.icmp_cksum,
				    pd->hdr.icmp.icmp_id, icmpid, 0);
				pd->hdr.icmp.icmp_id = icmpid;
				/* XXX TODO copyback. */
				rewrite = 1;
			}
		}
		break;
#endif /* INET */

#ifdef INET6
	case IPPROTO_ICMPV6:
		/* pf_translate() is also used when logging invalid packets */
		if (pd->af != AF_INET6)
			return (0);

		if (afto) {
			/* ip_sum will be recalculated in pf_translate_af */
			if (pf_translate_icmp_af(AF_INET, &pd->hdr.icmp6))
				return (0);
			pd->proto = IPPROTO_ICMP;
			rewrite = 1;
		}
		break;
#endif /* INET6 */

	default:
		break;
	}

	return (rewrite);
}

int
pf_translate_compat(struct pf_test_ctx *ctx)
{
	struct pf_pdesc		*pd = ctx->pd;
	struct pf_state_key	*nk = ctx->nk;
	struct tcphdr		*th = &pd->hdr.tcp;
	int 			 rewrite = 0;

	KASSERT(ctx->sk != NULL, ("%s: null sk", __func__));
	KASSERT(ctx->nk != NULL, ("%s: null nk", __func__));

	switch (pd->virtual_proto) {
	case IPPROTO_TCP:
		if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) ||
		    nk->port[pd->sidx] != pd->nsport) {
			pf_change_ap(pd, pd->src, &th->th_sport,
			    &nk->addr[pd->sidx], nk->port[pd->sidx]);
			pd->sport = &th->th_sport;
			pd->nsport = th->th_sport;
			pf_addrcpy(&pd->nsaddr, pd->src, pd->af);
		}

		if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) ||
		    nk->port[pd->didx] != pd->ndport) {
			pf_change_ap(pd, pd->dst, &th->th_dport,
			    &nk->addr[pd->didx], nk->port[pd->didx]);
			pd->dport = &th->th_dport;
			pd->ndport = th->th_dport;
			pf_addrcpy(&pd->ndaddr, pd->dst, pd->af);
		}
		rewrite++;
		break;
	case IPPROTO_UDP:
		if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) ||
		    nk->port[pd->sidx] != pd->nsport) {
			pf_change_ap(pd, pd->src,
			    &pd->hdr.udp.uh_sport,
			    &nk->addr[pd->sidx],
			    nk->port[pd->sidx]);
			pd->sport = &pd->hdr.udp.uh_sport;
			pd->nsport = pd->hdr.udp.uh_sport;
			pf_addrcpy(&pd->nsaddr, pd->src, pd->af);
		}

		if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) ||
		    nk->port[pd->didx] != pd->ndport) {
			pf_change_ap(pd, pd->dst,
			    &pd->hdr.udp.uh_dport,
			    &nk->addr[pd->didx],
			    nk->port[pd->didx]);
			pd->dport = &pd->hdr.udp.uh_dport;
			pd->ndport = pd->hdr.udp.uh_dport;
			pf_addrcpy(&pd->ndaddr, pd->dst, pd->af);
		}
		rewrite++;
		break;
	case IPPROTO_SCTP: {
		if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], pd->af) ||
		    nk->port[pd->sidx] != pd->nsport) {
			pf_change_ap(pd, pd->src,
			    &pd->hdr.sctp.src_port,
			    &nk->addr[pd->sidx],
			    nk->port[pd->sidx]);
			pd->sport = &pd->hdr.sctp.src_port;
			pd->nsport = pd->hdr.sctp.src_port;
			pf_addrcpy(&pd->nsaddr, pd->src, pd->af);
		}
		if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], pd->af) ||
		    nk->port[pd->didx] != pd->ndport) {
			pf_change_ap(pd, pd->dst,
			    &pd->hdr.sctp.dest_port,
			    &nk->addr[pd->didx],
			    nk->port[pd->didx]);
			pd->dport = &pd->hdr.sctp.dest_port;
			pd->ndport = pd->hdr.sctp.dest_port;
			pf_addrcpy(&pd->ndaddr, pd->dst, pd->af);
		}
		break;
	}
#ifdef INET
	case IPPROTO_ICMP:
		if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET)) {
			pf_change_a(&pd->src->v4.s_addr, pd->ip_sum,
			    nk->addr[pd->sidx].v4.s_addr, 0);
			pf_addrcpy(&pd->nsaddr, pd->src, pd->af);
		}

		if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET)) {
			pf_change_a(&pd->dst->v4.s_addr, pd->ip_sum,
			    nk->addr[pd->didx].v4.s_addr, 0);
			pf_addrcpy(&pd->ndaddr, pd->dst, pd->af);
		}

		if (ctx->virtual_type == htons(ICMP_ECHO) &&
		    nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) {
			pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
			    pd->hdr.icmp.icmp_cksum, pd->nsport,
			    nk->port[pd->sidx], 0);
			pd->hdr.icmp.icmp_id = nk->port[pd->sidx];
			pd->sport = &pd->hdr.icmp.icmp_id;
		}
		m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
		break;
#endif /* INET */
#ifdef INET6
	case IPPROTO_ICMPV6:
		if (PF_ANEQ(&pd->nsaddr, &nk->addr[pd->sidx], AF_INET6)) {
			pf_change_a6(pd->src, &pd->hdr.icmp6.icmp6_cksum,
			    &nk->addr[pd->sidx], 0);
			pf_addrcpy(&pd->nsaddr, pd->src, pd->af);
		}

		if (PF_ANEQ(&pd->ndaddr, &nk->addr[pd->didx], AF_INET6)) {
			pf_change_a6(pd->dst, &pd->hdr.icmp6.icmp6_cksum,
			    &nk->addr[pd->didx], 0);
			pf_addrcpy(&pd->ndaddr, pd->dst, pd->af);
		}
		rewrite++;
		break;
#endif /* INET */
	default:
		switch (pd->af) {
#ifdef INET
		case AF_INET:
			if (PF_ANEQ(&pd->nsaddr,
				&nk->addr[pd->sidx], AF_INET)) {
				pf_change_a(&pd->src->v4.s_addr,
				    pd->ip_sum,
				    nk->addr[pd->sidx].v4.s_addr, 0);
				pf_addrcpy(&pd->nsaddr, pd->src, pd->af);
			}

			if (PF_ANEQ(&pd->ndaddr,
				&nk->addr[pd->didx], AF_INET)) {
				pf_change_a(&pd->dst->v4.s_addr,
				    pd->ip_sum,
				    nk->addr[pd->didx].v4.s_addr, 0);
				pf_addrcpy(&pd->ndaddr, pd->dst, pd->af);
			}
			break;
#endif /* INET */
#ifdef INET6
		case AF_INET6:
			if (PF_ANEQ(&pd->nsaddr,
				&nk->addr[pd->sidx], AF_INET6)) {
				pf_addrcpy(&pd->nsaddr, &nk->addr[pd->sidx],
				    pd->af);
				pf_addrcpy(pd->src, &nk->addr[pd->sidx], pd->af);
			}

			if (PF_ANEQ(&pd->ndaddr,
				&nk->addr[pd->didx], AF_INET6)) {
				pf_addrcpy(&pd->ndaddr, &nk->addr[pd->didx],
				    pd->af);
				pf_addrcpy(pd->dst, &nk->addr[pd->didx],
				    pd->af);
			}
			break;
#endif /* INET6 */
		}
		break;
	}
	return (rewrite);
}

static int
pf_tcp_track_full(struct pf_kstate *state, struct pf_pdesc *pd,
    u_short *reason, int *copyback, struct pf_state_peer *src,
    struct pf_state_peer *dst, u_int8_t psrc, u_int8_t pdst)
{
	struct tcphdr		*th = &pd->hdr.tcp;
	u_int16_t		 win = ntohs(th->th_win);
	u_int32_t		 ack, end, data_end, seq, orig_seq;
	u_int8_t		 sws, dws;
	int			 ackskew;

	if (src->wscale && dst->wscale && !(tcp_get_flags(th) & TH_SYN)) {
		sws = src->wscale & PF_WSCALE_MASK;
		dws = dst->wscale & PF_WSCALE_MASK;
	} else
		sws = dws = 0;

	/*
	 * Sequence tracking algorithm from Guido van Rooij's paper:
	 *   http://www.madison-gurkha.com/publications/tcp_filtering/
	 *	tcp_filtering.ps
	 */

	orig_seq = seq = ntohl(th->th_seq);
	if (src->seqlo == 0) {
		/* First packet from this end. Set its state */

		if ((state->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) &&
		    src->scrub == NULL) {
			if (pf_normalize_tcp_init(pd, th, src)) {
				REASON_SET(reason, PFRES_MEMORY);
				return (PF_DROP);
			}
		}

		/* Deferred generation of sequence number modulator */
		if (dst->seqdiff && !src->seqdiff) {
			/* use random iss for the TCP server */
			while ((src->seqdiff = arc4random() - seq) == 0)
				;
			ack = ntohl(th->th_ack) - dst->seqdiff;
			pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq +
			    src->seqdiff), 0);
			pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0);
			*copyback = 1;
		} else {
			ack = ntohl(th->th_ack);
		}

		end = seq + pd->p_len;
		if (tcp_get_flags(th) & TH_SYN) {
			end++;
			if (dst->wscale & PF_WSCALE_FLAG) {
				src->wscale = pf_get_wscale(pd);
				if (src->wscale & PF_WSCALE_FLAG) {
					/* Remove scale factor from initial
					 * window */
					sws = src->wscale & PF_WSCALE_MASK;
					win = ((u_int32_t)win + (1 << sws) - 1)
					    >> sws;
					dws = dst->wscale & PF_WSCALE_MASK;
				} else {
					/* fixup other window */
					dst->max_win = MIN(TCP_MAXWIN,
					    (u_int32_t)dst->max_win <<
					    (dst->wscale & PF_WSCALE_MASK));
					/* in case of a retrans SYN|ACK */
					dst->wscale = 0;
				}
			}
		}
		data_end = end;
		if (tcp_get_flags(th) & TH_FIN)
			end++;

		src->seqlo = seq;
		if (src->state < TCPS_SYN_SENT)
			pf_set_protostate(state, psrc, TCPS_SYN_SENT);

		/*
		 * May need to slide the window (seqhi may have been set by
		 * the crappy stack check or if we picked up the connection
		 * after establishment)
		 */
		if (src->seqhi == 1 ||
		    SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
			src->seqhi = end + MAX(1, dst->max_win << dws);
		if (win > src->max_win)
			src->max_win = win;

	} else {
		ack = ntohl(th->th_ack) - dst->seqdiff;
		if (src->seqdiff) {
			/* Modulate sequence numbers */
			pf_change_proto_a(pd->m, &th->th_seq, &th->th_sum, htonl(seq +
			    src->seqdiff), 0);
			pf_change_proto_a(pd->m, &th->th_ack, &th->th_sum, htonl(ack), 0);
			*copyback = 1;
		}
		end = seq + pd->p_len;
		if (tcp_get_flags(th) & TH_SYN)
			end++;
		data_end = end;
		if (tcp_get_flags(th) & TH_FIN)
			end++;
	}

	if ((tcp_get_flags(th) & TH_ACK) == 0) {
		/* Let it pass through the ack skew check */
		ack = dst->seqlo;
	} else if ((ack == 0 &&
	    (tcp_get_flags(th) & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
	    /* broken tcp stacks do not set ack */
	    (dst->state < TCPS_SYN_SENT)) {
		/*
		 * Many stacks (ours included) will set the ACK number in an
		 * FIN|ACK if the SYN times out -- no sequence to ACK.
		 */
		ack = dst->seqlo;
	}

	if (seq == end) {
		/* Ease sequencing restrictions on no data packets */
		seq = src->seqlo;
		data_end = end = seq;
	}

	ackskew = dst->seqlo - ack;

	/*
	 * Need to demodulate the sequence numbers in any TCP SACK options
	 * (Selective ACK). We could optionally validate the SACK values
	 * against the current ACK window, either forwards or backwards, but
	 * I'm not confident that SACK has been implemented properly
	 * everywhere. It wouldn't surprise me if several stacks accidentally
	 * SACK too far backwards of previously ACKed data. There really aren't
	 * any security implications of bad SACKing unless the target stack
	 * doesn't validate the option length correctly. Someone trying to
	 * spoof into a TCP connection won't bother blindly sending SACK
	 * options anyway.
	 */
	if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
		if (pf_modulate_sack(pd, th, dst))
			*copyback = 1;
	}

#define	MAXACKWINDOW (0xffff + 1500)	/* 1500 is an arbitrary fudge factor */
	if (SEQ_GEQ(src->seqhi, data_end) &&
	    /* Last octet inside other's window space */
	    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
	    /* Retrans: not more than one window back */
	    (ackskew >= -MAXACKWINDOW) &&
	    /* Acking not more than one reassembled fragment backwards */
	    (ackskew <= (MAXACKWINDOW << sws)) &&
	    /* Acking not more than one window forward */
	    ((tcp_get_flags(th) & TH_RST) == 0 || orig_seq == src->seqlo ||
	    (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) {
	    /* Require an exact/+1 sequence match on resets when possible */

		if (dst->scrub || src->scrub) {
			if (pf_normalize_tcp_stateful(pd, reason, th,
			    state, src, dst, copyback))
				return (PF_DROP);
		}

		/* update max window */
		if (src->max_win < win)
			src->max_win = win;
		/* synchronize sequencing */
		if (SEQ_GT(end, src->seqlo))
			src->seqlo = end;
		/* slide the window of what the other end can send */
		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
			dst->seqhi = ack + MAX((win << sws), 1);

		/* update states */
		if (tcp_get_flags(th) & TH_SYN)
			if (src->state < TCPS_SYN_SENT)
				pf_set_protostate(state, psrc, TCPS_SYN_SENT);
		if (tcp_get_flags(th) & TH_FIN)
			if (src->state < TCPS_CLOSING)
				pf_set_protostate(state, psrc, TCPS_CLOSING);
		if (tcp_get_flags(th) & TH_ACK) {
			if (dst->state == TCPS_SYN_SENT) {
				pf_set_protostate(state, pdst,
				    TCPS_ESTABLISHED);
				if (src->state == TCPS_ESTABLISHED &&
				    state->sns[PF_SN_LIMIT] != NULL &&
				    pf_src_connlimit(state)) {
					REASON_SET(reason, PFRES_SRCLIMIT);
					return (PF_DROP);
				}
			} else if (dst->state == TCPS_CLOSING)
				pf_set_protostate(state, pdst,
				    TCPS_FIN_WAIT_2);
		}
		if (tcp_get_flags(th) & TH_RST)
			pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT);

		/* update expire time */
		state->expire = pf_get_uptime();
		if (src->state >= TCPS_FIN_WAIT_2 &&
		    dst->state >= TCPS_FIN_WAIT_2)
			state->timeout = PFTM_TCP_CLOSED;
		else if (src->state >= TCPS_CLOSING &&
		    dst->state >= TCPS_CLOSING)
			state->timeout = PFTM_TCP_FIN_WAIT;
		else if (src->state < TCPS_ESTABLISHED ||
		    dst->state < TCPS_ESTABLISHED)
			state->timeout = PFTM_TCP_OPENING;
		else if (src->state >= TCPS_CLOSING ||
		    dst->state >= TCPS_CLOSING)
			state->timeout = PFTM_TCP_CLOSING;
		else
			state->timeout = PFTM_TCP_ESTABLISHED;

		/* Fall through to PASS packet */

	} else if ((dst->state < TCPS_SYN_SENT ||
		dst->state >= TCPS_FIN_WAIT_2 ||
		src->state >= TCPS_FIN_WAIT_2) &&
	    SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) &&
	    /* Within a window forward of the originating packet */
	    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
	    /* Within a window backward of the originating packet */

		/*
		 * This currently handles three situations:
		 *  1) Stupid stacks will shotgun SYNs before their peer
		 *     replies.
		 *  2) When PF catches an already established stream (the
		 *     firewall rebooted, the state table was flushed, routes
		 *     changed...)
		 *  3) Packets get funky immediately after the connection
		 *     closes (this should catch Solaris spurious ACK|FINs
		 *     that web servers like to spew after a close)
		 *
		 * This must be a little more careful than the above code
		 * since packet floods will also be caught here. We don't
		 * update the TTL here to mitigate the damage of a packet
		 * flood and so the same code can handle awkward establishment
		 * and a loosened connection close.
		 * In the establishment case, a correct peer response will
		 * validate the connection, go through the normal state code
		 * and keep updating the state TTL.
		 */

		if (V_pf_status.debug >= PF_DEBUG_MISC) {
			printf("pf: loose state match: ");
			pf_print_state(state);
			pf_print_flags(tcp_get_flags(th));
			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
			    "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
			    pd->p_len, ackskew, (unsigned long long)state->packets[0],
			    (unsigned long long)state->packets[1],
			    pd->dir == PF_IN ? "in" : "out",
			    pd->dir == state->direction ? "fwd" : "rev");
		}

		if (dst->scrub || src->scrub) {
			if (pf_normalize_tcp_stateful(pd, reason, th,
			    state, src, dst, copyback))
				return (PF_DROP);
		}

		/* update max window */
		if (src->max_win < win)
			src->max_win = win;
		/* synchronize sequencing */
		if (SEQ_GT(end, src->seqlo))
			src->seqlo = end;
		/* slide the window of what the other end can send */
		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
			dst->seqhi = ack + MAX((win << sws), 1);

		/*
		 * Cannot set dst->seqhi here since this could be a shotgunned
		 * SYN and not an already established connection.
		 */

		if (tcp_get_flags(th) & TH_FIN)
			if (src->state < TCPS_CLOSING)
				pf_set_protostate(state, psrc, TCPS_CLOSING);
		if (tcp_get_flags(th) & TH_RST)
			pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT);

		/* Fall through to PASS packet */

	} else {
		if (state->dst.state == TCPS_SYN_SENT &&
		    state->src.state == TCPS_SYN_SENT) {
			/* Send RST for state mismatches during handshake */
			if (!(tcp_get_flags(th) & TH_RST))
				pf_send_tcp(state->rule, pd->af,
				    pd->dst, pd->src, th->th_dport,
				    th->th_sport, ntohl(th->th_ack), 0,
				    TH_RST, 0, 0,
				    state->rule->return_ttl, M_SKIP_FIREWALL,
				    0, 0, state->act.rtableid);
			src->seqlo = 0;
			src->seqhi = 1;
			src->max_win = 1;
		} else if (V_pf_status.debug >= PF_DEBUG_MISC) {
			printf("pf: BAD state: ");
			pf_print_state(state);
			pf_print_flags(tcp_get_flags(th));
			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
			    "pkts=%llu:%llu dir=%s,%s\n",
			    seq, orig_seq, ack, pd->p_len, ackskew,
			    (unsigned long long)state->packets[0],
			    (unsigned long long)state->packets[1],
			    pd->dir == PF_IN ? "in" : "out",
			    pd->dir == state->direction ? "fwd" : "rev");
			printf("pf: State failure on: %c %c %c %c | %c %c\n",
			    SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1',
			    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
			    ' ': '2',
			    (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
			    (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
			    SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5',
			    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
		}
		REASON_SET(reason, PFRES_BADSTATE);
		return (PF_DROP);
	}

	return (PF_PASS);
}

static int
pf_tcp_track_sloppy(struct pf_kstate *state, struct pf_pdesc *pd,
    u_short *reason, struct pf_state_peer *src, struct pf_state_peer *dst,
    u_int8_t psrc, u_int8_t pdst)
{
	struct tcphdr		*th = &pd->hdr.tcp;

	if (tcp_get_flags(th) & TH_SYN)
		if (src->state < TCPS_SYN_SENT)
			pf_set_protostate(state, psrc, TCPS_SYN_SENT);
	if (tcp_get_flags(th) & TH_FIN)
		if (src->state < TCPS_CLOSING)
			pf_set_protostate(state, psrc, TCPS_CLOSING);
	if (tcp_get_flags(th) & TH_ACK) {
		if (dst->state == TCPS_SYN_SENT) {
			pf_set_protostate(state, pdst, TCPS_ESTABLISHED);
			if (src->state == TCPS_ESTABLISHED &&
			    state->sns[PF_SN_LIMIT] != NULL &&
			    pf_src_connlimit(state)) {
				REASON_SET(reason, PFRES_SRCLIMIT);
				return (PF_DROP);
			}
		} else if (dst->state == TCPS_CLOSING) {
			pf_set_protostate(state, pdst, TCPS_FIN_WAIT_2);
		} else if (src->state == TCPS_SYN_SENT &&
		    dst->state < TCPS_SYN_SENT) {
			/*
			 * Handle a special sloppy case where we only see one
			 * half of the connection. If there is a ACK after
			 * the initial SYN without ever seeing a packet from
			 * the destination, set the connection to established.
			 */
			pf_set_protostate(state, PF_PEER_BOTH,
			    TCPS_ESTABLISHED);
			dst->state = src->state = TCPS_ESTABLISHED;
			if (state->sns[PF_SN_LIMIT] != NULL &&
			    pf_src_connlimit(state)) {
				REASON_SET(reason, PFRES_SRCLIMIT);
				return (PF_DROP);
			}
		} else if (src->state == TCPS_CLOSING &&
		    dst->state == TCPS_ESTABLISHED &&
		    dst->seqlo == 0) {
			/*
			 * Handle the closing of half connections where we
			 * don't see the full bidirectional FIN/ACK+ACK
			 * handshake.
			 */
			pf_set_protostate(state, pdst, TCPS_CLOSING);
		}
	}
	if (tcp_get_flags(th) & TH_RST)
		pf_set_protostate(state, PF_PEER_BOTH, TCPS_TIME_WAIT);

	/* update expire time */
	state->expire = pf_get_uptime();
	if (src->state >= TCPS_FIN_WAIT_2 &&
	    dst->state >= TCPS_FIN_WAIT_2)
		state->timeout = PFTM_TCP_CLOSED;
	else if (src->state >= TCPS_CLOSING &&
	    dst->state >= TCPS_CLOSING)
		state->timeout = PFTM_TCP_FIN_WAIT;
	else if (src->state < TCPS_ESTABLISHED ||
	    dst->state < TCPS_ESTABLISHED)
		state->timeout = PFTM_TCP_OPENING;
	else if (src->state >= TCPS_CLOSING ||
	    dst->state >= TCPS_CLOSING)
		state->timeout = PFTM_TCP_CLOSING;
	else
		state->timeout = PFTM_TCP_ESTABLISHED;

	return (PF_PASS);
}

static int
pf_synproxy(struct pf_pdesc *pd, struct pf_kstate *state, u_short *reason)
{
	struct pf_state_key	*sk = state->key[pd->didx];
	struct tcphdr		*th = &pd->hdr.tcp;

	if (state->src.state == PF_TCPS_PROXY_SRC) {
		if (pd->dir != state->direction) {
			REASON_SET(reason, PFRES_SYNPROXY);
			return (PF_SYNPROXY_DROP);
		}
		if (tcp_get_flags(th) & TH_SYN) {
			if (ntohl(th->th_seq) != state->src.seqlo) {
				REASON_SET(reason, PFRES_SYNPROXY);
				return (PF_DROP);
			}
			pf_send_tcp(state->rule, pd->af, pd->dst,
			    pd->src, th->th_dport, th->th_sport,
			    state->src.seqhi, ntohl(th->th_seq) + 1,
			    TH_SYN|TH_ACK, 0, state->src.mss, 0,
			    M_SKIP_FIREWALL, 0, 0, state->act.rtableid);
			REASON_SET(reason, PFRES_SYNPROXY);
			return (PF_SYNPROXY_DROP);
		} else if ((tcp_get_flags(th) & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK ||
		    (ntohl(th->th_ack) != state->src.seqhi + 1) ||
		    (ntohl(th->th_seq) != state->src.seqlo + 1)) {
			REASON_SET(reason, PFRES_SYNPROXY);
			return (PF_DROP);
		} else if (state->sns[PF_SN_LIMIT] != NULL &&
		    pf_src_connlimit(state)) {
			REASON_SET(reason, PFRES_SRCLIMIT);
			return (PF_DROP);
		} else
			pf_set_protostate(state, PF_PEER_SRC,
			    PF_TCPS_PROXY_DST);
	}
	if (state->src.state == PF_TCPS_PROXY_DST) {
		if (pd->dir == state->direction) {
			if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) != TH_ACK) ||
			    (ntohl(th->th_ack) != state->src.seqhi + 1) ||
			    (ntohl(th->th_seq) != state->src.seqlo + 1)) {
				REASON_SET(reason, PFRES_SYNPROXY);
				return (PF_DROP);
			}
			state->src.max_win = MAX(ntohs(th->th_win), 1);
			if (state->dst.seqhi == 1)
				state->dst.seqhi = arc4random();
			pf_send_tcp(state->rule, pd->af,
			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
			    sk->port[pd->sidx], sk->port[pd->didx],
			    state->dst.seqhi, 0, TH_SYN, 0,
			    state->src.mss, 0,
			    state->orig_kif->pfik_ifp == V_loif ? M_LOOP : 0,
			    state->tag, 0, state->act.rtableid);
			REASON_SET(reason, PFRES_SYNPROXY);
			return (PF_SYNPROXY_DROP);
		} else if (((tcp_get_flags(th) & (TH_SYN|TH_ACK)) !=
		    (TH_SYN|TH_ACK)) ||
		    (ntohl(th->th_ack) != state->dst.seqhi + 1)) {
			REASON_SET(reason, PFRES_SYNPROXY);
			return (PF_DROP);
		} else {
			state->dst.max_win = MAX(ntohs(th->th_win), 1);
			state->dst.seqlo = ntohl(th->th_seq);
			pf_send_tcp(state->rule, pd->af, pd->dst,
			    pd->src, th->th_dport, th->th_sport,
			    ntohl(th->th_ack), ntohl(th->th_seq) + 1,
			    TH_ACK, state->src.max_win, 0, 0, 0,
			    state->tag, 0, state->act.rtableid);
			pf_send_tcp(state->rule, pd->af,
			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
			    sk->port[pd->sidx], sk->port[pd->didx],
			    state->src.seqhi + 1, state->src.seqlo + 1,
			    TH_ACK, state->dst.max_win, 0, 0,
			    M_SKIP_FIREWALL, 0, 0, state->act.rtableid);
			state->src.seqdiff = state->dst.seqhi -
			    state->src.seqlo;
			state->dst.seqdiff = state->src.seqhi -
			    state->dst.seqlo;
			state->src.seqhi = state->src.seqlo +
			    state->dst.max_win;
			state->dst.seqhi = state->dst.seqlo +
			    state->src.max_win;
			state->src.wscale = state->dst.wscale = 0;
			pf_set_protostate(state, PF_PEER_BOTH,
			    TCPS_ESTABLISHED);
			REASON_SET(reason, PFRES_SYNPROXY);
			return (PF_SYNPROXY_DROP);
		}
	}

	return (PF_PASS);
}

static int
pf_test_state(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason)
{
	struct pf_state_key_cmp	 key;
	int			 copyback = 0;
	struct pf_state_peer	*src, *dst;
	uint8_t			 psrc, pdst;
	int			 action;

	bzero(&key, sizeof(key));
	key.af = pd->af;
	key.proto = pd->virtual_proto;
	pf_addrcpy(&key.addr[pd->sidx], pd->src, key.af);
	pf_addrcpy(&key.addr[pd->didx], pd->dst, key.af);
	key.port[pd->sidx] = pd->osport;
	key.port[pd->didx] = pd->odport;

	action = pf_find_state(pd, &key, state);
	if (action != PF_MATCH)
		return (action);

	action = PF_PASS;
	if (pd->dir == (*state)->direction) {
		if (PF_REVERSED_KEY(*state, pd->af)) {
			src = &(*state)->dst;
			dst = &(*state)->src;
			psrc = PF_PEER_DST;
			pdst = PF_PEER_SRC;
		} else {
			src = &(*state)->src;
			dst = &(*state)->dst;
			psrc = PF_PEER_SRC;
			pdst = PF_PEER_DST;
		}
	} else {
		if (PF_REVERSED_KEY(*state, pd->af)) {
			src = &(*state)->src;
			dst = &(*state)->dst;
			psrc = PF_PEER_SRC;
			pdst = PF_PEER_DST;
		} else {
			src = &(*state)->dst;
			dst = &(*state)->src;
			psrc = PF_PEER_DST;
			pdst = PF_PEER_SRC;
		}
	}

	switch (pd->virtual_proto) {
	case IPPROTO_TCP: {
		struct tcphdr		*th = &pd->hdr.tcp;

		if ((action = pf_synproxy(pd, *state, reason)) != PF_PASS)
			return (action);
		if (((tcp_get_flags(th) & (TH_SYN | TH_ACK)) == TH_SYN) ||
		    ((th->th_flags & (TH_SYN | TH_ACK | TH_RST)) == TH_ACK &&
		    pf_syncookie_check(pd) && pd->dir == PF_IN)) {
			if ((*state)->src.state >= TCPS_FIN_WAIT_2 &&
			    (*state)->dst.state >= TCPS_FIN_WAIT_2) {
				if (V_pf_status.debug >= PF_DEBUG_MISC) {
					printf("pf: state reuse ");
					pf_print_state(*state);
					pf_print_flags(tcp_get_flags(th));
					printf("\n");
				}
				/* XXX make sure it's the same direction ?? */
				pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED);
				pf_remove_state(*state);
				*state = NULL;
				return (PF_DROP);
			} else if ((*state)->src.state >= TCPS_ESTABLISHED &&
			    (*state)->dst.state >= TCPS_ESTABLISHED) {
				/*
				 * SYN matches existing state???
				 * Typically happens when sender boots up after
				 * sudden panic. Certain protocols (NFSv3) are
				 * always using same port numbers. Challenge
				 * ACK enables all parties (firewall and peers)
				 * to get in sync again.
				 */
				pf_send_challenge_ack(pd, *state, src, dst);
				return (PF_DROP);
			}
		}
		if ((*state)->state_flags & PFSTATE_SLOPPY) {
			if (pf_tcp_track_sloppy(*state, pd, reason, src, dst,
			    psrc, pdst) == PF_DROP)
				return (PF_DROP);
		} else {
			int	 ret;

			ret = pf_tcp_track_full(*state, pd, reason,
			    &copyback, src, dst, psrc, pdst);
			if (ret == PF_DROP)
				return (PF_DROP);
		}
		break;
	}
	case IPPROTO_UDP:
		/* update states */
		if (src->state < PFUDPS_SINGLE)
			pf_set_protostate(*state, psrc, PFUDPS_SINGLE);
		if (dst->state == PFUDPS_SINGLE)
			pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE);

		/* update expire time */
		(*state)->expire = pf_get_uptime();
		if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
			(*state)->timeout = PFTM_UDP_MULTIPLE;
		else
			(*state)->timeout = PFTM_UDP_SINGLE;
		break;
	case IPPROTO_SCTP:
		if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) &&
		    (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) &&
		    pd->sctp_flags & PFDESC_SCTP_INIT) {
			pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED);
			pf_remove_state(*state);
			*state = NULL;
			return (PF_DROP);
		}

		if (pf_sctp_track(*state, pd, reason) != PF_PASS)
			return (PF_DROP);

		/* Track state. */
		if (pd->sctp_flags & PFDESC_SCTP_INIT) {
			if (src->state < SCTP_COOKIE_WAIT) {
				pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT);
				(*state)->timeout = PFTM_SCTP_OPENING;
			}
		}
		if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) {
			MPASS(dst->scrub != NULL);
			if (dst->scrub->pfss_v_tag == 0)
				dst->scrub->pfss_v_tag = pd->sctp_initiate_tag;
		}

		/*
		 * Bind to the correct interface if we're if-bound. For multihomed
		 * extra associations we don't know which interface that will be until
		 * here, so we've inserted the state on V_pf_all. Fix that now.
		 */
		if ((*state)->kif == V_pfi_all &&
		    (*state)->rule->rule_flag & PFRULE_IFBOUND)
			(*state)->kif = pd->kif;

		if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) {
			if (src->state < SCTP_ESTABLISHED) {
				pf_set_protostate(*state, psrc, SCTP_ESTABLISHED);
				(*state)->timeout = PFTM_SCTP_ESTABLISHED;
			}
		}
		if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN |
		    PFDESC_SCTP_SHUTDOWN_COMPLETE)) {
			if (src->state < SCTP_SHUTDOWN_PENDING) {
				pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING);
				(*state)->timeout = PFTM_SCTP_CLOSING;
			}
		}
		if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE | PFDESC_SCTP_ABORT)) {
			pf_set_protostate(*state, psrc, SCTP_CLOSED);
			(*state)->timeout = PFTM_SCTP_CLOSED;
		}

		(*state)->expire = pf_get_uptime();
		break;
	default:
		/* update states */
		if (src->state < PFOTHERS_SINGLE)
			pf_set_protostate(*state, psrc, PFOTHERS_SINGLE);
		if (dst->state == PFOTHERS_SINGLE)
			pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE);

		/* update expire time */
		(*state)->expire = pf_get_uptime();
		if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
			(*state)->timeout = PFTM_OTHER_MULTIPLE;
		else
			(*state)->timeout = PFTM_OTHER_SINGLE;
		break;
	}

	/* translate source/destination address, if necessary */
	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
		struct pf_state_key	*nk;
		int			 afto, sidx, didx;

		if (PF_REVERSED_KEY(*state, pd->af))
			nk = (*state)->key[pd->sidx];
		else
			nk = (*state)->key[pd->didx];

		afto = pd->af != nk->af;

		if (afto && (*state)->direction == PF_IN) {
			sidx = pd->didx;
			didx = pd->sidx;
		} else {
			sidx = pd->sidx;
			didx = pd->didx;
		}

		if (afto) {
			pf_addrcpy(&pd->nsaddr, &nk->addr[sidx], nk->af);
			pf_addrcpy(&pd->ndaddr, &nk->addr[didx], nk->af);
			pd->naf = nk->af;
			action = PF_AFRT;
		}

		if (afto || PF_ANEQ(pd->src, &nk->addr[sidx], pd->af) ||
		    nk->port[sidx] != pd->osport)
			pf_change_ap(pd, pd->src, pd->sport,
			    &nk->addr[sidx], nk->port[sidx]);

		if (afto || PF_ANEQ(pd->dst, &nk->addr[didx], pd->af) ||
		    nk->port[didx] != pd->odport)
			pf_change_ap(pd, pd->dst, pd->dport,
			    &nk->addr[didx], nk->port[didx]);

		copyback = 1;
	}

	if (copyback && pd->hdrlen > 0)
		m_copyback(pd->m, pd->off, pd->hdrlen, pd->hdr.any);

	return (action);
}

static int
pf_sctp_track(struct pf_kstate *state, struct pf_pdesc *pd,
    u_short *reason)
{
	struct pf_state_peer	*src;
	if (pd->dir == state->direction) {
		if (PF_REVERSED_KEY(state, pd->af))
			src = &state->dst;
		else
			src = &state->src;
	} else {
		if (PF_REVERSED_KEY(state, pd->af))
			src = &state->src;
		else
			src = &state->dst;
	}

	if (src->scrub != NULL) {
		if (src->scrub->pfss_v_tag == 0)
			src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag;
		else  if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag)
			return (PF_DROP);
	}

	return (PF_PASS);
}

static void
pf_sctp_multihome_detach_addr(const struct pf_kstate *s)
{
	struct pf_sctp_endpoint key;
	struct pf_sctp_endpoint *ep;
	struct pf_state_key *sks = s->key[PF_SK_STACK];
	struct pf_sctp_source *i, *tmp;

	if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL)
		return;

	PF_SCTP_ENDPOINTS_LOCK();

	key.v_tag = s->dst.scrub->pfss_v_tag;
	ep  = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
	if (ep != NULL) {
		TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) {
			if (pf_addr_cmp(&i->addr,
			    &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT],
			    s->key[PF_SK_WIRE]->af) == 0) {
				SDT_PROBE3(pf, sctp, multihome, remove,
				    key.v_tag, s, i);
				TAILQ_REMOVE(&ep->sources, i, entry);
				free(i, M_PFTEMP);
				break;
			}
		}

		if (TAILQ_EMPTY(&ep->sources)) {
			RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
			free(ep, M_PFTEMP);
		}
	}

	/* Other direction. */
	key.v_tag = s->src.scrub->pfss_v_tag;
	ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
	if (ep != NULL) {
		TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) {
			if (pf_addr_cmp(&i->addr,
			    &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN],
			    s->key[PF_SK_WIRE]->af) == 0) {
				SDT_PROBE3(pf, sctp, multihome, remove,
				    key.v_tag, s, i);
				TAILQ_REMOVE(&ep->sources, i, entry);
				free(i, M_PFTEMP);
				break;
			}
		}

		if (TAILQ_EMPTY(&ep->sources)) {
			RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
			free(ep, M_PFTEMP);
		}
	}

	PF_SCTP_ENDPOINTS_UNLOCK();
}

static void
pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag)
{
	struct pf_sctp_endpoint key = {
		.v_tag = v_tag,
	};
	struct pf_sctp_source *i;
	struct pf_sctp_endpoint *ep;
	int count;

	PF_SCTP_ENDPOINTS_LOCK();

	ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
	if (ep == NULL) {
		ep = malloc(sizeof(struct pf_sctp_endpoint),
		    M_PFTEMP, M_NOWAIT);
		if (ep == NULL) {
			PF_SCTP_ENDPOINTS_UNLOCK();
			return;
		}

		ep->v_tag = v_tag;
		TAILQ_INIT(&ep->sources);
		RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
	}

	/* Avoid inserting duplicates. */
	count = 0;
	TAILQ_FOREACH(i, &ep->sources, entry) {
		count++;
		if (pf_addr_cmp(&i->addr, a, pd->af) == 0) {
			PF_SCTP_ENDPOINTS_UNLOCK();
			return;
		}
	}

	/* Limit the number of addresses per endpoint. */
	if (count >= PF_SCTP_MAX_ENDPOINTS) {
		PF_SCTP_ENDPOINTS_UNLOCK();
		return;
	}

	i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT);
	if (i == NULL) {
		PF_SCTP_ENDPOINTS_UNLOCK();
		return;
	}

	i->af = pd->af;
	memcpy(&i->addr, a, sizeof(*a));
	TAILQ_INSERT_TAIL(&ep->sources, i, entry);
	SDT_PROBE2(pf, sctp, multihome, add, v_tag, i);

	PF_SCTP_ENDPOINTS_UNLOCK();
}

static void
pf_sctp_multihome_delayed(struct pf_pdesc *pd, struct pfi_kkif *kif,
    struct pf_kstate *s, int action)
{
	struct pf_sctp_multihome_job	*j, *tmp;
	struct pf_sctp_source		*i;
	int			 ret;
	struct pf_kstate	*sm = NULL;
	struct pf_krule		*ra = NULL;
	struct pf_krule		*r = &V_pf_default_rule;
	struct pf_kruleset	*rs = NULL;
	u_short			 reason;
	bool do_extra = true;

	PF_RULES_RLOCK_TRACKER;

again:
	TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) {
		if (s == NULL || action != PF_PASS)
			goto free;

		/* Confirm we don't recurse here. */
		MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP));

		switch (j->op) {
		case  SCTP_ADD_IP_ADDRESS: {
			uint32_t v_tag = pd->sctp_initiate_tag;

			if (v_tag == 0) {
				if (s->direction == pd->dir)
					v_tag = s->src.scrub->pfss_v_tag;
				else
					v_tag = s->dst.scrub->pfss_v_tag;
			}

			/*
			 * Avoid duplicating states. We'll already have
			 * created a state based on the source address of
			 * the packet, but SCTP endpoints may also list this
			 * address again in the INIT(_ACK) parameters.
			 */
			if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) {
				break;
			}

			j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP;
			PF_RULES_RLOCK();
			sm = NULL;
			if (s->rule->rule_flag & PFRULE_ALLOW_RELATED) {
				j->pd.related_rule = s->rule;
			}
			ret = pf_test_rule(&r, &sm,
			    &j->pd, &ra, &rs, &reason, NULL);
			PF_RULES_RUNLOCK();
			SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->pd.m, ret);
			if (ret != PF_DROP && sm != NULL) {
				/* Inherit v_tag values. */
				if (sm->direction == s->direction) {
					sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag;
					sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag;
				} else {
					sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag;
					sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag;
				}
				PF_STATE_UNLOCK(sm);
			} else {
				/* If we try duplicate inserts? */
				break;
			}

			/* Only add the address if we've actually allowed the state. */
			pf_sctp_multihome_add_addr(pd, &j->src, v_tag);

			if (! do_extra) {
				break;
			}
			/*
			 * We need to do this for each of our source addresses.
			 * Find those based on the verification tag.
			 */
			struct pf_sctp_endpoint key = {
				.v_tag = pd->hdr.sctp.v_tag,
			};
			struct pf_sctp_endpoint *ep;

			PF_SCTP_ENDPOINTS_LOCK();
			ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
			if (ep == NULL) {
				PF_SCTP_ENDPOINTS_UNLOCK();
				break;
			}
			MPASS(ep != NULL);

			TAILQ_FOREACH(i, &ep->sources, entry) {
				struct pf_sctp_multihome_job *nj;

				/* SCTP can intermingle IPv4 and IPv6. */
				if (i->af != pd->af)
					continue;

				nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO);
				if (! nj) {
					continue;
				}
				memcpy(&nj->pd, &j->pd, sizeof(j->pd));
				memcpy(&nj->src, &j->src, sizeof(nj->src));
				nj->pd.src = &nj->src;
				// New destination address!
				memcpy(&nj->dst, &i->addr, sizeof(nj->dst));
				nj->pd.dst = &nj->dst;
				nj->pd.m = j->pd.m;
				nj->op = j->op;

				TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next);
			}
			PF_SCTP_ENDPOINTS_UNLOCK();

			break;
		}
		case SCTP_DEL_IP_ADDRESS: {
			struct pf_state_key_cmp key;
			uint8_t psrc;
			int action;

			bzero(&key, sizeof(key));
			key.af = j->pd.af;
			key.proto = IPPROTO_SCTP;
			if (j->pd.dir == PF_IN)	{	/* wire side, straight */
				pf_addrcpy(&key.addr[0], j->pd.src, key.af);
				pf_addrcpy(&key.addr[1], j->pd.dst, key.af);
				key.port[0] = j->pd.hdr.sctp.src_port;
				key.port[1] = j->pd.hdr.sctp.dest_port;
			} else {			/* stack side, reverse */
				pf_addrcpy(&key.addr[1], j->pd.src, key.af);
				pf_addrcpy(&key.addr[0], j->pd.dst, key.af);
				key.port[1] = j->pd.hdr.sctp.src_port;
				key.port[0] = j->pd.hdr.sctp.dest_port;
			}

			action = pf_find_state(&j->pd, &key, &sm);
			if (action == PF_MATCH) {
				PF_STATE_LOCK_ASSERT(sm);
				if (j->pd.dir == sm->direction) {
					psrc = PF_PEER_SRC;
				} else {
					psrc = PF_PEER_DST;
				}
				pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING);
				sm->timeout = PFTM_SCTP_CLOSING;
				PF_STATE_UNLOCK(sm);
			}
			break;
		default:
			panic("Unknown op %#x", j->op);
		}
	}

	free:
		TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next);
		free(j, M_PFTEMP);
	}

	/* We may have inserted extra work while processing the list. */
	if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) {
		do_extra = false;
		goto again;
	}
}

static int
pf_multihome_scan(int start, int len, struct pf_pdesc *pd, int op)
{
	int			 off = 0;
	struct pf_sctp_multihome_job	*job;

	SDT_PROBE4(pf, sctp, multihome_scan, entry, start, len, pd, op);

	while (off < len) {
		struct sctp_paramhdr h;

		if (!pf_pull_hdr(pd->m, start + off, &h, sizeof(h), NULL, NULL,
		    pd->af))
			return (PF_DROP);

		/* Parameters are at least 4 bytes. */
		if (ntohs(h.param_length) < 4)
			return (PF_DROP);

		SDT_PROBE2(pf, sctp, multihome_scan, param, ntohs(h.param_type),
		    ntohs(h.param_length));

		switch (ntohs(h.param_type)) {
		case  SCTP_IPV4_ADDRESS: {
			struct in_addr t;

			if (ntohs(h.param_length) !=
			    (sizeof(struct sctp_paramhdr) + sizeof(t)))
				return (PF_DROP);

			if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t),
			    NULL, NULL, pd->af))
				return (PF_DROP);

			if (in_nullhost(t))
				t.s_addr = pd->src->v4.s_addr;

			/*
			 * We hold the state lock (idhash) here, which means
			 * that we can't acquire the keyhash, or we'll get a
			 * LOR (and potentially double-lock things too). We also
			 * can't release the state lock here, so instead we'll
			 * enqueue this for async handling.
			 * There's a relatively small race here, in that a
			 * packet using the new addresses could arrive already,
			 * but that's just though luck for it.
			 */
			job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO);
			if (! job)
				return (PF_DROP);

			SDT_PROBE2(pf, sctp, multihome_scan, ipv4, &t, op);

			memcpy(&job->pd, pd, sizeof(*pd));

			// New source address!
			memcpy(&job->src, &t, sizeof(t));
			job->pd.src = &job->src;
			memcpy(&job->dst, pd->dst, sizeof(job->dst));
			job->pd.dst = &job->dst;
			job->pd.m = pd->m;
			job->op = op;

			TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next);
			break;
		}
#ifdef INET6
		case SCTP_IPV6_ADDRESS: {
			struct in6_addr t;

			if (ntohs(h.param_length) !=
			    (sizeof(struct sctp_paramhdr) + sizeof(t)))
				return (PF_DROP);

			if (!pf_pull_hdr(pd->m, start + off + sizeof(h), &t, sizeof(t),
			    NULL, NULL, pd->af))
				return (PF_DROP);
			if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0)
				break;
			if (memcmp(&t, &in6addr_any, sizeof(t)) == 0)
				memcpy(&t, &pd->src->v6, sizeof(t));

			job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO);
			if (! job)
				return (PF_DROP);

			SDT_PROBE2(pf, sctp, multihome_scan, ipv6, &t, op);

			memcpy(&job->pd, pd, sizeof(*pd));
			memcpy(&job->src, &t, sizeof(t));
			job->pd.src = &job->src;
			memcpy(&job->dst, pd->dst, sizeof(job->dst));
			job->pd.dst = &job->dst;
			job->pd.m = pd->m;
			job->op = op;

			TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next);
			break;
		}
#endif /* INET6 */
		case SCTP_ADD_IP_ADDRESS: {
			int ret;
			struct sctp_asconf_paramhdr ah;

			if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah),
			    NULL, NULL, pd->af))
				return (PF_DROP);

			ret = pf_multihome_scan(start + off + sizeof(ah),
			    ntohs(ah.ph.param_length) - sizeof(ah), pd,
			    SCTP_ADD_IP_ADDRESS);
			if (ret != PF_PASS)
				return (ret);
			break;
		}
		case SCTP_DEL_IP_ADDRESS: {
			int ret;
			struct sctp_asconf_paramhdr ah;

			if (!pf_pull_hdr(pd->m, start + off, &ah, sizeof(ah),
			    NULL, NULL, pd->af))
				return (PF_DROP);
			ret = pf_multihome_scan(start + off + sizeof(ah),
			    ntohs(ah.ph.param_length) - sizeof(ah), pd,
			    SCTP_DEL_IP_ADDRESS);
			if (ret != PF_PASS)
				return (ret);
			break;
		}
		default:
			break;
		}

		off += roundup(ntohs(h.param_length), 4);
	}

	return (PF_PASS);
}

int
pf_multihome_scan_init(int start, int len, struct pf_pdesc *pd)
{
	start += sizeof(struct sctp_init_chunk);
	len -= sizeof(struct sctp_init_chunk);

	return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS));
}

int
pf_multihome_scan_asconf(int start, int len, struct pf_pdesc *pd)
{
	start += sizeof(struct sctp_asconf_chunk);
	len -= sizeof(struct sctp_asconf_chunk);

	return (pf_multihome_scan(start, len, pd, SCTP_ADD_IP_ADDRESS));
}

int
pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd,
    struct pf_kstate **state, u_int16_t icmpid, u_int16_t type, int icmp_dir,
    int *iidx, int multi, int inner)
{
	int	 action, direction = pd->dir;

	key->af = pd->af;
	key->proto = pd->proto;
	if (icmp_dir == PF_IN) {
		*iidx = pd->sidx;
		key->port[pd->sidx] = icmpid;
		key->port[pd->didx] = type;
	} else {
		*iidx = pd->didx;
		key->port[pd->sidx] = type;
		key->port[pd->didx] = icmpid;
	}
	if (pf_state_key_addr_setup(pd, key, multi))
		return (PF_DROP);

	action = pf_find_state(pd, key, state);
	if (action != PF_MATCH)
		return (action);

	if ((*state)->state_flags & PFSTATE_SLOPPY)
		return (-1);

	/* Is this ICMP message flowing in right direction? */
	if ((*state)->key[PF_SK_WIRE]->af != (*state)->key[PF_SK_STACK]->af)
		direction = (pd->af == (*state)->key[PF_SK_WIRE]->af) ?
		    PF_IN : PF_OUT;
	else
		direction = (*state)->direction;
	if ((*state)->rule->type &&
	    (((!inner && direction == pd->dir) ||
	    (inner && direction != pd->dir)) ?
	    PF_IN : PF_OUT) != icmp_dir) {
		if (V_pf_status.debug >= PF_DEBUG_MISC) {
			printf("pf: icmp type %d in wrong direction (%d): ",
			    ntohs(type), icmp_dir);
			pf_print_state(*state);
			printf("\n");
		}
		PF_STATE_UNLOCK(*state);
		*state = NULL;
		return (PF_DROP);
	}
	return (-1);
}

static int
pf_test_state_icmp(struct pf_kstate **state, struct pf_pdesc *pd,
    u_short *reason)
{
	struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
	u_int16_t	*icmpsum, virtual_id, virtual_type;
	u_int8_t	 icmptype, icmpcode;
	int		 icmp_dir, iidx, ret;
	struct pf_state_key_cmp key;
#ifdef INET
	u_int16_t	 icmpid;
#endif /* INET*/

	MPASS(*state == NULL);

	bzero(&key, sizeof(key));
	switch (pd->proto) {
#ifdef INET
	case IPPROTO_ICMP:
		icmptype = pd->hdr.icmp.icmp_type;
		icmpcode = pd->hdr.icmp.icmp_code;
		icmpid = pd->hdr.icmp.icmp_id;
		icmpsum = &pd->hdr.icmp.icmp_cksum;
		break;
#endif /* INET */
#ifdef INET6
	case IPPROTO_ICMPV6:
		icmptype = pd->hdr.icmp6.icmp6_type;
		icmpcode = pd->hdr.icmp6.icmp6_code;
#ifdef INET
		icmpid = pd->hdr.icmp6.icmp6_id;
#endif /* INET */
		icmpsum = &pd->hdr.icmp6.icmp6_cksum;
		break;
#endif /* INET6 */
	default:
		panic("unhandled proto %d", pd->proto);
	}

	if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &virtual_id,
	    &virtual_type) == 0) {
		/*
		 * ICMP query/reply message not related to a TCP/UDP/SCTP
		 * packet. Search for an ICMP state.
		 */
		ret = pf_icmp_state_lookup(&key, pd, state, virtual_id,
		    virtual_type, icmp_dir, &iidx, 0, 0);
		/* IPv6? try matching a multicast address */
		if (ret == PF_DROP && pd->af == AF_INET6 && icmp_dir == PF_OUT) {
			MPASS(*state == NULL);
			ret = pf_icmp_state_lookup(&key, pd, state,
			    virtual_id, virtual_type,
			    icmp_dir, &iidx, 1, 0);
		}
		if (ret >= 0) {
			MPASS(*state == NULL);
			return (ret);
		}

		(*state)->expire = pf_get_uptime();
		(*state)->timeout = PFTM_ICMP_ERROR_REPLY;

		/* translate source/destination address, if necessary */
		if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
			struct pf_state_key	*nk;
			int			 afto, sidx, didx;

			if (PF_REVERSED_KEY(*state, pd->af))
				nk = (*state)->key[pd->sidx];
			else
				nk = (*state)->key[pd->didx];

			afto = pd->af != nk->af;

			if (afto && (*state)->direction == PF_IN) {
				sidx = pd->didx;
				didx = pd->sidx;
				iidx = !iidx;
			} else {
				sidx = pd->sidx;
				didx = pd->didx;
			}

			switch (pd->af) {
#ifdef INET
			case AF_INET:
#ifdef INET6
				if (afto) {
					if (pf_translate_icmp_af(AF_INET6,
					    &pd->hdr.icmp))
						return (PF_DROP);
					pd->proto = IPPROTO_ICMPV6;
				}
#endif /* INET6 */
				if (!afto &&
				    PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET))
					pf_change_a(&saddr->v4.s_addr,
					    pd->ip_sum,
					    nk->addr[sidx].v4.s_addr,
					    0);

				if (!afto && PF_ANEQ(pd->dst,
				    &nk->addr[didx], AF_INET))
					pf_change_a(&daddr->v4.s_addr,
					    pd->ip_sum,
					    nk->addr[didx].v4.s_addr, 0);

				if (nk->port[iidx] !=
				    pd->hdr.icmp.icmp_id) {
					pd->hdr.icmp.icmp_cksum =
					    pf_cksum_fixup(
					    pd->hdr.icmp.icmp_cksum, icmpid,
					    nk->port[iidx], 0);
					pd->hdr.icmp.icmp_id =
					    nk->port[iidx];
				}

				m_copyback(pd->m, pd->off, ICMP_MINLEN,
				    (caddr_t )&pd->hdr.icmp);
				break;
#endif /* INET */
#ifdef INET6
			case AF_INET6:
#ifdef INET
				if (afto) {
					if (pf_translate_icmp_af(AF_INET,
					    &pd->hdr.icmp6))
						return (PF_DROP);
					pd->proto = IPPROTO_ICMP;
				}
#endif /* INET */
				if (!afto &&
				    PF_ANEQ(pd->src, &nk->addr[sidx], AF_INET6))
					pf_change_a6(saddr,
					    &pd->hdr.icmp6.icmp6_cksum,
					    &nk->addr[sidx], 0);

				if (!afto && PF_ANEQ(pd->dst,
				    &nk->addr[didx], AF_INET6))
					pf_change_a6(daddr,
					    &pd->hdr.icmp6.icmp6_cksum,
					    &nk->addr[didx], 0);

				if (nk->port[iidx] != pd->hdr.icmp6.icmp6_id)
					pd->hdr.icmp6.icmp6_id =
					    nk->port[iidx];

				m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr),
				    (caddr_t )&pd->hdr.icmp6);
				break;
#endif /* INET6 */
			}
			if (afto) {
				pf_addrcpy(&pd->nsaddr, &nk->addr[sidx],
				    nk->af);
				pf_addrcpy(&pd->ndaddr, &nk->addr[didx],
				    nk->af);
				pd->naf = nk->af;
				return (PF_AFRT);
			}
		}
		return (PF_PASS);

	} else {
		/*
		 * ICMP error message in response to a TCP/UDP packet.
		 * Extract the inner TCP/UDP header and search for that state.
		 */

		struct pf_pdesc	pd2;
		bzero(&pd2, sizeof pd2);
#ifdef INET
		struct ip	h2;
#endif /* INET */
#ifdef INET6
		struct ip6_hdr	h2_6;
#endif /* INET6 */
		int		ipoff2 = 0;

		pd2.af = pd->af;
		pd2.dir = pd->dir;
		/* Payload packet is from the opposite direction. */
		pd2.sidx = (pd->dir == PF_IN) ? 1 : 0;
		pd2.didx = (pd->dir == PF_IN) ? 0 : 1;
		pd2.m = pd->m;
		pd2.pf_mtag = pd->pf_mtag;
		pd2.kif = pd->kif;
		switch (pd->af) {
#ifdef INET
		case AF_INET:
			/* offset of h2 in mbuf chain */
			ipoff2 = pd->off + ICMP_MINLEN;

			if (!pf_pull_hdr(pd->m, ipoff2, &h2, sizeof(h2),
			    NULL, reason, pd2.af)) {
				DPFPRINTF(PF_DEBUG_MISC,
				    "pf: ICMP error message too short "
				    "(ip)");
				return (PF_DROP);
			}
			/*
			 * ICMP error messages don't refer to non-first
			 * fragments
			 */
			if (h2.ip_off & htons(IP_OFFMASK)) {
				REASON_SET(reason, PFRES_FRAG);
				return (PF_DROP);
			}

			/* offset of protocol header that follows h2 */
			pd2.off = ipoff2;
			if (pf_walk_header(&pd2, &h2, reason) != PF_PASS)
				return (PF_DROP);

			pd2.tot_len = ntohs(h2.ip_len);
			pd2.ttl = h2.ip_ttl;
			pd2.src = (struct pf_addr *)&h2.ip_src;
			pd2.dst = (struct pf_addr *)&h2.ip_dst;
			pd2.ip_sum = &h2.ip_sum;
			break;
#endif /* INET */
#ifdef INET6
		case AF_INET6:
			ipoff2 = pd->off + sizeof(struct icmp6_hdr);

			if (!pf_pull_hdr(pd->m, ipoff2, &h2_6, sizeof(h2_6),
			    NULL, reason, pd2.af)) {
				DPFPRINTF(PF_DEBUG_MISC,
				    "pf: ICMP error message too short "
				    "(ip6)");
				return (PF_DROP);
			}
			pd2.off = ipoff2;
			if (pf_walk_header6(&pd2, &h2_6, reason) != PF_PASS)
				return (PF_DROP);

			pd2.tot_len = ntohs(h2_6.ip6_plen) +
			    sizeof(struct ip6_hdr);
			pd2.ttl = h2_6.ip6_hlim;
			pd2.src = (struct pf_addr *)&h2_6.ip6_src;
			pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
			pd2.ip_sum = NULL;
			break;
#endif /* INET6 */
		default:
			unhandled_af(pd->af);
		}

		if (PF_ANEQ(pd->dst, pd2.src, pd->af)) {
			if (V_pf_status.debug >= PF_DEBUG_MISC) {
				printf("pf: BAD ICMP %d:%d outer dst: ",
				    icmptype, icmpcode);
				pf_print_host(pd->src, 0, pd->af);
				printf(" -> ");
				pf_print_host(pd->dst, 0, pd->af);
				printf(" inner src: ");
				pf_print_host(pd2.src, 0, pd2.af);
				printf(" -> ");
				pf_print_host(pd2.dst, 0, pd2.af);
				printf("\n");
			}
			REASON_SET(reason, PFRES_BADSTATE);
			return (PF_DROP);
		}

		switch (pd2.proto) {
		case IPPROTO_TCP: {
			struct tcphdr		*th = &pd2.hdr.tcp;
			u_int32_t		 seq;
			struct pf_state_peer	*src, *dst;
			u_int8_t		 dws;
			int			 copyback = 0;
			int			 action;

			/*
			 * Only the first 8 bytes of the TCP header can be
			 * expected. Don't access any TCP header fields after
			 * th_seq, an ackskew test is not possible.
			 */
			if (!pf_pull_hdr(pd->m, pd2.off, th, 8, NULL, reason,
			    pd2.af)) {
				DPFPRINTF(PF_DEBUG_MISC,
				    "pf: ICMP error message too short "
				    "(tcp)");
				return (PF_DROP);
			}
			pd2.pcksum = &pd2.hdr.tcp.th_sum;

			key.af = pd2.af;
			key.proto = IPPROTO_TCP;
			pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af);
			pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af);
			key.port[pd2.sidx] = th->th_sport;
			key.port[pd2.didx] = th->th_dport;

			action = pf_find_state(&pd2, &key, state);
			if (action != PF_MATCH)
				return (action);

			if (pd->dir == (*state)->direction) {
				if (PF_REVERSED_KEY(*state, pd->af)) {
					src = &(*state)->src;
					dst = &(*state)->dst;
				} else {
					src = &(*state)->dst;
					dst = &(*state)->src;
				}
			} else {
				if (PF_REVERSED_KEY(*state, pd->af)) {
					src = &(*state)->dst;
					dst = &(*state)->src;
				} else {
					src = &(*state)->src;
					dst = &(*state)->dst;
				}
			}

			if (src->wscale && dst->wscale)
				dws = dst->wscale & PF_WSCALE_MASK;
			else
				dws = 0;

			/* Demodulate sequence number */
			seq = ntohl(th->th_seq) - src->seqdiff;
			if (src->seqdiff) {
				pf_change_a(&th->th_seq, icmpsum,
				    htonl(seq), 0);
				copyback = 1;
			}

			if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
			    (!SEQ_GEQ(src->seqhi, seq) ||
			    !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
				if (V_pf_status.debug >= PF_DEBUG_MISC) {
					printf("pf: BAD ICMP %d:%d ",
					    icmptype, icmpcode);
					pf_print_host(pd->src, 0, pd->af);
					printf(" -> ");
					pf_print_host(pd->dst, 0, pd->af);
					printf(" state: ");
					pf_print_state(*state);
					printf(" seq=%u\n", seq);
				}
				REASON_SET(reason, PFRES_BADSTATE);
				return (PF_DROP);
			} else {
				if (V_pf_status.debug >= PF_DEBUG_MISC) {
					printf("pf: OK ICMP %d:%d ",
					    icmptype, icmpcode);
					pf_print_host(pd->src, 0, pd->af);
					printf(" -> ");
					pf_print_host(pd->dst, 0, pd->af);
					printf(" state: ");
					pf_print_state(*state);
					printf(" seq=%u\n", seq);
				}
			}

			/* translate source/destination address, if necessary */
			if ((*state)->key[PF_SK_WIRE] !=
			    (*state)->key[PF_SK_STACK]) {

				struct pf_state_key	*nk;

				if (PF_REVERSED_KEY(*state, pd->af))
					nk = (*state)->key[pd->sidx];
				else
					nk = (*state)->key[pd->didx];

#if defined(INET) && defined(INET6)
				int		 afto, sidx, didx;

				afto = pd->af != nk->af;

				if (afto && (*state)->direction == PF_IN) {
					sidx = pd2.didx;
					didx = pd2.sidx;
				} else {
					sidx = pd2.sidx;
					didx = pd2.didx;
				}

				if (afto) {
					if (pf_translate_icmp_af(nk->af,
					    &pd->hdr.icmp))
						return (PF_DROP);
					m_copyback(pd->m, pd->off,
					    sizeof(struct icmp6_hdr),
					    (c_caddr_t)&pd->hdr.icmp6);
					if (pf_change_icmp_af(pd->m, ipoff2, pd,
					    &pd2, &nk->addr[sidx],
					    &nk->addr[didx], pd->af,
					    nk->af))
						return (PF_DROP);
					pf_addrcpy(&pd->nsaddr,
					    &nk->addr[pd2.sidx], nk->af);
					pf_addrcpy(&pd->ndaddr,
					    &nk->addr[pd2.didx], nk->af);
					if (nk->af == AF_INET) {
						pd->proto = IPPROTO_ICMP;
					} else {
						pd->proto = IPPROTO_ICMPV6;
						/*
						 * IPv4 becomes IPv6 so we must
						 * copy IPv4 src addr to least
						 * 32bits in IPv6 address to
						 * keep traceroute/icmp
						 * working.
						 */
						pd->nsaddr.addr32[3] =
						    pd->src->addr32[0];
					}
					pd->naf = pd2.naf = nk->af;
					pf_change_ap(&pd2, pd2.src, &th->th_sport,
					    &nk->addr[pd2.sidx], nk->port[sidx]);
					pf_change_ap(&pd2, pd2.dst, &th->th_dport,
					    &nk->addr[pd2.didx], nk->port[didx]);
					m_copyback(pd2.m, pd2.off, 8, (c_caddr_t)th);
					return (PF_AFRT);
				}
#endif /* INET && INET6 */

				if (PF_ANEQ(pd2.src,
				    &nk->addr[pd2.sidx], pd2.af) ||
				    nk->port[pd2.sidx] != th->th_sport)
					pf_change_icmp(pd2.src, &th->th_sport,
					    daddr, &nk->addr[pd2.sidx],
					    nk->port[pd2.sidx], NULL,
					    pd2.ip_sum, icmpsum,
					    pd->ip_sum, 0, pd2.af);

				if (PF_ANEQ(pd2.dst,
				    &nk->addr[pd2.didx], pd2.af) ||
				    nk->port[pd2.didx] != th->th_dport)
					pf_change_icmp(pd2.dst, &th->th_dport,
					    saddr, &nk->addr[pd2.didx],
					    nk->port[pd2.didx], NULL,
					    pd2.ip_sum, icmpsum,
					    pd->ip_sum, 0, pd2.af);
				copyback = 1;
			}

			if (copyback) {
				switch (pd2.af) {
#ifdef INET
				case AF_INET:
					m_copyback(pd->m, pd->off, ICMP_MINLEN,
					    (caddr_t )&pd->hdr.icmp);
					m_copyback(pd->m, ipoff2, sizeof(h2),
					    (caddr_t )&h2);
					break;
#endif /* INET */
#ifdef INET6
				case AF_INET6:
					m_copyback(pd->m, pd->off,
					    sizeof(struct icmp6_hdr),
					    (caddr_t )&pd->hdr.icmp6);
					m_copyback(pd->m, ipoff2, sizeof(h2_6),
					    (caddr_t )&h2_6);
					break;
#endif /* INET6 */
				default:
					unhandled_af(pd->af);
				}
				m_copyback(pd->m, pd2.off, 8, (caddr_t)th);
			}

			return (PF_PASS);
			break;
		}
		case IPPROTO_UDP: {
			struct udphdr		*uh = &pd2.hdr.udp;
			int			 action;

			if (!pf_pull_hdr(pd->m, pd2.off, uh, sizeof(*uh),
			    NULL, reason, pd2.af)) {
				DPFPRINTF(PF_DEBUG_MISC,
				    "pf: ICMP error message too short "
				    "(udp)");
				return (PF_DROP);
			}
			pd2.pcksum = &pd2.hdr.udp.uh_sum;

			key.af = pd2.af;
			key.proto = IPPROTO_UDP;
			pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af);
			pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af);
			key.port[pd2.sidx] = uh->uh_sport;
			key.port[pd2.didx] = uh->uh_dport;

			action = pf_find_state(&pd2, &key, state);
			if (action != PF_MATCH)
				return (action);

			/* translate source/destination address, if necessary */
			if ((*state)->key[PF_SK_WIRE] !=
			    (*state)->key[PF_SK_STACK]) {
				struct pf_state_key	*nk;

				if (PF_REVERSED_KEY(*state, pd->af))
					nk = (*state)->key[pd->sidx];
				else
					nk = (*state)->key[pd->didx];

#if defined(INET) && defined(INET6)
				int	 afto, sidx, didx;

				afto = pd->af != nk->af;

				if (afto && (*state)->direction == PF_IN) {
					sidx = pd2.didx;
					didx = pd2.sidx;
				} else {
					sidx = pd2.sidx;
					didx = pd2.didx;
				}

				if (afto) {
					if (pf_translate_icmp_af(nk->af,
					    &pd->hdr.icmp))
						return (PF_DROP);
					m_copyback(pd->m, pd->off,
					    sizeof(struct icmp6_hdr),
					    (c_caddr_t)&pd->hdr.icmp6);
					if (pf_change_icmp_af(pd->m, ipoff2, pd,
					    &pd2, &nk->addr[sidx],
					    &nk->addr[didx], pd->af,
					    nk->af))
						return (PF_DROP);
					pf_addrcpy(&pd->nsaddr,
					    &nk->addr[pd2.sidx], nk->af);
					pf_addrcpy(&pd->ndaddr,
					    &nk->addr[pd2.didx], nk->af);
					if (nk->af == AF_INET) {
						pd->proto = IPPROTO_ICMP;
					} else {
						pd->proto = IPPROTO_ICMPV6;
						/*
						 * IPv4 becomes IPv6 so we must
						 * copy IPv4 src addr to least
						 * 32bits in IPv6 address to
						 * keep traceroute/icmp
						 * working.
						 */
						pd->nsaddr.addr32[3] =
						    pd->src->addr32[0];
					}
					pd->naf = pd2.naf = nk->af;
					pf_change_ap(&pd2, pd2.src, &uh->uh_sport,
					    &nk->addr[pd2.sidx], nk->port[sidx]);
					pf_change_ap(&pd2, pd2.dst, &uh->uh_dport,
					    &nk->addr[pd2.didx], nk->port[didx]);
					m_copyback(pd2.m, pd2.off, sizeof(*uh),
					    (c_caddr_t)uh);
					return (PF_AFRT);
				}
#endif /* INET && INET6 */

				if (PF_ANEQ(pd2.src,
				    &nk->addr[pd2.sidx], pd2.af) ||
				    nk->port[pd2.sidx] != uh->uh_sport)
					pf_change_icmp(pd2.src, &uh->uh_sport,
					    daddr, &nk->addr[pd2.sidx],
					    nk->port[pd2.sidx], &uh->uh_sum,
					    pd2.ip_sum, icmpsum,
					    pd->ip_sum, 1, pd2.af);

				if (PF_ANEQ(pd2.dst,
				    &nk->addr[pd2.didx], pd2.af) ||
				    nk->port[pd2.didx] != uh->uh_dport)
					pf_change_icmp(pd2.dst, &uh->uh_dport,
					    saddr, &nk->addr[pd2.didx],
					    nk->port[pd2.didx], &uh->uh_sum,
					    pd2.ip_sum, icmpsum,
					    pd->ip_sum, 1, pd2.af);

				switch (pd2.af) {
#ifdef INET
				case AF_INET:
					m_copyback(pd->m, pd->off, ICMP_MINLEN,
					    (caddr_t )&pd->hdr.icmp);
					m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2);
					break;
#endif /* INET */
#ifdef INET6
				case AF_INET6:
					m_copyback(pd->m, pd->off,
					    sizeof(struct icmp6_hdr),
					    (caddr_t )&pd->hdr.icmp6);
					m_copyback(pd->m, ipoff2, sizeof(h2_6),
					    (caddr_t )&h2_6);
					break;
#endif /* INET6 */
				}
				m_copyback(pd->m, pd2.off, sizeof(*uh), (caddr_t)uh);
			}
			return (PF_PASS);
			break;
		}
#ifdef INET
		case IPPROTO_SCTP: {
			struct sctphdr		*sh = &pd2.hdr.sctp;
			struct pf_state_peer	*src;
			int			 copyback = 0;
			int			 action;

			if (! pf_pull_hdr(pd->m, pd2.off, sh, sizeof(*sh), NULL, reason,
			    pd2.af)) {
				DPFPRINTF(PF_DEBUG_MISC,
				    "pf: ICMP error message too short "
				    "(sctp)");
				return (PF_DROP);
			}
			pd2.pcksum = &pd2.sctp_dummy_sum;

			key.af = pd2.af;
			key.proto = IPPROTO_SCTP;
			pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af);
			pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af);
			key.port[pd2.sidx] = sh->src_port;
			key.port[pd2.didx] = sh->dest_port;

			action = pf_find_state(&pd2, &key, state);
			if (action != PF_MATCH)
				return (action);

			if (pd->dir == (*state)->direction) {
				if (PF_REVERSED_KEY(*state, pd->af))
					src = &(*state)->src;
				else
					src = &(*state)->dst;
			} else {
				if (PF_REVERSED_KEY(*state, pd->af))
					src = &(*state)->dst;
				else
					src = &(*state)->src;
			}

			if (src->scrub->pfss_v_tag != sh->v_tag) {
				DPFPRINTF(PF_DEBUG_MISC,
				    "pf: ICMP error message has incorrect "
				    "SCTP v_tag");
				return (PF_DROP);
			}

			/* translate source/destination address, if necessary */
			if ((*state)->key[PF_SK_WIRE] !=
			    (*state)->key[PF_SK_STACK]) {

				struct pf_state_key	*nk;

				if (PF_REVERSED_KEY(*state, pd->af))
					nk = (*state)->key[pd->sidx];
				else
					nk = (*state)->key[pd->didx];

#if defined(INET) && defined(INET6)
				int	 afto, sidx, didx;

				afto = pd->af != nk->af;

				if (afto && (*state)->direction == PF_IN) {
					sidx = pd2.didx;
					didx = pd2.sidx;
				} else {
					sidx = pd2.sidx;
					didx = pd2.didx;
				}

				if (afto) {
					if (pf_translate_icmp_af(nk->af,
					    &pd->hdr.icmp))
						return (PF_DROP);
					m_copyback(pd->m, pd->off,
					    sizeof(struct icmp6_hdr),
					    (c_caddr_t)&pd->hdr.icmp6);
					if (pf_change_icmp_af(pd->m, ipoff2, pd,
					    &pd2, &nk->addr[sidx],
					    &nk->addr[didx], pd->af,
					    nk->af))
						return (PF_DROP);
					sh->src_port = nk->port[sidx];
					sh->dest_port = nk->port[didx];
					m_copyback(pd2.m, pd2.off, sizeof(*sh), (c_caddr_t)sh);
					pf_addrcpy(&pd->nsaddr,
					    &nk->addr[pd2.sidx], nk->af);
					pf_addrcpy(&pd->ndaddr,
					    &nk->addr[pd2.didx], nk->af);
					if (nk->af == AF_INET) {
						pd->proto = IPPROTO_ICMP;
					} else {
						pd->proto = IPPROTO_ICMPV6;
						/*
						 * IPv4 becomes IPv6 so we must
						 * copy IPv4 src addr to least
						 * 32bits in IPv6 address to
						 * keep traceroute/icmp
						 * working.
						 */
						pd->nsaddr.addr32[3] =
						    pd->src->addr32[0];
					}
					pd->naf = nk->af;
					return (PF_AFRT);
				}
#endif /* INET && INET6 */

				if (PF_ANEQ(pd2.src,
				    &nk->addr[pd2.sidx], pd2.af) ||
				    nk->port[pd2.sidx] != sh->src_port)
					pf_change_icmp(pd2.src, &sh->src_port,
					    daddr, &nk->addr[pd2.sidx],
					    nk->port[pd2.sidx], NULL,
					    pd2.ip_sum, icmpsum,
					    pd->ip_sum, 0, pd2.af);

				if (PF_ANEQ(pd2.dst,
				    &nk->addr[pd2.didx], pd2.af) ||
				    nk->port[pd2.didx] != sh->dest_port)
					pf_change_icmp(pd2.dst, &sh->dest_port,
					    saddr, &nk->addr[pd2.didx],
					    nk->port[pd2.didx], NULL,
					    pd2.ip_sum, icmpsum,
					    pd->ip_sum, 0, pd2.af);
				copyback = 1;
			}

			if (copyback) {
				switch (pd2.af) {
#ifdef INET
				case AF_INET:
					m_copyback(pd->m, pd->off, ICMP_MINLEN,
					    (caddr_t )&pd->hdr.icmp);
					m_copyback(pd->m, ipoff2, sizeof(h2),
					    (caddr_t )&h2);
					break;
#endif /* INET */
#ifdef INET6
				case AF_INET6:
					m_copyback(pd->m, pd->off,
					    sizeof(struct icmp6_hdr),
					    (caddr_t )&pd->hdr.icmp6);
					m_copyback(pd->m, ipoff2, sizeof(h2_6),
					    (caddr_t )&h2_6);
					break;
#endif /* INET6 */
				}
				m_copyback(pd->m, pd2.off, sizeof(*sh), (caddr_t)sh);
			}

			return (PF_PASS);
			break;
		}
		case IPPROTO_ICMP: {
			struct icmp	*iih = &pd2.hdr.icmp;

			if (pd2.af != AF_INET) {
				REASON_SET(reason, PFRES_NORM);
				return (PF_DROP);
			}

			if (!pf_pull_hdr(pd->m, pd2.off, iih, ICMP_MINLEN,
			    NULL, reason, pd2.af)) {
				DPFPRINTF(PF_DEBUG_MISC,
				    "pf: ICMP error message too short i"
				    "(icmp)");
				return (PF_DROP);
			}
			pd2.pcksum = &pd2.hdr.icmp.icmp_cksum;

			icmpid = iih->icmp_id;
			pf_icmp_mapping(&pd2, iih->icmp_type,
			    &icmp_dir, &virtual_id, &virtual_type);

			ret = pf_icmp_state_lookup(&key, &pd2, state,
			    virtual_id, virtual_type, icmp_dir, &iidx, 0, 1);
			if (ret >= 0) {
				MPASS(*state == NULL);
				return (ret);
			}

			/* translate source/destination address, if necessary */
			if ((*state)->key[PF_SK_WIRE] !=
			    (*state)->key[PF_SK_STACK]) {
				struct pf_state_key	*nk;

				if (PF_REVERSED_KEY(*state, pd->af))
					nk = (*state)->key[pd->sidx];
				else
					nk = (*state)->key[pd->didx];

#if defined(INET) && defined(INET6)
				int	 afto, sidx, didx;

				afto = pd->af != nk->af;

				if (afto && (*state)->direction == PF_IN) {
					sidx = pd2.didx;
					didx = pd2.sidx;
					iidx = !iidx;
				} else {
					sidx = pd2.sidx;
					didx = pd2.didx;
				}

				if (afto) {
					if (nk->af != AF_INET6)
						return (PF_DROP);
					if (pf_translate_icmp_af(nk->af,
					    &pd->hdr.icmp))
						return (PF_DROP);
					m_copyback(pd->m, pd->off,
					    sizeof(struct icmp6_hdr),
					    (c_caddr_t)&pd->hdr.icmp6);
					if (pf_change_icmp_af(pd->m, ipoff2, pd,
					    &pd2, &nk->addr[sidx],
					    &nk->addr[didx], pd->af,
					    nk->af))
						return (PF_DROP);
					pd->proto = IPPROTO_ICMPV6;
					if (pf_translate_icmp_af(nk->af, iih))
						return (PF_DROP);
					if (virtual_type == htons(ICMP_ECHO) &&
					    nk->port[iidx] != iih->icmp_id)
						iih->icmp_id = nk->port[iidx];
					m_copyback(pd2.m, pd2.off, ICMP_MINLEN,
					    (c_caddr_t)iih);
					pf_addrcpy(&pd->nsaddr,
					    &nk->addr[pd2.sidx], nk->af);
					pf_addrcpy(&pd->ndaddr,
					    &nk->addr[pd2.didx], nk->af);
					/*
					 * IPv4 becomes IPv6 so we must copy
					 * IPv4 src addr to least 32bits in
					 * IPv6 address to keep traceroute
					 * working.
					 */
					pd->nsaddr.addr32[3] =
					    pd->src->addr32[0];
					pd->naf = nk->af;
					return (PF_AFRT);
				}
#endif /* INET && INET6 */

				if (PF_ANEQ(pd2.src,
				    &nk->addr[pd2.sidx], pd2.af) ||
				    (virtual_type == htons(ICMP_ECHO) &&
				    nk->port[iidx] != iih->icmp_id))
					pf_change_icmp(pd2.src,
					    (virtual_type == htons(ICMP_ECHO)) ?
					    &iih->icmp_id : NULL,
					    daddr, &nk->addr[pd2.sidx],
					    (virtual_type == htons(ICMP_ECHO)) ?
					    nk->port[iidx] : 0, NULL,
					    pd2.ip_sum, icmpsum,
					    pd->ip_sum, 0, AF_INET);

				if (PF_ANEQ(pd2.dst,
				    &nk->addr[pd2.didx], pd2.af))
					pf_change_icmp(pd2.dst, NULL, NULL,
					    &nk->addr[pd2.didx], 0, NULL,
					    pd2.ip_sum, icmpsum, pd->ip_sum, 0,
					    AF_INET);

				m_copyback(pd->m, pd->off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
				m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2);
				m_copyback(pd->m, pd2.off, ICMP_MINLEN, (caddr_t)iih);
			}
			return (PF_PASS);
			break;
		}
#endif /* INET */
#ifdef INET6
		case IPPROTO_ICMPV6: {
			struct icmp6_hdr	*iih = &pd2.hdr.icmp6;

			if (pd2.af != AF_INET6) {
				REASON_SET(reason, PFRES_NORM);
				return (PF_DROP);
			}

			if (!pf_pull_hdr(pd->m, pd2.off, iih,
			    sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
				DPFPRINTF(PF_DEBUG_MISC,
				    "pf: ICMP error message too short "
				    "(icmp6)");
				return (PF_DROP);
			}
			pd2.pcksum = &pd2.hdr.icmp6.icmp6_cksum;

			pf_icmp_mapping(&pd2, iih->icmp6_type,
			    &icmp_dir, &virtual_id, &virtual_type);

			ret = pf_icmp_state_lookup(&key, &pd2, state,
			    virtual_id, virtual_type, icmp_dir, &iidx, 0, 1);
			/* IPv6? try matching a multicast address */
			if (ret == PF_DROP && pd2.af == AF_INET6 &&
			    icmp_dir == PF_OUT) {
				MPASS(*state == NULL);
				ret = pf_icmp_state_lookup(&key, &pd2,
				    state, virtual_id, virtual_type,
				    icmp_dir, &iidx, 1, 1);
			}
			if (ret >= 0) {
				MPASS(*state == NULL);
				return (ret);
			}

			/* translate source/destination address, if necessary */
			if ((*state)->key[PF_SK_WIRE] !=
			    (*state)->key[PF_SK_STACK]) {
				struct pf_state_key	*nk;

				if (PF_REVERSED_KEY(*state, pd->af))
					nk = (*state)->key[pd->sidx];
				else
					nk = (*state)->key[pd->didx];

#if defined(INET) && defined(INET6)
				int	 afto, sidx, didx;

				afto = pd->af != nk->af;

				if (afto && (*state)->direction == PF_IN) {
					sidx = pd2.didx;
					didx = pd2.sidx;
					iidx = !iidx;
				} else {
					sidx = pd2.sidx;
					didx = pd2.didx;
				}

				if (afto) {
					if (nk->af != AF_INET)
						return (PF_DROP);
					if (pf_translate_icmp_af(nk->af,
					    &pd->hdr.icmp))
						return (PF_DROP);
					m_copyback(pd->m, pd->off,
					    sizeof(struct icmp6_hdr),
					    (c_caddr_t)&pd->hdr.icmp6);
					if (pf_change_icmp_af(pd->m, ipoff2, pd,
					    &pd2, &nk->addr[sidx],
					    &nk->addr[didx], pd->af,
					    nk->af))
						return (PF_DROP);
					pd->proto = IPPROTO_ICMP;
					if (pf_translate_icmp_af(nk->af, iih))
						return (PF_DROP);
					if (virtual_type ==
					    htons(ICMP6_ECHO_REQUEST) &&
					    nk->port[iidx] != iih->icmp6_id)
						iih->icmp6_id = nk->port[iidx];
					m_copyback(pd2.m, pd2.off,
					    sizeof(struct icmp6_hdr), (c_caddr_t)iih);
					pf_addrcpy(&pd->nsaddr,
					    &nk->addr[pd2.sidx], nk->af);
					pf_addrcpy(&pd->ndaddr,
					    &nk->addr[pd2.didx], nk->af);
					pd->naf = nk->af;
					return (PF_AFRT);
				}
#endif /* INET && INET6 */

				if (PF_ANEQ(pd2.src,
				    &nk->addr[pd2.sidx], pd2.af) ||
				    ((virtual_type == htons(ICMP6_ECHO_REQUEST)) &&
				    nk->port[pd2.sidx] != iih->icmp6_id))
					pf_change_icmp(pd2.src,
					    (virtual_type == htons(ICMP6_ECHO_REQUEST))
					    ? &iih->icmp6_id : NULL,
					    daddr, &nk->addr[pd2.sidx],
					    (virtual_type == htons(ICMP6_ECHO_REQUEST))
					    ? nk->port[iidx] : 0, NULL,
					    pd2.ip_sum, icmpsum,
					    pd->ip_sum, 0, AF_INET6);

				if (PF_ANEQ(pd2.dst,
				    &nk->addr[pd2.didx], pd2.af))
					pf_change_icmp(pd2.dst, NULL, NULL,
					    &nk->addr[pd2.didx], 0, NULL,
					    pd2.ip_sum, icmpsum,
					    pd->ip_sum, 0, AF_INET6);

				m_copyback(pd->m, pd->off, sizeof(struct icmp6_hdr),
				    (caddr_t)&pd->hdr.icmp6);
				m_copyback(pd->m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
				m_copyback(pd->m, pd2.off, sizeof(struct icmp6_hdr),
				    (caddr_t)iih);
			}
			return (PF_PASS);
			break;
		}
#endif /* INET6 */
		default: {
			int	action;

			key.af = pd2.af;
			key.proto = pd2.proto;
			pf_addrcpy(&key.addr[pd2.sidx], pd2.src, key.af);
			pf_addrcpy(&key.addr[pd2.didx], pd2.dst, key.af);
			key.port[0] = key.port[1] = 0;

			action = pf_find_state(&pd2, &key, state);
			if (action != PF_MATCH)
				return (action);

			/* translate source/destination address, if necessary */
			if ((*state)->key[PF_SK_WIRE] !=
			    (*state)->key[PF_SK_STACK]) {
				struct pf_state_key *nk =
				    (*state)->key[pd->didx];

				if (PF_ANEQ(pd2.src,
				    &nk->addr[pd2.sidx], pd2.af))
					pf_change_icmp(pd2.src, NULL, daddr,
					    &nk->addr[pd2.sidx], 0, NULL,
					    pd2.ip_sum, icmpsum,
					    pd->ip_sum, 0, pd2.af);

				if (PF_ANEQ(pd2.dst,
				    &nk->addr[pd2.didx], pd2.af))
					pf_change_icmp(pd2.dst, NULL, saddr,
					    &nk->addr[pd2.didx], 0, NULL,
					    pd2.ip_sum, icmpsum,
					    pd->ip_sum, 0, pd2.af);

				switch (pd2.af) {
#ifdef INET
				case AF_INET:
					m_copyback(pd->m, pd->off, ICMP_MINLEN,
					    (caddr_t)&pd->hdr.icmp);
					m_copyback(pd->m, ipoff2, sizeof(h2), (caddr_t)&h2);
					break;
#endif /* INET */
#ifdef INET6
				case AF_INET6:
					m_copyback(pd->m, pd->off,
					    sizeof(struct icmp6_hdr),
					    (caddr_t )&pd->hdr.icmp6);
					m_copyback(pd->m, ipoff2, sizeof(h2_6),
					    (caddr_t )&h2_6);
					break;
#endif /* INET6 */
				}
			}
			return (PF_PASS);
			break;
		}
		}
	}
}

/*
 * ipoff and off are measured from the start of the mbuf chain.
 * h must be at "ipoff" on the mbuf chain.
 */
void *
pf_pull_hdr(const struct mbuf *m, int off, void *p, int len,
    u_short *actionp, u_short *reasonp, sa_family_t af)
{
	int iplen = 0;
	switch (af) {
#ifdef INET
	case AF_INET: {
		const struct ip	*h = mtod(m, struct ip *);
		u_int16_t	 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;

		if (fragoff) {
			if (fragoff >= len)
				ACTION_SET(actionp, PF_PASS);
			else {
				ACTION_SET(actionp, PF_DROP);
				REASON_SET(reasonp, PFRES_FRAG);
			}
			return (NULL);
		}
		iplen = ntohs(h->ip_len);
		break;
	}
#endif /* INET */
#ifdef INET6
	case AF_INET6: {
		const struct ip6_hdr	*h = mtod(m, struct ip6_hdr *);

		iplen = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
		break;
	}
#endif /* INET6 */
	}
	if (m->m_pkthdr.len < off + len || iplen < off + len) {
		ACTION_SET(actionp, PF_DROP);
		REASON_SET(reasonp, PFRES_SHORT);
		return (NULL);
	}
	m_copydata(m, off, len, p);
	return (p);
}

int
pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif,
    int rtableid)
{
	struct ifnet		*ifp;

	/*
	 * Skip check for addresses with embedded interface scope,
	 * as they would always match anyway.
	 */
	if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
		return (1);

	if (af != AF_INET && af != AF_INET6)
		return (0);

	if (kif == V_pfi_all)
		return (1);

	/* Skip checks for ipsec interfaces */
	if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
		return (1);

	ifp = (kif != NULL) ? kif->pfik_ifp : NULL;

	switch (af) {
#ifdef INET6
	case AF_INET6:
		return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE,
		    ifp));
#endif /* INET6 */
#ifdef INET
	case AF_INET:
		return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE,
		    ifp));
#endif /* INET */
	}

	return (0);
}

#ifdef INET
static int
pf_route(struct pf_krule *r, struct ifnet *oifp,
    struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp)
{
	struct mbuf		*m0, *m1, *md;
	struct route_in6	 ro;
	union sockaddr_union	 rt_gw;
	const union sockaddr_union	*gw = (const union sockaddr_union *)&ro.ro_dst;
	union sockaddr_union	*dst;
	struct ip		*ip;
	struct ifnet		*ifp = NULL;
	int			 error = 0;
	uint16_t		 ip_len, ip_off;
	uint16_t		 tmp;
	int			 r_dir;
	bool			 skip_test = false;
	int			 action = PF_PASS;

	KASSERT(pd->m && r && oifp, ("%s: invalid parameters", __func__));

	SDT_PROBE4(pf, ip, route_to, entry, pd->m, pd, s, oifp);

	if (s) {
		r_dir = s->direction;
	} else {
		r_dir = r->direction;
	}

	KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT ||
	    r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction",
	    __func__));

	if ((pd->pf_mtag == NULL &&
	    ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) ||
	    pd->pf_mtag->routed++ > 3) {
		m0 = pd->m;
		pd->m = NULL;
		SDT_PROBE1(pf, ip, route_to, drop, __LINE__);
		action = PF_DROP;
		goto bad_locked;
	}

	if (pd->act.rt_kif != NULL)
		ifp = pd->act.rt_kif->pfik_ifp;

	if (pd->act.rt == PF_DUPTO) {
		if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) {
			if (s != NULL) {
				PF_STATE_UNLOCK(s);
			}
			if (ifp == oifp) {
				/* When the 2nd interface is not skipped */
				return (action);
			} else {
				m0 = pd->m;
				pd->m = NULL;
				SDT_PROBE1(pf, ip, route_to, drop, __LINE__);
				action = PF_DROP;
				goto bad;
			}
		} else {
			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED;
			if (((m0 = m_dup(pd->m, M_NOWAIT)) == NULL)) {
				if (s)
					PF_STATE_UNLOCK(s);
				return (action);
			}
		}
	} else {
		if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) {
			if (pd->af == pd->naf) {
				pf_dummynet(pd, s, r, &pd->m);
				if (s)
					PF_STATE_UNLOCK(s);
				return (action);
			} else {
				if (r_dir == PF_IN) {
					skip_test = true;
				}
			}
		}

		/*
		 * If we're actually doing route-to and af-to and are in the
		 * reply direction.
		 */
		if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp &&
		    pd->af != pd->naf) {
			if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET) {
				/* Un-set ifp so we do a plain route lookup. */
				ifp = NULL;
			}
			if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET6) {
				/* Un-set ifp so we do a plain route lookup. */
				ifp = NULL;
			}
		}
		m0 = pd->m;
	}

	ip = mtod(m0, struct ip *);

	bzero(&ro, sizeof(ro));
	dst = (union sockaddr_union *)&ro.ro_dst;
	dst->sin.sin_family = AF_INET;
	dst->sin.sin_len = sizeof(struct sockaddr_in);
	dst->sin.sin_addr = ip->ip_dst;
	if (ifp) { /* Only needed in forward direction and route-to */
		bzero(&rt_gw, sizeof(rt_gw));
		ro.ro_flags |= RT_HAS_GW;
		gw = &rt_gw;
		switch (pd->act.rt_af) {
#ifdef INET
		case AF_INET:
			rt_gw.sin.sin_family = AF_INET;
			rt_gw.sin.sin_len = sizeof(struct sockaddr_in);
			rt_gw.sin.sin_addr.s_addr = pd->act.rt_addr.v4.s_addr;
			break;
#endif /* INET */
#ifdef INET6
		case AF_INET6:
			rt_gw.sin6.sin6_family = AF_INET6;
			rt_gw.sin6.sin6_len = sizeof(struct sockaddr_in6);
			pf_addrcpy((struct pf_addr *)&rt_gw.sin6.sin6_addr,
			    &pd->act.rt_addr, AF_INET6);
			break;
#endif /* INET6 */
		default:
			/* Normal af-to without route-to */
			break;
		}
	}

	if (pd->dir == PF_IN) {
		if (ip->ip_ttl <= IPTTLDEC) {
			if (r->rt != PF_DUPTO)
				pf_send_icmp(m0, ICMP_TIMXCEED,
				    ICMP_TIMXCEED_INTRANS, 0, pd->af, r,
				    pd->act.rtableid);
			action = PF_DROP;
			goto bad_locked;
		}
		ip->ip_ttl -= IPTTLDEC;
	}

	if (s != NULL) {
		if (ifp == NULL && (pd->af != pd->naf)) {
			/* We're in the AFTO case. Do a route lookup. */
			const struct nhop_object *nh;
			nh = fib4_lookup(M_GETFIB(m0), ip->ip_dst, 0, NHR_NONE, 0);
			if (nh) {
				ifp = nh->nh_ifp;

				/* Use the gateway if needed. */
				if (nh->nh_flags & NHF_GATEWAY) {
					gw = (const union sockaddr_union *)&nh->gw_sa;
					ro.ro_flags |= RT_HAS_GW;
				} else {
					dst->sin.sin_addr = ip->ip_dst;
				}

				/*
				 * Bind to the correct interface if we're
				 * if-bound. We don't know which interface
				 * that will be until here, so we've inserted
				 * the state on V_pf_all. Fix that now.
				 */
				if (s->kif == V_pfi_all && ifp != NULL &&
				    r->rule_flag & PFRULE_IFBOUND)
					s->kif = ifp->if_pf_kif;
			}
		}

		if (r->rule_flag & PFRULE_IFBOUND &&
		    pd->act.rt == PF_REPLYTO &&
		    s->kif == V_pfi_all) {
			s->kif = pd->act.rt_kif;
			s->orig_kif = oifp->if_pf_kif;
		}

		PF_STATE_UNLOCK(s);
	}

	/* It must have been either set from rt_af or from fib4_lookup */
	KASSERT(gw->sin.sin_family != 0, ("%s: gw address family undetermined", __func__));

	if (ifp == NULL) {
		m0 = pd->m;
		pd->m = NULL;
		action = PF_DROP;
		SDT_PROBE1(pf, ip, route_to, drop, __LINE__);
		goto bad;
	}

	if (r->rt == PF_DUPTO)
		skip_test = true;

	if (pd->dir == PF_IN && !skip_test) {
		if (pf_test(AF_INET, PF_OUT, PFIL_FWD, ifp, &m0, inp,
		    &pd->act) != PF_PASS) {
			action = PF_DROP;
			SDT_PROBE1(pf, ip, route_to, drop, __LINE__);
			goto bad;
		} else if (m0 == NULL) {
			action = PF_DROP;
			SDT_PROBE1(pf, ip, route_to, drop, __LINE__);
			goto done;
		}
		if (m0->m_len < sizeof(struct ip)) {
			DPFPRINTF(PF_DEBUG_URGENT,
			    "%s: m0->m_len < sizeof(struct ip)", __func__);
			SDT_PROBE1(pf, ip, route_to, drop, __LINE__);
			action = PF_DROP;
			goto bad;
		}
		ip = mtod(m0, struct ip *);
	}

	if (ifp->if_flags & IFF_LOOPBACK)
		m0->m_flags |= M_SKIP_FIREWALL;

	ip_len = ntohs(ip->ip_len);
	ip_off = ntohs(ip->ip_off);

	/* Copied from FreeBSD 10.0-CURRENT ip_output. */
	m0->m_pkthdr.csum_flags |= CSUM_IP;
	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
		in_delayed_cksum(m0);
		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
	}
	if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
		pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2));
		m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
	}

	if (pd->dir == PF_IN) {
		/*
		 * Make sure dummynet gets the correct direction, in case it needs to
		 * re-inject later.
		 */
		pd->dir = PF_OUT;

		/*
		 * The following processing is actually the rest of the inbound processing, even
		 * though we've marked it as outbound (so we don't look through dummynet) and it
		 * happens after the outbound processing (pf_test(PF_OUT) above).
		 * Swap the dummynet pipe numbers, because it's going to come to the wrong
		 * conclusion about what direction it's processing, and we can't fix it or it
		 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect
		 * decision will pick the right pipe, and everything will mostly work as expected.
		 */
		tmp = pd->act.dnrpipe;
		pd->act.dnrpipe = pd->act.dnpipe;
		pd->act.dnpipe = tmp;
	}

	/*
	 * If small enough for interface, or the interface will take
	 * care of the fragmentation for us, we can just send directly.
	 */
	if (ip_len <= ifp->if_mtu ||
	    (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
		ip->ip_sum = 0;
		if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
			ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
			m0->m_pkthdr.csum_flags &= ~CSUM_IP;
		}
		m_clrprotoflags(m0);	/* Avoid confusing lower layers. */

		md = m0;
		error = pf_dummynet_route(pd, s, r, ifp,
		    (const struct sockaddr *)gw, &md);
		if (md != NULL) {
			error = (*ifp->if_output)(ifp, md,
			    (const struct sockaddr *)gw, (struct route *)&ro);
			SDT_PROBE2(pf, ip, route_to, output, ifp, error);
		}
		goto done;
	}

	/* Balk when DF bit is set or the interface didn't support TSO. */
	if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
		error = EMSGSIZE;
		KMOD_IPSTAT_INC(ips_cantfrag);
		if (pd->act.rt != PF_DUPTO) {
			if (s && s->nat_rule != NULL) {
				MPASS(m0 == pd->m);
				PACKET_UNDO_NAT(pd,
				    (ip->ip_hl << 2) + (ip_off & IP_OFFMASK),
				    s);
			}

			pf_send_icmp(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
			   ifp->if_mtu, pd->af, r, pd->act.rtableid);
		}
		SDT_PROBE1(pf, ip, route_to, drop, __LINE__);
		action = PF_DROP;
		goto bad;
	}

	error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
	if (error) {
		SDT_PROBE1(pf, ip, route_to, drop, __LINE__);
		action = PF_DROP;
		goto bad;
	}

	for (; m0; m0 = m1) {
		m1 = m0->m_nextpkt;
		m0->m_nextpkt = NULL;
		if (error == 0) {
			m_clrprotoflags(m0);
			md = m0;
			pd->pf_mtag = pf_find_mtag(md);
			error = pf_dummynet_route(pd, s, r, ifp,
			    (const struct sockaddr *)gw, &md);
			if (md != NULL) {
				error = (*ifp->if_output)(ifp, md,
				    (const struct sockaddr *)gw,
				    (struct route *)&ro);
				SDT_PROBE2(pf, ip, route_to, output, ifp, error);
			}
		} else
			m_freem(m0);
	}

	if (error == 0)
		KMOD_IPSTAT_INC(ips_fragmented);

done:
	if (pd->act.rt != PF_DUPTO)
		pd->m = NULL;
	else
		action = PF_PASS;
	return (action);

bad_locked:
	if (s)
		PF_STATE_UNLOCK(s);
bad:
	m_freem(m0);
	goto done;
}
#endif /* INET */

#ifdef INET6
static int
pf_route6(struct pf_krule *r, struct ifnet *oifp,
    struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp)
{
	struct mbuf		*m0, *md;
	struct m_tag		*mtag;
	struct sockaddr_in6	dst;
	struct ip6_hdr		*ip6;
	struct ifnet		*ifp = NULL;
	int			 r_dir;
	bool			 skip_test = false;
	int			 action = PF_PASS;

	KASSERT(pd->m && r && oifp, ("%s: invalid parameters", __func__));

	SDT_PROBE4(pf, ip6, route_to, entry, pd->m, pd, s, oifp);

	if (s) {
		r_dir = s->direction;
	} else {
		r_dir = r->direction;
	}

	KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT ||
	    r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction",
	    __func__));

	if ((pd->pf_mtag == NULL &&
	    ((pd->pf_mtag = pf_get_mtag(pd->m)) == NULL)) ||
	    pd->pf_mtag->routed++ > 3) {
		m0 = pd->m;
		pd->m = NULL;
		action = PF_DROP;
		SDT_PROBE1(pf, ip6, route_to, drop, __LINE__);
		goto bad_locked;
	}

	if (pd->act.rt_kif != NULL)
		ifp = pd->act.rt_kif->pfik_ifp;

	if (pd->act.rt == PF_DUPTO) {
		if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) {
			if (s != NULL) {
				PF_STATE_UNLOCK(s);
			}
			if (ifp == oifp) {
				/* When the 2nd interface is not skipped */
				return (action);
			} else {
				m0 = pd->m;
				pd->m = NULL;
				action = PF_DROP;
				SDT_PROBE1(pf, ip6, route_to, drop, __LINE__);
				goto bad;
			}
		} else {
			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED;
			if (((m0 = m_dup(pd->m, M_NOWAIT)) == NULL)) {
				if (s)
					PF_STATE_UNLOCK(s);
				return (action);
			}
		}
	} else {
		if ((pd->act.rt == PF_REPLYTO) == (r_dir == pd->dir)) {
			if (pd->af == pd->naf) {
				pf_dummynet(pd, s, r, &pd->m);
				if (s)
					PF_STATE_UNLOCK(s);
				return (action);
			} else {
				if (r_dir == PF_IN) {
					skip_test = true;
				}
			}
		}

		/*
		 * If we're actually doing route-to and af-to and are in the
		 * reply direction.
		 */
		if (pd->act.rt_kif && pd->act.rt_kif->pfik_ifp &&
		    pd->af != pd->naf) {
			if (pd->act.rt == PF_ROUTETO && r->naf != AF_INET6) {
				/* Un-set ifp so we do a plain route lookup. */
				ifp = NULL;
			}
			if (pd->act.rt == PF_REPLYTO && r->naf != AF_INET) {
				/* Un-set ifp so we do a plain route lookup. */
				ifp = NULL;
			}
		}
		m0 = pd->m;
	}

	ip6 = mtod(m0, struct ip6_hdr *);

	bzero(&dst, sizeof(dst));
	dst.sin6_family = AF_INET6;
	dst.sin6_len = sizeof(dst);
	pf_addrcpy((struct pf_addr *)&dst.sin6_addr, &pd->act.rt_addr,
	    AF_INET6);

	if (pd->dir == PF_IN) {
		if (ip6->ip6_hlim <= IPV6_HLIMDEC) {
			if (r->rt != PF_DUPTO)
				pf_send_icmp(m0, ICMP6_TIME_EXCEEDED,
				    ICMP6_TIME_EXCEED_TRANSIT, 0, pd->af, r,
				    pd->act.rtableid);
			action = PF_DROP;
			goto bad_locked;
		}
		ip6->ip6_hlim -= IPV6_HLIMDEC;
	}

	if (s != NULL) {
		if (ifp == NULL && (pd->af != pd->naf)) {
			const struct nhop_object *nh;
			nh = fib6_lookup(M_GETFIB(m0), &ip6->ip6_dst, 0, NHR_NONE, 0);
			if (nh) {
				ifp = nh->nh_ifp;

				/* Use the gateway if needed. */
				if (nh->nh_flags & NHF_GATEWAY)
					bcopy(&nh->gw6_sa.sin6_addr, &dst.sin6_addr,
					    sizeof(dst.sin6_addr));
				else
					dst.sin6_addr = ip6->ip6_dst;

				/*
				 * Bind to the correct interface if we're
				 * if-bound. We don't know which interface
				 * that will be until here, so we've inserted
				 * the state on V_pf_all. Fix that now.
				 */
				if (s->kif == V_pfi_all && ifp != NULL &&
				    r->rule_flag & PFRULE_IFBOUND)
					s->kif = ifp->if_pf_kif;
			}
		}

		if (r->rule_flag & PFRULE_IFBOUND &&
		    pd->act.rt == PF_REPLYTO &&
		    s->kif == V_pfi_all) {
			s->kif = pd->act.rt_kif;
			s->orig_kif = oifp->if_pf_kif;
		}

		PF_STATE_UNLOCK(s);
	}

	if (pd->af != pd->naf) {
		struct udphdr *uh = &pd->hdr.udp;

		if (pd->proto == IPPROTO_UDP && uh->uh_sum == 0) {
			uh->uh_sum = in6_cksum_pseudo(ip6,
			    ntohs(uh->uh_ulen), IPPROTO_UDP, 0);
			m_copyback(m0, pd->off, sizeof(*uh), pd->hdr.any);
		}
	}

	if (ifp == NULL) {
		m0 = pd->m;
		pd->m = NULL;
		action = PF_DROP;
		SDT_PROBE1(pf, ip6, route_to, drop, __LINE__);
		goto bad;
	}

	if (r->rt == PF_DUPTO)
		skip_test = true;

	if (pd->dir == PF_IN && !skip_test) {
		if (pf_test(AF_INET6, PF_OUT, PFIL_FWD | PF_PFIL_NOREFRAGMENT,
		    ifp, &m0, inp, &pd->act) != PF_PASS) {
			action = PF_DROP;
			SDT_PROBE1(pf, ip6, route_to, drop, __LINE__);
			goto bad;
		} else if (m0 == NULL) {
			action = PF_DROP;
			SDT_PROBE1(pf, ip6, route_to, drop, __LINE__);
			goto done;
		}
		if (m0->m_len < sizeof(struct ip6_hdr)) {
			DPFPRINTF(PF_DEBUG_URGENT,
			    "%s: m0->m_len < sizeof(struct ip6_hdr)",
			    __func__);
			action = PF_DROP;
			SDT_PROBE1(pf, ip6, route_to, drop, __LINE__);
			goto bad;
		}
		ip6 = mtod(m0, struct ip6_hdr *);
	}

	if (ifp->if_flags & IFF_LOOPBACK)
		m0->m_flags |= M_SKIP_FIREWALL;

	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
	    ~ifp->if_hwassist) {
		uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
		in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
	}

	if (pd->dir == PF_IN) {
		uint16_t	 tmp;
		/*
		 * Make sure dummynet gets the correct direction, in case it needs to
		 * re-inject later.
		 */
		pd->dir = PF_OUT;

		/*
		 * The following processing is actually the rest of the inbound processing, even
		 * though we've marked it as outbound (so we don't look through dummynet) and it
		 * happens after the outbound processing (pf_test(PF_OUT) above).
		 * Swap the dummynet pipe numbers, because it's going to come to the wrong
		 * conclusion about what direction it's processing, and we can't fix it or it
		 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect
		 * decision will pick the right pipe, and everything will mostly work as expected.
		 */
		tmp = pd->act.dnrpipe;
		pd->act.dnrpipe = pd->act.dnpipe;
		pd->act.dnpipe = tmp;
	}

	/*
	 * If the packet is too large for the outgoing interface,
	 * send back an icmp6 error.
	 */
	if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
		dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
	mtag = m_tag_find(m0, PACKET_TAG_PF_REASSEMBLED, NULL);
	if (mtag != NULL) {
		int ret __sdt_used;
		ret = pf_refragment6(ifp, &m0, mtag, ifp, true);
		SDT_PROBE2(pf, ip6, route_to, output, ifp, ret);
		goto done;
	}

	if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) {
		md = m0;
		pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md);
		if (md != NULL) {
			int ret __sdt_used;
			ret = nd6_output_ifp(ifp, ifp, md, &dst, NULL);
			SDT_PROBE2(pf, ip6, route_to, output, ifp, ret);
		}
	}
	else {
		in6_ifstat_inc(ifp, ifs6_in_toobig);
		if (pd->act.rt != PF_DUPTO) {
			if (s && s->nat_rule != NULL) {
				MPASS(m0 == pd->m);
				PACKET_UNDO_NAT(pd,
				    ((caddr_t)ip6 - m0->m_data) +
				    sizeof(struct ip6_hdr), s);
			}

			if (r->rt != PF_DUPTO)
				pf_send_icmp(m0, ICMP6_PACKET_TOO_BIG, 0,
				    ifp->if_mtu, pd->af, r, pd->act.rtableid);
		}
		action = PF_DROP;
		SDT_PROBE1(pf, ip6, route_to, drop, __LINE__);
		goto bad;
	}

done:
	if (pd->act.rt != PF_DUPTO)
		pd->m = NULL;
	else
		action = PF_PASS;
	return (action);

bad_locked:
	if (s)
		PF_STATE_UNLOCK(s);
bad:
	m_freem(m0);
	goto done;
}
#endif /* INET6 */

/*
 * FreeBSD supports cksum offloads for the following drivers.
 *  em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4)
 *
 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
 *  network driver performed cksum including pseudo header, need to verify
 *   csum_data
 * CSUM_DATA_VALID :
 *  network driver performed cksum, needs to additional pseudo header
 *  cksum computation with partial csum_data(i.e. lack of H/W support for
 *  pseudo header, for instance sk(4) and possibly gem(4))
 *
 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
 * TCP/UDP layer.
 * Also, set csum_data to 0xffff to force cksum validation.
 */
static int
pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
{
	u_int16_t sum = 0;
	int hw_assist = 0;
	struct ip *ip;

	if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
		return (1);
	if (m->m_pkthdr.len < off + len)
		return (1);

	switch (p) {
	case IPPROTO_TCP:
		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
				sum = m->m_pkthdr.csum_data;
			} else {
				ip = mtod(m, struct ip *);
				sum = in_pseudo(ip->ip_src.s_addr,
				ip->ip_dst.s_addr, htonl((u_short)len +
				m->m_pkthdr.csum_data + IPPROTO_TCP));
			}
			sum ^= 0xffff;
			++hw_assist;
		}
		break;
	case IPPROTO_UDP:
		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
				sum = m->m_pkthdr.csum_data;
			} else {
				ip = mtod(m, struct ip *);
				sum = in_pseudo(ip->ip_src.s_addr,
				ip->ip_dst.s_addr, htonl((u_short)len +
				m->m_pkthdr.csum_data + IPPROTO_UDP));
			}
			sum ^= 0xffff;
			++hw_assist;
		}
		break;
	case IPPROTO_ICMP:
#ifdef INET6
	case IPPROTO_ICMPV6:
#endif /* INET6 */
		break;
	default:
		return (1);
	}

	if (!hw_assist) {
		switch (af) {
		case AF_INET:
			if (m->m_len < sizeof(struct ip))
				return (1);
			sum = in4_cksum(m, (p == IPPROTO_ICMP ? 0 : p), off, len);
			break;
#ifdef INET6
		case AF_INET6:
			if (m->m_len < sizeof(struct ip6_hdr))
				return (1);
			sum = in6_cksum(m, p, off, len);
			break;
#endif /* INET6 */
		}
	}
	if (sum) {
		switch (p) {
		case IPPROTO_TCP:
		    {
			KMOD_TCPSTAT_INC(tcps_rcvbadsum);
			break;
		    }
		case IPPROTO_UDP:
		    {
			KMOD_UDPSTAT_INC(udps_badsum);
			break;
		    }
#ifdef INET
		case IPPROTO_ICMP:
		    {
			KMOD_ICMPSTAT_INC(icps_checksum);
			break;
		    }
#endif
#ifdef INET6
		case IPPROTO_ICMPV6:
		    {
			KMOD_ICMP6STAT_INC(icp6s_checksum);
			break;
		    }
#endif /* INET6 */
		}
		return (1);
	} else {
		if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
			m->m_pkthdr.csum_flags |=
			    (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
			m->m_pkthdr.csum_data = 0xffff;
		}
	}
	return (0);
}

static bool
pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r,
    const struct pf_kstate *s, struct ip_fw_args *dnflow)
{
	int dndir = r->direction;

	if (s && dndir == PF_INOUT) {
		dndir = s->direction;
	} else if (dndir == PF_INOUT) {
		/* Assume primary direction. Happens when we've set dnpipe in
		 * the ethernet level code. */
		dndir = pd->dir;
	}

	if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED)
		return (false);

	memset(dnflow, 0, sizeof(*dnflow));

	if (pd->dport != NULL)
		dnflow->f_id.dst_port = ntohs(*pd->dport);
	if (pd->sport != NULL)
		dnflow->f_id.src_port = ntohs(*pd->sport);

	if (pd->dir == PF_IN)
		dnflow->flags |= IPFW_ARGS_IN;
	else
		dnflow->flags |= IPFW_ARGS_OUT;

	if (pd->dir != dndir && pd->act.dnrpipe) {
		dnflow->rule.info = pd->act.dnrpipe;
	}
	else if (pd->dir == dndir && pd->act.dnpipe) {
		dnflow->rule.info = pd->act.dnpipe;
	}
	else {
		return (false);
	}

	dnflow->rule.info |= IPFW_IS_DUMMYNET;
	if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE)
		dnflow->rule.info |= IPFW_IS_PIPE;

	dnflow->f_id.proto = pd->proto;
	dnflow->f_id.extra = dnflow->rule.info;
	switch (pd->naf) {
	case AF_INET:
		dnflow->f_id.addr_type = 4;
		dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr);
		dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr);
		break;
	case AF_INET6:
		dnflow->flags |= IPFW_ARGS_IP6;
		dnflow->f_id.addr_type = 6;
		dnflow->f_id.src_ip6 = pd->src->v6;
		dnflow->f_id.dst_ip6 = pd->dst->v6;
		break;
	}

	return (true);
}

int
pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0,
    struct inpcb *inp)
{
	struct pfi_kkif		*kif;
	struct mbuf		*m = *m0;

	M_ASSERTPKTHDR(m);
	MPASS(ifp->if_vnet == curvnet);
	NET_EPOCH_ASSERT();

	if (!V_pf_status.running)
		return (PF_PASS);

	kif = (struct pfi_kkif *)ifp->if_pf_kif;

	if (kif == NULL) {
		DPFPRINTF(PF_DEBUG_URGENT,
		    "%s: kif == NULL, if_xname %s", __func__, ifp->if_xname);
		return (PF_DROP);
	}
	if (kif->pfik_flags & PFI_IFLAG_SKIP)
		return (PF_PASS);

	if (m->m_flags & M_SKIP_FIREWALL)
		return (PF_PASS);

	if (__predict_false(! M_WRITABLE(*m0))) {
		m = *m0 = m_unshare(*m0, M_NOWAIT);
		if (*m0 == NULL)
			return (PF_DROP);
	}

	/* Stateless! */
	return (pf_test_eth_rule(dir, kif, m0));
}

static __inline void
pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag)
{
	struct m_tag *mtag;

	pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET;

	/* dummynet adds this tag, but pf does not need it,
	 * and keeping it creates unexpected behavior,
	 * e.g. in case of divert(4) usage right after dummynet. */
	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
	if (mtag != NULL)
		m_tag_delete(m, mtag);
}

static int
pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s,
    struct pf_krule *r, struct mbuf **m0)
{
	return (pf_dummynet_route(pd, s, r, NULL, NULL, m0));
}

static int
pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s,
    struct pf_krule *r, struct ifnet *ifp, const struct sockaddr *sa,
    struct mbuf **m0)
{
	struct ip_fw_args dnflow;

	NET_EPOCH_ASSERT();

	if (pd->act.dnpipe == 0 && pd->act.dnrpipe == 0)
		return (0);

	if (ip_dn_io_ptr == NULL) {
		m_freem(*m0);
		*m0 = NULL;
		return (ENOMEM);
	}

	if (pd->pf_mtag == NULL &&
	    ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) {
		m_freem(*m0);
		*m0 = NULL;
		return (ENOMEM);
	}

	if (ifp != NULL) {
		pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO;

		pd->pf_mtag->if_index = ifp->if_index;
		pd->pf_mtag->if_idxgen = ifp->if_idxgen;

		MPASS(sa != NULL);

		switch (sa->sa_family) {
		case AF_INET:
			memcpy(&pd->pf_mtag->dst, sa,
			    sizeof(struct sockaddr_in));
			break;
		case AF_INET6:
			memcpy(&pd->pf_mtag->dst, sa,
			    sizeof(struct sockaddr_in6));
			break;
		}
	}

	if (s != NULL && s->nat_rule != NULL &&
	    s->nat_rule->action == PF_RDR &&
	    (
#ifdef INET
	    (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) ||
#endif /* INET */
	    (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) {
		/*
		 * If we're redirecting to loopback mark this packet
		 * as being local. Otherwise it might get dropped
		 * if dummynet re-injects.
		 */
		(*m0)->m_pkthdr.rcvif = V_loif;
	}

	if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) {
		pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET;
		pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED;
		ip_dn_io_ptr(m0, &dnflow);
		if (*m0 != NULL) {
			pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
			pf_dummynet_flag_remove(*m0, pd->pf_mtag);
		}
	}

	return (0);
}

static int
pf_walk_option(struct pf_pdesc *pd, struct ip *h, int off, int end,
    u_short *reason)
{
	uint8_t type, length, opts[15 * 4 - sizeof(struct ip)];

	/* IP header in payload of ICMP packet may be too short */
	if (pd->m->m_pkthdr.len < end) {
		DPFPRINTF(PF_DEBUG_MISC, "IP option too short");
		REASON_SET(reason, PFRES_SHORT);
		return (PF_DROP);
	}

	MPASS(end - off <= sizeof(opts));
	m_copydata(pd->m, off, end - off, opts);
	end -= off;
	off = 0;

	while (off < end) {
		type = opts[off];
		if (type == IPOPT_EOL)
			break;
		if (type == IPOPT_NOP) {
			off++;
			continue;
		}
		if (off + 2 > end) {
			DPFPRINTF(PF_DEBUG_MISC, "IP length opt");
			REASON_SET(reason, PFRES_IPOPTIONS);
			return (PF_DROP);
		}
		length = opts[off + 1];
		if (length < 2) {
			DPFPRINTF(PF_DEBUG_MISC, "IP short opt");
			REASON_SET(reason, PFRES_IPOPTIONS);
			return (PF_DROP);
		}
		if (off + length > end) {
			DPFPRINTF(PF_DEBUG_MISC, "IP long opt");
			REASON_SET(reason, PFRES_IPOPTIONS);
			return (PF_DROP);
		}
		switch (type) {
		case IPOPT_RA:
			pd->badopts |= PF_OPT_ROUTER_ALERT;
			break;
		default:
			pd->badopts |= PF_OPT_OTHER;
			break;
		}
		off += length;
	}

	return (PF_PASS);
}

static int
pf_walk_header(struct pf_pdesc *pd, struct ip *h, u_short *reason)
{
	struct ah	 ext;
	u_int32_t	 hlen, end;
	int		 hdr_cnt;

	hlen = h->ip_hl << 2;
	if (hlen < sizeof(struct ip) || hlen > ntohs(h->ip_len)) {
		REASON_SET(reason, PFRES_SHORT);
		return (PF_DROP);
	}
	if (hlen != sizeof(struct ip)) {
		if (pf_walk_option(pd, h, pd->off + sizeof(struct ip),
		    pd->off + hlen, reason) != PF_PASS)
			return (PF_DROP);
		/* header options which contain only padding is fishy */
		if (pd->badopts == 0)
			pd->badopts |= PF_OPT_OTHER;
	}
	end = pd->off + ntohs(h->ip_len);
	pd->off += hlen;
	pd->proto = h->ip_p;
	/* IGMP packets have router alert options, allow them */
	if (pd->proto == IPPROTO_IGMP) {
		/*
		 * According to RFC 1112 ttl must be set to 1 in all IGMP
		 * packets sent to 224.0.0.1
		 */
		if ((h->ip_ttl != 1) &&
		    (h->ip_dst.s_addr == INADDR_ALLHOSTS_GROUP)) {
			DPFPRINTF(PF_DEBUG_MISC, "Invalid IGMP");
			REASON_SET(reason, PFRES_IPOPTIONS);
			return (PF_DROP);
		}
		pd->badopts &= ~PF_OPT_ROUTER_ALERT;
	}
	/* stop walking over non initial fragments */
	if ((h->ip_off & htons(IP_OFFMASK)) != 0)
		return (PF_PASS);
	for (hdr_cnt = 0; hdr_cnt < PF_HDR_LIMIT; hdr_cnt++) {
		switch (pd->proto) {
		case IPPROTO_AH:
			/* fragments may be short */
			if ((h->ip_off & htons(IP_MF | IP_OFFMASK)) != 0 &&
			    end < pd->off + sizeof(ext))
				return (PF_PASS);
			if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext),
				NULL, reason, AF_INET)) {
				DPFPRINTF(PF_DEBUG_MISC, "IP short exthdr");
				return (PF_DROP);
			}
			pd->off += (ext.ah_len + 2) * 4;
			pd->proto = ext.ah_nxt;
			break;
		default:
			return (PF_PASS);
		}
	}
	DPFPRINTF(PF_DEBUG_MISC, "IPv4 nested authentication header limit");
	REASON_SET(reason, PFRES_IPOPTIONS);
	return (PF_DROP);
}

#ifdef INET6
static int
pf_walk_option6(struct pf_pdesc *pd, struct ip6_hdr *h, int off, int end,
    u_short *reason)
{
	struct ip6_opt		 opt;
	struct ip6_opt_jumbo	 jumbo;

	while (off < end) {
		if (!pf_pull_hdr(pd->m, off, &opt.ip6o_type,
		    sizeof(opt.ip6o_type), NULL, reason, AF_INET6)) {
			DPFPRINTF(PF_DEBUG_MISC, "IPv6 short opt type");
			return (PF_DROP);
		}
		if (opt.ip6o_type == IP6OPT_PAD1) {
			off++;
			continue;
		}
		if (!pf_pull_hdr(pd->m, off, &opt, sizeof(opt), NULL,
		    reason, AF_INET6)) {
			DPFPRINTF(PF_DEBUG_MISC, "IPv6 short opt");
			return (PF_DROP);
		}
		if (off + sizeof(opt) + opt.ip6o_len > end) {
			DPFPRINTF(PF_DEBUG_MISC, "IPv6 long opt");
			REASON_SET(reason, PFRES_IPOPTIONS);
			return (PF_DROP);
		}
		switch (opt.ip6o_type) {
		case IP6OPT_PADN:
			break;
		case IP6OPT_JUMBO:
			pd->badopts |= PF_OPT_JUMBO;
			if (pd->jumbolen != 0) {
				DPFPRINTF(PF_DEBUG_MISC, "IPv6 multiple jumbo");
				REASON_SET(reason, PFRES_IPOPTIONS);
				return (PF_DROP);
			}
			if (ntohs(h->ip6_plen) != 0) {
				DPFPRINTF(PF_DEBUG_MISC, "IPv6 bad jumbo plen");
				REASON_SET(reason, PFRES_IPOPTIONS);
				return (PF_DROP);
			}
			if (!pf_pull_hdr(pd->m, off, &jumbo, sizeof(jumbo), NULL,
				reason, AF_INET6)) {
				DPFPRINTF(PF_DEBUG_MISC, "IPv6 short jumbo");
				return (PF_DROP);
			}
			memcpy(&pd->jumbolen, jumbo.ip6oj_jumbo_len,
			    sizeof(pd->jumbolen));
			pd->jumbolen = ntohl(pd->jumbolen);
			if (pd->jumbolen < IPV6_MAXPACKET) {
				DPFPRINTF(PF_DEBUG_MISC, "IPv6 short jumbolen");
				REASON_SET(reason, PFRES_IPOPTIONS);
				return (PF_DROP);
			}
			break;
		case IP6OPT_ROUTER_ALERT:
			pd->badopts |= PF_OPT_ROUTER_ALERT;
			break;
		default:
			pd->badopts |= PF_OPT_OTHER;
			break;
		}
		off += sizeof(opt) + opt.ip6o_len;
	}

	return (PF_PASS);
}

int
pf_walk_header6(struct pf_pdesc *pd, struct ip6_hdr *h, u_short *reason)
{
	struct ip6_frag		 frag;
	struct ip6_ext		 ext;
	struct icmp6_hdr	 icmp6;
	struct ip6_rthdr	 rthdr;
	uint32_t		 end;
	int			 hdr_cnt, fraghdr_cnt = 0, rthdr_cnt = 0;

	pd->off += sizeof(struct ip6_hdr);
	end = pd->off + ntohs(h->ip6_plen);
	pd->fragoff = pd->extoff = pd->jumbolen = 0;
	pd->proto = h->ip6_nxt;
	for (hdr_cnt = 0; hdr_cnt < PF_HDR_LIMIT; hdr_cnt++) {
		switch (pd->proto) {
		case IPPROTO_ROUTING:
		case IPPROTO_DSTOPTS:
			pd->badopts |= PF_OPT_OTHER;
			break;
		case IPPROTO_HOPOPTS:
			if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext),
			    NULL, reason, AF_INET6)) {
				DPFPRINTF(PF_DEBUG_MISC, "IPv6 short exthdr");
				return (PF_DROP);
			}
			if (pf_walk_option6(pd, h, pd->off + sizeof(ext),
				pd->off + (ext.ip6e_len + 1) * 8,
				reason) != PF_PASS)
				return (PF_DROP);
			/* option header which contains only padding is fishy */
			if (pd->badopts == 0)
				pd->badopts |= PF_OPT_OTHER;
			break;
		}
		switch (pd->proto) {
		case IPPROTO_FRAGMENT:
			if (fraghdr_cnt++) {
				DPFPRINTF(PF_DEBUG_MISC, "IPv6 multiple fragment");
				REASON_SET(reason, PFRES_FRAG);
				return (PF_DROP);
			}
			/* jumbo payload packets cannot be fragmented */
			if (pd->jumbolen != 0) {
				DPFPRINTF(PF_DEBUG_MISC, "IPv6 fragmented jumbo");
				REASON_SET(reason, PFRES_FRAG);
				return (PF_DROP);
			}
			if (!pf_pull_hdr(pd->m, pd->off, &frag, sizeof(frag),
			    NULL, reason, AF_INET6)) {
				DPFPRINTF(PF_DEBUG_MISC, "IPv6 short fragment");
				return (PF_DROP);
			}
			/* stop walking over non initial fragments */
			if (ntohs((frag.ip6f_offlg & IP6F_OFF_MASK)) != 0) {
				pd->fragoff = pd->off;
				return (PF_PASS);
			}
			/* RFC6946:  reassemble only non atomic fragments */
			if (frag.ip6f_offlg & IP6F_MORE_FRAG)
				pd->fragoff = pd->off;
			pd->off += sizeof(frag);
			pd->proto = frag.ip6f_nxt;
			break;
		case IPPROTO_ROUTING:
			if (rthdr_cnt++) {
				DPFPRINTF(PF_DEBUG_MISC, "IPv6 multiple rthdr");
				REASON_SET(reason, PFRES_IPOPTIONS);
				return (PF_DROP);
			}
			/* fragments may be short */
			if (pd->fragoff != 0 && end < pd->off + sizeof(rthdr)) {
				pd->off = pd->fragoff;
				pd->proto = IPPROTO_FRAGMENT;
				return (PF_PASS);
			}
			if (!pf_pull_hdr(pd->m, pd->off, &rthdr, sizeof(rthdr),
			    NULL, reason, AF_INET6)) {
				DPFPRINTF(PF_DEBUG_MISC, "IPv6 short rthdr");
				return (PF_DROP);
			}
			if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
				DPFPRINTF(PF_DEBUG_MISC, "IPv6 rthdr0");
				REASON_SET(reason, PFRES_IPOPTIONS);
				return (PF_DROP);
			}
			/* FALLTHROUGH */
		case IPPROTO_HOPOPTS:
			/* RFC2460 4.1:  Hop-by-Hop only after IPv6 header */
			if (pd->proto == IPPROTO_HOPOPTS && hdr_cnt > 0) {
				DPFPRINTF(PF_DEBUG_MISC, "IPv6 hopopts not first");
				REASON_SET(reason, PFRES_IPOPTIONS);
				return (PF_DROP);
			}
			/* FALLTHROUGH */
		case IPPROTO_AH:
		case IPPROTO_DSTOPTS:
			if (!pf_pull_hdr(pd->m, pd->off, &ext, sizeof(ext),
			    NULL, reason, AF_INET6)) {
				DPFPRINTF(PF_DEBUG_MISC, "IPv6 short exthdr");
				return (PF_DROP);
			}
			/* fragments may be short */
			if (pd->fragoff != 0 && end < pd->off + sizeof(ext)) {
				pd->off = pd->fragoff;
				pd->proto = IPPROTO_FRAGMENT;
				return (PF_PASS);
			}
			/* reassembly needs the ext header before the frag */
			if (pd->fragoff == 0)
				pd->extoff = pd->off;
			if (pd->proto == IPPROTO_HOPOPTS && pd->fragoff == 0 &&
			    ntohs(h->ip6_plen) == 0 && pd->jumbolen != 0) {
				DPFPRINTF(PF_DEBUG_MISC, "IPv6 missing jumbo");
				REASON_SET(reason, PFRES_IPOPTIONS);
				return (PF_DROP);
			}
			if (pd->proto == IPPROTO_AH)
				pd->off += (ext.ip6e_len + 2) * 4;
			else
				pd->off += (ext.ip6e_len + 1) * 8;
			pd->proto = ext.ip6e_nxt;
			break;
		case IPPROTO_ICMPV6:
			/* fragments may be short, ignore inner header then */
			if (pd->fragoff != 0 && end < pd->off + sizeof(icmp6)) {
				pd->off = pd->fragoff;
				pd->proto = IPPROTO_FRAGMENT;
				return (PF_PASS);
			}
			if (!pf_pull_hdr(pd->m, pd->off, &icmp6, sizeof(icmp6),
				NULL, reason, AF_INET6)) {
				DPFPRINTF(PF_DEBUG_MISC,
				    "IPv6 short icmp6hdr");
				return (PF_DROP);
			}
			/* ICMP multicast packets have router alert options */
			switch (icmp6.icmp6_type) {
			case MLD_LISTENER_QUERY:
			case MLD_LISTENER_REPORT:
			case MLD_LISTENER_DONE:
			case MLDV2_LISTENER_REPORT:
				/*
				 * According to RFC 2710 all MLD messages are
				 * sent with hop-limit (ttl) set to 1, and link
				 * local source address.  If either one is
				 * missing then MLD message is invalid and
				 * should be discarded.
				 */
				if ((h->ip6_hlim != 1) ||
				    !IN6_IS_ADDR_LINKLOCAL(&h->ip6_src)) {
					DPFPRINTF(PF_DEBUG_MISC, "Invalid MLD");
					REASON_SET(reason, PFRES_IPOPTIONS);
					return (PF_DROP);
				}
				pd->badopts &= ~PF_OPT_ROUTER_ALERT;
				break;
			}
			return (PF_PASS);
		case IPPROTO_TCP:
		case IPPROTO_UDP:
		case IPPROTO_SCTP:
			/* fragments may be short, ignore inner header then */
			if (pd->fragoff != 0 && end < pd->off +
			    (pd->proto == IPPROTO_TCP ? sizeof(struct tcphdr) :
			    pd->proto == IPPROTO_UDP ? sizeof(struct udphdr) :
			    pd->proto == IPPROTO_SCTP ? sizeof(struct sctphdr) :
			    sizeof(struct icmp6_hdr))) {
				pd->off = pd->fragoff;
				pd->proto = IPPROTO_FRAGMENT;
			}
			/* FALLTHROUGH */
		default:
			return (PF_PASS);
		}
	}
	DPFPRINTF(PF_DEBUG_MISC, "IPv6 nested extension header limit");
	REASON_SET(reason, PFRES_IPOPTIONS);
	return (PF_DROP);
}
#endif /* INET6 */

static void
pf_init_pdesc(struct pf_pdesc *pd, struct mbuf *m)
{
	memset(pd, 0, sizeof(*pd));
	pd->pf_mtag = pf_find_mtag(m);
	pd->m = m;
}

static int
pf_setup_pdesc(sa_family_t af, int dir, struct pf_pdesc *pd, struct mbuf **m0,
    u_short *action, u_short *reason, struct pfi_kkif *kif,
    struct pf_rule_actions *default_actions)
{
	pd->dir = dir;
	pd->kif = kif;
	pd->m = *m0;
	pd->sidx = (dir == PF_IN) ? 0 : 1;
	pd->didx = (dir == PF_IN) ? 1 : 0;
	pd->af = pd->naf = af;

	PF_RULES_ASSERT();

	TAILQ_INIT(&pd->sctp_multihome_jobs);
	if (default_actions != NULL)
		memcpy(&pd->act, default_actions, sizeof(pd->act));

	if (pd->pf_mtag && pd->pf_mtag->dnpipe) {
		pd->act.dnpipe = pd->pf_mtag->dnpipe;
		pd->act.flags = pd->pf_mtag->dnflags;
	}

	switch (af) {
#ifdef INET
	case AF_INET: {
		struct ip *h;

		if (__predict_false((*m0)->m_len < sizeof(struct ip)) &&
		    (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) {
			DPFPRINTF(PF_DEBUG_URGENT,
			    "%s: m_len < sizeof(struct ip), pullup failed",
			    __func__);
			*action = PF_DROP;
			REASON_SET(reason, PFRES_SHORT);
			return (-1);
		}

		h = mtod(pd->m, struct ip *);
		if (pd->m->m_pkthdr.len < ntohs(h->ip_len)) {
			*action = PF_DROP;
			REASON_SET(reason, PFRES_SHORT);
			return (-1);
		}

		if (pf_normalize_ip(reason, pd) != PF_PASS) {
			/* We do IP header normalization and packet reassembly here */
			*m0 = pd->m;
			*action = PF_DROP;
			return (-1);
		}
		*m0 = pd->m;
		h = mtod(pd->m, struct ip *);

		if (pf_walk_header(pd, h, reason) != PF_PASS) {
			*action = PF_DROP;
			return (-1);
		}

		pd->src = (struct pf_addr *)&h->ip_src;
		pd->dst = (struct pf_addr *)&h->ip_dst;
		pf_addrcpy(&pd->osrc, pd->src, af);
		pf_addrcpy(&pd->odst, pd->dst, af);
		pd->ip_sum = &h->ip_sum;
		pd->tos = h->ip_tos & ~IPTOS_ECN_MASK;
		pd->ttl = h->ip_ttl;
		pd->tot_len = ntohs(h->ip_len);
		pd->act.rtableid = -1;
		pd->df = h->ip_off & htons(IP_DF);
		pd->virtual_proto = (h->ip_off & htons(IP_MF | IP_OFFMASK)) ?
		    PF_VPROTO_FRAGMENT : pd->proto;

		break;
	}
#endif /* INET */
#ifdef INET6
	case AF_INET6: {
		struct ip6_hdr *h;

		if (__predict_false((*m0)->m_len < sizeof(struct ip6_hdr)) &&
		    (pd->m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) {
			DPFPRINTF(PF_DEBUG_URGENT,
			    "%s: m_len < sizeof(struct ip6_hdr)"
			     ", pullup failed", __func__);
			*action = PF_DROP;
			REASON_SET(reason, PFRES_SHORT);
			return (-1);
		}

		h = mtod(pd->m, struct ip6_hdr *);
		if (pd->m->m_pkthdr.len <
		    sizeof(struct ip6_hdr) + ntohs(h->ip6_plen)) {
			*action = PF_DROP;
			REASON_SET(reason, PFRES_SHORT);
			return (-1);
		}

		/*
		 * we do not support jumbogram.  if we keep going, zero ip6_plen
		 * will do something bad, so drop the packet for now.
		 */
		if (htons(h->ip6_plen) == 0) {
			*action = PF_DROP;
			return (-1);
		}

		if (pf_walk_header6(pd, h, reason) != PF_PASS) {
			*action = PF_DROP;
			return (-1);
		}

		h = mtod(pd->m, struct ip6_hdr *);
		pd->src = (struct pf_addr *)&h->ip6_src;
		pd->dst = (struct pf_addr *)&h->ip6_dst;
		pf_addrcpy(&pd->osrc, pd->src, af);
		pf_addrcpy(&pd->odst, pd->dst, af);
		pd->ip_sum = NULL;
		pd->tos = IPV6_DSCP(h);
		pd->ttl = h->ip6_hlim;
		pd->tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
		pd->act.rtableid = -1;

		pd->virtual_proto = (pd->fragoff != 0) ?
		    PF_VPROTO_FRAGMENT : pd->proto;

		/* We do IP header normalization and packet reassembly here */
		if (pf_normalize_ip6(pd->fragoff, reason, pd) !=
		    PF_PASS) {
			*m0 = pd->m;
			*action = PF_DROP;
			return (-1);
		}
		*m0 = pd->m;
		if (pd->m == NULL) {
			/* packet sits in reassembly queue, no error */
			*action = PF_PASS;
			return (-1);
		}

		/* Update pointers into the packet. */
		h = mtod(pd->m, struct ip6_hdr *);
		pd->src = (struct pf_addr *)&h->ip6_src;
		pd->dst = (struct pf_addr *)&h->ip6_dst;

		pd->off = 0;

		if (pf_walk_header6(pd, h, reason) != PF_PASS) {
			*action = PF_DROP;
			return (-1);
		}

		if (m_tag_find(pd->m, PACKET_TAG_PF_REASSEMBLED, NULL) != NULL) {
			/*
			 * Reassembly may have changed the next protocol from
			 * fragment to something else, so update.
			 */
			pd->virtual_proto = pd->proto;
			MPASS(pd->fragoff == 0);
		}

		if (pd->fragoff != 0)
			pd->virtual_proto = PF_VPROTO_FRAGMENT;

		break;
	}
#endif /* INET6 */
	default:
		panic("pf_setup_pdesc called with illegal af %u", af);
	}

	switch (pd->virtual_proto) {
	case IPPROTO_TCP: {
		struct tcphdr *th = &pd->hdr.tcp;

		if (!pf_pull_hdr(pd->m, pd->off, th, sizeof(*th), action,
			reason, af)) {
			*action = PF_DROP;
			REASON_SET(reason, PFRES_SHORT);
			return (-1);
		}
		pd->hdrlen = sizeof(*th);
		pd->p_len = pd->tot_len - pd->off - (th->th_off << 2);
		pd->sport = &th->th_sport;
		pd->dport = &th->th_dport;
		pd->pcksum = &th->th_sum;
		break;
	}
	case IPPROTO_UDP: {
		struct udphdr *uh = &pd->hdr.udp;

		if (!pf_pull_hdr(pd->m, pd->off, uh, sizeof(*uh), action,
			reason, af)) {
			*action = PF_DROP;
			REASON_SET(reason, PFRES_SHORT);
			return (-1);
		}
		pd->hdrlen = sizeof(*uh);
		if (uh->uh_dport == 0 ||
		    ntohs(uh->uh_ulen) > pd->m->m_pkthdr.len - pd->off ||
		    ntohs(uh->uh_ulen) < sizeof(struct udphdr)) {
			*action = PF_DROP;
			REASON_SET(reason, PFRES_SHORT);
			return (-1);
		}
		pd->sport = &uh->uh_sport;
		pd->dport = &uh->uh_dport;
		pd->pcksum = &uh->uh_sum;
		break;
	}
	case IPPROTO_SCTP: {
		if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.sctp, sizeof(pd->hdr.sctp),
		    action, reason, af)) {
			*action = PF_DROP;
			REASON_SET(reason, PFRES_SHORT);
			return (-1);
		}
		pd->hdrlen = sizeof(pd->hdr.sctp);
		pd->p_len = pd->tot_len - pd->off;

		pd->sport = &pd->hdr.sctp.src_port;
		pd->dport = &pd->hdr.sctp.dest_port;
		if (pd->hdr.sctp.src_port == 0 || pd->hdr.sctp.dest_port == 0) {
			*action = PF_DROP;
			REASON_SET(reason, PFRES_SHORT);
			return (-1);
		}
		if (pf_scan_sctp(pd) != PF_PASS) {
			*action = PF_DROP;
			REASON_SET(reason, PFRES_SHORT);
			return (-1);
		}
		/*
		 * Placeholder. The SCTP checksum is 32-bits, but
		 * pf_test_state() expects to update a 16-bit checksum.
		 * Provide a dummy value which we'll subsequently ignore.
		 */
		pd->pcksum = &pd->sctp_dummy_sum;
		break;
	}
	case IPPROTO_ICMP: {
		if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp, ICMP_MINLEN,
			action, reason, af)) {
			*action = PF_DROP;
			REASON_SET(reason, PFRES_SHORT);
			return (-1);
		}
		pd->pcksum = &pd->hdr.icmp.icmp_cksum;
		pd->hdrlen = ICMP_MINLEN;
		break;
	}
#ifdef INET6
	case IPPROTO_ICMPV6: {
		size_t icmp_hlen = sizeof(struct icmp6_hdr);

		if (!pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen,
			action, reason, af)) {
			*action = PF_DROP;
			REASON_SET(reason, PFRES_SHORT);
			return (-1);
		}
		/* ICMP headers we look further into to match state */
		switch (pd->hdr.icmp6.icmp6_type) {
		case MLD_LISTENER_QUERY:
		case MLD_LISTENER_REPORT:
			icmp_hlen = sizeof(struct mld_hdr);
			break;
		case ND_NEIGHBOR_SOLICIT:
		case ND_NEIGHBOR_ADVERT:
			icmp_hlen = sizeof(struct nd_neighbor_solicit);
			/* FALLTHROUGH */
		case ND_ROUTER_SOLICIT:
		case ND_ROUTER_ADVERT:
		case ND_REDIRECT:
			if (pd->ttl != 255) {
				REASON_SET(reason, PFRES_NORM);
				return (PF_DROP);
			}
			break;
		}
		if (icmp_hlen > sizeof(struct icmp6_hdr) &&
		    !pf_pull_hdr(pd->m, pd->off, &pd->hdr.icmp6, icmp_hlen,
			action, reason, af)) {
			*action = PF_DROP;
			REASON_SET(reason, PFRES_SHORT);
			return (-1);
		}
		pd->hdrlen = icmp_hlen;
		pd->pcksum = &pd->hdr.icmp6.icmp6_cksum;
		break;
	}
#endif /* INET6 */
	}

	if (pd->sport)
		pd->osport = pd->nsport = *pd->sport;
	if (pd->dport)
		pd->odport = pd->ndport = *pd->dport;

	return (0);
}

static void
pf_counters_inc(int action, struct pf_pdesc *pd,
    struct pf_kstate *s, struct pf_krule *r, struct pf_krule *a)
{
	struct pf_krule		*tr;
	int			 dir = pd->dir;
	int			 dirndx;

	pf_counter_u64_critical_enter();
	pf_counter_u64_add_protected(
	    &pd->kif->pfik_bytes[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS],
	    pd->tot_len);
	pf_counter_u64_add_protected(
	    &pd->kif->pfik_packets[pd->af == AF_INET6][dir == PF_OUT][action != PF_PASS],
	    1);

	if (action == PF_PASS || action == PF_AFRT || r->action == PF_DROP) {
		dirndx = (dir == PF_OUT);
		pf_counter_u64_add_protected(&r->packets[dirndx], 1);
		pf_counter_u64_add_protected(&r->bytes[dirndx], pd->tot_len);
		pf_update_timestamp(r);

		if (a != NULL) {
			pf_counter_u64_add_protected(&a->packets[dirndx], 1);
			pf_counter_u64_add_protected(&a->bytes[dirndx], pd->tot_len);
		}
		if (s != NULL) {
			struct pf_krule_item	*ri;

			if (s->nat_rule != NULL) {
				pf_counter_u64_add_protected(&s->nat_rule->packets[dirndx],
				    1);
				pf_counter_u64_add_protected(&s->nat_rule->bytes[dirndx],
				    pd->tot_len);
			}
			/*
			 * Source nodes are accessed unlocked here.
			 * But since we are operating with stateful tracking
			 * and the state is locked, those SNs could not have
			 * been freed.
			 */
			for (pf_sn_types_t sn_type=0; sn_type<PF_SN_MAX; sn_type++) {
				if (s->sns[sn_type] != NULL) {
					counter_u64_add(
					    s->sns[sn_type]->packets[dirndx],
					    1);
					counter_u64_add(
					    s->sns[sn_type]->bytes[dirndx],
					    pd->tot_len);
				}
			}
			dirndx = (dir == s->direction) ? 0 : 1;
			s->packets[dirndx]++;
			s->bytes[dirndx] += pd->tot_len;

			SLIST_FOREACH(ri, &s->match_rules, entry) {
				pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1);
				pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd->tot_len);

				if (ri->r->src.addr.type == PF_ADDR_TABLE)
					pfr_update_stats(ri->r->src.addr.p.tbl,
					    (s == NULL) ? pd->src :
					    &s->key[(s->direction == PF_IN)]->
						addr[(s->direction == PF_OUT)],
					    pd->af, pd->tot_len, dir == PF_OUT,
					    r->action == PF_PASS, ri->r->src.neg);
				if (ri->r->dst.addr.type == PF_ADDR_TABLE)
					pfr_update_stats(ri->r->dst.addr.p.tbl,
					    (s == NULL) ? pd->dst :
					    &s->key[(s->direction == PF_IN)]->
						addr[(s->direction == PF_IN)],
					    pd->af, pd->tot_len, dir == PF_OUT,
					    r->action == PF_PASS, ri->r->dst.neg);
			}
		}

		tr = r;
		if (s != NULL && s->nat_rule != NULL &&
		    r == &V_pf_default_rule)
			tr = s->nat_rule;

		if (tr->src.addr.type == PF_ADDR_TABLE)
			pfr_update_stats(tr->src.addr.p.tbl,
			    (s == NULL) ? pd->src :
			    &s->key[(s->direction == PF_IN)]->
				addr[(s->direction == PF_OUT)],
			    pd->af, pd->tot_len, dir == PF_OUT,
			    r->action == PF_PASS, tr->src.neg);
		if (tr->dst.addr.type == PF_ADDR_TABLE)
			pfr_update_stats(tr->dst.addr.p.tbl,
			    (s == NULL) ? pd->dst :
			    &s->key[(s->direction == PF_IN)]->
				addr[(s->direction == PF_IN)],
			    pd->af, pd->tot_len, dir == PF_OUT,
			    r->action == PF_PASS, tr->dst.neg);
	}
	pf_counter_u64_critical_exit();
}
static void
pf_log_matches(struct pf_pdesc *pd, struct pf_krule *rm,
    struct pf_krule *am, struct pf_kruleset *ruleset,
    struct pf_krule_slist *matchrules)
{
	struct pf_krule_item	*ri;

	/* if this is the log(matches) rule, packet has been logged already */
	if (rm->log & PF_LOG_MATCHES)
		return;

	SLIST_FOREACH(ri, matchrules, entry)
		if (ri->r->log & PF_LOG_MATCHES)
			PFLOG_PACKET(rm->action, PFRES_MATCH, rm, am,
			    ruleset, pd, 1, ri->r);
}

#if defined(INET) || defined(INET6)
int
pf_test(sa_family_t af, int dir, int pflags, struct ifnet *ifp, struct mbuf **m0,
    struct inpcb *inp, struct pf_rule_actions *default_actions)
{
	struct pfi_kkif		*kif;
	u_short			 action, reason = 0;
	struct m_tag		*mtag;
	struct pf_krule		*a = NULL, *r = &V_pf_default_rule;
	struct pf_kstate	*s = NULL;
	struct pf_kruleset	*ruleset = NULL;
	struct pf_pdesc		 pd;
	int			 use_2nd_queue = 0;
	uint16_t		 tag;

	PF_RULES_RLOCK_TRACKER;
	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir));
	M_ASSERTPKTHDR(*m0);
	NET_EPOCH_ASSERT();

	if (!V_pf_status.running)
		return (PF_PASS);

	kif = (struct pfi_kkif *)ifp->if_pf_kif;

	if (__predict_false(kif == NULL)) {
		DPFPRINTF(PF_DEBUG_URGENT,
		    "%s: kif == NULL, if_xname %s",
		    __func__, ifp->if_xname);
		return (PF_DROP);
	}
	if (kif->pfik_flags & PFI_IFLAG_SKIP) {
		return (PF_PASS);
	}

	if ((*m0)->m_flags & M_SKIP_FIREWALL) {
		return (PF_PASS);
	}

	if (__predict_false(! M_WRITABLE(*m0))) {
		*m0 = m_unshare(*m0, M_NOWAIT);
		if (*m0 == NULL) {
			return (PF_DROP);
		}
	}

	pf_init_pdesc(&pd, *m0);

	if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) {
		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;

		ifp = ifnet_byindexgen(pd.pf_mtag->if_index,
		    pd.pf_mtag->if_idxgen);
		if (ifp == NULL || ifp->if_flags & IFF_DYING) {
			m_freem(*m0);
			*m0 = NULL;
			return (PF_PASS);
		}
		(ifp->if_output)(ifp, *m0, sintosa(&pd.pf_mtag->dst), NULL);
		*m0 = NULL;
		return (PF_PASS);
	}

	if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL &&
	    pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
		/* Dummynet re-injects packets after they've
		 * completed their delay. We've already
		 * processed them, so pass unconditionally. */

		/* But only once. We may see the packet multiple times (e.g.
		 * PFIL_IN/PFIL_OUT). */
		pf_dummynet_flag_remove(pd.m, pd.pf_mtag);

		return (PF_PASS);
	}

	PF_RULES_RLOCK();

	if (pf_setup_pdesc(af, dir, &pd, m0, &action, &reason,
		kif, default_actions) == -1) {
		if (action != PF_PASS)
			pd.act.log |= PF_LOG_FORCE;
		goto done;
	}

#ifdef INET
	if (af == AF_INET && dir == PF_OUT && pflags & PFIL_FWD &&
	    pd.df && (*m0)->m_pkthdr.len > ifp->if_mtu) {
		PF_RULES_RUNLOCK();
		icmp_error(*m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
			0, ifp->if_mtu);
		*m0 = NULL;
		return (PF_DROP);
	}
#endif /* INET */
#ifdef INET6
	/*
	 * If we end up changing IP addresses (e.g. binat) the stack may get
	 * confused and fail to send the icmp6 packet too big error. Just send
	 * it here, before we do any NAT.
	 */
	if (af == AF_INET6 && dir == PF_OUT && pflags & PFIL_FWD &&
	    IN6_LINKMTU(ifp) < pf_max_frag_size(*m0)) {
		PF_RULES_RUNLOCK();
		icmp6_error(*m0, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp));
		*m0 = NULL;
		return (PF_DROP);
	}
#endif /* INET6 */

	if (__predict_false(ip_divert_ptr != NULL) &&
	    ((mtag = m_tag_locate(pd.m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) {
		struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1);
		if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) ||
		    (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) {
			if (pd.pf_mtag == NULL &&
			    ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) {
				action = PF_DROP;
				goto done;
			}
			pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED;
		}
		if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) {
			pd.m->m_flags |= M_FASTFWD_OURS;
			pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT;
		}
		m_tag_delete(pd.m, mtag);

		mtag = m_tag_locate(pd.m, MTAG_IPFW_RULE, 0, NULL);
		if (mtag != NULL)
			m_tag_delete(pd.m, mtag);
	}

	switch (pd.virtual_proto) {
	case PF_VPROTO_FRAGMENT:
		/*
		 * handle fragments that aren't reassembled by
		 * normalization
		 */
		if (kif == NULL || r == NULL) /* pflog */
			action = PF_DROP;
		else
			action = pf_test_rule(&r, &s, &pd, &a,
			    &ruleset, &reason, inp);
		if (action != PF_PASS)
			REASON_SET(&reason, PFRES_FRAG);
		break;

	case IPPROTO_TCP: {
		/* Respond to SYN with a syncookie. */
		if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN &&
		    pd.dir == PF_IN && pf_synflood_check(&pd)) {
			pf_syncookie_send(&pd);
			action = PF_DROP;
			break;
		}

		if ((tcp_get_flags(&pd.hdr.tcp) & TH_ACK) && pd.p_len == 0)
			use_2nd_queue = 1;
		action = pf_normalize_tcp(&pd);
		if (action == PF_DROP)
			break;
		action = pf_test_state(&s, &pd, &reason);
		if (action == PF_PASS || action == PF_AFRT) {
			if (V_pfsync_update_state_ptr != NULL)
				V_pfsync_update_state_ptr(s);
			r = s->rule;
			a = s->anchor;
		} else if (s == NULL) {
			/* Validate remote SYN|ACK, re-create original SYN if
			 * valid. */
			if ((tcp_get_flags(&pd.hdr.tcp) & (TH_SYN|TH_ACK|TH_RST)) ==
			    TH_ACK && pf_syncookie_validate(&pd) &&
			    pd.dir == PF_IN) {
				struct mbuf *msyn;

				msyn = pf_syncookie_recreate_syn(&pd);
				if (msyn == NULL) {
					action = PF_DROP;
					break;
				}

				action = pf_test(af, dir, pflags, ifp, &msyn, inp,
				    &pd.act);
				m_freem(msyn);
				if (action != PF_PASS)
					break;

				action = pf_test_state(&s, &pd, &reason);
				if (action != PF_PASS || s == NULL) {
					action = PF_DROP;
					break;
				}

				s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1;
				s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1;
				pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST);
				action = pf_synproxy(&pd, s, &reason);
				break;
			} else {
				action = pf_test_rule(&r, &s, &pd,
				    &a, &ruleset, &reason, inp);
			}
		}
		break;
	}

	case IPPROTO_SCTP:
		action = pf_normalize_sctp(&pd);
		if (action == PF_DROP)
			break;
		/* fallthrough */
	case IPPROTO_UDP:
	default:
		action = pf_test_state(&s, &pd, &reason);
		if (action == PF_PASS || action == PF_AFRT) {
			if (V_pfsync_update_state_ptr != NULL)
				V_pfsync_update_state_ptr(s);
			r = s->rule;
			a = s->anchor;
		} else if (s == NULL) {
			action = pf_test_rule(&r, &s,
			    &pd, &a, &ruleset, &reason, inp);
		}
		break;

	case IPPROTO_ICMP:
	case IPPROTO_ICMPV6: {
		if (pd.virtual_proto == IPPROTO_ICMP && af != AF_INET) {
			action = PF_DROP;
			REASON_SET(&reason, PFRES_NORM);
			DPFPRINTF(PF_DEBUG_MISC,
			    "dropping IPv6 packet with ICMPv4 payload");
			break;
		}
		if (pd.virtual_proto == IPPROTO_ICMPV6 && af != AF_INET6) {
			action = PF_DROP;
			REASON_SET(&reason, PFRES_NORM);
			DPFPRINTF(PF_DEBUG_MISC,
			    "pf: dropping IPv4 packet with ICMPv6 payload");
			break;
		}
		action = pf_test_state_icmp(&s, &pd, &reason);
		if (action == PF_PASS || action == PF_AFRT) {
			if (V_pfsync_update_state_ptr != NULL)
				V_pfsync_update_state_ptr(s);
			r = s->rule;
			a = s->anchor;
		} else if (s == NULL)
			action = pf_test_rule(&r, &s, &pd,
			    &a, &ruleset, &reason, inp);
		break;
	}

	}

done:
	PF_RULES_RUNLOCK();

	if (pd.m == NULL)
		goto eat_pkt;

	if (s)
		memcpy(&pd.act, &s->act, sizeof(s->act));

	if (action == PF_PASS && pd.badopts != 0 && !pd.act.allow_opts) {
		action = PF_DROP;
		REASON_SET(&reason, PFRES_IPOPTIONS);
		pd.act.log = PF_LOG_FORCE;
		DPFPRINTF(PF_DEBUG_MISC,
		    "pf: dropping packet with dangerous headers");
	}

	if (pd.act.max_pkt_size && pd.act.max_pkt_size &&
	    pd.tot_len > pd.act.max_pkt_size) {
		action = PF_DROP;
		REASON_SET(&reason, PFRES_NORM);
		pd.act.log = PF_LOG_FORCE;
		DPFPRINTF(PF_DEBUG_MISC,
		    "pf: dropping overly long packet");
	}

	if (s) {
		uint8_t log = pd.act.log;
		memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions));
		pd.act.log |= log;
		tag = s->tag;
	} else {
		tag = r->tag;
	}

	if (tag > 0 && pf_tag_packet(&pd, tag)) {
		action = PF_DROP;
		REASON_SET(&reason, PFRES_MEMORY);
	}

	pf_scrub(&pd);
	if (pd.proto == IPPROTO_TCP && pd.act.max_mss)
		pf_normalize_mss(&pd);

	if (pd.act.rtableid >= 0)
		M_SETFIB(pd.m, pd.act.rtableid);

	if (pd.act.flags & PFSTATE_SETPRIO) {
		if (pd.tos & IPTOS_LOWDELAY)
			use_2nd_queue = 1;
		if (vlan_set_pcp(pd.m, pd.act.set_prio[use_2nd_queue])) {
			action = PF_DROP;
			REASON_SET(&reason, PFRES_MEMORY);
			pd.act.log = PF_LOG_FORCE;
			DPFPRINTF(PF_DEBUG_MISC,
			    "pf: failed to allocate 802.1q mtag");
		}
	}

#ifdef ALTQ
	if (action == PF_PASS && pd.act.qid) {
		if (pd.pf_mtag == NULL &&
		    ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) {
			action = PF_DROP;
			REASON_SET(&reason, PFRES_MEMORY);
		} else {
			if (s != NULL)
				pd.pf_mtag->qid_hash = pf_state_hash(s);
			if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY))
				pd.pf_mtag->qid = pd.act.pqid;
			else
				pd.pf_mtag->qid = pd.act.qid;
			/* Add hints for ecn. */
			pd.pf_mtag->hdr = mtod(pd.m, void *);
		}
	}
#endif /* ALTQ */

	/*
	 * connections redirected to loopback should not match sockets
	 * bound specifically to loopback due to security implications,
	 * see tcp_input() and in_pcblookup_listen().
	 */
	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule != NULL &&
	    (s->nat_rule->action == PF_RDR ||
	    s->nat_rule->action == PF_BINAT) &&
	    pf_is_loopback(af, pd.dst))
		pd.m->m_flags |= M_SKIP_FIREWALL;

	if (af == AF_INET && __predict_false(ip_divert_ptr != NULL) &&
	    action == PF_PASS && r->divert.port && !PACKET_LOOPED(&pd)) {
		mtag = m_tag_alloc(MTAG_PF_DIVERT, 0,
		    sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO);
		if (mtag != NULL) {
			((struct pf_divert_mtag *)(mtag+1))->port =
			    ntohs(r->divert.port);
			((struct pf_divert_mtag *)(mtag+1))->idir =
			    (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN :
			    PF_DIVERT_MTAG_DIR_OUT;

			if (s)
				PF_STATE_UNLOCK(s);

			m_tag_prepend(pd.m, mtag);
			if (pd.m->m_flags & M_FASTFWD_OURS) {
				if (pd.pf_mtag == NULL &&
				    ((pd.pf_mtag = pf_get_mtag(pd.m)) == NULL)) {
					action = PF_DROP;
					REASON_SET(&reason, PFRES_MEMORY);
					pd.act.log = PF_LOG_FORCE;
					DPFPRINTF(PF_DEBUG_MISC,
					    "pf: failed to allocate tag");
				} else {
					pd.pf_mtag->flags |=
					    PF_MTAG_FLAG_FASTFWD_OURS_PRESENT;
					pd.m->m_flags &= ~M_FASTFWD_OURS;
				}
			}
			ip_divert_ptr(*m0, dir == PF_IN);
			*m0 = NULL;

			return (action);
		} else {
			/* XXX: ipfw has the same behaviour! */
			action = PF_DROP;
			REASON_SET(&reason, PFRES_MEMORY);
			pd.act.log = PF_LOG_FORCE;
			DPFPRINTF(PF_DEBUG_MISC,
			    "pf: failed to allocate divert tag");
		}
	}
	/* XXX: Anybody working on it?! */
	if (af == AF_INET6 && r->divert.port)
		printf("pf: divert(9) is not supported for IPv6\n");

	/* this flag will need revising if the pkt is forwarded */
	if (pd.pf_mtag)
		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED;

	if (pd.act.log) {
		struct pf_krule		*lr;
		struct pf_krule_item	*ri;

		if (s != NULL && s->nat_rule != NULL &&
		    s->nat_rule->log & PF_LOG_ALL)
			lr = s->nat_rule;
		else
			lr = r;

		if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL)
			PFLOG_PACKET(action, reason, lr, a,
			    ruleset, &pd, (s == NULL), NULL);
		if (s) {
			SLIST_FOREACH(ri, &s->match_rules, entry)
				if (ri->r->log & PF_LOG_ALL)
					PFLOG_PACKET(action,
					    reason, ri->r, a, ruleset, &pd, 0, NULL);
		}
	}

	pf_counters_inc(action, &pd, s, r, a);

	switch (action) {
	case PF_SYNPROXY_DROP:
		m_freem(*m0);
	case PF_DEFER:
		*m0 = NULL;
		action = PF_PASS;
		break;
	case PF_DROP:
		m_freem(*m0);
		*m0 = NULL;
		break;
	case PF_AFRT:
		if (pf_translate_af(&pd)) {
			*m0 = pd.m;
			action = PF_DROP;
			break;
		}
#ifdef INET
		if (pd.naf == AF_INET) {
			action = pf_route(r, kif->pfik_ifp, s, &pd,
			    inp);
		}
#endif /* INET */
#ifdef INET6
		if (pd.naf == AF_INET6) {
			action = pf_route6(r, kif->pfik_ifp, s, &pd,
			    inp);
}
#endif /* INET6 */
		*m0 = pd.m;
		goto out;
		break;
	default:
		if (pd.act.rt) {
			switch (af) {
#ifdef INET
			case AF_INET:
				/* pf_route() returns unlocked. */
				action = pf_route(r, kif->pfik_ifp, s, &pd,
				    inp);
				break;
#endif /* INET */
#ifdef INET6
			case AF_INET6:
				/* pf_route6() returns unlocked. */
				action = pf_route6(r, kif->pfik_ifp, s, &pd,
				    inp);
				break;
#endif /* INET6 */
			}
			*m0 = pd.m;
			goto out;
		}
		if (pf_dummynet(&pd, s, r, m0) != 0) {
			action = PF_DROP;
			REASON_SET(&reason, PFRES_MEMORY);
		}
		break;
	}

eat_pkt:
	SDT_PROBE4(pf, ip, test, done, action, reason, r, s);

	if (s && action != PF_DROP) {
		if (!s->if_index_in && dir == PF_IN)
			s->if_index_in = ifp->if_index;
		else if (!s->if_index_out && dir == PF_OUT)
			s->if_index_out = ifp->if_index;
	}

	if (s)
		PF_STATE_UNLOCK(s);

out:
#ifdef INET6
	/* If reassembled packet passed, create new fragments. */
	if (af == AF_INET6 && action == PF_PASS && *m0 && dir == PF_OUT &&
	    (! (pflags & PF_PFIL_NOREFRAGMENT)) &&
	    (mtag = m_tag_find(pd.m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL)
		action = pf_refragment6(ifp, m0, mtag, NULL, pflags & PFIL_FWD);
#endif /* INET6 */

	pf_sctp_multihome_delayed(&pd, kif, s, action);

	return (action);
}
#endif /* INET || INET6 */