xref: /freebsd/sys/netpfil/pf/pf.c (revision 5f047ae0c73b4424b29e64d50b71346b9abd058e)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2001 Daniel Hartmeier
5  * Copyright (c) 2002 - 2008 Henning Brauer
6  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  *
13  *    - Redistributions of source code must retain the above copyright
14  *      notice, this list of conditions and the following disclaimer.
15  *    - Redistributions in binary form must reproduce the above
16  *      copyright notice, this list of conditions and the following
17  *      disclaimer in the documentation and/or other materials provided
18  *      with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
28  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
30  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31  * POSSIBILITY OF SUCH DAMAGE.
32  *
33  * Effort sponsored in part by the Defense Advanced Research Projects
34  * Agency (DARPA) and Air Force Research Laboratory, Air Force
35  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
36  *
37  *	$OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
38  */
39 
40 #include <sys/cdefs.h>
41 #include "opt_bpf.h"
42 #include "opt_inet.h"
43 #include "opt_inet6.h"
44 #include "opt_pf.h"
45 #include "opt_sctp.h"
46 
47 #include <sys/param.h>
48 #include <sys/bus.h>
49 #include <sys/endian.h>
50 #include <sys/gsb_crc32.h>
51 #include <sys/hash.h>
52 #include <sys/interrupt.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/limits.h>
56 #include <sys/mbuf.h>
57 #include <sys/md5.h>
58 #include <sys/random.h>
59 #include <sys/refcount.h>
60 #include <sys/sdt.h>
61 #include <sys/socket.h>
62 #include <sys/sysctl.h>
63 #include <sys/taskqueue.h>
64 #include <sys/ucred.h>
65 
66 #include <net/if.h>
67 #include <net/if_var.h>
68 #include <net/if_private.h>
69 #include <net/if_types.h>
70 #include <net/if_vlan_var.h>
71 #include <net/route.h>
72 #include <net/route/nhop.h>
73 #include <net/vnet.h>
74 
75 #include <net/pfil.h>
76 #include <net/pfvar.h>
77 #include <net/if_pflog.h>
78 #include <net/if_pfsync.h>
79 
80 #include <netinet/in_pcb.h>
81 #include <netinet/in_var.h>
82 #include <netinet/in_fib.h>
83 #include <netinet/ip.h>
84 #include <netinet/ip_fw.h>
85 #include <netinet/ip_icmp.h>
86 #include <netinet/icmp_var.h>
87 #include <netinet/ip_var.h>
88 #include <netinet/tcp.h>
89 #include <netinet/tcp_fsm.h>
90 #include <netinet/tcp_seq.h>
91 #include <netinet/tcp_timer.h>
92 #include <netinet/tcp_var.h>
93 #include <netinet/udp.h>
94 #include <netinet/udp_var.h>
95 
96 /* dummynet */
97 #include <netinet/ip_dummynet.h>
98 #include <netinet/ip_fw.h>
99 #include <netpfil/ipfw/dn_heap.h>
100 #include <netpfil/ipfw/ip_fw_private.h>
101 #include <netpfil/ipfw/ip_dn_private.h>
102 
103 #ifdef INET6
104 #include <netinet/ip6.h>
105 #include <netinet/icmp6.h>
106 #include <netinet6/nd6.h>
107 #include <netinet6/ip6_var.h>
108 #include <netinet6/in6_pcb.h>
109 #include <netinet6/in6_fib.h>
110 #include <netinet6/scope6_var.h>
111 #endif /* INET6 */
112 
113 #include <netinet/sctp_header.h>
114 #include <netinet/sctp_crc32.h>
115 
116 #include <machine/in_cksum.h>
117 #include <security/mac/mac_framework.h>
118 
119 #define	DPFPRINTF(n, x)	if (V_pf_status.debug >= (n)) printf x
120 
121 SDT_PROVIDER_DEFINE(pf);
122 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *",
123     "struct pf_kstate *");
124 SDT_PROBE_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *",
125     "struct pf_kstate *");
126 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *",
127     "struct pf_state_key_cmp *", "int", "struct pf_pdesc *",
128     "struct pf_kstate *");
129 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *",
130     "struct pf_krule *", "struct mbuf *", "int");
131 
132 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *",
133     "struct mbuf *");
134 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *");
135 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch,
136     "int", "struct pf_keth_rule *", "char *");
137 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *");
138 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match,
139     "int", "struct pf_keth_rule *");
140 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t");
141 
142 /*
143  * Global variables
144  */
145 
146 /* state tables */
147 VNET_DEFINE(struct pf_altqqueue,	 pf_altqs[4]);
148 VNET_DEFINE(struct pf_kpalist,		 pf_pabuf);
149 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_active);
150 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_active);
151 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_inactive);
152 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_inactive);
153 VNET_DEFINE(struct pf_kstatus,		 pf_status);
154 
155 VNET_DEFINE(u_int32_t,			 ticket_altqs_active);
156 VNET_DEFINE(u_int32_t,			 ticket_altqs_inactive);
157 VNET_DEFINE(int,			 altqs_inactive_open);
158 VNET_DEFINE(u_int32_t,			 ticket_pabuf);
159 
160 VNET_DEFINE(MD5_CTX,			 pf_tcp_secret_ctx);
161 #define	V_pf_tcp_secret_ctx		 VNET(pf_tcp_secret_ctx)
162 VNET_DEFINE(u_char,			 pf_tcp_secret[16]);
163 #define	V_pf_tcp_secret			 VNET(pf_tcp_secret)
164 VNET_DEFINE(int,			 pf_tcp_secret_init);
165 #define	V_pf_tcp_secret_init		 VNET(pf_tcp_secret_init)
166 VNET_DEFINE(int,			 pf_tcp_iss_off);
167 #define	V_pf_tcp_iss_off		 VNET(pf_tcp_iss_off)
168 VNET_DECLARE(int,			 pf_vnet_active);
169 #define	V_pf_vnet_active		 VNET(pf_vnet_active)
170 
171 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx);
172 #define V_pf_purge_idx	VNET(pf_purge_idx)
173 
174 #ifdef PF_WANT_32_TO_64_COUNTER
175 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter);
176 #define	V_pf_counter_periodic_iter	VNET(pf_counter_periodic_iter)
177 
178 VNET_DEFINE(struct allrulelist_head, pf_allrulelist);
179 VNET_DEFINE(size_t, pf_allrulecount);
180 VNET_DEFINE(struct pf_krule *, pf_rulemarker);
181 #endif
182 
183 struct pf_sctp_endpoint;
184 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint);
185 struct pf_sctp_source {
186 	sa_family_t			af;
187 	struct pf_addr			addr;
188 	TAILQ_ENTRY(pf_sctp_source)	entry;
189 };
190 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source);
191 struct pf_sctp_endpoint
192 {
193 	uint32_t		 v_tag;
194 	struct pf_sctp_sources	 sources;
195 	RB_ENTRY(pf_sctp_endpoint)	entry;
196 };
197 static int
198 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b)
199 {
200 	return (a->v_tag - b->v_tag);
201 }
202 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare);
203 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare);
204 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints);
205 #define V_pf_sctp_endpoints	VNET(pf_sctp_endpoints)
206 static struct mtx_padalign pf_sctp_endpoints_mtx;
207 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF);
208 #define	PF_SCTP_ENDPOINTS_LOCK()	mtx_lock(&pf_sctp_endpoints_mtx)
209 #define	PF_SCTP_ENDPOINTS_UNLOCK()	mtx_unlock(&pf_sctp_endpoints_mtx)
210 
211 /*
212  * Queue for pf_intr() sends.
213  */
214 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
215 struct pf_send_entry {
216 	STAILQ_ENTRY(pf_send_entry)	pfse_next;
217 	struct mbuf			*pfse_m;
218 	enum {
219 		PFSE_IP,
220 		PFSE_IP6,
221 		PFSE_ICMP,
222 		PFSE_ICMP6,
223 	}				pfse_type;
224 	struct {
225 		int		type;
226 		int		code;
227 		int		mtu;
228 	} icmpopts;
229 };
230 
231 STAILQ_HEAD(pf_send_head, pf_send_entry);
232 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue);
233 #define	V_pf_sendqueue	VNET(pf_sendqueue)
234 
235 static struct mtx_padalign pf_sendqueue_mtx;
236 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
237 #define	PF_SENDQ_LOCK()		mtx_lock(&pf_sendqueue_mtx)
238 #define	PF_SENDQ_UNLOCK()	mtx_unlock(&pf_sendqueue_mtx)
239 
240 /*
241  * Queue for pf_overload_task() tasks.
242  */
243 struct pf_overload_entry {
244 	SLIST_ENTRY(pf_overload_entry)	next;
245 	struct pf_addr  		addr;
246 	sa_family_t			af;
247 	uint8_t				dir;
248 	struct pf_krule  		*rule;
249 };
250 
251 SLIST_HEAD(pf_overload_head, pf_overload_entry);
252 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue);
253 #define V_pf_overloadqueue	VNET(pf_overloadqueue)
254 VNET_DEFINE_STATIC(struct task, pf_overloadtask);
255 #define	V_pf_overloadtask	VNET(pf_overloadtask)
256 
257 static struct mtx_padalign pf_overloadqueue_mtx;
258 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
259     "pf overload/flush queue", MTX_DEF);
260 #define	PF_OVERLOADQ_LOCK()	mtx_lock(&pf_overloadqueue_mtx)
261 #define	PF_OVERLOADQ_UNLOCK()	mtx_unlock(&pf_overloadqueue_mtx)
262 
263 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules);
264 struct mtx_padalign pf_unlnkdrules_mtx;
265 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
266     MTX_DEF);
267 
268 struct sx pf_config_lock;
269 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config");
270 
271 struct mtx_padalign pf_table_stats_lock;
272 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats",
273     MTX_DEF);
274 
275 VNET_DEFINE_STATIC(uma_zone_t,	pf_sources_z);
276 #define	V_pf_sources_z	VNET(pf_sources_z)
277 uma_zone_t		pf_mtag_z;
278 VNET_DEFINE(uma_zone_t,	 pf_state_z);
279 VNET_DEFINE(uma_zone_t,	 pf_state_key_z);
280 
281 VNET_DEFINE(struct unrhdr64, pf_stateid);
282 
283 static void		 pf_src_tree_remove_state(struct pf_kstate *);
284 static void		 pf_init_threshold(struct pf_threshold *, u_int32_t,
285 			    u_int32_t);
286 static void		 pf_add_threshold(struct pf_threshold *);
287 static int		 pf_check_threshold(struct pf_threshold *);
288 
289 static void		 pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
290 			    u_int16_t *, u_int16_t *, struct pf_addr *,
291 			    u_int16_t, u_int8_t, sa_family_t);
292 static int		 pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
293 			    struct tcphdr *, struct pf_state_peer *);
294 static void		 pf_change_icmp(struct pf_addr *, u_int16_t *,
295 			    struct pf_addr *, struct pf_addr *, u_int16_t,
296 			    u_int16_t *, u_int16_t *, u_int16_t *,
297 			    u_int16_t *, u_int8_t, sa_family_t);
298 static void		 pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
299 			    sa_family_t, struct pf_krule *, int);
300 static void		 pf_detach_state(struct pf_kstate *);
301 static int		 pf_state_key_attach(struct pf_state_key *,
302 			    struct pf_state_key *, struct pf_kstate *);
303 static void		 pf_state_key_detach(struct pf_kstate *, int);
304 static int		 pf_state_key_ctor(void *, int, void *, int);
305 static u_int32_t	 pf_tcp_iss(struct pf_pdesc *);
306 static __inline void	 pf_dummynet_flag_remove(struct mbuf *m,
307 			    struct pf_mtag *pf_mtag);
308 static int		 pf_dummynet(struct pf_pdesc *, struct pf_kstate *,
309 			    struct pf_krule *, struct mbuf **);
310 static int		 pf_dummynet_route(struct pf_pdesc *,
311 			    struct pf_kstate *, struct pf_krule *,
312 			    struct ifnet *, struct sockaddr *, struct mbuf **);
313 static int		 pf_test_eth_rule(int, struct pfi_kkif *,
314 			    struct mbuf **);
315 static int		 pf_test_rule(struct pf_krule **, struct pf_kstate **,
316 			    struct pfi_kkif *, struct mbuf *, int,
317 			    struct pf_pdesc *, struct pf_krule **,
318 			    struct pf_kruleset **, struct inpcb *);
319 static int		 pf_create_state(struct pf_krule *, struct pf_krule *,
320 			    struct pf_krule *, struct pf_pdesc *,
321 			    struct pf_ksrc_node *, struct pf_state_key *,
322 			    struct pf_state_key *, struct mbuf *, int,
323 			    u_int16_t, u_int16_t, int *, struct pfi_kkif *,
324 			    struct pf_kstate **, int, u_int16_t, u_int16_t,
325 			    int, struct pf_krule_slist *);
326 static int		 pf_test_fragment(struct pf_krule **, struct pfi_kkif *,
327 			    struct mbuf *, void *, struct pf_pdesc *,
328 			    struct pf_krule **, struct pf_kruleset **);
329 static int		 pf_tcp_track_full(struct pf_kstate **,
330 			    struct pfi_kkif *, struct mbuf *, int,
331 			    struct pf_pdesc *, u_short *, int *);
332 static int		 pf_tcp_track_sloppy(struct pf_kstate **,
333 			    struct pf_pdesc *, u_short *);
334 static int		 pf_test_state_tcp(struct pf_kstate **,
335 			    struct pfi_kkif *, struct mbuf *, int,
336 			    void *, struct pf_pdesc *, u_short *);
337 static int		 pf_test_state_udp(struct pf_kstate **,
338 			    struct pfi_kkif *, struct mbuf *, int,
339 			    void *, struct pf_pdesc *);
340 static int		 pf_test_state_icmp(struct pf_kstate **,
341 			    struct pfi_kkif *, struct mbuf *, int,
342 			    void *, struct pf_pdesc *, u_short *);
343 static void		 pf_sctp_multihome_detach_addr(const struct pf_kstate *);
344 static void		 pf_sctp_multihome_delayed(struct pf_pdesc *, int,
345 			    struct pfi_kkif *, struct pf_kstate *, int);
346 static int		 pf_test_state_sctp(struct pf_kstate **,
347 			    struct pfi_kkif *, struct mbuf *, int,
348 			    void *, struct pf_pdesc *, u_short *);
349 static int		 pf_test_state_other(struct pf_kstate **,
350 			    struct pfi_kkif *, struct mbuf *, struct pf_pdesc *);
351 static u_int16_t	 pf_calc_mss(struct pf_addr *, sa_family_t,
352 				int, u_int16_t);
353 static int		 pf_check_proto_cksum(struct mbuf *, int, int,
354 			    u_int8_t, sa_family_t);
355 static void		 pf_print_state_parts(struct pf_kstate *,
356 			    struct pf_state_key *, struct pf_state_key *);
357 static void		 pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t,
358 			    bool, u_int8_t);
359 static struct pf_kstate	*pf_find_state(struct pfi_kkif *,
360 			    struct pf_state_key_cmp *, u_int);
361 static int		 pf_src_connlimit(struct pf_kstate **);
362 static void		 pf_overload_task(void *v, int pending);
363 static u_short		 pf_insert_src_node(struct pf_ksrc_node **,
364 			    struct pf_krule *, struct pf_addr *, sa_family_t);
365 static u_int		 pf_purge_expired_states(u_int, int);
366 static void		 pf_purge_unlinked_rules(void);
367 static int		 pf_mtag_uminit(void *, int, int);
368 static void		 pf_mtag_free(struct m_tag *);
369 static void		 pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *,
370 			    int, struct pf_state_key *);
371 #ifdef INET
372 static void		 pf_route(struct mbuf **, struct pf_krule *,
373 			    struct ifnet *, struct pf_kstate *,
374 			    struct pf_pdesc *, struct inpcb *);
375 #endif /* INET */
376 #ifdef INET6
377 static void		 pf_change_a6(struct pf_addr *, u_int16_t *,
378 			    struct pf_addr *, u_int8_t);
379 static void		 pf_route6(struct mbuf **, struct pf_krule *,
380 			    struct ifnet *, struct pf_kstate *,
381 			    struct pf_pdesc *, struct inpcb *);
382 #endif /* INET6 */
383 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t);
384 
385 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
386 
387 extern int pf_end_threads;
388 extern struct proc *pf_purge_proc;
389 
390 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
391 
392 #define	PACKET_UNDO_NAT(_m, _pd, _off, _s)		\
393 	do {								\
394 		struct pf_state_key *nk;				\
395 		if ((pd->dir) == PF_OUT)					\
396 			nk = (_s)->key[PF_SK_STACK];			\
397 		else							\
398 			nk = (_s)->key[PF_SK_WIRE];			\
399 		pf_packet_rework_nat(_m, _pd, _off, nk);		\
400 	} while (0)
401 
402 #define	PACKET_LOOPED(pd)	((pd)->pf_mtag &&			\
403 				 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED)
404 
405 #define	STATE_LOOKUP(i, k, s, pd)					\
406 	do {								\
407 		(s) = pf_find_state((i), (k), (pd->dir));			\
408 		SDT_PROBE5(pf, ip, state, lookup, i, k, (pd->dir), pd, (s));	\
409 		if ((s) == NULL)					\
410 			return (PF_DROP);				\
411 		if (PACKET_LOOPED(pd))					\
412 			return (PF_PASS);				\
413 	} while (0)
414 
415 static struct pfi_kkif *
416 BOUND_IFACE(struct pf_krule *r, struct pfi_kkif *k, struct pf_pdesc *pd)
417 {
418 	/* Floating unless otherwise specified. */
419 	if (! (r->rule_flag & PFRULE_IFBOUND))
420 		return (V_pfi_all);
421 
422 	/* Don't overrule the interface for states created on incoming packets. */
423 	if (pd->dir == PF_IN)
424 		return (k);
425 
426 	/* No route-to, so don't overrrule. */
427 	if (r->rt != PF_ROUTETO)
428 		return (k);
429 
430 	if (r->rpool.cur == NULL)
431 		return (k);
432 
433 	/* Bind to the route-to interface. */
434 	return (r->rpool.cur->kif);
435 }
436 
437 #define	STATE_INC_COUNTERS(s)						\
438 	do {								\
439 		struct pf_krule_item *mrm;				\
440 		counter_u64_add(s->rule.ptr->states_cur, 1);		\
441 		counter_u64_add(s->rule.ptr->states_tot, 1);		\
442 		if (s->anchor.ptr != NULL) {				\
443 			counter_u64_add(s->anchor.ptr->states_cur, 1);	\
444 			counter_u64_add(s->anchor.ptr->states_tot, 1);	\
445 		}							\
446 		if (s->nat_rule.ptr != NULL) {				\
447 			counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
448 			counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
449 		}							\
450 		SLIST_FOREACH(mrm, &s->match_rules, entry) {		\
451 			counter_u64_add(mrm->r->states_cur, 1);		\
452 			counter_u64_add(mrm->r->states_tot, 1);		\
453 		}							\
454 	} while (0)
455 
456 #define	STATE_DEC_COUNTERS(s)						\
457 	do {								\
458 		struct pf_krule_item *mrm;				\
459 		if (s->nat_rule.ptr != NULL)				\
460 			counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
461 		if (s->anchor.ptr != NULL)				\
462 			counter_u64_add(s->anchor.ptr->states_cur, -1);	\
463 		counter_u64_add(s->rule.ptr->states_cur, -1);		\
464 		SLIST_FOREACH(mrm, &s->match_rules, entry)		\
465 			counter_u64_add(mrm->r->states_cur, -1);	\
466 	} while (0)
467 
468 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
469 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items");
470 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
471 VNET_DEFINE(struct pf_idhash *, pf_idhash);
472 VNET_DEFINE(struct pf_srchash *, pf_srchash);
473 
474 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
475     "pf(4)");
476 
477 u_long	pf_hashmask;
478 u_long	pf_srchashmask;
479 static u_long	pf_hashsize;
480 static u_long	pf_srchashsize;
481 u_long	pf_ioctl_maxcount = 65535;
482 
483 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
484     &pf_hashsize, 0, "Size of pf(4) states hashtable");
485 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
486     &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
487 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN,
488     &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call");
489 
490 VNET_DEFINE(void *, pf_swi_cookie);
491 VNET_DEFINE(struct intr_event *, pf_swi_ie);
492 
493 VNET_DEFINE(uint32_t, pf_hashseed);
494 #define	V_pf_hashseed	VNET(pf_hashseed)
495 
496 static void
497 pf_sctp_checksum(struct mbuf *m, int off)
498 {
499 	uint32_t sum = 0;
500 
501 	/* Zero out the checksum, to enable recalculation. */
502 	m_copyback(m, off + offsetof(struct sctphdr, checksum),
503 	    sizeof(sum), (caddr_t)&sum);
504 
505 	sum = sctp_calculate_cksum(m, off);
506 
507 	m_copyback(m, off + offsetof(struct sctphdr, checksum),
508 	    sizeof(sum), (caddr_t)&sum);
509 }
510 
511 int
512 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
513 {
514 
515 	switch (af) {
516 #ifdef INET
517 	case AF_INET:
518 		if (a->addr32[0] > b->addr32[0])
519 			return (1);
520 		if (a->addr32[0] < b->addr32[0])
521 			return (-1);
522 		break;
523 #endif /* INET */
524 #ifdef INET6
525 	case AF_INET6:
526 		if (a->addr32[3] > b->addr32[3])
527 			return (1);
528 		if (a->addr32[3] < b->addr32[3])
529 			return (-1);
530 		if (a->addr32[2] > b->addr32[2])
531 			return (1);
532 		if (a->addr32[2] < b->addr32[2])
533 			return (-1);
534 		if (a->addr32[1] > b->addr32[1])
535 			return (1);
536 		if (a->addr32[1] < b->addr32[1])
537 			return (-1);
538 		if (a->addr32[0] > b->addr32[0])
539 			return (1);
540 		if (a->addr32[0] < b->addr32[0])
541 			return (-1);
542 		break;
543 #endif /* INET6 */
544 	default:
545 		panic("%s: unknown address family %u", __func__, af);
546 	}
547 	return (0);
548 }
549 
550 static void
551 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off,
552 	struct pf_state_key *nk)
553 {
554 
555 	switch (pd->proto) {
556 	case IPPROTO_TCP: {
557 		struct tcphdr *th = &pd->hdr.tcp;
558 
559 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af))
560 			pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum,
561 			    &th->th_sum, &nk->addr[pd->sidx],
562 			    nk->port[pd->sidx], 0, pd->af);
563 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af))
564 			pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum,
565 			    &th->th_sum, &nk->addr[pd->didx],
566 			    nk->port[pd->didx], 0, pd->af);
567 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
568 		break;
569 	}
570 	case IPPROTO_UDP: {
571 		struct udphdr *uh = &pd->hdr.udp;
572 
573 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af))
574 			pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
575 			    &uh->uh_sum, &nk->addr[pd->sidx],
576 			    nk->port[pd->sidx], 1, pd->af);
577 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af))
578 			pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
579 			    &uh->uh_sum, &nk->addr[pd->didx],
580 			    nk->port[pd->didx], 1, pd->af);
581 		m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
582 		break;
583 	}
584 	case IPPROTO_SCTP: {
585 		struct sctphdr *sh = &pd->hdr.sctp;
586 		uint16_t checksum = 0;
587 
588 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) {
589 			pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum,
590 			    &checksum, &nk->addr[pd->sidx],
591 			    nk->port[pd->sidx], 1, pd->af);
592 		}
593 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) {
594 			pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum,
595 			    &checksum, &nk->addr[pd->didx],
596 			    nk->port[pd->didx], 1, pd->af);
597 		}
598 
599 		break;
600 	}
601 	case IPPROTO_ICMP: {
602 		struct icmp *ih = &pd->hdr.icmp;
603 
604 		if (nk->port[pd->sidx] != ih->icmp_id) {
605 			pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
606 			    ih->icmp_cksum, ih->icmp_id,
607 			    nk->port[pd->sidx], 0);
608 			ih->icmp_id = nk->port[pd->sidx];
609 			pd->sport = &ih->icmp_id;
610 
611 			m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih);
612 		}
613 		/* FALLTHROUGH */
614 	}
615 	default:
616 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) {
617 			switch (pd->af) {
618 			case AF_INET:
619 				pf_change_a(&pd->src->v4.s_addr,
620 				    pd->ip_sum, nk->addr[pd->sidx].v4.s_addr,
621 				    0);
622 				break;
623 			case AF_INET6:
624 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
625 				break;
626 			}
627 		}
628 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) {
629 			switch (pd->af) {
630 			case AF_INET:
631 				pf_change_a(&pd->dst->v4.s_addr,
632 				    pd->ip_sum, nk->addr[pd->didx].v4.s_addr,
633 				    0);
634 				break;
635 			case AF_INET6:
636 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
637 				break;
638 			}
639 		}
640 		break;
641 	}
642 }
643 
644 static __inline uint32_t
645 pf_hashkey(struct pf_state_key *sk)
646 {
647 	uint32_t h;
648 
649 	h = murmur3_32_hash32((uint32_t *)sk,
650 	    sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
651 	    V_pf_hashseed);
652 
653 	return (h & pf_hashmask);
654 }
655 
656 static __inline uint32_t
657 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
658 {
659 	uint32_t h;
660 
661 	switch (af) {
662 	case AF_INET:
663 		h = murmur3_32_hash32((uint32_t *)&addr->v4,
664 		    sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
665 		break;
666 	case AF_INET6:
667 		h = murmur3_32_hash32((uint32_t *)&addr->v6,
668 		    sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
669 		break;
670 	default:
671 		panic("%s: unknown address family %u", __func__, af);
672 	}
673 
674 	return (h & pf_srchashmask);
675 }
676 
677 #ifdef ALTQ
678 static int
679 pf_state_hash(struct pf_kstate *s)
680 {
681 	u_int32_t hv = (intptr_t)s / sizeof(*s);
682 
683 	hv ^= crc32(&s->src, sizeof(s->src));
684 	hv ^= crc32(&s->dst, sizeof(s->dst));
685 	if (hv == 0)
686 		hv = 1;
687 	return (hv);
688 }
689 #endif
690 
691 static __inline void
692 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate)
693 {
694 	if (which == PF_PEER_DST || which == PF_PEER_BOTH)
695 		s->dst.state = newstate;
696 	if (which == PF_PEER_DST)
697 		return;
698 	if (s->src.state == newstate)
699 		return;
700 	if (s->creatorid == V_pf_status.hostid &&
701 	    s->key[PF_SK_STACK] != NULL &&
702 	    s->key[PF_SK_STACK]->proto == IPPROTO_TCP &&
703 	    !(TCPS_HAVEESTABLISHED(s->src.state) ||
704 	    s->src.state == TCPS_CLOSED) &&
705 	    (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED))
706 		atomic_add_32(&V_pf_status.states_halfopen, -1);
707 
708 	s->src.state = newstate;
709 }
710 
711 #ifdef INET6
712 void
713 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
714 {
715 	switch (af) {
716 #ifdef INET
717 	case AF_INET:
718 		dst->addr32[0] = src->addr32[0];
719 		break;
720 #endif /* INET */
721 	case AF_INET6:
722 		dst->addr32[0] = src->addr32[0];
723 		dst->addr32[1] = src->addr32[1];
724 		dst->addr32[2] = src->addr32[2];
725 		dst->addr32[3] = src->addr32[3];
726 		break;
727 	}
728 }
729 #endif /* INET6 */
730 
731 static void
732 pf_init_threshold(struct pf_threshold *threshold,
733     u_int32_t limit, u_int32_t seconds)
734 {
735 	threshold->limit = limit * PF_THRESHOLD_MULT;
736 	threshold->seconds = seconds;
737 	threshold->count = 0;
738 	threshold->last = time_uptime;
739 }
740 
741 static void
742 pf_add_threshold(struct pf_threshold *threshold)
743 {
744 	u_int32_t t = time_uptime, diff = t - threshold->last;
745 
746 	if (diff >= threshold->seconds)
747 		threshold->count = 0;
748 	else
749 		threshold->count -= threshold->count * diff /
750 		    threshold->seconds;
751 	threshold->count += PF_THRESHOLD_MULT;
752 	threshold->last = t;
753 }
754 
755 static int
756 pf_check_threshold(struct pf_threshold *threshold)
757 {
758 	return (threshold->count > threshold->limit);
759 }
760 
761 static int
762 pf_src_connlimit(struct pf_kstate **state)
763 {
764 	struct pf_overload_entry *pfoe;
765 	int bad = 0;
766 
767 	PF_STATE_LOCK_ASSERT(*state);
768 	/*
769 	 * XXXKS: The src node is accessed unlocked!
770 	 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node);
771 	 */
772 
773 	(*state)->src_node->conn++;
774 	(*state)->src.tcp_est = 1;
775 	pf_add_threshold(&(*state)->src_node->conn_rate);
776 
777 	if ((*state)->rule.ptr->max_src_conn &&
778 	    (*state)->rule.ptr->max_src_conn <
779 	    (*state)->src_node->conn) {
780 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
781 		bad++;
782 	}
783 
784 	if ((*state)->rule.ptr->max_src_conn_rate.limit &&
785 	    pf_check_threshold(&(*state)->src_node->conn_rate)) {
786 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
787 		bad++;
788 	}
789 
790 	if (!bad)
791 		return (0);
792 
793 	/* Kill this state. */
794 	(*state)->timeout = PFTM_PURGE;
795 	pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED);
796 
797 	if ((*state)->rule.ptr->overload_tbl == NULL)
798 		return (1);
799 
800 	/* Schedule overloading and flushing task. */
801 	pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
802 	if (pfoe == NULL)
803 		return (1);	/* too bad :( */
804 
805 	bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
806 	pfoe->af = (*state)->key[PF_SK_WIRE]->af;
807 	pfoe->rule = (*state)->rule.ptr;
808 	pfoe->dir = (*state)->direction;
809 	PF_OVERLOADQ_LOCK();
810 	SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
811 	PF_OVERLOADQ_UNLOCK();
812 	taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
813 
814 	return (1);
815 }
816 
817 static void
818 pf_overload_task(void *v, int pending)
819 {
820 	struct pf_overload_head queue;
821 	struct pfr_addr p;
822 	struct pf_overload_entry *pfoe, *pfoe1;
823 	uint32_t killed = 0;
824 
825 	CURVNET_SET((struct vnet *)v);
826 
827 	PF_OVERLOADQ_LOCK();
828 	queue = V_pf_overloadqueue;
829 	SLIST_INIT(&V_pf_overloadqueue);
830 	PF_OVERLOADQ_UNLOCK();
831 
832 	bzero(&p, sizeof(p));
833 	SLIST_FOREACH(pfoe, &queue, next) {
834 		counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
835 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
836 			printf("%s: blocking address ", __func__);
837 			pf_print_host(&pfoe->addr, 0, pfoe->af);
838 			printf("\n");
839 		}
840 
841 		p.pfra_af = pfoe->af;
842 		switch (pfoe->af) {
843 #ifdef INET
844 		case AF_INET:
845 			p.pfra_net = 32;
846 			p.pfra_ip4addr = pfoe->addr.v4;
847 			break;
848 #endif
849 #ifdef INET6
850 		case AF_INET6:
851 			p.pfra_net = 128;
852 			p.pfra_ip6addr = pfoe->addr.v6;
853 			break;
854 #endif
855 		}
856 
857 		PF_RULES_WLOCK();
858 		pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
859 		PF_RULES_WUNLOCK();
860 	}
861 
862 	/*
863 	 * Remove those entries, that don't need flushing.
864 	 */
865 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
866 		if (pfoe->rule->flush == 0) {
867 			SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
868 			free(pfoe, M_PFTEMP);
869 		} else
870 			counter_u64_add(
871 			    V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
872 
873 	/* If nothing to flush, return. */
874 	if (SLIST_EMPTY(&queue)) {
875 		CURVNET_RESTORE();
876 		return;
877 	}
878 
879 	for (int i = 0; i <= pf_hashmask; i++) {
880 		struct pf_idhash *ih = &V_pf_idhash[i];
881 		struct pf_state_key *sk;
882 		struct pf_kstate *s;
883 
884 		PF_HASHROW_LOCK(ih);
885 		LIST_FOREACH(s, &ih->states, entry) {
886 		    sk = s->key[PF_SK_WIRE];
887 		    SLIST_FOREACH(pfoe, &queue, next)
888 			if (sk->af == pfoe->af &&
889 			    ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
890 			    pfoe->rule == s->rule.ptr) &&
891 			    ((pfoe->dir == PF_OUT &&
892 			    PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
893 			    (pfoe->dir == PF_IN &&
894 			    PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
895 				s->timeout = PFTM_PURGE;
896 				pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED);
897 				killed++;
898 			}
899 		}
900 		PF_HASHROW_UNLOCK(ih);
901 	}
902 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
903 		free(pfoe, M_PFTEMP);
904 	if (V_pf_status.debug >= PF_DEBUG_MISC)
905 		printf("%s: %u states killed", __func__, killed);
906 
907 	CURVNET_RESTORE();
908 }
909 
910 /*
911  * Can return locked on failure, so that we can consistently
912  * allocate and insert a new one.
913  */
914 struct pf_ksrc_node *
915 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af,
916 	struct pf_srchash **sh, bool returnlocked)
917 {
918 	struct pf_ksrc_node *n;
919 
920 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
921 
922 	*sh = &V_pf_srchash[pf_hashsrc(src, af)];
923 	PF_HASHROW_LOCK(*sh);
924 	LIST_FOREACH(n, &(*sh)->nodes, entry)
925 		if (n->rule.ptr == rule && n->af == af &&
926 		    ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
927 		    (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
928 			break;
929 
930 	if (n != NULL) {
931 		n->states++;
932 		PF_HASHROW_UNLOCK(*sh);
933 	} else if (returnlocked == false)
934 		PF_HASHROW_UNLOCK(*sh);
935 
936 	return (n);
937 }
938 
939 static void
940 pf_free_src_node(struct pf_ksrc_node *sn)
941 {
942 
943 	for (int i = 0; i < 2; i++) {
944 		counter_u64_free(sn->bytes[i]);
945 		counter_u64_free(sn->packets[i]);
946 	}
947 	uma_zfree(V_pf_sources_z, sn);
948 }
949 
950 static u_short
951 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule,
952     struct pf_addr *src, sa_family_t af)
953 {
954 	u_short			 reason = 0;
955 	struct pf_srchash	*sh = NULL;
956 
957 	KASSERT((rule->rule_flag & PFRULE_SRCTRACK ||
958 	    rule->rpool.opts & PF_POOL_STICKYADDR),
959 	    ("%s for non-tracking rule %p", __func__, rule));
960 
961 	if (*sn == NULL)
962 		*sn = pf_find_src_node(src, rule, af, &sh, true);
963 
964 	if (*sn == NULL) {
965 		PF_HASHROW_ASSERT(sh);
966 
967 		if (rule->max_src_nodes &&
968 		    counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) {
969 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1);
970 			PF_HASHROW_UNLOCK(sh);
971 			reason = PFRES_SRCLIMIT;
972 			goto done;
973 		}
974 
975 		(*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
976 		if ((*sn) == NULL) {
977 			PF_HASHROW_UNLOCK(sh);
978 			reason = PFRES_MEMORY;
979 			goto done;
980 		}
981 
982 		for (int i = 0; i < 2; i++) {
983 			(*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT);
984 			(*sn)->packets[i] = counter_u64_alloc(M_NOWAIT);
985 
986 			if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) {
987 				pf_free_src_node(*sn);
988 				PF_HASHROW_UNLOCK(sh);
989 				reason = PFRES_MEMORY;
990 				goto done;
991 			}
992 		}
993 
994 		pf_init_threshold(&(*sn)->conn_rate,
995 		    rule->max_src_conn_rate.limit,
996 		    rule->max_src_conn_rate.seconds);
997 
998 		MPASS((*sn)->lock == NULL);
999 		(*sn)->lock = &sh->lock;
1000 
1001 		(*sn)->af = af;
1002 		(*sn)->rule.ptr = rule;
1003 		PF_ACPY(&(*sn)->addr, src, af);
1004 		LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
1005 		(*sn)->creation = time_uptime;
1006 		(*sn)->ruletype = rule->action;
1007 		(*sn)->states = 1;
1008 		if ((*sn)->rule.ptr != NULL)
1009 			counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
1010 		PF_HASHROW_UNLOCK(sh);
1011 		counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
1012 	} else {
1013 		if (rule->max_src_states &&
1014 		    (*sn)->states >= rule->max_src_states) {
1015 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
1016 			    1);
1017 			reason = PFRES_SRCLIMIT;
1018 			goto done;
1019 		}
1020 	}
1021 done:
1022 	return (reason);
1023 }
1024 
1025 void
1026 pf_unlink_src_node(struct pf_ksrc_node *src)
1027 {
1028 	PF_SRC_NODE_LOCK_ASSERT(src);
1029 
1030 	LIST_REMOVE(src, entry);
1031 	if (src->rule.ptr)
1032 		counter_u64_add(src->rule.ptr->src_nodes, -1);
1033 }
1034 
1035 u_int
1036 pf_free_src_nodes(struct pf_ksrc_node_list *head)
1037 {
1038 	struct pf_ksrc_node *sn, *tmp;
1039 	u_int count = 0;
1040 
1041 	LIST_FOREACH_SAFE(sn, head, entry, tmp) {
1042 		pf_free_src_node(sn);
1043 		count++;
1044 	}
1045 
1046 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
1047 
1048 	return (count);
1049 }
1050 
1051 void
1052 pf_mtag_initialize(void)
1053 {
1054 
1055 	pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
1056 	    sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
1057 	    UMA_ALIGN_PTR, 0);
1058 }
1059 
1060 /* Per-vnet data storage structures initialization. */
1061 void
1062 pf_initialize(void)
1063 {
1064 	struct pf_keyhash	*kh;
1065 	struct pf_idhash	*ih;
1066 	struct pf_srchash	*sh;
1067 	u_int i;
1068 
1069 	if (pf_hashsize == 0 || !powerof2(pf_hashsize))
1070 		pf_hashsize = PF_HASHSIZ;
1071 	if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
1072 		pf_srchashsize = PF_SRCHASHSIZ;
1073 
1074 	V_pf_hashseed = arc4random();
1075 
1076 	/* States and state keys storage. */
1077 	V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate),
1078 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1079 	V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
1080 	uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
1081 	uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
1082 
1083 	V_pf_state_key_z = uma_zcreate("pf state keys",
1084 	    sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
1085 	    UMA_ALIGN_PTR, 0);
1086 
1087 	V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash),
1088 	    M_PFHASH, M_NOWAIT | M_ZERO);
1089 	V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash),
1090 	    M_PFHASH, M_NOWAIT | M_ZERO);
1091 	if (V_pf_keyhash == NULL || V_pf_idhash == NULL) {
1092 		printf("pf: Unable to allocate memory for "
1093 		    "state_hashsize %lu.\n", pf_hashsize);
1094 
1095 		free(V_pf_keyhash, M_PFHASH);
1096 		free(V_pf_idhash, M_PFHASH);
1097 
1098 		pf_hashsize = PF_HASHSIZ;
1099 		V_pf_keyhash = mallocarray(pf_hashsize,
1100 		    sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO);
1101 		V_pf_idhash = mallocarray(pf_hashsize,
1102 		    sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO);
1103 	}
1104 
1105 	pf_hashmask = pf_hashsize - 1;
1106 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
1107 	    i++, kh++, ih++) {
1108 		mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
1109 		mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
1110 	}
1111 
1112 	/* Source nodes. */
1113 	V_pf_sources_z = uma_zcreate("pf source nodes",
1114 	    sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
1115 	    0);
1116 	V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
1117 	uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
1118 	uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
1119 
1120 	V_pf_srchash = mallocarray(pf_srchashsize,
1121 	    sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO);
1122 	if (V_pf_srchash == NULL) {
1123 		printf("pf: Unable to allocate memory for "
1124 		    "source_hashsize %lu.\n", pf_srchashsize);
1125 
1126 		pf_srchashsize = PF_SRCHASHSIZ;
1127 		V_pf_srchash = mallocarray(pf_srchashsize,
1128 		    sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO);
1129 	}
1130 
1131 	pf_srchashmask = pf_srchashsize - 1;
1132 	for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
1133 		mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
1134 
1135 	/* ALTQ */
1136 	TAILQ_INIT(&V_pf_altqs[0]);
1137 	TAILQ_INIT(&V_pf_altqs[1]);
1138 	TAILQ_INIT(&V_pf_altqs[2]);
1139 	TAILQ_INIT(&V_pf_altqs[3]);
1140 	TAILQ_INIT(&V_pf_pabuf);
1141 	V_pf_altqs_active = &V_pf_altqs[0];
1142 	V_pf_altq_ifs_active = &V_pf_altqs[1];
1143 	V_pf_altqs_inactive = &V_pf_altqs[2];
1144 	V_pf_altq_ifs_inactive = &V_pf_altqs[3];
1145 
1146 	/* Send & overload+flush queues. */
1147 	STAILQ_INIT(&V_pf_sendqueue);
1148 	SLIST_INIT(&V_pf_overloadqueue);
1149 	TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
1150 
1151 	/* Unlinked, but may be referenced rules. */
1152 	TAILQ_INIT(&V_pf_unlinked_rules);
1153 }
1154 
1155 void
1156 pf_mtag_cleanup(void)
1157 {
1158 
1159 	uma_zdestroy(pf_mtag_z);
1160 }
1161 
1162 void
1163 pf_cleanup(void)
1164 {
1165 	struct pf_keyhash	*kh;
1166 	struct pf_idhash	*ih;
1167 	struct pf_srchash	*sh;
1168 	struct pf_send_entry	*pfse, *next;
1169 	u_int i;
1170 
1171 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
1172 	    i++, kh++, ih++) {
1173 		KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
1174 		    __func__));
1175 		KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
1176 		    __func__));
1177 		mtx_destroy(&kh->lock);
1178 		mtx_destroy(&ih->lock);
1179 	}
1180 	free(V_pf_keyhash, M_PFHASH);
1181 	free(V_pf_idhash, M_PFHASH);
1182 
1183 	for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1184 		KASSERT(LIST_EMPTY(&sh->nodes),
1185 		    ("%s: source node hash not empty", __func__));
1186 		mtx_destroy(&sh->lock);
1187 	}
1188 	free(V_pf_srchash, M_PFHASH);
1189 
1190 	STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
1191 		m_freem(pfse->pfse_m);
1192 		free(pfse, M_PFTEMP);
1193 	}
1194 	MPASS(RB_EMPTY(&V_pf_sctp_endpoints));
1195 
1196 	uma_zdestroy(V_pf_sources_z);
1197 	uma_zdestroy(V_pf_state_z);
1198 	uma_zdestroy(V_pf_state_key_z);
1199 }
1200 
1201 static int
1202 pf_mtag_uminit(void *mem, int size, int how)
1203 {
1204 	struct m_tag *t;
1205 
1206 	t = (struct m_tag *)mem;
1207 	t->m_tag_cookie = MTAG_ABI_COMPAT;
1208 	t->m_tag_id = PACKET_TAG_PF;
1209 	t->m_tag_len = sizeof(struct pf_mtag);
1210 	t->m_tag_free = pf_mtag_free;
1211 
1212 	return (0);
1213 }
1214 
1215 static void
1216 pf_mtag_free(struct m_tag *t)
1217 {
1218 
1219 	uma_zfree(pf_mtag_z, t);
1220 }
1221 
1222 struct pf_mtag *
1223 pf_get_mtag(struct mbuf *m)
1224 {
1225 	struct m_tag *mtag;
1226 
1227 	if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
1228 		return ((struct pf_mtag *)(mtag + 1));
1229 
1230 	mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
1231 	if (mtag == NULL)
1232 		return (NULL);
1233 	bzero(mtag + 1, sizeof(struct pf_mtag));
1234 	m_tag_prepend(m, mtag);
1235 
1236 	return ((struct pf_mtag *)(mtag + 1));
1237 }
1238 
1239 static int
1240 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
1241     struct pf_kstate *s)
1242 {
1243 	struct pf_keyhash	*khs, *khw, *kh;
1244 	struct pf_state_key	*sk, *cur;
1245 	struct pf_kstate	*si, *olds = NULL;
1246 	int idx;
1247 
1248 	NET_EPOCH_ASSERT();
1249 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1250 	KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
1251 	KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
1252 
1253 	/*
1254 	 * We need to lock hash slots of both keys. To avoid deadlock
1255 	 * we always lock the slot with lower address first. Unlock order
1256 	 * isn't important.
1257 	 *
1258 	 * We also need to lock ID hash slot before dropping key
1259 	 * locks. On success we return with ID hash slot locked.
1260 	 */
1261 
1262 	if (skw == sks) {
1263 		khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
1264 		PF_HASHROW_LOCK(khs);
1265 	} else {
1266 		khs = &V_pf_keyhash[pf_hashkey(sks)];
1267 		khw = &V_pf_keyhash[pf_hashkey(skw)];
1268 		if (khs == khw) {
1269 			PF_HASHROW_LOCK(khs);
1270 		} else if (khs < khw) {
1271 			PF_HASHROW_LOCK(khs);
1272 			PF_HASHROW_LOCK(khw);
1273 		} else {
1274 			PF_HASHROW_LOCK(khw);
1275 			PF_HASHROW_LOCK(khs);
1276 		}
1277 	}
1278 
1279 #define	KEYS_UNLOCK()	do {			\
1280 	if (khs != khw) {			\
1281 		PF_HASHROW_UNLOCK(khs);		\
1282 		PF_HASHROW_UNLOCK(khw);		\
1283 	} else					\
1284 		PF_HASHROW_UNLOCK(khs);		\
1285 } while (0)
1286 
1287 	/*
1288 	 * First run: start with wire key.
1289 	 */
1290 	sk = skw;
1291 	kh = khw;
1292 	idx = PF_SK_WIRE;
1293 
1294 	MPASS(s->lock == NULL);
1295 	s->lock = &V_pf_idhash[PF_IDHASH(s)].lock;
1296 
1297 keyattach:
1298 	LIST_FOREACH(cur, &kh->keys, entry)
1299 		if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
1300 			break;
1301 
1302 	if (cur != NULL) {
1303 		/* Key exists. Check for same kif, if none, add to key. */
1304 		TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
1305 			struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
1306 
1307 			PF_HASHROW_LOCK(ih);
1308 			if (si->kif == s->kif &&
1309 			    si->direction == s->direction) {
1310 				if (sk->proto == IPPROTO_TCP &&
1311 				    si->src.state >= TCPS_FIN_WAIT_2 &&
1312 				    si->dst.state >= TCPS_FIN_WAIT_2) {
1313 					/*
1314 					 * New state matches an old >FIN_WAIT_2
1315 					 * state. We can't drop key hash locks,
1316 					 * thus we can't unlink it properly.
1317 					 *
1318 					 * As a workaround we drop it into
1319 					 * TCPS_CLOSED state, schedule purge
1320 					 * ASAP and push it into the very end
1321 					 * of the slot TAILQ, so that it won't
1322 					 * conflict with our new state.
1323 					 */
1324 					pf_set_protostate(si, PF_PEER_BOTH,
1325 					    TCPS_CLOSED);
1326 					si->timeout = PFTM_PURGE;
1327 					olds = si;
1328 				} else {
1329 					if (V_pf_status.debug >= PF_DEBUG_MISC) {
1330 						printf("pf: %s key attach "
1331 						    "failed on %s: ",
1332 						    (idx == PF_SK_WIRE) ?
1333 						    "wire" : "stack",
1334 						    s->kif->pfik_name);
1335 						pf_print_state_parts(s,
1336 						    (idx == PF_SK_WIRE) ?
1337 						    sk : NULL,
1338 						    (idx == PF_SK_STACK) ?
1339 						    sk : NULL);
1340 						printf(", existing: ");
1341 						pf_print_state_parts(si,
1342 						    (idx == PF_SK_WIRE) ?
1343 						    sk : NULL,
1344 						    (idx == PF_SK_STACK) ?
1345 						    sk : NULL);
1346 						printf("\n");
1347 					}
1348 					PF_HASHROW_UNLOCK(ih);
1349 					KEYS_UNLOCK();
1350 					uma_zfree(V_pf_state_key_z, sk);
1351 					if (idx == PF_SK_STACK)
1352 						pf_detach_state(s);
1353 					return (EEXIST); /* collision! */
1354 				}
1355 			}
1356 			PF_HASHROW_UNLOCK(ih);
1357 		}
1358 		uma_zfree(V_pf_state_key_z, sk);
1359 		s->key[idx] = cur;
1360 	} else {
1361 		LIST_INSERT_HEAD(&kh->keys, sk, entry);
1362 		s->key[idx] = sk;
1363 	}
1364 
1365 stateattach:
1366 	/* List is sorted, if-bound states before floating. */
1367 	if (s->kif == V_pfi_all)
1368 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1369 	else
1370 		TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1371 
1372 	if (olds) {
1373 		TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1374 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1375 		    key_list[idx]);
1376 		olds = NULL;
1377 	}
1378 
1379 	/*
1380 	 * Attach done. See how should we (or should not?)
1381 	 * attach a second key.
1382 	 */
1383 	if (sks == skw) {
1384 		s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1385 		idx = PF_SK_STACK;
1386 		sks = NULL;
1387 		goto stateattach;
1388 	} else if (sks != NULL) {
1389 		/*
1390 		 * Continue attaching with stack key.
1391 		 */
1392 		sk = sks;
1393 		kh = khs;
1394 		idx = PF_SK_STACK;
1395 		sks = NULL;
1396 		goto keyattach;
1397 	}
1398 
1399 	PF_STATE_LOCK(s);
1400 	KEYS_UNLOCK();
1401 
1402 	KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1403 	    ("%s failure", __func__));
1404 
1405 	return (0);
1406 #undef	KEYS_UNLOCK
1407 }
1408 
1409 static void
1410 pf_detach_state(struct pf_kstate *s)
1411 {
1412 	struct pf_state_key *sks = s->key[PF_SK_STACK];
1413 	struct pf_keyhash *kh;
1414 
1415 	NET_EPOCH_ASSERT();
1416 
1417 	pf_sctp_multihome_detach_addr(s);
1418 
1419 	if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr)
1420 		V_pflow_export_state_ptr(s);
1421 
1422 	if (sks != NULL) {
1423 		kh = &V_pf_keyhash[pf_hashkey(sks)];
1424 		PF_HASHROW_LOCK(kh);
1425 		if (s->key[PF_SK_STACK] != NULL)
1426 			pf_state_key_detach(s, PF_SK_STACK);
1427 		/*
1428 		 * If both point to same key, then we are done.
1429 		 */
1430 		if (sks == s->key[PF_SK_WIRE]) {
1431 			pf_state_key_detach(s, PF_SK_WIRE);
1432 			PF_HASHROW_UNLOCK(kh);
1433 			return;
1434 		}
1435 		PF_HASHROW_UNLOCK(kh);
1436 	}
1437 
1438 	if (s->key[PF_SK_WIRE] != NULL) {
1439 		kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1440 		PF_HASHROW_LOCK(kh);
1441 		if (s->key[PF_SK_WIRE] != NULL)
1442 			pf_state_key_detach(s, PF_SK_WIRE);
1443 		PF_HASHROW_UNLOCK(kh);
1444 	}
1445 }
1446 
1447 static void
1448 pf_state_key_detach(struct pf_kstate *s, int idx)
1449 {
1450 	struct pf_state_key *sk = s->key[idx];
1451 #ifdef INVARIANTS
1452 	struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1453 
1454 	PF_HASHROW_ASSERT(kh);
1455 #endif
1456 	TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1457 	s->key[idx] = NULL;
1458 
1459 	if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1460 		LIST_REMOVE(sk, entry);
1461 		uma_zfree(V_pf_state_key_z, sk);
1462 	}
1463 }
1464 
1465 static int
1466 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1467 {
1468 	struct pf_state_key *sk = mem;
1469 
1470 	bzero(sk, sizeof(struct pf_state_key_cmp));
1471 	TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1472 	TAILQ_INIT(&sk->states[PF_SK_STACK]);
1473 
1474 	return (0);
1475 }
1476 
1477 struct pf_state_key *
1478 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1479 	struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1480 {
1481 	struct pf_state_key *sk;
1482 
1483 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1484 	if (sk == NULL)
1485 		return (NULL);
1486 
1487 	PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1488 	PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1489 	sk->port[pd->sidx] = sport;
1490 	sk->port[pd->didx] = dport;
1491 	sk->proto = pd->proto;
1492 	sk->af = pd->af;
1493 
1494 	return (sk);
1495 }
1496 
1497 struct pf_state_key *
1498 pf_state_key_clone(struct pf_state_key *orig)
1499 {
1500 	struct pf_state_key *sk;
1501 
1502 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1503 	if (sk == NULL)
1504 		return (NULL);
1505 
1506 	bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1507 
1508 	return (sk);
1509 }
1510 
1511 int
1512 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif,
1513     struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s)
1514 {
1515 	struct pf_idhash *ih;
1516 	struct pf_kstate *cur;
1517 	int error;
1518 
1519 	NET_EPOCH_ASSERT();
1520 
1521 	KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1522 	    ("%s: sks not pristine", __func__));
1523 	KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1524 	    ("%s: skw not pristine", __func__));
1525 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1526 
1527 	s->kif = kif;
1528 	s->orig_kif = orig_kif;
1529 
1530 	if (s->id == 0 && s->creatorid == 0) {
1531 		s->id = alloc_unr64(&V_pf_stateid);
1532 		s->id = htobe64(s->id);
1533 		s->creatorid = V_pf_status.hostid;
1534 	}
1535 
1536 	/* Returns with ID locked on success. */
1537 	if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1538 		return (error);
1539 
1540 	ih = &V_pf_idhash[PF_IDHASH(s)];
1541 	PF_HASHROW_ASSERT(ih);
1542 	LIST_FOREACH(cur, &ih->states, entry)
1543 		if (cur->id == s->id && cur->creatorid == s->creatorid)
1544 			break;
1545 
1546 	if (cur != NULL) {
1547 		PF_HASHROW_UNLOCK(ih);
1548 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1549 			printf("pf: state ID collision: "
1550 			    "id: %016llx creatorid: %08x\n",
1551 			    (unsigned long long)be64toh(s->id),
1552 			    ntohl(s->creatorid));
1553 		}
1554 		pf_detach_state(s);
1555 		return (EEXIST);
1556 	}
1557 	LIST_INSERT_HEAD(&ih->states, s, entry);
1558 	/* One for keys, one for ID hash. */
1559 	refcount_init(&s->refs, 2);
1560 
1561 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1562 	if (V_pfsync_insert_state_ptr != NULL)
1563 		V_pfsync_insert_state_ptr(s);
1564 
1565 	/* Returns locked. */
1566 	return (0);
1567 }
1568 
1569 /*
1570  * Find state by ID: returns with locked row on success.
1571  */
1572 struct pf_kstate *
1573 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1574 {
1575 	struct pf_idhash *ih;
1576 	struct pf_kstate *s;
1577 
1578 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1579 
1580 	ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1581 
1582 	PF_HASHROW_LOCK(ih);
1583 	LIST_FOREACH(s, &ih->states, entry)
1584 		if (s->id == id && s->creatorid == creatorid)
1585 			break;
1586 
1587 	if (s == NULL)
1588 		PF_HASHROW_UNLOCK(ih);
1589 
1590 	return (s);
1591 }
1592 
1593 /*
1594  * Find state by key.
1595  * Returns with ID hash slot locked on success.
1596  */
1597 static struct pf_kstate *
1598 pf_find_state(struct pfi_kkif *kif, struct pf_state_key_cmp *key, u_int dir)
1599 {
1600 	struct pf_keyhash	*kh;
1601 	struct pf_state_key	*sk;
1602 	struct pf_kstate	*s;
1603 	int idx;
1604 
1605 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1606 
1607 	kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1608 
1609 	PF_HASHROW_LOCK(kh);
1610 	LIST_FOREACH(sk, &kh->keys, entry)
1611 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1612 			break;
1613 	if (sk == NULL) {
1614 		PF_HASHROW_UNLOCK(kh);
1615 		return (NULL);
1616 	}
1617 
1618 	idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1619 
1620 	/* List is sorted, if-bound states before floating ones. */
1621 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1622 		if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) {
1623 			PF_STATE_LOCK(s);
1624 			PF_HASHROW_UNLOCK(kh);
1625 			if (__predict_false(s->timeout >= PFTM_MAX)) {
1626 				/*
1627 				 * State is either being processed by
1628 				 * pf_unlink_state() in an other thread, or
1629 				 * is scheduled for immediate expiry.
1630 				 */
1631 				PF_STATE_UNLOCK(s);
1632 				return (NULL);
1633 			}
1634 			return (s);
1635 		}
1636 	PF_HASHROW_UNLOCK(kh);
1637 
1638 	return (NULL);
1639 }
1640 
1641 /*
1642  * Returns with ID hash slot locked on success.
1643  */
1644 struct pf_kstate *
1645 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1646 {
1647 	struct pf_keyhash	*kh;
1648 	struct pf_state_key	*sk;
1649 	struct pf_kstate	*s, *ret = NULL;
1650 	int			 idx, inout = 0;
1651 
1652 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1653 
1654 	kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1655 
1656 	PF_HASHROW_LOCK(kh);
1657 	LIST_FOREACH(sk, &kh->keys, entry)
1658 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1659 			break;
1660 	if (sk == NULL) {
1661 		PF_HASHROW_UNLOCK(kh);
1662 		return (NULL);
1663 	}
1664 	switch (dir) {
1665 	case PF_IN:
1666 		idx = PF_SK_WIRE;
1667 		break;
1668 	case PF_OUT:
1669 		idx = PF_SK_STACK;
1670 		break;
1671 	case PF_INOUT:
1672 		idx = PF_SK_WIRE;
1673 		inout = 1;
1674 		break;
1675 	default:
1676 		panic("%s: dir %u", __func__, dir);
1677 	}
1678 second_run:
1679 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1680 		if (more == NULL) {
1681 			PF_STATE_LOCK(s);
1682 			PF_HASHROW_UNLOCK(kh);
1683 			return (s);
1684 		}
1685 
1686 		if (ret)
1687 			(*more)++;
1688 		else {
1689 			ret = s;
1690 			PF_STATE_LOCK(s);
1691 		}
1692 	}
1693 	if (inout == 1) {
1694 		inout = 0;
1695 		idx = PF_SK_STACK;
1696 		goto second_run;
1697 	}
1698 	PF_HASHROW_UNLOCK(kh);
1699 
1700 	return (ret);
1701 }
1702 
1703 /*
1704  * FIXME
1705  * This routine is inefficient -- locks the state only to unlock immediately on
1706  * return.
1707  * It is racy -- after the state is unlocked nothing stops other threads from
1708  * removing it.
1709  */
1710 bool
1711 pf_find_state_all_exists(struct pf_state_key_cmp *key, u_int dir)
1712 {
1713 	struct pf_kstate *s;
1714 
1715 	s = pf_find_state_all(key, dir, NULL);
1716 	if (s != NULL) {
1717 		PF_STATE_UNLOCK(s);
1718 		return (true);
1719 	}
1720 	return (false);
1721 }
1722 
1723 /* END state table stuff */
1724 
1725 static void
1726 pf_send(struct pf_send_entry *pfse)
1727 {
1728 
1729 	PF_SENDQ_LOCK();
1730 	STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1731 	PF_SENDQ_UNLOCK();
1732 	swi_sched(V_pf_swi_cookie, 0);
1733 }
1734 
1735 static bool
1736 pf_isforlocal(struct mbuf *m, int af)
1737 {
1738 	switch (af) {
1739 #ifdef INET
1740 	case AF_INET: {
1741 		struct ip *ip = mtod(m, struct ip *);
1742 
1743 		return (in_localip(ip->ip_dst));
1744 	}
1745 #endif
1746 #ifdef INET6
1747 	case AF_INET6: {
1748 		struct ip6_hdr *ip6;
1749 		struct in6_ifaddr *ia;
1750 		ip6 = mtod(m, struct ip6_hdr *);
1751 		ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1752 		if (ia == NULL)
1753 			return (false);
1754 		return (! (ia->ia6_flags & IN6_IFF_NOTREADY));
1755 	}
1756 #endif
1757 	default:
1758 		panic("Unsupported af %d", af);
1759 	}
1760 
1761 	return (false);
1762 }
1763 
1764 void
1765 pf_intr(void *v)
1766 {
1767 	struct epoch_tracker et;
1768 	struct pf_send_head queue;
1769 	struct pf_send_entry *pfse, *next;
1770 
1771 	CURVNET_SET((struct vnet *)v);
1772 
1773 	PF_SENDQ_LOCK();
1774 	queue = V_pf_sendqueue;
1775 	STAILQ_INIT(&V_pf_sendqueue);
1776 	PF_SENDQ_UNLOCK();
1777 
1778 	NET_EPOCH_ENTER(et);
1779 
1780 	STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1781 		switch (pfse->pfse_type) {
1782 #ifdef INET
1783 		case PFSE_IP: {
1784 			if (pf_isforlocal(pfse->pfse_m, AF_INET)) {
1785 				pfse->pfse_m->m_flags |= M_SKIP_FIREWALL;
1786 				pfse->pfse_m->m_pkthdr.csum_flags |=
1787 				    CSUM_IP_VALID | CSUM_IP_CHECKED;
1788 				ip_input(pfse->pfse_m);
1789 			} else {
1790 				ip_output(pfse->pfse_m, NULL, NULL, 0, NULL,
1791 				    NULL);
1792 			}
1793 			break;
1794 		}
1795 		case PFSE_ICMP:
1796 			icmp_error(pfse->pfse_m, pfse->icmpopts.type,
1797 			    pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
1798 			break;
1799 #endif /* INET */
1800 #ifdef INET6
1801 		case PFSE_IP6:
1802 			if (pf_isforlocal(pfse->pfse_m, AF_INET6)) {
1803 				pfse->pfse_m->m_flags |= M_SKIP_FIREWALL;
1804 				ip6_input(pfse->pfse_m);
1805 			} else {
1806 				ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL,
1807 				    NULL, NULL);
1808 			}
1809 			break;
1810 		case PFSE_ICMP6:
1811 			icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
1812 			    pfse->icmpopts.code, pfse->icmpopts.mtu);
1813 			break;
1814 #endif /* INET6 */
1815 		default:
1816 			panic("%s: unknown type", __func__);
1817 		}
1818 		free(pfse, M_PFTEMP);
1819 	}
1820 	NET_EPOCH_EXIT(et);
1821 	CURVNET_RESTORE();
1822 }
1823 
1824 #define	pf_purge_thread_period	(hz / 10)
1825 
1826 #ifdef PF_WANT_32_TO_64_COUNTER
1827 static void
1828 pf_status_counter_u64_periodic(void)
1829 {
1830 
1831 	PF_RULES_RASSERT();
1832 
1833 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) {
1834 		return;
1835 	}
1836 
1837 	for (int i = 0; i < FCNT_MAX; i++) {
1838 		pf_counter_u64_periodic(&V_pf_status.fcounters[i]);
1839 	}
1840 }
1841 
1842 static void
1843 pf_kif_counter_u64_periodic(void)
1844 {
1845 	struct pfi_kkif *kif;
1846 	size_t r, run;
1847 
1848 	PF_RULES_RASSERT();
1849 
1850 	if (__predict_false(V_pf_allkifcount == 0)) {
1851 		return;
1852 	}
1853 
1854 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) {
1855 		return;
1856 	}
1857 
1858 	run = V_pf_allkifcount / 10;
1859 	if (run < 5)
1860 		run = 5;
1861 
1862 	for (r = 0; r < run; r++) {
1863 		kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist);
1864 		if (kif == NULL) {
1865 			LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist);
1866 			LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist);
1867 			break;
1868 		}
1869 
1870 		LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist);
1871 		LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist);
1872 
1873 		for (int i = 0; i < 2; i++) {
1874 			for (int j = 0; j < 2; j++) {
1875 				for (int k = 0; k < 2; k++) {
1876 					pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]);
1877 					pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]);
1878 				}
1879 			}
1880 		}
1881 	}
1882 }
1883 
1884 static void
1885 pf_rule_counter_u64_periodic(void)
1886 {
1887 	struct pf_krule *rule;
1888 	size_t r, run;
1889 
1890 	PF_RULES_RASSERT();
1891 
1892 	if (__predict_false(V_pf_allrulecount == 0)) {
1893 		return;
1894 	}
1895 
1896 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) {
1897 		return;
1898 	}
1899 
1900 	run = V_pf_allrulecount / 10;
1901 	if (run < 5)
1902 		run = 5;
1903 
1904 	for (r = 0; r < run; r++) {
1905 		rule = LIST_NEXT(V_pf_rulemarker, allrulelist);
1906 		if (rule == NULL) {
1907 			LIST_REMOVE(V_pf_rulemarker, allrulelist);
1908 			LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist);
1909 			break;
1910 		}
1911 
1912 		LIST_REMOVE(V_pf_rulemarker, allrulelist);
1913 		LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist);
1914 
1915 		pf_counter_u64_periodic(&rule->evaluations);
1916 		for (int i = 0; i < 2; i++) {
1917 			pf_counter_u64_periodic(&rule->packets[i]);
1918 			pf_counter_u64_periodic(&rule->bytes[i]);
1919 		}
1920 	}
1921 }
1922 
1923 static void
1924 pf_counter_u64_periodic_main(void)
1925 {
1926 	PF_RULES_RLOCK_TRACKER;
1927 
1928 	V_pf_counter_periodic_iter++;
1929 
1930 	PF_RULES_RLOCK();
1931 	pf_counter_u64_critical_enter();
1932 	pf_status_counter_u64_periodic();
1933 	pf_kif_counter_u64_periodic();
1934 	pf_rule_counter_u64_periodic();
1935 	pf_counter_u64_critical_exit();
1936 	PF_RULES_RUNLOCK();
1937 }
1938 #else
1939 #define	pf_counter_u64_periodic_main()	do { } while (0)
1940 #endif
1941 
1942 void
1943 pf_purge_thread(void *unused __unused)
1944 {
1945 	struct epoch_tracker	 et;
1946 
1947 	VNET_ITERATOR_DECL(vnet_iter);
1948 
1949 	sx_xlock(&pf_end_lock);
1950 	while (pf_end_threads == 0) {
1951 		sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period);
1952 
1953 		VNET_LIST_RLOCK();
1954 		NET_EPOCH_ENTER(et);
1955 		VNET_FOREACH(vnet_iter) {
1956 			CURVNET_SET(vnet_iter);
1957 
1958 			/* Wait until V_pf_default_rule is initialized. */
1959 			if (V_pf_vnet_active == 0) {
1960 				CURVNET_RESTORE();
1961 				continue;
1962 			}
1963 
1964 			pf_counter_u64_periodic_main();
1965 
1966 			/*
1967 			 *  Process 1/interval fraction of the state
1968 			 * table every run.
1969 			 */
1970 			V_pf_purge_idx =
1971 			    pf_purge_expired_states(V_pf_purge_idx, pf_hashmask /
1972 			    (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1973 
1974 			/*
1975 			 * Purge other expired types every
1976 			 * PFTM_INTERVAL seconds.
1977 			 */
1978 			if (V_pf_purge_idx == 0) {
1979 				/*
1980 				 * Order is important:
1981 				 * - states and src nodes reference rules
1982 				 * - states and rules reference kifs
1983 				 */
1984 				pf_purge_expired_fragments();
1985 				pf_purge_expired_src_nodes();
1986 				pf_purge_unlinked_rules();
1987 				pfi_kkif_purge();
1988 			}
1989 			CURVNET_RESTORE();
1990 		}
1991 		NET_EPOCH_EXIT(et);
1992 		VNET_LIST_RUNLOCK();
1993 	}
1994 
1995 	pf_end_threads++;
1996 	sx_xunlock(&pf_end_lock);
1997 	kproc_exit(0);
1998 }
1999 
2000 void
2001 pf_unload_vnet_purge(void)
2002 {
2003 
2004 	/*
2005 	 * To cleanse up all kifs and rules we need
2006 	 * two runs: first one clears reference flags,
2007 	 * then pf_purge_expired_states() doesn't
2008 	 * raise them, and then second run frees.
2009 	 */
2010 	pf_purge_unlinked_rules();
2011 	pfi_kkif_purge();
2012 
2013 	/*
2014 	 * Now purge everything.
2015 	 */
2016 	pf_purge_expired_states(0, pf_hashmask);
2017 	pf_purge_fragments(UINT_MAX);
2018 	pf_purge_expired_src_nodes();
2019 
2020 	/*
2021 	 * Now all kifs & rules should be unreferenced,
2022 	 * thus should be successfully freed.
2023 	 */
2024 	pf_purge_unlinked_rules();
2025 	pfi_kkif_purge();
2026 }
2027 
2028 u_int32_t
2029 pf_state_expires(const struct pf_kstate *state)
2030 {
2031 	u_int32_t	timeout;
2032 	u_int32_t	start;
2033 	u_int32_t	end;
2034 	u_int32_t	states;
2035 
2036 	/* handle all PFTM_* > PFTM_MAX here */
2037 	if (state->timeout == PFTM_PURGE)
2038 		return (time_uptime);
2039 	KASSERT(state->timeout != PFTM_UNLINKED,
2040 	    ("pf_state_expires: timeout == PFTM_UNLINKED"));
2041 	KASSERT((state->timeout < PFTM_MAX),
2042 	    ("pf_state_expires: timeout > PFTM_MAX"));
2043 	timeout = state->rule.ptr->timeout[state->timeout];
2044 	if (!timeout)
2045 		timeout = V_pf_default_rule.timeout[state->timeout];
2046 	start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
2047 	if (start && state->rule.ptr != &V_pf_default_rule) {
2048 		end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
2049 		states = counter_u64_fetch(state->rule.ptr->states_cur);
2050 	} else {
2051 		start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
2052 		end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
2053 		states = V_pf_status.states;
2054 	}
2055 	if (end && states > start && start < end) {
2056 		if (states < end) {
2057 			timeout = (u_int64_t)timeout * (end - states) /
2058 			    (end - start);
2059 			return ((state->expire / 1000) + timeout);
2060 		}
2061 		else
2062 			return (time_uptime);
2063 	}
2064 	return ((state->expire / 1000) + timeout);
2065 }
2066 
2067 void
2068 pf_purge_expired_src_nodes(void)
2069 {
2070 	struct pf_ksrc_node_list	 freelist;
2071 	struct pf_srchash	*sh;
2072 	struct pf_ksrc_node	*cur, *next;
2073 	int i;
2074 
2075 	LIST_INIT(&freelist);
2076 	for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
2077 	    PF_HASHROW_LOCK(sh);
2078 	    LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
2079 		if (cur->states == 0 && cur->expire <= time_uptime) {
2080 			pf_unlink_src_node(cur);
2081 			LIST_INSERT_HEAD(&freelist, cur, entry);
2082 		} else if (cur->rule.ptr != NULL)
2083 			cur->rule.ptr->rule_ref |= PFRULE_REFS;
2084 	    PF_HASHROW_UNLOCK(sh);
2085 	}
2086 
2087 	pf_free_src_nodes(&freelist);
2088 
2089 	V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
2090 }
2091 
2092 static void
2093 pf_src_tree_remove_state(struct pf_kstate *s)
2094 {
2095 	struct pf_ksrc_node *sn;
2096 	uint32_t timeout;
2097 
2098 	timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
2099 	    s->rule.ptr->timeout[PFTM_SRC_NODE] :
2100 	    V_pf_default_rule.timeout[PFTM_SRC_NODE];
2101 
2102 	if (s->src_node != NULL) {
2103 		sn = s->src_node;
2104 		PF_SRC_NODE_LOCK(sn);
2105 		if (s->src.tcp_est)
2106 			--sn->conn;
2107 		if (--sn->states == 0)
2108 			sn->expire = time_uptime + timeout;
2109 		PF_SRC_NODE_UNLOCK(sn);
2110 	}
2111 	if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
2112 		sn = s->nat_src_node;
2113 		PF_SRC_NODE_LOCK(sn);
2114 		if (--sn->states == 0)
2115 			sn->expire = time_uptime + timeout;
2116 		PF_SRC_NODE_UNLOCK(sn);
2117 	}
2118 	s->src_node = s->nat_src_node = NULL;
2119 }
2120 
2121 /*
2122  * Unlink and potentilly free a state. Function may be
2123  * called with ID hash row locked, but always returns
2124  * unlocked, since it needs to go through key hash locking.
2125  */
2126 int
2127 pf_unlink_state(struct pf_kstate *s)
2128 {
2129 	struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
2130 
2131 	NET_EPOCH_ASSERT();
2132 	PF_HASHROW_ASSERT(ih);
2133 
2134 	if (s->timeout == PFTM_UNLINKED) {
2135 		/*
2136 		 * State is being processed
2137 		 * by pf_unlink_state() in
2138 		 * an other thread.
2139 		 */
2140 		PF_HASHROW_UNLOCK(ih);
2141 		return (0);	/* XXXGL: undefined actually */
2142 	}
2143 
2144 	if (s->src.state == PF_TCPS_PROXY_DST) {
2145 		/* XXX wire key the right one? */
2146 		pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af,
2147 		    &s->key[PF_SK_WIRE]->addr[1],
2148 		    &s->key[PF_SK_WIRE]->addr[0],
2149 		    s->key[PF_SK_WIRE]->port[1],
2150 		    s->key[PF_SK_WIRE]->port[0],
2151 		    s->src.seqhi, s->src.seqlo + 1,
2152 		    TH_RST|TH_ACK, 0, 0, 0, true, s->tag, 0, s->act.rtableid);
2153 	}
2154 
2155 	LIST_REMOVE(s, entry);
2156 	pf_src_tree_remove_state(s);
2157 
2158 	if (V_pfsync_delete_state_ptr != NULL)
2159 		V_pfsync_delete_state_ptr(s);
2160 
2161 	STATE_DEC_COUNTERS(s);
2162 
2163 	s->timeout = PFTM_UNLINKED;
2164 
2165 	/* Ensure we remove it from the list of halfopen states, if needed. */
2166 	if (s->key[PF_SK_STACK] != NULL &&
2167 	    s->key[PF_SK_STACK]->proto == IPPROTO_TCP)
2168 		pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED);
2169 
2170 	PF_HASHROW_UNLOCK(ih);
2171 
2172 	pf_detach_state(s);
2173 	/* pf_state_insert() initialises refs to 2 */
2174 	return (pf_release_staten(s, 2));
2175 }
2176 
2177 struct pf_kstate *
2178 pf_alloc_state(int flags)
2179 {
2180 
2181 	return (uma_zalloc(V_pf_state_z, flags | M_ZERO));
2182 }
2183 
2184 void
2185 pf_free_state(struct pf_kstate *cur)
2186 {
2187 	struct pf_krule_item *ri;
2188 
2189 	KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
2190 	KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
2191 	    cur->timeout));
2192 
2193 	while ((ri = SLIST_FIRST(&cur->match_rules))) {
2194 		SLIST_REMOVE_HEAD(&cur->match_rules, entry);
2195 		free(ri, M_PF_RULE_ITEM);
2196 	}
2197 
2198 	pf_normalize_tcp_cleanup(cur);
2199 	uma_zfree(V_pf_state_z, cur);
2200 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
2201 }
2202 
2203 /*
2204  * Called only from pf_purge_thread(), thus serialized.
2205  */
2206 static u_int
2207 pf_purge_expired_states(u_int i, int maxcheck)
2208 {
2209 	struct pf_idhash *ih;
2210 	struct pf_kstate *s;
2211 	struct pf_krule_item *mrm;
2212 	size_t count __unused;
2213 
2214 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2215 
2216 	/*
2217 	 * Go through hash and unlink states that expire now.
2218 	 */
2219 	while (maxcheck > 0) {
2220 		count = 0;
2221 		ih = &V_pf_idhash[i];
2222 
2223 		/* only take the lock if we expect to do work */
2224 		if (!LIST_EMPTY(&ih->states)) {
2225 relock:
2226 			PF_HASHROW_LOCK(ih);
2227 			LIST_FOREACH(s, &ih->states, entry) {
2228 				if (pf_state_expires(s) <= time_uptime) {
2229 					V_pf_status.states -=
2230 					    pf_unlink_state(s);
2231 					goto relock;
2232 				}
2233 				s->rule.ptr->rule_ref |= PFRULE_REFS;
2234 				if (s->nat_rule.ptr != NULL)
2235 					s->nat_rule.ptr->rule_ref |= PFRULE_REFS;
2236 				if (s->anchor.ptr != NULL)
2237 					s->anchor.ptr->rule_ref |= PFRULE_REFS;
2238 				s->kif->pfik_flags |= PFI_IFLAG_REFS;
2239 				SLIST_FOREACH(mrm, &s->match_rules, entry)
2240 					mrm->r->rule_ref |= PFRULE_REFS;
2241 				if (s->rt_kif)
2242 					s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
2243 				count++;
2244 			}
2245 			PF_HASHROW_UNLOCK(ih);
2246 		}
2247 
2248 		SDT_PROBE2(pf, purge, state, rowcount, i, count);
2249 
2250 		/* Return when we hit end of hash. */
2251 		if (++i > pf_hashmask) {
2252 			V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2253 			return (0);
2254 		}
2255 
2256 		maxcheck--;
2257 	}
2258 
2259 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2260 
2261 	return (i);
2262 }
2263 
2264 static void
2265 pf_purge_unlinked_rules(void)
2266 {
2267 	struct pf_krulequeue tmpq;
2268 	struct pf_krule *r, *r1;
2269 
2270 	/*
2271 	 * If we have overloading task pending, then we'd
2272 	 * better skip purging this time. There is a tiny
2273 	 * probability that overloading task references
2274 	 * an already unlinked rule.
2275 	 */
2276 	PF_OVERLOADQ_LOCK();
2277 	if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
2278 		PF_OVERLOADQ_UNLOCK();
2279 		return;
2280 	}
2281 	PF_OVERLOADQ_UNLOCK();
2282 
2283 	/*
2284 	 * Do naive mark-and-sweep garbage collecting of old rules.
2285 	 * Reference flag is raised by pf_purge_expired_states()
2286 	 * and pf_purge_expired_src_nodes().
2287 	 *
2288 	 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
2289 	 * use a temporary queue.
2290 	 */
2291 	TAILQ_INIT(&tmpq);
2292 	PF_UNLNKDRULES_LOCK();
2293 	TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
2294 		if (!(r->rule_ref & PFRULE_REFS)) {
2295 			TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
2296 			TAILQ_INSERT_TAIL(&tmpq, r, entries);
2297 		} else
2298 			r->rule_ref &= ~PFRULE_REFS;
2299 	}
2300 	PF_UNLNKDRULES_UNLOCK();
2301 
2302 	if (!TAILQ_EMPTY(&tmpq)) {
2303 		PF_CONFIG_LOCK();
2304 		PF_RULES_WLOCK();
2305 		TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
2306 			TAILQ_REMOVE(&tmpq, r, entries);
2307 			pf_free_rule(r);
2308 		}
2309 		PF_RULES_WUNLOCK();
2310 		PF_CONFIG_UNLOCK();
2311 	}
2312 }
2313 
2314 void
2315 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
2316 {
2317 	switch (af) {
2318 #ifdef INET
2319 	case AF_INET: {
2320 		u_int32_t a = ntohl(addr->addr32[0]);
2321 		printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
2322 		    (a>>8)&255, a&255);
2323 		if (p) {
2324 			p = ntohs(p);
2325 			printf(":%u", p);
2326 		}
2327 		break;
2328 	}
2329 #endif /* INET */
2330 #ifdef INET6
2331 	case AF_INET6: {
2332 		u_int16_t b;
2333 		u_int8_t i, curstart, curend, maxstart, maxend;
2334 		curstart = curend = maxstart = maxend = 255;
2335 		for (i = 0; i < 8; i++) {
2336 			if (!addr->addr16[i]) {
2337 				if (curstart == 255)
2338 					curstart = i;
2339 				curend = i;
2340 			} else {
2341 				if ((curend - curstart) >
2342 				    (maxend - maxstart)) {
2343 					maxstart = curstart;
2344 					maxend = curend;
2345 				}
2346 				curstart = curend = 255;
2347 			}
2348 		}
2349 		if ((curend - curstart) >
2350 		    (maxend - maxstart)) {
2351 			maxstart = curstart;
2352 			maxend = curend;
2353 		}
2354 		for (i = 0; i < 8; i++) {
2355 			if (i >= maxstart && i <= maxend) {
2356 				if (i == 0)
2357 					printf(":");
2358 				if (i == maxend)
2359 					printf(":");
2360 			} else {
2361 				b = ntohs(addr->addr16[i]);
2362 				printf("%x", b);
2363 				if (i < 7)
2364 					printf(":");
2365 			}
2366 		}
2367 		if (p) {
2368 			p = ntohs(p);
2369 			printf("[%u]", p);
2370 		}
2371 		break;
2372 	}
2373 #endif /* INET6 */
2374 	}
2375 }
2376 
2377 void
2378 pf_print_state(struct pf_kstate *s)
2379 {
2380 	pf_print_state_parts(s, NULL, NULL);
2381 }
2382 
2383 static void
2384 pf_print_state_parts(struct pf_kstate *s,
2385     struct pf_state_key *skwp, struct pf_state_key *sksp)
2386 {
2387 	struct pf_state_key *skw, *sks;
2388 	u_int8_t proto, dir;
2389 
2390 	/* Do our best to fill these, but they're skipped if NULL */
2391 	skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
2392 	sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
2393 	proto = skw ? skw->proto : (sks ? sks->proto : 0);
2394 	dir = s ? s->direction : 0;
2395 
2396 	switch (proto) {
2397 	case IPPROTO_IPV4:
2398 		printf("IPv4");
2399 		break;
2400 	case IPPROTO_IPV6:
2401 		printf("IPv6");
2402 		break;
2403 	case IPPROTO_TCP:
2404 		printf("TCP");
2405 		break;
2406 	case IPPROTO_UDP:
2407 		printf("UDP");
2408 		break;
2409 	case IPPROTO_ICMP:
2410 		printf("ICMP");
2411 		break;
2412 	case IPPROTO_ICMPV6:
2413 		printf("ICMPv6");
2414 		break;
2415 	default:
2416 		printf("%u", proto);
2417 		break;
2418 	}
2419 	switch (dir) {
2420 	case PF_IN:
2421 		printf(" in");
2422 		break;
2423 	case PF_OUT:
2424 		printf(" out");
2425 		break;
2426 	}
2427 	if (skw) {
2428 		printf(" wire: ");
2429 		pf_print_host(&skw->addr[0], skw->port[0], skw->af);
2430 		printf(" ");
2431 		pf_print_host(&skw->addr[1], skw->port[1], skw->af);
2432 	}
2433 	if (sks) {
2434 		printf(" stack: ");
2435 		if (sks != skw) {
2436 			pf_print_host(&sks->addr[0], sks->port[0], sks->af);
2437 			printf(" ");
2438 			pf_print_host(&sks->addr[1], sks->port[1], sks->af);
2439 		} else
2440 			printf("-");
2441 	}
2442 	if (s) {
2443 		if (proto == IPPROTO_TCP) {
2444 			printf(" [lo=%u high=%u win=%u modulator=%u",
2445 			    s->src.seqlo, s->src.seqhi,
2446 			    s->src.max_win, s->src.seqdiff);
2447 			if (s->src.wscale && s->dst.wscale)
2448 				printf(" wscale=%u",
2449 				    s->src.wscale & PF_WSCALE_MASK);
2450 			printf("]");
2451 			printf(" [lo=%u high=%u win=%u modulator=%u",
2452 			    s->dst.seqlo, s->dst.seqhi,
2453 			    s->dst.max_win, s->dst.seqdiff);
2454 			if (s->src.wscale && s->dst.wscale)
2455 				printf(" wscale=%u",
2456 				s->dst.wscale & PF_WSCALE_MASK);
2457 			printf("]");
2458 		}
2459 		printf(" %u:%u", s->src.state, s->dst.state);
2460 	}
2461 }
2462 
2463 void
2464 pf_print_flags(u_int8_t f)
2465 {
2466 	if (f)
2467 		printf(" ");
2468 	if (f & TH_FIN)
2469 		printf("F");
2470 	if (f & TH_SYN)
2471 		printf("S");
2472 	if (f & TH_RST)
2473 		printf("R");
2474 	if (f & TH_PUSH)
2475 		printf("P");
2476 	if (f & TH_ACK)
2477 		printf("A");
2478 	if (f & TH_URG)
2479 		printf("U");
2480 	if (f & TH_ECE)
2481 		printf("E");
2482 	if (f & TH_CWR)
2483 		printf("W");
2484 }
2485 
2486 #define	PF_SET_SKIP_STEPS(i)					\
2487 	do {							\
2488 		while (head[i] != cur) {			\
2489 			head[i]->skip[i].ptr = cur;		\
2490 			head[i] = TAILQ_NEXT(head[i], entries);	\
2491 		}						\
2492 	} while (0)
2493 
2494 void
2495 pf_calc_skip_steps(struct pf_krulequeue *rules)
2496 {
2497 	struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT];
2498 	int i;
2499 
2500 	cur = TAILQ_FIRST(rules);
2501 	prev = cur;
2502 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2503 		head[i] = cur;
2504 	while (cur != NULL) {
2505 		if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
2506 			PF_SET_SKIP_STEPS(PF_SKIP_IFP);
2507 		if (cur->direction != prev->direction)
2508 			PF_SET_SKIP_STEPS(PF_SKIP_DIR);
2509 		if (cur->af != prev->af)
2510 			PF_SET_SKIP_STEPS(PF_SKIP_AF);
2511 		if (cur->proto != prev->proto)
2512 			PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
2513 		if (cur->src.neg != prev->src.neg ||
2514 		    pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
2515 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
2516 		if (cur->src.port[0] != prev->src.port[0] ||
2517 		    cur->src.port[1] != prev->src.port[1] ||
2518 		    cur->src.port_op != prev->src.port_op)
2519 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
2520 		if (cur->dst.neg != prev->dst.neg ||
2521 		    pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
2522 			PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
2523 		if (cur->dst.port[0] != prev->dst.port[0] ||
2524 		    cur->dst.port[1] != prev->dst.port[1] ||
2525 		    cur->dst.port_op != prev->dst.port_op)
2526 			PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
2527 
2528 		prev = cur;
2529 		cur = TAILQ_NEXT(cur, entries);
2530 	}
2531 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2532 		PF_SET_SKIP_STEPS(i);
2533 }
2534 
2535 int
2536 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
2537 {
2538 	if (aw1->type != aw2->type)
2539 		return (1);
2540 	switch (aw1->type) {
2541 	case PF_ADDR_ADDRMASK:
2542 	case PF_ADDR_RANGE:
2543 		if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
2544 			return (1);
2545 		if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
2546 			return (1);
2547 		return (0);
2548 	case PF_ADDR_DYNIFTL:
2549 		return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
2550 	case PF_ADDR_NOROUTE:
2551 	case PF_ADDR_URPFFAILED:
2552 		return (0);
2553 	case PF_ADDR_TABLE:
2554 		return (aw1->p.tbl != aw2->p.tbl);
2555 	default:
2556 		printf("invalid address type: %d\n", aw1->type);
2557 		return (1);
2558 	}
2559 }
2560 
2561 /**
2562  * Checksum updates are a little complicated because the checksum in the TCP/UDP
2563  * header isn't always a full checksum. In some cases (i.e. output) it's a
2564  * pseudo-header checksum, which is a partial checksum over src/dst IP
2565  * addresses, protocol number and length.
2566  *
2567  * That means we have the following cases:
2568  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2569  *  	checksums, we need to update the checksum whenever we change anything.
2570  *  * Output (i.e. the checksum is a pseudo-header checksum):
2571  *  	x The field being updated is src/dst address or affects the length of
2572  *  	the packet. We need to update the pseudo-header checksum (note that this
2573  *  	checksum is not ones' complement).
2574  *  	x Some other field is being modified (e.g. src/dst port numbers): We
2575  *  	don't have to update anything.
2576  **/
2577 u_int16_t
2578 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2579 {
2580 	u_int32_t x;
2581 
2582 	x = cksum + old - new;
2583 	x = (x + (x >> 16)) & 0xffff;
2584 
2585 	/* optimise: eliminate a branch when not udp */
2586 	if (udp && cksum == 0x0000)
2587 		return cksum;
2588 	if (udp && x == 0x0000)
2589 		x = 0xffff;
2590 
2591 	return (u_int16_t)(x);
2592 }
2593 
2594 static void
2595 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi,
2596     u_int8_t udp)
2597 {
2598 	u_int16_t old = htons(hi ? (*f << 8) : *f);
2599 	u_int16_t new = htons(hi ? ( v << 8) :  v);
2600 
2601 	if (*f == v)
2602 		return;
2603 
2604 	*f = v;
2605 
2606 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2607 		return;
2608 
2609 	*cksum = pf_cksum_fixup(*cksum, old, new, udp);
2610 }
2611 
2612 void
2613 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v,
2614     bool hi, u_int8_t udp)
2615 {
2616 	u_int8_t *fb = (u_int8_t *)f;
2617 	u_int8_t *vb = (u_int8_t *)&v;
2618 
2619 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2620 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2621 }
2622 
2623 void
2624 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v,
2625     bool hi, u_int8_t udp)
2626 {
2627 	u_int8_t *fb = (u_int8_t *)f;
2628 	u_int8_t *vb = (u_int8_t *)&v;
2629 
2630 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2631 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2632 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2633 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2634 }
2635 
2636 u_int16_t
2637 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2638         u_int16_t new, u_int8_t udp)
2639 {
2640 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2641 		return (cksum);
2642 
2643 	return (pf_cksum_fixup(cksum, old, new, udp));
2644 }
2645 
2646 static void
2647 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2648         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2649         sa_family_t af)
2650 {
2651 	struct pf_addr	ao;
2652 	u_int16_t	po = *p;
2653 
2654 	PF_ACPY(&ao, a, af);
2655 	PF_ACPY(a, an, af);
2656 
2657 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2658 		*pc = ~*pc;
2659 
2660 	*p = pn;
2661 
2662 	switch (af) {
2663 #ifdef INET
2664 	case AF_INET:
2665 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2666 		    ao.addr16[0], an->addr16[0], 0),
2667 		    ao.addr16[1], an->addr16[1], 0);
2668 		*p = pn;
2669 
2670 		*pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2671 		    ao.addr16[0], an->addr16[0], u),
2672 		    ao.addr16[1], an->addr16[1], u);
2673 
2674 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2675 		break;
2676 #endif /* INET */
2677 #ifdef INET6
2678 	case AF_INET6:
2679 		*pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2680 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2681 		    pf_cksum_fixup(pf_cksum_fixup(*pc,
2682 		    ao.addr16[0], an->addr16[0], u),
2683 		    ao.addr16[1], an->addr16[1], u),
2684 		    ao.addr16[2], an->addr16[2], u),
2685 		    ao.addr16[3], an->addr16[3], u),
2686 		    ao.addr16[4], an->addr16[4], u),
2687 		    ao.addr16[5], an->addr16[5], u),
2688 		    ao.addr16[6], an->addr16[6], u),
2689 		    ao.addr16[7], an->addr16[7], u);
2690 
2691 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2692 		break;
2693 #endif /* INET6 */
2694 	}
2695 
2696 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2697 	    CSUM_DELAY_DATA_IPV6)) {
2698 		*pc = ~*pc;
2699 		if (! *pc)
2700 			*pc = 0xffff;
2701 	}
2702 }
2703 
2704 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2705 void
2706 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2707 {
2708 	u_int32_t	ao;
2709 
2710 	memcpy(&ao, a, sizeof(ao));
2711 	memcpy(a, &an, sizeof(u_int32_t));
2712 	*c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2713 	    ao % 65536, an % 65536, u);
2714 }
2715 
2716 void
2717 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2718 {
2719 	u_int32_t	ao;
2720 
2721 	memcpy(&ao, a, sizeof(ao));
2722 	memcpy(a, &an, sizeof(u_int32_t));
2723 
2724 	*c = pf_proto_cksum_fixup(m,
2725 	    pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2726 	    ao % 65536, an % 65536, udp);
2727 }
2728 
2729 #ifdef INET6
2730 static void
2731 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2732 {
2733 	struct pf_addr	ao;
2734 
2735 	PF_ACPY(&ao, a, AF_INET6);
2736 	PF_ACPY(a, an, AF_INET6);
2737 
2738 	*c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2739 	    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2740 	    pf_cksum_fixup(pf_cksum_fixup(*c,
2741 	    ao.addr16[0], an->addr16[0], u),
2742 	    ao.addr16[1], an->addr16[1], u),
2743 	    ao.addr16[2], an->addr16[2], u),
2744 	    ao.addr16[3], an->addr16[3], u),
2745 	    ao.addr16[4], an->addr16[4], u),
2746 	    ao.addr16[5], an->addr16[5], u),
2747 	    ao.addr16[6], an->addr16[6], u),
2748 	    ao.addr16[7], an->addr16[7], u);
2749 }
2750 #endif /* INET6 */
2751 
2752 static void
2753 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2754     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2755     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2756 {
2757 	struct pf_addr	oia, ooa;
2758 
2759 	PF_ACPY(&oia, ia, af);
2760 	if (oa)
2761 		PF_ACPY(&ooa, oa, af);
2762 
2763 	/* Change inner protocol port, fix inner protocol checksum. */
2764 	if (ip != NULL) {
2765 		u_int16_t	oip = *ip;
2766 		u_int32_t	opc;
2767 
2768 		if (pc != NULL)
2769 			opc = *pc;
2770 		*ip = np;
2771 		if (pc != NULL)
2772 			*pc = pf_cksum_fixup(*pc, oip, *ip, u);
2773 		*ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2774 		if (pc != NULL)
2775 			*ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2776 	}
2777 	/* Change inner ip address, fix inner ip and icmp checksums. */
2778 	PF_ACPY(ia, na, af);
2779 	switch (af) {
2780 #ifdef INET
2781 	case AF_INET: {
2782 		u_int32_t	 oh2c = *h2c;
2783 
2784 		*h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2785 		    oia.addr16[0], ia->addr16[0], 0),
2786 		    oia.addr16[1], ia->addr16[1], 0);
2787 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2788 		    oia.addr16[0], ia->addr16[0], 0),
2789 		    oia.addr16[1], ia->addr16[1], 0);
2790 		*ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2791 		break;
2792 	}
2793 #endif /* INET */
2794 #ifdef INET6
2795 	case AF_INET6:
2796 		*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2797 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2798 		    pf_cksum_fixup(pf_cksum_fixup(*ic,
2799 		    oia.addr16[0], ia->addr16[0], u),
2800 		    oia.addr16[1], ia->addr16[1], u),
2801 		    oia.addr16[2], ia->addr16[2], u),
2802 		    oia.addr16[3], ia->addr16[3], u),
2803 		    oia.addr16[4], ia->addr16[4], u),
2804 		    oia.addr16[5], ia->addr16[5], u),
2805 		    oia.addr16[6], ia->addr16[6], u),
2806 		    oia.addr16[7], ia->addr16[7], u);
2807 		break;
2808 #endif /* INET6 */
2809 	}
2810 	/* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2811 	if (oa) {
2812 		PF_ACPY(oa, na, af);
2813 		switch (af) {
2814 #ifdef INET
2815 		case AF_INET:
2816 			*hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2817 			    ooa.addr16[0], oa->addr16[0], 0),
2818 			    ooa.addr16[1], oa->addr16[1], 0);
2819 			break;
2820 #endif /* INET */
2821 #ifdef INET6
2822 		case AF_INET6:
2823 			*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2824 			    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2825 			    pf_cksum_fixup(pf_cksum_fixup(*ic,
2826 			    ooa.addr16[0], oa->addr16[0], u),
2827 			    ooa.addr16[1], oa->addr16[1], u),
2828 			    ooa.addr16[2], oa->addr16[2], u),
2829 			    ooa.addr16[3], oa->addr16[3], u),
2830 			    ooa.addr16[4], oa->addr16[4], u),
2831 			    ooa.addr16[5], oa->addr16[5], u),
2832 			    ooa.addr16[6], oa->addr16[6], u),
2833 			    ooa.addr16[7], oa->addr16[7], u);
2834 			break;
2835 #endif /* INET6 */
2836 		}
2837 	}
2838 }
2839 
2840 /*
2841  * Need to modulate the sequence numbers in the TCP SACK option
2842  * (credits to Krzysztof Pfaff for report and patch)
2843  */
2844 static int
2845 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2846     struct tcphdr *th, struct pf_state_peer *dst)
2847 {
2848 	int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2849 	u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2850 	int copyback = 0, i, olen;
2851 	struct sackblk sack;
2852 
2853 #define	TCPOLEN_SACKLEN	(TCPOLEN_SACK + 2)
2854 	if (hlen < TCPOLEN_SACKLEN ||
2855 	    !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2856 		return 0;
2857 
2858 	while (hlen >= TCPOLEN_SACKLEN) {
2859 		size_t startoff = opt - opts;
2860 		olen = opt[1];
2861 		switch (*opt) {
2862 		case TCPOPT_EOL:	/* FALLTHROUGH */
2863 		case TCPOPT_NOP:
2864 			opt++;
2865 			hlen--;
2866 			break;
2867 		case TCPOPT_SACK:
2868 			if (olen > hlen)
2869 				olen = hlen;
2870 			if (olen >= TCPOLEN_SACKLEN) {
2871 				for (i = 2; i + TCPOLEN_SACK <= olen;
2872 				    i += TCPOLEN_SACK) {
2873 					memcpy(&sack, &opt[i], sizeof(sack));
2874 					pf_patch_32_unaligned(m,
2875 					    &th->th_sum, &sack.start,
2876 					    htonl(ntohl(sack.start) - dst->seqdiff),
2877 					    PF_ALGNMNT(startoff),
2878 					    0);
2879 					pf_patch_32_unaligned(m, &th->th_sum,
2880 					    &sack.end,
2881 					    htonl(ntohl(sack.end) - dst->seqdiff),
2882 					    PF_ALGNMNT(startoff),
2883 					    0);
2884 					memcpy(&opt[i], &sack, sizeof(sack));
2885 				}
2886 				copyback = 1;
2887 			}
2888 			/* FALLTHROUGH */
2889 		default:
2890 			if (olen < 2)
2891 				olen = 2;
2892 			hlen -= olen;
2893 			opt += olen;
2894 		}
2895 	}
2896 
2897 	if (copyback)
2898 		m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2899 	return (copyback);
2900 }
2901 
2902 struct mbuf *
2903 pf_build_tcp(const struct pf_krule *r, sa_family_t af,
2904     const struct pf_addr *saddr, const struct pf_addr *daddr,
2905     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2906     u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl,
2907     bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid)
2908 {
2909 	struct mbuf	*m;
2910 	int		 len, tlen;
2911 #ifdef INET
2912 	struct ip	*h = NULL;
2913 #endif /* INET */
2914 #ifdef INET6
2915 	struct ip6_hdr	*h6 = NULL;
2916 #endif /* INET6 */
2917 	struct tcphdr	*th;
2918 	char		*opt;
2919 	struct pf_mtag  *pf_mtag;
2920 
2921 	len = 0;
2922 	th = NULL;
2923 
2924 	/* maximum segment size tcp option */
2925 	tlen = sizeof(struct tcphdr);
2926 	if (mss)
2927 		tlen += 4;
2928 
2929 	switch (af) {
2930 #ifdef INET
2931 	case AF_INET:
2932 		len = sizeof(struct ip) + tlen;
2933 		break;
2934 #endif /* INET */
2935 #ifdef INET6
2936 	case AF_INET6:
2937 		len = sizeof(struct ip6_hdr) + tlen;
2938 		break;
2939 #endif /* INET6 */
2940 	default:
2941 		panic("%s: unsupported af %d", __func__, af);
2942 	}
2943 
2944 	m = m_gethdr(M_NOWAIT, MT_DATA);
2945 	if (m == NULL)
2946 		return (NULL);
2947 
2948 #ifdef MAC
2949 	mac_netinet_firewall_send(m);
2950 #endif
2951 	if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2952 		m_freem(m);
2953 		return (NULL);
2954 	}
2955 	if (skip_firewall)
2956 		m->m_flags |= M_SKIP_FIREWALL;
2957 	pf_mtag->tag = mtag_tag;
2958 	pf_mtag->flags = mtag_flags;
2959 
2960 	if (rtableid >= 0)
2961 		M_SETFIB(m, rtableid);
2962 
2963 #ifdef ALTQ
2964 	if (r != NULL && r->qid) {
2965 		pf_mtag->qid = r->qid;
2966 
2967 		/* add hints for ecn */
2968 		pf_mtag->hdr = mtod(m, struct ip *);
2969 	}
2970 #endif /* ALTQ */
2971 	m->m_data += max_linkhdr;
2972 	m->m_pkthdr.len = m->m_len = len;
2973 	/* The rest of the stack assumes a rcvif, so provide one.
2974 	 * This is a locally generated packet, so .. close enough. */
2975 	m->m_pkthdr.rcvif = V_loif;
2976 	bzero(m->m_data, len);
2977 	switch (af) {
2978 #ifdef INET
2979 	case AF_INET:
2980 		h = mtod(m, struct ip *);
2981 
2982 		/* IP header fields included in the TCP checksum */
2983 		h->ip_p = IPPROTO_TCP;
2984 		h->ip_len = htons(tlen);
2985 		h->ip_src.s_addr = saddr->v4.s_addr;
2986 		h->ip_dst.s_addr = daddr->v4.s_addr;
2987 
2988 		th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2989 		break;
2990 #endif /* INET */
2991 #ifdef INET6
2992 	case AF_INET6:
2993 		h6 = mtod(m, struct ip6_hdr *);
2994 
2995 		/* IP header fields included in the TCP checksum */
2996 		h6->ip6_nxt = IPPROTO_TCP;
2997 		h6->ip6_plen = htons(tlen);
2998 		memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2999 		memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
3000 
3001 		th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
3002 		break;
3003 #endif /* INET6 */
3004 	}
3005 
3006 	/* TCP header */
3007 	th->th_sport = sport;
3008 	th->th_dport = dport;
3009 	th->th_seq = htonl(seq);
3010 	th->th_ack = htonl(ack);
3011 	th->th_off = tlen >> 2;
3012 	th->th_flags = tcp_flags;
3013 	th->th_win = htons(win);
3014 
3015 	if (mss) {
3016 		opt = (char *)(th + 1);
3017 		opt[0] = TCPOPT_MAXSEG;
3018 		opt[1] = 4;
3019 		HTONS(mss);
3020 		bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
3021 	}
3022 
3023 	switch (af) {
3024 #ifdef INET
3025 	case AF_INET:
3026 		/* TCP checksum */
3027 		th->th_sum = in_cksum(m, len);
3028 
3029 		/* Finish the IP header */
3030 		h->ip_v = 4;
3031 		h->ip_hl = sizeof(*h) >> 2;
3032 		h->ip_tos = IPTOS_LOWDELAY;
3033 		h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
3034 		h->ip_len = htons(len);
3035 		h->ip_ttl = ttl ? ttl : V_ip_defttl;
3036 		h->ip_sum = 0;
3037 		break;
3038 #endif /* INET */
3039 #ifdef INET6
3040 	case AF_INET6:
3041 		/* TCP checksum */
3042 		th->th_sum = in6_cksum(m, IPPROTO_TCP,
3043 		    sizeof(struct ip6_hdr), tlen);
3044 
3045 		h6->ip6_vfc |= IPV6_VERSION;
3046 		h6->ip6_hlim = IPV6_DEFHLIM;
3047 		break;
3048 #endif /* INET6 */
3049 	}
3050 
3051 	return (m);
3052 }
3053 
3054 static void
3055 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd,
3056     uint8_t ttl, int rtableid)
3057 {
3058 	struct mbuf		*m;
3059 #ifdef INET
3060 	struct ip		*h = NULL;
3061 #endif /* INET */
3062 #ifdef INET6
3063 	struct ip6_hdr		*h6 = NULL;
3064 #endif /* INET6 */
3065 	struct sctphdr		*hdr;
3066 	struct sctp_chunkhdr	*chunk;
3067 	struct pf_send_entry	*pfse;
3068 	int			 off = 0;
3069 
3070 	MPASS(af == pd->af);
3071 
3072 	m = m_gethdr(M_NOWAIT, MT_DATA);
3073 	if (m == NULL)
3074 		return;
3075 
3076 	m->m_data += max_linkhdr;
3077 	m->m_flags |= M_SKIP_FIREWALL;
3078 	/* The rest of the stack assumes a rcvif, so provide one.
3079 	 * This is a locally generated packet, so .. close enough. */
3080 	m->m_pkthdr.rcvif = V_loif;
3081 
3082 	/* IPv4|6 header */
3083 	switch (af) {
3084 #ifdef INET
3085 	case AF_INET:
3086 		bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk));
3087 
3088 		h = mtod(m, struct ip *);
3089 
3090 		/* IP header fields included in the TCP checksum */
3091 
3092 		h->ip_p = IPPROTO_SCTP;
3093 		h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk));
3094 		h->ip_ttl = ttl ? ttl : V_ip_defttl;
3095 		h->ip_src = pd->dst->v4;
3096 		h->ip_dst = pd->src->v4;
3097 
3098 		off += sizeof(struct ip);
3099 		break;
3100 #endif /* INET */
3101 #ifdef INET6
3102 	case AF_INET6:
3103 		bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk));
3104 
3105 		h6 = mtod(m, struct ip6_hdr *);
3106 
3107 		/* IP header fields included in the TCP checksum */
3108 		h6->ip6_vfc |= IPV6_VERSION;
3109 		h6->ip6_nxt = IPPROTO_SCTP;
3110 		h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk));
3111 		h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim;
3112 		memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr));
3113 		memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr));
3114 
3115 		off += sizeof(struct ip6_hdr);
3116 		break;
3117 #endif /* INET6 */
3118 	}
3119 
3120 	/* SCTP header */
3121 	hdr = mtodo(m, off);
3122 
3123 	hdr->src_port = pd->hdr.sctp.dest_port;
3124 	hdr->dest_port = pd->hdr.sctp.src_port;
3125 	hdr->v_tag = pd->sctp_initiate_tag;
3126 	hdr->checksum = 0;
3127 
3128 	/* Abort chunk. */
3129 	off += sizeof(struct sctphdr);
3130 	chunk = mtodo(m, off);
3131 
3132 	chunk->chunk_type = SCTP_ABORT_ASSOCIATION;
3133 	chunk->chunk_length = htons(sizeof(*chunk));
3134 
3135 	/* SCTP checksum */
3136 	off += sizeof(*chunk);
3137 	m->m_pkthdr.len = m->m_len = off;
3138 
3139 	pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk));;
3140 
3141 	if (rtableid >= 0)
3142 		M_SETFIB(m, rtableid);
3143 
3144 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3145 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3146 	if (pfse == NULL) {
3147 		m_freem(m);
3148 		return;
3149 	}
3150 
3151 	switch (af) {
3152 #ifdef INET
3153 	case AF_INET:
3154 		pfse->pfse_type = PFSE_IP;
3155 		break;
3156 #endif /* INET */
3157 #ifdef INET6
3158 	case AF_INET6:
3159 		pfse->pfse_type = PFSE_IP6;
3160 		break;
3161 #endif /* INET6 */
3162 	}
3163 
3164 	pfse->pfse_m = m;
3165 	pf_send(pfse);
3166 }
3167 
3168 void
3169 pf_send_tcp(const struct pf_krule *r, sa_family_t af,
3170     const struct pf_addr *saddr, const struct pf_addr *daddr,
3171     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
3172     u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl,
3173     bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid)
3174 {
3175 	struct pf_send_entry *pfse;
3176 	struct mbuf	*m;
3177 
3178 	m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags,
3179 	    win, mss, ttl, skip_firewall, mtag_tag, mtag_flags, rtableid);
3180 	if (m == NULL)
3181 		return;
3182 
3183 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3184 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3185 	if (pfse == NULL) {
3186 		m_freem(m);
3187 		return;
3188 	}
3189 
3190 	switch (af) {
3191 #ifdef INET
3192 	case AF_INET:
3193 		pfse->pfse_type = PFSE_IP;
3194 		break;
3195 #endif /* INET */
3196 #ifdef INET6
3197 	case AF_INET6:
3198 		pfse->pfse_type = PFSE_IP6;
3199 		break;
3200 #endif /* INET6 */
3201 	}
3202 
3203 	pfse->pfse_m = m;
3204 	pf_send(pfse);
3205 }
3206 
3207 static void
3208 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd,
3209     struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th,
3210     struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
3211     u_short *reason, int rtableid)
3212 {
3213 	struct pf_addr	* const saddr = pd->src;
3214 	struct pf_addr	* const daddr = pd->dst;
3215 	sa_family_t	 af = pd->af;
3216 
3217 	/* undo NAT changes, if they have taken place */
3218 	if (nr != NULL) {
3219 		PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3220 		PF_ACPY(daddr, &sk->addr[pd->didx], af);
3221 		if (pd->sport)
3222 			*pd->sport = sk->port[pd->sidx];
3223 		if (pd->dport)
3224 			*pd->dport = sk->port[pd->didx];
3225 		if (pd->proto_sum)
3226 			*pd->proto_sum = bproto_sum;
3227 		if (pd->ip_sum)
3228 			*pd->ip_sum = bip_sum;
3229 		m_copyback(m, off, hdrlen, pd->hdr.any);
3230 	}
3231 	if (pd->proto == IPPROTO_TCP &&
3232 	    ((r->rule_flag & PFRULE_RETURNRST) ||
3233 	    (r->rule_flag & PFRULE_RETURN)) &&
3234 	    !(th->th_flags & TH_RST)) {
3235 		u_int32_t	 ack = ntohl(th->th_seq) + pd->p_len;
3236 		int		 len = 0;
3237 #ifdef INET
3238 		struct ip	*h4;
3239 #endif
3240 #ifdef INET6
3241 		struct ip6_hdr	*h6;
3242 #endif
3243 
3244 		switch (af) {
3245 #ifdef INET
3246 		case AF_INET:
3247 			h4 = mtod(m, struct ip *);
3248 			len = ntohs(h4->ip_len) - off;
3249 			break;
3250 #endif
3251 #ifdef INET6
3252 		case AF_INET6:
3253 			h6 = mtod(m, struct ip6_hdr *);
3254 			len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3255 			break;
3256 #endif
3257 		}
3258 
3259 		if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3260 			REASON_SET(reason, PFRES_PROTCKSUM);
3261 		else {
3262 			if (th->th_flags & TH_SYN)
3263 				ack++;
3264 			if (th->th_flags & TH_FIN)
3265 				ack++;
3266 			pf_send_tcp(r, af, pd->dst,
3267 				pd->src, th->th_dport, th->th_sport,
3268 				ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3269 				r->return_ttl, true, 0, 0, rtableid);
3270 		}
3271 	} else if (pd->proto == IPPROTO_SCTP &&
3272 	    (r->rule_flag & PFRULE_RETURN)) {
3273 		pf_send_sctp_abort(af, pd, r->return_ttl, rtableid);
3274 	} else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3275 		r->return_icmp)
3276 		pf_send_icmp(m, r->return_icmp >> 8,
3277 			r->return_icmp & 255, af, r, rtableid);
3278 	else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3279 		r->return_icmp6)
3280 		pf_send_icmp(m, r->return_icmp6 >> 8,
3281 			r->return_icmp6 & 255, af, r, rtableid);
3282 }
3283 
3284 static int
3285 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
3286 {
3287 	struct m_tag *mtag;
3288 	u_int8_t mpcp;
3289 
3290 	mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
3291 	if (mtag == NULL)
3292 		return (0);
3293 
3294 	if (prio == PF_PRIO_ZERO)
3295 		prio = 0;
3296 
3297 	mpcp = *(uint8_t *)(mtag + 1);
3298 
3299 	return (mpcp == prio);
3300 }
3301 
3302 static int
3303 pf_icmp_to_bandlim(uint8_t type)
3304 {
3305 	switch (type) {
3306 		case ICMP_ECHO:
3307 		case ICMP_ECHOREPLY:
3308 			return (BANDLIM_ICMP_ECHO);
3309 		case ICMP_TSTAMP:
3310 		case ICMP_TSTAMPREPLY:
3311 			return (BANDLIM_ICMP_TSTAMP);
3312 		case ICMP_UNREACH:
3313 		default:
3314 			return (BANDLIM_ICMP_UNREACH);
3315 	}
3316 }
3317 
3318 static void
3319 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
3320     struct pf_krule *r, int rtableid)
3321 {
3322 	struct pf_send_entry *pfse;
3323 	struct mbuf *m0;
3324 	struct pf_mtag *pf_mtag;
3325 
3326 	/* ICMP packet rate limitation. */
3327 #ifdef INET6
3328 	if (af == AF_INET6) {
3329 		if (icmp6_ratelimit(NULL, type, code))
3330 			return;
3331 	}
3332 #endif
3333 #ifdef INET
3334 	if (af == AF_INET) {
3335 		if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0)
3336 			return;
3337 	}
3338 #endif
3339 
3340 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3341 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3342 	if (pfse == NULL)
3343 		return;
3344 
3345 	if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
3346 		free(pfse, M_PFTEMP);
3347 		return;
3348 	}
3349 
3350 	if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
3351 		free(pfse, M_PFTEMP);
3352 		return;
3353 	}
3354 	/* XXX: revisit */
3355 	m0->m_flags |= M_SKIP_FIREWALL;
3356 
3357 	if (rtableid >= 0)
3358 		M_SETFIB(m0, rtableid);
3359 
3360 #ifdef ALTQ
3361 	if (r->qid) {
3362 		pf_mtag->qid = r->qid;
3363 		/* add hints for ecn */
3364 		pf_mtag->hdr = mtod(m0, struct ip *);
3365 	}
3366 #endif /* ALTQ */
3367 
3368 	switch (af) {
3369 #ifdef INET
3370 	case AF_INET:
3371 		pfse->pfse_type = PFSE_ICMP;
3372 		break;
3373 #endif /* INET */
3374 #ifdef INET6
3375 	case AF_INET6:
3376 		pfse->pfse_type = PFSE_ICMP6;
3377 		break;
3378 #endif /* INET6 */
3379 	}
3380 	pfse->pfse_m = m0;
3381 	pfse->icmpopts.type = type;
3382 	pfse->icmpopts.code = code;
3383 	pf_send(pfse);
3384 }
3385 
3386 /*
3387  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
3388  * If n is 0, they match if they are equal. If n is != 0, they match if they
3389  * are different.
3390  */
3391 int
3392 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
3393     struct pf_addr *b, sa_family_t af)
3394 {
3395 	int	match = 0;
3396 
3397 	switch (af) {
3398 #ifdef INET
3399 	case AF_INET:
3400 		if ((a->addr32[0] & m->addr32[0]) ==
3401 		    (b->addr32[0] & m->addr32[0]))
3402 			match++;
3403 		break;
3404 #endif /* INET */
3405 #ifdef INET6
3406 	case AF_INET6:
3407 		if (((a->addr32[0] & m->addr32[0]) ==
3408 		     (b->addr32[0] & m->addr32[0])) &&
3409 		    ((a->addr32[1] & m->addr32[1]) ==
3410 		     (b->addr32[1] & m->addr32[1])) &&
3411 		    ((a->addr32[2] & m->addr32[2]) ==
3412 		     (b->addr32[2] & m->addr32[2])) &&
3413 		    ((a->addr32[3] & m->addr32[3]) ==
3414 		     (b->addr32[3] & m->addr32[3])))
3415 			match++;
3416 		break;
3417 #endif /* INET6 */
3418 	}
3419 	if (match) {
3420 		if (n)
3421 			return (0);
3422 		else
3423 			return (1);
3424 	} else {
3425 		if (n)
3426 			return (1);
3427 		else
3428 			return (0);
3429 	}
3430 }
3431 
3432 /*
3433  * Return 1 if b <= a <= e, otherwise return 0.
3434  */
3435 int
3436 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
3437     struct pf_addr *a, sa_family_t af)
3438 {
3439 	switch (af) {
3440 #ifdef INET
3441 	case AF_INET:
3442 		if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
3443 		    (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
3444 			return (0);
3445 		break;
3446 #endif /* INET */
3447 #ifdef INET6
3448 	case AF_INET6: {
3449 		int	i;
3450 
3451 		/* check a >= b */
3452 		for (i = 0; i < 4; ++i)
3453 			if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
3454 				break;
3455 			else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
3456 				return (0);
3457 		/* check a <= e */
3458 		for (i = 0; i < 4; ++i)
3459 			if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
3460 				break;
3461 			else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
3462 				return (0);
3463 		break;
3464 	}
3465 #endif /* INET6 */
3466 	}
3467 	return (1);
3468 }
3469 
3470 static int
3471 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
3472 {
3473 	switch (op) {
3474 	case PF_OP_IRG:
3475 		return ((p > a1) && (p < a2));
3476 	case PF_OP_XRG:
3477 		return ((p < a1) || (p > a2));
3478 	case PF_OP_RRG:
3479 		return ((p >= a1) && (p <= a2));
3480 	case PF_OP_EQ:
3481 		return (p == a1);
3482 	case PF_OP_NE:
3483 		return (p != a1);
3484 	case PF_OP_LT:
3485 		return (p < a1);
3486 	case PF_OP_LE:
3487 		return (p <= a1);
3488 	case PF_OP_GT:
3489 		return (p > a1);
3490 	case PF_OP_GE:
3491 		return (p >= a1);
3492 	}
3493 	return (0); /* never reached */
3494 }
3495 
3496 int
3497 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
3498 {
3499 	NTOHS(a1);
3500 	NTOHS(a2);
3501 	NTOHS(p);
3502 	return (pf_match(op, a1, a2, p));
3503 }
3504 
3505 static int
3506 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
3507 {
3508 	if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
3509 		return (0);
3510 	return (pf_match(op, a1, a2, u));
3511 }
3512 
3513 static int
3514 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
3515 {
3516 	if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
3517 		return (0);
3518 	return (pf_match(op, a1, a2, g));
3519 }
3520 
3521 int
3522 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag)
3523 {
3524 	if (*tag == -1)
3525 		*tag = mtag;
3526 
3527 	return ((!r->match_tag_not && r->match_tag == *tag) ||
3528 	    (r->match_tag_not && r->match_tag != *tag));
3529 }
3530 
3531 int
3532 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
3533 {
3534 
3535 	KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
3536 
3537 	if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
3538 		return (ENOMEM);
3539 
3540 	pd->pf_mtag->tag = tag;
3541 
3542 	return (0);
3543 }
3544 
3545 #define	PF_ANCHOR_STACKSIZE	32
3546 struct pf_kanchor_stackframe {
3547 	struct pf_kruleset	*rs;
3548 	struct pf_krule		*r;	/* XXX: + match bit */
3549 	struct pf_kanchor	*child;
3550 };
3551 
3552 /*
3553  * XXX: We rely on malloc(9) returning pointer aligned addresses.
3554  */
3555 #define	PF_ANCHORSTACK_MATCH	0x00000001
3556 #define	PF_ANCHORSTACK_MASK	(PF_ANCHORSTACK_MATCH)
3557 
3558 #define	PF_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
3559 #define	PF_ANCHOR_RULE(f)	(struct pf_krule *)			\
3560 				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
3561 #define	PF_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 			\
3562 				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
3563 } while (0)
3564 
3565 void
3566 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth,
3567     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
3568     int *match)
3569 {
3570 	struct pf_kanchor_stackframe	*f;
3571 
3572 	PF_RULES_RASSERT();
3573 
3574 	if (match)
3575 		*match = 0;
3576 	if (*depth >= PF_ANCHOR_STACKSIZE) {
3577 		printf("%s: anchor stack overflow on %s\n",
3578 		    __func__, (*r)->anchor->name);
3579 		*r = TAILQ_NEXT(*r, entries);
3580 		return;
3581 	} else if (*depth == 0 && a != NULL)
3582 		*a = *r;
3583 	f = stack + (*depth)++;
3584 	f->rs = *rs;
3585 	f->r = *r;
3586 	if ((*r)->anchor_wildcard) {
3587 		struct pf_kanchor_node *parent = &(*r)->anchor->children;
3588 
3589 		if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) {
3590 			*r = NULL;
3591 			return;
3592 		}
3593 		*rs = &f->child->ruleset;
3594 	} else {
3595 		f->child = NULL;
3596 		*rs = &(*r)->anchor->ruleset;
3597 	}
3598 	*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
3599 }
3600 
3601 int
3602 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth,
3603     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
3604     int *match)
3605 {
3606 	struct pf_kanchor_stackframe	*f;
3607 	struct pf_krule *fr;
3608 	int quick = 0;
3609 
3610 	PF_RULES_RASSERT();
3611 
3612 	do {
3613 		if (*depth <= 0)
3614 			break;
3615 		f = stack + *depth - 1;
3616 		fr = PF_ANCHOR_RULE(f);
3617 		if (f->child != NULL) {
3618 			/*
3619 			 * This block traverses through
3620 			 * a wildcard anchor.
3621 			 */
3622 			if (match != NULL && *match) {
3623 				/*
3624 				 * If any of "*" matched, then
3625 				 * "foo/ *" matched, mark frame
3626 				 * appropriately.
3627 				 */
3628 				PF_ANCHOR_SET_MATCH(f);
3629 				*match = 0;
3630 			}
3631 			f->child = RB_NEXT(pf_kanchor_node,
3632 			    &fr->anchor->children, f->child);
3633 			if (f->child != NULL) {
3634 				*rs = &f->child->ruleset;
3635 				*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
3636 				if (*r == NULL)
3637 					continue;
3638 				else
3639 					break;
3640 			}
3641 		}
3642 		(*depth)--;
3643 		if (*depth == 0 && a != NULL)
3644 			*a = NULL;
3645 		*rs = f->rs;
3646 		if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
3647 			quick = fr->quick;
3648 		*r = TAILQ_NEXT(fr, entries);
3649 	} while (*r == NULL);
3650 
3651 	return (quick);
3652 }
3653 
3654 struct pf_keth_anchor_stackframe {
3655 	struct pf_keth_ruleset	*rs;
3656 	struct pf_keth_rule	*r;	/* XXX: + match bit */
3657 	struct pf_keth_anchor	*child;
3658 };
3659 
3660 #define	PF_ETH_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
3661 #define	PF_ETH_ANCHOR_RULE(f)	(struct pf_keth_rule *)			\
3662 				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
3663 #define	PF_ETH_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 		\
3664 				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
3665 } while (0)
3666 
3667 void
3668 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth,
3669     struct pf_keth_ruleset **rs, struct pf_keth_rule **r,
3670     struct pf_keth_rule **a, int *match)
3671 {
3672 	struct pf_keth_anchor_stackframe	*f;
3673 
3674 	NET_EPOCH_ASSERT();
3675 
3676 	if (match)
3677 		*match = 0;
3678 	if (*depth >= PF_ANCHOR_STACKSIZE) {
3679 		printf("%s: anchor stack overflow on %s\n",
3680 		    __func__, (*r)->anchor->name);
3681 		*r = TAILQ_NEXT(*r, entries);
3682 		return;
3683 	} else if (*depth == 0 && a != NULL)
3684 		*a = *r;
3685 	f = stack + (*depth)++;
3686 	f->rs = *rs;
3687 	f->r = *r;
3688 	if ((*r)->anchor_wildcard) {
3689 		struct pf_keth_anchor_node *parent = &(*r)->anchor->children;
3690 
3691 		if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) {
3692 			*r = NULL;
3693 			return;
3694 		}
3695 		*rs = &f->child->ruleset;
3696 	} else {
3697 		f->child = NULL;
3698 		*rs = &(*r)->anchor->ruleset;
3699 	}
3700 	*r = TAILQ_FIRST((*rs)->active.rules);
3701 }
3702 
3703 int
3704 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth,
3705     struct pf_keth_ruleset **rs, struct pf_keth_rule **r,
3706     struct pf_keth_rule **a, int *match)
3707 {
3708 	struct pf_keth_anchor_stackframe	*f;
3709 	struct pf_keth_rule *fr;
3710 	int quick = 0;
3711 
3712 	NET_EPOCH_ASSERT();
3713 
3714 	do {
3715 		if (*depth <= 0)
3716 			break;
3717 		f = stack + *depth - 1;
3718 		fr = PF_ETH_ANCHOR_RULE(f);
3719 		if (f->child != NULL) {
3720 			/*
3721 			 * This block traverses through
3722 			 * a wildcard anchor.
3723 			 */
3724 			if (match != NULL && *match) {
3725 				/*
3726 				 * If any of "*" matched, then
3727 				 * "foo/ *" matched, mark frame
3728 				 * appropriately.
3729 				 */
3730 				PF_ETH_ANCHOR_SET_MATCH(f);
3731 				*match = 0;
3732 			}
3733 			f->child = RB_NEXT(pf_keth_anchor_node,
3734 			    &fr->anchor->children, f->child);
3735 			if (f->child != NULL) {
3736 				*rs = &f->child->ruleset;
3737 				*r = TAILQ_FIRST((*rs)->active.rules);
3738 				if (*r == NULL)
3739 					continue;
3740 				else
3741 					break;
3742 			}
3743 		}
3744 		(*depth)--;
3745 		if (*depth == 0 && a != NULL)
3746 			*a = NULL;
3747 		*rs = f->rs;
3748 		if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match))
3749 			quick = fr->quick;
3750 		*r = TAILQ_NEXT(fr, entries);
3751 	} while (*r == NULL);
3752 
3753 	return (quick);
3754 }
3755 
3756 #ifdef INET6
3757 void
3758 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
3759     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
3760 {
3761 	switch (af) {
3762 #ifdef INET
3763 	case AF_INET:
3764 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
3765 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
3766 		break;
3767 #endif /* INET */
3768 	case AF_INET6:
3769 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
3770 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
3771 		naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
3772 		((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
3773 		naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
3774 		((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
3775 		naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
3776 		((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
3777 		break;
3778 	}
3779 }
3780 
3781 void
3782 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
3783 {
3784 	switch (af) {
3785 #ifdef INET
3786 	case AF_INET:
3787 		addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
3788 		break;
3789 #endif /* INET */
3790 	case AF_INET6:
3791 		if (addr->addr32[3] == 0xffffffff) {
3792 			addr->addr32[3] = 0;
3793 			if (addr->addr32[2] == 0xffffffff) {
3794 				addr->addr32[2] = 0;
3795 				if (addr->addr32[1] == 0xffffffff) {
3796 					addr->addr32[1] = 0;
3797 					addr->addr32[0] =
3798 					    htonl(ntohl(addr->addr32[0]) + 1);
3799 				} else
3800 					addr->addr32[1] =
3801 					    htonl(ntohl(addr->addr32[1]) + 1);
3802 			} else
3803 				addr->addr32[2] =
3804 				    htonl(ntohl(addr->addr32[2]) + 1);
3805 		} else
3806 			addr->addr32[3] =
3807 			    htonl(ntohl(addr->addr32[3]) + 1);
3808 		break;
3809 	}
3810 }
3811 #endif /* INET6 */
3812 
3813 void
3814 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a)
3815 {
3816 	/*
3817 	 * Modern rules use the same flags in rules as they do in states.
3818 	 */
3819 	a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID|
3820 	    PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO));
3821 
3822 	/*
3823 	 * Old-style scrub rules have different flags which need to be translated.
3824 	 */
3825 	if (r->rule_flag & PFRULE_RANDOMID)
3826 		a->flags |= PFSTATE_RANDOMID;
3827 	if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) {
3828 		a->flags |= PFSTATE_SETTOS;
3829 		a->set_tos = r->set_tos;
3830 	}
3831 
3832 	if (r->qid)
3833 		a->qid = r->qid;
3834 	if (r->pqid)
3835 		a->pqid = r->pqid;
3836 	if (r->rtableid >= 0)
3837 		a->rtableid = r->rtableid;
3838 	a->log |= r->log;
3839 	if (r->min_ttl)
3840 		a->min_ttl = r->min_ttl;
3841 	if (r->max_mss)
3842 		a->max_mss = r->max_mss;
3843 	if (r->dnpipe)
3844 		a->dnpipe = r->dnpipe;
3845 	if (r->dnrpipe)
3846 		a->dnrpipe = r->dnrpipe;
3847 	if (r->dnpipe || r->dnrpipe) {
3848 		if (r->free_flags & PFRULE_DN_IS_PIPE)
3849 			a->flags |= PFSTATE_DN_IS_PIPE;
3850 		else
3851 			a->flags &= ~PFSTATE_DN_IS_PIPE;
3852 	}
3853 	if (r->scrub_flags & PFSTATE_SETPRIO) {
3854 		a->set_prio[0] = r->set_prio[0];
3855 		a->set_prio[1] = r->set_prio[1];
3856 	}
3857 }
3858 
3859 int
3860 pf_socket_lookup(struct pf_pdesc *pd, struct mbuf *m)
3861 {
3862 	struct pf_addr		*saddr, *daddr;
3863 	u_int16_t		 sport, dport;
3864 	struct inpcbinfo	*pi;
3865 	struct inpcb		*inp;
3866 
3867 	pd->lookup.uid = UID_MAX;
3868 	pd->lookup.gid = GID_MAX;
3869 
3870 	switch (pd->proto) {
3871 	case IPPROTO_TCP:
3872 		sport = pd->hdr.tcp.th_sport;
3873 		dport = pd->hdr.tcp.th_dport;
3874 		pi = &V_tcbinfo;
3875 		break;
3876 	case IPPROTO_UDP:
3877 		sport = pd->hdr.udp.uh_sport;
3878 		dport = pd->hdr.udp.uh_dport;
3879 		pi = &V_udbinfo;
3880 		break;
3881 	default:
3882 		return (-1);
3883 	}
3884 	if (pd->dir == PF_IN) {
3885 		saddr = pd->src;
3886 		daddr = pd->dst;
3887 	} else {
3888 		u_int16_t	p;
3889 
3890 		p = sport;
3891 		sport = dport;
3892 		dport = p;
3893 		saddr = pd->dst;
3894 		daddr = pd->src;
3895 	}
3896 	switch (pd->af) {
3897 #ifdef INET
3898 	case AF_INET:
3899 		inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
3900 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
3901 		if (inp == NULL) {
3902 			inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
3903 			   daddr->v4, dport, INPLOOKUP_WILDCARD |
3904 			   INPLOOKUP_RLOCKPCB, NULL, m);
3905 			if (inp == NULL)
3906 				return (-1);
3907 		}
3908 		break;
3909 #endif /* INET */
3910 #ifdef INET6
3911 	case AF_INET6:
3912 		inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
3913 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
3914 		if (inp == NULL) {
3915 			inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
3916 			    &daddr->v6, dport, INPLOOKUP_WILDCARD |
3917 			    INPLOOKUP_RLOCKPCB, NULL, m);
3918 			if (inp == NULL)
3919 				return (-1);
3920 		}
3921 		break;
3922 #endif /* INET6 */
3923 
3924 	default:
3925 		return (-1);
3926 	}
3927 	INP_RLOCK_ASSERT(inp);
3928 	pd->lookup.uid = inp->inp_cred->cr_uid;
3929 	pd->lookup.gid = inp->inp_cred->cr_groups[0];
3930 	INP_RUNLOCK(inp);
3931 
3932 	return (1);
3933 }
3934 
3935 u_int8_t
3936 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3937 {
3938 	int		 hlen;
3939 	u_int8_t	 hdr[60];
3940 	u_int8_t	*opt, optlen;
3941 	u_int8_t	 wscale = 0;
3942 
3943 	hlen = th_off << 2;		/* hlen <= sizeof(hdr) */
3944 	if (hlen <= sizeof(struct tcphdr))
3945 		return (0);
3946 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3947 		return (0);
3948 	opt = hdr + sizeof(struct tcphdr);
3949 	hlen -= sizeof(struct tcphdr);
3950 	while (hlen >= 3) {
3951 		switch (*opt) {
3952 		case TCPOPT_EOL:
3953 		case TCPOPT_NOP:
3954 			++opt;
3955 			--hlen;
3956 			break;
3957 		case TCPOPT_WINDOW:
3958 			wscale = opt[2];
3959 			if (wscale > TCP_MAX_WINSHIFT)
3960 				wscale = TCP_MAX_WINSHIFT;
3961 			wscale |= PF_WSCALE_FLAG;
3962 			/* FALLTHROUGH */
3963 		default:
3964 			optlen = opt[1];
3965 			if (optlen < 2)
3966 				optlen = 2;
3967 			hlen -= optlen;
3968 			opt += optlen;
3969 			break;
3970 		}
3971 	}
3972 	return (wscale);
3973 }
3974 
3975 u_int16_t
3976 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3977 {
3978 	int		 hlen;
3979 	u_int8_t	 hdr[60];
3980 	u_int8_t	*opt, optlen;
3981 	u_int16_t	 mss = V_tcp_mssdflt;
3982 
3983 	hlen = th_off << 2;	/* hlen <= sizeof(hdr) */
3984 	if (hlen <= sizeof(struct tcphdr))
3985 		return (0);
3986 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3987 		return (0);
3988 	opt = hdr + sizeof(struct tcphdr);
3989 	hlen -= sizeof(struct tcphdr);
3990 	while (hlen >= TCPOLEN_MAXSEG) {
3991 		switch (*opt) {
3992 		case TCPOPT_EOL:
3993 		case TCPOPT_NOP:
3994 			++opt;
3995 			--hlen;
3996 			break;
3997 		case TCPOPT_MAXSEG:
3998 			bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
3999 			NTOHS(mss);
4000 			/* FALLTHROUGH */
4001 		default:
4002 			optlen = opt[1];
4003 			if (optlen < 2)
4004 				optlen = 2;
4005 			hlen -= optlen;
4006 			opt += optlen;
4007 			break;
4008 		}
4009 	}
4010 	return (mss);
4011 }
4012 
4013 static u_int16_t
4014 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
4015 {
4016 	struct nhop_object *nh;
4017 #ifdef INET6
4018 	struct in6_addr		dst6;
4019 	uint32_t		scopeid;
4020 #endif /* INET6 */
4021 	int			 hlen = 0;
4022 	uint16_t		 mss = 0;
4023 
4024 	NET_EPOCH_ASSERT();
4025 
4026 	switch (af) {
4027 #ifdef INET
4028 	case AF_INET:
4029 		hlen = sizeof(struct ip);
4030 		nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0);
4031 		if (nh != NULL)
4032 			mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
4033 		break;
4034 #endif /* INET */
4035 #ifdef INET6
4036 	case AF_INET6:
4037 		hlen = sizeof(struct ip6_hdr);
4038 		in6_splitscope(&addr->v6, &dst6, &scopeid);
4039 		nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0);
4040 		if (nh != NULL)
4041 			mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
4042 		break;
4043 #endif /* INET6 */
4044 	}
4045 
4046 	mss = max(V_tcp_mssdflt, mss);
4047 	mss = min(mss, offer);
4048 	mss = max(mss, 64);		/* sanity - at least max opt space */
4049 	return (mss);
4050 }
4051 
4052 static u_int32_t
4053 pf_tcp_iss(struct pf_pdesc *pd)
4054 {
4055 	MD5_CTX ctx;
4056 	u_int32_t digest[4];
4057 
4058 	if (V_pf_tcp_secret_init == 0) {
4059 		arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
4060 		MD5Init(&V_pf_tcp_secret_ctx);
4061 		MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
4062 		    sizeof(V_pf_tcp_secret));
4063 		V_pf_tcp_secret_init = 1;
4064 	}
4065 
4066 	ctx = V_pf_tcp_secret_ctx;
4067 
4068 	MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short));
4069 	MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short));
4070 	if (pd->af == AF_INET6) {
4071 		MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
4072 		MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
4073 	} else {
4074 		MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
4075 		MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
4076 	}
4077 	MD5Final((u_char *)digest, &ctx);
4078 	V_pf_tcp_iss_off += 4096;
4079 #define	ISN_RANDOM_INCREMENT (4096 - 1)
4080 	return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
4081 	    V_pf_tcp_iss_off);
4082 #undef	ISN_RANDOM_INCREMENT
4083 }
4084 
4085 static bool
4086 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r)
4087 {
4088 	bool match = true;
4089 
4090 	/* Always matches if not set */
4091 	if (! r->isset)
4092 		return (!r->neg);
4093 
4094 	for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4095 		if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) {
4096 			match = false;
4097 			break;
4098 		}
4099 	}
4100 
4101 	return (match ^ r->neg);
4102 }
4103 
4104 static int
4105 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag)
4106 {
4107 	if (*tag == -1)
4108 		*tag = mtag;
4109 
4110 	return ((!r->match_tag_not && r->match_tag == *tag) ||
4111 	    (r->match_tag_not && r->match_tag != *tag));
4112 }
4113 
4114 static void
4115 pf_bridge_to(struct ifnet *ifp, struct mbuf *m)
4116 {
4117 	/* If we don't have the interface drop the packet. */
4118 	if (ifp == NULL) {
4119 		m_freem(m);
4120 		return;
4121 	}
4122 
4123 	switch (ifp->if_type) {
4124 	case IFT_ETHER:
4125 	case IFT_XETHER:
4126 	case IFT_L2VLAN:
4127 	case IFT_BRIDGE:
4128 	case IFT_IEEE8023ADLAG:
4129 		break;
4130 	default:
4131 		m_freem(m);
4132 		return;
4133 	}
4134 
4135 	ifp->if_transmit(ifp, m);
4136 }
4137 
4138 static int
4139 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0)
4140 {
4141 #ifdef INET
4142 	struct ip ip;
4143 #endif
4144 #ifdef INET6
4145 	struct ip6_hdr ip6;
4146 #endif
4147 	struct mbuf *m = *m0;
4148 	struct ether_header *e;
4149 	struct pf_keth_rule *r, *rm, *a = NULL;
4150 	struct pf_keth_ruleset *ruleset = NULL;
4151 	struct pf_mtag *mtag;
4152 	struct pf_keth_ruleq *rules;
4153 	struct pf_addr *src = NULL, *dst = NULL;
4154 	struct pfi_kkif *bridge_to;
4155 	sa_family_t af = 0;
4156 	uint16_t proto;
4157 	int asd = 0, match = 0;
4158 	int tag = -1;
4159 	uint8_t action;
4160 	struct pf_keth_anchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
4161 
4162 	MPASS(kif->pfik_ifp->if_vnet == curvnet);
4163 	NET_EPOCH_ASSERT();
4164 
4165 	PF_RULES_RLOCK_TRACKER;
4166 
4167 	SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m);
4168 
4169 	mtag = pf_find_mtag(m);
4170 	if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
4171 		/* Dummynet re-injects packets after they've
4172 		 * completed their delay. We've already
4173 		 * processed them, so pass unconditionally. */
4174 
4175 		/* But only once. We may see the packet multiple times (e.g.
4176 		 * PFIL_IN/PFIL_OUT). */
4177 		pf_dummynet_flag_remove(m, mtag);
4178 
4179 		return (PF_PASS);
4180 	}
4181 
4182 	ruleset = V_pf_keth;
4183 	rules = ck_pr_load_ptr(&ruleset->active.rules);
4184 	r = TAILQ_FIRST(rules);
4185 	rm = NULL;
4186 
4187 	e = mtod(m, struct ether_header *);
4188 	proto = ntohs(e->ether_type);
4189 
4190 	switch (proto) {
4191 #ifdef INET
4192 	case ETHERTYPE_IP: {
4193 		if (m_length(m, NULL) < (sizeof(struct ether_header) +
4194 		    sizeof(ip)))
4195 			return (PF_DROP);
4196 
4197 		af = AF_INET;
4198 		m_copydata(m, sizeof(struct ether_header), sizeof(ip),
4199 		    (caddr_t)&ip);
4200 		src = (struct pf_addr *)&ip.ip_src;
4201 		dst = (struct pf_addr *)&ip.ip_dst;
4202 		break;
4203 	}
4204 #endif /* INET */
4205 #ifdef INET6
4206 	case ETHERTYPE_IPV6: {
4207 		if (m_length(m, NULL) < (sizeof(struct ether_header) +
4208 		    sizeof(ip6)))
4209 			return (PF_DROP);
4210 
4211 		af = AF_INET6;
4212 		m_copydata(m, sizeof(struct ether_header), sizeof(ip6),
4213 		    (caddr_t)&ip6);
4214 		src = (struct pf_addr *)&ip6.ip6_src;
4215 		dst = (struct pf_addr *)&ip6.ip6_dst;
4216 		break;
4217 	}
4218 #endif /* INET6 */
4219 	}
4220 
4221 	PF_RULES_RLOCK();
4222 
4223 	while (r != NULL) {
4224 		counter_u64_add(r->evaluations, 1);
4225 		SDT_PROBE2(pf, eth, test_rule, test, r->nr, r);
4226 
4227 		if (pfi_kkif_match(r->kif, kif) == r->ifnot) {
4228 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4229 			    "kif");
4230 			r = r->skip[PFE_SKIP_IFP].ptr;
4231 		}
4232 		else if (r->direction && r->direction != dir) {
4233 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4234 			    "dir");
4235 			r = r->skip[PFE_SKIP_DIR].ptr;
4236 		}
4237 		else if (r->proto && r->proto != proto) {
4238 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4239 			    "proto");
4240 			r = r->skip[PFE_SKIP_PROTO].ptr;
4241 		}
4242 		else if (! pf_match_eth_addr(e->ether_shost, &r->src)) {
4243 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4244 			    "src");
4245 			r = r->skip[PFE_SKIP_SRC_ADDR].ptr;
4246 		}
4247 		else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) {
4248 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4249 			    "dst");
4250 			r = r->skip[PFE_SKIP_DST_ADDR].ptr;
4251 		}
4252 		else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af,
4253 		    r->ipsrc.neg, kif, M_GETFIB(m))) {
4254 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4255 			    "ip_src");
4256 			r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr;
4257 		}
4258 		else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af,
4259 		    r->ipdst.neg, kif, M_GETFIB(m))) {
4260 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4261 			    "ip_dst");
4262 			r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr;
4263 		}
4264 		else if (r->match_tag && !pf_match_eth_tag(m, r, &tag,
4265 		    mtag ? mtag->tag : 0)) {
4266 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4267 			    "match_tag");
4268 			r = TAILQ_NEXT(r, entries);
4269 		}
4270 		else {
4271 			if (r->tag)
4272 				tag = r->tag;
4273 			if (r->anchor == NULL) {
4274 				/* Rule matches */
4275 				rm = r;
4276 
4277 				SDT_PROBE2(pf, eth, test_rule, match, r->nr, r);
4278 
4279 				if (r->quick)
4280 					break;
4281 
4282 				r = TAILQ_NEXT(r, entries);
4283 			} else {
4284 				pf_step_into_keth_anchor(anchor_stack, &asd,
4285 				    &ruleset, &r, &a, &match);
4286 			}
4287 		}
4288 		if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd,
4289 		    &ruleset, &r, &a, &match))
4290 			break;
4291 	}
4292 
4293 	r = rm;
4294 
4295 	SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r);
4296 
4297 	/* Default to pass. */
4298 	if (r == NULL) {
4299 		PF_RULES_RUNLOCK();
4300 		return (PF_PASS);
4301 	}
4302 
4303 	/* Execute action. */
4304 	counter_u64_add(r->packets[dir == PF_OUT], 1);
4305 	counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL));
4306 	pf_update_timestamp(r);
4307 
4308 	/* Shortcut. Don't tag if we're just going to drop anyway. */
4309 	if (r->action == PF_DROP) {
4310 		PF_RULES_RUNLOCK();
4311 		return (PF_DROP);
4312 	}
4313 
4314 	if (tag > 0) {
4315 		if (mtag == NULL)
4316 			mtag = pf_get_mtag(m);
4317 		if (mtag == NULL) {
4318 			PF_RULES_RUNLOCK();
4319 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4320 			return (PF_DROP);
4321 		}
4322 		mtag->tag = tag;
4323 	}
4324 
4325 	if (r->qid != 0) {
4326 		if (mtag == NULL)
4327 			mtag = pf_get_mtag(m);
4328 		if (mtag == NULL) {
4329 			PF_RULES_RUNLOCK();
4330 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4331 			return (PF_DROP);
4332 		}
4333 		mtag->qid = r->qid;
4334 	}
4335 
4336 	action = r->action;
4337 	bridge_to = r->bridge_to;
4338 
4339 	/* Dummynet */
4340 	if (r->dnpipe) {
4341 		struct ip_fw_args dnflow;
4342 
4343 		/* Drop packet if dummynet is not loaded. */
4344 		if (ip_dn_io_ptr == NULL) {
4345 			PF_RULES_RUNLOCK();
4346 			m_freem(m);
4347 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4348 			return (PF_DROP);
4349 		}
4350 		if (mtag == NULL)
4351 			mtag = pf_get_mtag(m);
4352 		if (mtag == NULL) {
4353 			PF_RULES_RUNLOCK();
4354 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4355 			return (PF_DROP);
4356 		}
4357 
4358 		bzero(&dnflow, sizeof(dnflow));
4359 
4360 		/* We don't have port numbers here, so we set 0.  That means
4361 		 * that we'll be somewhat limited in distinguishing flows (i.e.
4362 		 * only based on IP addresses, not based on port numbers), but
4363 		 * it's better than nothing. */
4364 		dnflow.f_id.dst_port = 0;
4365 		dnflow.f_id.src_port = 0;
4366 		dnflow.f_id.proto = 0;
4367 
4368 		dnflow.rule.info = r->dnpipe;
4369 		dnflow.rule.info |= IPFW_IS_DUMMYNET;
4370 		if (r->dnflags & PFRULE_DN_IS_PIPE)
4371 			dnflow.rule.info |= IPFW_IS_PIPE;
4372 
4373 		dnflow.f_id.extra = dnflow.rule.info;
4374 
4375 		dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT;
4376 		dnflow.flags |= IPFW_ARGS_ETHER;
4377 		dnflow.ifp = kif->pfik_ifp;
4378 
4379 		switch (af) {
4380 		case AF_INET:
4381 			dnflow.f_id.addr_type = 4;
4382 			dnflow.f_id.src_ip = src->v4.s_addr;
4383 			dnflow.f_id.dst_ip = dst->v4.s_addr;
4384 			break;
4385 		case AF_INET6:
4386 			dnflow.flags |= IPFW_ARGS_IP6;
4387 			dnflow.f_id.addr_type = 6;
4388 			dnflow.f_id.src_ip6 = src->v6;
4389 			dnflow.f_id.dst_ip6 = dst->v6;
4390 			break;
4391 		}
4392 
4393 		PF_RULES_RUNLOCK();
4394 
4395 		mtag->flags |= PF_MTAG_FLAG_DUMMYNET;
4396 		ip_dn_io_ptr(m0, &dnflow);
4397 		if (*m0 != NULL)
4398 			pf_dummynet_flag_remove(m, mtag);
4399 	} else {
4400 		PF_RULES_RUNLOCK();
4401 	}
4402 
4403 	if (action == PF_PASS && bridge_to) {
4404 		pf_bridge_to(bridge_to->pfik_ifp, *m0);
4405 		*m0 = NULL; /* We've eaten the packet. */
4406 	}
4407 
4408 	return (action);
4409 }
4410 
4411 static int
4412 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, struct pfi_kkif *kif,
4413     struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_krule **am,
4414     struct pf_kruleset **rsm, struct inpcb *inp)
4415 {
4416 	struct pf_krule		*nr = NULL;
4417 	struct pf_addr		* const saddr = pd->src;
4418 	struct pf_addr		* const daddr = pd->dst;
4419 	sa_family_t		 af = pd->af;
4420 	struct pf_krule		*r, *a = NULL;
4421 	struct pf_kruleset	*ruleset = NULL;
4422 	struct pf_krule_slist	 match_rules;
4423 	struct pf_krule_item	*ri;
4424 	struct pf_ksrc_node	*nsn = NULL;
4425 	struct tcphdr		*th = &pd->hdr.tcp;
4426 	struct pf_state_key	*sk = NULL, *nk = NULL;
4427 	u_short			 reason;
4428 	int			 rewrite = 0, hdrlen = 0;
4429 	int			 tag = -1;
4430 	int			 asd = 0;
4431 	int			 match = 0;
4432 	int			 state_icmp = 0;
4433 	u_int16_t		 sport = 0, dport = 0;
4434 	u_int16_t		 bproto_sum = 0, bip_sum = 0;
4435 	u_int8_t		 icmptype = 0, icmpcode = 0;
4436 	struct pf_kanchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
4437 
4438 	PF_RULES_RASSERT();
4439 
4440 	if (inp != NULL) {
4441 		INP_LOCK_ASSERT(inp);
4442 		pd->lookup.uid = inp->inp_cred->cr_uid;
4443 		pd->lookup.gid = inp->inp_cred->cr_groups[0];
4444 		pd->lookup.done = 1;
4445 	}
4446 
4447 	switch (pd->proto) {
4448 	case IPPROTO_TCP:
4449 		sport = th->th_sport;
4450 		dport = th->th_dport;
4451 		hdrlen = sizeof(*th);
4452 		break;
4453 	case IPPROTO_UDP:
4454 		sport = pd->hdr.udp.uh_sport;
4455 		dport = pd->hdr.udp.uh_dport;
4456 		hdrlen = sizeof(pd->hdr.udp);
4457 		break;
4458 	case IPPROTO_SCTP:
4459 		sport = pd->hdr.sctp.src_port;
4460 		dport = pd->hdr.sctp.dest_port;
4461 		hdrlen = sizeof(pd->hdr.sctp);
4462 		break;
4463 #ifdef INET
4464 	case IPPROTO_ICMP:
4465 		if (pd->af != AF_INET)
4466 			break;
4467 		sport = dport = pd->hdr.icmp.icmp_id;
4468 		hdrlen = sizeof(pd->hdr.icmp);
4469 		icmptype = pd->hdr.icmp.icmp_type;
4470 		icmpcode = pd->hdr.icmp.icmp_code;
4471 
4472 		if (icmptype == ICMP_UNREACH ||
4473 		    icmptype == ICMP_SOURCEQUENCH ||
4474 		    icmptype == ICMP_REDIRECT ||
4475 		    icmptype == ICMP_TIMXCEED ||
4476 		    icmptype == ICMP_PARAMPROB)
4477 			state_icmp++;
4478 		break;
4479 #endif /* INET */
4480 #ifdef INET6
4481 	case IPPROTO_ICMPV6:
4482 		if (af != AF_INET6)
4483 			break;
4484 		sport = dport = pd->hdr.icmp6.icmp6_id;
4485 		hdrlen = sizeof(pd->hdr.icmp6);
4486 		icmptype = pd->hdr.icmp6.icmp6_type;
4487 		icmpcode = pd->hdr.icmp6.icmp6_code;
4488 
4489 		if (icmptype == ICMP6_DST_UNREACH ||
4490 		    icmptype == ICMP6_PACKET_TOO_BIG ||
4491 		    icmptype == ICMP6_TIME_EXCEEDED ||
4492 		    icmptype == ICMP6_PARAM_PROB)
4493 			state_icmp++;
4494 		break;
4495 #endif /* INET6 */
4496 	default:
4497 		sport = dport = hdrlen = 0;
4498 		break;
4499 	}
4500 
4501 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
4502 
4503 	/* check packet for BINAT/NAT/RDR */
4504 	if ((nr = pf_get_translation(pd, m, off, kif, &nsn, &sk,
4505 	    &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
4506 		KASSERT(sk != NULL, ("%s: null sk", __func__));
4507 		KASSERT(nk != NULL, ("%s: null nk", __func__));
4508 
4509 		if (nr->log) {
4510 			PFLOG_PACKET(kif, m, af, PF_PASS, PFRES_MATCH, nr, a,
4511 			    ruleset, pd, 1);
4512 		}
4513 
4514 		if (pd->ip_sum)
4515 			bip_sum = *pd->ip_sum;
4516 
4517 		switch (pd->proto) {
4518 		case IPPROTO_TCP:
4519 			bproto_sum = th->th_sum;
4520 			pd->proto_sum = &th->th_sum;
4521 
4522 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
4523 			    nk->port[pd->sidx] != sport) {
4524 				pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
4525 				    &th->th_sum, &nk->addr[pd->sidx],
4526 				    nk->port[pd->sidx], 0, af);
4527 				pd->sport = &th->th_sport;
4528 				sport = th->th_sport;
4529 			}
4530 
4531 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
4532 			    nk->port[pd->didx] != dport) {
4533 				pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
4534 				    &th->th_sum, &nk->addr[pd->didx],
4535 				    nk->port[pd->didx], 0, af);
4536 				dport = th->th_dport;
4537 				pd->dport = &th->th_dport;
4538 			}
4539 			rewrite++;
4540 			break;
4541 		case IPPROTO_UDP:
4542 			bproto_sum = pd->hdr.udp.uh_sum;
4543 			pd->proto_sum = &pd->hdr.udp.uh_sum;
4544 
4545 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
4546 			    nk->port[pd->sidx] != sport) {
4547 				pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport,
4548 				    pd->ip_sum, &pd->hdr.udp.uh_sum,
4549 				    &nk->addr[pd->sidx],
4550 				    nk->port[pd->sidx], 1, af);
4551 				sport = pd->hdr.udp.uh_sport;
4552 				pd->sport = &pd->hdr.udp.uh_sport;
4553 			}
4554 
4555 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
4556 			    nk->port[pd->didx] != dport) {
4557 				pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport,
4558 				    pd->ip_sum, &pd->hdr.udp.uh_sum,
4559 				    &nk->addr[pd->didx],
4560 				    nk->port[pd->didx], 1, af);
4561 				dport = pd->hdr.udp.uh_dport;
4562 				pd->dport = &pd->hdr.udp.uh_dport;
4563 			}
4564 			rewrite++;
4565 			break;
4566 		case IPPROTO_SCTP: {
4567 			uint16_t checksum = 0;
4568 
4569 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
4570 			    nk->port[pd->sidx] != sport) {
4571 				pf_change_ap(m, saddr, &pd->hdr.sctp.src_port,
4572 				    pd->ip_sum, &checksum,
4573 				    &nk->addr[pd->sidx],
4574 				    nk->port[pd->sidx], 1, af);
4575 			}
4576 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
4577 			    nk->port[pd->didx] != dport) {
4578 				pf_change_ap(m, daddr, &pd->hdr.sctp.dest_port,
4579 				    pd->ip_sum, &checksum,
4580 				    &nk->addr[pd->didx],
4581 				    nk->port[pd->didx], 1, af);
4582 			}
4583 			break;
4584 		}
4585 #ifdef INET
4586 		case IPPROTO_ICMP:
4587 			nk->port[0] = nk->port[1];
4588 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
4589 				pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
4590 				    nk->addr[pd->sidx].v4.s_addr, 0);
4591 
4592 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
4593 				pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
4594 				    nk->addr[pd->didx].v4.s_addr, 0);
4595 
4596 			if (nk->port[1] != pd->hdr.icmp.icmp_id) {
4597 				pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
4598 				    pd->hdr.icmp.icmp_cksum, sport,
4599 				    nk->port[1], 0);
4600 				pd->hdr.icmp.icmp_id = nk->port[1];
4601 				pd->sport = &pd->hdr.icmp.icmp_id;
4602 			}
4603 			m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
4604 			break;
4605 #endif /* INET */
4606 #ifdef INET6
4607 		case IPPROTO_ICMPV6:
4608 			nk->port[0] = nk->port[1];
4609 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
4610 				pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum,
4611 				    &nk->addr[pd->sidx], 0);
4612 
4613 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
4614 				pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum,
4615 				    &nk->addr[pd->didx], 0);
4616 			rewrite++;
4617 			break;
4618 #endif /* INET */
4619 		default:
4620 			switch (af) {
4621 #ifdef INET
4622 			case AF_INET:
4623 				if (PF_ANEQ(saddr,
4624 				    &nk->addr[pd->sidx], AF_INET))
4625 					pf_change_a(&saddr->v4.s_addr,
4626 					    pd->ip_sum,
4627 					    nk->addr[pd->sidx].v4.s_addr, 0);
4628 
4629 				if (PF_ANEQ(daddr,
4630 				    &nk->addr[pd->didx], AF_INET))
4631 					pf_change_a(&daddr->v4.s_addr,
4632 					    pd->ip_sum,
4633 					    nk->addr[pd->didx].v4.s_addr, 0);
4634 				break;
4635 #endif /* INET */
4636 #ifdef INET6
4637 			case AF_INET6:
4638 				if (PF_ANEQ(saddr,
4639 				    &nk->addr[pd->sidx], AF_INET6))
4640 					PF_ACPY(saddr, &nk->addr[pd->sidx], af);
4641 
4642 				if (PF_ANEQ(daddr,
4643 				    &nk->addr[pd->didx], AF_INET6))
4644 					PF_ACPY(daddr, &nk->addr[pd->didx], af);
4645 				break;
4646 #endif /* INET */
4647 			}
4648 			break;
4649 		}
4650 		if (nr->natpass)
4651 			r = NULL;
4652 		pd->nat_rule = nr;
4653 	}
4654 
4655 	SLIST_INIT(&match_rules);
4656 	while (r != NULL) {
4657 		pf_counter_u64_add(&r->evaluations, 1);
4658 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
4659 			r = r->skip[PF_SKIP_IFP].ptr;
4660 		else if (r->direction && r->direction != pd->dir)
4661 			r = r->skip[PF_SKIP_DIR].ptr;
4662 		else if (r->af && r->af != af)
4663 			r = r->skip[PF_SKIP_AF].ptr;
4664 		else if (r->proto && r->proto != pd->proto)
4665 			r = r->skip[PF_SKIP_PROTO].ptr;
4666 		else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
4667 		    r->src.neg, kif, M_GETFIB(m)))
4668 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
4669 		/* tcp/udp only. port_op always 0 in other cases */
4670 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
4671 		    r->src.port[0], r->src.port[1], sport))
4672 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
4673 		else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
4674 		    r->dst.neg, NULL, M_GETFIB(m)))
4675 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
4676 		/* tcp/udp only. port_op always 0 in other cases */
4677 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
4678 		    r->dst.port[0], r->dst.port[1], dport))
4679 			r = r->skip[PF_SKIP_DST_PORT].ptr;
4680 		/* icmp only. type always 0 in other cases */
4681 		else if (r->type && r->type != icmptype + 1)
4682 			r = TAILQ_NEXT(r, entries);
4683 		/* icmp only. type always 0 in other cases */
4684 		else if (r->code && r->code != icmpcode + 1)
4685 			r = TAILQ_NEXT(r, entries);
4686 		else if (r->tos && !(r->tos == pd->tos))
4687 			r = TAILQ_NEXT(r, entries);
4688 		else if (r->rule_flag & PFRULE_FRAGMENT)
4689 			r = TAILQ_NEXT(r, entries);
4690 		else if (pd->proto == IPPROTO_TCP &&
4691 		    (r->flagset & th->th_flags) != r->flags)
4692 			r = TAILQ_NEXT(r, entries);
4693 		/* tcp/udp only. uid.op always 0 in other cases */
4694 		else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
4695 		    pf_socket_lookup(pd, m), 1)) &&
4696 		    !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
4697 		    pd->lookup.uid))
4698 			r = TAILQ_NEXT(r, entries);
4699 		/* tcp/udp only. gid.op always 0 in other cases */
4700 		else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
4701 		    pf_socket_lookup(pd, m), 1)) &&
4702 		    !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
4703 		    pd->lookup.gid))
4704 			r = TAILQ_NEXT(r, entries);
4705 		else if (r->prio &&
4706 		    !pf_match_ieee8021q_pcp(r->prio, m))
4707 			r = TAILQ_NEXT(r, entries);
4708 		else if (r->prob &&
4709 		    r->prob <= arc4random())
4710 			r = TAILQ_NEXT(r, entries);
4711 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
4712 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
4713 			r = TAILQ_NEXT(r, entries);
4714 		else if (r->os_fingerprint != PF_OSFP_ANY &&
4715 		    (pd->proto != IPPROTO_TCP || !pf_osfp_match(
4716 		    pf_osfp_fingerprint(pd, m, off, th),
4717 		    r->os_fingerprint)))
4718 			r = TAILQ_NEXT(r, entries);
4719 		else {
4720 			if (r->tag)
4721 				tag = r->tag;
4722 			if (r->anchor == NULL) {
4723 				if (r->action == PF_MATCH) {
4724 					ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO);
4725 					if (ri == NULL) {
4726 						REASON_SET(&reason, PFRES_MEMORY);
4727 						goto cleanup;
4728 					}
4729 					ri->r = r;
4730 					SLIST_INSERT_HEAD(&match_rules, ri, entry);
4731 					pf_counter_u64_critical_enter();
4732 					pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
4733 					pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
4734 					pf_counter_u64_critical_exit();
4735 					pf_rule_to_actions(r, &pd->act);
4736 					if (r->log)
4737 						PFLOG_PACKET(kif, m, af,
4738 						    r->action, PFRES_MATCH, r,
4739 						    a, ruleset, pd, 1);
4740 				} else {
4741 					match = 1;
4742 					*rm = r;
4743 					*am = a;
4744 					*rsm = ruleset;
4745 				}
4746 				if ((*rm)->quick)
4747 					break;
4748 				r = TAILQ_NEXT(r, entries);
4749 			} else
4750 				pf_step_into_anchor(anchor_stack, &asd,
4751 				    &ruleset, PF_RULESET_FILTER, &r, &a,
4752 				    &match);
4753 		}
4754 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
4755 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
4756 			break;
4757 	}
4758 	r = *rm;
4759 	a = *am;
4760 	ruleset = *rsm;
4761 
4762 	REASON_SET(&reason, PFRES_MATCH);
4763 
4764 	/* apply actions for last matching pass/block rule */
4765 	pf_rule_to_actions(r, &pd->act);
4766 
4767 	if (r->log) {
4768 		if (rewrite)
4769 			m_copyback(m, off, hdrlen, pd->hdr.any);
4770 		PFLOG_PACKET(kif, m, af, r->action, reason, r, a, ruleset, pd, 1);
4771 	}
4772 
4773 	if ((r->action == PF_DROP) &&
4774 	    ((r->rule_flag & PFRULE_RETURNRST) ||
4775 	    (r->rule_flag & PFRULE_RETURNICMP) ||
4776 	    (r->rule_flag & PFRULE_RETURN))) {
4777 		pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum,
4778 		    bip_sum, hdrlen, &reason, r->rtableid);
4779 	}
4780 
4781 	if (r->action == PF_DROP)
4782 		goto cleanup;
4783 
4784 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
4785 		REASON_SET(&reason, PFRES_MEMORY);
4786 		goto cleanup;
4787 	}
4788 	if (pd->act.rtableid >= 0)
4789 		M_SETFIB(m, pd->act.rtableid);
4790 
4791 	if (!state_icmp && (r->keep_state || nr != NULL ||
4792 	    (pd->flags & PFDESC_TCP_NORM))) {
4793 		int action;
4794 		action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
4795 		    sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
4796 		    hdrlen, &match_rules);
4797 		if (action != PF_PASS) {
4798 			if (action == PF_DROP &&
4799 			    (r->rule_flag & PFRULE_RETURN))
4800 				pf_return(r, nr, pd, sk, off, m, th, kif,
4801 				    bproto_sum, bip_sum, hdrlen, &reason,
4802 				    pd->act.rtableid);
4803 			return (action);
4804 		}
4805 	} else {
4806 		while ((ri = SLIST_FIRST(&match_rules))) {
4807 			SLIST_REMOVE_HEAD(&match_rules, entry);
4808 			free(ri, M_PF_RULE_ITEM);
4809 		}
4810 
4811 		uma_zfree(V_pf_state_key_z, sk);
4812 		uma_zfree(V_pf_state_key_z, nk);
4813 	}
4814 
4815 	/* copy back packet headers if we performed NAT operations */
4816 	if (rewrite)
4817 		m_copyback(m, off, hdrlen, pd->hdr.any);
4818 
4819 	if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
4820 	    pd->dir == PF_OUT &&
4821 	    V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m))
4822 		/*
4823 		 * We want the state created, but we dont
4824 		 * want to send this in case a partner
4825 		 * firewall has to know about it to allow
4826 		 * replies through it.
4827 		 */
4828 		return (PF_DEFER);
4829 
4830 	return (PF_PASS);
4831 
4832 cleanup:
4833 	while ((ri = SLIST_FIRST(&match_rules))) {
4834 		SLIST_REMOVE_HEAD(&match_rules, entry);
4835 		free(ri, M_PF_RULE_ITEM);
4836 	}
4837 
4838 	uma_zfree(V_pf_state_key_z, sk);
4839 	uma_zfree(V_pf_state_key_z, nk);
4840 	return (PF_DROP);
4841 }
4842 
4843 static int
4844 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a,
4845     struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk,
4846     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
4847     u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm,
4848     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
4849     struct pf_krule_slist *match_rules)
4850 {
4851 	struct pf_kstate	*s = NULL;
4852 	struct pf_ksrc_node	*sn = NULL;
4853 	struct tcphdr		*th = &pd->hdr.tcp;
4854 	u_int16_t		 mss = V_tcp_mssdflt;
4855 	u_short			 reason, sn_reason;
4856 	struct pf_krule_item	*ri;
4857 
4858 	/* check maximums */
4859 	if (r->max_states &&
4860 	    (counter_u64_fetch(r->states_cur) >= r->max_states)) {
4861 		counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
4862 		REASON_SET(&reason, PFRES_MAXSTATES);
4863 		goto csfailed;
4864 	}
4865 	/* src node for filter rule */
4866 	if ((r->rule_flag & PFRULE_SRCTRACK ||
4867 	    r->rpool.opts & PF_POOL_STICKYADDR) &&
4868 	    (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) {
4869 		REASON_SET(&reason, sn_reason);
4870 		goto csfailed;
4871 	}
4872 	/* src node for translation rule */
4873 	if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
4874 	    (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx],
4875 	    pd->af)) != 0 ) {
4876 		REASON_SET(&reason, sn_reason);
4877 		goto csfailed;
4878 	}
4879 	s = pf_alloc_state(M_NOWAIT);
4880 	if (s == NULL) {
4881 		REASON_SET(&reason, PFRES_MEMORY);
4882 		goto csfailed;
4883 	}
4884 	s->rule.ptr = r;
4885 	s->nat_rule.ptr = nr;
4886 	s->anchor.ptr = a;
4887 	bcopy(match_rules, &s->match_rules, sizeof(s->match_rules));
4888 	memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions));
4889 
4890 	STATE_INC_COUNTERS(s);
4891 	if (r->allow_opts)
4892 		s->state_flags |= PFSTATE_ALLOWOPTS;
4893 	if (r->rule_flag & PFRULE_STATESLOPPY)
4894 		s->state_flags |= PFSTATE_SLOPPY;
4895 	if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */
4896 		s->state_flags |= PFSTATE_SCRUB_TCP;
4897 	if ((r->rule_flag & PFRULE_PFLOW) ||
4898 	    (nr != NULL && nr->rule_flag & PFRULE_PFLOW))
4899 		s->state_flags |= PFSTATE_PFLOW;
4900 
4901 	s->act.log = pd->act.log & PF_LOG_ALL;
4902 	s->sync_state = PFSYNC_S_NONE;
4903 	s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */
4904 
4905 	if (nr != NULL)
4906 		s->act.log |= nr->log & PF_LOG_ALL;
4907 	switch (pd->proto) {
4908 	case IPPROTO_TCP:
4909 		s->src.seqlo = ntohl(th->th_seq);
4910 		s->src.seqhi = s->src.seqlo + pd->p_len + 1;
4911 		if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
4912 		    r->keep_state == PF_STATE_MODULATE) {
4913 			/* Generate sequence number modulator */
4914 			if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
4915 			    0)
4916 				s->src.seqdiff = 1;
4917 			pf_change_proto_a(m, &th->th_seq, &th->th_sum,
4918 			    htonl(s->src.seqlo + s->src.seqdiff), 0);
4919 			*rewrite = 1;
4920 		} else
4921 			s->src.seqdiff = 0;
4922 		if (th->th_flags & TH_SYN) {
4923 			s->src.seqhi++;
4924 			s->src.wscale = pf_get_wscale(m, off,
4925 			    th->th_off, pd->af);
4926 		}
4927 		s->src.max_win = MAX(ntohs(th->th_win), 1);
4928 		if (s->src.wscale & PF_WSCALE_MASK) {
4929 			/* Remove scale factor from initial window */
4930 			int win = s->src.max_win;
4931 			win += 1 << (s->src.wscale & PF_WSCALE_MASK);
4932 			s->src.max_win = (win - 1) >>
4933 			    (s->src.wscale & PF_WSCALE_MASK);
4934 		}
4935 		if (th->th_flags & TH_FIN)
4936 			s->src.seqhi++;
4937 		s->dst.seqhi = 1;
4938 		s->dst.max_win = 1;
4939 		pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT);
4940 		pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED);
4941 		s->timeout = PFTM_TCP_FIRST_PACKET;
4942 		atomic_add_32(&V_pf_status.states_halfopen, 1);
4943 		break;
4944 	case IPPROTO_UDP:
4945 		pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE);
4946 		pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC);
4947 		s->timeout = PFTM_UDP_FIRST_PACKET;
4948 		break;
4949 	case IPPROTO_SCTP:
4950 		pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT);
4951 		pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED);
4952 		s->timeout = PFTM_SCTP_FIRST_PACKET;
4953 		break;
4954 	case IPPROTO_ICMP:
4955 #ifdef INET6
4956 	case IPPROTO_ICMPV6:
4957 #endif
4958 		s->timeout = PFTM_ICMP_FIRST_PACKET;
4959 		break;
4960 	default:
4961 		pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE);
4962 		pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC);
4963 		s->timeout = PFTM_OTHER_FIRST_PACKET;
4964 	}
4965 
4966 	if (r->rt) {
4967 		/* pf_map_addr increases the reason counters */
4968 		if ((reason = pf_map_addr(pd->af, r, pd->src, &s->rt_addr,
4969 		    &s->rt_kif, NULL, &sn)) != 0)
4970 			goto csfailed;
4971 		s->rt = r->rt;
4972 	}
4973 
4974 	s->creation = s->expire = pf_get_uptime();
4975 
4976 	if (sn != NULL)
4977 		s->src_node = sn;
4978 	if (nsn != NULL) {
4979 		/* XXX We only modify one side for now. */
4980 		PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
4981 		s->nat_src_node = nsn;
4982 	}
4983 	if (pd->proto == IPPROTO_TCP) {
4984 		if (s->state_flags & PFSTATE_SCRUB_TCP &&
4985 		    pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) {
4986 			REASON_SET(&reason, PFRES_MEMORY);
4987 			goto drop;
4988 		}
4989 		if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub &&
4990 		    pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
4991 		    &s->src, &s->dst, rewrite)) {
4992 			/* This really shouldn't happen!!! */
4993 			DPFPRINTF(PF_DEBUG_URGENT,
4994 			    ("pf_normalize_tcp_stateful failed on first "
4995 			     "pkt\n"));
4996 			goto drop;
4997 		}
4998 	} else if (pd->proto == IPPROTO_SCTP) {
4999 		if (pf_normalize_sctp_init(m, off, pd, &s->src, &s->dst))
5000 			goto drop;
5001 		if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP)))
5002 			goto drop;
5003 	}
5004 	s->direction = pd->dir;
5005 
5006 	/*
5007 	 * sk/nk could already been setup by pf_get_translation().
5008 	 */
5009 	if (nr == NULL) {
5010 		KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
5011 		    __func__, nr, sk, nk));
5012 		sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
5013 		if (sk == NULL)
5014 			goto csfailed;
5015 		nk = sk;
5016 	} else
5017 		KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
5018 		    __func__, nr, sk, nk));
5019 
5020 	/* Swap sk/nk for PF_OUT. */
5021 	if (pf_state_insert(BOUND_IFACE(r, kif, pd), kif,
5022 	    (pd->dir == PF_IN) ? sk : nk,
5023 	    (pd->dir == PF_IN) ? nk : sk, s)) {
5024 		REASON_SET(&reason, PFRES_STATEINS);
5025 		goto drop;
5026 	} else
5027 		*sm = s;
5028 
5029 	if (tag > 0)
5030 		s->tag = tag;
5031 	if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
5032 	    TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
5033 		pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC);
5034 		/* undo NAT changes, if they have taken place */
5035 		if (nr != NULL) {
5036 			struct pf_state_key *skt = s->key[PF_SK_WIRE];
5037 			if (pd->dir == PF_OUT)
5038 				skt = s->key[PF_SK_STACK];
5039 			PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
5040 			PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
5041 			if (pd->sport)
5042 				*pd->sport = skt->port[pd->sidx];
5043 			if (pd->dport)
5044 				*pd->dport = skt->port[pd->didx];
5045 			if (pd->proto_sum)
5046 				*pd->proto_sum = bproto_sum;
5047 			if (pd->ip_sum)
5048 				*pd->ip_sum = bip_sum;
5049 			m_copyback(m, off, hdrlen, pd->hdr.any);
5050 		}
5051 		s->src.seqhi = htonl(arc4random());
5052 		/* Find mss option */
5053 		int rtid = M_GETFIB(m);
5054 		mss = pf_get_mss(m, off, th->th_off, pd->af);
5055 		mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
5056 		mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
5057 		s->src.mss = mss;
5058 		pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport,
5059 		    th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
5060 		    TH_SYN|TH_ACK, 0, s->src.mss, 0, true, 0, 0,
5061 		    pd->act.rtableid);
5062 		REASON_SET(&reason, PFRES_SYNPROXY);
5063 		return (PF_SYNPROXY_DROP);
5064 	}
5065 
5066 	return (PF_PASS);
5067 
5068 csfailed:
5069 	while ((ri = SLIST_FIRST(match_rules))) {
5070 		SLIST_REMOVE_HEAD(match_rules, entry);
5071 		free(ri, M_PF_RULE_ITEM);
5072 	}
5073 
5074 	uma_zfree(V_pf_state_key_z, sk);
5075 	uma_zfree(V_pf_state_key_z, nk);
5076 
5077 	if (sn != NULL) {
5078 		PF_SRC_NODE_LOCK(sn);
5079 		if (--sn->states == 0 && sn->expire == 0) {
5080 			pf_unlink_src_node(sn);
5081 			uma_zfree(V_pf_sources_z, sn);
5082 			counter_u64_add(
5083 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
5084 		}
5085 		PF_SRC_NODE_UNLOCK(sn);
5086 	}
5087 
5088 	if (nsn != sn && nsn != NULL) {
5089 		PF_SRC_NODE_LOCK(nsn);
5090 		if (--nsn->states == 0 && nsn->expire == 0) {
5091 			pf_unlink_src_node(nsn);
5092 			uma_zfree(V_pf_sources_z, nsn);
5093 			counter_u64_add(
5094 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
5095 		}
5096 		PF_SRC_NODE_UNLOCK(nsn);
5097 	}
5098 
5099 drop:
5100 	if (s != NULL) {
5101 		pf_src_tree_remove_state(s);
5102 		s->timeout = PFTM_UNLINKED;
5103 		STATE_DEC_COUNTERS(s);
5104 		pf_free_state(s);
5105 	}
5106 
5107 	return (PF_DROP);
5108 }
5109 
5110 static int
5111 pf_test_fragment(struct pf_krule **rm, struct pfi_kkif *kif,
5112     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am,
5113     struct pf_kruleset **rsm)
5114 {
5115 	struct pf_krule		*r, *a = NULL;
5116 	struct pf_kruleset	*ruleset = NULL;
5117 	struct pf_krule_slist	 match_rules;
5118 	struct pf_krule_item	*ri;
5119 	sa_family_t		 af = pd->af;
5120 	u_short			 reason;
5121 	int			 tag = -1;
5122 	int			 asd = 0;
5123 	int			 match = 0;
5124 	struct pf_kanchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
5125 
5126 	PF_RULES_RASSERT();
5127 
5128 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
5129 	SLIST_INIT(&match_rules);
5130 	while (r != NULL) {
5131 		pf_counter_u64_add(&r->evaluations, 1);
5132 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
5133 			r = r->skip[PF_SKIP_IFP].ptr;
5134 		else if (r->direction && r->direction != pd->dir)
5135 			r = r->skip[PF_SKIP_DIR].ptr;
5136 		else if (r->af && r->af != af)
5137 			r = r->skip[PF_SKIP_AF].ptr;
5138 		else if (r->proto && r->proto != pd->proto)
5139 			r = r->skip[PF_SKIP_PROTO].ptr;
5140 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
5141 		    r->src.neg, kif, M_GETFIB(m)))
5142 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
5143 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
5144 		    r->dst.neg, NULL, M_GETFIB(m)))
5145 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
5146 		else if (r->tos && !(r->tos == pd->tos))
5147 			r = TAILQ_NEXT(r, entries);
5148 		else if (r->os_fingerprint != PF_OSFP_ANY)
5149 			r = TAILQ_NEXT(r, entries);
5150 		else if (pd->proto == IPPROTO_UDP &&
5151 		    (r->src.port_op || r->dst.port_op))
5152 			r = TAILQ_NEXT(r, entries);
5153 		else if (pd->proto == IPPROTO_TCP &&
5154 		    (r->src.port_op || r->dst.port_op || r->flagset))
5155 			r = TAILQ_NEXT(r, entries);
5156 		else if ((pd->proto == IPPROTO_ICMP ||
5157 		    pd->proto == IPPROTO_ICMPV6) &&
5158 		    (r->type || r->code))
5159 			r = TAILQ_NEXT(r, entries);
5160 		else if (r->prio &&
5161 		    !pf_match_ieee8021q_pcp(r->prio, m))
5162 			r = TAILQ_NEXT(r, entries);
5163 		else if (r->prob && r->prob <=
5164 		    (arc4random() % (UINT_MAX - 1) + 1))
5165 			r = TAILQ_NEXT(r, entries);
5166 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
5167 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
5168 			r = TAILQ_NEXT(r, entries);
5169 		else {
5170 			if (r->anchor == NULL) {
5171 				if (r->action == PF_MATCH) {
5172 					ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO);
5173 					if (ri == NULL) {
5174 						REASON_SET(&reason, PFRES_MEMORY);
5175 						goto cleanup;
5176 					}
5177 					ri->r = r;
5178 					SLIST_INSERT_HEAD(&match_rules, ri, entry);
5179 					pf_counter_u64_critical_enter();
5180 					pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
5181 					pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
5182 					pf_counter_u64_critical_exit();
5183 					pf_rule_to_actions(r, &pd->act);
5184 					if (r->log)
5185 						PFLOG_PACKET(kif, m, af,
5186 						    r->action, PFRES_MATCH, r,
5187 						    a, ruleset, pd, 1);
5188 				} else {
5189 					match = 1;
5190 					*rm = r;
5191 					*am = a;
5192 					*rsm = ruleset;
5193 				}
5194 				if ((*rm)->quick)
5195 					break;
5196 				r = TAILQ_NEXT(r, entries);
5197 			} else
5198 				pf_step_into_anchor(anchor_stack, &asd,
5199 				    &ruleset, PF_RULESET_FILTER, &r, &a,
5200 				    &match);
5201 		}
5202 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
5203 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
5204 			break;
5205 	}
5206 	r = *rm;
5207 	a = *am;
5208 	ruleset = *rsm;
5209 
5210 	REASON_SET(&reason, PFRES_MATCH);
5211 
5212 	/* apply actions for last matching pass/block rule */
5213 	pf_rule_to_actions(r, &pd->act);
5214 
5215 	if (r->log)
5216 		PFLOG_PACKET(kif, m, af, r->action, reason, r, a, ruleset, pd, 1);
5217 
5218 	if (r->action != PF_PASS)
5219 		return (PF_DROP);
5220 
5221 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
5222 		REASON_SET(&reason, PFRES_MEMORY);
5223 		goto cleanup;
5224 	}
5225 
5226 	return (PF_PASS);
5227 
5228 cleanup:
5229 	while ((ri = SLIST_FIRST(&match_rules))) {
5230 		SLIST_REMOVE_HEAD(&match_rules, entry);
5231 		free(ri, M_PF_RULE_ITEM);
5232 	}
5233 
5234 	return (PF_DROP);
5235 }
5236 
5237 static int
5238 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif,
5239     struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason,
5240     int *copyback)
5241 {
5242 	struct tcphdr		*th = &pd->hdr.tcp;
5243 	struct pf_state_peer	*src, *dst;
5244 	u_int16_t		 win = ntohs(th->th_win);
5245 	u_int32_t		 ack, end, seq, orig_seq;
5246 	u_int8_t		 sws, dws, psrc, pdst;
5247 	int			 ackskew;
5248 
5249 	if (pd->dir == (*state)->direction) {
5250 		src = &(*state)->src;
5251 		dst = &(*state)->dst;
5252 		psrc = PF_PEER_SRC;
5253 		pdst = PF_PEER_DST;
5254 	} else {
5255 		src = &(*state)->dst;
5256 		dst = &(*state)->src;
5257 		psrc = PF_PEER_DST;
5258 		pdst = PF_PEER_SRC;
5259 	}
5260 
5261 	if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
5262 		sws = src->wscale & PF_WSCALE_MASK;
5263 		dws = dst->wscale & PF_WSCALE_MASK;
5264 	} else
5265 		sws = dws = 0;
5266 
5267 	/*
5268 	 * Sequence tracking algorithm from Guido van Rooij's paper:
5269 	 *   http://www.madison-gurkha.com/publications/tcp_filtering/
5270 	 *	tcp_filtering.ps
5271 	 */
5272 
5273 	orig_seq = seq = ntohl(th->th_seq);
5274 	if (src->seqlo == 0) {
5275 		/* First packet from this end. Set its state */
5276 
5277 		if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) &&
5278 		    src->scrub == NULL) {
5279 			if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
5280 				REASON_SET(reason, PFRES_MEMORY);
5281 				return (PF_DROP);
5282 			}
5283 		}
5284 
5285 		/* Deferred generation of sequence number modulator */
5286 		if (dst->seqdiff && !src->seqdiff) {
5287 			/* use random iss for the TCP server */
5288 			while ((src->seqdiff = arc4random() - seq) == 0)
5289 				;
5290 			ack = ntohl(th->th_ack) - dst->seqdiff;
5291 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
5292 			    src->seqdiff), 0);
5293 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
5294 			*copyback = 1;
5295 		} else {
5296 			ack = ntohl(th->th_ack);
5297 		}
5298 
5299 		end = seq + pd->p_len;
5300 		if (th->th_flags & TH_SYN) {
5301 			end++;
5302 			if (dst->wscale & PF_WSCALE_FLAG) {
5303 				src->wscale = pf_get_wscale(m, off, th->th_off,
5304 				    pd->af);
5305 				if (src->wscale & PF_WSCALE_FLAG) {
5306 					/* Remove scale factor from initial
5307 					 * window */
5308 					sws = src->wscale & PF_WSCALE_MASK;
5309 					win = ((u_int32_t)win + (1 << sws) - 1)
5310 					    >> sws;
5311 					dws = dst->wscale & PF_WSCALE_MASK;
5312 				} else {
5313 					/* fixup other window */
5314 					dst->max_win <<= dst->wscale &
5315 					    PF_WSCALE_MASK;
5316 					/* in case of a retrans SYN|ACK */
5317 					dst->wscale = 0;
5318 				}
5319 			}
5320 		}
5321 		if (th->th_flags & TH_FIN)
5322 			end++;
5323 
5324 		src->seqlo = seq;
5325 		if (src->state < TCPS_SYN_SENT)
5326 			pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
5327 
5328 		/*
5329 		 * May need to slide the window (seqhi may have been set by
5330 		 * the crappy stack check or if we picked up the connection
5331 		 * after establishment)
5332 		 */
5333 		if (src->seqhi == 1 ||
5334 		    SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
5335 			src->seqhi = end + MAX(1, dst->max_win << dws);
5336 		if (win > src->max_win)
5337 			src->max_win = win;
5338 
5339 	} else {
5340 		ack = ntohl(th->th_ack) - dst->seqdiff;
5341 		if (src->seqdiff) {
5342 			/* Modulate sequence numbers */
5343 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
5344 			    src->seqdiff), 0);
5345 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
5346 			*copyback = 1;
5347 		}
5348 		end = seq + pd->p_len;
5349 		if (th->th_flags & TH_SYN)
5350 			end++;
5351 		if (th->th_flags & TH_FIN)
5352 			end++;
5353 	}
5354 
5355 	if ((th->th_flags & TH_ACK) == 0) {
5356 		/* Let it pass through the ack skew check */
5357 		ack = dst->seqlo;
5358 	} else if ((ack == 0 &&
5359 	    (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
5360 	    /* broken tcp stacks do not set ack */
5361 	    (dst->state < TCPS_SYN_SENT)) {
5362 		/*
5363 		 * Many stacks (ours included) will set the ACK number in an
5364 		 * FIN|ACK if the SYN times out -- no sequence to ACK.
5365 		 */
5366 		ack = dst->seqlo;
5367 	}
5368 
5369 	if (seq == end) {
5370 		/* Ease sequencing restrictions on no data packets */
5371 		seq = src->seqlo;
5372 		end = seq;
5373 	}
5374 
5375 	ackskew = dst->seqlo - ack;
5376 
5377 	/*
5378 	 * Need to demodulate the sequence numbers in any TCP SACK options
5379 	 * (Selective ACK). We could optionally validate the SACK values
5380 	 * against the current ACK window, either forwards or backwards, but
5381 	 * I'm not confident that SACK has been implemented properly
5382 	 * everywhere. It wouldn't surprise me if several stacks accidentally
5383 	 * SACK too far backwards of previously ACKed data. There really aren't
5384 	 * any security implications of bad SACKing unless the target stack
5385 	 * doesn't validate the option length correctly. Someone trying to
5386 	 * spoof into a TCP connection won't bother blindly sending SACK
5387 	 * options anyway.
5388 	 */
5389 	if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
5390 		if (pf_modulate_sack(m, off, pd, th, dst))
5391 			*copyback = 1;
5392 	}
5393 
5394 #define	MAXACKWINDOW (0xffff + 1500)	/* 1500 is an arbitrary fudge factor */
5395 	if (SEQ_GEQ(src->seqhi, end) &&
5396 	    /* Last octet inside other's window space */
5397 	    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
5398 	    /* Retrans: not more than one window back */
5399 	    (ackskew >= -MAXACKWINDOW) &&
5400 	    /* Acking not more than one reassembled fragment backwards */
5401 	    (ackskew <= (MAXACKWINDOW << sws)) &&
5402 	    /* Acking not more than one window forward */
5403 	    ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
5404 	    (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) {
5405 	    /* Require an exact/+1 sequence match on resets when possible */
5406 
5407 		if (dst->scrub || src->scrub) {
5408 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
5409 			    *state, src, dst, copyback))
5410 				return (PF_DROP);
5411 		}
5412 
5413 		/* update max window */
5414 		if (src->max_win < win)
5415 			src->max_win = win;
5416 		/* synchronize sequencing */
5417 		if (SEQ_GT(end, src->seqlo))
5418 			src->seqlo = end;
5419 		/* slide the window of what the other end can send */
5420 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
5421 			dst->seqhi = ack + MAX((win << sws), 1);
5422 
5423 		/* update states */
5424 		if (th->th_flags & TH_SYN)
5425 			if (src->state < TCPS_SYN_SENT)
5426 				pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
5427 		if (th->th_flags & TH_FIN)
5428 			if (src->state < TCPS_CLOSING)
5429 				pf_set_protostate(*state, psrc, TCPS_CLOSING);
5430 		if (th->th_flags & TH_ACK) {
5431 			if (dst->state == TCPS_SYN_SENT) {
5432 				pf_set_protostate(*state, pdst,
5433 				    TCPS_ESTABLISHED);
5434 				if (src->state == TCPS_ESTABLISHED &&
5435 				    (*state)->src_node != NULL &&
5436 				    pf_src_connlimit(state)) {
5437 					REASON_SET(reason, PFRES_SRCLIMIT);
5438 					return (PF_DROP);
5439 				}
5440 			} else if (dst->state == TCPS_CLOSING)
5441 				pf_set_protostate(*state, pdst,
5442 				    TCPS_FIN_WAIT_2);
5443 		}
5444 		if (th->th_flags & TH_RST)
5445 			pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
5446 
5447 		/* update expire time */
5448 		(*state)->expire = pf_get_uptime();
5449 		if (src->state >= TCPS_FIN_WAIT_2 &&
5450 		    dst->state >= TCPS_FIN_WAIT_2)
5451 			(*state)->timeout = PFTM_TCP_CLOSED;
5452 		else if (src->state >= TCPS_CLOSING &&
5453 		    dst->state >= TCPS_CLOSING)
5454 			(*state)->timeout = PFTM_TCP_FIN_WAIT;
5455 		else if (src->state < TCPS_ESTABLISHED ||
5456 		    dst->state < TCPS_ESTABLISHED)
5457 			(*state)->timeout = PFTM_TCP_OPENING;
5458 		else if (src->state >= TCPS_CLOSING ||
5459 		    dst->state >= TCPS_CLOSING)
5460 			(*state)->timeout = PFTM_TCP_CLOSING;
5461 		else
5462 			(*state)->timeout = PFTM_TCP_ESTABLISHED;
5463 
5464 		/* Fall through to PASS packet */
5465 
5466 	} else if ((dst->state < TCPS_SYN_SENT ||
5467 		dst->state >= TCPS_FIN_WAIT_2 ||
5468 		src->state >= TCPS_FIN_WAIT_2) &&
5469 	    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
5470 	    /* Within a window forward of the originating packet */
5471 	    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
5472 	    /* Within a window backward of the originating packet */
5473 
5474 		/*
5475 		 * This currently handles three situations:
5476 		 *  1) Stupid stacks will shotgun SYNs before their peer
5477 		 *     replies.
5478 		 *  2) When PF catches an already established stream (the
5479 		 *     firewall rebooted, the state table was flushed, routes
5480 		 *     changed...)
5481 		 *  3) Packets get funky immediately after the connection
5482 		 *     closes (this should catch Solaris spurious ACK|FINs
5483 		 *     that web servers like to spew after a close)
5484 		 *
5485 		 * This must be a little more careful than the above code
5486 		 * since packet floods will also be caught here. We don't
5487 		 * update the TTL here to mitigate the damage of a packet
5488 		 * flood and so the same code can handle awkward establishment
5489 		 * and a loosened connection close.
5490 		 * In the establishment case, a correct peer response will
5491 		 * validate the connection, go through the normal state code
5492 		 * and keep updating the state TTL.
5493 		 */
5494 
5495 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
5496 			printf("pf: loose state match: ");
5497 			pf_print_state(*state);
5498 			pf_print_flags(th->th_flags);
5499 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
5500 			    "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
5501 			    pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
5502 			    (unsigned long long)(*state)->packets[1],
5503 			    pd->dir == PF_IN ? "in" : "out",
5504 			    pd->dir == (*state)->direction ? "fwd" : "rev");
5505 		}
5506 
5507 		if (dst->scrub || src->scrub) {
5508 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
5509 			    *state, src, dst, copyback))
5510 				return (PF_DROP);
5511 		}
5512 
5513 		/* update max window */
5514 		if (src->max_win < win)
5515 			src->max_win = win;
5516 		/* synchronize sequencing */
5517 		if (SEQ_GT(end, src->seqlo))
5518 			src->seqlo = end;
5519 		/* slide the window of what the other end can send */
5520 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
5521 			dst->seqhi = ack + MAX((win << sws), 1);
5522 
5523 		/*
5524 		 * Cannot set dst->seqhi here since this could be a shotgunned
5525 		 * SYN and not an already established connection.
5526 		 */
5527 
5528 		if (th->th_flags & TH_FIN)
5529 			if (src->state < TCPS_CLOSING)
5530 				pf_set_protostate(*state, psrc, TCPS_CLOSING);
5531 		if (th->th_flags & TH_RST)
5532 			pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
5533 
5534 		/* Fall through to PASS packet */
5535 
5536 	} else {
5537 		if ((*state)->dst.state == TCPS_SYN_SENT &&
5538 		    (*state)->src.state == TCPS_SYN_SENT) {
5539 			/* Send RST for state mismatches during handshake */
5540 			if (!(th->th_flags & TH_RST))
5541 				pf_send_tcp((*state)->rule.ptr, pd->af,
5542 				    pd->dst, pd->src, th->th_dport,
5543 				    th->th_sport, ntohl(th->th_ack), 0,
5544 				    TH_RST, 0, 0,
5545 				    (*state)->rule.ptr->return_ttl, true, 0, 0,
5546 				    (*state)->act.rtableid);
5547 			src->seqlo = 0;
5548 			src->seqhi = 1;
5549 			src->max_win = 1;
5550 		} else if (V_pf_status.debug >= PF_DEBUG_MISC) {
5551 			printf("pf: BAD state: ");
5552 			pf_print_state(*state);
5553 			pf_print_flags(th->th_flags);
5554 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
5555 			    "pkts=%llu:%llu dir=%s,%s\n",
5556 			    seq, orig_seq, ack, pd->p_len, ackskew,
5557 			    (unsigned long long)(*state)->packets[0],
5558 			    (unsigned long long)(*state)->packets[1],
5559 			    pd->dir == PF_IN ? "in" : "out",
5560 			    pd->dir == (*state)->direction ? "fwd" : "rev");
5561 			printf("pf: State failure on: %c %c %c %c | %c %c\n",
5562 			    SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
5563 			    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
5564 			    ' ': '2',
5565 			    (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
5566 			    (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
5567 			    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
5568 			    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
5569 		}
5570 		REASON_SET(reason, PFRES_BADSTATE);
5571 		return (PF_DROP);
5572 	}
5573 
5574 	return (PF_PASS);
5575 }
5576 
5577 static int
5578 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason)
5579 {
5580 	struct tcphdr		*th = &pd->hdr.tcp;
5581 	struct pf_state_peer	*src, *dst;
5582 	u_int8_t		 psrc, pdst;
5583 
5584 	if (pd->dir == (*state)->direction) {
5585 		src = &(*state)->src;
5586 		dst = &(*state)->dst;
5587 		psrc = PF_PEER_SRC;
5588 		pdst = PF_PEER_DST;
5589 	} else {
5590 		src = &(*state)->dst;
5591 		dst = &(*state)->src;
5592 		psrc = PF_PEER_DST;
5593 		pdst = PF_PEER_SRC;
5594 	}
5595 
5596 	if (th->th_flags & TH_SYN)
5597 		if (src->state < TCPS_SYN_SENT)
5598 			pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
5599 	if (th->th_flags & TH_FIN)
5600 		if (src->state < TCPS_CLOSING)
5601 			pf_set_protostate(*state, psrc, TCPS_CLOSING);
5602 	if (th->th_flags & TH_ACK) {
5603 		if (dst->state == TCPS_SYN_SENT) {
5604 			pf_set_protostate(*state, pdst, TCPS_ESTABLISHED);
5605 			if (src->state == TCPS_ESTABLISHED &&
5606 			    (*state)->src_node != NULL &&
5607 			    pf_src_connlimit(state)) {
5608 				REASON_SET(reason, PFRES_SRCLIMIT);
5609 				return (PF_DROP);
5610 			}
5611 		} else if (dst->state == TCPS_CLOSING) {
5612 			pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2);
5613 		} else if (src->state == TCPS_SYN_SENT &&
5614 		    dst->state < TCPS_SYN_SENT) {
5615 			/*
5616 			 * Handle a special sloppy case where we only see one
5617 			 * half of the connection. If there is a ACK after
5618 			 * the initial SYN without ever seeing a packet from
5619 			 * the destination, set the connection to established.
5620 			 */
5621 			pf_set_protostate(*state, PF_PEER_BOTH,
5622 			    TCPS_ESTABLISHED);
5623 			dst->state = src->state = TCPS_ESTABLISHED;
5624 			if ((*state)->src_node != NULL &&
5625 			    pf_src_connlimit(state)) {
5626 				REASON_SET(reason, PFRES_SRCLIMIT);
5627 				return (PF_DROP);
5628 			}
5629 		} else if (src->state == TCPS_CLOSING &&
5630 		    dst->state == TCPS_ESTABLISHED &&
5631 		    dst->seqlo == 0) {
5632 			/*
5633 			 * Handle the closing of half connections where we
5634 			 * don't see the full bidirectional FIN/ACK+ACK
5635 			 * handshake.
5636 			 */
5637 			pf_set_protostate(*state, pdst, TCPS_CLOSING);
5638 		}
5639 	}
5640 	if (th->th_flags & TH_RST)
5641 		pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
5642 
5643 	/* update expire time */
5644 	(*state)->expire = pf_get_uptime();
5645 	if (src->state >= TCPS_FIN_WAIT_2 &&
5646 	    dst->state >= TCPS_FIN_WAIT_2)
5647 		(*state)->timeout = PFTM_TCP_CLOSED;
5648 	else if (src->state >= TCPS_CLOSING &&
5649 	    dst->state >= TCPS_CLOSING)
5650 		(*state)->timeout = PFTM_TCP_FIN_WAIT;
5651 	else if (src->state < TCPS_ESTABLISHED ||
5652 	    dst->state < TCPS_ESTABLISHED)
5653 		(*state)->timeout = PFTM_TCP_OPENING;
5654 	else if (src->state >= TCPS_CLOSING ||
5655 	    dst->state >= TCPS_CLOSING)
5656 		(*state)->timeout = PFTM_TCP_CLOSING;
5657 	else
5658 		(*state)->timeout = PFTM_TCP_ESTABLISHED;
5659 
5660 	return (PF_PASS);
5661 }
5662 
5663 static int
5664 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason)
5665 {
5666 	struct pf_state_key	*sk = (*state)->key[pd->didx];
5667 	struct tcphdr		*th = &pd->hdr.tcp;
5668 
5669 	if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
5670 		if (pd->dir != (*state)->direction) {
5671 			REASON_SET(reason, PFRES_SYNPROXY);
5672 			return (PF_SYNPROXY_DROP);
5673 		}
5674 		if (th->th_flags & TH_SYN) {
5675 			if (ntohl(th->th_seq) != (*state)->src.seqlo) {
5676 				REASON_SET(reason, PFRES_SYNPROXY);
5677 				return (PF_DROP);
5678 			}
5679 			pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
5680 			    pd->src, th->th_dport, th->th_sport,
5681 			    (*state)->src.seqhi, ntohl(th->th_seq) + 1,
5682 			    TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, true, 0, 0,
5683 			    (*state)->act.rtableid);
5684 			REASON_SET(reason, PFRES_SYNPROXY);
5685 			return (PF_SYNPROXY_DROP);
5686 		} else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK ||
5687 		    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
5688 		    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
5689 			REASON_SET(reason, PFRES_SYNPROXY);
5690 			return (PF_DROP);
5691 		} else if ((*state)->src_node != NULL &&
5692 		    pf_src_connlimit(state)) {
5693 			REASON_SET(reason, PFRES_SRCLIMIT);
5694 			return (PF_DROP);
5695 		} else
5696 			pf_set_protostate(*state, PF_PEER_SRC,
5697 			    PF_TCPS_PROXY_DST);
5698 	}
5699 	if ((*state)->src.state == PF_TCPS_PROXY_DST) {
5700 		if (pd->dir == (*state)->direction) {
5701 			if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
5702 			    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
5703 			    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
5704 				REASON_SET(reason, PFRES_SYNPROXY);
5705 				return (PF_DROP);
5706 			}
5707 			(*state)->src.max_win = MAX(ntohs(th->th_win), 1);
5708 			if ((*state)->dst.seqhi == 1)
5709 				(*state)->dst.seqhi = htonl(arc4random());
5710 			pf_send_tcp((*state)->rule.ptr, pd->af,
5711 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
5712 			    sk->port[pd->sidx], sk->port[pd->didx],
5713 			    (*state)->dst.seqhi, 0, TH_SYN, 0,
5714 			    (*state)->src.mss, 0, false, (*state)->tag, 0,
5715 			    (*state)->act.rtableid);
5716 			REASON_SET(reason, PFRES_SYNPROXY);
5717 			return (PF_SYNPROXY_DROP);
5718 		} else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
5719 		    (TH_SYN|TH_ACK)) ||
5720 		    (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
5721 			REASON_SET(reason, PFRES_SYNPROXY);
5722 			return (PF_DROP);
5723 		} else {
5724 			(*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
5725 			(*state)->dst.seqlo = ntohl(th->th_seq);
5726 			pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
5727 			    pd->src, th->th_dport, th->th_sport,
5728 			    ntohl(th->th_ack), ntohl(th->th_seq) + 1,
5729 			    TH_ACK, (*state)->src.max_win, 0, 0, false,
5730 			    (*state)->tag, 0, (*state)->act.rtableid);
5731 			pf_send_tcp((*state)->rule.ptr, pd->af,
5732 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
5733 			    sk->port[pd->sidx], sk->port[pd->didx],
5734 			    (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
5735 			    TH_ACK, (*state)->dst.max_win, 0, 0, true, 0, 0,
5736 			    (*state)->act.rtableid);
5737 			(*state)->src.seqdiff = (*state)->dst.seqhi -
5738 			    (*state)->src.seqlo;
5739 			(*state)->dst.seqdiff = (*state)->src.seqhi -
5740 			    (*state)->dst.seqlo;
5741 			(*state)->src.seqhi = (*state)->src.seqlo +
5742 			    (*state)->dst.max_win;
5743 			(*state)->dst.seqhi = (*state)->dst.seqlo +
5744 			    (*state)->src.max_win;
5745 			(*state)->src.wscale = (*state)->dst.wscale = 0;
5746 			pf_set_protostate(*state, PF_PEER_BOTH,
5747 			    TCPS_ESTABLISHED);
5748 			REASON_SET(reason, PFRES_SYNPROXY);
5749 			return (PF_SYNPROXY_DROP);
5750 		}
5751 	}
5752 
5753 	return (PF_PASS);
5754 }
5755 
5756 static int
5757 pf_test_state_tcp(struct pf_kstate **state, struct pfi_kkif *kif,
5758     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
5759     u_short *reason)
5760 {
5761 	struct pf_state_key_cmp	 key;
5762 	struct tcphdr		*th = &pd->hdr.tcp;
5763 	int			 copyback = 0;
5764 	int			 action;
5765 	struct pf_state_peer	*src, *dst;
5766 
5767 	bzero(&key, sizeof(key));
5768 	key.af = pd->af;
5769 	key.proto = IPPROTO_TCP;
5770 	if (pd->dir == PF_IN)	{	/* wire side, straight */
5771 		PF_ACPY(&key.addr[0], pd->src, key.af);
5772 		PF_ACPY(&key.addr[1], pd->dst, key.af);
5773 		key.port[0] = th->th_sport;
5774 		key.port[1] = th->th_dport;
5775 	} else {			/* stack side, reverse */
5776 		PF_ACPY(&key.addr[1], pd->src, key.af);
5777 		PF_ACPY(&key.addr[0], pd->dst, key.af);
5778 		key.port[1] = th->th_sport;
5779 		key.port[0] = th->th_dport;
5780 	}
5781 
5782 	STATE_LOOKUP(kif, &key, *state, pd);
5783 
5784 	if (pd->dir == (*state)->direction) {
5785 		src = &(*state)->src;
5786 		dst = &(*state)->dst;
5787 	} else {
5788 		src = &(*state)->dst;
5789 		dst = &(*state)->src;
5790 	}
5791 
5792 	if ((action = pf_synproxy(pd, state, reason)) != PF_PASS)
5793 		return (action);
5794 
5795 	if (dst->state >= TCPS_FIN_WAIT_2 &&
5796 	    src->state >= TCPS_FIN_WAIT_2 &&
5797 	    (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) ||
5798 	    ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK &&
5799 	    pf_syncookie_check(pd) && pd->dir == PF_IN))) {
5800 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
5801 			printf("pf: state reuse ");
5802 			pf_print_state(*state);
5803 			pf_print_flags(th->th_flags);
5804 			printf("\n");
5805 		}
5806 		/* XXX make sure it's the same direction ?? */
5807 		pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED);
5808 		pf_unlink_state(*state);
5809 		*state = NULL;
5810 		return (PF_DROP);
5811 	}
5812 
5813 	if ((*state)->state_flags & PFSTATE_SLOPPY) {
5814 		if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP)
5815 			return (PF_DROP);
5816 	} else {
5817 		if (pf_tcp_track_full(state, kif, m, off, pd, reason,
5818 		    &copyback) == PF_DROP)
5819 			return (PF_DROP);
5820 	}
5821 
5822 	/* translate source/destination address, if necessary */
5823 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5824 		struct pf_state_key *nk = (*state)->key[pd->didx];
5825 
5826 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
5827 		    nk->port[pd->sidx] != th->th_sport)
5828 			pf_change_ap(m, pd->src, &th->th_sport,
5829 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
5830 			    nk->port[pd->sidx], 0, pd->af);
5831 
5832 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
5833 		    nk->port[pd->didx] != th->th_dport)
5834 			pf_change_ap(m, pd->dst, &th->th_dport,
5835 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
5836 			    nk->port[pd->didx], 0, pd->af);
5837 		copyback = 1;
5838 	}
5839 
5840 	/* Copyback sequence modulation or stateful scrub changes if needed */
5841 	if (copyback)
5842 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
5843 
5844 	return (PF_PASS);
5845 }
5846 
5847 static int
5848 pf_test_state_udp(struct pf_kstate **state, struct pfi_kkif *kif,
5849     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
5850 {
5851 	struct pf_state_peer	*src, *dst;
5852 	struct pf_state_key_cmp	 key;
5853 	struct udphdr		*uh = &pd->hdr.udp;
5854 	uint8_t			 psrc, pdst;
5855 
5856 	bzero(&key, sizeof(key));
5857 	key.af = pd->af;
5858 	key.proto = IPPROTO_UDP;
5859 	if (pd->dir == PF_IN)	{	/* wire side, straight */
5860 		PF_ACPY(&key.addr[0], pd->src, key.af);
5861 		PF_ACPY(&key.addr[1], pd->dst, key.af);
5862 		key.port[0] = uh->uh_sport;
5863 		key.port[1] = uh->uh_dport;
5864 	} else {			/* stack side, reverse */
5865 		PF_ACPY(&key.addr[1], pd->src, key.af);
5866 		PF_ACPY(&key.addr[0], pd->dst, key.af);
5867 		key.port[1] = uh->uh_sport;
5868 		key.port[0] = uh->uh_dport;
5869 	}
5870 
5871 	STATE_LOOKUP(kif, &key, *state, pd);
5872 
5873 	if (pd->dir == (*state)->direction) {
5874 		src = &(*state)->src;
5875 		dst = &(*state)->dst;
5876 		psrc = PF_PEER_SRC;
5877 		pdst = PF_PEER_DST;
5878 	} else {
5879 		src = &(*state)->dst;
5880 		dst = &(*state)->src;
5881 		psrc = PF_PEER_DST;
5882 		pdst = PF_PEER_SRC;
5883 	}
5884 
5885 	/* update states */
5886 	if (src->state < PFUDPS_SINGLE)
5887 		pf_set_protostate(*state, psrc, PFUDPS_SINGLE);
5888 	if (dst->state == PFUDPS_SINGLE)
5889 		pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE);
5890 
5891 	/* update expire time */
5892 	(*state)->expire = pf_get_uptime();
5893 	if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
5894 		(*state)->timeout = PFTM_UDP_MULTIPLE;
5895 	else
5896 		(*state)->timeout = PFTM_UDP_SINGLE;
5897 
5898 	/* translate source/destination address, if necessary */
5899 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5900 		struct pf_state_key *nk = (*state)->key[pd->didx];
5901 
5902 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
5903 		    nk->port[pd->sidx] != uh->uh_sport)
5904 			pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
5905 			    &uh->uh_sum, &nk->addr[pd->sidx],
5906 			    nk->port[pd->sidx], 1, pd->af);
5907 
5908 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
5909 		    nk->port[pd->didx] != uh->uh_dport)
5910 			pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
5911 			    &uh->uh_sum, &nk->addr[pd->didx],
5912 			    nk->port[pd->didx], 1, pd->af);
5913 		m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
5914 	}
5915 
5916 	return (PF_PASS);
5917 }
5918 
5919 static int
5920 pf_test_state_sctp(struct pf_kstate **state, struct pfi_kkif *kif,
5921     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
5922 {
5923 	struct pf_state_key_cmp	 key;
5924 	struct pf_state_peer	*src, *dst;
5925 	struct sctphdr		*sh = &pd->hdr.sctp;
5926 	u_int8_t		 psrc; //, pdst;
5927 
5928 	bzero(&key, sizeof(key));
5929 	key.af = pd->af;
5930 	key.proto = IPPROTO_SCTP;
5931 	if (pd->dir == PF_IN)	{	/* wire side, straight */
5932 		PF_ACPY(&key.addr[0], pd->src, key.af);
5933 		PF_ACPY(&key.addr[1], pd->dst, key.af);
5934 		key.port[0] = sh->src_port;
5935 		key.port[1] = sh->dest_port;
5936 	} else {			/* stack side, reverse */
5937 		PF_ACPY(&key.addr[1], pd->src, key.af);
5938 		PF_ACPY(&key.addr[0], pd->dst, key.af);
5939 		key.port[1] = sh->src_port;
5940 		key.port[0] = sh->dest_port;
5941 	}
5942 
5943 	STATE_LOOKUP(kif, &key, *state, pd);
5944 
5945 	if (pd->dir == (*state)->direction) {
5946 		src = &(*state)->src;
5947 		dst = &(*state)->dst;
5948 		psrc = PF_PEER_SRC;
5949 	} else {
5950 		src = &(*state)->dst;
5951 		dst = &(*state)->src;
5952 		psrc = PF_PEER_DST;
5953 	}
5954 
5955 	/* Track state. */
5956 	if (pd->sctp_flags & PFDESC_SCTP_INIT) {
5957 		if (src->state < SCTP_COOKIE_WAIT) {
5958 			pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT);
5959 			(*state)->timeout = PFTM_SCTP_OPENING;
5960 		}
5961 	}
5962 	if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) {
5963 		MPASS(dst->scrub != NULL);
5964 		if (dst->scrub->pfss_v_tag == 0)
5965 			dst->scrub->pfss_v_tag = pd->sctp_initiate_tag;
5966 	}
5967 
5968 	if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) {
5969 		if (src->state < SCTP_ESTABLISHED) {
5970 			pf_set_protostate(*state, psrc, SCTP_ESTABLISHED);
5971 			(*state)->timeout = PFTM_SCTP_ESTABLISHED;
5972 		}
5973 	}
5974 	if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT |
5975 	    PFDESC_SCTP_SHUTDOWN_COMPLETE)) {
5976 		if (src->state < SCTP_SHUTDOWN_PENDING) {
5977 			pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING);
5978 			(*state)->timeout = PFTM_SCTP_CLOSING;
5979 		}
5980 	}
5981 	if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) {
5982 		pf_set_protostate(*state, psrc, SCTP_CLOSED);
5983 		(*state)->timeout = PFTM_SCTP_CLOSED;
5984 	}
5985 
5986 	if (src->scrub != NULL) {
5987 		if (src->scrub->pfss_v_tag == 0) {
5988 			src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag;
5989 		} else  if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag)
5990 			return (PF_DROP);
5991 	}
5992 
5993 	(*state)->expire = pf_get_uptime();
5994 
5995 	/* translate source/destination address, if necessary */
5996 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5997 		uint16_t checksum = 0;
5998 		struct pf_state_key *nk = (*state)->key[pd->didx];
5999 
6000 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
6001 		    nk->port[pd->sidx] != pd->hdr.sctp.src_port) {
6002 			pf_change_ap(m, pd->src, &pd->hdr.sctp.src_port,
6003 			    pd->ip_sum, &checksum, &nk->addr[pd->sidx],
6004 			    nk->port[pd->sidx], 1, pd->af);
6005 		}
6006 
6007 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
6008 		    nk->port[pd->didx] != pd->hdr.sctp.dest_port) {
6009 			pf_change_ap(m, pd->dst, &pd->hdr.sctp.dest_port,
6010 			    pd->ip_sum, &checksum, &nk->addr[pd->didx],
6011 			    nk->port[pd->didx], 1, pd->af);
6012 		}
6013 	}
6014 
6015 	return (PF_PASS);
6016 }
6017 
6018 static void
6019 pf_sctp_multihome_detach_addr(const struct pf_kstate *s)
6020 {
6021 	struct pf_sctp_endpoint key;
6022 	struct pf_sctp_endpoint *ep;
6023 	struct pf_state_key *sks = s->key[PF_SK_STACK];
6024 	struct pf_sctp_source *i, *tmp;
6025 
6026 	if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL)
6027 		return;
6028 
6029 	PF_SCTP_ENDPOINTS_LOCK();
6030 
6031 	key.v_tag = s->dst.scrub->pfss_v_tag;
6032 	ep  = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6033 	if (ep != NULL) {
6034 		/* XXX Actually remove! */
6035 		TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) {
6036 			if (pf_addr_cmp(&i->addr,
6037 			    &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT],
6038 			    s->key[PF_SK_WIRE]->af) == 0) {
6039 				TAILQ_REMOVE(&ep->sources, i, entry);
6040 				free(i, M_PFTEMP);
6041 				break;
6042 			}
6043 		}
6044 
6045 		if (TAILQ_EMPTY(&ep->sources)) {
6046 			RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
6047 			free(ep, M_PFTEMP);
6048 		}
6049 	}
6050 
6051 	/* Other direction. */
6052 	key.v_tag = s->src.scrub->pfss_v_tag;
6053 	ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6054 	if (ep != NULL) {
6055 		TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) {
6056 			if (pf_addr_cmp(&i->addr,
6057 			    &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN],
6058 			    s->key[PF_SK_WIRE]->af) == 0) {
6059 				TAILQ_REMOVE(&ep->sources, i, entry);
6060 				free(i, M_PFTEMP);
6061 				break;
6062 			}
6063 		}
6064 
6065 		if (TAILQ_EMPTY(&ep->sources)) {
6066 			RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
6067 			free(ep, M_PFTEMP);
6068 		}
6069 	}
6070 
6071 	PF_SCTP_ENDPOINTS_UNLOCK();
6072 }
6073 
6074 static void
6075 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag)
6076 {
6077 	struct pf_sctp_endpoint key = {
6078 		.v_tag = v_tag,
6079 	};
6080 	struct pf_sctp_source *i;
6081 	struct pf_sctp_endpoint *ep;
6082 
6083 	PF_SCTP_ENDPOINTS_LOCK();
6084 
6085 	ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6086 	if (ep == NULL) {
6087 		ep = malloc(sizeof(struct pf_sctp_endpoint),
6088 		    M_PFTEMP, M_NOWAIT);
6089 		if (ep == NULL) {
6090 			PF_SCTP_ENDPOINTS_UNLOCK();
6091 			return;
6092 		}
6093 
6094 		ep->v_tag = v_tag;
6095 		TAILQ_INIT(&ep->sources);
6096 		RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
6097 	}
6098 
6099 	/* Avoid inserting duplicates. */
6100 	TAILQ_FOREACH(i, &ep->sources, entry) {
6101 		if (pf_addr_cmp(&i->addr, a, pd->af) == 0) {
6102 			PF_SCTP_ENDPOINTS_UNLOCK();
6103 			return;
6104 		}
6105 	}
6106 
6107 	i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT);
6108 	if (i == NULL) {
6109 		PF_SCTP_ENDPOINTS_UNLOCK();
6110 		return;
6111 	}
6112 
6113 	i->af = pd->af;
6114 	memcpy(&i->addr, a, sizeof(*a));
6115 	TAILQ_INSERT_TAIL(&ep->sources, i, entry);
6116 
6117 	PF_SCTP_ENDPOINTS_UNLOCK();
6118 }
6119 
6120 static void
6121 pf_sctp_multihome_delayed(struct pf_pdesc *pd, int off, struct pfi_kkif *kif,
6122     struct pf_kstate *s, int action)
6123 {
6124 	struct pf_sctp_multihome_job	*j, *tmp;
6125 	struct pf_sctp_source		*i;
6126 	int			 ret __unused;;
6127 	struct pf_kstate	*sm = NULL;
6128 	struct pf_krule		*ra = NULL;
6129 	struct pf_krule		*r = &V_pf_default_rule;
6130 	struct pf_kruleset	*rs = NULL;
6131 	bool do_extra = true;
6132 
6133 	PF_RULES_RLOCK_TRACKER;
6134 
6135 again:
6136 	TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) {
6137 		if (s == NULL || action != PF_PASS)
6138 			goto free;
6139 
6140 		/* Confirm we don't recurse here. */
6141 		MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP));
6142 
6143 		switch (j->op) {
6144 		case  SCTP_ADD_IP_ADDRESS: {
6145 			uint32_t v_tag = pd->sctp_initiate_tag;
6146 
6147 			if (v_tag == 0) {
6148 				if (s->direction == pd->dir)
6149 					v_tag = s->src.scrub->pfss_v_tag;
6150 				else
6151 					v_tag = s->dst.scrub->pfss_v_tag;
6152 			}
6153 
6154 			/*
6155 			 * Avoid duplicating states. We'll already have
6156 			 * created a state based on the source address of
6157 			 * the packet, but SCTP endpoints may also list this
6158 			 * address again in the INIT(_ACK) parameters.
6159 			 */
6160 			if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) {
6161 				break;
6162 			}
6163 
6164 			j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP;
6165 			PF_RULES_RLOCK();
6166 			sm = NULL;
6167 			/*
6168 			 * New connections need to be floating, because
6169 			 * we cannot know what interfaces it will use.
6170 			 * That's why we pass V_pfi_all rather than kif.
6171 			 */
6172 			ret = pf_test_rule(&r, &sm, V_pfi_all,
6173 			    j->m, off, &j->pd, &ra, &rs, NULL);
6174 			PF_RULES_RUNLOCK();
6175 			SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->m, ret);
6176 			if (ret != PF_DROP && sm != NULL) {
6177 				/* Inherit v_tag values. */
6178 				if (sm->direction == s->direction) {
6179 					sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag;
6180 					sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag;
6181 				} else {
6182 					sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag;
6183 					sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag;
6184 				}
6185 				PF_STATE_UNLOCK(sm);
6186 			} else {
6187 				/* If we try duplicate inserts? */
6188 				break;
6189 			}
6190 
6191 			/* Only add the addres if we've actually allowed the state. */
6192 			pf_sctp_multihome_add_addr(pd, &j->src, v_tag);
6193 
6194 			if (! do_extra) {
6195 				break;
6196 			}
6197 			/*
6198 			 * We need to do this for each of our source addresses.
6199 			 * Find those based on the verification tag.
6200 			 */
6201 			struct pf_sctp_endpoint key = {
6202 				.v_tag = pd->hdr.sctp.v_tag,
6203 			};
6204 			struct pf_sctp_endpoint *ep;
6205 
6206 			PF_SCTP_ENDPOINTS_LOCK();
6207 			ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6208 			if (ep == NULL) {
6209 				PF_SCTP_ENDPOINTS_UNLOCK();
6210 				break;
6211 			}
6212 			MPASS(ep != NULL);
6213 
6214 			TAILQ_FOREACH(i, &ep->sources, entry) {
6215 				struct pf_sctp_multihome_job *nj;
6216 
6217 				/* SCTP can intermingle IPv4 and IPv6. */
6218 				if (i->af != pd->af)
6219 					continue;
6220 
6221 				nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO);
6222 				if (! nj) {
6223 					continue;
6224 				}
6225 				memcpy(&nj->pd, &j->pd, sizeof(j->pd));
6226 				memcpy(&nj->src, &j->src, sizeof(nj->src));
6227 				nj->pd.src = &nj->src;
6228 				// New destination address!
6229 				memcpy(&nj->dst, &i->addr, sizeof(nj->dst));
6230 				nj->pd.dst = &nj->dst;
6231 				nj->m = j->m;
6232 				nj->op = j->op;
6233 
6234 				TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next);
6235 			}
6236 			PF_SCTP_ENDPOINTS_UNLOCK();
6237 
6238 			break;
6239 		}
6240 		case SCTP_DEL_IP_ADDRESS: {
6241 			struct pf_state_key_cmp key;
6242 			uint8_t psrc;
6243 
6244 			bzero(&key, sizeof(key));
6245 			key.af = j->pd.af;
6246 			key.proto = IPPROTO_SCTP;
6247 			if (j->pd.dir == PF_IN)	{	/* wire side, straight */
6248 				PF_ACPY(&key.addr[0], j->pd.src, key.af);
6249 				PF_ACPY(&key.addr[1], j->pd.dst, key.af);
6250 				key.port[0] = j->pd.hdr.sctp.src_port;
6251 				key.port[1] = j->pd.hdr.sctp.dest_port;
6252 			} else {			/* stack side, reverse */
6253 				PF_ACPY(&key.addr[1], j->pd.src, key.af);
6254 				PF_ACPY(&key.addr[0], j->pd.dst, key.af);
6255 				key.port[1] = j->pd.hdr.sctp.src_port;
6256 				key.port[0] = j->pd.hdr.sctp.dest_port;
6257 			}
6258 
6259 			sm = pf_find_state(kif, &key, j->pd.dir);
6260 			if (sm != NULL) {
6261 				PF_STATE_LOCK_ASSERT(sm);
6262 				if (j->pd.dir == sm->direction) {
6263 					psrc = PF_PEER_SRC;
6264 				} else {
6265 					psrc = PF_PEER_DST;
6266 				}
6267 				pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING);
6268 				sm->timeout = PFTM_SCTP_CLOSING;
6269 				PF_STATE_UNLOCK(sm);
6270 			}
6271 			break;
6272 		default:
6273 			panic("Unknown op %#x", j->op);
6274 		}
6275 	}
6276 
6277 	free:
6278 		TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next);
6279 		free(j, M_PFTEMP);
6280 	}
6281 
6282 	/* We may have inserted extra work while processing the list. */
6283 	if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) {
6284 		do_extra = false;
6285 		goto again;
6286 	}
6287 }
6288 
6289 static int
6290 pf_multihome_scan(struct mbuf *m, int start, int len, struct pf_pdesc *pd,
6291     struct pfi_kkif *kif, int op)
6292 {
6293 	int			 off = 0;
6294 	struct pf_sctp_multihome_job	*job;
6295 
6296 	while (off < len) {
6297 		struct sctp_paramhdr h;
6298 
6299 		if (!pf_pull_hdr(m, start + off, &h, sizeof(h), NULL, NULL,
6300 		    pd->af))
6301 			return (PF_DROP);
6302 
6303 		/* Parameters are at least 4 bytes. */
6304 		if (ntohs(h.param_length) < 4)
6305 			return (PF_DROP);
6306 
6307 		switch (ntohs(h.param_type)) {
6308 		case  SCTP_IPV4_ADDRESS: {
6309 			struct in_addr t;
6310 
6311 			if (ntohs(h.param_length) !=
6312 			    (sizeof(struct sctp_paramhdr) + sizeof(t)))
6313 				return (PF_DROP);
6314 
6315 			if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t),
6316 			    NULL, NULL, pd->af))
6317 				return (PF_DROP);
6318 
6319 			if (in_nullhost(t))
6320 				t.s_addr = pd->src->v4.s_addr;
6321 
6322 			/*
6323 			 * We hold the state lock (idhash) here, which means
6324 			 * that we can't acquire the keyhash, or we'll get a
6325 			 * LOR (and potentially double-lock things too). We also
6326 			 * can't release the state lock here, so instead we'll
6327 			 * enqueue this for async handling.
6328 			 * There's a relatively small race here, in that a
6329 			 * packet using the new addresses could arrive already,
6330 			 * but that's just though luck for it.
6331 			 */
6332 			job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO);
6333 			if (! job)
6334 				return (PF_DROP);
6335 
6336 			memcpy(&job->pd, pd, sizeof(*pd));
6337 
6338 			// New source address!
6339 			memcpy(&job->src, &t, sizeof(t));
6340 			job->pd.src = &job->src;
6341 			memcpy(&job->dst, pd->dst, sizeof(job->dst));
6342 			job->pd.dst = &job->dst;
6343 			job->m = m;
6344 			job->op = op;
6345 
6346 			TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next);
6347 			break;
6348 		}
6349 #ifdef INET6
6350 		case SCTP_IPV6_ADDRESS: {
6351 			struct in6_addr t;
6352 
6353 			if (ntohs(h.param_length) !=
6354 			    (sizeof(struct sctp_paramhdr) + sizeof(t)))
6355 				return (PF_DROP);
6356 
6357 			if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t),
6358 			    NULL, NULL, pd->af))
6359 				return (PF_DROP);
6360 			if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0)
6361 				break;
6362 			if (memcmp(&t, &in6addr_any, sizeof(t)) == 0)
6363 				memcpy(&t, &pd->src->v6, sizeof(t));
6364 
6365 			job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO);
6366 			if (! job)
6367 				return (PF_DROP);
6368 
6369 			memcpy(&job->pd, pd, sizeof(*pd));
6370 			memcpy(&job->src, &t, sizeof(t));
6371 			job->pd.src = &job->src;
6372 			memcpy(&job->dst, pd->dst, sizeof(job->dst));
6373 			job->pd.dst = &job->dst;
6374 			job->m = m;
6375 			job->op = op;
6376 
6377 			TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next);
6378 			break;
6379 		}
6380 #endif
6381 		case SCTP_ADD_IP_ADDRESS: {
6382 			int ret;
6383 			struct sctp_asconf_paramhdr ah;
6384 
6385 			if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah),
6386 			    NULL, NULL, pd->af))
6387 				return (PF_DROP);
6388 
6389 			ret = pf_multihome_scan(m, start + off + sizeof(ah),
6390 			    ntohs(ah.ph.param_length) - sizeof(ah), pd, kif,
6391 			    SCTP_ADD_IP_ADDRESS);
6392 			if (ret != PF_PASS)
6393 				return (ret);
6394 			break;
6395 		}
6396 		case SCTP_DEL_IP_ADDRESS: {
6397 			int ret;
6398 			struct sctp_asconf_paramhdr ah;
6399 
6400 			if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah),
6401 			    NULL, NULL, pd->af))
6402 				return (PF_DROP);
6403 			ret = pf_multihome_scan(m, start + off + sizeof(ah),
6404 			    ntohs(ah.ph.param_length) - sizeof(ah), pd, kif,
6405 			    SCTP_DEL_IP_ADDRESS);
6406 			if (ret != PF_PASS)
6407 				return (ret);
6408 			break;
6409 		}
6410 		default:
6411 			break;
6412 		}
6413 
6414 		off += roundup(ntohs(h.param_length), 4);
6415 	}
6416 
6417 	return (PF_PASS);
6418 }
6419 int
6420 pf_multihome_scan_init(struct mbuf *m, int start, int len, struct pf_pdesc *pd,
6421     struct pfi_kkif *kif)
6422 {
6423 	start += sizeof(struct sctp_init_chunk);
6424 	len -= sizeof(struct sctp_init_chunk);
6425 
6426 	return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS));
6427 }
6428 
6429 int
6430 pf_multihome_scan_asconf(struct mbuf *m, int start, int len,
6431     struct pf_pdesc *pd, struct pfi_kkif *kif)
6432 {
6433 	start += sizeof(struct sctp_asconf_chunk);
6434 	len -= sizeof(struct sctp_asconf_chunk);
6435 
6436 	return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS));
6437 }
6438 
6439 static int
6440 pf_test_state_icmp(struct pf_kstate **state, struct pfi_kkif *kif,
6441     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
6442 {
6443 	struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
6444 	u_int16_t	 icmpid = 0, *icmpsum;
6445 	u_int8_t	 icmptype, icmpcode;
6446 	int		 state_icmp = 0;
6447 	struct pf_state_key_cmp key;
6448 
6449 	bzero(&key, sizeof(key));
6450 	switch (pd->proto) {
6451 #ifdef INET
6452 	case IPPROTO_ICMP:
6453 		icmptype = pd->hdr.icmp.icmp_type;
6454 		icmpcode = pd->hdr.icmp.icmp_code;
6455 		icmpid = pd->hdr.icmp.icmp_id;
6456 		icmpsum = &pd->hdr.icmp.icmp_cksum;
6457 
6458 		if (icmptype == ICMP_UNREACH ||
6459 		    icmptype == ICMP_SOURCEQUENCH ||
6460 		    icmptype == ICMP_REDIRECT ||
6461 		    icmptype == ICMP_TIMXCEED ||
6462 		    icmptype == ICMP_PARAMPROB)
6463 			state_icmp++;
6464 		break;
6465 #endif /* INET */
6466 #ifdef INET6
6467 	case IPPROTO_ICMPV6:
6468 		icmptype = pd->hdr.icmp6.icmp6_type;
6469 		icmpcode = pd->hdr.icmp6.icmp6_code;
6470 		icmpid = pd->hdr.icmp6.icmp6_id;
6471 		icmpsum = &pd->hdr.icmp6.icmp6_cksum;
6472 
6473 		if (icmptype == ICMP6_DST_UNREACH ||
6474 		    icmptype == ICMP6_PACKET_TOO_BIG ||
6475 		    icmptype == ICMP6_TIME_EXCEEDED ||
6476 		    icmptype == ICMP6_PARAM_PROB)
6477 			state_icmp++;
6478 		break;
6479 #endif /* INET6 */
6480 	}
6481 
6482 	if (!state_icmp) {
6483 		/*
6484 		 * ICMP query/reply message not related to a TCP/UDP packet.
6485 		 * Search for an ICMP state.
6486 		 */
6487 		key.af = pd->af;
6488 		key.proto = pd->proto;
6489 		key.port[0] = key.port[1] = icmpid;
6490 		if (pd->dir == PF_IN)	{	/* wire side, straight */
6491 			PF_ACPY(&key.addr[0], pd->src, key.af);
6492 			PF_ACPY(&key.addr[1], pd->dst, key.af);
6493 		} else {			/* stack side, reverse */
6494 			PF_ACPY(&key.addr[1], pd->src, key.af);
6495 			PF_ACPY(&key.addr[0], pd->dst, key.af);
6496 		}
6497 
6498 		STATE_LOOKUP(kif, &key, *state, pd);
6499 
6500 		(*state)->expire = pf_get_uptime();
6501 		(*state)->timeout = PFTM_ICMP_ERROR_REPLY;
6502 
6503 		/* translate source/destination address, if necessary */
6504 		if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
6505 			struct pf_state_key *nk = (*state)->key[pd->didx];
6506 
6507 			switch (pd->af) {
6508 #ifdef INET
6509 			case AF_INET:
6510 				if (PF_ANEQ(pd->src,
6511 				    &nk->addr[pd->sidx], AF_INET))
6512 					pf_change_a(&saddr->v4.s_addr,
6513 					    pd->ip_sum,
6514 					    nk->addr[pd->sidx].v4.s_addr, 0);
6515 
6516 				if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
6517 				    AF_INET))
6518 					pf_change_a(&daddr->v4.s_addr,
6519 					    pd->ip_sum,
6520 					    nk->addr[pd->didx].v4.s_addr, 0);
6521 
6522 				if (nk->port[0] !=
6523 				    pd->hdr.icmp.icmp_id) {
6524 					pd->hdr.icmp.icmp_cksum =
6525 					    pf_cksum_fixup(
6526 					    pd->hdr.icmp.icmp_cksum, icmpid,
6527 					    nk->port[pd->sidx], 0);
6528 					pd->hdr.icmp.icmp_id =
6529 					    nk->port[pd->sidx];
6530 				}
6531 
6532 				m_copyback(m, off, ICMP_MINLEN,
6533 				    (caddr_t )&pd->hdr.icmp);
6534 				break;
6535 #endif /* INET */
6536 #ifdef INET6
6537 			case AF_INET6:
6538 				if (PF_ANEQ(pd->src,
6539 				    &nk->addr[pd->sidx], AF_INET6))
6540 					pf_change_a6(saddr,
6541 					    &pd->hdr.icmp6.icmp6_cksum,
6542 					    &nk->addr[pd->sidx], 0);
6543 
6544 				if (PF_ANEQ(pd->dst,
6545 				    &nk->addr[pd->didx], AF_INET6))
6546 					pf_change_a6(daddr,
6547 					    &pd->hdr.icmp6.icmp6_cksum,
6548 					    &nk->addr[pd->didx], 0);
6549 
6550 				m_copyback(m, off, sizeof(struct icmp6_hdr),
6551 				    (caddr_t )&pd->hdr.icmp6);
6552 				break;
6553 #endif /* INET6 */
6554 			}
6555 		}
6556 		return (PF_PASS);
6557 
6558 	} else {
6559 		/*
6560 		 * ICMP error message in response to a TCP/UDP packet.
6561 		 * Extract the inner TCP/UDP header and search for that state.
6562 		 */
6563 
6564 		struct pf_pdesc	pd2;
6565 		bzero(&pd2, sizeof pd2);
6566 #ifdef INET
6567 		struct ip	h2;
6568 #endif /* INET */
6569 #ifdef INET6
6570 		struct ip6_hdr	h2_6;
6571 		int		terminal = 0;
6572 #endif /* INET6 */
6573 		int		ipoff2 = 0;
6574 		int		off2 = 0;
6575 
6576 		pd2.af = pd->af;
6577 		/* Payload packet is from the opposite direction. */
6578 		pd2.sidx = (pd->dir == PF_IN) ? 1 : 0;
6579 		pd2.didx = (pd->dir == PF_IN) ? 0 : 1;
6580 		switch (pd->af) {
6581 #ifdef INET
6582 		case AF_INET:
6583 			/* offset of h2 in mbuf chain */
6584 			ipoff2 = off + ICMP_MINLEN;
6585 
6586 			if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
6587 			    NULL, reason, pd2.af)) {
6588 				DPFPRINTF(PF_DEBUG_MISC,
6589 				    ("pf: ICMP error message too short "
6590 				    "(ip)\n"));
6591 				return (PF_DROP);
6592 			}
6593 			/*
6594 			 * ICMP error messages don't refer to non-first
6595 			 * fragments
6596 			 */
6597 			if (h2.ip_off & htons(IP_OFFMASK)) {
6598 				REASON_SET(reason, PFRES_FRAG);
6599 				return (PF_DROP);
6600 			}
6601 
6602 			/* offset of protocol header that follows h2 */
6603 			off2 = ipoff2 + (h2.ip_hl << 2);
6604 
6605 			pd2.proto = h2.ip_p;
6606 			pd2.src = (struct pf_addr *)&h2.ip_src;
6607 			pd2.dst = (struct pf_addr *)&h2.ip_dst;
6608 			pd2.ip_sum = &h2.ip_sum;
6609 			break;
6610 #endif /* INET */
6611 #ifdef INET6
6612 		case AF_INET6:
6613 			ipoff2 = off + sizeof(struct icmp6_hdr);
6614 
6615 			if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
6616 			    NULL, reason, pd2.af)) {
6617 				DPFPRINTF(PF_DEBUG_MISC,
6618 				    ("pf: ICMP error message too short "
6619 				    "(ip6)\n"));
6620 				return (PF_DROP);
6621 			}
6622 			pd2.proto = h2_6.ip6_nxt;
6623 			pd2.src = (struct pf_addr *)&h2_6.ip6_src;
6624 			pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
6625 			pd2.ip_sum = NULL;
6626 			off2 = ipoff2 + sizeof(h2_6);
6627 			do {
6628 				switch (pd2.proto) {
6629 				case IPPROTO_FRAGMENT:
6630 					/*
6631 					 * ICMPv6 error messages for
6632 					 * non-first fragments
6633 					 */
6634 					REASON_SET(reason, PFRES_FRAG);
6635 					return (PF_DROP);
6636 				case IPPROTO_AH:
6637 				case IPPROTO_HOPOPTS:
6638 				case IPPROTO_ROUTING:
6639 				case IPPROTO_DSTOPTS: {
6640 					/* get next header and header length */
6641 					struct ip6_ext opt6;
6642 
6643 					if (!pf_pull_hdr(m, off2, &opt6,
6644 					    sizeof(opt6), NULL, reason,
6645 					    pd2.af)) {
6646 						DPFPRINTF(PF_DEBUG_MISC,
6647 						    ("pf: ICMPv6 short opt\n"));
6648 						return (PF_DROP);
6649 					}
6650 					if (pd2.proto == IPPROTO_AH)
6651 						off2 += (opt6.ip6e_len + 2) * 4;
6652 					else
6653 						off2 += (opt6.ip6e_len + 1) * 8;
6654 					pd2.proto = opt6.ip6e_nxt;
6655 					/* goto the next header */
6656 					break;
6657 				}
6658 				default:
6659 					terminal++;
6660 					break;
6661 				}
6662 			} while (!terminal);
6663 			break;
6664 #endif /* INET6 */
6665 		}
6666 
6667 		if (PF_ANEQ(pd->dst, pd2.src, pd->af)) {
6668 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
6669 				printf("pf: BAD ICMP %d:%d outer dst: ",
6670 				    icmptype, icmpcode);
6671 				pf_print_host(pd->src, 0, pd->af);
6672 				printf(" -> ");
6673 				pf_print_host(pd->dst, 0, pd->af);
6674 				printf(" inner src: ");
6675 				pf_print_host(pd2.src, 0, pd2.af);
6676 				printf(" -> ");
6677 				pf_print_host(pd2.dst, 0, pd2.af);
6678 				printf("\n");
6679 			}
6680 			REASON_SET(reason, PFRES_BADSTATE);
6681 			return (PF_DROP);
6682 		}
6683 
6684 		switch (pd2.proto) {
6685 		case IPPROTO_TCP: {
6686 			struct tcphdr		 th;
6687 			u_int32_t		 seq;
6688 			struct pf_state_peer	*src, *dst;
6689 			u_int8_t		 dws;
6690 			int			 copyback = 0;
6691 
6692 			/*
6693 			 * Only the first 8 bytes of the TCP header can be
6694 			 * expected. Don't access any TCP header fields after
6695 			 * th_seq, an ackskew test is not possible.
6696 			 */
6697 			if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
6698 			    pd2.af)) {
6699 				DPFPRINTF(PF_DEBUG_MISC,
6700 				    ("pf: ICMP error message too short "
6701 				    "(tcp)\n"));
6702 				return (PF_DROP);
6703 			}
6704 
6705 			key.af = pd2.af;
6706 			key.proto = IPPROTO_TCP;
6707 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
6708 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
6709 			key.port[pd2.sidx] = th.th_sport;
6710 			key.port[pd2.didx] = th.th_dport;
6711 
6712 			STATE_LOOKUP(kif, &key, *state, pd);
6713 
6714 			if (pd->dir == (*state)->direction) {
6715 				src = &(*state)->dst;
6716 				dst = &(*state)->src;
6717 			} else {
6718 				src = &(*state)->src;
6719 				dst = &(*state)->dst;
6720 			}
6721 
6722 			if (src->wscale && dst->wscale)
6723 				dws = dst->wscale & PF_WSCALE_MASK;
6724 			else
6725 				dws = 0;
6726 
6727 			/* Demodulate sequence number */
6728 			seq = ntohl(th.th_seq) - src->seqdiff;
6729 			if (src->seqdiff) {
6730 				pf_change_a(&th.th_seq, icmpsum,
6731 				    htonl(seq), 0);
6732 				copyback = 1;
6733 			}
6734 
6735 			if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
6736 			    (!SEQ_GEQ(src->seqhi, seq) ||
6737 			    !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
6738 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
6739 					printf("pf: BAD ICMP %d:%d ",
6740 					    icmptype, icmpcode);
6741 					pf_print_host(pd->src, 0, pd->af);
6742 					printf(" -> ");
6743 					pf_print_host(pd->dst, 0, pd->af);
6744 					printf(" state: ");
6745 					pf_print_state(*state);
6746 					printf(" seq=%u\n", seq);
6747 				}
6748 				REASON_SET(reason, PFRES_BADSTATE);
6749 				return (PF_DROP);
6750 			} else {
6751 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
6752 					printf("pf: OK ICMP %d:%d ",
6753 					    icmptype, icmpcode);
6754 					pf_print_host(pd->src, 0, pd->af);
6755 					printf(" -> ");
6756 					pf_print_host(pd->dst, 0, pd->af);
6757 					printf(" state: ");
6758 					pf_print_state(*state);
6759 					printf(" seq=%u\n", seq);
6760 				}
6761 			}
6762 
6763 			/* translate source/destination address, if necessary */
6764 			if ((*state)->key[PF_SK_WIRE] !=
6765 			    (*state)->key[PF_SK_STACK]) {
6766 				struct pf_state_key *nk =
6767 				    (*state)->key[pd->didx];
6768 
6769 				if (PF_ANEQ(pd2.src,
6770 				    &nk->addr[pd2.sidx], pd2.af) ||
6771 				    nk->port[pd2.sidx] != th.th_sport)
6772 					pf_change_icmp(pd2.src, &th.th_sport,
6773 					    daddr, &nk->addr[pd2.sidx],
6774 					    nk->port[pd2.sidx], NULL,
6775 					    pd2.ip_sum, icmpsum,
6776 					    pd->ip_sum, 0, pd2.af);
6777 
6778 				if (PF_ANEQ(pd2.dst,
6779 				    &nk->addr[pd2.didx], pd2.af) ||
6780 				    nk->port[pd2.didx] != th.th_dport)
6781 					pf_change_icmp(pd2.dst, &th.th_dport,
6782 					    saddr, &nk->addr[pd2.didx],
6783 					    nk->port[pd2.didx], NULL,
6784 					    pd2.ip_sum, icmpsum,
6785 					    pd->ip_sum, 0, pd2.af);
6786 				copyback = 1;
6787 			}
6788 
6789 			if (copyback) {
6790 				switch (pd2.af) {
6791 #ifdef INET
6792 				case AF_INET:
6793 					m_copyback(m, off, ICMP_MINLEN,
6794 					    (caddr_t )&pd->hdr.icmp);
6795 					m_copyback(m, ipoff2, sizeof(h2),
6796 					    (caddr_t )&h2);
6797 					break;
6798 #endif /* INET */
6799 #ifdef INET6
6800 				case AF_INET6:
6801 					m_copyback(m, off,
6802 					    sizeof(struct icmp6_hdr),
6803 					    (caddr_t )&pd->hdr.icmp6);
6804 					m_copyback(m, ipoff2, sizeof(h2_6),
6805 					    (caddr_t )&h2_6);
6806 					break;
6807 #endif /* INET6 */
6808 				}
6809 				m_copyback(m, off2, 8, (caddr_t)&th);
6810 			}
6811 
6812 			return (PF_PASS);
6813 			break;
6814 		}
6815 		case IPPROTO_UDP: {
6816 			struct udphdr		uh;
6817 
6818 			if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
6819 			    NULL, reason, pd2.af)) {
6820 				DPFPRINTF(PF_DEBUG_MISC,
6821 				    ("pf: ICMP error message too short "
6822 				    "(udp)\n"));
6823 				return (PF_DROP);
6824 			}
6825 
6826 			key.af = pd2.af;
6827 			key.proto = IPPROTO_UDP;
6828 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
6829 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
6830 			key.port[pd2.sidx] = uh.uh_sport;
6831 			key.port[pd2.didx] = uh.uh_dport;
6832 
6833 			STATE_LOOKUP(kif, &key, *state, pd);
6834 
6835 			/* translate source/destination address, if necessary */
6836 			if ((*state)->key[PF_SK_WIRE] !=
6837 			    (*state)->key[PF_SK_STACK]) {
6838 				struct pf_state_key *nk =
6839 				    (*state)->key[pd->didx];
6840 
6841 				if (PF_ANEQ(pd2.src,
6842 				    &nk->addr[pd2.sidx], pd2.af) ||
6843 				    nk->port[pd2.sidx] != uh.uh_sport)
6844 					pf_change_icmp(pd2.src, &uh.uh_sport,
6845 					    daddr, &nk->addr[pd2.sidx],
6846 					    nk->port[pd2.sidx], &uh.uh_sum,
6847 					    pd2.ip_sum, icmpsum,
6848 					    pd->ip_sum, 1, pd2.af);
6849 
6850 				if (PF_ANEQ(pd2.dst,
6851 				    &nk->addr[pd2.didx], pd2.af) ||
6852 				    nk->port[pd2.didx] != uh.uh_dport)
6853 					pf_change_icmp(pd2.dst, &uh.uh_dport,
6854 					    saddr, &nk->addr[pd2.didx],
6855 					    nk->port[pd2.didx], &uh.uh_sum,
6856 					    pd2.ip_sum, icmpsum,
6857 					    pd->ip_sum, 1, pd2.af);
6858 
6859 				switch (pd2.af) {
6860 #ifdef INET
6861 				case AF_INET:
6862 					m_copyback(m, off, ICMP_MINLEN,
6863 					    (caddr_t )&pd->hdr.icmp);
6864 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
6865 					break;
6866 #endif /* INET */
6867 #ifdef INET6
6868 				case AF_INET6:
6869 					m_copyback(m, off,
6870 					    sizeof(struct icmp6_hdr),
6871 					    (caddr_t )&pd->hdr.icmp6);
6872 					m_copyback(m, ipoff2, sizeof(h2_6),
6873 					    (caddr_t )&h2_6);
6874 					break;
6875 #endif /* INET6 */
6876 				}
6877 				m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
6878 			}
6879 			return (PF_PASS);
6880 			break;
6881 		}
6882 #ifdef INET
6883 		case IPPROTO_ICMP: {
6884 			struct icmp		iih;
6885 
6886 			if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
6887 			    NULL, reason, pd2.af)) {
6888 				DPFPRINTF(PF_DEBUG_MISC,
6889 				    ("pf: ICMP error message too short i"
6890 				    "(icmp)\n"));
6891 				return (PF_DROP);
6892 			}
6893 
6894 			key.af = pd2.af;
6895 			key.proto = IPPROTO_ICMP;
6896 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
6897 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
6898 			key.port[0] = key.port[1] = iih.icmp_id;
6899 
6900 			STATE_LOOKUP(kif, &key, *state, pd);
6901 
6902 			/* translate source/destination address, if necessary */
6903 			if ((*state)->key[PF_SK_WIRE] !=
6904 			    (*state)->key[PF_SK_STACK]) {
6905 				struct pf_state_key *nk =
6906 				    (*state)->key[pd->didx];
6907 
6908 				if (PF_ANEQ(pd2.src,
6909 				    &nk->addr[pd2.sidx], pd2.af) ||
6910 				    nk->port[pd2.sidx] != iih.icmp_id)
6911 					pf_change_icmp(pd2.src, &iih.icmp_id,
6912 					    daddr, &nk->addr[pd2.sidx],
6913 					    nk->port[pd2.sidx], NULL,
6914 					    pd2.ip_sum, icmpsum,
6915 					    pd->ip_sum, 0, AF_INET);
6916 
6917 				if (PF_ANEQ(pd2.dst,
6918 				    &nk->addr[pd2.didx], pd2.af) ||
6919 				    nk->port[pd2.didx] != iih.icmp_id)
6920 					pf_change_icmp(pd2.dst, &iih.icmp_id,
6921 					    saddr, &nk->addr[pd2.didx],
6922 					    nk->port[pd2.didx], NULL,
6923 					    pd2.ip_sum, icmpsum,
6924 					    pd->ip_sum, 0, AF_INET);
6925 
6926 				m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
6927 				m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
6928 				m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
6929 			}
6930 			return (PF_PASS);
6931 			break;
6932 		}
6933 #endif /* INET */
6934 #ifdef INET6
6935 		case IPPROTO_ICMPV6: {
6936 			struct icmp6_hdr	iih;
6937 
6938 			if (!pf_pull_hdr(m, off2, &iih,
6939 			    sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
6940 				DPFPRINTF(PF_DEBUG_MISC,
6941 				    ("pf: ICMP error message too short "
6942 				    "(icmp6)\n"));
6943 				return (PF_DROP);
6944 			}
6945 
6946 			key.af = pd2.af;
6947 			key.proto = IPPROTO_ICMPV6;
6948 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
6949 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
6950 			key.port[0] = key.port[1] = iih.icmp6_id;
6951 
6952 			STATE_LOOKUP(kif, &key, *state, pd);
6953 
6954 			/* translate source/destination address, if necessary */
6955 			if ((*state)->key[PF_SK_WIRE] !=
6956 			    (*state)->key[PF_SK_STACK]) {
6957 				struct pf_state_key *nk =
6958 				    (*state)->key[pd->didx];
6959 
6960 				if (PF_ANEQ(pd2.src,
6961 				    &nk->addr[pd2.sidx], pd2.af) ||
6962 				    nk->port[pd2.sidx] != iih.icmp6_id)
6963 					pf_change_icmp(pd2.src, &iih.icmp6_id,
6964 					    daddr, &nk->addr[pd2.sidx],
6965 					    nk->port[pd2.sidx], NULL,
6966 					    pd2.ip_sum, icmpsum,
6967 					    pd->ip_sum, 0, AF_INET6);
6968 
6969 				if (PF_ANEQ(pd2.dst,
6970 				    &nk->addr[pd2.didx], pd2.af) ||
6971 				    nk->port[pd2.didx] != iih.icmp6_id)
6972 					pf_change_icmp(pd2.dst, &iih.icmp6_id,
6973 					    saddr, &nk->addr[pd2.didx],
6974 					    nk->port[pd2.didx], NULL,
6975 					    pd2.ip_sum, icmpsum,
6976 					    pd->ip_sum, 0, AF_INET6);
6977 
6978 				m_copyback(m, off, sizeof(struct icmp6_hdr),
6979 				    (caddr_t)&pd->hdr.icmp6);
6980 				m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
6981 				m_copyback(m, off2, sizeof(struct icmp6_hdr),
6982 				    (caddr_t)&iih);
6983 			}
6984 			return (PF_PASS);
6985 			break;
6986 		}
6987 #endif /* INET6 */
6988 		default: {
6989 			key.af = pd2.af;
6990 			key.proto = pd2.proto;
6991 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
6992 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
6993 			key.port[0] = key.port[1] = 0;
6994 
6995 			STATE_LOOKUP(kif, &key, *state, pd);
6996 
6997 			/* translate source/destination address, if necessary */
6998 			if ((*state)->key[PF_SK_WIRE] !=
6999 			    (*state)->key[PF_SK_STACK]) {
7000 				struct pf_state_key *nk =
7001 				    (*state)->key[pd->didx];
7002 
7003 				if (PF_ANEQ(pd2.src,
7004 				    &nk->addr[pd2.sidx], pd2.af))
7005 					pf_change_icmp(pd2.src, NULL, daddr,
7006 					    &nk->addr[pd2.sidx], 0, NULL,
7007 					    pd2.ip_sum, icmpsum,
7008 					    pd->ip_sum, 0, pd2.af);
7009 
7010 				if (PF_ANEQ(pd2.dst,
7011 				    &nk->addr[pd2.didx], pd2.af))
7012 					pf_change_icmp(pd2.dst, NULL, saddr,
7013 					    &nk->addr[pd2.didx], 0, NULL,
7014 					    pd2.ip_sum, icmpsum,
7015 					    pd->ip_sum, 0, pd2.af);
7016 
7017 				switch (pd2.af) {
7018 #ifdef INET
7019 				case AF_INET:
7020 					m_copyback(m, off, ICMP_MINLEN,
7021 					    (caddr_t)&pd->hdr.icmp);
7022 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
7023 					break;
7024 #endif /* INET */
7025 #ifdef INET6
7026 				case AF_INET6:
7027 					m_copyback(m, off,
7028 					    sizeof(struct icmp6_hdr),
7029 					    (caddr_t )&pd->hdr.icmp6);
7030 					m_copyback(m, ipoff2, sizeof(h2_6),
7031 					    (caddr_t )&h2_6);
7032 					break;
7033 #endif /* INET6 */
7034 				}
7035 			}
7036 			return (PF_PASS);
7037 			break;
7038 		}
7039 		}
7040 	}
7041 }
7042 
7043 static int
7044 pf_test_state_other(struct pf_kstate **state, struct pfi_kkif *kif,
7045     struct mbuf *m, struct pf_pdesc *pd)
7046 {
7047 	struct pf_state_peer	*src, *dst;
7048 	struct pf_state_key_cmp	 key;
7049 	uint8_t			 psrc, pdst;
7050 
7051 	bzero(&key, sizeof(key));
7052 	key.af = pd->af;
7053 	key.proto = pd->proto;
7054 	if (pd->dir == PF_IN)	{
7055 		PF_ACPY(&key.addr[0], pd->src, key.af);
7056 		PF_ACPY(&key.addr[1], pd->dst, key.af);
7057 		key.port[0] = key.port[1] = 0;
7058 	} else {
7059 		PF_ACPY(&key.addr[1], pd->src, key.af);
7060 		PF_ACPY(&key.addr[0], pd->dst, key.af);
7061 		key.port[1] = key.port[0] = 0;
7062 	}
7063 
7064 	STATE_LOOKUP(kif, &key, *state, pd);
7065 
7066 	if (pd->dir == (*state)->direction) {
7067 		src = &(*state)->src;
7068 		dst = &(*state)->dst;
7069 		psrc = PF_PEER_SRC;
7070 		pdst = PF_PEER_DST;
7071 	} else {
7072 		src = &(*state)->dst;
7073 		dst = &(*state)->src;
7074 		psrc = PF_PEER_DST;
7075 		pdst = PF_PEER_SRC;
7076 	}
7077 
7078 	/* update states */
7079 	if (src->state < PFOTHERS_SINGLE)
7080 		pf_set_protostate(*state, psrc, PFOTHERS_SINGLE);
7081 	if (dst->state == PFOTHERS_SINGLE)
7082 		pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE);
7083 
7084 	/* update expire time */
7085 	(*state)->expire = pf_get_uptime();
7086 	if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
7087 		(*state)->timeout = PFTM_OTHER_MULTIPLE;
7088 	else
7089 		(*state)->timeout = PFTM_OTHER_SINGLE;
7090 
7091 	/* translate source/destination address, if necessary */
7092 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
7093 		struct pf_state_key *nk = (*state)->key[pd->didx];
7094 
7095 		KASSERT(nk, ("%s: nk is null", __func__));
7096 		KASSERT(pd, ("%s: pd is null", __func__));
7097 		KASSERT(pd->src, ("%s: pd->src is null", __func__));
7098 		KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
7099 		switch (pd->af) {
7100 #ifdef INET
7101 		case AF_INET:
7102 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
7103 				pf_change_a(&pd->src->v4.s_addr,
7104 				    pd->ip_sum,
7105 				    nk->addr[pd->sidx].v4.s_addr,
7106 				    0);
7107 
7108 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
7109 				pf_change_a(&pd->dst->v4.s_addr,
7110 				    pd->ip_sum,
7111 				    nk->addr[pd->didx].v4.s_addr,
7112 				    0);
7113 
7114 			break;
7115 #endif /* INET */
7116 #ifdef INET6
7117 		case AF_INET6:
7118 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
7119 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
7120 
7121 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
7122 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
7123 #endif /* INET6 */
7124 		}
7125 	}
7126 	return (PF_PASS);
7127 }
7128 
7129 /*
7130  * ipoff and off are measured from the start of the mbuf chain.
7131  * h must be at "ipoff" on the mbuf chain.
7132  */
7133 void *
7134 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
7135     u_short *actionp, u_short *reasonp, sa_family_t af)
7136 {
7137 	switch (af) {
7138 #ifdef INET
7139 	case AF_INET: {
7140 		struct ip	*h = mtod(m, struct ip *);
7141 		u_int16_t	 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
7142 
7143 		if (fragoff) {
7144 			if (fragoff >= len)
7145 				ACTION_SET(actionp, PF_PASS);
7146 			else {
7147 				ACTION_SET(actionp, PF_DROP);
7148 				REASON_SET(reasonp, PFRES_FRAG);
7149 			}
7150 			return (NULL);
7151 		}
7152 		if (m->m_pkthdr.len < off + len ||
7153 		    ntohs(h->ip_len) < off + len) {
7154 			ACTION_SET(actionp, PF_DROP);
7155 			REASON_SET(reasonp, PFRES_SHORT);
7156 			return (NULL);
7157 		}
7158 		break;
7159 	}
7160 #endif /* INET */
7161 #ifdef INET6
7162 	case AF_INET6: {
7163 		struct ip6_hdr	*h = mtod(m, struct ip6_hdr *);
7164 
7165 		if (m->m_pkthdr.len < off + len ||
7166 		    (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
7167 		    (unsigned)(off + len)) {
7168 			ACTION_SET(actionp, PF_DROP);
7169 			REASON_SET(reasonp, PFRES_SHORT);
7170 			return (NULL);
7171 		}
7172 		break;
7173 	}
7174 #endif /* INET6 */
7175 	}
7176 	m_copydata(m, off, len, p);
7177 	return (p);
7178 }
7179 
7180 int
7181 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif,
7182     int rtableid)
7183 {
7184 	struct ifnet		*ifp;
7185 
7186 	/*
7187 	 * Skip check for addresses with embedded interface scope,
7188 	 * as they would always match anyway.
7189 	 */
7190 	if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
7191 		return (1);
7192 
7193 	if (af != AF_INET && af != AF_INET6)
7194 		return (0);
7195 
7196 	if (kif == V_pfi_all)
7197 		return (1);
7198 
7199 	/* Skip checks for ipsec interfaces */
7200 	if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
7201 		return (1);
7202 
7203 	ifp = (kif != NULL) ? kif->pfik_ifp : NULL;
7204 
7205 	switch (af) {
7206 #ifdef INET6
7207 	case AF_INET6:
7208 		return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE,
7209 		    ifp));
7210 #endif
7211 #ifdef INET
7212 	case AF_INET:
7213 		return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE,
7214 		    ifp));
7215 #endif
7216 	}
7217 
7218 	return (0);
7219 }
7220 
7221 #ifdef INET
7222 static void
7223 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp,
7224     struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp)
7225 {
7226 	struct mbuf		*m0, *m1, *md;
7227 	struct sockaddr_in	dst;
7228 	struct ip		*ip;
7229 	struct pfi_kkif		*nkif = NULL;
7230 	struct ifnet		*ifp = NULL;
7231 	struct pf_addr		 naddr;
7232 	struct pf_ksrc_node	*sn = NULL;
7233 	int			 error = 0;
7234 	uint16_t		 ip_len, ip_off;
7235 	int			 r_rt, r_dir;
7236 
7237 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
7238 
7239 	if (s) {
7240 		r_rt = s->rt;
7241 		r_dir = s->direction;
7242 	} else {
7243 		r_rt = r->rt;
7244 		r_dir = r->direction;
7245 	}
7246 
7247 	KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT ||
7248 	    r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction",
7249 	    __func__));
7250 
7251 	if ((pd->pf_mtag == NULL &&
7252 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
7253 	    pd->pf_mtag->routed++ > 3) {
7254 		m0 = *m;
7255 		*m = NULL;
7256 		goto bad_locked;
7257 	}
7258 
7259 	if (r_rt == PF_DUPTO) {
7260 		if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) {
7261 			if (s == NULL) {
7262 				ifp = r->rpool.cur->kif ?
7263 				    r->rpool.cur->kif->pfik_ifp : NULL;
7264 			} else {
7265 				ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7266 				/* If pfsync'd */
7267 				if (ifp == NULL && r->rpool.cur != NULL)
7268 					ifp = r->rpool.cur->kif ?
7269 					    r->rpool.cur->kif->pfik_ifp : NULL;
7270 				PF_STATE_UNLOCK(s);
7271 			}
7272 			if (ifp == oifp) {
7273 				/* When the 2nd interface is not skipped */
7274 				return;
7275 			} else {
7276 				m0 = *m;
7277 				*m = NULL;
7278 				goto bad;
7279 			}
7280 		} else {
7281 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED;
7282 			if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) {
7283 				if (s)
7284 					PF_STATE_UNLOCK(s);
7285 				return;
7286 			}
7287 		}
7288 	} else {
7289 		if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) {
7290 			pf_dummynet(pd, s, r, m);
7291 			if (s)
7292 				PF_STATE_UNLOCK(s);
7293 			return;
7294 		}
7295 		m0 = *m;
7296 	}
7297 
7298 	ip = mtod(m0, struct ip *);
7299 
7300 	bzero(&dst, sizeof(dst));
7301 	dst.sin_family = AF_INET;
7302 	dst.sin_len = sizeof(dst);
7303 	dst.sin_addr = ip->ip_dst;
7304 
7305 	bzero(&naddr, sizeof(naddr));
7306 
7307 	if (s == NULL) {
7308 		if (TAILQ_EMPTY(&r->rpool.list)) {
7309 			DPFPRINTF(PF_DEBUG_URGENT,
7310 			    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
7311 			goto bad_locked;
7312 		}
7313 		pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
7314 		    &naddr, &nkif, NULL, &sn);
7315 		if (!PF_AZERO(&naddr, AF_INET))
7316 			dst.sin_addr.s_addr = naddr.v4.s_addr;
7317 		ifp = nkif ? nkif->pfik_ifp : NULL;
7318 	} else {
7319 		if (!PF_AZERO(&s->rt_addr, AF_INET))
7320 			dst.sin_addr.s_addr =
7321 			    s->rt_addr.v4.s_addr;
7322 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7323 		/* If pfsync'd */
7324 		if (ifp == NULL && r->rpool.cur != NULL) {
7325 			ifp = r->rpool.cur->kif ?
7326 			    r->rpool.cur->kif->pfik_ifp : NULL;
7327 		}
7328 		PF_STATE_UNLOCK(s);
7329 	}
7330 
7331 	if (ifp == NULL)
7332 		goto bad;
7333 
7334 	if (pd->dir == PF_IN) {
7335 		if (pf_test(PF_OUT, 0, ifp, &m0, inp, &pd->act) != PF_PASS)
7336 			goto bad;
7337 		else if (m0 == NULL)
7338 			goto done;
7339 		if (m0->m_len < sizeof(struct ip)) {
7340 			DPFPRINTF(PF_DEBUG_URGENT,
7341 			    ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
7342 			goto bad;
7343 		}
7344 		ip = mtod(m0, struct ip *);
7345 	}
7346 
7347 	if (ifp->if_flags & IFF_LOOPBACK)
7348 		m0->m_flags |= M_SKIP_FIREWALL;
7349 
7350 	ip_len = ntohs(ip->ip_len);
7351 	ip_off = ntohs(ip->ip_off);
7352 
7353 	/* Copied from FreeBSD 10.0-CURRENT ip_output. */
7354 	m0->m_pkthdr.csum_flags |= CSUM_IP;
7355 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
7356 		in_delayed_cksum(m0);
7357 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
7358 	}
7359 	if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
7360 		pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2));
7361 		m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
7362 	}
7363 
7364 	/*
7365 	 * If small enough for interface, or the interface will take
7366 	 * care of the fragmentation for us, we can just send directly.
7367 	 */
7368 	if (ip_len <= ifp->if_mtu ||
7369 	    (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
7370 		ip->ip_sum = 0;
7371 		if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
7372 			ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
7373 			m0->m_pkthdr.csum_flags &= ~CSUM_IP;
7374 		}
7375 		m_clrprotoflags(m0);	/* Avoid confusing lower layers. */
7376 
7377 		md = m0;
7378 		error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md);
7379 		if (md != NULL)
7380 			error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL);
7381 		goto done;
7382 	}
7383 
7384 	/* Balk when DF bit is set or the interface didn't support TSO. */
7385 	if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
7386 		error = EMSGSIZE;
7387 		KMOD_IPSTAT_INC(ips_cantfrag);
7388 		if (r_rt != PF_DUPTO) {
7389 			if (s && pd->nat_rule != NULL)
7390 				PACKET_UNDO_NAT(m0, pd,
7391 				    (ip->ip_hl << 2) + (ip_off & IP_OFFMASK),
7392 				    s);
7393 
7394 			icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
7395 			    ifp->if_mtu);
7396 			goto done;
7397 		} else
7398 			goto bad;
7399 	}
7400 
7401 	error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
7402 	if (error)
7403 		goto bad;
7404 
7405 	for (; m0; m0 = m1) {
7406 		m1 = m0->m_nextpkt;
7407 		m0->m_nextpkt = NULL;
7408 		if (error == 0) {
7409 			m_clrprotoflags(m0);
7410 			md = m0;
7411 			error = pf_dummynet_route(pd, s, r, ifp,
7412 			    sintosa(&dst), &md);
7413 			if (md != NULL)
7414 				error = (*ifp->if_output)(ifp, md,
7415 				    sintosa(&dst), NULL);
7416 		} else
7417 			m_freem(m0);
7418 	}
7419 
7420 	if (error == 0)
7421 		KMOD_IPSTAT_INC(ips_fragmented);
7422 
7423 done:
7424 	if (r_rt != PF_DUPTO)
7425 		*m = NULL;
7426 	return;
7427 
7428 bad_locked:
7429 	if (s)
7430 		PF_STATE_UNLOCK(s);
7431 bad:
7432 	m_freem(m0);
7433 	goto done;
7434 }
7435 #endif /* INET */
7436 
7437 #ifdef INET6
7438 static void
7439 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp,
7440     struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp)
7441 {
7442 	struct mbuf		*m0, *md;
7443 	struct sockaddr_in6	dst;
7444 	struct ip6_hdr		*ip6;
7445 	struct pfi_kkif		*nkif = NULL;
7446 	struct ifnet		*ifp = NULL;
7447 	struct pf_addr		 naddr;
7448 	struct pf_ksrc_node	*sn = NULL;
7449 	int			 r_rt, r_dir;
7450 
7451 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
7452 
7453 	if (s) {
7454 		r_rt = s->rt;
7455 		r_dir = s->direction;
7456 	} else {
7457 		r_rt = r->rt;
7458 		r_dir = r->direction;
7459 	}
7460 
7461 	KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT ||
7462 	    r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction",
7463 	    __func__));
7464 
7465 	if ((pd->pf_mtag == NULL &&
7466 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
7467 	    pd->pf_mtag->routed++ > 3) {
7468 		m0 = *m;
7469 		*m = NULL;
7470 		goto bad_locked;
7471 	}
7472 
7473 	if (r_rt == PF_DUPTO) {
7474 		if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) {
7475 			if (s == NULL) {
7476 				ifp = r->rpool.cur->kif ?
7477 				    r->rpool.cur->kif->pfik_ifp : NULL;
7478 			} else {
7479 				ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7480 				/* If pfsync'd */
7481 				if (ifp == NULL && r->rpool.cur != NULL)
7482 					ifp = r->rpool.cur->kif ?
7483 					    r->rpool.cur->kif->pfik_ifp : NULL;
7484 				PF_STATE_UNLOCK(s);
7485 			}
7486 			if (ifp == oifp) {
7487 				/* When the 2nd interface is not skipped */
7488 				return;
7489 			} else {
7490 				m0 = *m;
7491 				*m = NULL;
7492 				goto bad;
7493 			}
7494 		} else {
7495 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED;
7496 			if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) {
7497 				if (s)
7498 					PF_STATE_UNLOCK(s);
7499 				return;
7500 			}
7501 		}
7502 	} else {
7503 		if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) {
7504 			pf_dummynet(pd, s, r, m);
7505 			if (s)
7506 				PF_STATE_UNLOCK(s);
7507 			return;
7508 		}
7509 		m0 = *m;
7510 	}
7511 
7512 	ip6 = mtod(m0, struct ip6_hdr *);
7513 
7514 	bzero(&dst, sizeof(dst));
7515 	dst.sin6_family = AF_INET6;
7516 	dst.sin6_len = sizeof(dst);
7517 	dst.sin6_addr = ip6->ip6_dst;
7518 
7519 	bzero(&naddr, sizeof(naddr));
7520 
7521 	if (s == NULL) {
7522 		if (TAILQ_EMPTY(&r->rpool.list)) {
7523 			DPFPRINTF(PF_DEBUG_URGENT,
7524 			    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
7525 			goto bad_locked;
7526 		}
7527 		pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
7528 		    &naddr, &nkif, NULL, &sn);
7529 		if (!PF_AZERO(&naddr, AF_INET6))
7530 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
7531 			    &naddr, AF_INET6);
7532 		ifp = nkif ? nkif->pfik_ifp : NULL;
7533 	} else {
7534 		if (!PF_AZERO(&s->rt_addr, AF_INET6))
7535 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
7536 			    &s->rt_addr, AF_INET6);
7537 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7538 		/* If pfsync'd */
7539 		if (ifp == NULL && r->rpool.cur != NULL)
7540 			ifp = r->rpool.cur->kif ?
7541 			    r->rpool.cur->kif->pfik_ifp : NULL;
7542 	}
7543 
7544 	if (s)
7545 		PF_STATE_UNLOCK(s);
7546 
7547 	if (ifp == NULL)
7548 		goto bad;
7549 
7550 	if (pd->dir == PF_IN) {
7551 		if (pf_test6(PF_OUT, 0, ifp, &m0, inp, &pd->act) != PF_PASS)
7552 			goto bad;
7553 		else if (m0 == NULL)
7554 			goto done;
7555 		if (m0->m_len < sizeof(struct ip6_hdr)) {
7556 			DPFPRINTF(PF_DEBUG_URGENT,
7557 			    ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
7558 			    __func__));
7559 			goto bad;
7560 		}
7561 		ip6 = mtod(m0, struct ip6_hdr *);
7562 	}
7563 
7564 	if (ifp->if_flags & IFF_LOOPBACK)
7565 		m0->m_flags |= M_SKIP_FIREWALL;
7566 
7567 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
7568 	    ~ifp->if_hwassist) {
7569 		uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
7570 		in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
7571 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
7572 	}
7573 
7574 	/*
7575 	 * If the packet is too large for the outgoing interface,
7576 	 * send back an icmp6 error.
7577 	 */
7578 	if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
7579 		dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
7580 	if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) {
7581 		md = m0;
7582 		pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md);
7583 		if (md != NULL)
7584 			nd6_output_ifp(ifp, ifp, md, &dst, NULL);
7585 	}
7586 	else {
7587 		in6_ifstat_inc(ifp, ifs6_in_toobig);
7588 		if (r_rt != PF_DUPTO) {
7589 			if (s && pd->nat_rule != NULL)
7590 				PACKET_UNDO_NAT(m0, pd,
7591 				    ((caddr_t)ip6 - m0->m_data) +
7592 				    sizeof(struct ip6_hdr), s);
7593 
7594 			icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
7595 		} else
7596 			goto bad;
7597 	}
7598 
7599 done:
7600 	if (r_rt != PF_DUPTO)
7601 		*m = NULL;
7602 	return;
7603 
7604 bad_locked:
7605 	if (s)
7606 		PF_STATE_UNLOCK(s);
7607 bad:
7608 	m_freem(m0);
7609 	goto done;
7610 }
7611 #endif /* INET6 */
7612 
7613 /*
7614  * FreeBSD supports cksum offloads for the following drivers.
7615  *  em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4)
7616  *
7617  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
7618  *  network driver performed cksum including pseudo header, need to verify
7619  *   csum_data
7620  * CSUM_DATA_VALID :
7621  *  network driver performed cksum, needs to additional pseudo header
7622  *  cksum computation with partial csum_data(i.e. lack of H/W support for
7623  *  pseudo header, for instance sk(4) and possibly gem(4))
7624  *
7625  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
7626  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
7627  * TCP/UDP layer.
7628  * Also, set csum_data to 0xffff to force cksum validation.
7629  */
7630 static int
7631 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
7632 {
7633 	u_int16_t sum = 0;
7634 	int hw_assist = 0;
7635 	struct ip *ip;
7636 
7637 	if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
7638 		return (1);
7639 	if (m->m_pkthdr.len < off + len)
7640 		return (1);
7641 
7642 	switch (p) {
7643 	case IPPROTO_TCP:
7644 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
7645 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
7646 				sum = m->m_pkthdr.csum_data;
7647 			} else {
7648 				ip = mtod(m, struct ip *);
7649 				sum = in_pseudo(ip->ip_src.s_addr,
7650 				ip->ip_dst.s_addr, htonl((u_short)len +
7651 				m->m_pkthdr.csum_data + IPPROTO_TCP));
7652 			}
7653 			sum ^= 0xffff;
7654 			++hw_assist;
7655 		}
7656 		break;
7657 	case IPPROTO_UDP:
7658 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
7659 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
7660 				sum = m->m_pkthdr.csum_data;
7661 			} else {
7662 				ip = mtod(m, struct ip *);
7663 				sum = in_pseudo(ip->ip_src.s_addr,
7664 				ip->ip_dst.s_addr, htonl((u_short)len +
7665 				m->m_pkthdr.csum_data + IPPROTO_UDP));
7666 			}
7667 			sum ^= 0xffff;
7668 			++hw_assist;
7669 		}
7670 		break;
7671 	case IPPROTO_ICMP:
7672 #ifdef INET6
7673 	case IPPROTO_ICMPV6:
7674 #endif /* INET6 */
7675 		break;
7676 	default:
7677 		return (1);
7678 	}
7679 
7680 	if (!hw_assist) {
7681 		switch (af) {
7682 		case AF_INET:
7683 			if (p == IPPROTO_ICMP) {
7684 				if (m->m_len < off)
7685 					return (1);
7686 				m->m_data += off;
7687 				m->m_len -= off;
7688 				sum = in_cksum(m, len);
7689 				m->m_data -= off;
7690 				m->m_len += off;
7691 			} else {
7692 				if (m->m_len < sizeof(struct ip))
7693 					return (1);
7694 				sum = in4_cksum(m, p, off, len);
7695 			}
7696 			break;
7697 #ifdef INET6
7698 		case AF_INET6:
7699 			if (m->m_len < sizeof(struct ip6_hdr))
7700 				return (1);
7701 			sum = in6_cksum(m, p, off, len);
7702 			break;
7703 #endif /* INET6 */
7704 		default:
7705 			return (1);
7706 		}
7707 	}
7708 	if (sum) {
7709 		switch (p) {
7710 		case IPPROTO_TCP:
7711 		    {
7712 			KMOD_TCPSTAT_INC(tcps_rcvbadsum);
7713 			break;
7714 		    }
7715 		case IPPROTO_UDP:
7716 		    {
7717 			KMOD_UDPSTAT_INC(udps_badsum);
7718 			break;
7719 		    }
7720 #ifdef INET
7721 		case IPPROTO_ICMP:
7722 		    {
7723 			KMOD_ICMPSTAT_INC(icps_checksum);
7724 			break;
7725 		    }
7726 #endif
7727 #ifdef INET6
7728 		case IPPROTO_ICMPV6:
7729 		    {
7730 			KMOD_ICMP6STAT_INC(icp6s_checksum);
7731 			break;
7732 		    }
7733 #endif /* INET6 */
7734 		}
7735 		return (1);
7736 	} else {
7737 		if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
7738 			m->m_pkthdr.csum_flags |=
7739 			    (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
7740 			m->m_pkthdr.csum_data = 0xffff;
7741 		}
7742 	}
7743 	return (0);
7744 }
7745 
7746 static bool
7747 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r,
7748     const struct pf_kstate *s, struct ip_fw_args *dnflow)
7749 {
7750 	int dndir = r->direction;
7751 
7752 	if (s && dndir == PF_INOUT) {
7753 		dndir = s->direction;
7754 	} else if (dndir == PF_INOUT) {
7755 		/* Assume primary direction. Happens when we've set dnpipe in
7756 		 * the ethernet level code. */
7757 		dndir = pd->dir;
7758 	}
7759 
7760 	memset(dnflow, 0, sizeof(*dnflow));
7761 
7762 	if (pd->dport != NULL)
7763 		dnflow->f_id.dst_port = ntohs(*pd->dport);
7764 	if (pd->sport != NULL)
7765 		dnflow->f_id.src_port = ntohs(*pd->sport);
7766 
7767 	if (pd->dir == PF_IN)
7768 		dnflow->flags |= IPFW_ARGS_IN;
7769 	else
7770 		dnflow->flags |= IPFW_ARGS_OUT;
7771 
7772 	if (pd->dir != dndir && pd->act.dnrpipe) {
7773 		dnflow->rule.info = pd->act.dnrpipe;
7774 	}
7775 	else if (pd->dir == dndir && pd->act.dnpipe) {
7776 		dnflow->rule.info = pd->act.dnpipe;
7777 	}
7778 	else {
7779 		return (false);
7780 	}
7781 
7782 	dnflow->rule.info |= IPFW_IS_DUMMYNET;
7783 	if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE)
7784 		dnflow->rule.info |= IPFW_IS_PIPE;
7785 
7786 	dnflow->f_id.proto = pd->proto;
7787 	dnflow->f_id.extra = dnflow->rule.info;
7788 	switch (pd->af) {
7789 	case AF_INET:
7790 		dnflow->f_id.addr_type = 4;
7791 		dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr);
7792 		dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr);
7793 		break;
7794 	case AF_INET6:
7795 		dnflow->flags |= IPFW_ARGS_IP6;
7796 		dnflow->f_id.addr_type = 6;
7797 		dnflow->f_id.src_ip6 = pd->src->v6;
7798 		dnflow->f_id.dst_ip6 = pd->dst->v6;
7799 		break;
7800 	default:
7801 		panic("Invalid AF");
7802 		break;
7803 	}
7804 
7805 	return (true);
7806 }
7807 
7808 int
7809 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0,
7810     struct inpcb *inp)
7811 {
7812 	struct pfi_kkif		*kif;
7813 	struct mbuf		*m = *m0;
7814 
7815 	M_ASSERTPKTHDR(m);
7816 	MPASS(ifp->if_vnet == curvnet);
7817 	NET_EPOCH_ASSERT();
7818 
7819 	if (!V_pf_status.running)
7820 		return (PF_PASS);
7821 
7822 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
7823 
7824 	if (kif == NULL) {
7825 		DPFPRINTF(PF_DEBUG_URGENT,
7826 		    ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname));
7827 		return (PF_DROP);
7828 	}
7829 	if (kif->pfik_flags & PFI_IFLAG_SKIP)
7830 		return (PF_PASS);
7831 
7832 	if (m->m_flags & M_SKIP_FIREWALL)
7833 		return (PF_PASS);
7834 
7835 	/* Stateless! */
7836 	return (pf_test_eth_rule(dir, kif, m0));
7837 }
7838 
7839 static __inline void
7840 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag)
7841 {
7842 	struct m_tag *mtag;
7843 
7844 	pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET;
7845 
7846 	/* dummynet adds this tag, but pf does not need it,
7847 	 * and keeping it creates unexpected behavior,
7848 	 * e.g. in case of divert(4) usage right after dummynet. */
7849 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
7850 	if (mtag != NULL)
7851 		m_tag_delete(m, mtag);
7852 }
7853 
7854 static int
7855 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s,
7856     struct pf_krule *r, struct mbuf **m0)
7857 {
7858 	return (pf_dummynet_route(pd, s, r, NULL, NULL, m0));
7859 }
7860 
7861 static int
7862 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s,
7863     struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa,
7864     struct mbuf **m0)
7865 {
7866 	NET_EPOCH_ASSERT();
7867 
7868 	if (pd->act.dnpipe || pd->act.dnrpipe) {
7869 		struct ip_fw_args dnflow;
7870 		if (ip_dn_io_ptr == NULL) {
7871 			m_freem(*m0);
7872 			*m0 = NULL;
7873 			return (ENOMEM);
7874 		}
7875 
7876 		if (pd->pf_mtag == NULL &&
7877 		    ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) {
7878 			m_freem(*m0);
7879 			*m0 = NULL;
7880 			return (ENOMEM);
7881 		}
7882 
7883 		if (ifp != NULL) {
7884 			pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO;
7885 
7886 			pd->pf_mtag->if_index = ifp->if_index;
7887 			pd->pf_mtag->if_idxgen = ifp->if_idxgen;
7888 
7889 			MPASS(sa != NULL);
7890 
7891 			if (pd->af == AF_INET)
7892 				memcpy(&pd->pf_mtag->dst, sa,
7893 				    sizeof(struct sockaddr_in));
7894 			else
7895 				memcpy(&pd->pf_mtag->dst, sa,
7896 				    sizeof(struct sockaddr_in6));
7897 		}
7898 
7899 		if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) {
7900 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET;
7901 			ip_dn_io_ptr(m0, &dnflow);
7902 			if (*m0 != NULL) {
7903 				pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
7904 				pf_dummynet_flag_remove(*m0, pd->pf_mtag);
7905 			}
7906 		}
7907 	}
7908 
7909 	return (0);
7910 }
7911 
7912 #ifdef INET
7913 int
7914 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0,
7915     struct inpcb *inp, struct pf_rule_actions *default_actions)
7916 {
7917 	struct pfi_kkif		*kif;
7918 	u_short			 action, reason = 0;
7919 	struct mbuf		*m = *m0;
7920 	struct ip		*h = NULL;
7921 	struct m_tag		*mtag;
7922 	struct pf_krule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
7923 	struct pf_kstate	*s = NULL;
7924 	struct pf_kruleset	*ruleset = NULL;
7925 	struct pf_pdesc		 pd;
7926 	int			 off, dirndx, use_2nd_queue = 0;
7927 	uint16_t		 tag;
7928 	uint8_t			 rt;
7929 
7930 	PF_RULES_RLOCK_TRACKER;
7931 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir));
7932 	M_ASSERTPKTHDR(m);
7933 
7934 	if (!V_pf_status.running)
7935 		return (PF_PASS);
7936 
7937 	PF_RULES_RLOCK();
7938 
7939 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
7940 
7941 	if (__predict_false(kif == NULL)) {
7942 		DPFPRINTF(PF_DEBUG_URGENT,
7943 		    ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
7944 		PF_RULES_RUNLOCK();
7945 		return (PF_DROP);
7946 	}
7947 	if (kif->pfik_flags & PFI_IFLAG_SKIP) {
7948 		PF_RULES_RUNLOCK();
7949 		return (PF_PASS);
7950 	}
7951 
7952 	if (m->m_flags & M_SKIP_FIREWALL) {
7953 		PF_RULES_RUNLOCK();
7954 		return (PF_PASS);
7955 	}
7956 
7957 	memset(&pd, 0, sizeof(pd));
7958 	TAILQ_INIT(&pd.sctp_multihome_jobs);
7959 	if (default_actions != NULL)
7960 		memcpy(&pd.act, default_actions, sizeof(pd.act));
7961 	pd.pf_mtag = pf_find_mtag(m);
7962 
7963 	if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) {
7964 		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
7965 
7966 		ifp = ifnet_byindexgen(pd.pf_mtag->if_index,
7967 		    pd.pf_mtag->if_idxgen);
7968 		if (ifp == NULL || ifp->if_flags & IFF_DYING) {
7969 			PF_RULES_RUNLOCK();
7970 			m_freem(*m0);
7971 			*m0 = NULL;
7972 			return (PF_PASS);
7973 		}
7974 		PF_RULES_RUNLOCK();
7975 		(ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL);
7976 		*m0 = NULL;
7977 		return (PF_PASS);
7978 	}
7979 
7980 	if (pd.pf_mtag && pd.pf_mtag->dnpipe) {
7981 		pd.act.dnpipe = pd.pf_mtag->dnpipe;
7982 		pd.act.flags = pd.pf_mtag->dnflags;
7983 	}
7984 
7985 	if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL &&
7986 	    pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
7987 		/* Dummynet re-injects packets after they've
7988 		 * completed their delay. We've already
7989 		 * processed them, so pass unconditionally. */
7990 
7991 		/* But only once. We may see the packet multiple times (e.g.
7992 		 * PFIL_IN/PFIL_OUT). */
7993 		pf_dummynet_flag_remove(m, pd.pf_mtag);
7994 		PF_RULES_RUNLOCK();
7995 
7996 		return (PF_PASS);
7997 	}
7998 
7999 	pd.sport = pd.dport = NULL;
8000 	pd.proto_sum = NULL;
8001 	pd.dir = dir;
8002 	pd.sidx = (dir == PF_IN) ? 0 : 1;
8003 	pd.didx = (dir == PF_IN) ? 1 : 0;
8004 	pd.af = AF_INET;
8005 	pd.act.rtableid = -1;
8006 
8007 	h = mtod(m, struct ip *);
8008 	off = h->ip_hl << 2;
8009 
8010 	if (__predict_false(ip_divert_ptr != NULL) &&
8011 	    ((mtag = m_tag_locate(m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) {
8012 		struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1);
8013 		if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) ||
8014 		    (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) {
8015 			if (pd.pf_mtag == NULL &&
8016 			    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8017 				action = PF_DROP;
8018 				goto done;
8019 			}
8020 			pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED;
8021 		}
8022 		if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) {
8023 			m->m_flags |= M_FASTFWD_OURS;
8024 			pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT;
8025 		}
8026 		m_tag_delete(m, mtag);
8027 
8028 		mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
8029 		if (mtag != NULL)
8030 			m_tag_delete(m, mtag);
8031 	} else if (pf_normalize_ip(m0, kif, &reason, &pd) != PF_PASS) {
8032 		/* We do IP header normalization and packet reassembly here */
8033 		action = PF_DROP;
8034 		goto done;
8035 	}
8036 	m = *m0;	/* pf_normalize messes with m0 */
8037 	h = mtod(m, struct ip *);
8038 
8039 	off = h->ip_hl << 2;
8040 	if (off < (int)sizeof(struct ip)) {
8041 		action = PF_DROP;
8042 		REASON_SET(&reason, PFRES_SHORT);
8043 		pd.act.log = PF_LOG_FORCE;
8044 		goto done;
8045 	}
8046 
8047 	pd.src = (struct pf_addr *)&h->ip_src;
8048 	pd.dst = (struct pf_addr *)&h->ip_dst;
8049 	pd.ip_sum = &h->ip_sum;
8050 	pd.proto = h->ip_p;
8051 	pd.tos = h->ip_tos & ~IPTOS_ECN_MASK;
8052 	pd.tot_len = ntohs(h->ip_len);
8053 
8054 	/* handle fragments that didn't get reassembled by normalization */
8055 	if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
8056 		action = pf_test_fragment(&r, kif, m, h, &pd, &a, &ruleset);
8057 		goto done;
8058 	}
8059 
8060 	switch (h->ip_p) {
8061 	case IPPROTO_TCP: {
8062 		if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp),
8063 		    &action, &reason, AF_INET)) {
8064 			if (action != PF_PASS)
8065 				pd.act.log = PF_LOG_FORCE;
8066 			goto done;
8067 		}
8068 		pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2);
8069 
8070 		pd.sport = &pd.hdr.tcp.th_sport;
8071 		pd.dport = &pd.hdr.tcp.th_dport;
8072 
8073 		/* Respond to SYN with a syncookie. */
8074 		if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN &&
8075 		    pd.dir == PF_IN && pf_synflood_check(&pd)) {
8076 			pf_syncookie_send(m, off, &pd);
8077 			action = PF_DROP;
8078 			break;
8079 		}
8080 
8081 		if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0)
8082 			use_2nd_queue = 1;
8083 		action = pf_normalize_tcp(kif, m, 0, off, h, &pd);
8084 		if (action == PF_DROP)
8085 			goto done;
8086 		action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason);
8087 		if (action == PF_PASS) {
8088 			if (V_pfsync_update_state_ptr != NULL)
8089 				V_pfsync_update_state_ptr(s);
8090 			r = s->rule.ptr;
8091 			a = s->anchor.ptr;
8092 		} else if (s == NULL) {
8093 			/* Validate remote SYN|ACK, re-create original SYN if
8094 			 * valid. */
8095 			if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) ==
8096 			    TH_ACK && pf_syncookie_validate(&pd) &&
8097 			    pd.dir == PF_IN) {
8098 				struct mbuf *msyn;
8099 
8100 				msyn = pf_syncookie_recreate_syn(h->ip_ttl, off,
8101 				    &pd);
8102 				if (msyn == NULL) {
8103 					action = PF_DROP;
8104 					break;
8105 				}
8106 
8107 				action = pf_test(dir, pflags, ifp, &msyn, inp,
8108 				    &pd.act);
8109 				m_freem(msyn);
8110 				if (action != PF_PASS)
8111 					break;
8112 
8113 				action = pf_test_state_tcp(&s, kif, m, off, h,
8114 				    &pd, &reason);
8115 				if (action != PF_PASS || s == NULL) {
8116 					action = PF_DROP;
8117 					break;
8118 				}
8119 
8120 				s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1;
8121 				s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1;
8122 				pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST);
8123 				action = pf_synproxy(&pd, &s, &reason);
8124 				break;
8125 			} else {
8126 				action = pf_test_rule(&r, &s, kif, m, off, &pd,
8127 				    &a, &ruleset, inp);
8128 			}
8129 		}
8130 		break;
8131 	}
8132 
8133 	case IPPROTO_UDP: {
8134 		if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp),
8135 		    &action, &reason, AF_INET)) {
8136 			if (action != PF_PASS)
8137 				pd.act.log = PF_LOG_FORCE;
8138 			goto done;
8139 		}
8140 		pd.sport = &pd.hdr.udp.uh_sport;
8141 		pd.dport = &pd.hdr.udp.uh_dport;
8142 		if (pd.hdr.udp.uh_dport == 0 ||
8143 		    ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off ||
8144 		    ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) {
8145 			action = PF_DROP;
8146 			REASON_SET(&reason, PFRES_SHORT);
8147 			goto done;
8148 		}
8149 		action = pf_test_state_udp(&s, kif, m, off, h, &pd);
8150 		if (action == PF_PASS) {
8151 			if (V_pfsync_update_state_ptr != NULL)
8152 				V_pfsync_update_state_ptr(s);
8153 			r = s->rule.ptr;
8154 			a = s->anchor.ptr;
8155 		} else if (s == NULL)
8156 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8157 			    &a, &ruleset, inp);
8158 		break;
8159 	}
8160 
8161 	case IPPROTO_SCTP: {
8162 		if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp),
8163 		    &action, &reason, AF_INET)) {
8164 			if (action != PF_PASS)
8165 				pd.act.log |= PF_LOG_FORCE;
8166 			goto done;
8167 		}
8168 		pd.p_len = pd.tot_len - off;
8169 
8170 		pd.sport = &pd.hdr.sctp.src_port;
8171 		pd.dport = &pd.hdr.sctp.dest_port;
8172 		if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) {
8173 			action = PF_DROP;
8174 			REASON_SET(&reason, PFRES_SHORT);
8175 			goto done;
8176 		}
8177 		action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd);
8178 		if (action == PF_DROP)
8179 			goto done;
8180 		action = pf_test_state_sctp(&s, kif, m, off, h, &pd,
8181 		    &reason);
8182 		if (action == PF_PASS) {
8183 			if (V_pfsync_update_state_ptr != NULL)
8184 				V_pfsync_update_state_ptr(s);
8185 			r = s->rule.ptr;
8186 			a = s->anchor.ptr;
8187 		} else {
8188 			action = pf_test_rule(&r, &s, kif, m, off,
8189 			    &pd, &a, &ruleset, inp);
8190 		}
8191 		break;
8192 	}
8193 
8194 	case IPPROTO_ICMP: {
8195 		if (!pf_pull_hdr(m, off, &pd.hdr.icmp, ICMP_MINLEN,
8196 		    &action, &reason, AF_INET)) {
8197 			if (action != PF_PASS)
8198 				pd.act.log = PF_LOG_FORCE;
8199 			goto done;
8200 		}
8201 		action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason);
8202 		if (action == PF_PASS) {
8203 			if (V_pfsync_update_state_ptr != NULL)
8204 				V_pfsync_update_state_ptr(s);
8205 			r = s->rule.ptr;
8206 			a = s->anchor.ptr;
8207 		} else if (s == NULL)
8208 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8209 			    &a, &ruleset, inp);
8210 		break;
8211 	}
8212 
8213 #ifdef INET6
8214 	case IPPROTO_ICMPV6: {
8215 		action = PF_DROP;
8216 		DPFPRINTF(PF_DEBUG_MISC,
8217 		    ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
8218 		goto done;
8219 	}
8220 #endif
8221 
8222 	default:
8223 		action = pf_test_state_other(&s, kif, m, &pd);
8224 		if (action == PF_PASS) {
8225 			if (V_pfsync_update_state_ptr != NULL)
8226 				V_pfsync_update_state_ptr(s);
8227 			r = s->rule.ptr;
8228 			a = s->anchor.ptr;
8229 		} else if (s == NULL)
8230 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8231 			    &a, &ruleset, inp);
8232 		break;
8233 	}
8234 
8235 done:
8236 	PF_RULES_RUNLOCK();
8237 	if (action == PF_PASS && h->ip_hl > 5 &&
8238 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
8239 		action = PF_DROP;
8240 		REASON_SET(&reason, PFRES_IPOPTIONS);
8241 		pd.act.log = PF_LOG_FORCE;
8242 		DPFPRINTF(PF_DEBUG_MISC,
8243 		    ("pf: dropping packet with ip options\n"));
8244 	}
8245 
8246 	if (s) {
8247 		uint8_t log = pd.act.log;
8248 		memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions));
8249 		pd.act.log |= log;
8250 		tag = s->tag;
8251 		rt = s->rt;
8252 	} else {
8253 		tag = r->tag;
8254 		rt = r->rt;
8255 	}
8256 
8257 	if (tag > 0 && pf_tag_packet(m, &pd, tag)) {
8258 		action = PF_DROP;
8259 		REASON_SET(&reason, PFRES_MEMORY);
8260 	}
8261 
8262 	pf_scrub_ip(&m, &pd);
8263 	if (pd.proto == IPPROTO_TCP && pd.act.max_mss)
8264 		pf_normalize_mss(m, off, &pd);
8265 
8266 	if (pd.act.rtableid >= 0)
8267 		M_SETFIB(m, pd.act.rtableid);
8268 
8269 	if (pd.act.flags & PFSTATE_SETPRIO) {
8270 		if (pd.tos & IPTOS_LOWDELAY)
8271 			use_2nd_queue = 1;
8272 		if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) {
8273 			action = PF_DROP;
8274 			REASON_SET(&reason, PFRES_MEMORY);
8275 			pd.act.log = PF_LOG_FORCE;
8276 			DPFPRINTF(PF_DEBUG_MISC,
8277 			    ("pf: failed to allocate 802.1q mtag\n"));
8278 		}
8279 	}
8280 
8281 #ifdef ALTQ
8282 	if (action == PF_PASS && pd.act.qid) {
8283 		if (pd.pf_mtag == NULL &&
8284 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8285 			action = PF_DROP;
8286 			REASON_SET(&reason, PFRES_MEMORY);
8287 		} else {
8288 			if (s != NULL)
8289 				pd.pf_mtag->qid_hash = pf_state_hash(s);
8290 			if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY))
8291 				pd.pf_mtag->qid = pd.act.pqid;
8292 			else
8293 				pd.pf_mtag->qid = pd.act.qid;
8294 			/* Add hints for ecn. */
8295 			pd.pf_mtag->hdr = h;
8296 		}
8297 	}
8298 #endif /* ALTQ */
8299 
8300 	/*
8301 	 * connections redirected to loopback should not match sockets
8302 	 * bound specifically to loopback due to security implications,
8303 	 * see tcp_input() and in_pcblookup_listen().
8304 	 */
8305 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
8306 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
8307 	    (s->nat_rule.ptr->action == PF_RDR ||
8308 	    s->nat_rule.ptr->action == PF_BINAT) &&
8309 	    IN_LOOPBACK(ntohl(pd.dst->v4.s_addr)))
8310 		m->m_flags |= M_SKIP_FIREWALL;
8311 
8312 	if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS &&
8313 	    r->divert.port && !PACKET_LOOPED(&pd)) {
8314 		mtag = m_tag_alloc(MTAG_PF_DIVERT, 0,
8315 		    sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO);
8316 		if (mtag != NULL) {
8317 			((struct pf_divert_mtag *)(mtag+1))->port =
8318 			    ntohs(r->divert.port);
8319 			((struct pf_divert_mtag *)(mtag+1))->idir =
8320 			    (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN :
8321 			    PF_DIVERT_MTAG_DIR_OUT;
8322 
8323 			if (s)
8324 				PF_STATE_UNLOCK(s);
8325 
8326 			m_tag_prepend(m, mtag);
8327 			if (m->m_flags & M_FASTFWD_OURS) {
8328 				if (pd.pf_mtag == NULL &&
8329 				    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8330 					action = PF_DROP;
8331 					REASON_SET(&reason, PFRES_MEMORY);
8332 					pd.act.log = PF_LOG_FORCE;
8333 					DPFPRINTF(PF_DEBUG_MISC,
8334 					    ("pf: failed to allocate tag\n"));
8335 				} else {
8336 					pd.pf_mtag->flags |=
8337 					    PF_MTAG_FLAG_FASTFWD_OURS_PRESENT;
8338 					m->m_flags &= ~M_FASTFWD_OURS;
8339 				}
8340 			}
8341 			ip_divert_ptr(*m0, dir == PF_IN);
8342 			*m0 = NULL;
8343 
8344 			return (action);
8345 		} else {
8346 			/* XXX: ipfw has the same behaviour! */
8347 			action = PF_DROP;
8348 			REASON_SET(&reason, PFRES_MEMORY);
8349 			pd.act.log = PF_LOG_FORCE;
8350 			DPFPRINTF(PF_DEBUG_MISC,
8351 			    ("pf: failed to allocate divert tag\n"));
8352 		}
8353 	}
8354 	/* this flag will need revising if the pkt is forwarded */
8355 	if (pd.pf_mtag)
8356 		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED;
8357 
8358 	if (pd.act.log) {
8359 		struct pf_krule		*lr;
8360 		struct pf_krule_item	*ri;
8361 
8362 		if (s != NULL && s->nat_rule.ptr != NULL &&
8363 		    s->nat_rule.ptr->log & PF_LOG_ALL)
8364 			lr = s->nat_rule.ptr;
8365 		else
8366 			lr = r;
8367 
8368 		if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL)
8369 			PFLOG_PACKET(kif, m, AF_INET, action, reason, lr, a,
8370 			    ruleset, &pd, (s == NULL));
8371 		if (s) {
8372 			SLIST_FOREACH(ri, &s->match_rules, entry)
8373 				if (ri->r->log & PF_LOG_ALL)
8374 					PFLOG_PACKET(kif, m, AF_INET, action,
8375 					    reason, ri->r, a, ruleset, &pd, 0);
8376 		}
8377 	}
8378 
8379 	pf_counter_u64_critical_enter();
8380 	pf_counter_u64_add_protected(&kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS],
8381 	    pd.tot_len);
8382 	pf_counter_u64_add_protected(&kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS],
8383 	    1);
8384 
8385 	if (action == PF_PASS || r->action == PF_DROP) {
8386 		dirndx = (dir == PF_OUT);
8387 		pf_counter_u64_add_protected(&r->packets[dirndx], 1);
8388 		pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len);
8389 		pf_update_timestamp(r);
8390 
8391 		if (a != NULL) {
8392 			pf_counter_u64_add_protected(&a->packets[dirndx], 1);
8393 			pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len);
8394 		}
8395 		if (s != NULL) {
8396 			struct pf_krule_item	*ri;
8397 
8398 			if (s->nat_rule.ptr != NULL) {
8399 				pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx],
8400 				    1);
8401 				pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx],
8402 				    pd.tot_len);
8403 			}
8404 			if (s->src_node != NULL) {
8405 				counter_u64_add(s->src_node->packets[dirndx],
8406 				    1);
8407 				counter_u64_add(s->src_node->bytes[dirndx],
8408 				    pd.tot_len);
8409 			}
8410 			if (s->nat_src_node != NULL) {
8411 				counter_u64_add(s->nat_src_node->packets[dirndx],
8412 				    1);
8413 				counter_u64_add(s->nat_src_node->bytes[dirndx],
8414 				    pd.tot_len);
8415 			}
8416 			dirndx = (dir == s->direction) ? 0 : 1;
8417 			s->packets[dirndx]++;
8418 			s->bytes[dirndx] += pd.tot_len;
8419 			SLIST_FOREACH(ri, &s->match_rules, entry) {
8420 				pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1);
8421 				pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd.tot_len);
8422 			}
8423 		}
8424 		tr = r;
8425 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
8426 		if (nr != NULL && r == &V_pf_default_rule)
8427 			tr = nr;
8428 		if (tr->src.addr.type == PF_ADDR_TABLE)
8429 			pfr_update_stats(tr->src.addr.p.tbl,
8430 			    (s == NULL) ? pd.src :
8431 			    &s->key[(s->direction == PF_IN)]->
8432 				addr[(s->direction == PF_OUT)],
8433 			    pd.af, pd.tot_len, dir == PF_OUT,
8434 			    r->action == PF_PASS, tr->src.neg);
8435 		if (tr->dst.addr.type == PF_ADDR_TABLE)
8436 			pfr_update_stats(tr->dst.addr.p.tbl,
8437 			    (s == NULL) ? pd.dst :
8438 			    &s->key[(s->direction == PF_IN)]->
8439 				addr[(s->direction == PF_IN)],
8440 			    pd.af, pd.tot_len, dir == PF_OUT,
8441 			    r->action == PF_PASS, tr->dst.neg);
8442 	}
8443 	pf_counter_u64_critical_exit();
8444 
8445 	switch (action) {
8446 	case PF_SYNPROXY_DROP:
8447 		m_freem(*m0);
8448 	case PF_DEFER:
8449 		*m0 = NULL;
8450 		action = PF_PASS;
8451 		break;
8452 	case PF_DROP:
8453 		m_freem(*m0);
8454 		*m0 = NULL;
8455 		break;
8456 	default:
8457 		/* pf_route() returns unlocked. */
8458 		if (rt) {
8459 			pf_route(m0, r, kif->pfik_ifp, s, &pd, inp);
8460 			goto out;
8461 		}
8462 		if (pf_dummynet(&pd, s, r, m0) != 0) {
8463 			action = PF_DROP;
8464 			REASON_SET(&reason, PFRES_MEMORY);
8465 		}
8466 		break;
8467 	}
8468 
8469 	SDT_PROBE4(pf, ip, test, done, action, reason, r, s);
8470 
8471 	if (s && action != PF_DROP) {
8472 		if (!s->if_index_in && dir == PF_IN)
8473 			s->if_index_in = ifp->if_index;
8474 		else if (!s->if_index_out && dir == PF_OUT)
8475 			s->if_index_out = ifp->if_index;
8476 	}
8477 
8478 	if (s)
8479 		PF_STATE_UNLOCK(s);
8480 
8481 out:
8482 	pf_sctp_multihome_delayed(&pd, off, kif, s, action);
8483 
8484 	return (action);
8485 }
8486 #endif /* INET */
8487 
8488 #ifdef INET6
8489 int
8490 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp,
8491     struct pf_rule_actions *default_actions)
8492 {
8493 	struct pfi_kkif		*kif;
8494 	u_short			 action, reason = 0;
8495 	struct mbuf		*m = *m0, *n = NULL;
8496 	struct m_tag		*mtag;
8497 	struct ip6_hdr		*h = NULL;
8498 	struct pf_krule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
8499 	struct pf_kstate	*s = NULL;
8500 	struct pf_kruleset	*ruleset = NULL;
8501 	struct pf_pdesc		 pd;
8502 	int			 off, terminal = 0, dirndx, rh_cnt = 0, use_2nd_queue = 0;
8503 	uint16_t		 tag;
8504 	uint8_t			 rt;
8505 
8506 	PF_RULES_RLOCK_TRACKER;
8507 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir));
8508 	M_ASSERTPKTHDR(m);
8509 
8510 	if (!V_pf_status.running)
8511 		return (PF_PASS);
8512 
8513 	PF_RULES_RLOCK();
8514 
8515 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
8516 	if (__predict_false(kif == NULL)) {
8517 		DPFPRINTF(PF_DEBUG_URGENT,
8518 		    ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
8519 		PF_RULES_RUNLOCK();
8520 		return (PF_DROP);
8521 	}
8522 	if (kif->pfik_flags & PFI_IFLAG_SKIP) {
8523 		PF_RULES_RUNLOCK();
8524 		return (PF_PASS);
8525 	}
8526 
8527 	if (m->m_flags & M_SKIP_FIREWALL) {
8528 		PF_RULES_RUNLOCK();
8529 		return (PF_PASS);
8530 	}
8531 
8532 	/*
8533 	 * If we end up changing IP addresses (e.g. binat) the stack may get
8534 	 * confused and fail to send the icmp6 packet too big error. Just send
8535 	 * it here, before we do any NAT.
8536 	 */
8537 	if (dir == PF_OUT && pflags & PFIL_FWD && IN6_LINKMTU(ifp) < pf_max_frag_size(m)) {
8538 		PF_RULES_RUNLOCK();
8539 		*m0 = NULL;
8540 		icmp6_error(m, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp));
8541 		return (PF_DROP);
8542 	}
8543 
8544 	memset(&pd, 0, sizeof(pd));
8545 	TAILQ_INIT(&pd.sctp_multihome_jobs);
8546 	if (default_actions != NULL)
8547 		memcpy(&pd.act, default_actions, sizeof(pd.act));
8548 	pd.pf_mtag = pf_find_mtag(m);
8549 
8550 	if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) {
8551 		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
8552 
8553 		ifp = ifnet_byindexgen(pd.pf_mtag->if_index,
8554 		    pd.pf_mtag->if_idxgen);
8555 		if (ifp == NULL || ifp->if_flags & IFF_DYING) {
8556 			PF_RULES_RUNLOCK();
8557 			m_freem(*m0);
8558 			*m0 = NULL;
8559 			return (PF_PASS);
8560 		}
8561 		PF_RULES_RUNLOCK();
8562 		nd6_output_ifp(ifp, ifp, m,
8563                     (struct sockaddr_in6 *)&pd.pf_mtag->dst, NULL);
8564 		*m0 = NULL;
8565 		return (PF_PASS);
8566 	}
8567 
8568 	if (pd.pf_mtag && pd.pf_mtag->dnpipe) {
8569 		pd.act.dnpipe = pd.pf_mtag->dnpipe;
8570 		pd.act.flags = pd.pf_mtag->dnflags;
8571 	}
8572 
8573 	if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL &&
8574 	    pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
8575 		pf_dummynet_flag_remove(m, pd.pf_mtag);
8576 		/* Dummynet re-injects packets after they've
8577 		 * completed their delay. We've already
8578 		 * processed them, so pass unconditionally. */
8579 		PF_RULES_RUNLOCK();
8580 		return (PF_PASS);
8581 	}
8582 
8583 	pd.sport = pd.dport = NULL;
8584 	pd.ip_sum = NULL;
8585 	pd.proto_sum = NULL;
8586 	pd.dir = dir;
8587 	pd.sidx = (dir == PF_IN) ? 0 : 1;
8588 	pd.didx = (dir == PF_IN) ? 1 : 0;
8589 	pd.af = AF_INET6;
8590 	pd.act.rtableid = -1;
8591 
8592 	h = mtod(m, struct ip6_hdr *);
8593 	off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
8594 
8595 	/* We do IP header normalization and packet reassembly here */
8596 	if (pf_normalize_ip6(m0, kif, &reason, &pd) != PF_PASS) {
8597 		action = PF_DROP;
8598 		goto done;
8599 	}
8600 	m = *m0;	/* pf_normalize messes with m0 */
8601 	h = mtod(m, struct ip6_hdr *);
8602 	off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
8603 
8604 	/*
8605 	 * we do not support jumbogram.  if we keep going, zero ip6_plen
8606 	 * will do something bad, so drop the packet for now.
8607 	 */
8608 	if (htons(h->ip6_plen) == 0) {
8609 		action = PF_DROP;
8610 		REASON_SET(&reason, PFRES_NORM);	/*XXX*/
8611 		goto done;
8612 	}
8613 
8614 	pd.src = (struct pf_addr *)&h->ip6_src;
8615 	pd.dst = (struct pf_addr *)&h->ip6_dst;
8616 	pd.tos = IPV6_DSCP(h);
8617 	pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
8618 
8619 	pd.proto = h->ip6_nxt;
8620 	do {
8621 		switch (pd.proto) {
8622 		case IPPROTO_FRAGMENT:
8623 			action = pf_test_fragment(&r, kif, m, h, &pd, &a,
8624 			    &ruleset);
8625 			if (action == PF_DROP)
8626 				REASON_SET(&reason, PFRES_FRAG);
8627 			goto done;
8628 		case IPPROTO_ROUTING: {
8629 			struct ip6_rthdr rthdr;
8630 
8631 			if (rh_cnt++) {
8632 				DPFPRINTF(PF_DEBUG_MISC,
8633 				    ("pf: IPv6 more than one rthdr\n"));
8634 				action = PF_DROP;
8635 				REASON_SET(&reason, PFRES_IPOPTIONS);
8636 				pd.act.log = PF_LOG_FORCE;
8637 				goto done;
8638 			}
8639 			if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
8640 			    &reason, pd.af)) {
8641 				DPFPRINTF(PF_DEBUG_MISC,
8642 				    ("pf: IPv6 short rthdr\n"));
8643 				action = PF_DROP;
8644 				REASON_SET(&reason, PFRES_SHORT);
8645 				pd.act.log = PF_LOG_FORCE;
8646 				goto done;
8647 			}
8648 			if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
8649 				DPFPRINTF(PF_DEBUG_MISC,
8650 				    ("pf: IPv6 rthdr0\n"));
8651 				action = PF_DROP;
8652 				REASON_SET(&reason, PFRES_IPOPTIONS);
8653 				pd.act.log = PF_LOG_FORCE;
8654 				goto done;
8655 			}
8656 			/* FALLTHROUGH */
8657 		}
8658 		case IPPROTO_AH:
8659 		case IPPROTO_HOPOPTS:
8660 		case IPPROTO_DSTOPTS: {
8661 			/* get next header and header length */
8662 			struct ip6_ext	opt6;
8663 
8664 			if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
8665 			    NULL, &reason, pd.af)) {
8666 				DPFPRINTF(PF_DEBUG_MISC,
8667 				    ("pf: IPv6 short opt\n"));
8668 				action = PF_DROP;
8669 				pd.act.log = PF_LOG_FORCE;
8670 				goto done;
8671 			}
8672 			if (pd.proto == IPPROTO_AH)
8673 				off += (opt6.ip6e_len + 2) * 4;
8674 			else
8675 				off += (opt6.ip6e_len + 1) * 8;
8676 			pd.proto = opt6.ip6e_nxt;
8677 			/* goto the next header */
8678 			break;
8679 		}
8680 		default:
8681 			terminal++;
8682 			break;
8683 		}
8684 	} while (!terminal);
8685 
8686 	/* if there's no routing header, use unmodified mbuf for checksumming */
8687 	if (!n)
8688 		n = m;
8689 
8690 	switch (pd.proto) {
8691 	case IPPROTO_TCP: {
8692 		if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp),
8693 		    &action, &reason, AF_INET6)) {
8694 			if (action != PF_PASS)
8695 				pd.act.log |= PF_LOG_FORCE;
8696 			goto done;
8697 		}
8698 		pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2);
8699 		pd.sport = &pd.hdr.tcp.th_sport;
8700 		pd.dport = &pd.hdr.tcp.th_dport;
8701 
8702 		/* Respond to SYN with a syncookie. */
8703 		if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN &&
8704 		    pd.dir == PF_IN && pf_synflood_check(&pd)) {
8705 			pf_syncookie_send(m, off, &pd);
8706 			action = PF_DROP;
8707 			break;
8708 		}
8709 
8710 		action = pf_normalize_tcp(kif, m, 0, off, h, &pd);
8711 		if (action == PF_DROP)
8712 			goto done;
8713 		action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason);
8714 		if (action == PF_PASS) {
8715 			if (V_pfsync_update_state_ptr != NULL)
8716 				V_pfsync_update_state_ptr(s);
8717 			r = s->rule.ptr;
8718 			a = s->anchor.ptr;
8719 		} else if (s == NULL) {
8720 			/* Validate remote SYN|ACK, re-create original SYN if
8721 			 * valid. */
8722 			if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) ==
8723 			    TH_ACK && pf_syncookie_validate(&pd) &&
8724 			    pd.dir == PF_IN) {
8725 				struct mbuf *msyn;
8726 
8727 				msyn = pf_syncookie_recreate_syn(h->ip6_hlim,
8728 				    off, &pd);
8729 				if (msyn == NULL) {
8730 					action = PF_DROP;
8731 					break;
8732 				}
8733 
8734 				action = pf_test6(dir, pflags, ifp, &msyn, inp,
8735 				    &pd.act);
8736 				m_freem(msyn);
8737 				if (action != PF_PASS)
8738 					break;
8739 
8740 				action = pf_test_state_tcp(&s, kif, m, off, h,
8741 				    &pd, &reason);
8742 				if (action != PF_PASS || s == NULL) {
8743 					action = PF_DROP;
8744 					break;
8745 				}
8746 
8747 				s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1;
8748 				s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1;
8749 				pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST);
8750 
8751 				action = pf_synproxy(&pd, &s, &reason);
8752 				break;
8753 			} else {
8754 				action = pf_test_rule(&r, &s, kif, m, off, &pd,
8755 				    &a, &ruleset, inp);
8756 			}
8757 		}
8758 		break;
8759 	}
8760 
8761 	case IPPROTO_UDP: {
8762 		if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp),
8763 		    &action, &reason, AF_INET6)) {
8764 			if (action != PF_PASS)
8765 				pd.act.log |= PF_LOG_FORCE;
8766 			goto done;
8767 		}
8768 		pd.sport = &pd.hdr.udp.uh_sport;
8769 		pd.dport = &pd.hdr.udp.uh_dport;
8770 		if (pd.hdr.udp.uh_dport == 0 ||
8771 		    ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off ||
8772 		    ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) {
8773 			action = PF_DROP;
8774 			REASON_SET(&reason, PFRES_SHORT);
8775 			goto done;
8776 		}
8777 		action = pf_test_state_udp(&s, kif, m, off, h, &pd);
8778 		if (action == PF_PASS) {
8779 			if (V_pfsync_update_state_ptr != NULL)
8780 				V_pfsync_update_state_ptr(s);
8781 			r = s->rule.ptr;
8782 			a = s->anchor.ptr;
8783 		} else if (s == NULL)
8784 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8785 			    &a, &ruleset, inp);
8786 		break;
8787 	}
8788 
8789 	case IPPROTO_SCTP: {
8790 		if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp),
8791 		    &action, &reason, AF_INET6)) {
8792 			if (action != PF_PASS)
8793 				pd.act.log |= PF_LOG_FORCE;
8794 			goto done;
8795 		}
8796 		pd.sport = &pd.hdr.sctp.src_port;
8797 		pd.dport = &pd.hdr.sctp.dest_port;
8798 		if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) {
8799 			action = PF_DROP;
8800 			REASON_SET(&reason, PFRES_SHORT);
8801 			goto done;
8802 		}
8803 		action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd);
8804 		if (action == PF_DROP)
8805 			goto done;
8806 		action = pf_test_state_sctp(&s, kif, m, off, h, &pd,
8807 		    &reason);
8808 		if (action == PF_PASS) {
8809 			if (V_pfsync_update_state_ptr != NULL)
8810 				V_pfsync_update_state_ptr(s);
8811 			r = s->rule.ptr;
8812 			a = s->anchor.ptr;
8813 		} else {
8814 			action = pf_test_rule(&r, &s, kif, m, off,
8815 			    &pd, &a, &ruleset, inp);
8816 		}
8817 		break;
8818 	}
8819 
8820 	case IPPROTO_ICMP: {
8821 		action = PF_DROP;
8822 		DPFPRINTF(PF_DEBUG_MISC,
8823 		    ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
8824 		goto done;
8825 	}
8826 
8827 	case IPPROTO_ICMPV6: {
8828 		if (!pf_pull_hdr(m, off, &pd.hdr.icmp6, sizeof(pd.hdr.icmp6),
8829 		    &action, &reason, AF_INET6)) {
8830 			if (action != PF_PASS)
8831 				pd.act.log |= PF_LOG_FORCE;
8832 			goto done;
8833 		}
8834 		action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason);
8835 		if (action == PF_PASS) {
8836 			if (V_pfsync_update_state_ptr != NULL)
8837 				V_pfsync_update_state_ptr(s);
8838 			r = s->rule.ptr;
8839 			a = s->anchor.ptr;
8840 		} else if (s == NULL)
8841 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8842 			    &a, &ruleset, inp);
8843 		break;
8844 	}
8845 
8846 	default:
8847 		action = pf_test_state_other(&s, kif, m, &pd);
8848 		if (action == PF_PASS) {
8849 			if (V_pfsync_update_state_ptr != NULL)
8850 				V_pfsync_update_state_ptr(s);
8851 			r = s->rule.ptr;
8852 			a = s->anchor.ptr;
8853 		} else if (s == NULL)
8854 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8855 			    &a, &ruleset, inp);
8856 		break;
8857 	}
8858 
8859 done:
8860 	PF_RULES_RUNLOCK();
8861 	if (n != m) {
8862 		m_freem(n);
8863 		n = NULL;
8864 	}
8865 
8866 	/* handle dangerous IPv6 extension headers. */
8867 	if (action == PF_PASS && rh_cnt &&
8868 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
8869 		action = PF_DROP;
8870 		REASON_SET(&reason, PFRES_IPOPTIONS);
8871 		pd.act.log = r->log;
8872 		DPFPRINTF(PF_DEBUG_MISC,
8873 		    ("pf: dropping packet with dangerous v6 headers\n"));
8874 	}
8875 
8876 	if (s) {
8877 		uint8_t log = pd.act.log;
8878 		memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions));
8879 		pd.act.log |= log;
8880 		tag = s->tag;
8881 		rt = s->rt;
8882 	} else {
8883 		tag = r->tag;
8884 		rt = r->rt;
8885 	}
8886 
8887 	if (tag > 0 && pf_tag_packet(m, &pd, tag)) {
8888 		action = PF_DROP;
8889 		REASON_SET(&reason, PFRES_MEMORY);
8890 	}
8891 
8892 	pf_scrub_ip6(&m, &pd);
8893 	if (pd.proto == IPPROTO_TCP && pd.act.max_mss)
8894 		pf_normalize_mss(m, off, &pd);
8895 
8896 	if (pd.act.rtableid >= 0)
8897 		M_SETFIB(m, pd.act.rtableid);
8898 
8899 	if (pd.act.flags & PFSTATE_SETPRIO) {
8900 		if (pd.tos & IPTOS_LOWDELAY)
8901 			use_2nd_queue = 1;
8902 		if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) {
8903 			action = PF_DROP;
8904 			REASON_SET(&reason, PFRES_MEMORY);
8905 			pd.act.log = PF_LOG_FORCE;
8906 			DPFPRINTF(PF_DEBUG_MISC,
8907 			    ("pf: failed to allocate 802.1q mtag\n"));
8908 		}
8909 	}
8910 
8911 #ifdef ALTQ
8912 	if (action == PF_PASS && pd.act.qid) {
8913 		if (pd.pf_mtag == NULL &&
8914 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8915 			action = PF_DROP;
8916 			REASON_SET(&reason, PFRES_MEMORY);
8917 		} else {
8918 			if (s != NULL)
8919 				pd.pf_mtag->qid_hash = pf_state_hash(s);
8920 			if (pd.tos & IPTOS_LOWDELAY)
8921 				pd.pf_mtag->qid = pd.act.pqid;
8922 			else
8923 				pd.pf_mtag->qid = pd.act.qid;
8924 			/* Add hints for ecn. */
8925 			pd.pf_mtag->hdr = h;
8926 		}
8927 	}
8928 #endif /* ALTQ */
8929 
8930 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
8931 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
8932 	    (s->nat_rule.ptr->action == PF_RDR ||
8933 	    s->nat_rule.ptr->action == PF_BINAT) &&
8934 	    IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
8935 		m->m_flags |= M_SKIP_FIREWALL;
8936 
8937 	/* XXX: Anybody working on it?! */
8938 	if (r->divert.port)
8939 		printf("pf: divert(9) is not supported for IPv6\n");
8940 
8941 	if (pd.act.log) {
8942 		struct pf_krule		*lr;
8943 		struct pf_krule_item	*ri;
8944 
8945 		if (s != NULL && s->nat_rule.ptr != NULL &&
8946 		    s->nat_rule.ptr->log & PF_LOG_ALL)
8947 			lr = s->nat_rule.ptr;
8948 		else
8949 			lr = r;
8950 
8951 		if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL)
8952 			PFLOG_PACKET(kif, m, AF_INET6, action, reason, lr, a, ruleset,
8953 			    &pd, (s == NULL));
8954 		if (s) {
8955 			SLIST_FOREACH(ri, &s->match_rules, entry)
8956 				if (ri->r->log & PF_LOG_ALL)
8957 					PFLOG_PACKET(kif, m, AF_INET6, action, reason,
8958 					    ri->r, a, ruleset, &pd, 0);
8959 		}
8960 	}
8961 
8962 	pf_counter_u64_critical_enter();
8963 	pf_counter_u64_add_protected(&kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS],
8964 	    pd.tot_len);
8965 	pf_counter_u64_add_protected(&kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS],
8966 	    1);
8967 
8968 	if (action == PF_PASS || r->action == PF_DROP) {
8969 		dirndx = (dir == PF_OUT);
8970 		pf_counter_u64_add_protected(&r->packets[dirndx], 1);
8971 		pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len);
8972 		if (a != NULL) {
8973 			pf_counter_u64_add_protected(&a->packets[dirndx], 1);
8974 			pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len);
8975 		}
8976 		if (s != NULL) {
8977 			if (s->nat_rule.ptr != NULL) {
8978 				pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx],
8979 				    1);
8980 				pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx],
8981 				    pd.tot_len);
8982 			}
8983 			if (s->src_node != NULL) {
8984 				counter_u64_add(s->src_node->packets[dirndx],
8985 				    1);
8986 				counter_u64_add(s->src_node->bytes[dirndx],
8987 				    pd.tot_len);
8988 			}
8989 			if (s->nat_src_node != NULL) {
8990 				counter_u64_add(s->nat_src_node->packets[dirndx],
8991 				    1);
8992 				counter_u64_add(s->nat_src_node->bytes[dirndx],
8993 				    pd.tot_len);
8994 			}
8995 			dirndx = (dir == s->direction) ? 0 : 1;
8996 			s->packets[dirndx]++;
8997 			s->bytes[dirndx] += pd.tot_len;
8998 		}
8999 		tr = r;
9000 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
9001 		if (nr != NULL && r == &V_pf_default_rule)
9002 			tr = nr;
9003 		if (tr->src.addr.type == PF_ADDR_TABLE)
9004 			pfr_update_stats(tr->src.addr.p.tbl,
9005 			    (s == NULL) ? pd.src :
9006 			    &s->key[(s->direction == PF_IN)]->addr[0],
9007 			    pd.af, pd.tot_len, dir == PF_OUT,
9008 			    r->action == PF_PASS, tr->src.neg);
9009 		if (tr->dst.addr.type == PF_ADDR_TABLE)
9010 			pfr_update_stats(tr->dst.addr.p.tbl,
9011 			    (s == NULL) ? pd.dst :
9012 			    &s->key[(s->direction == PF_IN)]->addr[1],
9013 			    pd.af, pd.tot_len, dir == PF_OUT,
9014 			    r->action == PF_PASS, tr->dst.neg);
9015 	}
9016 	pf_counter_u64_critical_exit();
9017 
9018 	switch (action) {
9019 	case PF_SYNPROXY_DROP:
9020 		m_freem(*m0);
9021 	case PF_DEFER:
9022 		*m0 = NULL;
9023 		action = PF_PASS;
9024 		break;
9025 	case PF_DROP:
9026 		m_freem(*m0);
9027 		*m0 = NULL;
9028 		break;
9029 	default:
9030 		/* pf_route6() returns unlocked. */
9031 		if (rt) {
9032 			pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp);
9033 			goto out;
9034 		}
9035 		if (pf_dummynet(&pd, s, r, m0) != 0) {
9036 			action = PF_DROP;
9037 			REASON_SET(&reason, PFRES_MEMORY);
9038 		}
9039 		break;
9040 	}
9041 
9042 	if (s && action != PF_DROP) {
9043 		if (!s->if_index_in && dir == PF_IN)
9044 			s->if_index_in = ifp->if_index;
9045 		else if (!s->if_index_out && dir == PF_OUT)
9046 			s->if_index_out = ifp->if_index;
9047 	}
9048 
9049 	if (s)
9050 		PF_STATE_UNLOCK(s);
9051 
9052 	/* If reassembled packet passed, create new fragments. */
9053 	if (action == PF_PASS && *m0 && dir == PF_OUT &&
9054 	    (mtag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL)
9055 		action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD);
9056 
9057 out:
9058 	SDT_PROBE4(pf, ip, test6, done, action, reason, r, s);
9059 
9060 	pf_sctp_multihome_delayed(&pd, off, kif, s, action);
9061 
9062 	return (action);
9063 }
9064 #endif /* INET6 */
9065