xref: /freebsd/sys/netpfil/pf/pf.c (revision 8aac90f18aef7c9eea906c3ff9a001ca7b94f375)
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_DEFINE2(pf, ip, , bound_iface, "struct pf_kstate *",
130     "struct pfi_kkif *");
131 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *",
132     "struct pf_krule *", "struct mbuf *", "int");
133 
134 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *",
135     "struct mbuf *");
136 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *");
137 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch,
138     "int", "struct pf_keth_rule *", "char *");
139 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *");
140 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match,
141     "int", "struct pf_keth_rule *");
142 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t");
143 
144 /*
145  * Global variables
146  */
147 
148 /* state tables */
149 VNET_DEFINE(struct pf_altqqueue,	 pf_altqs[4]);
150 VNET_DEFINE(struct pf_kpalist,		 pf_pabuf);
151 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_active);
152 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_active);
153 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_inactive);
154 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_inactive);
155 VNET_DEFINE(struct pf_kstatus,		 pf_status);
156 
157 VNET_DEFINE(u_int32_t,			 ticket_altqs_active);
158 VNET_DEFINE(u_int32_t,			 ticket_altqs_inactive);
159 VNET_DEFINE(int,			 altqs_inactive_open);
160 VNET_DEFINE(u_int32_t,			 ticket_pabuf);
161 
162 VNET_DEFINE(MD5_CTX,			 pf_tcp_secret_ctx);
163 #define	V_pf_tcp_secret_ctx		 VNET(pf_tcp_secret_ctx)
164 VNET_DEFINE(u_char,			 pf_tcp_secret[16]);
165 #define	V_pf_tcp_secret			 VNET(pf_tcp_secret)
166 VNET_DEFINE(int,			 pf_tcp_secret_init);
167 #define	V_pf_tcp_secret_init		 VNET(pf_tcp_secret_init)
168 VNET_DEFINE(int,			 pf_tcp_iss_off);
169 #define	V_pf_tcp_iss_off		 VNET(pf_tcp_iss_off)
170 VNET_DECLARE(int,			 pf_vnet_active);
171 #define	V_pf_vnet_active		 VNET(pf_vnet_active)
172 
173 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx);
174 #define V_pf_purge_idx	VNET(pf_purge_idx)
175 
176 #ifdef PF_WANT_32_TO_64_COUNTER
177 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter);
178 #define	V_pf_counter_periodic_iter	VNET(pf_counter_periodic_iter)
179 
180 VNET_DEFINE(struct allrulelist_head, pf_allrulelist);
181 VNET_DEFINE(size_t, pf_allrulecount);
182 VNET_DEFINE(struct pf_krule *, pf_rulemarker);
183 #endif
184 
185 struct pf_sctp_endpoint;
186 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint);
187 struct pf_sctp_source {
188 	sa_family_t			af;
189 	struct pf_addr			addr;
190 	TAILQ_ENTRY(pf_sctp_source)	entry;
191 };
192 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source);
193 struct pf_sctp_endpoint
194 {
195 	uint32_t		 v_tag;
196 	struct pf_sctp_sources	 sources;
197 	RB_ENTRY(pf_sctp_endpoint)	entry;
198 };
199 static int
200 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b)
201 {
202 	return (a->v_tag - b->v_tag);
203 }
204 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare);
205 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare);
206 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints);
207 #define V_pf_sctp_endpoints	VNET(pf_sctp_endpoints)
208 static struct mtx_padalign pf_sctp_endpoints_mtx;
209 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF);
210 #define	PF_SCTP_ENDPOINTS_LOCK()	mtx_lock(&pf_sctp_endpoints_mtx)
211 #define	PF_SCTP_ENDPOINTS_UNLOCK()	mtx_unlock(&pf_sctp_endpoints_mtx)
212 
213 /*
214  * Queue for pf_intr() sends.
215  */
216 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
217 struct pf_send_entry {
218 	STAILQ_ENTRY(pf_send_entry)	pfse_next;
219 	struct mbuf			*pfse_m;
220 	enum {
221 		PFSE_IP,
222 		PFSE_IP6,
223 		PFSE_ICMP,
224 		PFSE_ICMP6,
225 	}				pfse_type;
226 	struct {
227 		int		type;
228 		int		code;
229 		int		mtu;
230 	} icmpopts;
231 };
232 
233 STAILQ_HEAD(pf_send_head, pf_send_entry);
234 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue);
235 #define	V_pf_sendqueue	VNET(pf_sendqueue)
236 
237 static struct mtx_padalign pf_sendqueue_mtx;
238 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
239 #define	PF_SENDQ_LOCK()		mtx_lock(&pf_sendqueue_mtx)
240 #define	PF_SENDQ_UNLOCK()	mtx_unlock(&pf_sendqueue_mtx)
241 
242 /*
243  * Queue for pf_overload_task() tasks.
244  */
245 struct pf_overload_entry {
246 	SLIST_ENTRY(pf_overload_entry)	next;
247 	struct pf_addr  		addr;
248 	sa_family_t			af;
249 	uint8_t				dir;
250 	struct pf_krule  		*rule;
251 };
252 
253 SLIST_HEAD(pf_overload_head, pf_overload_entry);
254 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue);
255 #define V_pf_overloadqueue	VNET(pf_overloadqueue)
256 VNET_DEFINE_STATIC(struct task, pf_overloadtask);
257 #define	V_pf_overloadtask	VNET(pf_overloadtask)
258 
259 static struct mtx_padalign pf_overloadqueue_mtx;
260 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
261     "pf overload/flush queue", MTX_DEF);
262 #define	PF_OVERLOADQ_LOCK()	mtx_lock(&pf_overloadqueue_mtx)
263 #define	PF_OVERLOADQ_UNLOCK()	mtx_unlock(&pf_overloadqueue_mtx)
264 
265 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules);
266 struct mtx_padalign pf_unlnkdrules_mtx;
267 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
268     MTX_DEF);
269 
270 struct sx pf_config_lock;
271 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config");
272 
273 struct mtx_padalign pf_table_stats_lock;
274 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats",
275     MTX_DEF);
276 
277 VNET_DEFINE_STATIC(uma_zone_t,	pf_sources_z);
278 #define	V_pf_sources_z	VNET(pf_sources_z)
279 uma_zone_t		pf_mtag_z;
280 VNET_DEFINE(uma_zone_t,	 pf_state_z);
281 VNET_DEFINE(uma_zone_t,	 pf_state_key_z);
282 
283 VNET_DEFINE(struct unrhdr64, pf_stateid);
284 
285 static void		 pf_src_tree_remove_state(struct pf_kstate *);
286 static void		 pf_init_threshold(struct pf_threshold *, u_int32_t,
287 			    u_int32_t);
288 static void		 pf_add_threshold(struct pf_threshold *);
289 static int		 pf_check_threshold(struct pf_threshold *);
290 
291 static void		 pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
292 			    u_int16_t *, u_int16_t *, struct pf_addr *,
293 			    u_int16_t, u_int8_t, sa_family_t);
294 static int		 pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
295 			    struct tcphdr *, struct pf_state_peer *);
296 static void		 pf_change_icmp(struct pf_addr *, u_int16_t *,
297 			    struct pf_addr *, struct pf_addr *, u_int16_t,
298 			    u_int16_t *, u_int16_t *, u_int16_t *,
299 			    u_int16_t *, u_int8_t, sa_family_t);
300 static void		 pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
301 			    sa_family_t, struct pf_krule *, int);
302 static void		 pf_detach_state(struct pf_kstate *);
303 static int		 pf_state_key_attach(struct pf_state_key *,
304 			    struct pf_state_key *, struct pf_kstate *);
305 static void		 pf_state_key_detach(struct pf_kstate *, int);
306 static int		 pf_state_key_ctor(void *, int, void *, int);
307 static u_int32_t	 pf_tcp_iss(struct pf_pdesc *);
308 static __inline void	 pf_dummynet_flag_remove(struct mbuf *m,
309 			    struct pf_mtag *pf_mtag);
310 static int		 pf_dummynet(struct pf_pdesc *, struct pf_kstate *,
311 			    struct pf_krule *, struct mbuf **);
312 static int		 pf_dummynet_route(struct pf_pdesc *,
313 			    struct pf_kstate *, struct pf_krule *,
314 			    struct ifnet *, struct sockaddr *, struct mbuf **);
315 static int		 pf_test_eth_rule(int, struct pfi_kkif *,
316 			    struct mbuf **);
317 static int		 pf_test_rule(struct pf_krule **, struct pf_kstate **,
318 			    struct pfi_kkif *, struct mbuf *, int,
319 			    struct pf_pdesc *, struct pf_krule **,
320 			    struct pf_kruleset **, struct inpcb *);
321 static int		 pf_create_state(struct pf_krule *, struct pf_krule *,
322 			    struct pf_krule *, struct pf_pdesc *,
323 			    struct pf_ksrc_node *, struct pf_state_key *,
324 			    struct pf_state_key *, struct mbuf *, int,
325 			    u_int16_t, u_int16_t, int *, struct pfi_kkif *,
326 			    struct pf_kstate **, int, u_int16_t, u_int16_t,
327 			    int, struct pf_krule_slist *);
328 static int		 pf_test_fragment(struct pf_krule **, struct pfi_kkif *,
329 			    struct mbuf *, void *, struct pf_pdesc *,
330 			    struct pf_krule **, struct pf_kruleset **);
331 static int		 pf_tcp_track_full(struct pf_kstate **,
332 			    struct pfi_kkif *, struct mbuf *, int,
333 			    struct pf_pdesc *, u_short *, int *);
334 static int		 pf_tcp_track_sloppy(struct pf_kstate **,
335 			    struct pf_pdesc *, u_short *);
336 static int		 pf_test_state_tcp(struct pf_kstate **,
337 			    struct pfi_kkif *, struct mbuf *, int,
338 			    void *, struct pf_pdesc *, u_short *);
339 static int		 pf_test_state_udp(struct pf_kstate **,
340 			    struct pfi_kkif *, struct mbuf *, int,
341 			    void *, struct pf_pdesc *);
342 static int		 pf_test_state_icmp(struct pf_kstate **,
343 			    struct pfi_kkif *, struct mbuf *, int,
344 			    void *, struct pf_pdesc *, u_short *);
345 static void		 pf_sctp_multihome_detach_addr(const struct pf_kstate *);
346 static void		 pf_sctp_multihome_delayed(struct pf_pdesc *, int,
347 			    struct pfi_kkif *, struct pf_kstate *, int);
348 static int		 pf_test_state_sctp(struct pf_kstate **,
349 			    struct pfi_kkif *, struct mbuf *, int,
350 			    void *, struct pf_pdesc *, u_short *);
351 static int		 pf_test_state_other(struct pf_kstate **,
352 			    struct pfi_kkif *, struct mbuf *, struct pf_pdesc *);
353 static u_int16_t	 pf_calc_mss(struct pf_addr *, sa_family_t,
354 				int, u_int16_t);
355 static int		 pf_check_proto_cksum(struct mbuf *, int, int,
356 			    u_int8_t, sa_family_t);
357 static void		 pf_print_state_parts(struct pf_kstate *,
358 			    struct pf_state_key *, struct pf_state_key *);
359 static void		 pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t,
360 			    bool, u_int8_t);
361 static struct pf_kstate	*pf_find_state(struct pfi_kkif *,
362 			    struct pf_state_key_cmp *, u_int);
363 static int		 pf_src_connlimit(struct pf_kstate **);
364 static void		 pf_overload_task(void *v, int pending);
365 static u_short		 pf_insert_src_node(struct pf_ksrc_node **,
366 			    struct pf_krule *, struct pf_addr *, sa_family_t);
367 static u_int		 pf_purge_expired_states(u_int, int);
368 static void		 pf_purge_unlinked_rules(void);
369 static int		 pf_mtag_uminit(void *, int, int);
370 static void		 pf_mtag_free(struct m_tag *);
371 static void		 pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *,
372 			    int, struct pf_state_key *);
373 #ifdef INET
374 static void		 pf_route(struct mbuf **, struct pf_krule *,
375 			    struct ifnet *, struct pf_kstate *,
376 			    struct pf_pdesc *, struct inpcb *);
377 #endif /* INET */
378 #ifdef INET6
379 static void		 pf_change_a6(struct pf_addr *, u_int16_t *,
380 			    struct pf_addr *, u_int8_t);
381 static void		 pf_route6(struct mbuf **, struct pf_krule *,
382 			    struct ifnet *, struct pf_kstate *,
383 			    struct pf_pdesc *, struct inpcb *);
384 #endif /* INET6 */
385 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t);
386 
387 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
388 
389 extern int pf_end_threads;
390 extern struct proc *pf_purge_proc;
391 
392 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
393 
394 #define	PACKET_UNDO_NAT(_m, _pd, _off, _s)		\
395 	do {								\
396 		struct pf_state_key *nk;				\
397 		if ((pd->dir) == PF_OUT)					\
398 			nk = (_s)->key[PF_SK_STACK];			\
399 		else							\
400 			nk = (_s)->key[PF_SK_WIRE];			\
401 		pf_packet_rework_nat(_m, _pd, _off, nk);		\
402 	} while (0)
403 
404 #define	PACKET_LOOPED(pd)	((pd)->pf_mtag &&			\
405 				 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED)
406 
407 #define	STATE_LOOKUP(i, k, s, pd)					\
408 	do {								\
409 		(s) = pf_find_state((i), (k), (pd->dir));			\
410 		SDT_PROBE5(pf, ip, state, lookup, i, k, (pd->dir), pd, (s));	\
411 		if ((s) == NULL)					\
412 			return (PF_DROP);				\
413 		if (PACKET_LOOPED(pd))					\
414 			return (PF_PASS);				\
415 	} while (0)
416 
417 static struct pfi_kkif *
418 BOUND_IFACE(struct pf_kstate *st, struct pfi_kkif *k)
419 {
420 	SDT_PROBE2(pf, ip, , bound_iface, st, k);
421 
422 	/* Floating unless otherwise specified. */
423 	if (! (st->rule.ptr->rule_flag & PFRULE_IFBOUND))
424 		return (V_pfi_all);
425 
426 	/*
427 	 * Initially set to all, because we don't know what interface we'll be
428 	 * sending this out when we create the state.
429 	 */
430 	if (st->rule.ptr->rt == PF_REPLYTO)
431 		return (V_pfi_all);
432 
433 	/* Don't overrule the interface for states created on incoming packets. */
434 	if (st->direction == PF_IN)
435 		return (k);
436 
437 	/* No route-to, so don't overrule. */
438 	if (st->rt != PF_ROUTETO)
439 		return (k);
440 
441 	/* Bind to the route-to interface. */
442 	return (st->rt_kif);
443 }
444 
445 #define	STATE_INC_COUNTERS(s)						\
446 	do {								\
447 		struct pf_krule_item *mrm;				\
448 		counter_u64_add(s->rule.ptr->states_cur, 1);		\
449 		counter_u64_add(s->rule.ptr->states_tot, 1);		\
450 		if (s->anchor.ptr != NULL) {				\
451 			counter_u64_add(s->anchor.ptr->states_cur, 1);	\
452 			counter_u64_add(s->anchor.ptr->states_tot, 1);	\
453 		}							\
454 		if (s->nat_rule.ptr != NULL) {				\
455 			counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
456 			counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
457 		}							\
458 		SLIST_FOREACH(mrm, &s->match_rules, entry) {		\
459 			counter_u64_add(mrm->r->states_cur, 1);		\
460 			counter_u64_add(mrm->r->states_tot, 1);		\
461 		}							\
462 	} while (0)
463 
464 #define	STATE_DEC_COUNTERS(s)						\
465 	do {								\
466 		struct pf_krule_item *mrm;				\
467 		if (s->nat_rule.ptr != NULL)				\
468 			counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
469 		if (s->anchor.ptr != NULL)				\
470 			counter_u64_add(s->anchor.ptr->states_cur, -1);	\
471 		counter_u64_add(s->rule.ptr->states_cur, -1);		\
472 		SLIST_FOREACH(mrm, &s->match_rules, entry)		\
473 			counter_u64_add(mrm->r->states_cur, -1);	\
474 	} while (0)
475 
476 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
477 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items");
478 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
479 VNET_DEFINE(struct pf_idhash *, pf_idhash);
480 VNET_DEFINE(struct pf_srchash *, pf_srchash);
481 
482 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
483     "pf(4)");
484 
485 u_long	pf_hashmask;
486 u_long	pf_srchashmask;
487 static u_long	pf_hashsize;
488 static u_long	pf_srchashsize;
489 u_long	pf_ioctl_maxcount = 65535;
490 
491 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
492     &pf_hashsize, 0, "Size of pf(4) states hashtable");
493 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
494     &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
495 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN,
496     &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call");
497 
498 VNET_DEFINE(void *, pf_swi_cookie);
499 VNET_DEFINE(struct intr_event *, pf_swi_ie);
500 
501 VNET_DEFINE(uint32_t, pf_hashseed);
502 #define	V_pf_hashseed	VNET(pf_hashseed)
503 
504 static void
505 pf_sctp_checksum(struct mbuf *m, int off)
506 {
507 	uint32_t sum = 0;
508 
509 	/* Zero out the checksum, to enable recalculation. */
510 	m_copyback(m, off + offsetof(struct sctphdr, checksum),
511 	    sizeof(sum), (caddr_t)&sum);
512 
513 	sum = sctp_calculate_cksum(m, off);
514 
515 	m_copyback(m, off + offsetof(struct sctphdr, checksum),
516 	    sizeof(sum), (caddr_t)&sum);
517 }
518 
519 int
520 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
521 {
522 
523 	switch (af) {
524 #ifdef INET
525 	case AF_INET:
526 		if (a->addr32[0] > b->addr32[0])
527 			return (1);
528 		if (a->addr32[0] < b->addr32[0])
529 			return (-1);
530 		break;
531 #endif /* INET */
532 #ifdef INET6
533 	case AF_INET6:
534 		if (a->addr32[3] > b->addr32[3])
535 			return (1);
536 		if (a->addr32[3] < b->addr32[3])
537 			return (-1);
538 		if (a->addr32[2] > b->addr32[2])
539 			return (1);
540 		if (a->addr32[2] < b->addr32[2])
541 			return (-1);
542 		if (a->addr32[1] > b->addr32[1])
543 			return (1);
544 		if (a->addr32[1] < b->addr32[1])
545 			return (-1);
546 		if (a->addr32[0] > b->addr32[0])
547 			return (1);
548 		if (a->addr32[0] < b->addr32[0])
549 			return (-1);
550 		break;
551 #endif /* INET6 */
552 	default:
553 		panic("%s: unknown address family %u", __func__, af);
554 	}
555 	return (0);
556 }
557 
558 static void
559 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off,
560 	struct pf_state_key *nk)
561 {
562 
563 	switch (pd->proto) {
564 	case IPPROTO_TCP: {
565 		struct tcphdr *th = &pd->hdr.tcp;
566 
567 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af))
568 			pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum,
569 			    &th->th_sum, &nk->addr[pd->sidx],
570 			    nk->port[pd->sidx], 0, pd->af);
571 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af))
572 			pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum,
573 			    &th->th_sum, &nk->addr[pd->didx],
574 			    nk->port[pd->didx], 0, pd->af);
575 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
576 		break;
577 	}
578 	case IPPROTO_UDP: {
579 		struct udphdr *uh = &pd->hdr.udp;
580 
581 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af))
582 			pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
583 			    &uh->uh_sum, &nk->addr[pd->sidx],
584 			    nk->port[pd->sidx], 1, pd->af);
585 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af))
586 			pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
587 			    &uh->uh_sum, &nk->addr[pd->didx],
588 			    nk->port[pd->didx], 1, pd->af);
589 		m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
590 		break;
591 	}
592 	case IPPROTO_SCTP: {
593 		struct sctphdr *sh = &pd->hdr.sctp;
594 		uint16_t checksum = 0;
595 
596 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) {
597 			pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum,
598 			    &checksum, &nk->addr[pd->sidx],
599 			    nk->port[pd->sidx], 1, pd->af);
600 		}
601 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) {
602 			pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum,
603 			    &checksum, &nk->addr[pd->didx],
604 			    nk->port[pd->didx], 1, pd->af);
605 		}
606 
607 		break;
608 	}
609 	case IPPROTO_ICMP: {
610 		struct icmp *ih = &pd->hdr.icmp;
611 
612 		if (nk->port[pd->sidx] != ih->icmp_id) {
613 			pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
614 			    ih->icmp_cksum, ih->icmp_id,
615 			    nk->port[pd->sidx], 0);
616 			ih->icmp_id = nk->port[pd->sidx];
617 			pd->sport = &ih->icmp_id;
618 
619 			m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih);
620 		}
621 		/* FALLTHROUGH */
622 	}
623 	default:
624 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) {
625 			switch (pd->af) {
626 			case AF_INET:
627 				pf_change_a(&pd->src->v4.s_addr,
628 				    pd->ip_sum, nk->addr[pd->sidx].v4.s_addr,
629 				    0);
630 				break;
631 			case AF_INET6:
632 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
633 				break;
634 			}
635 		}
636 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) {
637 			switch (pd->af) {
638 			case AF_INET:
639 				pf_change_a(&pd->dst->v4.s_addr,
640 				    pd->ip_sum, nk->addr[pd->didx].v4.s_addr,
641 				    0);
642 				break;
643 			case AF_INET6:
644 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
645 				break;
646 			}
647 		}
648 		break;
649 	}
650 }
651 
652 static __inline uint32_t
653 pf_hashkey(struct pf_state_key *sk)
654 {
655 	uint32_t h;
656 
657 	h = murmur3_32_hash32((uint32_t *)sk,
658 	    sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
659 	    V_pf_hashseed);
660 
661 	return (h & pf_hashmask);
662 }
663 
664 static __inline uint32_t
665 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
666 {
667 	uint32_t h;
668 
669 	switch (af) {
670 	case AF_INET:
671 		h = murmur3_32_hash32((uint32_t *)&addr->v4,
672 		    sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
673 		break;
674 	case AF_INET6:
675 		h = murmur3_32_hash32((uint32_t *)&addr->v6,
676 		    sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
677 		break;
678 	default:
679 		panic("%s: unknown address family %u", __func__, af);
680 	}
681 
682 	return (h & pf_srchashmask);
683 }
684 
685 #ifdef ALTQ
686 static int
687 pf_state_hash(struct pf_kstate *s)
688 {
689 	u_int32_t hv = (intptr_t)s / sizeof(*s);
690 
691 	hv ^= crc32(&s->src, sizeof(s->src));
692 	hv ^= crc32(&s->dst, sizeof(s->dst));
693 	if (hv == 0)
694 		hv = 1;
695 	return (hv);
696 }
697 #endif
698 
699 static __inline void
700 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate)
701 {
702 	if (which == PF_PEER_DST || which == PF_PEER_BOTH)
703 		s->dst.state = newstate;
704 	if (which == PF_PEER_DST)
705 		return;
706 	if (s->src.state == newstate)
707 		return;
708 	if (s->creatorid == V_pf_status.hostid &&
709 	    s->key[PF_SK_STACK] != NULL &&
710 	    s->key[PF_SK_STACK]->proto == IPPROTO_TCP &&
711 	    !(TCPS_HAVEESTABLISHED(s->src.state) ||
712 	    s->src.state == TCPS_CLOSED) &&
713 	    (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED))
714 		atomic_add_32(&V_pf_status.states_halfopen, -1);
715 
716 	s->src.state = newstate;
717 }
718 
719 #ifdef INET6
720 void
721 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
722 {
723 	switch (af) {
724 #ifdef INET
725 	case AF_INET:
726 		dst->addr32[0] = src->addr32[0];
727 		break;
728 #endif /* INET */
729 	case AF_INET6:
730 		dst->addr32[0] = src->addr32[0];
731 		dst->addr32[1] = src->addr32[1];
732 		dst->addr32[2] = src->addr32[2];
733 		dst->addr32[3] = src->addr32[3];
734 		break;
735 	}
736 }
737 #endif /* INET6 */
738 
739 static void
740 pf_init_threshold(struct pf_threshold *threshold,
741     u_int32_t limit, u_int32_t seconds)
742 {
743 	threshold->limit = limit * PF_THRESHOLD_MULT;
744 	threshold->seconds = seconds;
745 	threshold->count = 0;
746 	threshold->last = time_uptime;
747 }
748 
749 static void
750 pf_add_threshold(struct pf_threshold *threshold)
751 {
752 	u_int32_t t = time_uptime, diff = t - threshold->last;
753 
754 	if (diff >= threshold->seconds)
755 		threshold->count = 0;
756 	else
757 		threshold->count -= threshold->count * diff /
758 		    threshold->seconds;
759 	threshold->count += PF_THRESHOLD_MULT;
760 	threshold->last = t;
761 }
762 
763 static int
764 pf_check_threshold(struct pf_threshold *threshold)
765 {
766 	return (threshold->count > threshold->limit);
767 }
768 
769 static int
770 pf_src_connlimit(struct pf_kstate **state)
771 {
772 	struct pf_overload_entry *pfoe;
773 	int bad = 0;
774 
775 	PF_STATE_LOCK_ASSERT(*state);
776 	/*
777 	 * XXXKS: The src node is accessed unlocked!
778 	 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node);
779 	 */
780 
781 	(*state)->src_node->conn++;
782 	(*state)->src.tcp_est = 1;
783 	pf_add_threshold(&(*state)->src_node->conn_rate);
784 
785 	if ((*state)->rule.ptr->max_src_conn &&
786 	    (*state)->rule.ptr->max_src_conn <
787 	    (*state)->src_node->conn) {
788 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
789 		bad++;
790 	}
791 
792 	if ((*state)->rule.ptr->max_src_conn_rate.limit &&
793 	    pf_check_threshold(&(*state)->src_node->conn_rate)) {
794 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
795 		bad++;
796 	}
797 
798 	if (!bad)
799 		return (0);
800 
801 	/* Kill this state. */
802 	(*state)->timeout = PFTM_PURGE;
803 	pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED);
804 
805 	if ((*state)->rule.ptr->overload_tbl == NULL)
806 		return (1);
807 
808 	/* Schedule overloading and flushing task. */
809 	pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
810 	if (pfoe == NULL)
811 		return (1);	/* too bad :( */
812 
813 	bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
814 	pfoe->af = (*state)->key[PF_SK_WIRE]->af;
815 	pfoe->rule = (*state)->rule.ptr;
816 	pfoe->dir = (*state)->direction;
817 	PF_OVERLOADQ_LOCK();
818 	SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
819 	PF_OVERLOADQ_UNLOCK();
820 	taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
821 
822 	return (1);
823 }
824 
825 static void
826 pf_overload_task(void *v, int pending)
827 {
828 	struct pf_overload_head queue;
829 	struct pfr_addr p;
830 	struct pf_overload_entry *pfoe, *pfoe1;
831 	uint32_t killed = 0;
832 
833 	CURVNET_SET((struct vnet *)v);
834 
835 	PF_OVERLOADQ_LOCK();
836 	queue = V_pf_overloadqueue;
837 	SLIST_INIT(&V_pf_overloadqueue);
838 	PF_OVERLOADQ_UNLOCK();
839 
840 	bzero(&p, sizeof(p));
841 	SLIST_FOREACH(pfoe, &queue, next) {
842 		counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
843 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
844 			printf("%s: blocking address ", __func__);
845 			pf_print_host(&pfoe->addr, 0, pfoe->af);
846 			printf("\n");
847 		}
848 
849 		p.pfra_af = pfoe->af;
850 		switch (pfoe->af) {
851 #ifdef INET
852 		case AF_INET:
853 			p.pfra_net = 32;
854 			p.pfra_ip4addr = pfoe->addr.v4;
855 			break;
856 #endif
857 #ifdef INET6
858 		case AF_INET6:
859 			p.pfra_net = 128;
860 			p.pfra_ip6addr = pfoe->addr.v6;
861 			break;
862 #endif
863 		}
864 
865 		PF_RULES_WLOCK();
866 		pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
867 		PF_RULES_WUNLOCK();
868 	}
869 
870 	/*
871 	 * Remove those entries, that don't need flushing.
872 	 */
873 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
874 		if (pfoe->rule->flush == 0) {
875 			SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
876 			free(pfoe, M_PFTEMP);
877 		} else
878 			counter_u64_add(
879 			    V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
880 
881 	/* If nothing to flush, return. */
882 	if (SLIST_EMPTY(&queue)) {
883 		CURVNET_RESTORE();
884 		return;
885 	}
886 
887 	for (int i = 0; i <= pf_hashmask; i++) {
888 		struct pf_idhash *ih = &V_pf_idhash[i];
889 		struct pf_state_key *sk;
890 		struct pf_kstate *s;
891 
892 		PF_HASHROW_LOCK(ih);
893 		LIST_FOREACH(s, &ih->states, entry) {
894 		    sk = s->key[PF_SK_WIRE];
895 		    SLIST_FOREACH(pfoe, &queue, next)
896 			if (sk->af == pfoe->af &&
897 			    ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
898 			    pfoe->rule == s->rule.ptr) &&
899 			    ((pfoe->dir == PF_OUT &&
900 			    PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
901 			    (pfoe->dir == PF_IN &&
902 			    PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
903 				s->timeout = PFTM_PURGE;
904 				pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED);
905 				killed++;
906 			}
907 		}
908 		PF_HASHROW_UNLOCK(ih);
909 	}
910 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
911 		free(pfoe, M_PFTEMP);
912 	if (V_pf_status.debug >= PF_DEBUG_MISC)
913 		printf("%s: %u states killed", __func__, killed);
914 
915 	CURVNET_RESTORE();
916 }
917 
918 /*
919  * Can return locked on failure, so that we can consistently
920  * allocate and insert a new one.
921  */
922 struct pf_ksrc_node *
923 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af,
924 	struct pf_srchash **sh, bool returnlocked)
925 {
926 	struct pf_ksrc_node *n;
927 
928 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
929 
930 	*sh = &V_pf_srchash[pf_hashsrc(src, af)];
931 	PF_HASHROW_LOCK(*sh);
932 	LIST_FOREACH(n, &(*sh)->nodes, entry)
933 		if (n->rule.ptr == rule && n->af == af &&
934 		    ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
935 		    (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
936 			break;
937 
938 	if (n != NULL) {
939 		n->states++;
940 		PF_HASHROW_UNLOCK(*sh);
941 	} else if (returnlocked == false)
942 		PF_HASHROW_UNLOCK(*sh);
943 
944 	return (n);
945 }
946 
947 static void
948 pf_free_src_node(struct pf_ksrc_node *sn)
949 {
950 
951 	for (int i = 0; i < 2; i++) {
952 		counter_u64_free(sn->bytes[i]);
953 		counter_u64_free(sn->packets[i]);
954 	}
955 	uma_zfree(V_pf_sources_z, sn);
956 }
957 
958 static u_short
959 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule,
960     struct pf_addr *src, sa_family_t af)
961 {
962 	u_short			 reason = 0;
963 	struct pf_srchash	*sh = NULL;
964 
965 	KASSERT((rule->rule_flag & PFRULE_SRCTRACK ||
966 	    rule->rpool.opts & PF_POOL_STICKYADDR),
967 	    ("%s for non-tracking rule %p", __func__, rule));
968 
969 	if (*sn == NULL)
970 		*sn = pf_find_src_node(src, rule, af, &sh, true);
971 
972 	if (*sn == NULL) {
973 		PF_HASHROW_ASSERT(sh);
974 
975 		if (rule->max_src_nodes &&
976 		    counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) {
977 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1);
978 			PF_HASHROW_UNLOCK(sh);
979 			reason = PFRES_SRCLIMIT;
980 			goto done;
981 		}
982 
983 		(*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
984 		if ((*sn) == NULL) {
985 			PF_HASHROW_UNLOCK(sh);
986 			reason = PFRES_MEMORY;
987 			goto done;
988 		}
989 
990 		for (int i = 0; i < 2; i++) {
991 			(*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT);
992 			(*sn)->packets[i] = counter_u64_alloc(M_NOWAIT);
993 
994 			if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) {
995 				pf_free_src_node(*sn);
996 				PF_HASHROW_UNLOCK(sh);
997 				reason = PFRES_MEMORY;
998 				goto done;
999 			}
1000 		}
1001 
1002 		pf_init_threshold(&(*sn)->conn_rate,
1003 		    rule->max_src_conn_rate.limit,
1004 		    rule->max_src_conn_rate.seconds);
1005 
1006 		MPASS((*sn)->lock == NULL);
1007 		(*sn)->lock = &sh->lock;
1008 
1009 		(*sn)->af = af;
1010 		(*sn)->rule.ptr = rule;
1011 		PF_ACPY(&(*sn)->addr, src, af);
1012 		LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
1013 		(*sn)->creation = time_uptime;
1014 		(*sn)->ruletype = rule->action;
1015 		(*sn)->states = 1;
1016 		if ((*sn)->rule.ptr != NULL)
1017 			counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
1018 		PF_HASHROW_UNLOCK(sh);
1019 		counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
1020 	} else {
1021 		if (rule->max_src_states &&
1022 		    (*sn)->states >= rule->max_src_states) {
1023 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
1024 			    1);
1025 			reason = PFRES_SRCLIMIT;
1026 			goto done;
1027 		}
1028 	}
1029 done:
1030 	return (reason);
1031 }
1032 
1033 void
1034 pf_unlink_src_node(struct pf_ksrc_node *src)
1035 {
1036 	PF_SRC_NODE_LOCK_ASSERT(src);
1037 
1038 	LIST_REMOVE(src, entry);
1039 	if (src->rule.ptr)
1040 		counter_u64_add(src->rule.ptr->src_nodes, -1);
1041 }
1042 
1043 u_int
1044 pf_free_src_nodes(struct pf_ksrc_node_list *head)
1045 {
1046 	struct pf_ksrc_node *sn, *tmp;
1047 	u_int count = 0;
1048 
1049 	LIST_FOREACH_SAFE(sn, head, entry, tmp) {
1050 		pf_free_src_node(sn);
1051 		count++;
1052 	}
1053 
1054 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
1055 
1056 	return (count);
1057 }
1058 
1059 void
1060 pf_mtag_initialize(void)
1061 {
1062 
1063 	pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
1064 	    sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
1065 	    UMA_ALIGN_PTR, 0);
1066 }
1067 
1068 /* Per-vnet data storage structures initialization. */
1069 void
1070 pf_initialize(void)
1071 {
1072 	struct pf_keyhash	*kh;
1073 	struct pf_idhash	*ih;
1074 	struct pf_srchash	*sh;
1075 	u_int i;
1076 
1077 	if (pf_hashsize == 0 || !powerof2(pf_hashsize))
1078 		pf_hashsize = PF_HASHSIZ;
1079 	if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
1080 		pf_srchashsize = PF_SRCHASHSIZ;
1081 
1082 	V_pf_hashseed = arc4random();
1083 
1084 	/* States and state keys storage. */
1085 	V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate),
1086 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1087 	V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
1088 	uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
1089 	uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
1090 
1091 	V_pf_state_key_z = uma_zcreate("pf state keys",
1092 	    sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
1093 	    UMA_ALIGN_PTR, 0);
1094 
1095 	V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash),
1096 	    M_PFHASH, M_NOWAIT | M_ZERO);
1097 	V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash),
1098 	    M_PFHASH, M_NOWAIT | M_ZERO);
1099 	if (V_pf_keyhash == NULL || V_pf_idhash == NULL) {
1100 		printf("pf: Unable to allocate memory for "
1101 		    "state_hashsize %lu.\n", pf_hashsize);
1102 
1103 		free(V_pf_keyhash, M_PFHASH);
1104 		free(V_pf_idhash, M_PFHASH);
1105 
1106 		pf_hashsize = PF_HASHSIZ;
1107 		V_pf_keyhash = mallocarray(pf_hashsize,
1108 		    sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO);
1109 		V_pf_idhash = mallocarray(pf_hashsize,
1110 		    sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO);
1111 	}
1112 
1113 	pf_hashmask = pf_hashsize - 1;
1114 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
1115 	    i++, kh++, ih++) {
1116 		mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
1117 		mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
1118 	}
1119 
1120 	/* Source nodes. */
1121 	V_pf_sources_z = uma_zcreate("pf source nodes",
1122 	    sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
1123 	    0);
1124 	V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
1125 	uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
1126 	uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
1127 
1128 	V_pf_srchash = mallocarray(pf_srchashsize,
1129 	    sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO);
1130 	if (V_pf_srchash == NULL) {
1131 		printf("pf: Unable to allocate memory for "
1132 		    "source_hashsize %lu.\n", pf_srchashsize);
1133 
1134 		pf_srchashsize = PF_SRCHASHSIZ;
1135 		V_pf_srchash = mallocarray(pf_srchashsize,
1136 		    sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO);
1137 	}
1138 
1139 	pf_srchashmask = pf_srchashsize - 1;
1140 	for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
1141 		mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
1142 
1143 	/* ALTQ */
1144 	TAILQ_INIT(&V_pf_altqs[0]);
1145 	TAILQ_INIT(&V_pf_altqs[1]);
1146 	TAILQ_INIT(&V_pf_altqs[2]);
1147 	TAILQ_INIT(&V_pf_altqs[3]);
1148 	TAILQ_INIT(&V_pf_pabuf);
1149 	V_pf_altqs_active = &V_pf_altqs[0];
1150 	V_pf_altq_ifs_active = &V_pf_altqs[1];
1151 	V_pf_altqs_inactive = &V_pf_altqs[2];
1152 	V_pf_altq_ifs_inactive = &V_pf_altqs[3];
1153 
1154 	/* Send & overload+flush queues. */
1155 	STAILQ_INIT(&V_pf_sendqueue);
1156 	SLIST_INIT(&V_pf_overloadqueue);
1157 	TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
1158 
1159 	/* Unlinked, but may be referenced rules. */
1160 	TAILQ_INIT(&V_pf_unlinked_rules);
1161 }
1162 
1163 void
1164 pf_mtag_cleanup(void)
1165 {
1166 
1167 	uma_zdestroy(pf_mtag_z);
1168 }
1169 
1170 void
1171 pf_cleanup(void)
1172 {
1173 	struct pf_keyhash	*kh;
1174 	struct pf_idhash	*ih;
1175 	struct pf_srchash	*sh;
1176 	struct pf_send_entry	*pfse, *next;
1177 	u_int i;
1178 
1179 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
1180 	    i++, kh++, ih++) {
1181 		KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
1182 		    __func__));
1183 		KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
1184 		    __func__));
1185 		mtx_destroy(&kh->lock);
1186 		mtx_destroy(&ih->lock);
1187 	}
1188 	free(V_pf_keyhash, M_PFHASH);
1189 	free(V_pf_idhash, M_PFHASH);
1190 
1191 	for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1192 		KASSERT(LIST_EMPTY(&sh->nodes),
1193 		    ("%s: source node hash not empty", __func__));
1194 		mtx_destroy(&sh->lock);
1195 	}
1196 	free(V_pf_srchash, M_PFHASH);
1197 
1198 	STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
1199 		m_freem(pfse->pfse_m);
1200 		free(pfse, M_PFTEMP);
1201 	}
1202 	MPASS(RB_EMPTY(&V_pf_sctp_endpoints));
1203 
1204 	uma_zdestroy(V_pf_sources_z);
1205 	uma_zdestroy(V_pf_state_z);
1206 	uma_zdestroy(V_pf_state_key_z);
1207 }
1208 
1209 static int
1210 pf_mtag_uminit(void *mem, int size, int how)
1211 {
1212 	struct m_tag *t;
1213 
1214 	t = (struct m_tag *)mem;
1215 	t->m_tag_cookie = MTAG_ABI_COMPAT;
1216 	t->m_tag_id = PACKET_TAG_PF;
1217 	t->m_tag_len = sizeof(struct pf_mtag);
1218 	t->m_tag_free = pf_mtag_free;
1219 
1220 	return (0);
1221 }
1222 
1223 static void
1224 pf_mtag_free(struct m_tag *t)
1225 {
1226 
1227 	uma_zfree(pf_mtag_z, t);
1228 }
1229 
1230 struct pf_mtag *
1231 pf_get_mtag(struct mbuf *m)
1232 {
1233 	struct m_tag *mtag;
1234 
1235 	if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
1236 		return ((struct pf_mtag *)(mtag + 1));
1237 
1238 	mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
1239 	if (mtag == NULL)
1240 		return (NULL);
1241 	bzero(mtag + 1, sizeof(struct pf_mtag));
1242 	m_tag_prepend(m, mtag);
1243 
1244 	return ((struct pf_mtag *)(mtag + 1));
1245 }
1246 
1247 static int
1248 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
1249     struct pf_kstate *s)
1250 {
1251 	struct pf_keyhash	*khs, *khw, *kh;
1252 	struct pf_state_key	*sk, *cur;
1253 	struct pf_kstate	*si, *olds = NULL;
1254 	int idx;
1255 
1256 	NET_EPOCH_ASSERT();
1257 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1258 	KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
1259 	KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
1260 
1261 	/*
1262 	 * We need to lock hash slots of both keys. To avoid deadlock
1263 	 * we always lock the slot with lower address first. Unlock order
1264 	 * isn't important.
1265 	 *
1266 	 * We also need to lock ID hash slot before dropping key
1267 	 * locks. On success we return with ID hash slot locked.
1268 	 */
1269 
1270 	if (skw == sks) {
1271 		khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
1272 		PF_HASHROW_LOCK(khs);
1273 	} else {
1274 		khs = &V_pf_keyhash[pf_hashkey(sks)];
1275 		khw = &V_pf_keyhash[pf_hashkey(skw)];
1276 		if (khs == khw) {
1277 			PF_HASHROW_LOCK(khs);
1278 		} else if (khs < khw) {
1279 			PF_HASHROW_LOCK(khs);
1280 			PF_HASHROW_LOCK(khw);
1281 		} else {
1282 			PF_HASHROW_LOCK(khw);
1283 			PF_HASHROW_LOCK(khs);
1284 		}
1285 	}
1286 
1287 #define	KEYS_UNLOCK()	do {			\
1288 	if (khs != khw) {			\
1289 		PF_HASHROW_UNLOCK(khs);		\
1290 		PF_HASHROW_UNLOCK(khw);		\
1291 	} else					\
1292 		PF_HASHROW_UNLOCK(khs);		\
1293 } while (0)
1294 
1295 	/*
1296 	 * First run: start with wire key.
1297 	 */
1298 	sk = skw;
1299 	kh = khw;
1300 	idx = PF_SK_WIRE;
1301 
1302 	MPASS(s->lock == NULL);
1303 	s->lock = &V_pf_idhash[PF_IDHASH(s)].lock;
1304 
1305 keyattach:
1306 	LIST_FOREACH(cur, &kh->keys, entry)
1307 		if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
1308 			break;
1309 
1310 	if (cur != NULL) {
1311 		/* Key exists. Check for same kif, if none, add to key. */
1312 		TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
1313 			struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
1314 
1315 			PF_HASHROW_LOCK(ih);
1316 			if (si->kif == s->kif &&
1317 			    si->direction == s->direction) {
1318 				if (sk->proto == IPPROTO_TCP &&
1319 				    si->src.state >= TCPS_FIN_WAIT_2 &&
1320 				    si->dst.state >= TCPS_FIN_WAIT_2) {
1321 					/*
1322 					 * New state matches an old >FIN_WAIT_2
1323 					 * state. We can't drop key hash locks,
1324 					 * thus we can't unlink it properly.
1325 					 *
1326 					 * As a workaround we drop it into
1327 					 * TCPS_CLOSED state, schedule purge
1328 					 * ASAP and push it into the very end
1329 					 * of the slot TAILQ, so that it won't
1330 					 * conflict with our new state.
1331 					 */
1332 					pf_set_protostate(si, PF_PEER_BOTH,
1333 					    TCPS_CLOSED);
1334 					si->timeout = PFTM_PURGE;
1335 					olds = si;
1336 				} else {
1337 					if (V_pf_status.debug >= PF_DEBUG_MISC) {
1338 						printf("pf: %s key attach "
1339 						    "failed on %s: ",
1340 						    (idx == PF_SK_WIRE) ?
1341 						    "wire" : "stack",
1342 						    s->kif->pfik_name);
1343 						pf_print_state_parts(s,
1344 						    (idx == PF_SK_WIRE) ?
1345 						    sk : NULL,
1346 						    (idx == PF_SK_STACK) ?
1347 						    sk : NULL);
1348 						printf(", existing: ");
1349 						pf_print_state_parts(si,
1350 						    (idx == PF_SK_WIRE) ?
1351 						    sk : NULL,
1352 						    (idx == PF_SK_STACK) ?
1353 						    sk : NULL);
1354 						printf("\n");
1355 					}
1356 					s->timeout = PFTM_UNLINKED;
1357 					PF_HASHROW_UNLOCK(ih);
1358 					KEYS_UNLOCK();
1359 					uma_zfree(V_pf_state_key_z, sk);
1360 					if (idx == PF_SK_STACK)
1361 						pf_detach_state(s);
1362 					return (EEXIST); /* collision! */
1363 				}
1364 			}
1365 			PF_HASHROW_UNLOCK(ih);
1366 		}
1367 		uma_zfree(V_pf_state_key_z, sk);
1368 		s->key[idx] = cur;
1369 	} else {
1370 		LIST_INSERT_HEAD(&kh->keys, sk, entry);
1371 		s->key[idx] = sk;
1372 	}
1373 
1374 stateattach:
1375 	/* List is sorted, if-bound states before floating. */
1376 	if (s->kif == V_pfi_all)
1377 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1378 	else
1379 		TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1380 
1381 	if (olds) {
1382 		TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1383 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1384 		    key_list[idx]);
1385 		olds = NULL;
1386 	}
1387 
1388 	/*
1389 	 * Attach done. See how should we (or should not?)
1390 	 * attach a second key.
1391 	 */
1392 	if (sks == skw) {
1393 		s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1394 		idx = PF_SK_STACK;
1395 		sks = NULL;
1396 		goto stateattach;
1397 	} else if (sks != NULL) {
1398 		/*
1399 		 * Continue attaching with stack key.
1400 		 */
1401 		sk = sks;
1402 		kh = khs;
1403 		idx = PF_SK_STACK;
1404 		sks = NULL;
1405 		goto keyattach;
1406 	}
1407 
1408 	PF_STATE_LOCK(s);
1409 	KEYS_UNLOCK();
1410 
1411 	KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1412 	    ("%s failure", __func__));
1413 
1414 	return (0);
1415 #undef	KEYS_UNLOCK
1416 }
1417 
1418 static void
1419 pf_detach_state(struct pf_kstate *s)
1420 {
1421 	struct pf_state_key *sks = s->key[PF_SK_STACK];
1422 	struct pf_keyhash *kh;
1423 
1424 	NET_EPOCH_ASSERT();
1425 	MPASS(s->timeout >= PFTM_MAX);
1426 
1427 	pf_sctp_multihome_detach_addr(s);
1428 
1429 	if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr)
1430 		V_pflow_export_state_ptr(s);
1431 
1432 	if (sks != NULL) {
1433 		kh = &V_pf_keyhash[pf_hashkey(sks)];
1434 		PF_HASHROW_LOCK(kh);
1435 		if (s->key[PF_SK_STACK] != NULL)
1436 			pf_state_key_detach(s, PF_SK_STACK);
1437 		/*
1438 		 * If both point to same key, then we are done.
1439 		 */
1440 		if (sks == s->key[PF_SK_WIRE]) {
1441 			pf_state_key_detach(s, PF_SK_WIRE);
1442 			PF_HASHROW_UNLOCK(kh);
1443 			return;
1444 		}
1445 		PF_HASHROW_UNLOCK(kh);
1446 	}
1447 
1448 	if (s->key[PF_SK_WIRE] != NULL) {
1449 		kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1450 		PF_HASHROW_LOCK(kh);
1451 		if (s->key[PF_SK_WIRE] != NULL)
1452 			pf_state_key_detach(s, PF_SK_WIRE);
1453 		PF_HASHROW_UNLOCK(kh);
1454 	}
1455 }
1456 
1457 static void
1458 pf_state_key_detach(struct pf_kstate *s, int idx)
1459 {
1460 	struct pf_state_key *sk = s->key[idx];
1461 #ifdef INVARIANTS
1462 	struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1463 
1464 	PF_HASHROW_ASSERT(kh);
1465 #endif
1466 	TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1467 	s->key[idx] = NULL;
1468 
1469 	if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1470 		LIST_REMOVE(sk, entry);
1471 		uma_zfree(V_pf_state_key_z, sk);
1472 	}
1473 }
1474 
1475 static int
1476 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1477 {
1478 	struct pf_state_key *sk = mem;
1479 
1480 	bzero(sk, sizeof(struct pf_state_key_cmp));
1481 	TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1482 	TAILQ_INIT(&sk->states[PF_SK_STACK]);
1483 
1484 	return (0);
1485 }
1486 
1487 struct pf_state_key *
1488 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1489 	struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1490 {
1491 	struct pf_state_key *sk;
1492 
1493 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1494 	if (sk == NULL)
1495 		return (NULL);
1496 
1497 	PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1498 	PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1499 	sk->port[pd->sidx] = sport;
1500 	sk->port[pd->didx] = dport;
1501 	sk->proto = pd->proto;
1502 	sk->af = pd->af;
1503 
1504 	return (sk);
1505 }
1506 
1507 struct pf_state_key *
1508 pf_state_key_clone(struct pf_state_key *orig)
1509 {
1510 	struct pf_state_key *sk;
1511 
1512 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1513 	if (sk == NULL)
1514 		return (NULL);
1515 
1516 	bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1517 
1518 	return (sk);
1519 }
1520 
1521 int
1522 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif,
1523     struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s)
1524 {
1525 	struct pf_idhash *ih;
1526 	struct pf_kstate *cur;
1527 	int error;
1528 
1529 	NET_EPOCH_ASSERT();
1530 
1531 	KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1532 	    ("%s: sks not pristine", __func__));
1533 	KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1534 	    ("%s: skw not pristine", __func__));
1535 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1536 
1537 	s->kif = kif;
1538 	s->orig_kif = orig_kif;
1539 
1540 	if (s->id == 0 && s->creatorid == 0) {
1541 		s->id = alloc_unr64(&V_pf_stateid);
1542 		s->id = htobe64(s->id);
1543 		s->creatorid = V_pf_status.hostid;
1544 	}
1545 
1546 	/* Returns with ID locked on success. */
1547 	if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1548 		return (error);
1549 
1550 	ih = &V_pf_idhash[PF_IDHASH(s)];
1551 	PF_HASHROW_ASSERT(ih);
1552 	LIST_FOREACH(cur, &ih->states, entry)
1553 		if (cur->id == s->id && cur->creatorid == s->creatorid)
1554 			break;
1555 
1556 	if (cur != NULL) {
1557 		s->timeout = PFTM_UNLINKED;
1558 		PF_HASHROW_UNLOCK(ih);
1559 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1560 			printf("pf: state ID collision: "
1561 			    "id: %016llx creatorid: %08x\n",
1562 			    (unsigned long long)be64toh(s->id),
1563 			    ntohl(s->creatorid));
1564 		}
1565 		pf_detach_state(s);
1566 		return (EEXIST);
1567 	}
1568 	LIST_INSERT_HEAD(&ih->states, s, entry);
1569 	/* One for keys, one for ID hash. */
1570 	refcount_init(&s->refs, 2);
1571 
1572 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1573 	if (V_pfsync_insert_state_ptr != NULL)
1574 		V_pfsync_insert_state_ptr(s);
1575 
1576 	/* Returns locked. */
1577 	return (0);
1578 }
1579 
1580 /*
1581  * Find state by ID: returns with locked row on success.
1582  */
1583 struct pf_kstate *
1584 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1585 {
1586 	struct pf_idhash *ih;
1587 	struct pf_kstate *s;
1588 
1589 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1590 
1591 	ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1592 
1593 	PF_HASHROW_LOCK(ih);
1594 	LIST_FOREACH(s, &ih->states, entry)
1595 		if (s->id == id && s->creatorid == creatorid)
1596 			break;
1597 
1598 	if (s == NULL)
1599 		PF_HASHROW_UNLOCK(ih);
1600 
1601 	return (s);
1602 }
1603 
1604 /*
1605  * Find state by key.
1606  * Returns with ID hash slot locked on success.
1607  */
1608 static struct pf_kstate *
1609 pf_find_state(struct pfi_kkif *kif, struct pf_state_key_cmp *key, u_int dir)
1610 {
1611 	struct pf_keyhash	*kh;
1612 	struct pf_state_key	*sk;
1613 	struct pf_kstate	*s;
1614 	int idx;
1615 
1616 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1617 
1618 	kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1619 
1620 	PF_HASHROW_LOCK(kh);
1621 	LIST_FOREACH(sk, &kh->keys, entry)
1622 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1623 			break;
1624 	if (sk == NULL) {
1625 		PF_HASHROW_UNLOCK(kh);
1626 		return (NULL);
1627 	}
1628 
1629 	idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1630 
1631 	/* List is sorted, if-bound states before floating ones. */
1632 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1633 		if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) {
1634 			PF_STATE_LOCK(s);
1635 			PF_HASHROW_UNLOCK(kh);
1636 			if (__predict_false(s->timeout >= PFTM_MAX)) {
1637 				/*
1638 				 * State is either being processed by
1639 				 * pf_unlink_state() in an other thread, or
1640 				 * is scheduled for immediate expiry.
1641 				 */
1642 				PF_STATE_UNLOCK(s);
1643 				return (NULL);
1644 			}
1645 			return (s);
1646 		}
1647 	PF_HASHROW_UNLOCK(kh);
1648 
1649 	return (NULL);
1650 }
1651 
1652 /*
1653  * Returns with ID hash slot locked on success.
1654  */
1655 struct pf_kstate *
1656 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1657 {
1658 	struct pf_keyhash	*kh;
1659 	struct pf_state_key	*sk;
1660 	struct pf_kstate	*s, *ret = NULL;
1661 	int			 idx, inout = 0;
1662 
1663 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1664 
1665 	kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1666 
1667 	PF_HASHROW_LOCK(kh);
1668 	LIST_FOREACH(sk, &kh->keys, entry)
1669 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1670 			break;
1671 	if (sk == NULL) {
1672 		PF_HASHROW_UNLOCK(kh);
1673 		return (NULL);
1674 	}
1675 	switch (dir) {
1676 	case PF_IN:
1677 		idx = PF_SK_WIRE;
1678 		break;
1679 	case PF_OUT:
1680 		idx = PF_SK_STACK;
1681 		break;
1682 	case PF_INOUT:
1683 		idx = PF_SK_WIRE;
1684 		inout = 1;
1685 		break;
1686 	default:
1687 		panic("%s: dir %u", __func__, dir);
1688 	}
1689 second_run:
1690 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1691 		if (more == NULL) {
1692 			PF_STATE_LOCK(s);
1693 			PF_HASHROW_UNLOCK(kh);
1694 			return (s);
1695 		}
1696 
1697 		if (ret)
1698 			(*more)++;
1699 		else {
1700 			ret = s;
1701 			PF_STATE_LOCK(s);
1702 		}
1703 	}
1704 	if (inout == 1) {
1705 		inout = 0;
1706 		idx = PF_SK_STACK;
1707 		goto second_run;
1708 	}
1709 	PF_HASHROW_UNLOCK(kh);
1710 
1711 	return (ret);
1712 }
1713 
1714 /*
1715  * FIXME
1716  * This routine is inefficient -- locks the state only to unlock immediately on
1717  * return.
1718  * It is racy -- after the state is unlocked nothing stops other threads from
1719  * removing it.
1720  */
1721 bool
1722 pf_find_state_all_exists(struct pf_state_key_cmp *key, u_int dir)
1723 {
1724 	struct pf_kstate *s;
1725 
1726 	s = pf_find_state_all(key, dir, NULL);
1727 	if (s != NULL) {
1728 		PF_STATE_UNLOCK(s);
1729 		return (true);
1730 	}
1731 	return (false);
1732 }
1733 
1734 /* END state table stuff */
1735 
1736 static void
1737 pf_send(struct pf_send_entry *pfse)
1738 {
1739 
1740 	PF_SENDQ_LOCK();
1741 	STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1742 	PF_SENDQ_UNLOCK();
1743 	swi_sched(V_pf_swi_cookie, 0);
1744 }
1745 
1746 static bool
1747 pf_isforlocal(struct mbuf *m, int af)
1748 {
1749 	switch (af) {
1750 #ifdef INET
1751 	case AF_INET: {
1752 		struct ip *ip = mtod(m, struct ip *);
1753 
1754 		return (in_localip(ip->ip_dst));
1755 	}
1756 #endif
1757 #ifdef INET6
1758 	case AF_INET6: {
1759 		struct ip6_hdr *ip6;
1760 		struct in6_ifaddr *ia;
1761 		ip6 = mtod(m, struct ip6_hdr *);
1762 		ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1763 		if (ia == NULL)
1764 			return (false);
1765 		return (! (ia->ia6_flags & IN6_IFF_NOTREADY));
1766 	}
1767 #endif
1768 	default:
1769 		panic("Unsupported af %d", af);
1770 	}
1771 
1772 	return (false);
1773 }
1774 
1775 void
1776 pf_intr(void *v)
1777 {
1778 	struct epoch_tracker et;
1779 	struct pf_send_head queue;
1780 	struct pf_send_entry *pfse, *next;
1781 
1782 	CURVNET_SET((struct vnet *)v);
1783 
1784 	PF_SENDQ_LOCK();
1785 	queue = V_pf_sendqueue;
1786 	STAILQ_INIT(&V_pf_sendqueue);
1787 	PF_SENDQ_UNLOCK();
1788 
1789 	NET_EPOCH_ENTER(et);
1790 
1791 	STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1792 		switch (pfse->pfse_type) {
1793 #ifdef INET
1794 		case PFSE_IP: {
1795 			if (pf_isforlocal(pfse->pfse_m, AF_INET)) {
1796 				pfse->pfse_m->m_flags |= M_SKIP_FIREWALL;
1797 				pfse->pfse_m->m_pkthdr.csum_flags |=
1798 				    CSUM_IP_VALID | CSUM_IP_CHECKED;
1799 				ip_input(pfse->pfse_m);
1800 			} else {
1801 				ip_output(pfse->pfse_m, NULL, NULL, 0, NULL,
1802 				    NULL);
1803 			}
1804 			break;
1805 		}
1806 		case PFSE_ICMP:
1807 			icmp_error(pfse->pfse_m, pfse->icmpopts.type,
1808 			    pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
1809 			break;
1810 #endif /* INET */
1811 #ifdef INET6
1812 		case PFSE_IP6:
1813 			if (pf_isforlocal(pfse->pfse_m, AF_INET6)) {
1814 				pfse->pfse_m->m_flags |= M_SKIP_FIREWALL;
1815 				ip6_input(pfse->pfse_m);
1816 			} else {
1817 				ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL,
1818 				    NULL, NULL);
1819 			}
1820 			break;
1821 		case PFSE_ICMP6:
1822 			icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
1823 			    pfse->icmpopts.code, pfse->icmpopts.mtu);
1824 			break;
1825 #endif /* INET6 */
1826 		default:
1827 			panic("%s: unknown type", __func__);
1828 		}
1829 		free(pfse, M_PFTEMP);
1830 	}
1831 	NET_EPOCH_EXIT(et);
1832 	CURVNET_RESTORE();
1833 }
1834 
1835 #define	pf_purge_thread_period	(hz / 10)
1836 
1837 #ifdef PF_WANT_32_TO_64_COUNTER
1838 static void
1839 pf_status_counter_u64_periodic(void)
1840 {
1841 
1842 	PF_RULES_RASSERT();
1843 
1844 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) {
1845 		return;
1846 	}
1847 
1848 	for (int i = 0; i < FCNT_MAX; i++) {
1849 		pf_counter_u64_periodic(&V_pf_status.fcounters[i]);
1850 	}
1851 }
1852 
1853 static void
1854 pf_kif_counter_u64_periodic(void)
1855 {
1856 	struct pfi_kkif *kif;
1857 	size_t r, run;
1858 
1859 	PF_RULES_RASSERT();
1860 
1861 	if (__predict_false(V_pf_allkifcount == 0)) {
1862 		return;
1863 	}
1864 
1865 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) {
1866 		return;
1867 	}
1868 
1869 	run = V_pf_allkifcount / 10;
1870 	if (run < 5)
1871 		run = 5;
1872 
1873 	for (r = 0; r < run; r++) {
1874 		kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist);
1875 		if (kif == NULL) {
1876 			LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist);
1877 			LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist);
1878 			break;
1879 		}
1880 
1881 		LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist);
1882 		LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist);
1883 
1884 		for (int i = 0; i < 2; i++) {
1885 			for (int j = 0; j < 2; j++) {
1886 				for (int k = 0; k < 2; k++) {
1887 					pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]);
1888 					pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]);
1889 				}
1890 			}
1891 		}
1892 	}
1893 }
1894 
1895 static void
1896 pf_rule_counter_u64_periodic(void)
1897 {
1898 	struct pf_krule *rule;
1899 	size_t r, run;
1900 
1901 	PF_RULES_RASSERT();
1902 
1903 	if (__predict_false(V_pf_allrulecount == 0)) {
1904 		return;
1905 	}
1906 
1907 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) {
1908 		return;
1909 	}
1910 
1911 	run = V_pf_allrulecount / 10;
1912 	if (run < 5)
1913 		run = 5;
1914 
1915 	for (r = 0; r < run; r++) {
1916 		rule = LIST_NEXT(V_pf_rulemarker, allrulelist);
1917 		if (rule == NULL) {
1918 			LIST_REMOVE(V_pf_rulemarker, allrulelist);
1919 			LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist);
1920 			break;
1921 		}
1922 
1923 		LIST_REMOVE(V_pf_rulemarker, allrulelist);
1924 		LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist);
1925 
1926 		pf_counter_u64_periodic(&rule->evaluations);
1927 		for (int i = 0; i < 2; i++) {
1928 			pf_counter_u64_periodic(&rule->packets[i]);
1929 			pf_counter_u64_periodic(&rule->bytes[i]);
1930 		}
1931 	}
1932 }
1933 
1934 static void
1935 pf_counter_u64_periodic_main(void)
1936 {
1937 	PF_RULES_RLOCK_TRACKER;
1938 
1939 	V_pf_counter_periodic_iter++;
1940 
1941 	PF_RULES_RLOCK();
1942 	pf_counter_u64_critical_enter();
1943 	pf_status_counter_u64_periodic();
1944 	pf_kif_counter_u64_periodic();
1945 	pf_rule_counter_u64_periodic();
1946 	pf_counter_u64_critical_exit();
1947 	PF_RULES_RUNLOCK();
1948 }
1949 #else
1950 #define	pf_counter_u64_periodic_main()	do { } while (0)
1951 #endif
1952 
1953 void
1954 pf_purge_thread(void *unused __unused)
1955 {
1956 	struct epoch_tracker	 et;
1957 
1958 	VNET_ITERATOR_DECL(vnet_iter);
1959 
1960 	sx_xlock(&pf_end_lock);
1961 	while (pf_end_threads == 0) {
1962 		sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period);
1963 
1964 		VNET_LIST_RLOCK();
1965 		NET_EPOCH_ENTER(et);
1966 		VNET_FOREACH(vnet_iter) {
1967 			CURVNET_SET(vnet_iter);
1968 
1969 			/* Wait until V_pf_default_rule is initialized. */
1970 			if (V_pf_vnet_active == 0) {
1971 				CURVNET_RESTORE();
1972 				continue;
1973 			}
1974 
1975 			pf_counter_u64_periodic_main();
1976 
1977 			/*
1978 			 *  Process 1/interval fraction of the state
1979 			 * table every run.
1980 			 */
1981 			V_pf_purge_idx =
1982 			    pf_purge_expired_states(V_pf_purge_idx, pf_hashmask /
1983 			    (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1984 
1985 			/*
1986 			 * Purge other expired types every
1987 			 * PFTM_INTERVAL seconds.
1988 			 */
1989 			if (V_pf_purge_idx == 0) {
1990 				/*
1991 				 * Order is important:
1992 				 * - states and src nodes reference rules
1993 				 * - states and rules reference kifs
1994 				 */
1995 				pf_purge_expired_fragments();
1996 				pf_purge_expired_src_nodes();
1997 				pf_purge_unlinked_rules();
1998 				pfi_kkif_purge();
1999 			}
2000 			CURVNET_RESTORE();
2001 		}
2002 		NET_EPOCH_EXIT(et);
2003 		VNET_LIST_RUNLOCK();
2004 	}
2005 
2006 	pf_end_threads++;
2007 	sx_xunlock(&pf_end_lock);
2008 	kproc_exit(0);
2009 }
2010 
2011 void
2012 pf_unload_vnet_purge(void)
2013 {
2014 
2015 	/*
2016 	 * To cleanse up all kifs and rules we need
2017 	 * two runs: first one clears reference flags,
2018 	 * then pf_purge_expired_states() doesn't
2019 	 * raise them, and then second run frees.
2020 	 */
2021 	pf_purge_unlinked_rules();
2022 	pfi_kkif_purge();
2023 
2024 	/*
2025 	 * Now purge everything.
2026 	 */
2027 	pf_purge_expired_states(0, pf_hashmask);
2028 	pf_purge_fragments(UINT_MAX);
2029 	pf_purge_expired_src_nodes();
2030 
2031 	/*
2032 	 * Now all kifs & rules should be unreferenced,
2033 	 * thus should be successfully freed.
2034 	 */
2035 	pf_purge_unlinked_rules();
2036 	pfi_kkif_purge();
2037 }
2038 
2039 u_int32_t
2040 pf_state_expires(const struct pf_kstate *state)
2041 {
2042 	u_int32_t	timeout;
2043 	u_int32_t	start;
2044 	u_int32_t	end;
2045 	u_int32_t	states;
2046 
2047 	/* handle all PFTM_* > PFTM_MAX here */
2048 	if (state->timeout == PFTM_PURGE)
2049 		return (time_uptime);
2050 	KASSERT(state->timeout != PFTM_UNLINKED,
2051 	    ("pf_state_expires: timeout == PFTM_UNLINKED"));
2052 	KASSERT((state->timeout < PFTM_MAX),
2053 	    ("pf_state_expires: timeout > PFTM_MAX"));
2054 	timeout = state->rule.ptr->timeout[state->timeout];
2055 	if (!timeout)
2056 		timeout = V_pf_default_rule.timeout[state->timeout];
2057 	start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
2058 	if (start && state->rule.ptr != &V_pf_default_rule) {
2059 		end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
2060 		states = counter_u64_fetch(state->rule.ptr->states_cur);
2061 	} else {
2062 		start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
2063 		end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
2064 		states = V_pf_status.states;
2065 	}
2066 	if (end && states > start && start < end) {
2067 		if (states < end) {
2068 			timeout = (u_int64_t)timeout * (end - states) /
2069 			    (end - start);
2070 			return ((state->expire / 1000) + timeout);
2071 		}
2072 		else
2073 			return (time_uptime);
2074 	}
2075 	return ((state->expire / 1000) + timeout);
2076 }
2077 
2078 void
2079 pf_purge_expired_src_nodes(void)
2080 {
2081 	struct pf_ksrc_node_list	 freelist;
2082 	struct pf_srchash	*sh;
2083 	struct pf_ksrc_node	*cur, *next;
2084 	int i;
2085 
2086 	LIST_INIT(&freelist);
2087 	for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
2088 	    PF_HASHROW_LOCK(sh);
2089 	    LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
2090 		if (cur->states == 0 && cur->expire <= time_uptime) {
2091 			pf_unlink_src_node(cur);
2092 			LIST_INSERT_HEAD(&freelist, cur, entry);
2093 		} else if (cur->rule.ptr != NULL)
2094 			cur->rule.ptr->rule_ref |= PFRULE_REFS;
2095 	    PF_HASHROW_UNLOCK(sh);
2096 	}
2097 
2098 	pf_free_src_nodes(&freelist);
2099 
2100 	V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
2101 }
2102 
2103 static void
2104 pf_src_tree_remove_state(struct pf_kstate *s)
2105 {
2106 	struct pf_ksrc_node *sn;
2107 	uint32_t timeout;
2108 
2109 	timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
2110 	    s->rule.ptr->timeout[PFTM_SRC_NODE] :
2111 	    V_pf_default_rule.timeout[PFTM_SRC_NODE];
2112 
2113 	if (s->src_node != NULL) {
2114 		sn = s->src_node;
2115 		PF_SRC_NODE_LOCK(sn);
2116 		if (s->src.tcp_est)
2117 			--sn->conn;
2118 		if (--sn->states == 0)
2119 			sn->expire = time_uptime + timeout;
2120 		PF_SRC_NODE_UNLOCK(sn);
2121 	}
2122 	if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
2123 		sn = s->nat_src_node;
2124 		PF_SRC_NODE_LOCK(sn);
2125 		if (--sn->states == 0)
2126 			sn->expire = time_uptime + timeout;
2127 		PF_SRC_NODE_UNLOCK(sn);
2128 	}
2129 	s->src_node = s->nat_src_node = NULL;
2130 }
2131 
2132 /*
2133  * Unlink and potentilly free a state. Function may be
2134  * called with ID hash row locked, but always returns
2135  * unlocked, since it needs to go through key hash locking.
2136  */
2137 int
2138 pf_unlink_state(struct pf_kstate *s)
2139 {
2140 	struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
2141 
2142 	NET_EPOCH_ASSERT();
2143 	PF_HASHROW_ASSERT(ih);
2144 
2145 	if (s->timeout == PFTM_UNLINKED) {
2146 		/*
2147 		 * State is being processed
2148 		 * by pf_unlink_state() in
2149 		 * an other thread.
2150 		 */
2151 		PF_HASHROW_UNLOCK(ih);
2152 		return (0);	/* XXXGL: undefined actually */
2153 	}
2154 
2155 	if (s->src.state == PF_TCPS_PROXY_DST) {
2156 		/* XXX wire key the right one? */
2157 		pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af,
2158 		    &s->key[PF_SK_WIRE]->addr[1],
2159 		    &s->key[PF_SK_WIRE]->addr[0],
2160 		    s->key[PF_SK_WIRE]->port[1],
2161 		    s->key[PF_SK_WIRE]->port[0],
2162 		    s->src.seqhi, s->src.seqlo + 1,
2163 		    TH_RST|TH_ACK, 0, 0, 0, true, s->tag, 0, s->act.rtableid);
2164 	}
2165 
2166 	LIST_REMOVE(s, entry);
2167 	pf_src_tree_remove_state(s);
2168 
2169 	if (V_pfsync_delete_state_ptr != NULL)
2170 		V_pfsync_delete_state_ptr(s);
2171 
2172 	STATE_DEC_COUNTERS(s);
2173 
2174 	s->timeout = PFTM_UNLINKED;
2175 
2176 	/* Ensure we remove it from the list of halfopen states, if needed. */
2177 	if (s->key[PF_SK_STACK] != NULL &&
2178 	    s->key[PF_SK_STACK]->proto == IPPROTO_TCP)
2179 		pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED);
2180 
2181 	PF_HASHROW_UNLOCK(ih);
2182 
2183 	pf_detach_state(s);
2184 	/* pf_state_insert() initialises refs to 2 */
2185 	return (pf_release_staten(s, 2));
2186 }
2187 
2188 struct pf_kstate *
2189 pf_alloc_state(int flags)
2190 {
2191 
2192 	return (uma_zalloc(V_pf_state_z, flags | M_ZERO));
2193 }
2194 
2195 void
2196 pf_free_state(struct pf_kstate *cur)
2197 {
2198 	struct pf_krule_item *ri;
2199 
2200 	KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
2201 	KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
2202 	    cur->timeout));
2203 
2204 	while ((ri = SLIST_FIRST(&cur->match_rules))) {
2205 		SLIST_REMOVE_HEAD(&cur->match_rules, entry);
2206 		free(ri, M_PF_RULE_ITEM);
2207 	}
2208 
2209 	pf_normalize_tcp_cleanup(cur);
2210 	uma_zfree(V_pf_state_z, cur);
2211 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
2212 }
2213 
2214 /*
2215  * Called only from pf_purge_thread(), thus serialized.
2216  */
2217 static u_int
2218 pf_purge_expired_states(u_int i, int maxcheck)
2219 {
2220 	struct pf_idhash *ih;
2221 	struct pf_kstate *s;
2222 	struct pf_krule_item *mrm;
2223 	size_t count __unused;
2224 
2225 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2226 
2227 	/*
2228 	 * Go through hash and unlink states that expire now.
2229 	 */
2230 	while (maxcheck > 0) {
2231 		count = 0;
2232 		ih = &V_pf_idhash[i];
2233 
2234 		/* only take the lock if we expect to do work */
2235 		if (!LIST_EMPTY(&ih->states)) {
2236 relock:
2237 			PF_HASHROW_LOCK(ih);
2238 			LIST_FOREACH(s, &ih->states, entry) {
2239 				if (pf_state_expires(s) <= time_uptime) {
2240 					V_pf_status.states -=
2241 					    pf_unlink_state(s);
2242 					goto relock;
2243 				}
2244 				s->rule.ptr->rule_ref |= PFRULE_REFS;
2245 				if (s->nat_rule.ptr != NULL)
2246 					s->nat_rule.ptr->rule_ref |= PFRULE_REFS;
2247 				if (s->anchor.ptr != NULL)
2248 					s->anchor.ptr->rule_ref |= PFRULE_REFS;
2249 				s->kif->pfik_flags |= PFI_IFLAG_REFS;
2250 				SLIST_FOREACH(mrm, &s->match_rules, entry)
2251 					mrm->r->rule_ref |= PFRULE_REFS;
2252 				if (s->rt_kif)
2253 					s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
2254 				count++;
2255 			}
2256 			PF_HASHROW_UNLOCK(ih);
2257 		}
2258 
2259 		SDT_PROBE2(pf, purge, state, rowcount, i, count);
2260 
2261 		/* Return when we hit end of hash. */
2262 		if (++i > pf_hashmask) {
2263 			V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2264 			return (0);
2265 		}
2266 
2267 		maxcheck--;
2268 	}
2269 
2270 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2271 
2272 	return (i);
2273 }
2274 
2275 static void
2276 pf_purge_unlinked_rules(void)
2277 {
2278 	struct pf_krulequeue tmpq;
2279 	struct pf_krule *r, *r1;
2280 
2281 	/*
2282 	 * If we have overloading task pending, then we'd
2283 	 * better skip purging this time. There is a tiny
2284 	 * probability that overloading task references
2285 	 * an already unlinked rule.
2286 	 */
2287 	PF_OVERLOADQ_LOCK();
2288 	if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
2289 		PF_OVERLOADQ_UNLOCK();
2290 		return;
2291 	}
2292 	PF_OVERLOADQ_UNLOCK();
2293 
2294 	/*
2295 	 * Do naive mark-and-sweep garbage collecting of old rules.
2296 	 * Reference flag is raised by pf_purge_expired_states()
2297 	 * and pf_purge_expired_src_nodes().
2298 	 *
2299 	 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
2300 	 * use a temporary queue.
2301 	 */
2302 	TAILQ_INIT(&tmpq);
2303 	PF_UNLNKDRULES_LOCK();
2304 	TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
2305 		if (!(r->rule_ref & PFRULE_REFS)) {
2306 			TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
2307 			TAILQ_INSERT_TAIL(&tmpq, r, entries);
2308 		} else
2309 			r->rule_ref &= ~PFRULE_REFS;
2310 	}
2311 	PF_UNLNKDRULES_UNLOCK();
2312 
2313 	if (!TAILQ_EMPTY(&tmpq)) {
2314 		PF_CONFIG_LOCK();
2315 		PF_RULES_WLOCK();
2316 		TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
2317 			TAILQ_REMOVE(&tmpq, r, entries);
2318 			pf_free_rule(r);
2319 		}
2320 		PF_RULES_WUNLOCK();
2321 		PF_CONFIG_UNLOCK();
2322 	}
2323 }
2324 
2325 void
2326 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
2327 {
2328 	switch (af) {
2329 #ifdef INET
2330 	case AF_INET: {
2331 		u_int32_t a = ntohl(addr->addr32[0]);
2332 		printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
2333 		    (a>>8)&255, a&255);
2334 		if (p) {
2335 			p = ntohs(p);
2336 			printf(":%u", p);
2337 		}
2338 		break;
2339 	}
2340 #endif /* INET */
2341 #ifdef INET6
2342 	case AF_INET6: {
2343 		u_int16_t b;
2344 		u_int8_t i, curstart, curend, maxstart, maxend;
2345 		curstart = curend = maxstart = maxend = 255;
2346 		for (i = 0; i < 8; i++) {
2347 			if (!addr->addr16[i]) {
2348 				if (curstart == 255)
2349 					curstart = i;
2350 				curend = i;
2351 			} else {
2352 				if ((curend - curstart) >
2353 				    (maxend - maxstart)) {
2354 					maxstart = curstart;
2355 					maxend = curend;
2356 				}
2357 				curstart = curend = 255;
2358 			}
2359 		}
2360 		if ((curend - curstart) >
2361 		    (maxend - maxstart)) {
2362 			maxstart = curstart;
2363 			maxend = curend;
2364 		}
2365 		for (i = 0; i < 8; i++) {
2366 			if (i >= maxstart && i <= maxend) {
2367 				if (i == 0)
2368 					printf(":");
2369 				if (i == maxend)
2370 					printf(":");
2371 			} else {
2372 				b = ntohs(addr->addr16[i]);
2373 				printf("%x", b);
2374 				if (i < 7)
2375 					printf(":");
2376 			}
2377 		}
2378 		if (p) {
2379 			p = ntohs(p);
2380 			printf("[%u]", p);
2381 		}
2382 		break;
2383 	}
2384 #endif /* INET6 */
2385 	}
2386 }
2387 
2388 void
2389 pf_print_state(struct pf_kstate *s)
2390 {
2391 	pf_print_state_parts(s, NULL, NULL);
2392 }
2393 
2394 static void
2395 pf_print_state_parts(struct pf_kstate *s,
2396     struct pf_state_key *skwp, struct pf_state_key *sksp)
2397 {
2398 	struct pf_state_key *skw, *sks;
2399 	u_int8_t proto, dir;
2400 
2401 	/* Do our best to fill these, but they're skipped if NULL */
2402 	skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
2403 	sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
2404 	proto = skw ? skw->proto : (sks ? sks->proto : 0);
2405 	dir = s ? s->direction : 0;
2406 
2407 	switch (proto) {
2408 	case IPPROTO_IPV4:
2409 		printf("IPv4");
2410 		break;
2411 	case IPPROTO_IPV6:
2412 		printf("IPv6");
2413 		break;
2414 	case IPPROTO_TCP:
2415 		printf("TCP");
2416 		break;
2417 	case IPPROTO_UDP:
2418 		printf("UDP");
2419 		break;
2420 	case IPPROTO_ICMP:
2421 		printf("ICMP");
2422 		break;
2423 	case IPPROTO_ICMPV6:
2424 		printf("ICMPv6");
2425 		break;
2426 	default:
2427 		printf("%u", proto);
2428 		break;
2429 	}
2430 	switch (dir) {
2431 	case PF_IN:
2432 		printf(" in");
2433 		break;
2434 	case PF_OUT:
2435 		printf(" out");
2436 		break;
2437 	}
2438 	if (skw) {
2439 		printf(" wire: ");
2440 		pf_print_host(&skw->addr[0], skw->port[0], skw->af);
2441 		printf(" ");
2442 		pf_print_host(&skw->addr[1], skw->port[1], skw->af);
2443 	}
2444 	if (sks) {
2445 		printf(" stack: ");
2446 		if (sks != skw) {
2447 			pf_print_host(&sks->addr[0], sks->port[0], sks->af);
2448 			printf(" ");
2449 			pf_print_host(&sks->addr[1], sks->port[1], sks->af);
2450 		} else
2451 			printf("-");
2452 	}
2453 	if (s) {
2454 		if (proto == IPPROTO_TCP) {
2455 			printf(" [lo=%u high=%u win=%u modulator=%u",
2456 			    s->src.seqlo, s->src.seqhi,
2457 			    s->src.max_win, s->src.seqdiff);
2458 			if (s->src.wscale && s->dst.wscale)
2459 				printf(" wscale=%u",
2460 				    s->src.wscale & PF_WSCALE_MASK);
2461 			printf("]");
2462 			printf(" [lo=%u high=%u win=%u modulator=%u",
2463 			    s->dst.seqlo, s->dst.seqhi,
2464 			    s->dst.max_win, s->dst.seqdiff);
2465 			if (s->src.wscale && s->dst.wscale)
2466 				printf(" wscale=%u",
2467 				s->dst.wscale & PF_WSCALE_MASK);
2468 			printf("]");
2469 		}
2470 		printf(" %u:%u", s->src.state, s->dst.state);
2471 	}
2472 }
2473 
2474 void
2475 pf_print_flags(u_int8_t f)
2476 {
2477 	if (f)
2478 		printf(" ");
2479 	if (f & TH_FIN)
2480 		printf("F");
2481 	if (f & TH_SYN)
2482 		printf("S");
2483 	if (f & TH_RST)
2484 		printf("R");
2485 	if (f & TH_PUSH)
2486 		printf("P");
2487 	if (f & TH_ACK)
2488 		printf("A");
2489 	if (f & TH_URG)
2490 		printf("U");
2491 	if (f & TH_ECE)
2492 		printf("E");
2493 	if (f & TH_CWR)
2494 		printf("W");
2495 }
2496 
2497 #define	PF_SET_SKIP_STEPS(i)					\
2498 	do {							\
2499 		while (head[i] != cur) {			\
2500 			head[i]->skip[i].ptr = cur;		\
2501 			head[i] = TAILQ_NEXT(head[i], entries);	\
2502 		}						\
2503 	} while (0)
2504 
2505 void
2506 pf_calc_skip_steps(struct pf_krulequeue *rules)
2507 {
2508 	struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT];
2509 	int i;
2510 
2511 	cur = TAILQ_FIRST(rules);
2512 	prev = cur;
2513 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2514 		head[i] = cur;
2515 	while (cur != NULL) {
2516 		if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
2517 			PF_SET_SKIP_STEPS(PF_SKIP_IFP);
2518 		if (cur->direction != prev->direction)
2519 			PF_SET_SKIP_STEPS(PF_SKIP_DIR);
2520 		if (cur->af != prev->af)
2521 			PF_SET_SKIP_STEPS(PF_SKIP_AF);
2522 		if (cur->proto != prev->proto)
2523 			PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
2524 		if (cur->src.neg != prev->src.neg ||
2525 		    pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
2526 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
2527 		if (cur->src.port[0] != prev->src.port[0] ||
2528 		    cur->src.port[1] != prev->src.port[1] ||
2529 		    cur->src.port_op != prev->src.port_op)
2530 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
2531 		if (cur->dst.neg != prev->dst.neg ||
2532 		    pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
2533 			PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
2534 		if (cur->dst.port[0] != prev->dst.port[0] ||
2535 		    cur->dst.port[1] != prev->dst.port[1] ||
2536 		    cur->dst.port_op != prev->dst.port_op)
2537 			PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
2538 
2539 		prev = cur;
2540 		cur = TAILQ_NEXT(cur, entries);
2541 	}
2542 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2543 		PF_SET_SKIP_STEPS(i);
2544 }
2545 
2546 int
2547 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
2548 {
2549 	if (aw1->type != aw2->type)
2550 		return (1);
2551 	switch (aw1->type) {
2552 	case PF_ADDR_ADDRMASK:
2553 	case PF_ADDR_RANGE:
2554 		if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
2555 			return (1);
2556 		if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
2557 			return (1);
2558 		return (0);
2559 	case PF_ADDR_DYNIFTL:
2560 		return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
2561 	case PF_ADDR_NOROUTE:
2562 	case PF_ADDR_URPFFAILED:
2563 		return (0);
2564 	case PF_ADDR_TABLE:
2565 		return (aw1->p.tbl != aw2->p.tbl);
2566 	default:
2567 		printf("invalid address type: %d\n", aw1->type);
2568 		return (1);
2569 	}
2570 }
2571 
2572 /**
2573  * Checksum updates are a little complicated because the checksum in the TCP/UDP
2574  * header isn't always a full checksum. In some cases (i.e. output) it's a
2575  * pseudo-header checksum, which is a partial checksum over src/dst IP
2576  * addresses, protocol number and length.
2577  *
2578  * That means we have the following cases:
2579  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2580  *  	checksums, we need to update the checksum whenever we change anything.
2581  *  * Output (i.e. the checksum is a pseudo-header checksum):
2582  *  	x The field being updated is src/dst address or affects the length of
2583  *  	the packet. We need to update the pseudo-header checksum (note that this
2584  *  	checksum is not ones' complement).
2585  *  	x Some other field is being modified (e.g. src/dst port numbers): We
2586  *  	don't have to update anything.
2587  **/
2588 u_int16_t
2589 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2590 {
2591 	u_int32_t x;
2592 
2593 	x = cksum + old - new;
2594 	x = (x + (x >> 16)) & 0xffff;
2595 
2596 	/* optimise: eliminate a branch when not udp */
2597 	if (udp && cksum == 0x0000)
2598 		return cksum;
2599 	if (udp && x == 0x0000)
2600 		x = 0xffff;
2601 
2602 	return (u_int16_t)(x);
2603 }
2604 
2605 static void
2606 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi,
2607     u_int8_t udp)
2608 {
2609 	u_int16_t old = htons(hi ? (*f << 8) : *f);
2610 	u_int16_t new = htons(hi ? ( v << 8) :  v);
2611 
2612 	if (*f == v)
2613 		return;
2614 
2615 	*f = v;
2616 
2617 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2618 		return;
2619 
2620 	*cksum = pf_cksum_fixup(*cksum, old, new, udp);
2621 }
2622 
2623 void
2624 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_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 }
2633 
2634 void
2635 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v,
2636     bool hi, u_int8_t udp)
2637 {
2638 	u_int8_t *fb = (u_int8_t *)f;
2639 	u_int8_t *vb = (u_int8_t *)&v;
2640 
2641 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2642 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2643 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2644 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2645 }
2646 
2647 u_int16_t
2648 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2649         u_int16_t new, u_int8_t udp)
2650 {
2651 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2652 		return (cksum);
2653 
2654 	return (pf_cksum_fixup(cksum, old, new, udp));
2655 }
2656 
2657 static void
2658 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2659         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2660         sa_family_t af)
2661 {
2662 	struct pf_addr	ao;
2663 	u_int16_t	po = *p;
2664 
2665 	PF_ACPY(&ao, a, af);
2666 	PF_ACPY(a, an, af);
2667 
2668 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2669 		*pc = ~*pc;
2670 
2671 	*p = pn;
2672 
2673 	switch (af) {
2674 #ifdef INET
2675 	case AF_INET:
2676 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2677 		    ao.addr16[0], an->addr16[0], 0),
2678 		    ao.addr16[1], an->addr16[1], 0);
2679 		*p = pn;
2680 
2681 		*pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2682 		    ao.addr16[0], an->addr16[0], u),
2683 		    ao.addr16[1], an->addr16[1], u);
2684 
2685 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2686 		break;
2687 #endif /* INET */
2688 #ifdef INET6
2689 	case AF_INET6:
2690 		*pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2691 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2692 		    pf_cksum_fixup(pf_cksum_fixup(*pc,
2693 		    ao.addr16[0], an->addr16[0], u),
2694 		    ao.addr16[1], an->addr16[1], u),
2695 		    ao.addr16[2], an->addr16[2], u),
2696 		    ao.addr16[3], an->addr16[3], u),
2697 		    ao.addr16[4], an->addr16[4], u),
2698 		    ao.addr16[5], an->addr16[5], u),
2699 		    ao.addr16[6], an->addr16[6], u),
2700 		    ao.addr16[7], an->addr16[7], u);
2701 
2702 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2703 		break;
2704 #endif /* INET6 */
2705 	}
2706 
2707 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2708 	    CSUM_DELAY_DATA_IPV6)) {
2709 		*pc = ~*pc;
2710 		if (! *pc)
2711 			*pc = 0xffff;
2712 	}
2713 }
2714 
2715 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2716 void
2717 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2718 {
2719 	u_int32_t	ao;
2720 
2721 	memcpy(&ao, a, sizeof(ao));
2722 	memcpy(a, &an, sizeof(u_int32_t));
2723 	*c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2724 	    ao % 65536, an % 65536, u);
2725 }
2726 
2727 void
2728 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2729 {
2730 	u_int32_t	ao;
2731 
2732 	memcpy(&ao, a, sizeof(ao));
2733 	memcpy(a, &an, sizeof(u_int32_t));
2734 
2735 	*c = pf_proto_cksum_fixup(m,
2736 	    pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2737 	    ao % 65536, an % 65536, udp);
2738 }
2739 
2740 #ifdef INET6
2741 static void
2742 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2743 {
2744 	struct pf_addr	ao;
2745 
2746 	PF_ACPY(&ao, a, AF_INET6);
2747 	PF_ACPY(a, an, AF_INET6);
2748 
2749 	*c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2750 	    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2751 	    pf_cksum_fixup(pf_cksum_fixup(*c,
2752 	    ao.addr16[0], an->addr16[0], u),
2753 	    ao.addr16[1], an->addr16[1], u),
2754 	    ao.addr16[2], an->addr16[2], u),
2755 	    ao.addr16[3], an->addr16[3], u),
2756 	    ao.addr16[4], an->addr16[4], u),
2757 	    ao.addr16[5], an->addr16[5], u),
2758 	    ao.addr16[6], an->addr16[6], u),
2759 	    ao.addr16[7], an->addr16[7], u);
2760 }
2761 #endif /* INET6 */
2762 
2763 static void
2764 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2765     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2766     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2767 {
2768 	struct pf_addr	oia, ooa;
2769 
2770 	PF_ACPY(&oia, ia, af);
2771 	if (oa)
2772 		PF_ACPY(&ooa, oa, af);
2773 
2774 	/* Change inner protocol port, fix inner protocol checksum. */
2775 	if (ip != NULL) {
2776 		u_int16_t	oip = *ip;
2777 		u_int32_t	opc;
2778 
2779 		if (pc != NULL)
2780 			opc = *pc;
2781 		*ip = np;
2782 		if (pc != NULL)
2783 			*pc = pf_cksum_fixup(*pc, oip, *ip, u);
2784 		*ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2785 		if (pc != NULL)
2786 			*ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2787 	}
2788 	/* Change inner ip address, fix inner ip and icmp checksums. */
2789 	PF_ACPY(ia, na, af);
2790 	switch (af) {
2791 #ifdef INET
2792 	case AF_INET: {
2793 		u_int32_t	 oh2c = *h2c;
2794 
2795 		*h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2796 		    oia.addr16[0], ia->addr16[0], 0),
2797 		    oia.addr16[1], ia->addr16[1], 0);
2798 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2799 		    oia.addr16[0], ia->addr16[0], 0),
2800 		    oia.addr16[1], ia->addr16[1], 0);
2801 		*ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2802 		break;
2803 	}
2804 #endif /* INET */
2805 #ifdef INET6
2806 	case AF_INET6:
2807 		*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2808 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2809 		    pf_cksum_fixup(pf_cksum_fixup(*ic,
2810 		    oia.addr16[0], ia->addr16[0], u),
2811 		    oia.addr16[1], ia->addr16[1], u),
2812 		    oia.addr16[2], ia->addr16[2], u),
2813 		    oia.addr16[3], ia->addr16[3], u),
2814 		    oia.addr16[4], ia->addr16[4], u),
2815 		    oia.addr16[5], ia->addr16[5], u),
2816 		    oia.addr16[6], ia->addr16[6], u),
2817 		    oia.addr16[7], ia->addr16[7], u);
2818 		break;
2819 #endif /* INET6 */
2820 	}
2821 	/* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2822 	if (oa) {
2823 		PF_ACPY(oa, na, af);
2824 		switch (af) {
2825 #ifdef INET
2826 		case AF_INET:
2827 			*hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2828 			    ooa.addr16[0], oa->addr16[0], 0),
2829 			    ooa.addr16[1], oa->addr16[1], 0);
2830 			break;
2831 #endif /* INET */
2832 #ifdef INET6
2833 		case AF_INET6:
2834 			*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2835 			    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2836 			    pf_cksum_fixup(pf_cksum_fixup(*ic,
2837 			    ooa.addr16[0], oa->addr16[0], u),
2838 			    ooa.addr16[1], oa->addr16[1], u),
2839 			    ooa.addr16[2], oa->addr16[2], u),
2840 			    ooa.addr16[3], oa->addr16[3], u),
2841 			    ooa.addr16[4], oa->addr16[4], u),
2842 			    ooa.addr16[5], oa->addr16[5], u),
2843 			    ooa.addr16[6], oa->addr16[6], u),
2844 			    ooa.addr16[7], oa->addr16[7], u);
2845 			break;
2846 #endif /* INET6 */
2847 		}
2848 	}
2849 }
2850 
2851 /*
2852  * Need to modulate the sequence numbers in the TCP SACK option
2853  * (credits to Krzysztof Pfaff for report and patch)
2854  */
2855 static int
2856 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2857     struct tcphdr *th, struct pf_state_peer *dst)
2858 {
2859 	int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2860 	u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2861 	int copyback = 0, i, olen;
2862 	struct sackblk sack;
2863 
2864 #define	TCPOLEN_SACKLEN	(TCPOLEN_SACK + 2)
2865 	if (hlen < TCPOLEN_SACKLEN ||
2866 	    !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2867 		return 0;
2868 
2869 	while (hlen >= TCPOLEN_SACKLEN) {
2870 		size_t startoff = opt - opts;
2871 		olen = opt[1];
2872 		switch (*opt) {
2873 		case TCPOPT_EOL:	/* FALLTHROUGH */
2874 		case TCPOPT_NOP:
2875 			opt++;
2876 			hlen--;
2877 			break;
2878 		case TCPOPT_SACK:
2879 			if (olen > hlen)
2880 				olen = hlen;
2881 			if (olen >= TCPOLEN_SACKLEN) {
2882 				for (i = 2; i + TCPOLEN_SACK <= olen;
2883 				    i += TCPOLEN_SACK) {
2884 					memcpy(&sack, &opt[i], sizeof(sack));
2885 					pf_patch_32_unaligned(m,
2886 					    &th->th_sum, &sack.start,
2887 					    htonl(ntohl(sack.start) - dst->seqdiff),
2888 					    PF_ALGNMNT(startoff),
2889 					    0);
2890 					pf_patch_32_unaligned(m, &th->th_sum,
2891 					    &sack.end,
2892 					    htonl(ntohl(sack.end) - dst->seqdiff),
2893 					    PF_ALGNMNT(startoff),
2894 					    0);
2895 					memcpy(&opt[i], &sack, sizeof(sack));
2896 				}
2897 				copyback = 1;
2898 			}
2899 			/* FALLTHROUGH */
2900 		default:
2901 			if (olen < 2)
2902 				olen = 2;
2903 			hlen -= olen;
2904 			opt += olen;
2905 		}
2906 	}
2907 
2908 	if (copyback)
2909 		m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2910 	return (copyback);
2911 }
2912 
2913 struct mbuf *
2914 pf_build_tcp(const struct pf_krule *r, sa_family_t af,
2915     const struct pf_addr *saddr, const struct pf_addr *daddr,
2916     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2917     u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl,
2918     bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid)
2919 {
2920 	struct mbuf	*m;
2921 	int		 len, tlen;
2922 #ifdef INET
2923 	struct ip	*h = NULL;
2924 #endif /* INET */
2925 #ifdef INET6
2926 	struct ip6_hdr	*h6 = NULL;
2927 #endif /* INET6 */
2928 	struct tcphdr	*th;
2929 	char		*opt;
2930 	struct pf_mtag  *pf_mtag;
2931 
2932 	len = 0;
2933 	th = NULL;
2934 
2935 	/* maximum segment size tcp option */
2936 	tlen = sizeof(struct tcphdr);
2937 	if (mss)
2938 		tlen += 4;
2939 
2940 	switch (af) {
2941 #ifdef INET
2942 	case AF_INET:
2943 		len = sizeof(struct ip) + tlen;
2944 		break;
2945 #endif /* INET */
2946 #ifdef INET6
2947 	case AF_INET6:
2948 		len = sizeof(struct ip6_hdr) + tlen;
2949 		break;
2950 #endif /* INET6 */
2951 	default:
2952 		panic("%s: unsupported af %d", __func__, af);
2953 	}
2954 
2955 	m = m_gethdr(M_NOWAIT, MT_DATA);
2956 	if (m == NULL)
2957 		return (NULL);
2958 
2959 #ifdef MAC
2960 	mac_netinet_firewall_send(m);
2961 #endif
2962 	if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2963 		m_freem(m);
2964 		return (NULL);
2965 	}
2966 	if (skip_firewall)
2967 		m->m_flags |= M_SKIP_FIREWALL;
2968 	pf_mtag->tag = mtag_tag;
2969 	pf_mtag->flags = mtag_flags;
2970 
2971 	if (rtableid >= 0)
2972 		M_SETFIB(m, rtableid);
2973 
2974 #ifdef ALTQ
2975 	if (r != NULL && r->qid) {
2976 		pf_mtag->qid = r->qid;
2977 
2978 		/* add hints for ecn */
2979 		pf_mtag->hdr = mtod(m, struct ip *);
2980 	}
2981 #endif /* ALTQ */
2982 	m->m_data += max_linkhdr;
2983 	m->m_pkthdr.len = m->m_len = len;
2984 	/* The rest of the stack assumes a rcvif, so provide one.
2985 	 * This is a locally generated packet, so .. close enough. */
2986 	m->m_pkthdr.rcvif = V_loif;
2987 	bzero(m->m_data, len);
2988 	switch (af) {
2989 #ifdef INET
2990 	case AF_INET:
2991 		h = mtod(m, struct ip *);
2992 
2993 		/* IP header fields included in the TCP checksum */
2994 		h->ip_p = IPPROTO_TCP;
2995 		h->ip_len = htons(tlen);
2996 		h->ip_src.s_addr = saddr->v4.s_addr;
2997 		h->ip_dst.s_addr = daddr->v4.s_addr;
2998 
2999 		th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
3000 		break;
3001 #endif /* INET */
3002 #ifdef INET6
3003 	case AF_INET6:
3004 		h6 = mtod(m, struct ip6_hdr *);
3005 
3006 		/* IP header fields included in the TCP checksum */
3007 		h6->ip6_nxt = IPPROTO_TCP;
3008 		h6->ip6_plen = htons(tlen);
3009 		memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
3010 		memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
3011 
3012 		th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
3013 		break;
3014 #endif /* INET6 */
3015 	}
3016 
3017 	/* TCP header */
3018 	th->th_sport = sport;
3019 	th->th_dport = dport;
3020 	th->th_seq = htonl(seq);
3021 	th->th_ack = htonl(ack);
3022 	th->th_off = tlen >> 2;
3023 	th->th_flags = tcp_flags;
3024 	th->th_win = htons(win);
3025 
3026 	if (mss) {
3027 		opt = (char *)(th + 1);
3028 		opt[0] = TCPOPT_MAXSEG;
3029 		opt[1] = 4;
3030 		HTONS(mss);
3031 		bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
3032 	}
3033 
3034 	switch (af) {
3035 #ifdef INET
3036 	case AF_INET:
3037 		/* TCP checksum */
3038 		th->th_sum = in_cksum(m, len);
3039 
3040 		/* Finish the IP header */
3041 		h->ip_v = 4;
3042 		h->ip_hl = sizeof(*h) >> 2;
3043 		h->ip_tos = IPTOS_LOWDELAY;
3044 		h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
3045 		h->ip_len = htons(len);
3046 		h->ip_ttl = ttl ? ttl : V_ip_defttl;
3047 		h->ip_sum = 0;
3048 		break;
3049 #endif /* INET */
3050 #ifdef INET6
3051 	case AF_INET6:
3052 		/* TCP checksum */
3053 		th->th_sum = in6_cksum(m, IPPROTO_TCP,
3054 		    sizeof(struct ip6_hdr), tlen);
3055 
3056 		h6->ip6_vfc |= IPV6_VERSION;
3057 		h6->ip6_hlim = IPV6_DEFHLIM;
3058 		break;
3059 #endif /* INET6 */
3060 	}
3061 
3062 	return (m);
3063 }
3064 
3065 static void
3066 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd,
3067     uint8_t ttl, int rtableid)
3068 {
3069 	struct mbuf		*m;
3070 #ifdef INET
3071 	struct ip		*h = NULL;
3072 #endif /* INET */
3073 #ifdef INET6
3074 	struct ip6_hdr		*h6 = NULL;
3075 #endif /* INET6 */
3076 	struct sctphdr		*hdr;
3077 	struct sctp_chunkhdr	*chunk;
3078 	struct pf_send_entry	*pfse;
3079 	int			 off = 0;
3080 
3081 	MPASS(af == pd->af);
3082 
3083 	m = m_gethdr(M_NOWAIT, MT_DATA);
3084 	if (m == NULL)
3085 		return;
3086 
3087 	m->m_data += max_linkhdr;
3088 	m->m_flags |= M_SKIP_FIREWALL;
3089 	/* The rest of the stack assumes a rcvif, so provide one.
3090 	 * This is a locally generated packet, so .. close enough. */
3091 	m->m_pkthdr.rcvif = V_loif;
3092 
3093 	/* IPv4|6 header */
3094 	switch (af) {
3095 #ifdef INET
3096 	case AF_INET:
3097 		bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk));
3098 
3099 		h = mtod(m, struct ip *);
3100 
3101 		/* IP header fields included in the TCP checksum */
3102 
3103 		h->ip_p = IPPROTO_SCTP;
3104 		h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk));
3105 		h->ip_ttl = ttl ? ttl : V_ip_defttl;
3106 		h->ip_src = pd->dst->v4;
3107 		h->ip_dst = pd->src->v4;
3108 
3109 		off += sizeof(struct ip);
3110 		break;
3111 #endif /* INET */
3112 #ifdef INET6
3113 	case AF_INET6:
3114 		bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk));
3115 
3116 		h6 = mtod(m, struct ip6_hdr *);
3117 
3118 		/* IP header fields included in the TCP checksum */
3119 		h6->ip6_vfc |= IPV6_VERSION;
3120 		h6->ip6_nxt = IPPROTO_SCTP;
3121 		h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk));
3122 		h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim;
3123 		memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr));
3124 		memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr));
3125 
3126 		off += sizeof(struct ip6_hdr);
3127 		break;
3128 #endif /* INET6 */
3129 	}
3130 
3131 	/* SCTP header */
3132 	hdr = mtodo(m, off);
3133 
3134 	hdr->src_port = pd->hdr.sctp.dest_port;
3135 	hdr->dest_port = pd->hdr.sctp.src_port;
3136 	hdr->v_tag = pd->sctp_initiate_tag;
3137 	hdr->checksum = 0;
3138 
3139 	/* Abort chunk. */
3140 	off += sizeof(struct sctphdr);
3141 	chunk = mtodo(m, off);
3142 
3143 	chunk->chunk_type = SCTP_ABORT_ASSOCIATION;
3144 	chunk->chunk_length = htons(sizeof(*chunk));
3145 
3146 	/* SCTP checksum */
3147 	off += sizeof(*chunk);
3148 	m->m_pkthdr.len = m->m_len = off;
3149 
3150 	pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk));
3151 
3152 	if (rtableid >= 0)
3153 		M_SETFIB(m, rtableid);
3154 
3155 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3156 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3157 	if (pfse == NULL) {
3158 		m_freem(m);
3159 		return;
3160 	}
3161 
3162 	switch (af) {
3163 #ifdef INET
3164 	case AF_INET:
3165 		pfse->pfse_type = PFSE_IP;
3166 		break;
3167 #endif /* INET */
3168 #ifdef INET6
3169 	case AF_INET6:
3170 		pfse->pfse_type = PFSE_IP6;
3171 		break;
3172 #endif /* INET6 */
3173 	}
3174 
3175 	pfse->pfse_m = m;
3176 	pf_send(pfse);
3177 }
3178 
3179 void
3180 pf_send_tcp(const struct pf_krule *r, sa_family_t af,
3181     const struct pf_addr *saddr, const struct pf_addr *daddr,
3182     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
3183     u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl,
3184     bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid)
3185 {
3186 	struct pf_send_entry *pfse;
3187 	struct mbuf	*m;
3188 
3189 	m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags,
3190 	    win, mss, ttl, skip_firewall, mtag_tag, mtag_flags, rtableid);
3191 	if (m == NULL)
3192 		return;
3193 
3194 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3195 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3196 	if (pfse == NULL) {
3197 		m_freem(m);
3198 		return;
3199 	}
3200 
3201 	switch (af) {
3202 #ifdef INET
3203 	case AF_INET:
3204 		pfse->pfse_type = PFSE_IP;
3205 		break;
3206 #endif /* INET */
3207 #ifdef INET6
3208 	case AF_INET6:
3209 		pfse->pfse_type = PFSE_IP6;
3210 		break;
3211 #endif /* INET6 */
3212 	}
3213 
3214 	pfse->pfse_m = m;
3215 	pf_send(pfse);
3216 }
3217 
3218 static void
3219 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd,
3220     struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th,
3221     struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
3222     u_short *reason, int rtableid)
3223 {
3224 	struct pf_addr	* const saddr = pd->src;
3225 	struct pf_addr	* const daddr = pd->dst;
3226 	sa_family_t	 af = pd->af;
3227 
3228 	/* undo NAT changes, if they have taken place */
3229 	if (nr != NULL) {
3230 		PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3231 		PF_ACPY(daddr, &sk->addr[pd->didx], af);
3232 		if (pd->sport)
3233 			*pd->sport = sk->port[pd->sidx];
3234 		if (pd->dport)
3235 			*pd->dport = sk->port[pd->didx];
3236 		if (pd->proto_sum)
3237 			*pd->proto_sum = bproto_sum;
3238 		if (pd->ip_sum)
3239 			*pd->ip_sum = bip_sum;
3240 		m_copyback(m, off, hdrlen, pd->hdr.any);
3241 	}
3242 	if (pd->proto == IPPROTO_TCP &&
3243 	    ((r->rule_flag & PFRULE_RETURNRST) ||
3244 	    (r->rule_flag & PFRULE_RETURN)) &&
3245 	    !(th->th_flags & TH_RST)) {
3246 		u_int32_t	 ack = ntohl(th->th_seq) + pd->p_len;
3247 		int		 len = 0;
3248 #ifdef INET
3249 		struct ip	*h4;
3250 #endif
3251 #ifdef INET6
3252 		struct ip6_hdr	*h6;
3253 #endif
3254 
3255 		switch (af) {
3256 #ifdef INET
3257 		case AF_INET:
3258 			h4 = mtod(m, struct ip *);
3259 			len = ntohs(h4->ip_len) - off;
3260 			break;
3261 #endif
3262 #ifdef INET6
3263 		case AF_INET6:
3264 			h6 = mtod(m, struct ip6_hdr *);
3265 			len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3266 			break;
3267 #endif
3268 		}
3269 
3270 		if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3271 			REASON_SET(reason, PFRES_PROTCKSUM);
3272 		else {
3273 			if (th->th_flags & TH_SYN)
3274 				ack++;
3275 			if (th->th_flags & TH_FIN)
3276 				ack++;
3277 			pf_send_tcp(r, af, pd->dst,
3278 				pd->src, th->th_dport, th->th_sport,
3279 				ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3280 				r->return_ttl, true, 0, 0, rtableid);
3281 		}
3282 	} else if (pd->proto == IPPROTO_SCTP &&
3283 	    (r->rule_flag & PFRULE_RETURN)) {
3284 		pf_send_sctp_abort(af, pd, r->return_ttl, rtableid);
3285 	} else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3286 		r->return_icmp)
3287 		pf_send_icmp(m, r->return_icmp >> 8,
3288 			r->return_icmp & 255, af, r, rtableid);
3289 	else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3290 		r->return_icmp6)
3291 		pf_send_icmp(m, r->return_icmp6 >> 8,
3292 			r->return_icmp6 & 255, af, r, rtableid);
3293 }
3294 
3295 static int
3296 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
3297 {
3298 	struct m_tag *mtag;
3299 	u_int8_t mpcp;
3300 
3301 	mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
3302 	if (mtag == NULL)
3303 		return (0);
3304 
3305 	if (prio == PF_PRIO_ZERO)
3306 		prio = 0;
3307 
3308 	mpcp = *(uint8_t *)(mtag + 1);
3309 
3310 	return (mpcp == prio);
3311 }
3312 
3313 static int
3314 pf_icmp_to_bandlim(uint8_t type)
3315 {
3316 	switch (type) {
3317 		case ICMP_ECHO:
3318 		case ICMP_ECHOREPLY:
3319 			return (BANDLIM_ICMP_ECHO);
3320 		case ICMP_TSTAMP:
3321 		case ICMP_TSTAMPREPLY:
3322 			return (BANDLIM_ICMP_TSTAMP);
3323 		case ICMP_UNREACH:
3324 		default:
3325 			return (BANDLIM_ICMP_UNREACH);
3326 	}
3327 }
3328 
3329 static void
3330 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
3331     struct pf_krule *r, int rtableid)
3332 {
3333 	struct pf_send_entry *pfse;
3334 	struct mbuf *m0;
3335 	struct pf_mtag *pf_mtag;
3336 
3337 	/* ICMP packet rate limitation. */
3338 #ifdef INET6
3339 	if (af == AF_INET6) {
3340 		if (icmp6_ratelimit(NULL, type, code))
3341 			return;
3342 	}
3343 #endif
3344 #ifdef INET
3345 	if (af == AF_INET) {
3346 		if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0)
3347 			return;
3348 	}
3349 #endif
3350 
3351 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3352 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3353 	if (pfse == NULL)
3354 		return;
3355 
3356 	if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
3357 		free(pfse, M_PFTEMP);
3358 		return;
3359 	}
3360 
3361 	if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
3362 		free(pfse, M_PFTEMP);
3363 		return;
3364 	}
3365 	/* XXX: revisit */
3366 	m0->m_flags |= M_SKIP_FIREWALL;
3367 
3368 	if (rtableid >= 0)
3369 		M_SETFIB(m0, rtableid);
3370 
3371 #ifdef ALTQ
3372 	if (r->qid) {
3373 		pf_mtag->qid = r->qid;
3374 		/* add hints for ecn */
3375 		pf_mtag->hdr = mtod(m0, struct ip *);
3376 	}
3377 #endif /* ALTQ */
3378 
3379 	switch (af) {
3380 #ifdef INET
3381 	case AF_INET:
3382 		pfse->pfse_type = PFSE_ICMP;
3383 		break;
3384 #endif /* INET */
3385 #ifdef INET6
3386 	case AF_INET6:
3387 		pfse->pfse_type = PFSE_ICMP6;
3388 		break;
3389 #endif /* INET6 */
3390 	}
3391 	pfse->pfse_m = m0;
3392 	pfse->icmpopts.type = type;
3393 	pfse->icmpopts.code = code;
3394 	pf_send(pfse);
3395 }
3396 
3397 /*
3398  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
3399  * If n is 0, they match if they are equal. If n is != 0, they match if they
3400  * are different.
3401  */
3402 int
3403 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
3404     struct pf_addr *b, sa_family_t af)
3405 {
3406 	int	match = 0;
3407 
3408 	switch (af) {
3409 #ifdef INET
3410 	case AF_INET:
3411 		if ((a->addr32[0] & m->addr32[0]) ==
3412 		    (b->addr32[0] & m->addr32[0]))
3413 			match++;
3414 		break;
3415 #endif /* INET */
3416 #ifdef INET6
3417 	case AF_INET6:
3418 		if (((a->addr32[0] & m->addr32[0]) ==
3419 		     (b->addr32[0] & m->addr32[0])) &&
3420 		    ((a->addr32[1] & m->addr32[1]) ==
3421 		     (b->addr32[1] & m->addr32[1])) &&
3422 		    ((a->addr32[2] & m->addr32[2]) ==
3423 		     (b->addr32[2] & m->addr32[2])) &&
3424 		    ((a->addr32[3] & m->addr32[3]) ==
3425 		     (b->addr32[3] & m->addr32[3])))
3426 			match++;
3427 		break;
3428 #endif /* INET6 */
3429 	}
3430 	if (match) {
3431 		if (n)
3432 			return (0);
3433 		else
3434 			return (1);
3435 	} else {
3436 		if (n)
3437 			return (1);
3438 		else
3439 			return (0);
3440 	}
3441 }
3442 
3443 /*
3444  * Return 1 if b <= a <= e, otherwise return 0.
3445  */
3446 int
3447 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
3448     struct pf_addr *a, sa_family_t af)
3449 {
3450 	switch (af) {
3451 #ifdef INET
3452 	case AF_INET:
3453 		if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
3454 		    (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
3455 			return (0);
3456 		break;
3457 #endif /* INET */
3458 #ifdef INET6
3459 	case AF_INET6: {
3460 		int	i;
3461 
3462 		/* check a >= b */
3463 		for (i = 0; i < 4; ++i)
3464 			if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
3465 				break;
3466 			else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
3467 				return (0);
3468 		/* check a <= e */
3469 		for (i = 0; i < 4; ++i)
3470 			if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
3471 				break;
3472 			else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
3473 				return (0);
3474 		break;
3475 	}
3476 #endif /* INET6 */
3477 	}
3478 	return (1);
3479 }
3480 
3481 static int
3482 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
3483 {
3484 	switch (op) {
3485 	case PF_OP_IRG:
3486 		return ((p > a1) && (p < a2));
3487 	case PF_OP_XRG:
3488 		return ((p < a1) || (p > a2));
3489 	case PF_OP_RRG:
3490 		return ((p >= a1) && (p <= a2));
3491 	case PF_OP_EQ:
3492 		return (p == a1);
3493 	case PF_OP_NE:
3494 		return (p != a1);
3495 	case PF_OP_LT:
3496 		return (p < a1);
3497 	case PF_OP_LE:
3498 		return (p <= a1);
3499 	case PF_OP_GT:
3500 		return (p > a1);
3501 	case PF_OP_GE:
3502 		return (p >= a1);
3503 	}
3504 	return (0); /* never reached */
3505 }
3506 
3507 int
3508 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
3509 {
3510 	NTOHS(a1);
3511 	NTOHS(a2);
3512 	NTOHS(p);
3513 	return (pf_match(op, a1, a2, p));
3514 }
3515 
3516 static int
3517 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
3518 {
3519 	if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
3520 		return (0);
3521 	return (pf_match(op, a1, a2, u));
3522 }
3523 
3524 static int
3525 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
3526 {
3527 	if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
3528 		return (0);
3529 	return (pf_match(op, a1, a2, g));
3530 }
3531 
3532 int
3533 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag)
3534 {
3535 	if (*tag == -1)
3536 		*tag = mtag;
3537 
3538 	return ((!r->match_tag_not && r->match_tag == *tag) ||
3539 	    (r->match_tag_not && r->match_tag != *tag));
3540 }
3541 
3542 int
3543 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
3544 {
3545 
3546 	KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
3547 
3548 	if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
3549 		return (ENOMEM);
3550 
3551 	pd->pf_mtag->tag = tag;
3552 
3553 	return (0);
3554 }
3555 
3556 #define	PF_ANCHOR_STACKSIZE	32
3557 struct pf_kanchor_stackframe {
3558 	struct pf_kruleset	*rs;
3559 	struct pf_krule		*r;	/* XXX: + match bit */
3560 	struct pf_kanchor	*child;
3561 };
3562 
3563 /*
3564  * XXX: We rely on malloc(9) returning pointer aligned addresses.
3565  */
3566 #define	PF_ANCHORSTACK_MATCH	0x00000001
3567 #define	PF_ANCHORSTACK_MASK	(PF_ANCHORSTACK_MATCH)
3568 
3569 #define	PF_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
3570 #define	PF_ANCHOR_RULE(f)	(struct pf_krule *)			\
3571 				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
3572 #define	PF_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 			\
3573 				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
3574 } while (0)
3575 
3576 void
3577 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth,
3578     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
3579     int *match)
3580 {
3581 	struct pf_kanchor_stackframe	*f;
3582 
3583 	PF_RULES_RASSERT();
3584 
3585 	if (match)
3586 		*match = 0;
3587 	if (*depth >= PF_ANCHOR_STACKSIZE) {
3588 		printf("%s: anchor stack overflow on %s\n",
3589 		    __func__, (*r)->anchor->name);
3590 		*r = TAILQ_NEXT(*r, entries);
3591 		return;
3592 	} else if (*depth == 0 && a != NULL)
3593 		*a = *r;
3594 	f = stack + (*depth)++;
3595 	f->rs = *rs;
3596 	f->r = *r;
3597 	if ((*r)->anchor_wildcard) {
3598 		struct pf_kanchor_node *parent = &(*r)->anchor->children;
3599 
3600 		if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) {
3601 			*r = NULL;
3602 			return;
3603 		}
3604 		*rs = &f->child->ruleset;
3605 	} else {
3606 		f->child = NULL;
3607 		*rs = &(*r)->anchor->ruleset;
3608 	}
3609 	*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
3610 }
3611 
3612 int
3613 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth,
3614     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
3615     int *match)
3616 {
3617 	struct pf_kanchor_stackframe	*f;
3618 	struct pf_krule *fr;
3619 	int quick = 0;
3620 
3621 	PF_RULES_RASSERT();
3622 
3623 	do {
3624 		if (*depth <= 0)
3625 			break;
3626 		f = stack + *depth - 1;
3627 		fr = PF_ANCHOR_RULE(f);
3628 		if (f->child != NULL) {
3629 			/*
3630 			 * This block traverses through
3631 			 * a wildcard anchor.
3632 			 */
3633 			if (match != NULL && *match) {
3634 				/*
3635 				 * If any of "*" matched, then
3636 				 * "foo/ *" matched, mark frame
3637 				 * appropriately.
3638 				 */
3639 				PF_ANCHOR_SET_MATCH(f);
3640 				*match = 0;
3641 			}
3642 			f->child = RB_NEXT(pf_kanchor_node,
3643 			    &fr->anchor->children, f->child);
3644 			if (f->child != NULL) {
3645 				*rs = &f->child->ruleset;
3646 				*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
3647 				if (*r == NULL)
3648 					continue;
3649 				else
3650 					break;
3651 			}
3652 		}
3653 		(*depth)--;
3654 		if (*depth == 0 && a != NULL)
3655 			*a = NULL;
3656 		*rs = f->rs;
3657 		if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
3658 			quick = fr->quick;
3659 		*r = TAILQ_NEXT(fr, entries);
3660 	} while (*r == NULL);
3661 
3662 	return (quick);
3663 }
3664 
3665 struct pf_keth_anchor_stackframe {
3666 	struct pf_keth_ruleset	*rs;
3667 	struct pf_keth_rule	*r;	/* XXX: + match bit */
3668 	struct pf_keth_anchor	*child;
3669 };
3670 
3671 #define	PF_ETH_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
3672 #define	PF_ETH_ANCHOR_RULE(f)	(struct pf_keth_rule *)			\
3673 				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
3674 #define	PF_ETH_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 		\
3675 				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
3676 } while (0)
3677 
3678 void
3679 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth,
3680     struct pf_keth_ruleset **rs, struct pf_keth_rule **r,
3681     struct pf_keth_rule **a, int *match)
3682 {
3683 	struct pf_keth_anchor_stackframe	*f;
3684 
3685 	NET_EPOCH_ASSERT();
3686 
3687 	if (match)
3688 		*match = 0;
3689 	if (*depth >= PF_ANCHOR_STACKSIZE) {
3690 		printf("%s: anchor stack overflow on %s\n",
3691 		    __func__, (*r)->anchor->name);
3692 		*r = TAILQ_NEXT(*r, entries);
3693 		return;
3694 	} else if (*depth == 0 && a != NULL)
3695 		*a = *r;
3696 	f = stack + (*depth)++;
3697 	f->rs = *rs;
3698 	f->r = *r;
3699 	if ((*r)->anchor_wildcard) {
3700 		struct pf_keth_anchor_node *parent = &(*r)->anchor->children;
3701 
3702 		if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) {
3703 			*r = NULL;
3704 			return;
3705 		}
3706 		*rs = &f->child->ruleset;
3707 	} else {
3708 		f->child = NULL;
3709 		*rs = &(*r)->anchor->ruleset;
3710 	}
3711 	*r = TAILQ_FIRST((*rs)->active.rules);
3712 }
3713 
3714 int
3715 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth,
3716     struct pf_keth_ruleset **rs, struct pf_keth_rule **r,
3717     struct pf_keth_rule **a, int *match)
3718 {
3719 	struct pf_keth_anchor_stackframe	*f;
3720 	struct pf_keth_rule *fr;
3721 	int quick = 0;
3722 
3723 	NET_EPOCH_ASSERT();
3724 
3725 	do {
3726 		if (*depth <= 0)
3727 			break;
3728 		f = stack + *depth - 1;
3729 		fr = PF_ETH_ANCHOR_RULE(f);
3730 		if (f->child != NULL) {
3731 			/*
3732 			 * This block traverses through
3733 			 * a wildcard anchor.
3734 			 */
3735 			if (match != NULL && *match) {
3736 				/*
3737 				 * If any of "*" matched, then
3738 				 * "foo/ *" matched, mark frame
3739 				 * appropriately.
3740 				 */
3741 				PF_ETH_ANCHOR_SET_MATCH(f);
3742 				*match = 0;
3743 			}
3744 			f->child = RB_NEXT(pf_keth_anchor_node,
3745 			    &fr->anchor->children, f->child);
3746 			if (f->child != NULL) {
3747 				*rs = &f->child->ruleset;
3748 				*r = TAILQ_FIRST((*rs)->active.rules);
3749 				if (*r == NULL)
3750 					continue;
3751 				else
3752 					break;
3753 			}
3754 		}
3755 		(*depth)--;
3756 		if (*depth == 0 && a != NULL)
3757 			*a = NULL;
3758 		*rs = f->rs;
3759 		if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match))
3760 			quick = fr->quick;
3761 		*r = TAILQ_NEXT(fr, entries);
3762 	} while (*r == NULL);
3763 
3764 	return (quick);
3765 }
3766 
3767 #ifdef INET6
3768 void
3769 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
3770     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
3771 {
3772 	switch (af) {
3773 #ifdef INET
3774 	case AF_INET:
3775 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
3776 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
3777 		break;
3778 #endif /* INET */
3779 	case AF_INET6:
3780 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
3781 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
3782 		naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
3783 		((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
3784 		naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
3785 		((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
3786 		naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
3787 		((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
3788 		break;
3789 	}
3790 }
3791 
3792 void
3793 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
3794 {
3795 	switch (af) {
3796 #ifdef INET
3797 	case AF_INET:
3798 		addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
3799 		break;
3800 #endif /* INET */
3801 	case AF_INET6:
3802 		if (addr->addr32[3] == 0xffffffff) {
3803 			addr->addr32[3] = 0;
3804 			if (addr->addr32[2] == 0xffffffff) {
3805 				addr->addr32[2] = 0;
3806 				if (addr->addr32[1] == 0xffffffff) {
3807 					addr->addr32[1] = 0;
3808 					addr->addr32[0] =
3809 					    htonl(ntohl(addr->addr32[0]) + 1);
3810 				} else
3811 					addr->addr32[1] =
3812 					    htonl(ntohl(addr->addr32[1]) + 1);
3813 			} else
3814 				addr->addr32[2] =
3815 				    htonl(ntohl(addr->addr32[2]) + 1);
3816 		} else
3817 			addr->addr32[3] =
3818 			    htonl(ntohl(addr->addr32[3]) + 1);
3819 		break;
3820 	}
3821 }
3822 #endif /* INET6 */
3823 
3824 void
3825 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a)
3826 {
3827 	/*
3828 	 * Modern rules use the same flags in rules as they do in states.
3829 	 */
3830 	a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID|
3831 	    PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO));
3832 
3833 	/*
3834 	 * Old-style scrub rules have different flags which need to be translated.
3835 	 */
3836 	if (r->rule_flag & PFRULE_RANDOMID)
3837 		a->flags |= PFSTATE_RANDOMID;
3838 	if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) {
3839 		a->flags |= PFSTATE_SETTOS;
3840 		a->set_tos = r->set_tos;
3841 	}
3842 
3843 	if (r->qid)
3844 		a->qid = r->qid;
3845 	if (r->pqid)
3846 		a->pqid = r->pqid;
3847 	if (r->rtableid >= 0)
3848 		a->rtableid = r->rtableid;
3849 	a->log |= r->log;
3850 	if (r->min_ttl)
3851 		a->min_ttl = r->min_ttl;
3852 	if (r->max_mss)
3853 		a->max_mss = r->max_mss;
3854 	if (r->dnpipe)
3855 		a->dnpipe = r->dnpipe;
3856 	if (r->dnrpipe)
3857 		a->dnrpipe = r->dnrpipe;
3858 	if (r->dnpipe || r->dnrpipe) {
3859 		if (r->free_flags & PFRULE_DN_IS_PIPE)
3860 			a->flags |= PFSTATE_DN_IS_PIPE;
3861 		else
3862 			a->flags &= ~PFSTATE_DN_IS_PIPE;
3863 	}
3864 	if (r->scrub_flags & PFSTATE_SETPRIO) {
3865 		a->set_prio[0] = r->set_prio[0];
3866 		a->set_prio[1] = r->set_prio[1];
3867 	}
3868 }
3869 
3870 int
3871 pf_socket_lookup(struct pf_pdesc *pd, struct mbuf *m)
3872 {
3873 	struct pf_addr		*saddr, *daddr;
3874 	u_int16_t		 sport, dport;
3875 	struct inpcbinfo	*pi;
3876 	struct inpcb		*inp;
3877 
3878 	pd->lookup.uid = UID_MAX;
3879 	pd->lookup.gid = GID_MAX;
3880 
3881 	switch (pd->proto) {
3882 	case IPPROTO_TCP:
3883 		sport = pd->hdr.tcp.th_sport;
3884 		dport = pd->hdr.tcp.th_dport;
3885 		pi = &V_tcbinfo;
3886 		break;
3887 	case IPPROTO_UDP:
3888 		sport = pd->hdr.udp.uh_sport;
3889 		dport = pd->hdr.udp.uh_dport;
3890 		pi = &V_udbinfo;
3891 		break;
3892 	default:
3893 		return (-1);
3894 	}
3895 	if (pd->dir == PF_IN) {
3896 		saddr = pd->src;
3897 		daddr = pd->dst;
3898 	} else {
3899 		u_int16_t	p;
3900 
3901 		p = sport;
3902 		sport = dport;
3903 		dport = p;
3904 		saddr = pd->dst;
3905 		daddr = pd->src;
3906 	}
3907 	switch (pd->af) {
3908 #ifdef INET
3909 	case AF_INET:
3910 		inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
3911 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
3912 		if (inp == NULL) {
3913 			inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
3914 			   daddr->v4, dport, INPLOOKUP_WILDCARD |
3915 			   INPLOOKUP_RLOCKPCB, NULL, m);
3916 			if (inp == NULL)
3917 				return (-1);
3918 		}
3919 		break;
3920 #endif /* INET */
3921 #ifdef INET6
3922 	case AF_INET6:
3923 		inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
3924 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
3925 		if (inp == NULL) {
3926 			inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
3927 			    &daddr->v6, dport, INPLOOKUP_WILDCARD |
3928 			    INPLOOKUP_RLOCKPCB, NULL, m);
3929 			if (inp == NULL)
3930 				return (-1);
3931 		}
3932 		break;
3933 #endif /* INET6 */
3934 
3935 	default:
3936 		return (-1);
3937 	}
3938 	INP_RLOCK_ASSERT(inp);
3939 	pd->lookup.uid = inp->inp_cred->cr_uid;
3940 	pd->lookup.gid = inp->inp_cred->cr_groups[0];
3941 	INP_RUNLOCK(inp);
3942 
3943 	return (1);
3944 }
3945 
3946 u_int8_t
3947 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3948 {
3949 	int		 hlen;
3950 	u_int8_t	 hdr[60];
3951 	u_int8_t	*opt, optlen;
3952 	u_int8_t	 wscale = 0;
3953 
3954 	hlen = th_off << 2;		/* hlen <= sizeof(hdr) */
3955 	if (hlen <= sizeof(struct tcphdr))
3956 		return (0);
3957 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3958 		return (0);
3959 	opt = hdr + sizeof(struct tcphdr);
3960 	hlen -= sizeof(struct tcphdr);
3961 	while (hlen >= 3) {
3962 		switch (*opt) {
3963 		case TCPOPT_EOL:
3964 		case TCPOPT_NOP:
3965 			++opt;
3966 			--hlen;
3967 			break;
3968 		case TCPOPT_WINDOW:
3969 			wscale = opt[2];
3970 			if (wscale > TCP_MAX_WINSHIFT)
3971 				wscale = TCP_MAX_WINSHIFT;
3972 			wscale |= PF_WSCALE_FLAG;
3973 			/* FALLTHROUGH */
3974 		default:
3975 			optlen = opt[1];
3976 			if (optlen < 2)
3977 				optlen = 2;
3978 			hlen -= optlen;
3979 			opt += optlen;
3980 			break;
3981 		}
3982 	}
3983 	return (wscale);
3984 }
3985 
3986 u_int16_t
3987 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3988 {
3989 	int		 hlen;
3990 	u_int8_t	 hdr[60];
3991 	u_int8_t	*opt, optlen;
3992 	u_int16_t	 mss = V_tcp_mssdflt;
3993 
3994 	hlen = th_off << 2;	/* hlen <= sizeof(hdr) */
3995 	if (hlen <= sizeof(struct tcphdr))
3996 		return (0);
3997 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3998 		return (0);
3999 	opt = hdr + sizeof(struct tcphdr);
4000 	hlen -= sizeof(struct tcphdr);
4001 	while (hlen >= TCPOLEN_MAXSEG) {
4002 		switch (*opt) {
4003 		case TCPOPT_EOL:
4004 		case TCPOPT_NOP:
4005 			++opt;
4006 			--hlen;
4007 			break;
4008 		case TCPOPT_MAXSEG:
4009 			bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
4010 			NTOHS(mss);
4011 			/* FALLTHROUGH */
4012 		default:
4013 			optlen = opt[1];
4014 			if (optlen < 2)
4015 				optlen = 2;
4016 			hlen -= optlen;
4017 			opt += optlen;
4018 			break;
4019 		}
4020 	}
4021 	return (mss);
4022 }
4023 
4024 static u_int16_t
4025 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
4026 {
4027 	struct nhop_object *nh;
4028 #ifdef INET6
4029 	struct in6_addr		dst6;
4030 	uint32_t		scopeid;
4031 #endif /* INET6 */
4032 	int			 hlen = 0;
4033 	uint16_t		 mss = 0;
4034 
4035 	NET_EPOCH_ASSERT();
4036 
4037 	switch (af) {
4038 #ifdef INET
4039 	case AF_INET:
4040 		hlen = sizeof(struct ip);
4041 		nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0);
4042 		if (nh != NULL)
4043 			mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
4044 		break;
4045 #endif /* INET */
4046 #ifdef INET6
4047 	case AF_INET6:
4048 		hlen = sizeof(struct ip6_hdr);
4049 		in6_splitscope(&addr->v6, &dst6, &scopeid);
4050 		nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0);
4051 		if (nh != NULL)
4052 			mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
4053 		break;
4054 #endif /* INET6 */
4055 	}
4056 
4057 	mss = max(V_tcp_mssdflt, mss);
4058 	mss = min(mss, offer);
4059 	mss = max(mss, 64);		/* sanity - at least max opt space */
4060 	return (mss);
4061 }
4062 
4063 static u_int32_t
4064 pf_tcp_iss(struct pf_pdesc *pd)
4065 {
4066 	MD5_CTX ctx;
4067 	u_int32_t digest[4];
4068 
4069 	if (V_pf_tcp_secret_init == 0) {
4070 		arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
4071 		MD5Init(&V_pf_tcp_secret_ctx);
4072 		MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
4073 		    sizeof(V_pf_tcp_secret));
4074 		V_pf_tcp_secret_init = 1;
4075 	}
4076 
4077 	ctx = V_pf_tcp_secret_ctx;
4078 
4079 	MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short));
4080 	MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short));
4081 	if (pd->af == AF_INET6) {
4082 		MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
4083 		MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
4084 	} else {
4085 		MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
4086 		MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
4087 	}
4088 	MD5Final((u_char *)digest, &ctx);
4089 	V_pf_tcp_iss_off += 4096;
4090 #define	ISN_RANDOM_INCREMENT (4096 - 1)
4091 	return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
4092 	    V_pf_tcp_iss_off);
4093 #undef	ISN_RANDOM_INCREMENT
4094 }
4095 
4096 static bool
4097 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r)
4098 {
4099 	bool match = true;
4100 
4101 	/* Always matches if not set */
4102 	if (! r->isset)
4103 		return (!r->neg);
4104 
4105 	for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4106 		if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) {
4107 			match = false;
4108 			break;
4109 		}
4110 	}
4111 
4112 	return (match ^ r->neg);
4113 }
4114 
4115 static int
4116 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag)
4117 {
4118 	if (*tag == -1)
4119 		*tag = mtag;
4120 
4121 	return ((!r->match_tag_not && r->match_tag == *tag) ||
4122 	    (r->match_tag_not && r->match_tag != *tag));
4123 }
4124 
4125 static void
4126 pf_bridge_to(struct ifnet *ifp, struct mbuf *m)
4127 {
4128 	/* If we don't have the interface drop the packet. */
4129 	if (ifp == NULL) {
4130 		m_freem(m);
4131 		return;
4132 	}
4133 
4134 	switch (ifp->if_type) {
4135 	case IFT_ETHER:
4136 	case IFT_XETHER:
4137 	case IFT_L2VLAN:
4138 	case IFT_BRIDGE:
4139 	case IFT_IEEE8023ADLAG:
4140 		break;
4141 	default:
4142 		m_freem(m);
4143 		return;
4144 	}
4145 
4146 	ifp->if_transmit(ifp, m);
4147 }
4148 
4149 static int
4150 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0)
4151 {
4152 #ifdef INET
4153 	struct ip ip;
4154 #endif
4155 #ifdef INET6
4156 	struct ip6_hdr ip6;
4157 #endif
4158 	struct mbuf *m = *m0;
4159 	struct ether_header *e;
4160 	struct pf_keth_rule *r, *rm, *a = NULL;
4161 	struct pf_keth_ruleset *ruleset = NULL;
4162 	struct pf_mtag *mtag;
4163 	struct pf_keth_ruleq *rules;
4164 	struct pf_addr *src = NULL, *dst = NULL;
4165 	struct pfi_kkif *bridge_to;
4166 	sa_family_t af = 0;
4167 	uint16_t proto;
4168 	int asd = 0, match = 0;
4169 	int tag = -1;
4170 	uint8_t action;
4171 	struct pf_keth_anchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
4172 
4173 	MPASS(kif->pfik_ifp->if_vnet == curvnet);
4174 	NET_EPOCH_ASSERT();
4175 
4176 	PF_RULES_RLOCK_TRACKER;
4177 
4178 	SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m);
4179 
4180 	mtag = pf_find_mtag(m);
4181 	if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
4182 		/* Dummynet re-injects packets after they've
4183 		 * completed their delay. We've already
4184 		 * processed them, so pass unconditionally. */
4185 
4186 		/* But only once. We may see the packet multiple times (e.g.
4187 		 * PFIL_IN/PFIL_OUT). */
4188 		pf_dummynet_flag_remove(m, mtag);
4189 
4190 		return (PF_PASS);
4191 	}
4192 
4193 	ruleset = V_pf_keth;
4194 	rules = ck_pr_load_ptr(&ruleset->active.rules);
4195 	r = TAILQ_FIRST(rules);
4196 	rm = NULL;
4197 
4198 	e = mtod(m, struct ether_header *);
4199 	proto = ntohs(e->ether_type);
4200 
4201 	switch (proto) {
4202 #ifdef INET
4203 	case ETHERTYPE_IP: {
4204 		if (m_length(m, NULL) < (sizeof(struct ether_header) +
4205 		    sizeof(ip)))
4206 			return (PF_DROP);
4207 
4208 		af = AF_INET;
4209 		m_copydata(m, sizeof(struct ether_header), sizeof(ip),
4210 		    (caddr_t)&ip);
4211 		src = (struct pf_addr *)&ip.ip_src;
4212 		dst = (struct pf_addr *)&ip.ip_dst;
4213 		break;
4214 	}
4215 #endif /* INET */
4216 #ifdef INET6
4217 	case ETHERTYPE_IPV6: {
4218 		if (m_length(m, NULL) < (sizeof(struct ether_header) +
4219 		    sizeof(ip6)))
4220 			return (PF_DROP);
4221 
4222 		af = AF_INET6;
4223 		m_copydata(m, sizeof(struct ether_header), sizeof(ip6),
4224 		    (caddr_t)&ip6);
4225 		src = (struct pf_addr *)&ip6.ip6_src;
4226 		dst = (struct pf_addr *)&ip6.ip6_dst;
4227 		break;
4228 	}
4229 #endif /* INET6 */
4230 	}
4231 
4232 	PF_RULES_RLOCK();
4233 
4234 	while (r != NULL) {
4235 		counter_u64_add(r->evaluations, 1);
4236 		SDT_PROBE2(pf, eth, test_rule, test, r->nr, r);
4237 
4238 		if (pfi_kkif_match(r->kif, kif) == r->ifnot) {
4239 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4240 			    "kif");
4241 			r = r->skip[PFE_SKIP_IFP].ptr;
4242 		}
4243 		else if (r->direction && r->direction != dir) {
4244 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4245 			    "dir");
4246 			r = r->skip[PFE_SKIP_DIR].ptr;
4247 		}
4248 		else if (r->proto && r->proto != proto) {
4249 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4250 			    "proto");
4251 			r = r->skip[PFE_SKIP_PROTO].ptr;
4252 		}
4253 		else if (! pf_match_eth_addr(e->ether_shost, &r->src)) {
4254 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4255 			    "src");
4256 			r = r->skip[PFE_SKIP_SRC_ADDR].ptr;
4257 		}
4258 		else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) {
4259 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4260 			    "dst");
4261 			r = r->skip[PFE_SKIP_DST_ADDR].ptr;
4262 		}
4263 		else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af,
4264 		    r->ipsrc.neg, kif, M_GETFIB(m))) {
4265 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4266 			    "ip_src");
4267 			r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr;
4268 		}
4269 		else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af,
4270 		    r->ipdst.neg, kif, M_GETFIB(m))) {
4271 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4272 			    "ip_dst");
4273 			r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr;
4274 		}
4275 		else if (r->match_tag && !pf_match_eth_tag(m, r, &tag,
4276 		    mtag ? mtag->tag : 0)) {
4277 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4278 			    "match_tag");
4279 			r = TAILQ_NEXT(r, entries);
4280 		}
4281 		else {
4282 			if (r->tag)
4283 				tag = r->tag;
4284 			if (r->anchor == NULL) {
4285 				/* Rule matches */
4286 				rm = r;
4287 
4288 				SDT_PROBE2(pf, eth, test_rule, match, r->nr, r);
4289 
4290 				if (r->quick)
4291 					break;
4292 
4293 				r = TAILQ_NEXT(r, entries);
4294 			} else {
4295 				pf_step_into_keth_anchor(anchor_stack, &asd,
4296 				    &ruleset, &r, &a, &match);
4297 			}
4298 		}
4299 		if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd,
4300 		    &ruleset, &r, &a, &match))
4301 			break;
4302 	}
4303 
4304 	r = rm;
4305 
4306 	SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r);
4307 
4308 	/* Default to pass. */
4309 	if (r == NULL) {
4310 		PF_RULES_RUNLOCK();
4311 		return (PF_PASS);
4312 	}
4313 
4314 	/* Execute action. */
4315 	counter_u64_add(r->packets[dir == PF_OUT], 1);
4316 	counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL));
4317 	pf_update_timestamp(r);
4318 
4319 	/* Shortcut. Don't tag if we're just going to drop anyway. */
4320 	if (r->action == PF_DROP) {
4321 		PF_RULES_RUNLOCK();
4322 		return (PF_DROP);
4323 	}
4324 
4325 	if (tag > 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->tag = tag;
4334 	}
4335 
4336 	if (r->qid != 0) {
4337 		if (mtag == NULL)
4338 			mtag = pf_get_mtag(m);
4339 		if (mtag == NULL) {
4340 			PF_RULES_RUNLOCK();
4341 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4342 			return (PF_DROP);
4343 		}
4344 		mtag->qid = r->qid;
4345 	}
4346 
4347 	action = r->action;
4348 	bridge_to = r->bridge_to;
4349 
4350 	/* Dummynet */
4351 	if (r->dnpipe) {
4352 		struct ip_fw_args dnflow;
4353 
4354 		/* Drop packet if dummynet is not loaded. */
4355 		if (ip_dn_io_ptr == NULL) {
4356 			PF_RULES_RUNLOCK();
4357 			m_freem(m);
4358 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4359 			return (PF_DROP);
4360 		}
4361 		if (mtag == NULL)
4362 			mtag = pf_get_mtag(m);
4363 		if (mtag == NULL) {
4364 			PF_RULES_RUNLOCK();
4365 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4366 			return (PF_DROP);
4367 		}
4368 
4369 		bzero(&dnflow, sizeof(dnflow));
4370 
4371 		/* We don't have port numbers here, so we set 0.  That means
4372 		 * that we'll be somewhat limited in distinguishing flows (i.e.
4373 		 * only based on IP addresses, not based on port numbers), but
4374 		 * it's better than nothing. */
4375 		dnflow.f_id.dst_port = 0;
4376 		dnflow.f_id.src_port = 0;
4377 		dnflow.f_id.proto = 0;
4378 
4379 		dnflow.rule.info = r->dnpipe;
4380 		dnflow.rule.info |= IPFW_IS_DUMMYNET;
4381 		if (r->dnflags & PFRULE_DN_IS_PIPE)
4382 			dnflow.rule.info |= IPFW_IS_PIPE;
4383 
4384 		dnflow.f_id.extra = dnflow.rule.info;
4385 
4386 		dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT;
4387 		dnflow.flags |= IPFW_ARGS_ETHER;
4388 		dnflow.ifp = kif->pfik_ifp;
4389 
4390 		switch (af) {
4391 		case AF_INET:
4392 			dnflow.f_id.addr_type = 4;
4393 			dnflow.f_id.src_ip = src->v4.s_addr;
4394 			dnflow.f_id.dst_ip = dst->v4.s_addr;
4395 			break;
4396 		case AF_INET6:
4397 			dnflow.flags |= IPFW_ARGS_IP6;
4398 			dnflow.f_id.addr_type = 6;
4399 			dnflow.f_id.src_ip6 = src->v6;
4400 			dnflow.f_id.dst_ip6 = dst->v6;
4401 			break;
4402 		}
4403 
4404 		PF_RULES_RUNLOCK();
4405 
4406 		mtag->flags |= PF_MTAG_FLAG_DUMMYNET;
4407 		ip_dn_io_ptr(m0, &dnflow);
4408 		if (*m0 != NULL)
4409 			pf_dummynet_flag_remove(m, mtag);
4410 	} else {
4411 		PF_RULES_RUNLOCK();
4412 	}
4413 
4414 	if (action == PF_PASS && bridge_to) {
4415 		pf_bridge_to(bridge_to->pfik_ifp, *m0);
4416 		*m0 = NULL; /* We've eaten the packet. */
4417 	}
4418 
4419 	return (action);
4420 }
4421 
4422 static int
4423 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, struct pfi_kkif *kif,
4424     struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_krule **am,
4425     struct pf_kruleset **rsm, struct inpcb *inp)
4426 {
4427 	struct pf_krule		*nr = NULL;
4428 	struct pf_addr		* const saddr = pd->src;
4429 	struct pf_addr		* const daddr = pd->dst;
4430 	sa_family_t		 af = pd->af;
4431 	struct pf_krule		*r, *a = NULL;
4432 	struct pf_kruleset	*ruleset = NULL;
4433 	struct pf_krule_slist	 match_rules;
4434 	struct pf_krule_item	*ri;
4435 	struct pf_ksrc_node	*nsn = NULL;
4436 	struct tcphdr		*th = &pd->hdr.tcp;
4437 	struct pf_state_key	*sk = NULL, *nk = NULL;
4438 	u_short			 reason;
4439 	int			 rewrite = 0, hdrlen = 0;
4440 	int			 tag = -1;
4441 	int			 asd = 0;
4442 	int			 match = 0;
4443 	int			 state_icmp = 0;
4444 	u_int16_t		 sport = 0, dport = 0;
4445 	u_int16_t		 bproto_sum = 0, bip_sum = 0;
4446 	u_int8_t		 icmptype = 0, icmpcode = 0;
4447 	struct pf_kanchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
4448 
4449 	PF_RULES_RASSERT();
4450 
4451 	if (inp != NULL) {
4452 		INP_LOCK_ASSERT(inp);
4453 		pd->lookup.uid = inp->inp_cred->cr_uid;
4454 		pd->lookup.gid = inp->inp_cred->cr_groups[0];
4455 		pd->lookup.done = 1;
4456 	}
4457 
4458 	switch (pd->proto) {
4459 	case IPPROTO_TCP:
4460 		sport = th->th_sport;
4461 		dport = th->th_dport;
4462 		hdrlen = sizeof(*th);
4463 		break;
4464 	case IPPROTO_UDP:
4465 		sport = pd->hdr.udp.uh_sport;
4466 		dport = pd->hdr.udp.uh_dport;
4467 		hdrlen = sizeof(pd->hdr.udp);
4468 		break;
4469 	case IPPROTO_SCTP:
4470 		sport = pd->hdr.sctp.src_port;
4471 		dport = pd->hdr.sctp.dest_port;
4472 		hdrlen = sizeof(pd->hdr.sctp);
4473 		break;
4474 #ifdef INET
4475 	case IPPROTO_ICMP:
4476 		if (pd->af != AF_INET)
4477 			break;
4478 		sport = dport = pd->hdr.icmp.icmp_id;
4479 		hdrlen = sizeof(pd->hdr.icmp);
4480 		icmptype = pd->hdr.icmp.icmp_type;
4481 		icmpcode = pd->hdr.icmp.icmp_code;
4482 
4483 		if (icmptype == ICMP_UNREACH ||
4484 		    icmptype == ICMP_SOURCEQUENCH ||
4485 		    icmptype == ICMP_REDIRECT ||
4486 		    icmptype == ICMP_TIMXCEED ||
4487 		    icmptype == ICMP_PARAMPROB)
4488 			state_icmp++;
4489 		break;
4490 #endif /* INET */
4491 #ifdef INET6
4492 	case IPPROTO_ICMPV6:
4493 		if (af != AF_INET6)
4494 			break;
4495 		sport = dport = pd->hdr.icmp6.icmp6_id;
4496 		hdrlen = sizeof(pd->hdr.icmp6);
4497 		icmptype = pd->hdr.icmp6.icmp6_type;
4498 		icmpcode = pd->hdr.icmp6.icmp6_code;
4499 
4500 		if (icmptype == ICMP6_DST_UNREACH ||
4501 		    icmptype == ICMP6_PACKET_TOO_BIG ||
4502 		    icmptype == ICMP6_TIME_EXCEEDED ||
4503 		    icmptype == ICMP6_PARAM_PROB)
4504 			state_icmp++;
4505 		break;
4506 #endif /* INET6 */
4507 	default:
4508 		sport = dport = hdrlen = 0;
4509 		break;
4510 	}
4511 
4512 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
4513 
4514 	/* check packet for BINAT/NAT/RDR */
4515 	if ((nr = pf_get_translation(pd, m, off, kif, &nsn, &sk,
4516 	    &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
4517 		KASSERT(sk != NULL, ("%s: null sk", __func__));
4518 		KASSERT(nk != NULL, ("%s: null nk", __func__));
4519 
4520 		if (nr->log) {
4521 			PFLOG_PACKET(kif, m, af, PF_PASS, PFRES_MATCH, nr, a,
4522 			    ruleset, pd, 1);
4523 		}
4524 
4525 		if (pd->ip_sum)
4526 			bip_sum = *pd->ip_sum;
4527 
4528 		switch (pd->proto) {
4529 		case IPPROTO_TCP:
4530 			bproto_sum = th->th_sum;
4531 			pd->proto_sum = &th->th_sum;
4532 
4533 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
4534 			    nk->port[pd->sidx] != sport) {
4535 				pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
4536 				    &th->th_sum, &nk->addr[pd->sidx],
4537 				    nk->port[pd->sidx], 0, af);
4538 				pd->sport = &th->th_sport;
4539 				sport = th->th_sport;
4540 			}
4541 
4542 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
4543 			    nk->port[pd->didx] != dport) {
4544 				pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
4545 				    &th->th_sum, &nk->addr[pd->didx],
4546 				    nk->port[pd->didx], 0, af);
4547 				dport = th->th_dport;
4548 				pd->dport = &th->th_dport;
4549 			}
4550 			rewrite++;
4551 			break;
4552 		case IPPROTO_UDP:
4553 			bproto_sum = pd->hdr.udp.uh_sum;
4554 			pd->proto_sum = &pd->hdr.udp.uh_sum;
4555 
4556 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
4557 			    nk->port[pd->sidx] != sport) {
4558 				pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport,
4559 				    pd->ip_sum, &pd->hdr.udp.uh_sum,
4560 				    &nk->addr[pd->sidx],
4561 				    nk->port[pd->sidx], 1, af);
4562 				sport = pd->hdr.udp.uh_sport;
4563 				pd->sport = &pd->hdr.udp.uh_sport;
4564 			}
4565 
4566 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
4567 			    nk->port[pd->didx] != dport) {
4568 				pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport,
4569 				    pd->ip_sum, &pd->hdr.udp.uh_sum,
4570 				    &nk->addr[pd->didx],
4571 				    nk->port[pd->didx], 1, af);
4572 				dport = pd->hdr.udp.uh_dport;
4573 				pd->dport = &pd->hdr.udp.uh_dport;
4574 			}
4575 			rewrite++;
4576 			break;
4577 		case IPPROTO_SCTP: {
4578 			uint16_t checksum = 0;
4579 
4580 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
4581 			    nk->port[pd->sidx] != sport) {
4582 				pf_change_ap(m, saddr, &pd->hdr.sctp.src_port,
4583 				    pd->ip_sum, &checksum,
4584 				    &nk->addr[pd->sidx],
4585 				    nk->port[pd->sidx], 1, af);
4586 			}
4587 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
4588 			    nk->port[pd->didx] != dport) {
4589 				pf_change_ap(m, daddr, &pd->hdr.sctp.dest_port,
4590 				    pd->ip_sum, &checksum,
4591 				    &nk->addr[pd->didx],
4592 				    nk->port[pd->didx], 1, af);
4593 			}
4594 			break;
4595 		}
4596 #ifdef INET
4597 		case IPPROTO_ICMP:
4598 			nk->port[0] = nk->port[1];
4599 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
4600 				pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
4601 				    nk->addr[pd->sidx].v4.s_addr, 0);
4602 
4603 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
4604 				pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
4605 				    nk->addr[pd->didx].v4.s_addr, 0);
4606 
4607 			if (nk->port[1] != pd->hdr.icmp.icmp_id) {
4608 				pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
4609 				    pd->hdr.icmp.icmp_cksum, sport,
4610 				    nk->port[1], 0);
4611 				pd->hdr.icmp.icmp_id = nk->port[1];
4612 				pd->sport = &pd->hdr.icmp.icmp_id;
4613 			}
4614 			m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
4615 			break;
4616 #endif /* INET */
4617 #ifdef INET6
4618 		case IPPROTO_ICMPV6:
4619 			nk->port[0] = nk->port[1];
4620 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
4621 				pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum,
4622 				    &nk->addr[pd->sidx], 0);
4623 
4624 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
4625 				pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum,
4626 				    &nk->addr[pd->didx], 0);
4627 			rewrite++;
4628 			break;
4629 #endif /* INET */
4630 		default:
4631 			switch (af) {
4632 #ifdef INET
4633 			case AF_INET:
4634 				if (PF_ANEQ(saddr,
4635 				    &nk->addr[pd->sidx], AF_INET))
4636 					pf_change_a(&saddr->v4.s_addr,
4637 					    pd->ip_sum,
4638 					    nk->addr[pd->sidx].v4.s_addr, 0);
4639 
4640 				if (PF_ANEQ(daddr,
4641 				    &nk->addr[pd->didx], AF_INET))
4642 					pf_change_a(&daddr->v4.s_addr,
4643 					    pd->ip_sum,
4644 					    nk->addr[pd->didx].v4.s_addr, 0);
4645 				break;
4646 #endif /* INET */
4647 #ifdef INET6
4648 			case AF_INET6:
4649 				if (PF_ANEQ(saddr,
4650 				    &nk->addr[pd->sidx], AF_INET6))
4651 					PF_ACPY(saddr, &nk->addr[pd->sidx], af);
4652 
4653 				if (PF_ANEQ(daddr,
4654 				    &nk->addr[pd->didx], AF_INET6))
4655 					PF_ACPY(daddr, &nk->addr[pd->didx], af);
4656 				break;
4657 #endif /* INET */
4658 			}
4659 			break;
4660 		}
4661 		if (nr->natpass)
4662 			r = NULL;
4663 		pd->nat_rule = nr;
4664 	}
4665 
4666 	SLIST_INIT(&match_rules);
4667 	while (r != NULL) {
4668 		pf_counter_u64_add(&r->evaluations, 1);
4669 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
4670 			r = r->skip[PF_SKIP_IFP].ptr;
4671 		else if (r->direction && r->direction != pd->dir)
4672 			r = r->skip[PF_SKIP_DIR].ptr;
4673 		else if (r->af && r->af != af)
4674 			r = r->skip[PF_SKIP_AF].ptr;
4675 		else if (r->proto && r->proto != pd->proto)
4676 			r = r->skip[PF_SKIP_PROTO].ptr;
4677 		else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
4678 		    r->src.neg, kif, M_GETFIB(m)))
4679 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
4680 		/* tcp/udp only. port_op always 0 in other cases */
4681 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
4682 		    r->src.port[0], r->src.port[1], sport))
4683 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
4684 		else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
4685 		    r->dst.neg, NULL, M_GETFIB(m)))
4686 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
4687 		/* tcp/udp only. port_op always 0 in other cases */
4688 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
4689 		    r->dst.port[0], r->dst.port[1], dport))
4690 			r = r->skip[PF_SKIP_DST_PORT].ptr;
4691 		/* icmp only. type always 0 in other cases */
4692 		else if (r->type && r->type != icmptype + 1)
4693 			r = TAILQ_NEXT(r, entries);
4694 		/* icmp only. type always 0 in other cases */
4695 		else if (r->code && r->code != icmpcode + 1)
4696 			r = TAILQ_NEXT(r, entries);
4697 		else if (r->tos && !(r->tos == pd->tos))
4698 			r = TAILQ_NEXT(r, entries);
4699 		else if (r->rule_flag & PFRULE_FRAGMENT)
4700 			r = TAILQ_NEXT(r, entries);
4701 		else if (pd->proto == IPPROTO_TCP &&
4702 		    (r->flagset & th->th_flags) != r->flags)
4703 			r = TAILQ_NEXT(r, entries);
4704 		/* tcp/udp only. uid.op always 0 in other cases */
4705 		else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
4706 		    pf_socket_lookup(pd, m), 1)) &&
4707 		    !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
4708 		    pd->lookup.uid))
4709 			r = TAILQ_NEXT(r, entries);
4710 		/* tcp/udp only. gid.op always 0 in other cases */
4711 		else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
4712 		    pf_socket_lookup(pd, m), 1)) &&
4713 		    !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
4714 		    pd->lookup.gid))
4715 			r = TAILQ_NEXT(r, entries);
4716 		else if (r->prio &&
4717 		    !pf_match_ieee8021q_pcp(r->prio, m))
4718 			r = TAILQ_NEXT(r, entries);
4719 		else if (r->prob &&
4720 		    r->prob <= arc4random())
4721 			r = TAILQ_NEXT(r, entries);
4722 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
4723 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
4724 			r = TAILQ_NEXT(r, entries);
4725 		else if (r->os_fingerprint != PF_OSFP_ANY &&
4726 		    (pd->proto != IPPROTO_TCP || !pf_osfp_match(
4727 		    pf_osfp_fingerprint(pd, m, off, th),
4728 		    r->os_fingerprint)))
4729 			r = TAILQ_NEXT(r, entries);
4730 		else {
4731 			if (r->tag)
4732 				tag = r->tag;
4733 			if (r->anchor == NULL) {
4734 				if (r->action == PF_MATCH) {
4735 					ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO);
4736 					if (ri == NULL) {
4737 						REASON_SET(&reason, PFRES_MEMORY);
4738 						goto cleanup;
4739 					}
4740 					ri->r = r;
4741 					SLIST_INSERT_HEAD(&match_rules, ri, entry);
4742 					pf_counter_u64_critical_enter();
4743 					pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
4744 					pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
4745 					pf_counter_u64_critical_exit();
4746 					pf_rule_to_actions(r, &pd->act);
4747 					if (r->log)
4748 						PFLOG_PACKET(kif, m, af,
4749 						    r->action, PFRES_MATCH, r,
4750 						    a, ruleset, pd, 1);
4751 				} else {
4752 					match = 1;
4753 					*rm = r;
4754 					*am = a;
4755 					*rsm = ruleset;
4756 				}
4757 				if ((*rm)->quick)
4758 					break;
4759 				r = TAILQ_NEXT(r, entries);
4760 			} else
4761 				pf_step_into_anchor(anchor_stack, &asd,
4762 				    &ruleset, PF_RULESET_FILTER, &r, &a,
4763 				    &match);
4764 		}
4765 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
4766 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
4767 			break;
4768 	}
4769 	r = *rm;
4770 	a = *am;
4771 	ruleset = *rsm;
4772 
4773 	REASON_SET(&reason, PFRES_MATCH);
4774 
4775 	/* apply actions for last matching pass/block rule */
4776 	pf_rule_to_actions(r, &pd->act);
4777 
4778 	if (r->log) {
4779 		if (rewrite)
4780 			m_copyback(m, off, hdrlen, pd->hdr.any);
4781 		PFLOG_PACKET(kif, m, af, r->action, reason, r, a, ruleset, pd, 1);
4782 	}
4783 
4784 	if ((r->action == PF_DROP) &&
4785 	    ((r->rule_flag & PFRULE_RETURNRST) ||
4786 	    (r->rule_flag & PFRULE_RETURNICMP) ||
4787 	    (r->rule_flag & PFRULE_RETURN))) {
4788 		pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum,
4789 		    bip_sum, hdrlen, &reason, r->rtableid);
4790 	}
4791 
4792 	if (r->action == PF_DROP)
4793 		goto cleanup;
4794 
4795 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
4796 		REASON_SET(&reason, PFRES_MEMORY);
4797 		goto cleanup;
4798 	}
4799 	if (pd->act.rtableid >= 0)
4800 		M_SETFIB(m, pd->act.rtableid);
4801 
4802 	if (!state_icmp && (r->keep_state || nr != NULL ||
4803 	    (pd->flags & PFDESC_TCP_NORM))) {
4804 		int action;
4805 		action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
4806 		    sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
4807 		    hdrlen, &match_rules);
4808 		if (action != PF_PASS) {
4809 			if (action == PF_DROP &&
4810 			    (r->rule_flag & PFRULE_RETURN))
4811 				pf_return(r, nr, pd, sk, off, m, th, kif,
4812 				    bproto_sum, bip_sum, hdrlen, &reason,
4813 				    pd->act.rtableid);
4814 			return (action);
4815 		}
4816 	} else {
4817 		while ((ri = SLIST_FIRST(&match_rules))) {
4818 			SLIST_REMOVE_HEAD(&match_rules, entry);
4819 			free(ri, M_PF_RULE_ITEM);
4820 		}
4821 
4822 		uma_zfree(V_pf_state_key_z, sk);
4823 		uma_zfree(V_pf_state_key_z, nk);
4824 	}
4825 
4826 	/* copy back packet headers if we performed NAT operations */
4827 	if (rewrite)
4828 		m_copyback(m, off, hdrlen, pd->hdr.any);
4829 
4830 	if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
4831 	    pd->dir == PF_OUT &&
4832 	    V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m))
4833 		/*
4834 		 * We want the state created, but we dont
4835 		 * want to send this in case a partner
4836 		 * firewall has to know about it to allow
4837 		 * replies through it.
4838 		 */
4839 		return (PF_DEFER);
4840 
4841 	return (PF_PASS);
4842 
4843 cleanup:
4844 	while ((ri = SLIST_FIRST(&match_rules))) {
4845 		SLIST_REMOVE_HEAD(&match_rules, entry);
4846 		free(ri, M_PF_RULE_ITEM);
4847 	}
4848 
4849 	uma_zfree(V_pf_state_key_z, sk);
4850 	uma_zfree(V_pf_state_key_z, nk);
4851 	return (PF_DROP);
4852 }
4853 
4854 static int
4855 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a,
4856     struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk,
4857     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
4858     u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm,
4859     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
4860     struct pf_krule_slist *match_rules)
4861 {
4862 	struct pf_kstate	*s = NULL;
4863 	struct pf_ksrc_node	*sn = NULL;
4864 	struct tcphdr		*th = &pd->hdr.tcp;
4865 	u_int16_t		 mss = V_tcp_mssdflt;
4866 	u_short			 reason, sn_reason;
4867 	struct pf_krule_item	*ri;
4868 
4869 	/* check maximums */
4870 	if (r->max_states &&
4871 	    (counter_u64_fetch(r->states_cur) >= r->max_states)) {
4872 		counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
4873 		REASON_SET(&reason, PFRES_MAXSTATES);
4874 		goto csfailed;
4875 	}
4876 	/* src node for filter rule */
4877 	if ((r->rule_flag & PFRULE_SRCTRACK ||
4878 	    r->rpool.opts & PF_POOL_STICKYADDR) &&
4879 	    (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) {
4880 		REASON_SET(&reason, sn_reason);
4881 		goto csfailed;
4882 	}
4883 	/* src node for translation rule */
4884 	if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
4885 	    (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx],
4886 	    pd->af)) != 0 ) {
4887 		REASON_SET(&reason, sn_reason);
4888 		goto csfailed;
4889 	}
4890 	s = pf_alloc_state(M_NOWAIT);
4891 	if (s == NULL) {
4892 		REASON_SET(&reason, PFRES_MEMORY);
4893 		goto csfailed;
4894 	}
4895 	s->rule.ptr = r;
4896 	s->nat_rule.ptr = nr;
4897 	s->anchor.ptr = a;
4898 	bcopy(match_rules, &s->match_rules, sizeof(s->match_rules));
4899 	memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions));
4900 
4901 	STATE_INC_COUNTERS(s);
4902 	if (r->allow_opts)
4903 		s->state_flags |= PFSTATE_ALLOWOPTS;
4904 	if (r->rule_flag & PFRULE_STATESLOPPY)
4905 		s->state_flags |= PFSTATE_SLOPPY;
4906 	if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */
4907 		s->state_flags |= PFSTATE_SCRUB_TCP;
4908 	if ((r->rule_flag & PFRULE_PFLOW) ||
4909 	    (nr != NULL && nr->rule_flag & PFRULE_PFLOW))
4910 		s->state_flags |= PFSTATE_PFLOW;
4911 
4912 	s->act.log = pd->act.log & PF_LOG_ALL;
4913 	s->sync_state = PFSYNC_S_NONE;
4914 	s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */
4915 
4916 	if (nr != NULL)
4917 		s->act.log |= nr->log & PF_LOG_ALL;
4918 	switch (pd->proto) {
4919 	case IPPROTO_TCP:
4920 		s->src.seqlo = ntohl(th->th_seq);
4921 		s->src.seqhi = s->src.seqlo + pd->p_len + 1;
4922 		if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
4923 		    r->keep_state == PF_STATE_MODULATE) {
4924 			/* Generate sequence number modulator */
4925 			if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
4926 			    0)
4927 				s->src.seqdiff = 1;
4928 			pf_change_proto_a(m, &th->th_seq, &th->th_sum,
4929 			    htonl(s->src.seqlo + s->src.seqdiff), 0);
4930 			*rewrite = 1;
4931 		} else
4932 			s->src.seqdiff = 0;
4933 		if (th->th_flags & TH_SYN) {
4934 			s->src.seqhi++;
4935 			s->src.wscale = pf_get_wscale(m, off,
4936 			    th->th_off, pd->af);
4937 		}
4938 		s->src.max_win = MAX(ntohs(th->th_win), 1);
4939 		if (s->src.wscale & PF_WSCALE_MASK) {
4940 			/* Remove scale factor from initial window */
4941 			int win = s->src.max_win;
4942 			win += 1 << (s->src.wscale & PF_WSCALE_MASK);
4943 			s->src.max_win = (win - 1) >>
4944 			    (s->src.wscale & PF_WSCALE_MASK);
4945 		}
4946 		if (th->th_flags & TH_FIN)
4947 			s->src.seqhi++;
4948 		s->dst.seqhi = 1;
4949 		s->dst.max_win = 1;
4950 		pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT);
4951 		pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED);
4952 		s->timeout = PFTM_TCP_FIRST_PACKET;
4953 		atomic_add_32(&V_pf_status.states_halfopen, 1);
4954 		break;
4955 	case IPPROTO_UDP:
4956 		pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE);
4957 		pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC);
4958 		s->timeout = PFTM_UDP_FIRST_PACKET;
4959 		break;
4960 	case IPPROTO_SCTP:
4961 		pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT);
4962 		pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED);
4963 		s->timeout = PFTM_SCTP_FIRST_PACKET;
4964 		break;
4965 	case IPPROTO_ICMP:
4966 #ifdef INET6
4967 	case IPPROTO_ICMPV6:
4968 #endif
4969 		s->timeout = PFTM_ICMP_FIRST_PACKET;
4970 		break;
4971 	default:
4972 		pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE);
4973 		pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC);
4974 		s->timeout = PFTM_OTHER_FIRST_PACKET;
4975 	}
4976 
4977 	if (r->rt) {
4978 		/* pf_map_addr increases the reason counters */
4979 		if ((reason = pf_map_addr(pd->af, r, pd->src, &s->rt_addr,
4980 		    &s->rt_kif, NULL, &sn)) != 0)
4981 			goto csfailed;
4982 		s->rt = r->rt;
4983 	}
4984 
4985 	s->creation = s->expire = pf_get_uptime();
4986 
4987 	if (sn != NULL)
4988 		s->src_node = sn;
4989 	if (nsn != NULL) {
4990 		/* XXX We only modify one side for now. */
4991 		PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
4992 		s->nat_src_node = nsn;
4993 	}
4994 	if (pd->proto == IPPROTO_TCP) {
4995 		if (s->state_flags & PFSTATE_SCRUB_TCP &&
4996 		    pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) {
4997 			REASON_SET(&reason, PFRES_MEMORY);
4998 			goto drop;
4999 		}
5000 		if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub &&
5001 		    pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
5002 		    &s->src, &s->dst, rewrite)) {
5003 			/* This really shouldn't happen!!! */
5004 			DPFPRINTF(PF_DEBUG_URGENT,
5005 			    ("pf_normalize_tcp_stateful failed on first "
5006 			     "pkt\n"));
5007 			goto drop;
5008 		}
5009 	} else if (pd->proto == IPPROTO_SCTP) {
5010 		if (pf_normalize_sctp_init(m, off, pd, &s->src, &s->dst))
5011 			goto drop;
5012 		if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP)))
5013 			goto drop;
5014 	}
5015 	s->direction = pd->dir;
5016 
5017 	/*
5018 	 * sk/nk could already been setup by pf_get_translation().
5019 	 */
5020 	if (nr == NULL) {
5021 		KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
5022 		    __func__, nr, sk, nk));
5023 		sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
5024 		if (sk == NULL)
5025 			goto csfailed;
5026 		nk = sk;
5027 	} else
5028 		KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
5029 		    __func__, nr, sk, nk));
5030 
5031 	/* Swap sk/nk for PF_OUT. */
5032 	if (pf_state_insert(BOUND_IFACE(s, kif), kif,
5033 	    (pd->dir == PF_IN) ? sk : nk,
5034 	    (pd->dir == PF_IN) ? nk : sk, s)) {
5035 		REASON_SET(&reason, PFRES_STATEINS);
5036 		goto drop;
5037 	} else
5038 		*sm = s;
5039 
5040 	if (tag > 0)
5041 		s->tag = tag;
5042 	if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
5043 	    TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
5044 		pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC);
5045 		/* undo NAT changes, if they have taken place */
5046 		if (nr != NULL) {
5047 			struct pf_state_key *skt = s->key[PF_SK_WIRE];
5048 			if (pd->dir == PF_OUT)
5049 				skt = s->key[PF_SK_STACK];
5050 			PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
5051 			PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
5052 			if (pd->sport)
5053 				*pd->sport = skt->port[pd->sidx];
5054 			if (pd->dport)
5055 				*pd->dport = skt->port[pd->didx];
5056 			if (pd->proto_sum)
5057 				*pd->proto_sum = bproto_sum;
5058 			if (pd->ip_sum)
5059 				*pd->ip_sum = bip_sum;
5060 			m_copyback(m, off, hdrlen, pd->hdr.any);
5061 		}
5062 		s->src.seqhi = htonl(arc4random());
5063 		/* Find mss option */
5064 		int rtid = M_GETFIB(m);
5065 		mss = pf_get_mss(m, off, th->th_off, pd->af);
5066 		mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
5067 		mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
5068 		s->src.mss = mss;
5069 		pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport,
5070 		    th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
5071 		    TH_SYN|TH_ACK, 0, s->src.mss, 0, true, 0, 0,
5072 		    pd->act.rtableid);
5073 		REASON_SET(&reason, PFRES_SYNPROXY);
5074 		return (PF_SYNPROXY_DROP);
5075 	}
5076 
5077 	return (PF_PASS);
5078 
5079 csfailed:
5080 	while ((ri = SLIST_FIRST(match_rules))) {
5081 		SLIST_REMOVE_HEAD(match_rules, entry);
5082 		free(ri, M_PF_RULE_ITEM);
5083 	}
5084 
5085 	uma_zfree(V_pf_state_key_z, sk);
5086 	uma_zfree(V_pf_state_key_z, nk);
5087 
5088 	if (sn != NULL) {
5089 		PF_SRC_NODE_LOCK(sn);
5090 		if (--sn->states == 0 && sn->expire == 0) {
5091 			pf_unlink_src_node(sn);
5092 			uma_zfree(V_pf_sources_z, sn);
5093 			counter_u64_add(
5094 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
5095 		}
5096 		PF_SRC_NODE_UNLOCK(sn);
5097 	}
5098 
5099 	if (nsn != sn && nsn != NULL) {
5100 		PF_SRC_NODE_LOCK(nsn);
5101 		if (--nsn->states == 0 && nsn->expire == 0) {
5102 			pf_unlink_src_node(nsn);
5103 			uma_zfree(V_pf_sources_z, nsn);
5104 			counter_u64_add(
5105 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
5106 		}
5107 		PF_SRC_NODE_UNLOCK(nsn);
5108 	}
5109 
5110 drop:
5111 	if (s != NULL) {
5112 		pf_src_tree_remove_state(s);
5113 		s->timeout = PFTM_UNLINKED;
5114 		STATE_DEC_COUNTERS(s);
5115 		pf_free_state(s);
5116 	}
5117 
5118 	return (PF_DROP);
5119 }
5120 
5121 static int
5122 pf_test_fragment(struct pf_krule **rm, struct pfi_kkif *kif,
5123     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am,
5124     struct pf_kruleset **rsm)
5125 {
5126 	struct pf_krule		*r, *a = NULL;
5127 	struct pf_kruleset	*ruleset = NULL;
5128 	struct pf_krule_slist	 match_rules;
5129 	struct pf_krule_item	*ri;
5130 	sa_family_t		 af = pd->af;
5131 	u_short			 reason;
5132 	int			 tag = -1;
5133 	int			 asd = 0;
5134 	int			 match = 0;
5135 	struct pf_kanchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
5136 
5137 	PF_RULES_RASSERT();
5138 
5139 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
5140 	SLIST_INIT(&match_rules);
5141 	while (r != NULL) {
5142 		pf_counter_u64_add(&r->evaluations, 1);
5143 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
5144 			r = r->skip[PF_SKIP_IFP].ptr;
5145 		else if (r->direction && r->direction != pd->dir)
5146 			r = r->skip[PF_SKIP_DIR].ptr;
5147 		else if (r->af && r->af != af)
5148 			r = r->skip[PF_SKIP_AF].ptr;
5149 		else if (r->proto && r->proto != pd->proto)
5150 			r = r->skip[PF_SKIP_PROTO].ptr;
5151 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
5152 		    r->src.neg, kif, M_GETFIB(m)))
5153 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
5154 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
5155 		    r->dst.neg, NULL, M_GETFIB(m)))
5156 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
5157 		else if (r->tos && !(r->tos == pd->tos))
5158 			r = TAILQ_NEXT(r, entries);
5159 		else if (r->os_fingerprint != PF_OSFP_ANY)
5160 			r = TAILQ_NEXT(r, entries);
5161 		else if (pd->proto == IPPROTO_UDP &&
5162 		    (r->src.port_op || r->dst.port_op))
5163 			r = TAILQ_NEXT(r, entries);
5164 		else if (pd->proto == IPPROTO_TCP &&
5165 		    (r->src.port_op || r->dst.port_op || r->flagset))
5166 			r = TAILQ_NEXT(r, entries);
5167 		else if ((pd->proto == IPPROTO_ICMP ||
5168 		    pd->proto == IPPROTO_ICMPV6) &&
5169 		    (r->type || r->code))
5170 			r = TAILQ_NEXT(r, entries);
5171 		else if (r->prio &&
5172 		    !pf_match_ieee8021q_pcp(r->prio, m))
5173 			r = TAILQ_NEXT(r, entries);
5174 		else if (r->prob && r->prob <=
5175 		    (arc4random() % (UINT_MAX - 1) + 1))
5176 			r = TAILQ_NEXT(r, entries);
5177 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
5178 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
5179 			r = TAILQ_NEXT(r, entries);
5180 		else {
5181 			if (r->anchor == NULL) {
5182 				if (r->action == PF_MATCH) {
5183 					ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO);
5184 					if (ri == NULL) {
5185 						REASON_SET(&reason, PFRES_MEMORY);
5186 						goto cleanup;
5187 					}
5188 					ri->r = r;
5189 					SLIST_INSERT_HEAD(&match_rules, ri, entry);
5190 					pf_counter_u64_critical_enter();
5191 					pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
5192 					pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
5193 					pf_counter_u64_critical_exit();
5194 					pf_rule_to_actions(r, &pd->act);
5195 					if (r->log)
5196 						PFLOG_PACKET(kif, m, af,
5197 						    r->action, PFRES_MATCH, r,
5198 						    a, ruleset, pd, 1);
5199 				} else {
5200 					match = 1;
5201 					*rm = r;
5202 					*am = a;
5203 					*rsm = ruleset;
5204 				}
5205 				if ((*rm)->quick)
5206 					break;
5207 				r = TAILQ_NEXT(r, entries);
5208 			} else
5209 				pf_step_into_anchor(anchor_stack, &asd,
5210 				    &ruleset, PF_RULESET_FILTER, &r, &a,
5211 				    &match);
5212 		}
5213 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
5214 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
5215 			break;
5216 	}
5217 	r = *rm;
5218 	a = *am;
5219 	ruleset = *rsm;
5220 
5221 	REASON_SET(&reason, PFRES_MATCH);
5222 
5223 	/* apply actions for last matching pass/block rule */
5224 	pf_rule_to_actions(r, &pd->act);
5225 
5226 	if (r->log)
5227 		PFLOG_PACKET(kif, m, af, r->action, reason, r, a, ruleset, pd, 1);
5228 
5229 	if (r->action != PF_PASS)
5230 		return (PF_DROP);
5231 
5232 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
5233 		REASON_SET(&reason, PFRES_MEMORY);
5234 		goto cleanup;
5235 	}
5236 
5237 	return (PF_PASS);
5238 
5239 cleanup:
5240 	while ((ri = SLIST_FIRST(&match_rules))) {
5241 		SLIST_REMOVE_HEAD(&match_rules, entry);
5242 		free(ri, M_PF_RULE_ITEM);
5243 	}
5244 
5245 	return (PF_DROP);
5246 }
5247 
5248 static int
5249 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif,
5250     struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason,
5251     int *copyback)
5252 {
5253 	struct tcphdr		*th = &pd->hdr.tcp;
5254 	struct pf_state_peer	*src, *dst;
5255 	u_int16_t		 win = ntohs(th->th_win);
5256 	u_int32_t		 ack, end, seq, orig_seq;
5257 	u_int8_t		 sws, dws, psrc, pdst;
5258 	int			 ackskew;
5259 
5260 	if (pd->dir == (*state)->direction) {
5261 		src = &(*state)->src;
5262 		dst = &(*state)->dst;
5263 		psrc = PF_PEER_SRC;
5264 		pdst = PF_PEER_DST;
5265 	} else {
5266 		src = &(*state)->dst;
5267 		dst = &(*state)->src;
5268 		psrc = PF_PEER_DST;
5269 		pdst = PF_PEER_SRC;
5270 	}
5271 
5272 	if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
5273 		sws = src->wscale & PF_WSCALE_MASK;
5274 		dws = dst->wscale & PF_WSCALE_MASK;
5275 	} else
5276 		sws = dws = 0;
5277 
5278 	/*
5279 	 * Sequence tracking algorithm from Guido van Rooij's paper:
5280 	 *   http://www.madison-gurkha.com/publications/tcp_filtering/
5281 	 *	tcp_filtering.ps
5282 	 */
5283 
5284 	orig_seq = seq = ntohl(th->th_seq);
5285 	if (src->seqlo == 0) {
5286 		/* First packet from this end. Set its state */
5287 
5288 		if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) &&
5289 		    src->scrub == NULL) {
5290 			if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
5291 				REASON_SET(reason, PFRES_MEMORY);
5292 				return (PF_DROP);
5293 			}
5294 		}
5295 
5296 		/* Deferred generation of sequence number modulator */
5297 		if (dst->seqdiff && !src->seqdiff) {
5298 			/* use random iss for the TCP server */
5299 			while ((src->seqdiff = arc4random() - seq) == 0)
5300 				;
5301 			ack = ntohl(th->th_ack) - dst->seqdiff;
5302 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
5303 			    src->seqdiff), 0);
5304 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
5305 			*copyback = 1;
5306 		} else {
5307 			ack = ntohl(th->th_ack);
5308 		}
5309 
5310 		end = seq + pd->p_len;
5311 		if (th->th_flags & TH_SYN) {
5312 			end++;
5313 			if (dst->wscale & PF_WSCALE_FLAG) {
5314 				src->wscale = pf_get_wscale(m, off, th->th_off,
5315 				    pd->af);
5316 				if (src->wscale & PF_WSCALE_FLAG) {
5317 					/* Remove scale factor from initial
5318 					 * window */
5319 					sws = src->wscale & PF_WSCALE_MASK;
5320 					win = ((u_int32_t)win + (1 << sws) - 1)
5321 					    >> sws;
5322 					dws = dst->wscale & PF_WSCALE_MASK;
5323 				} else {
5324 					/* fixup other window */
5325 					dst->max_win <<= dst->wscale &
5326 					    PF_WSCALE_MASK;
5327 					/* in case of a retrans SYN|ACK */
5328 					dst->wscale = 0;
5329 				}
5330 			}
5331 		}
5332 		if (th->th_flags & TH_FIN)
5333 			end++;
5334 
5335 		src->seqlo = seq;
5336 		if (src->state < TCPS_SYN_SENT)
5337 			pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
5338 
5339 		/*
5340 		 * May need to slide the window (seqhi may have been set by
5341 		 * the crappy stack check or if we picked up the connection
5342 		 * after establishment)
5343 		 */
5344 		if (src->seqhi == 1 ||
5345 		    SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
5346 			src->seqhi = end + MAX(1, dst->max_win << dws);
5347 		if (win > src->max_win)
5348 			src->max_win = win;
5349 
5350 	} else {
5351 		ack = ntohl(th->th_ack) - dst->seqdiff;
5352 		if (src->seqdiff) {
5353 			/* Modulate sequence numbers */
5354 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
5355 			    src->seqdiff), 0);
5356 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
5357 			*copyback = 1;
5358 		}
5359 		end = seq + pd->p_len;
5360 		if (th->th_flags & TH_SYN)
5361 			end++;
5362 		if (th->th_flags & TH_FIN)
5363 			end++;
5364 	}
5365 
5366 	if ((th->th_flags & TH_ACK) == 0) {
5367 		/* Let it pass through the ack skew check */
5368 		ack = dst->seqlo;
5369 	} else if ((ack == 0 &&
5370 	    (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
5371 	    /* broken tcp stacks do not set ack */
5372 	    (dst->state < TCPS_SYN_SENT)) {
5373 		/*
5374 		 * Many stacks (ours included) will set the ACK number in an
5375 		 * FIN|ACK if the SYN times out -- no sequence to ACK.
5376 		 */
5377 		ack = dst->seqlo;
5378 	}
5379 
5380 	if (seq == end) {
5381 		/* Ease sequencing restrictions on no data packets */
5382 		seq = src->seqlo;
5383 		end = seq;
5384 	}
5385 
5386 	ackskew = dst->seqlo - ack;
5387 
5388 	/*
5389 	 * Need to demodulate the sequence numbers in any TCP SACK options
5390 	 * (Selective ACK). We could optionally validate the SACK values
5391 	 * against the current ACK window, either forwards or backwards, but
5392 	 * I'm not confident that SACK has been implemented properly
5393 	 * everywhere. It wouldn't surprise me if several stacks accidentally
5394 	 * SACK too far backwards of previously ACKed data. There really aren't
5395 	 * any security implications of bad SACKing unless the target stack
5396 	 * doesn't validate the option length correctly. Someone trying to
5397 	 * spoof into a TCP connection won't bother blindly sending SACK
5398 	 * options anyway.
5399 	 */
5400 	if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
5401 		if (pf_modulate_sack(m, off, pd, th, dst))
5402 			*copyback = 1;
5403 	}
5404 
5405 #define	MAXACKWINDOW (0xffff + 1500)	/* 1500 is an arbitrary fudge factor */
5406 	if (SEQ_GEQ(src->seqhi, end) &&
5407 	    /* Last octet inside other's window space */
5408 	    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
5409 	    /* Retrans: not more than one window back */
5410 	    (ackskew >= -MAXACKWINDOW) &&
5411 	    /* Acking not more than one reassembled fragment backwards */
5412 	    (ackskew <= (MAXACKWINDOW << sws)) &&
5413 	    /* Acking not more than one window forward */
5414 	    ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
5415 	    (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) {
5416 	    /* Require an exact/+1 sequence match on resets when possible */
5417 
5418 		if (dst->scrub || src->scrub) {
5419 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
5420 			    *state, src, dst, copyback))
5421 				return (PF_DROP);
5422 		}
5423 
5424 		/* update max window */
5425 		if (src->max_win < win)
5426 			src->max_win = win;
5427 		/* synchronize sequencing */
5428 		if (SEQ_GT(end, src->seqlo))
5429 			src->seqlo = end;
5430 		/* slide the window of what the other end can send */
5431 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
5432 			dst->seqhi = ack + MAX((win << sws), 1);
5433 
5434 		/* update states */
5435 		if (th->th_flags & TH_SYN)
5436 			if (src->state < TCPS_SYN_SENT)
5437 				pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
5438 		if (th->th_flags & TH_FIN)
5439 			if (src->state < TCPS_CLOSING)
5440 				pf_set_protostate(*state, psrc, TCPS_CLOSING);
5441 		if (th->th_flags & TH_ACK) {
5442 			if (dst->state == TCPS_SYN_SENT) {
5443 				pf_set_protostate(*state, pdst,
5444 				    TCPS_ESTABLISHED);
5445 				if (src->state == TCPS_ESTABLISHED &&
5446 				    (*state)->src_node != NULL &&
5447 				    pf_src_connlimit(state)) {
5448 					REASON_SET(reason, PFRES_SRCLIMIT);
5449 					return (PF_DROP);
5450 				}
5451 			} else if (dst->state == TCPS_CLOSING)
5452 				pf_set_protostate(*state, pdst,
5453 				    TCPS_FIN_WAIT_2);
5454 		}
5455 		if (th->th_flags & TH_RST)
5456 			pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
5457 
5458 		/* update expire time */
5459 		(*state)->expire = pf_get_uptime();
5460 		if (src->state >= TCPS_FIN_WAIT_2 &&
5461 		    dst->state >= TCPS_FIN_WAIT_2)
5462 			(*state)->timeout = PFTM_TCP_CLOSED;
5463 		else if (src->state >= TCPS_CLOSING &&
5464 		    dst->state >= TCPS_CLOSING)
5465 			(*state)->timeout = PFTM_TCP_FIN_WAIT;
5466 		else if (src->state < TCPS_ESTABLISHED ||
5467 		    dst->state < TCPS_ESTABLISHED)
5468 			(*state)->timeout = PFTM_TCP_OPENING;
5469 		else if (src->state >= TCPS_CLOSING ||
5470 		    dst->state >= TCPS_CLOSING)
5471 			(*state)->timeout = PFTM_TCP_CLOSING;
5472 		else
5473 			(*state)->timeout = PFTM_TCP_ESTABLISHED;
5474 
5475 		/* Fall through to PASS packet */
5476 
5477 	} else if ((dst->state < TCPS_SYN_SENT ||
5478 		dst->state >= TCPS_FIN_WAIT_2 ||
5479 		src->state >= TCPS_FIN_WAIT_2) &&
5480 	    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
5481 	    /* Within a window forward of the originating packet */
5482 	    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
5483 	    /* Within a window backward of the originating packet */
5484 
5485 		/*
5486 		 * This currently handles three situations:
5487 		 *  1) Stupid stacks will shotgun SYNs before their peer
5488 		 *     replies.
5489 		 *  2) When PF catches an already established stream (the
5490 		 *     firewall rebooted, the state table was flushed, routes
5491 		 *     changed...)
5492 		 *  3) Packets get funky immediately after the connection
5493 		 *     closes (this should catch Solaris spurious ACK|FINs
5494 		 *     that web servers like to spew after a close)
5495 		 *
5496 		 * This must be a little more careful than the above code
5497 		 * since packet floods will also be caught here. We don't
5498 		 * update the TTL here to mitigate the damage of a packet
5499 		 * flood and so the same code can handle awkward establishment
5500 		 * and a loosened connection close.
5501 		 * In the establishment case, a correct peer response will
5502 		 * validate the connection, go through the normal state code
5503 		 * and keep updating the state TTL.
5504 		 */
5505 
5506 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
5507 			printf("pf: loose state match: ");
5508 			pf_print_state(*state);
5509 			pf_print_flags(th->th_flags);
5510 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
5511 			    "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
5512 			    pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
5513 			    (unsigned long long)(*state)->packets[1],
5514 			    pd->dir == PF_IN ? "in" : "out",
5515 			    pd->dir == (*state)->direction ? "fwd" : "rev");
5516 		}
5517 
5518 		if (dst->scrub || src->scrub) {
5519 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
5520 			    *state, src, dst, copyback))
5521 				return (PF_DROP);
5522 		}
5523 
5524 		/* update max window */
5525 		if (src->max_win < win)
5526 			src->max_win = win;
5527 		/* synchronize sequencing */
5528 		if (SEQ_GT(end, src->seqlo))
5529 			src->seqlo = end;
5530 		/* slide the window of what the other end can send */
5531 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
5532 			dst->seqhi = ack + MAX((win << sws), 1);
5533 
5534 		/*
5535 		 * Cannot set dst->seqhi here since this could be a shotgunned
5536 		 * SYN and not an already established connection.
5537 		 */
5538 
5539 		if (th->th_flags & TH_FIN)
5540 			if (src->state < TCPS_CLOSING)
5541 				pf_set_protostate(*state, psrc, TCPS_CLOSING);
5542 		if (th->th_flags & TH_RST)
5543 			pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
5544 
5545 		/* Fall through to PASS packet */
5546 
5547 	} else {
5548 		if ((*state)->dst.state == TCPS_SYN_SENT &&
5549 		    (*state)->src.state == TCPS_SYN_SENT) {
5550 			/* Send RST for state mismatches during handshake */
5551 			if (!(th->th_flags & TH_RST))
5552 				pf_send_tcp((*state)->rule.ptr, pd->af,
5553 				    pd->dst, pd->src, th->th_dport,
5554 				    th->th_sport, ntohl(th->th_ack), 0,
5555 				    TH_RST, 0, 0,
5556 				    (*state)->rule.ptr->return_ttl, true, 0, 0,
5557 				    (*state)->act.rtableid);
5558 			src->seqlo = 0;
5559 			src->seqhi = 1;
5560 			src->max_win = 1;
5561 		} else if (V_pf_status.debug >= PF_DEBUG_MISC) {
5562 			printf("pf: BAD state: ");
5563 			pf_print_state(*state);
5564 			pf_print_flags(th->th_flags);
5565 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
5566 			    "pkts=%llu:%llu dir=%s,%s\n",
5567 			    seq, orig_seq, ack, pd->p_len, ackskew,
5568 			    (unsigned long long)(*state)->packets[0],
5569 			    (unsigned long long)(*state)->packets[1],
5570 			    pd->dir == PF_IN ? "in" : "out",
5571 			    pd->dir == (*state)->direction ? "fwd" : "rev");
5572 			printf("pf: State failure on: %c %c %c %c | %c %c\n",
5573 			    SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
5574 			    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
5575 			    ' ': '2',
5576 			    (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
5577 			    (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
5578 			    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
5579 			    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
5580 		}
5581 		REASON_SET(reason, PFRES_BADSTATE);
5582 		return (PF_DROP);
5583 	}
5584 
5585 	return (PF_PASS);
5586 }
5587 
5588 static int
5589 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason)
5590 {
5591 	struct tcphdr		*th = &pd->hdr.tcp;
5592 	struct pf_state_peer	*src, *dst;
5593 	u_int8_t		 psrc, pdst;
5594 
5595 	if (pd->dir == (*state)->direction) {
5596 		src = &(*state)->src;
5597 		dst = &(*state)->dst;
5598 		psrc = PF_PEER_SRC;
5599 		pdst = PF_PEER_DST;
5600 	} else {
5601 		src = &(*state)->dst;
5602 		dst = &(*state)->src;
5603 		psrc = PF_PEER_DST;
5604 		pdst = PF_PEER_SRC;
5605 	}
5606 
5607 	if (th->th_flags & TH_SYN)
5608 		if (src->state < TCPS_SYN_SENT)
5609 			pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
5610 	if (th->th_flags & TH_FIN)
5611 		if (src->state < TCPS_CLOSING)
5612 			pf_set_protostate(*state, psrc, TCPS_CLOSING);
5613 	if (th->th_flags & TH_ACK) {
5614 		if (dst->state == TCPS_SYN_SENT) {
5615 			pf_set_protostate(*state, pdst, TCPS_ESTABLISHED);
5616 			if (src->state == TCPS_ESTABLISHED &&
5617 			    (*state)->src_node != NULL &&
5618 			    pf_src_connlimit(state)) {
5619 				REASON_SET(reason, PFRES_SRCLIMIT);
5620 				return (PF_DROP);
5621 			}
5622 		} else if (dst->state == TCPS_CLOSING) {
5623 			pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2);
5624 		} else if (src->state == TCPS_SYN_SENT &&
5625 		    dst->state < TCPS_SYN_SENT) {
5626 			/*
5627 			 * Handle a special sloppy case where we only see one
5628 			 * half of the connection. If there is a ACK after
5629 			 * the initial SYN without ever seeing a packet from
5630 			 * the destination, set the connection to established.
5631 			 */
5632 			pf_set_protostate(*state, PF_PEER_BOTH,
5633 			    TCPS_ESTABLISHED);
5634 			dst->state = src->state = TCPS_ESTABLISHED;
5635 			if ((*state)->src_node != NULL &&
5636 			    pf_src_connlimit(state)) {
5637 				REASON_SET(reason, PFRES_SRCLIMIT);
5638 				return (PF_DROP);
5639 			}
5640 		} else if (src->state == TCPS_CLOSING &&
5641 		    dst->state == TCPS_ESTABLISHED &&
5642 		    dst->seqlo == 0) {
5643 			/*
5644 			 * Handle the closing of half connections where we
5645 			 * don't see the full bidirectional FIN/ACK+ACK
5646 			 * handshake.
5647 			 */
5648 			pf_set_protostate(*state, pdst, TCPS_CLOSING);
5649 		}
5650 	}
5651 	if (th->th_flags & TH_RST)
5652 		pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
5653 
5654 	/* update expire time */
5655 	(*state)->expire = pf_get_uptime();
5656 	if (src->state >= TCPS_FIN_WAIT_2 &&
5657 	    dst->state >= TCPS_FIN_WAIT_2)
5658 		(*state)->timeout = PFTM_TCP_CLOSED;
5659 	else if (src->state >= TCPS_CLOSING &&
5660 	    dst->state >= TCPS_CLOSING)
5661 		(*state)->timeout = PFTM_TCP_FIN_WAIT;
5662 	else if (src->state < TCPS_ESTABLISHED ||
5663 	    dst->state < TCPS_ESTABLISHED)
5664 		(*state)->timeout = PFTM_TCP_OPENING;
5665 	else if (src->state >= TCPS_CLOSING ||
5666 	    dst->state >= TCPS_CLOSING)
5667 		(*state)->timeout = PFTM_TCP_CLOSING;
5668 	else
5669 		(*state)->timeout = PFTM_TCP_ESTABLISHED;
5670 
5671 	return (PF_PASS);
5672 }
5673 
5674 static int
5675 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason)
5676 {
5677 	struct pf_state_key	*sk = (*state)->key[pd->didx];
5678 	struct tcphdr		*th = &pd->hdr.tcp;
5679 
5680 	if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
5681 		if (pd->dir != (*state)->direction) {
5682 			REASON_SET(reason, PFRES_SYNPROXY);
5683 			return (PF_SYNPROXY_DROP);
5684 		}
5685 		if (th->th_flags & TH_SYN) {
5686 			if (ntohl(th->th_seq) != (*state)->src.seqlo) {
5687 				REASON_SET(reason, PFRES_SYNPROXY);
5688 				return (PF_DROP);
5689 			}
5690 			pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
5691 			    pd->src, th->th_dport, th->th_sport,
5692 			    (*state)->src.seqhi, ntohl(th->th_seq) + 1,
5693 			    TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, true, 0, 0,
5694 			    (*state)->act.rtableid);
5695 			REASON_SET(reason, PFRES_SYNPROXY);
5696 			return (PF_SYNPROXY_DROP);
5697 		} else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK ||
5698 		    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
5699 		    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
5700 			REASON_SET(reason, PFRES_SYNPROXY);
5701 			return (PF_DROP);
5702 		} else if ((*state)->src_node != NULL &&
5703 		    pf_src_connlimit(state)) {
5704 			REASON_SET(reason, PFRES_SRCLIMIT);
5705 			return (PF_DROP);
5706 		} else
5707 			pf_set_protostate(*state, PF_PEER_SRC,
5708 			    PF_TCPS_PROXY_DST);
5709 	}
5710 	if ((*state)->src.state == PF_TCPS_PROXY_DST) {
5711 		if (pd->dir == (*state)->direction) {
5712 			if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
5713 			    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
5714 			    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
5715 				REASON_SET(reason, PFRES_SYNPROXY);
5716 				return (PF_DROP);
5717 			}
5718 			(*state)->src.max_win = MAX(ntohs(th->th_win), 1);
5719 			if ((*state)->dst.seqhi == 1)
5720 				(*state)->dst.seqhi = htonl(arc4random());
5721 			pf_send_tcp((*state)->rule.ptr, pd->af,
5722 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
5723 			    sk->port[pd->sidx], sk->port[pd->didx],
5724 			    (*state)->dst.seqhi, 0, TH_SYN, 0,
5725 			    (*state)->src.mss, 0, false, (*state)->tag, 0,
5726 			    (*state)->act.rtableid);
5727 			REASON_SET(reason, PFRES_SYNPROXY);
5728 			return (PF_SYNPROXY_DROP);
5729 		} else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
5730 		    (TH_SYN|TH_ACK)) ||
5731 		    (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
5732 			REASON_SET(reason, PFRES_SYNPROXY);
5733 			return (PF_DROP);
5734 		} else {
5735 			(*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
5736 			(*state)->dst.seqlo = ntohl(th->th_seq);
5737 			pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
5738 			    pd->src, th->th_dport, th->th_sport,
5739 			    ntohl(th->th_ack), ntohl(th->th_seq) + 1,
5740 			    TH_ACK, (*state)->src.max_win, 0, 0, false,
5741 			    (*state)->tag, 0, (*state)->act.rtableid);
5742 			pf_send_tcp((*state)->rule.ptr, pd->af,
5743 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
5744 			    sk->port[pd->sidx], sk->port[pd->didx],
5745 			    (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
5746 			    TH_ACK, (*state)->dst.max_win, 0, 0, true, 0, 0,
5747 			    (*state)->act.rtableid);
5748 			(*state)->src.seqdiff = (*state)->dst.seqhi -
5749 			    (*state)->src.seqlo;
5750 			(*state)->dst.seqdiff = (*state)->src.seqhi -
5751 			    (*state)->dst.seqlo;
5752 			(*state)->src.seqhi = (*state)->src.seqlo +
5753 			    (*state)->dst.max_win;
5754 			(*state)->dst.seqhi = (*state)->dst.seqlo +
5755 			    (*state)->src.max_win;
5756 			(*state)->src.wscale = (*state)->dst.wscale = 0;
5757 			pf_set_protostate(*state, PF_PEER_BOTH,
5758 			    TCPS_ESTABLISHED);
5759 			REASON_SET(reason, PFRES_SYNPROXY);
5760 			return (PF_SYNPROXY_DROP);
5761 		}
5762 	}
5763 
5764 	return (PF_PASS);
5765 }
5766 
5767 static int
5768 pf_test_state_tcp(struct pf_kstate **state, struct pfi_kkif *kif,
5769     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
5770     u_short *reason)
5771 {
5772 	struct pf_state_key_cmp	 key;
5773 	struct tcphdr		*th = &pd->hdr.tcp;
5774 	int			 copyback = 0;
5775 	int			 action;
5776 	struct pf_state_peer	*src, *dst;
5777 
5778 	bzero(&key, sizeof(key));
5779 	key.af = pd->af;
5780 	key.proto = IPPROTO_TCP;
5781 	if (pd->dir == PF_IN)	{	/* wire side, straight */
5782 		PF_ACPY(&key.addr[0], pd->src, key.af);
5783 		PF_ACPY(&key.addr[1], pd->dst, key.af);
5784 		key.port[0] = th->th_sport;
5785 		key.port[1] = th->th_dport;
5786 	} else {			/* stack side, reverse */
5787 		PF_ACPY(&key.addr[1], pd->src, key.af);
5788 		PF_ACPY(&key.addr[0], pd->dst, key.af);
5789 		key.port[1] = th->th_sport;
5790 		key.port[0] = th->th_dport;
5791 	}
5792 
5793 	STATE_LOOKUP(kif, &key, *state, pd);
5794 
5795 	if (pd->dir == (*state)->direction) {
5796 		src = &(*state)->src;
5797 		dst = &(*state)->dst;
5798 	} else {
5799 		src = &(*state)->dst;
5800 		dst = &(*state)->src;
5801 	}
5802 
5803 	if ((action = pf_synproxy(pd, state, reason)) != PF_PASS)
5804 		return (action);
5805 
5806 	if (dst->state >= TCPS_FIN_WAIT_2 &&
5807 	    src->state >= TCPS_FIN_WAIT_2 &&
5808 	    (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) ||
5809 	    ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK &&
5810 	    pf_syncookie_check(pd) && pd->dir == PF_IN))) {
5811 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
5812 			printf("pf: state reuse ");
5813 			pf_print_state(*state);
5814 			pf_print_flags(th->th_flags);
5815 			printf("\n");
5816 		}
5817 		/* XXX make sure it's the same direction ?? */
5818 		pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED);
5819 		pf_unlink_state(*state);
5820 		*state = NULL;
5821 		return (PF_DROP);
5822 	}
5823 
5824 	if ((*state)->state_flags & PFSTATE_SLOPPY) {
5825 		if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP)
5826 			return (PF_DROP);
5827 	} else {
5828 		if (pf_tcp_track_full(state, kif, m, off, pd, reason,
5829 		    &copyback) == PF_DROP)
5830 			return (PF_DROP);
5831 	}
5832 
5833 	/* translate source/destination address, if necessary */
5834 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5835 		struct pf_state_key *nk = (*state)->key[pd->didx];
5836 
5837 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
5838 		    nk->port[pd->sidx] != th->th_sport)
5839 			pf_change_ap(m, pd->src, &th->th_sport,
5840 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
5841 			    nk->port[pd->sidx], 0, pd->af);
5842 
5843 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
5844 		    nk->port[pd->didx] != th->th_dport)
5845 			pf_change_ap(m, pd->dst, &th->th_dport,
5846 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
5847 			    nk->port[pd->didx], 0, pd->af);
5848 		copyback = 1;
5849 	}
5850 
5851 	/* Copyback sequence modulation or stateful scrub changes if needed */
5852 	if (copyback)
5853 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
5854 
5855 	return (PF_PASS);
5856 }
5857 
5858 static int
5859 pf_test_state_udp(struct pf_kstate **state, struct pfi_kkif *kif,
5860     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
5861 {
5862 	struct pf_state_peer	*src, *dst;
5863 	struct pf_state_key_cmp	 key;
5864 	struct udphdr		*uh = &pd->hdr.udp;
5865 	uint8_t			 psrc, pdst;
5866 
5867 	bzero(&key, sizeof(key));
5868 	key.af = pd->af;
5869 	key.proto = IPPROTO_UDP;
5870 	if (pd->dir == PF_IN)	{	/* wire side, straight */
5871 		PF_ACPY(&key.addr[0], pd->src, key.af);
5872 		PF_ACPY(&key.addr[1], pd->dst, key.af);
5873 		key.port[0] = uh->uh_sport;
5874 		key.port[1] = uh->uh_dport;
5875 	} else {			/* stack side, reverse */
5876 		PF_ACPY(&key.addr[1], pd->src, key.af);
5877 		PF_ACPY(&key.addr[0], pd->dst, key.af);
5878 		key.port[1] = uh->uh_sport;
5879 		key.port[0] = uh->uh_dport;
5880 	}
5881 
5882 	STATE_LOOKUP(kif, &key, *state, pd);
5883 
5884 	if (pd->dir == (*state)->direction) {
5885 		src = &(*state)->src;
5886 		dst = &(*state)->dst;
5887 		psrc = PF_PEER_SRC;
5888 		pdst = PF_PEER_DST;
5889 	} else {
5890 		src = &(*state)->dst;
5891 		dst = &(*state)->src;
5892 		psrc = PF_PEER_DST;
5893 		pdst = PF_PEER_SRC;
5894 	}
5895 
5896 	/* update states */
5897 	if (src->state < PFUDPS_SINGLE)
5898 		pf_set_protostate(*state, psrc, PFUDPS_SINGLE);
5899 	if (dst->state == PFUDPS_SINGLE)
5900 		pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE);
5901 
5902 	/* update expire time */
5903 	(*state)->expire = pf_get_uptime();
5904 	if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
5905 		(*state)->timeout = PFTM_UDP_MULTIPLE;
5906 	else
5907 		(*state)->timeout = PFTM_UDP_SINGLE;
5908 
5909 	/* translate source/destination address, if necessary */
5910 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5911 		struct pf_state_key *nk = (*state)->key[pd->didx];
5912 
5913 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
5914 		    nk->port[pd->sidx] != uh->uh_sport)
5915 			pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
5916 			    &uh->uh_sum, &nk->addr[pd->sidx],
5917 			    nk->port[pd->sidx], 1, pd->af);
5918 
5919 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
5920 		    nk->port[pd->didx] != uh->uh_dport)
5921 			pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
5922 			    &uh->uh_sum, &nk->addr[pd->didx],
5923 			    nk->port[pd->didx], 1, pd->af);
5924 		m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
5925 	}
5926 
5927 	return (PF_PASS);
5928 }
5929 
5930 static int
5931 pf_test_state_sctp(struct pf_kstate **state, struct pfi_kkif *kif,
5932     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
5933 {
5934 	struct pf_state_key_cmp	 key;
5935 	struct pf_state_peer	*src, *dst;
5936 	struct sctphdr		*sh = &pd->hdr.sctp;
5937 	u_int8_t		 psrc; //, pdst;
5938 
5939 	bzero(&key, sizeof(key));
5940 	key.af = pd->af;
5941 	key.proto = IPPROTO_SCTP;
5942 	if (pd->dir == PF_IN)	{	/* wire side, straight */
5943 		PF_ACPY(&key.addr[0], pd->src, key.af);
5944 		PF_ACPY(&key.addr[1], pd->dst, key.af);
5945 		key.port[0] = sh->src_port;
5946 		key.port[1] = sh->dest_port;
5947 	} else {			/* stack side, reverse */
5948 		PF_ACPY(&key.addr[1], pd->src, key.af);
5949 		PF_ACPY(&key.addr[0], pd->dst, key.af);
5950 		key.port[1] = sh->src_port;
5951 		key.port[0] = sh->dest_port;
5952 	}
5953 
5954 	STATE_LOOKUP(kif, &key, *state, pd);
5955 
5956 	if (pd->dir == (*state)->direction) {
5957 		src = &(*state)->src;
5958 		dst = &(*state)->dst;
5959 		psrc = PF_PEER_SRC;
5960 	} else {
5961 		src = &(*state)->dst;
5962 		dst = &(*state)->src;
5963 		psrc = PF_PEER_DST;
5964 	}
5965 
5966 	/* Track state. */
5967 	if (pd->sctp_flags & PFDESC_SCTP_INIT) {
5968 		if (src->state < SCTP_COOKIE_WAIT) {
5969 			pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT);
5970 			(*state)->timeout = PFTM_SCTP_OPENING;
5971 		}
5972 	}
5973 	if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) {
5974 		MPASS(dst->scrub != NULL);
5975 		if (dst->scrub->pfss_v_tag == 0)
5976 			dst->scrub->pfss_v_tag = pd->sctp_initiate_tag;
5977 	}
5978 
5979 	if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) {
5980 		if (src->state < SCTP_ESTABLISHED) {
5981 			pf_set_protostate(*state, psrc, SCTP_ESTABLISHED);
5982 			(*state)->timeout = PFTM_SCTP_ESTABLISHED;
5983 		}
5984 	}
5985 	if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT |
5986 	    PFDESC_SCTP_SHUTDOWN_COMPLETE)) {
5987 		if (src->state < SCTP_SHUTDOWN_PENDING) {
5988 			pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING);
5989 			(*state)->timeout = PFTM_SCTP_CLOSING;
5990 		}
5991 	}
5992 	if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) {
5993 		pf_set_protostate(*state, psrc, SCTP_CLOSED);
5994 		(*state)->timeout = PFTM_SCTP_CLOSED;
5995 	}
5996 
5997 	if (src->scrub != NULL) {
5998 		if (src->scrub->pfss_v_tag == 0) {
5999 			src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag;
6000 		} else  if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag)
6001 			return (PF_DROP);
6002 	}
6003 
6004 	(*state)->expire = pf_get_uptime();
6005 
6006 	/* translate source/destination address, if necessary */
6007 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
6008 		uint16_t checksum = 0;
6009 		struct pf_state_key *nk = (*state)->key[pd->didx];
6010 
6011 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
6012 		    nk->port[pd->sidx] != pd->hdr.sctp.src_port) {
6013 			pf_change_ap(m, pd->src, &pd->hdr.sctp.src_port,
6014 			    pd->ip_sum, &checksum, &nk->addr[pd->sidx],
6015 			    nk->port[pd->sidx], 1, pd->af);
6016 		}
6017 
6018 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
6019 		    nk->port[pd->didx] != pd->hdr.sctp.dest_port) {
6020 			pf_change_ap(m, pd->dst, &pd->hdr.sctp.dest_port,
6021 			    pd->ip_sum, &checksum, &nk->addr[pd->didx],
6022 			    nk->port[pd->didx], 1, pd->af);
6023 		}
6024 	}
6025 
6026 	return (PF_PASS);
6027 }
6028 
6029 static void
6030 pf_sctp_multihome_detach_addr(const struct pf_kstate *s)
6031 {
6032 	struct pf_sctp_endpoint key;
6033 	struct pf_sctp_endpoint *ep;
6034 	struct pf_state_key *sks = s->key[PF_SK_STACK];
6035 	struct pf_sctp_source *i, *tmp;
6036 
6037 	if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL)
6038 		return;
6039 
6040 	PF_SCTP_ENDPOINTS_LOCK();
6041 
6042 	key.v_tag = s->dst.scrub->pfss_v_tag;
6043 	ep  = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6044 	if (ep != NULL) {
6045 		/* XXX Actually remove! */
6046 		TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) {
6047 			if (pf_addr_cmp(&i->addr,
6048 			    &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT],
6049 			    s->key[PF_SK_WIRE]->af) == 0) {
6050 				TAILQ_REMOVE(&ep->sources, i, entry);
6051 				free(i, M_PFTEMP);
6052 				break;
6053 			}
6054 		}
6055 
6056 		if (TAILQ_EMPTY(&ep->sources)) {
6057 			RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
6058 			free(ep, M_PFTEMP);
6059 		}
6060 	}
6061 
6062 	/* Other direction. */
6063 	key.v_tag = s->src.scrub->pfss_v_tag;
6064 	ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6065 	if (ep != NULL) {
6066 		TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) {
6067 			if (pf_addr_cmp(&i->addr,
6068 			    &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN],
6069 			    s->key[PF_SK_WIRE]->af) == 0) {
6070 				TAILQ_REMOVE(&ep->sources, i, entry);
6071 				free(i, M_PFTEMP);
6072 				break;
6073 			}
6074 		}
6075 
6076 		if (TAILQ_EMPTY(&ep->sources)) {
6077 			RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
6078 			free(ep, M_PFTEMP);
6079 		}
6080 	}
6081 
6082 	PF_SCTP_ENDPOINTS_UNLOCK();
6083 }
6084 
6085 static void
6086 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag)
6087 {
6088 	struct pf_sctp_endpoint key = {
6089 		.v_tag = v_tag,
6090 	};
6091 	struct pf_sctp_source *i;
6092 	struct pf_sctp_endpoint *ep;
6093 
6094 	PF_SCTP_ENDPOINTS_LOCK();
6095 
6096 	ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6097 	if (ep == NULL) {
6098 		ep = malloc(sizeof(struct pf_sctp_endpoint),
6099 		    M_PFTEMP, M_NOWAIT);
6100 		if (ep == NULL) {
6101 			PF_SCTP_ENDPOINTS_UNLOCK();
6102 			return;
6103 		}
6104 
6105 		ep->v_tag = v_tag;
6106 		TAILQ_INIT(&ep->sources);
6107 		RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
6108 	}
6109 
6110 	/* Avoid inserting duplicates. */
6111 	TAILQ_FOREACH(i, &ep->sources, entry) {
6112 		if (pf_addr_cmp(&i->addr, a, pd->af) == 0) {
6113 			PF_SCTP_ENDPOINTS_UNLOCK();
6114 			return;
6115 		}
6116 	}
6117 
6118 	i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT);
6119 	if (i == NULL) {
6120 		PF_SCTP_ENDPOINTS_UNLOCK();
6121 		return;
6122 	}
6123 
6124 	i->af = pd->af;
6125 	memcpy(&i->addr, a, sizeof(*a));
6126 	TAILQ_INSERT_TAIL(&ep->sources, i, entry);
6127 
6128 	PF_SCTP_ENDPOINTS_UNLOCK();
6129 }
6130 
6131 static void
6132 pf_sctp_multihome_delayed(struct pf_pdesc *pd, int off, struct pfi_kkif *kif,
6133     struct pf_kstate *s, int action)
6134 {
6135 	struct pf_sctp_multihome_job	*j, *tmp;
6136 	struct pf_sctp_source		*i;
6137 	int			 ret __unused;
6138 	struct pf_kstate	*sm = NULL;
6139 	struct pf_krule		*ra = NULL;
6140 	struct pf_krule		*r = &V_pf_default_rule;
6141 	struct pf_kruleset	*rs = NULL;
6142 	bool do_extra = true;
6143 
6144 	PF_RULES_RLOCK_TRACKER;
6145 
6146 again:
6147 	TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) {
6148 		if (s == NULL || action != PF_PASS)
6149 			goto free;
6150 
6151 		/* Confirm we don't recurse here. */
6152 		MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP));
6153 
6154 		switch (j->op) {
6155 		case  SCTP_ADD_IP_ADDRESS: {
6156 			uint32_t v_tag = pd->sctp_initiate_tag;
6157 
6158 			if (v_tag == 0) {
6159 				if (s->direction == pd->dir)
6160 					v_tag = s->src.scrub->pfss_v_tag;
6161 				else
6162 					v_tag = s->dst.scrub->pfss_v_tag;
6163 			}
6164 
6165 			/*
6166 			 * Avoid duplicating states. We'll already have
6167 			 * created a state based on the source address of
6168 			 * the packet, but SCTP endpoints may also list this
6169 			 * address again in the INIT(_ACK) parameters.
6170 			 */
6171 			if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) {
6172 				break;
6173 			}
6174 
6175 			j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP;
6176 			PF_RULES_RLOCK();
6177 			sm = NULL;
6178 			/*
6179 			 * New connections need to be floating, because
6180 			 * we cannot know what interfaces it will use.
6181 			 * That's why we pass V_pfi_all rather than kif.
6182 			 */
6183 			ret = pf_test_rule(&r, &sm, V_pfi_all,
6184 			    j->m, off, &j->pd, &ra, &rs, NULL);
6185 			PF_RULES_RUNLOCK();
6186 			SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->m, ret);
6187 			if (ret != PF_DROP && sm != NULL) {
6188 				/* Inherit v_tag values. */
6189 				if (sm->direction == s->direction) {
6190 					sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag;
6191 					sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag;
6192 				} else {
6193 					sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag;
6194 					sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag;
6195 				}
6196 				PF_STATE_UNLOCK(sm);
6197 			} else {
6198 				/* If we try duplicate inserts? */
6199 				break;
6200 			}
6201 
6202 			/* Only add the address if we've actually allowed the state. */
6203 			pf_sctp_multihome_add_addr(pd, &j->src, v_tag);
6204 
6205 			if (! do_extra) {
6206 				break;
6207 			}
6208 			/*
6209 			 * We need to do this for each of our source addresses.
6210 			 * Find those based on the verification tag.
6211 			 */
6212 			struct pf_sctp_endpoint key = {
6213 				.v_tag = pd->hdr.sctp.v_tag,
6214 			};
6215 			struct pf_sctp_endpoint *ep;
6216 
6217 			PF_SCTP_ENDPOINTS_LOCK();
6218 			ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6219 			if (ep == NULL) {
6220 				PF_SCTP_ENDPOINTS_UNLOCK();
6221 				break;
6222 			}
6223 			MPASS(ep != NULL);
6224 
6225 			TAILQ_FOREACH(i, &ep->sources, entry) {
6226 				struct pf_sctp_multihome_job *nj;
6227 
6228 				/* SCTP can intermingle IPv4 and IPv6. */
6229 				if (i->af != pd->af)
6230 					continue;
6231 
6232 				nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO);
6233 				if (! nj) {
6234 					continue;
6235 				}
6236 				memcpy(&nj->pd, &j->pd, sizeof(j->pd));
6237 				memcpy(&nj->src, &j->src, sizeof(nj->src));
6238 				nj->pd.src = &nj->src;
6239 				// New destination address!
6240 				memcpy(&nj->dst, &i->addr, sizeof(nj->dst));
6241 				nj->pd.dst = &nj->dst;
6242 				nj->m = j->m;
6243 				nj->op = j->op;
6244 
6245 				TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next);
6246 			}
6247 			PF_SCTP_ENDPOINTS_UNLOCK();
6248 
6249 			break;
6250 		}
6251 		case SCTP_DEL_IP_ADDRESS: {
6252 			struct pf_state_key_cmp key;
6253 			uint8_t psrc;
6254 
6255 			bzero(&key, sizeof(key));
6256 			key.af = j->pd.af;
6257 			key.proto = IPPROTO_SCTP;
6258 			if (j->pd.dir == PF_IN)	{	/* wire side, straight */
6259 				PF_ACPY(&key.addr[0], j->pd.src, key.af);
6260 				PF_ACPY(&key.addr[1], j->pd.dst, key.af);
6261 				key.port[0] = j->pd.hdr.sctp.src_port;
6262 				key.port[1] = j->pd.hdr.sctp.dest_port;
6263 			} else {			/* stack side, reverse */
6264 				PF_ACPY(&key.addr[1], j->pd.src, key.af);
6265 				PF_ACPY(&key.addr[0], j->pd.dst, key.af);
6266 				key.port[1] = j->pd.hdr.sctp.src_port;
6267 				key.port[0] = j->pd.hdr.sctp.dest_port;
6268 			}
6269 
6270 			sm = pf_find_state(kif, &key, j->pd.dir);
6271 			if (sm != NULL) {
6272 				PF_STATE_LOCK_ASSERT(sm);
6273 				if (j->pd.dir == sm->direction) {
6274 					psrc = PF_PEER_SRC;
6275 				} else {
6276 					psrc = PF_PEER_DST;
6277 				}
6278 				pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING);
6279 				sm->timeout = PFTM_SCTP_CLOSING;
6280 				PF_STATE_UNLOCK(sm);
6281 			}
6282 			break;
6283 		default:
6284 			panic("Unknown op %#x", j->op);
6285 		}
6286 	}
6287 
6288 	free:
6289 		TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next);
6290 		free(j, M_PFTEMP);
6291 	}
6292 
6293 	/* We may have inserted extra work while processing the list. */
6294 	if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) {
6295 		do_extra = false;
6296 		goto again;
6297 	}
6298 }
6299 
6300 static int
6301 pf_multihome_scan(struct mbuf *m, int start, int len, struct pf_pdesc *pd,
6302     struct pfi_kkif *kif, int op)
6303 {
6304 	int			 off = 0;
6305 	struct pf_sctp_multihome_job	*job;
6306 
6307 	while (off < len) {
6308 		struct sctp_paramhdr h;
6309 
6310 		if (!pf_pull_hdr(m, start + off, &h, sizeof(h), NULL, NULL,
6311 		    pd->af))
6312 			return (PF_DROP);
6313 
6314 		/* Parameters are at least 4 bytes. */
6315 		if (ntohs(h.param_length) < 4)
6316 			return (PF_DROP);
6317 
6318 		switch (ntohs(h.param_type)) {
6319 		case  SCTP_IPV4_ADDRESS: {
6320 			struct in_addr t;
6321 
6322 			if (ntohs(h.param_length) !=
6323 			    (sizeof(struct sctp_paramhdr) + sizeof(t)))
6324 				return (PF_DROP);
6325 
6326 			if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t),
6327 			    NULL, NULL, pd->af))
6328 				return (PF_DROP);
6329 
6330 			if (in_nullhost(t))
6331 				t.s_addr = pd->src->v4.s_addr;
6332 
6333 			/*
6334 			 * We hold the state lock (idhash) here, which means
6335 			 * that we can't acquire the keyhash, or we'll get a
6336 			 * LOR (and potentially double-lock things too). We also
6337 			 * can't release the state lock here, so instead we'll
6338 			 * enqueue this for async handling.
6339 			 * There's a relatively small race here, in that a
6340 			 * packet using the new addresses could arrive already,
6341 			 * but that's just though luck for it.
6342 			 */
6343 			job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO);
6344 			if (! job)
6345 				return (PF_DROP);
6346 
6347 			memcpy(&job->pd, pd, sizeof(*pd));
6348 
6349 			// New source address!
6350 			memcpy(&job->src, &t, sizeof(t));
6351 			job->pd.src = &job->src;
6352 			memcpy(&job->dst, pd->dst, sizeof(job->dst));
6353 			job->pd.dst = &job->dst;
6354 			job->m = m;
6355 			job->op = op;
6356 
6357 			TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next);
6358 			break;
6359 		}
6360 #ifdef INET6
6361 		case SCTP_IPV6_ADDRESS: {
6362 			struct in6_addr t;
6363 
6364 			if (ntohs(h.param_length) !=
6365 			    (sizeof(struct sctp_paramhdr) + sizeof(t)))
6366 				return (PF_DROP);
6367 
6368 			if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t),
6369 			    NULL, NULL, pd->af))
6370 				return (PF_DROP);
6371 			if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0)
6372 				break;
6373 			if (memcmp(&t, &in6addr_any, sizeof(t)) == 0)
6374 				memcpy(&t, &pd->src->v6, sizeof(t));
6375 
6376 			job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO);
6377 			if (! job)
6378 				return (PF_DROP);
6379 
6380 			memcpy(&job->pd, pd, sizeof(*pd));
6381 			memcpy(&job->src, &t, sizeof(t));
6382 			job->pd.src = &job->src;
6383 			memcpy(&job->dst, pd->dst, sizeof(job->dst));
6384 			job->pd.dst = &job->dst;
6385 			job->m = m;
6386 			job->op = op;
6387 
6388 			TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next);
6389 			break;
6390 		}
6391 #endif
6392 		case SCTP_ADD_IP_ADDRESS: {
6393 			int ret;
6394 			struct sctp_asconf_paramhdr ah;
6395 
6396 			if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah),
6397 			    NULL, NULL, pd->af))
6398 				return (PF_DROP);
6399 
6400 			ret = pf_multihome_scan(m, start + off + sizeof(ah),
6401 			    ntohs(ah.ph.param_length) - sizeof(ah), pd, kif,
6402 			    SCTP_ADD_IP_ADDRESS);
6403 			if (ret != PF_PASS)
6404 				return (ret);
6405 			break;
6406 		}
6407 		case SCTP_DEL_IP_ADDRESS: {
6408 			int ret;
6409 			struct sctp_asconf_paramhdr ah;
6410 
6411 			if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah),
6412 			    NULL, NULL, pd->af))
6413 				return (PF_DROP);
6414 			ret = pf_multihome_scan(m, start + off + sizeof(ah),
6415 			    ntohs(ah.ph.param_length) - sizeof(ah), pd, kif,
6416 			    SCTP_DEL_IP_ADDRESS);
6417 			if (ret != PF_PASS)
6418 				return (ret);
6419 			break;
6420 		}
6421 		default:
6422 			break;
6423 		}
6424 
6425 		off += roundup(ntohs(h.param_length), 4);
6426 	}
6427 
6428 	return (PF_PASS);
6429 }
6430 int
6431 pf_multihome_scan_init(struct mbuf *m, int start, int len, struct pf_pdesc *pd,
6432     struct pfi_kkif *kif)
6433 {
6434 	start += sizeof(struct sctp_init_chunk);
6435 	len -= sizeof(struct sctp_init_chunk);
6436 
6437 	return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS));
6438 }
6439 
6440 int
6441 pf_multihome_scan_asconf(struct mbuf *m, int start, int len,
6442     struct pf_pdesc *pd, struct pfi_kkif *kif)
6443 {
6444 	start += sizeof(struct sctp_asconf_chunk);
6445 	len -= sizeof(struct sctp_asconf_chunk);
6446 
6447 	return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS));
6448 }
6449 
6450 static int
6451 pf_test_state_icmp(struct pf_kstate **state, struct pfi_kkif *kif,
6452     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
6453 {
6454 	struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
6455 	u_int16_t	 icmpid = 0, *icmpsum;
6456 	u_int8_t	 icmptype, icmpcode;
6457 	int		 state_icmp = 0;
6458 	struct pf_state_key_cmp key;
6459 
6460 	bzero(&key, sizeof(key));
6461 	switch (pd->proto) {
6462 #ifdef INET
6463 	case IPPROTO_ICMP:
6464 		icmptype = pd->hdr.icmp.icmp_type;
6465 		icmpcode = pd->hdr.icmp.icmp_code;
6466 		icmpid = pd->hdr.icmp.icmp_id;
6467 		icmpsum = &pd->hdr.icmp.icmp_cksum;
6468 
6469 		if (icmptype == ICMP_UNREACH ||
6470 		    icmptype == ICMP_SOURCEQUENCH ||
6471 		    icmptype == ICMP_REDIRECT ||
6472 		    icmptype == ICMP_TIMXCEED ||
6473 		    icmptype == ICMP_PARAMPROB)
6474 			state_icmp++;
6475 		break;
6476 #endif /* INET */
6477 #ifdef INET6
6478 	case IPPROTO_ICMPV6:
6479 		icmptype = pd->hdr.icmp6.icmp6_type;
6480 		icmpcode = pd->hdr.icmp6.icmp6_code;
6481 		icmpid = pd->hdr.icmp6.icmp6_id;
6482 		icmpsum = &pd->hdr.icmp6.icmp6_cksum;
6483 
6484 		if (icmptype == ICMP6_DST_UNREACH ||
6485 		    icmptype == ICMP6_PACKET_TOO_BIG ||
6486 		    icmptype == ICMP6_TIME_EXCEEDED ||
6487 		    icmptype == ICMP6_PARAM_PROB)
6488 			state_icmp++;
6489 		break;
6490 #endif /* INET6 */
6491 	}
6492 
6493 	if (!state_icmp) {
6494 		/*
6495 		 * ICMP query/reply message not related to a TCP/UDP packet.
6496 		 * Search for an ICMP state.
6497 		 */
6498 		key.af = pd->af;
6499 		key.proto = pd->proto;
6500 		key.port[0] = key.port[1] = icmpid;
6501 		if (pd->dir == PF_IN)	{	/* wire side, straight */
6502 			PF_ACPY(&key.addr[0], pd->src, key.af);
6503 			PF_ACPY(&key.addr[1], pd->dst, key.af);
6504 		} else {			/* stack side, reverse */
6505 			PF_ACPY(&key.addr[1], pd->src, key.af);
6506 			PF_ACPY(&key.addr[0], pd->dst, key.af);
6507 		}
6508 
6509 		STATE_LOOKUP(kif, &key, *state, pd);
6510 
6511 		(*state)->expire = pf_get_uptime();
6512 		(*state)->timeout = PFTM_ICMP_ERROR_REPLY;
6513 
6514 		/* translate source/destination address, if necessary */
6515 		if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
6516 			struct pf_state_key *nk = (*state)->key[pd->didx];
6517 
6518 			switch (pd->af) {
6519 #ifdef INET
6520 			case AF_INET:
6521 				if (PF_ANEQ(pd->src,
6522 				    &nk->addr[pd->sidx], AF_INET))
6523 					pf_change_a(&saddr->v4.s_addr,
6524 					    pd->ip_sum,
6525 					    nk->addr[pd->sidx].v4.s_addr, 0);
6526 
6527 				if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
6528 				    AF_INET))
6529 					pf_change_a(&daddr->v4.s_addr,
6530 					    pd->ip_sum,
6531 					    nk->addr[pd->didx].v4.s_addr, 0);
6532 
6533 				if (nk->port[0] !=
6534 				    pd->hdr.icmp.icmp_id) {
6535 					pd->hdr.icmp.icmp_cksum =
6536 					    pf_cksum_fixup(
6537 					    pd->hdr.icmp.icmp_cksum, icmpid,
6538 					    nk->port[pd->sidx], 0);
6539 					pd->hdr.icmp.icmp_id =
6540 					    nk->port[pd->sidx];
6541 				}
6542 
6543 				m_copyback(m, off, ICMP_MINLEN,
6544 				    (caddr_t )&pd->hdr.icmp);
6545 				break;
6546 #endif /* INET */
6547 #ifdef INET6
6548 			case AF_INET6:
6549 				if (PF_ANEQ(pd->src,
6550 				    &nk->addr[pd->sidx], AF_INET6))
6551 					pf_change_a6(saddr,
6552 					    &pd->hdr.icmp6.icmp6_cksum,
6553 					    &nk->addr[pd->sidx], 0);
6554 
6555 				if (PF_ANEQ(pd->dst,
6556 				    &nk->addr[pd->didx], AF_INET6))
6557 					pf_change_a6(daddr,
6558 					    &pd->hdr.icmp6.icmp6_cksum,
6559 					    &nk->addr[pd->didx], 0);
6560 
6561 				m_copyback(m, off, sizeof(struct icmp6_hdr),
6562 				    (caddr_t )&pd->hdr.icmp6);
6563 				break;
6564 #endif /* INET6 */
6565 			}
6566 		}
6567 		return (PF_PASS);
6568 
6569 	} else {
6570 		/*
6571 		 * ICMP error message in response to a TCP/UDP packet.
6572 		 * Extract the inner TCP/UDP header and search for that state.
6573 		 */
6574 
6575 		struct pf_pdesc	pd2;
6576 		bzero(&pd2, sizeof pd2);
6577 #ifdef INET
6578 		struct ip	h2;
6579 #endif /* INET */
6580 #ifdef INET6
6581 		struct ip6_hdr	h2_6;
6582 		int		terminal = 0;
6583 #endif /* INET6 */
6584 		int		ipoff2 = 0;
6585 		int		off2 = 0;
6586 
6587 		pd2.af = pd->af;
6588 		/* Payload packet is from the opposite direction. */
6589 		pd2.sidx = (pd->dir == PF_IN) ? 1 : 0;
6590 		pd2.didx = (pd->dir == PF_IN) ? 0 : 1;
6591 		switch (pd->af) {
6592 #ifdef INET
6593 		case AF_INET:
6594 			/* offset of h2 in mbuf chain */
6595 			ipoff2 = off + ICMP_MINLEN;
6596 
6597 			if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
6598 			    NULL, reason, pd2.af)) {
6599 				DPFPRINTF(PF_DEBUG_MISC,
6600 				    ("pf: ICMP error message too short "
6601 				    "(ip)\n"));
6602 				return (PF_DROP);
6603 			}
6604 			/*
6605 			 * ICMP error messages don't refer to non-first
6606 			 * fragments
6607 			 */
6608 			if (h2.ip_off & htons(IP_OFFMASK)) {
6609 				REASON_SET(reason, PFRES_FRAG);
6610 				return (PF_DROP);
6611 			}
6612 
6613 			/* offset of protocol header that follows h2 */
6614 			off2 = ipoff2 + (h2.ip_hl << 2);
6615 
6616 			pd2.proto = h2.ip_p;
6617 			pd2.src = (struct pf_addr *)&h2.ip_src;
6618 			pd2.dst = (struct pf_addr *)&h2.ip_dst;
6619 			pd2.ip_sum = &h2.ip_sum;
6620 			break;
6621 #endif /* INET */
6622 #ifdef INET6
6623 		case AF_INET6:
6624 			ipoff2 = off + sizeof(struct icmp6_hdr);
6625 
6626 			if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
6627 			    NULL, reason, pd2.af)) {
6628 				DPFPRINTF(PF_DEBUG_MISC,
6629 				    ("pf: ICMP error message too short "
6630 				    "(ip6)\n"));
6631 				return (PF_DROP);
6632 			}
6633 			pd2.proto = h2_6.ip6_nxt;
6634 			pd2.src = (struct pf_addr *)&h2_6.ip6_src;
6635 			pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
6636 			pd2.ip_sum = NULL;
6637 			off2 = ipoff2 + sizeof(h2_6);
6638 			do {
6639 				switch (pd2.proto) {
6640 				case IPPROTO_FRAGMENT:
6641 					/*
6642 					 * ICMPv6 error messages for
6643 					 * non-first fragments
6644 					 */
6645 					REASON_SET(reason, PFRES_FRAG);
6646 					return (PF_DROP);
6647 				case IPPROTO_AH:
6648 				case IPPROTO_HOPOPTS:
6649 				case IPPROTO_ROUTING:
6650 				case IPPROTO_DSTOPTS: {
6651 					/* get next header and header length */
6652 					struct ip6_ext opt6;
6653 
6654 					if (!pf_pull_hdr(m, off2, &opt6,
6655 					    sizeof(opt6), NULL, reason,
6656 					    pd2.af)) {
6657 						DPFPRINTF(PF_DEBUG_MISC,
6658 						    ("pf: ICMPv6 short opt\n"));
6659 						return (PF_DROP);
6660 					}
6661 					if (pd2.proto == IPPROTO_AH)
6662 						off2 += (opt6.ip6e_len + 2) * 4;
6663 					else
6664 						off2 += (opt6.ip6e_len + 1) * 8;
6665 					pd2.proto = opt6.ip6e_nxt;
6666 					/* goto the next header */
6667 					break;
6668 				}
6669 				default:
6670 					terminal++;
6671 					break;
6672 				}
6673 			} while (!terminal);
6674 			break;
6675 #endif /* INET6 */
6676 		}
6677 
6678 		if (PF_ANEQ(pd->dst, pd2.src, pd->af)) {
6679 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
6680 				printf("pf: BAD ICMP %d:%d outer dst: ",
6681 				    icmptype, icmpcode);
6682 				pf_print_host(pd->src, 0, pd->af);
6683 				printf(" -> ");
6684 				pf_print_host(pd->dst, 0, pd->af);
6685 				printf(" inner src: ");
6686 				pf_print_host(pd2.src, 0, pd2.af);
6687 				printf(" -> ");
6688 				pf_print_host(pd2.dst, 0, pd2.af);
6689 				printf("\n");
6690 			}
6691 			REASON_SET(reason, PFRES_BADSTATE);
6692 			return (PF_DROP);
6693 		}
6694 
6695 		switch (pd2.proto) {
6696 		case IPPROTO_TCP: {
6697 			struct tcphdr		 th;
6698 			u_int32_t		 seq;
6699 			struct pf_state_peer	*src, *dst;
6700 			u_int8_t		 dws;
6701 			int			 copyback = 0;
6702 
6703 			/*
6704 			 * Only the first 8 bytes of the TCP header can be
6705 			 * expected. Don't access any TCP header fields after
6706 			 * th_seq, an ackskew test is not possible.
6707 			 */
6708 			if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
6709 			    pd2.af)) {
6710 				DPFPRINTF(PF_DEBUG_MISC,
6711 				    ("pf: ICMP error message too short "
6712 				    "(tcp)\n"));
6713 				return (PF_DROP);
6714 			}
6715 
6716 			key.af = pd2.af;
6717 			key.proto = IPPROTO_TCP;
6718 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
6719 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
6720 			key.port[pd2.sidx] = th.th_sport;
6721 			key.port[pd2.didx] = th.th_dport;
6722 
6723 			STATE_LOOKUP(kif, &key, *state, pd);
6724 
6725 			if (pd->dir == (*state)->direction) {
6726 				src = &(*state)->dst;
6727 				dst = &(*state)->src;
6728 			} else {
6729 				src = &(*state)->src;
6730 				dst = &(*state)->dst;
6731 			}
6732 
6733 			if (src->wscale && dst->wscale)
6734 				dws = dst->wscale & PF_WSCALE_MASK;
6735 			else
6736 				dws = 0;
6737 
6738 			/* Demodulate sequence number */
6739 			seq = ntohl(th.th_seq) - src->seqdiff;
6740 			if (src->seqdiff) {
6741 				pf_change_a(&th.th_seq, icmpsum,
6742 				    htonl(seq), 0);
6743 				copyback = 1;
6744 			}
6745 
6746 			if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
6747 			    (!SEQ_GEQ(src->seqhi, seq) ||
6748 			    !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
6749 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
6750 					printf("pf: BAD ICMP %d:%d ",
6751 					    icmptype, icmpcode);
6752 					pf_print_host(pd->src, 0, pd->af);
6753 					printf(" -> ");
6754 					pf_print_host(pd->dst, 0, pd->af);
6755 					printf(" state: ");
6756 					pf_print_state(*state);
6757 					printf(" seq=%u\n", seq);
6758 				}
6759 				REASON_SET(reason, PFRES_BADSTATE);
6760 				return (PF_DROP);
6761 			} else {
6762 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
6763 					printf("pf: OK ICMP %d:%d ",
6764 					    icmptype, icmpcode);
6765 					pf_print_host(pd->src, 0, pd->af);
6766 					printf(" -> ");
6767 					pf_print_host(pd->dst, 0, pd->af);
6768 					printf(" state: ");
6769 					pf_print_state(*state);
6770 					printf(" seq=%u\n", seq);
6771 				}
6772 			}
6773 
6774 			/* translate source/destination address, if necessary */
6775 			if ((*state)->key[PF_SK_WIRE] !=
6776 			    (*state)->key[PF_SK_STACK]) {
6777 				struct pf_state_key *nk =
6778 				    (*state)->key[pd->didx];
6779 
6780 				if (PF_ANEQ(pd2.src,
6781 				    &nk->addr[pd2.sidx], pd2.af) ||
6782 				    nk->port[pd2.sidx] != th.th_sport)
6783 					pf_change_icmp(pd2.src, &th.th_sport,
6784 					    daddr, &nk->addr[pd2.sidx],
6785 					    nk->port[pd2.sidx], NULL,
6786 					    pd2.ip_sum, icmpsum,
6787 					    pd->ip_sum, 0, pd2.af);
6788 
6789 				if (PF_ANEQ(pd2.dst,
6790 				    &nk->addr[pd2.didx], pd2.af) ||
6791 				    nk->port[pd2.didx] != th.th_dport)
6792 					pf_change_icmp(pd2.dst, &th.th_dport,
6793 					    saddr, &nk->addr[pd2.didx],
6794 					    nk->port[pd2.didx], NULL,
6795 					    pd2.ip_sum, icmpsum,
6796 					    pd->ip_sum, 0, pd2.af);
6797 				copyback = 1;
6798 			}
6799 
6800 			if (copyback) {
6801 				switch (pd2.af) {
6802 #ifdef INET
6803 				case AF_INET:
6804 					m_copyback(m, off, ICMP_MINLEN,
6805 					    (caddr_t )&pd->hdr.icmp);
6806 					m_copyback(m, ipoff2, sizeof(h2),
6807 					    (caddr_t )&h2);
6808 					break;
6809 #endif /* INET */
6810 #ifdef INET6
6811 				case AF_INET6:
6812 					m_copyback(m, off,
6813 					    sizeof(struct icmp6_hdr),
6814 					    (caddr_t )&pd->hdr.icmp6);
6815 					m_copyback(m, ipoff2, sizeof(h2_6),
6816 					    (caddr_t )&h2_6);
6817 					break;
6818 #endif /* INET6 */
6819 				}
6820 				m_copyback(m, off2, 8, (caddr_t)&th);
6821 			}
6822 
6823 			return (PF_PASS);
6824 			break;
6825 		}
6826 		case IPPROTO_UDP: {
6827 			struct udphdr		uh;
6828 
6829 			if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
6830 			    NULL, reason, pd2.af)) {
6831 				DPFPRINTF(PF_DEBUG_MISC,
6832 				    ("pf: ICMP error message too short "
6833 				    "(udp)\n"));
6834 				return (PF_DROP);
6835 			}
6836 
6837 			key.af = pd2.af;
6838 			key.proto = IPPROTO_UDP;
6839 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
6840 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
6841 			key.port[pd2.sidx] = uh.uh_sport;
6842 			key.port[pd2.didx] = uh.uh_dport;
6843 
6844 			STATE_LOOKUP(kif, &key, *state, pd);
6845 
6846 			/* translate source/destination address, if necessary */
6847 			if ((*state)->key[PF_SK_WIRE] !=
6848 			    (*state)->key[PF_SK_STACK]) {
6849 				struct pf_state_key *nk =
6850 				    (*state)->key[pd->didx];
6851 
6852 				if (PF_ANEQ(pd2.src,
6853 				    &nk->addr[pd2.sidx], pd2.af) ||
6854 				    nk->port[pd2.sidx] != uh.uh_sport)
6855 					pf_change_icmp(pd2.src, &uh.uh_sport,
6856 					    daddr, &nk->addr[pd2.sidx],
6857 					    nk->port[pd2.sidx], &uh.uh_sum,
6858 					    pd2.ip_sum, icmpsum,
6859 					    pd->ip_sum, 1, pd2.af);
6860 
6861 				if (PF_ANEQ(pd2.dst,
6862 				    &nk->addr[pd2.didx], pd2.af) ||
6863 				    nk->port[pd2.didx] != uh.uh_dport)
6864 					pf_change_icmp(pd2.dst, &uh.uh_dport,
6865 					    saddr, &nk->addr[pd2.didx],
6866 					    nk->port[pd2.didx], &uh.uh_sum,
6867 					    pd2.ip_sum, icmpsum,
6868 					    pd->ip_sum, 1, pd2.af);
6869 
6870 				switch (pd2.af) {
6871 #ifdef INET
6872 				case AF_INET:
6873 					m_copyback(m, off, ICMP_MINLEN,
6874 					    (caddr_t )&pd->hdr.icmp);
6875 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
6876 					break;
6877 #endif /* INET */
6878 #ifdef INET6
6879 				case AF_INET6:
6880 					m_copyback(m, off,
6881 					    sizeof(struct icmp6_hdr),
6882 					    (caddr_t )&pd->hdr.icmp6);
6883 					m_copyback(m, ipoff2, sizeof(h2_6),
6884 					    (caddr_t )&h2_6);
6885 					break;
6886 #endif /* INET6 */
6887 				}
6888 				m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
6889 			}
6890 			return (PF_PASS);
6891 			break;
6892 		}
6893 #ifdef INET
6894 		case IPPROTO_ICMP: {
6895 			struct icmp		iih;
6896 
6897 			if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
6898 			    NULL, reason, pd2.af)) {
6899 				DPFPRINTF(PF_DEBUG_MISC,
6900 				    ("pf: ICMP error message too short i"
6901 				    "(icmp)\n"));
6902 				return (PF_DROP);
6903 			}
6904 
6905 			key.af = pd2.af;
6906 			key.proto = IPPROTO_ICMP;
6907 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
6908 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
6909 			key.port[0] = key.port[1] = iih.icmp_id;
6910 
6911 			STATE_LOOKUP(kif, &key, *state, pd);
6912 
6913 			/* translate source/destination address, if necessary */
6914 			if ((*state)->key[PF_SK_WIRE] !=
6915 			    (*state)->key[PF_SK_STACK]) {
6916 				struct pf_state_key *nk =
6917 				    (*state)->key[pd->didx];
6918 
6919 				if (PF_ANEQ(pd2.src,
6920 				    &nk->addr[pd2.sidx], pd2.af) ||
6921 				    nk->port[pd2.sidx] != iih.icmp_id)
6922 					pf_change_icmp(pd2.src, &iih.icmp_id,
6923 					    daddr, &nk->addr[pd2.sidx],
6924 					    nk->port[pd2.sidx], NULL,
6925 					    pd2.ip_sum, icmpsum,
6926 					    pd->ip_sum, 0, AF_INET);
6927 
6928 				if (PF_ANEQ(pd2.dst,
6929 				    &nk->addr[pd2.didx], pd2.af) ||
6930 				    nk->port[pd2.didx] != iih.icmp_id)
6931 					pf_change_icmp(pd2.dst, &iih.icmp_id,
6932 					    saddr, &nk->addr[pd2.didx],
6933 					    nk->port[pd2.didx], NULL,
6934 					    pd2.ip_sum, icmpsum,
6935 					    pd->ip_sum, 0, AF_INET);
6936 
6937 				m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
6938 				m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
6939 				m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
6940 			}
6941 			return (PF_PASS);
6942 			break;
6943 		}
6944 #endif /* INET */
6945 #ifdef INET6
6946 		case IPPROTO_ICMPV6: {
6947 			struct icmp6_hdr	iih;
6948 
6949 			if (!pf_pull_hdr(m, off2, &iih,
6950 			    sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
6951 				DPFPRINTF(PF_DEBUG_MISC,
6952 				    ("pf: ICMP error message too short "
6953 				    "(icmp6)\n"));
6954 				return (PF_DROP);
6955 			}
6956 
6957 			key.af = pd2.af;
6958 			key.proto = IPPROTO_ICMPV6;
6959 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
6960 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
6961 			key.port[0] = key.port[1] = iih.icmp6_id;
6962 
6963 			STATE_LOOKUP(kif, &key, *state, pd);
6964 
6965 			/* translate source/destination address, if necessary */
6966 			if ((*state)->key[PF_SK_WIRE] !=
6967 			    (*state)->key[PF_SK_STACK]) {
6968 				struct pf_state_key *nk =
6969 				    (*state)->key[pd->didx];
6970 
6971 				if (PF_ANEQ(pd2.src,
6972 				    &nk->addr[pd2.sidx], pd2.af) ||
6973 				    nk->port[pd2.sidx] != iih.icmp6_id)
6974 					pf_change_icmp(pd2.src, &iih.icmp6_id,
6975 					    daddr, &nk->addr[pd2.sidx],
6976 					    nk->port[pd2.sidx], NULL,
6977 					    pd2.ip_sum, icmpsum,
6978 					    pd->ip_sum, 0, AF_INET6);
6979 
6980 				if (PF_ANEQ(pd2.dst,
6981 				    &nk->addr[pd2.didx], pd2.af) ||
6982 				    nk->port[pd2.didx] != iih.icmp6_id)
6983 					pf_change_icmp(pd2.dst, &iih.icmp6_id,
6984 					    saddr, &nk->addr[pd2.didx],
6985 					    nk->port[pd2.didx], NULL,
6986 					    pd2.ip_sum, icmpsum,
6987 					    pd->ip_sum, 0, AF_INET6);
6988 
6989 				m_copyback(m, off, sizeof(struct icmp6_hdr),
6990 				    (caddr_t)&pd->hdr.icmp6);
6991 				m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
6992 				m_copyback(m, off2, sizeof(struct icmp6_hdr),
6993 				    (caddr_t)&iih);
6994 			}
6995 			return (PF_PASS);
6996 			break;
6997 		}
6998 #endif /* INET6 */
6999 		default: {
7000 			key.af = pd2.af;
7001 			key.proto = pd2.proto;
7002 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
7003 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
7004 			key.port[0] = key.port[1] = 0;
7005 
7006 			STATE_LOOKUP(kif, &key, *state, pd);
7007 
7008 			/* translate source/destination address, if necessary */
7009 			if ((*state)->key[PF_SK_WIRE] !=
7010 			    (*state)->key[PF_SK_STACK]) {
7011 				struct pf_state_key *nk =
7012 				    (*state)->key[pd->didx];
7013 
7014 				if (PF_ANEQ(pd2.src,
7015 				    &nk->addr[pd2.sidx], pd2.af))
7016 					pf_change_icmp(pd2.src, NULL, daddr,
7017 					    &nk->addr[pd2.sidx], 0, NULL,
7018 					    pd2.ip_sum, icmpsum,
7019 					    pd->ip_sum, 0, pd2.af);
7020 
7021 				if (PF_ANEQ(pd2.dst,
7022 				    &nk->addr[pd2.didx], pd2.af))
7023 					pf_change_icmp(pd2.dst, NULL, saddr,
7024 					    &nk->addr[pd2.didx], 0, NULL,
7025 					    pd2.ip_sum, icmpsum,
7026 					    pd->ip_sum, 0, pd2.af);
7027 
7028 				switch (pd2.af) {
7029 #ifdef INET
7030 				case AF_INET:
7031 					m_copyback(m, off, ICMP_MINLEN,
7032 					    (caddr_t)&pd->hdr.icmp);
7033 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
7034 					break;
7035 #endif /* INET */
7036 #ifdef INET6
7037 				case AF_INET6:
7038 					m_copyback(m, off,
7039 					    sizeof(struct icmp6_hdr),
7040 					    (caddr_t )&pd->hdr.icmp6);
7041 					m_copyback(m, ipoff2, sizeof(h2_6),
7042 					    (caddr_t )&h2_6);
7043 					break;
7044 #endif /* INET6 */
7045 				}
7046 			}
7047 			return (PF_PASS);
7048 			break;
7049 		}
7050 		}
7051 	}
7052 }
7053 
7054 static int
7055 pf_test_state_other(struct pf_kstate **state, struct pfi_kkif *kif,
7056     struct mbuf *m, struct pf_pdesc *pd)
7057 {
7058 	struct pf_state_peer	*src, *dst;
7059 	struct pf_state_key_cmp	 key;
7060 	uint8_t			 psrc, pdst;
7061 
7062 	bzero(&key, sizeof(key));
7063 	key.af = pd->af;
7064 	key.proto = pd->proto;
7065 	if (pd->dir == PF_IN)	{
7066 		PF_ACPY(&key.addr[0], pd->src, key.af);
7067 		PF_ACPY(&key.addr[1], pd->dst, key.af);
7068 		key.port[0] = key.port[1] = 0;
7069 	} else {
7070 		PF_ACPY(&key.addr[1], pd->src, key.af);
7071 		PF_ACPY(&key.addr[0], pd->dst, key.af);
7072 		key.port[1] = key.port[0] = 0;
7073 	}
7074 
7075 	STATE_LOOKUP(kif, &key, *state, pd);
7076 
7077 	if (pd->dir == (*state)->direction) {
7078 		src = &(*state)->src;
7079 		dst = &(*state)->dst;
7080 		psrc = PF_PEER_SRC;
7081 		pdst = PF_PEER_DST;
7082 	} else {
7083 		src = &(*state)->dst;
7084 		dst = &(*state)->src;
7085 		psrc = PF_PEER_DST;
7086 		pdst = PF_PEER_SRC;
7087 	}
7088 
7089 	/* update states */
7090 	if (src->state < PFOTHERS_SINGLE)
7091 		pf_set_protostate(*state, psrc, PFOTHERS_SINGLE);
7092 	if (dst->state == PFOTHERS_SINGLE)
7093 		pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE);
7094 
7095 	/* update expire time */
7096 	(*state)->expire = pf_get_uptime();
7097 	if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
7098 		(*state)->timeout = PFTM_OTHER_MULTIPLE;
7099 	else
7100 		(*state)->timeout = PFTM_OTHER_SINGLE;
7101 
7102 	/* translate source/destination address, if necessary */
7103 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
7104 		struct pf_state_key *nk = (*state)->key[pd->didx];
7105 
7106 		KASSERT(nk, ("%s: nk is null", __func__));
7107 		KASSERT(pd, ("%s: pd is null", __func__));
7108 		KASSERT(pd->src, ("%s: pd->src is null", __func__));
7109 		KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
7110 		switch (pd->af) {
7111 #ifdef INET
7112 		case AF_INET:
7113 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
7114 				pf_change_a(&pd->src->v4.s_addr,
7115 				    pd->ip_sum,
7116 				    nk->addr[pd->sidx].v4.s_addr,
7117 				    0);
7118 
7119 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
7120 				pf_change_a(&pd->dst->v4.s_addr,
7121 				    pd->ip_sum,
7122 				    nk->addr[pd->didx].v4.s_addr,
7123 				    0);
7124 
7125 			break;
7126 #endif /* INET */
7127 #ifdef INET6
7128 		case AF_INET6:
7129 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
7130 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
7131 
7132 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
7133 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
7134 #endif /* INET6 */
7135 		}
7136 	}
7137 	return (PF_PASS);
7138 }
7139 
7140 /*
7141  * ipoff and off are measured from the start of the mbuf chain.
7142  * h must be at "ipoff" on the mbuf chain.
7143  */
7144 void *
7145 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
7146     u_short *actionp, u_short *reasonp, sa_family_t af)
7147 {
7148 	switch (af) {
7149 #ifdef INET
7150 	case AF_INET: {
7151 		struct ip	*h = mtod(m, struct ip *);
7152 		u_int16_t	 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
7153 
7154 		if (fragoff) {
7155 			if (fragoff >= len)
7156 				ACTION_SET(actionp, PF_PASS);
7157 			else {
7158 				ACTION_SET(actionp, PF_DROP);
7159 				REASON_SET(reasonp, PFRES_FRAG);
7160 			}
7161 			return (NULL);
7162 		}
7163 		if (m->m_pkthdr.len < off + len ||
7164 		    ntohs(h->ip_len) < off + len) {
7165 			ACTION_SET(actionp, PF_DROP);
7166 			REASON_SET(reasonp, PFRES_SHORT);
7167 			return (NULL);
7168 		}
7169 		break;
7170 	}
7171 #endif /* INET */
7172 #ifdef INET6
7173 	case AF_INET6: {
7174 		struct ip6_hdr	*h = mtod(m, struct ip6_hdr *);
7175 
7176 		if (m->m_pkthdr.len < off + len ||
7177 		    (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
7178 		    (unsigned)(off + len)) {
7179 			ACTION_SET(actionp, PF_DROP);
7180 			REASON_SET(reasonp, PFRES_SHORT);
7181 			return (NULL);
7182 		}
7183 		break;
7184 	}
7185 #endif /* INET6 */
7186 	}
7187 	m_copydata(m, off, len, p);
7188 	return (p);
7189 }
7190 
7191 int
7192 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif,
7193     int rtableid)
7194 {
7195 	struct ifnet		*ifp;
7196 
7197 	/*
7198 	 * Skip check for addresses with embedded interface scope,
7199 	 * as they would always match anyway.
7200 	 */
7201 	if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
7202 		return (1);
7203 
7204 	if (af != AF_INET && af != AF_INET6)
7205 		return (0);
7206 
7207 	if (kif == V_pfi_all)
7208 		return (1);
7209 
7210 	/* Skip checks for ipsec interfaces */
7211 	if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
7212 		return (1);
7213 
7214 	ifp = (kif != NULL) ? kif->pfik_ifp : NULL;
7215 
7216 	switch (af) {
7217 #ifdef INET6
7218 	case AF_INET6:
7219 		return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE,
7220 		    ifp));
7221 #endif
7222 #ifdef INET
7223 	case AF_INET:
7224 		return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE,
7225 		    ifp));
7226 #endif
7227 	}
7228 
7229 	return (0);
7230 }
7231 
7232 #ifdef INET
7233 static void
7234 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp,
7235     struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp)
7236 {
7237 	struct mbuf		*m0, *m1, *md;
7238 	struct sockaddr_in	dst;
7239 	struct ip		*ip;
7240 	struct pfi_kkif		*nkif = NULL;
7241 	struct ifnet		*ifp = NULL;
7242 	struct pf_addr		 naddr;
7243 	struct pf_ksrc_node	*sn = NULL;
7244 	int			 error = 0;
7245 	uint16_t		 ip_len, ip_off;
7246 	uint16_t		 tmp;
7247 	int			 r_rt, r_dir;
7248 
7249 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
7250 
7251 	if (s) {
7252 		r_rt = s->rt;
7253 		r_dir = s->direction;
7254 	} else {
7255 		r_rt = r->rt;
7256 		r_dir = r->direction;
7257 	}
7258 
7259 	KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT ||
7260 	    r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction",
7261 	    __func__));
7262 
7263 	if ((pd->pf_mtag == NULL &&
7264 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
7265 	    pd->pf_mtag->routed++ > 3) {
7266 		m0 = *m;
7267 		*m = NULL;
7268 		goto bad_locked;
7269 	}
7270 
7271 	if (r_rt == PF_DUPTO) {
7272 		if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) {
7273 			if (s == NULL) {
7274 				ifp = r->rpool.cur->kif ?
7275 				    r->rpool.cur->kif->pfik_ifp : NULL;
7276 			} else {
7277 				ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7278 				/* If pfsync'd */
7279 				if (ifp == NULL && r->rpool.cur != NULL)
7280 					ifp = r->rpool.cur->kif ?
7281 					    r->rpool.cur->kif->pfik_ifp : NULL;
7282 				PF_STATE_UNLOCK(s);
7283 			}
7284 			if (ifp == oifp) {
7285 				/* When the 2nd interface is not skipped */
7286 				return;
7287 			} else {
7288 				m0 = *m;
7289 				*m = NULL;
7290 				goto bad;
7291 			}
7292 		} else {
7293 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED;
7294 			if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) {
7295 				if (s)
7296 					PF_STATE_UNLOCK(s);
7297 				return;
7298 			}
7299 		}
7300 	} else {
7301 		if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) {
7302 			pf_dummynet(pd, s, r, m);
7303 			if (s)
7304 				PF_STATE_UNLOCK(s);
7305 			return;
7306 		}
7307 		m0 = *m;
7308 	}
7309 
7310 	ip = mtod(m0, struct ip *);
7311 
7312 	bzero(&dst, sizeof(dst));
7313 	dst.sin_family = AF_INET;
7314 	dst.sin_len = sizeof(dst);
7315 	dst.sin_addr = ip->ip_dst;
7316 
7317 	bzero(&naddr, sizeof(naddr));
7318 
7319 	if (s == NULL) {
7320 		if (TAILQ_EMPTY(&r->rpool.list)) {
7321 			DPFPRINTF(PF_DEBUG_URGENT,
7322 			    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
7323 			goto bad_locked;
7324 		}
7325 		pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
7326 		    &naddr, &nkif, NULL, &sn);
7327 		if (!PF_AZERO(&naddr, AF_INET))
7328 			dst.sin_addr.s_addr = naddr.v4.s_addr;
7329 		ifp = nkif ? nkif->pfik_ifp : NULL;
7330 	} else {
7331 		struct pfi_kkif *kif;
7332 
7333 		if (!PF_AZERO(&s->rt_addr, AF_INET))
7334 			dst.sin_addr.s_addr =
7335 			    s->rt_addr.v4.s_addr;
7336 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7337 		kif = s->rt_kif;
7338 		/* If pfsync'd */
7339 		if (ifp == NULL && r->rpool.cur != NULL) {
7340 			ifp = r->rpool.cur->kif ?
7341 			    r->rpool.cur->kif->pfik_ifp : NULL;
7342 			kif = r->rpool.cur->kif;
7343 		}
7344 		if (ifp != NULL && kif != NULL &&
7345 		    r->rule_flag & PFRULE_IFBOUND &&
7346 		    r->rt == PF_REPLYTO &&
7347 		    s->kif == V_pfi_all) {
7348 			s->kif = kif;
7349 			s->orig_kif = oifp->if_pf_kif;
7350 		}
7351 
7352 		PF_STATE_UNLOCK(s);
7353 	}
7354 
7355 	if (ifp == NULL)
7356 		goto bad;
7357 
7358 	if (pd->dir == PF_IN) {
7359 		if (pf_test(PF_OUT, PFIL_FWD, ifp, &m0, inp, &pd->act) != PF_PASS)
7360 			goto bad;
7361 		else if (m0 == NULL)
7362 			goto done;
7363 		if (m0->m_len < sizeof(struct ip)) {
7364 			DPFPRINTF(PF_DEBUG_URGENT,
7365 			    ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
7366 			goto bad;
7367 		}
7368 		ip = mtod(m0, struct ip *);
7369 	}
7370 
7371 	if (ifp->if_flags & IFF_LOOPBACK)
7372 		m0->m_flags |= M_SKIP_FIREWALL;
7373 
7374 	ip_len = ntohs(ip->ip_len);
7375 	ip_off = ntohs(ip->ip_off);
7376 
7377 	/* Copied from FreeBSD 10.0-CURRENT ip_output. */
7378 	m0->m_pkthdr.csum_flags |= CSUM_IP;
7379 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
7380 		in_delayed_cksum(m0);
7381 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
7382 	}
7383 	if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
7384 		pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2));
7385 		m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
7386 	}
7387 
7388 	if (pd->dir == PF_IN) {
7389 		/*
7390 		 * Make sure dummynet gets the correct direction, in case it needs to
7391 		 * re-inject later.
7392 		 */
7393 		pd->dir = PF_OUT;
7394 
7395 		/*
7396 		 * The following processing is actually the rest of the inbound processing, even
7397 		 * though we've marked it as outbound (so we don't look through dummynet) and it
7398 		 * happens after the outbound processing (pf_test(PF_OUT) above).
7399 		 * Swap the dummynet pipe numbers, because it's going to come to the wrong
7400 		 * conclusion about what direction it's processing, and we can't fix it or it
7401 		 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect
7402 		 * decision will pick the right pipe, and everything will mostly work as expected.
7403 		 */
7404 		tmp = pd->act.dnrpipe;
7405 		pd->act.dnrpipe = pd->act.dnpipe;
7406 		pd->act.dnpipe = tmp;
7407 	}
7408 
7409 	/*
7410 	 * If small enough for interface, or the interface will take
7411 	 * care of the fragmentation for us, we can just send directly.
7412 	 */
7413 	if (ip_len <= ifp->if_mtu ||
7414 	    (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
7415 		ip->ip_sum = 0;
7416 		if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
7417 			ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
7418 			m0->m_pkthdr.csum_flags &= ~CSUM_IP;
7419 		}
7420 		m_clrprotoflags(m0);	/* Avoid confusing lower layers. */
7421 
7422 		md = m0;
7423 		error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md);
7424 		if (md != NULL)
7425 			error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL);
7426 		goto done;
7427 	}
7428 
7429 	/* Balk when DF bit is set or the interface didn't support TSO. */
7430 	if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
7431 		error = EMSGSIZE;
7432 		KMOD_IPSTAT_INC(ips_cantfrag);
7433 		if (r_rt != PF_DUPTO) {
7434 			if (s && pd->nat_rule != NULL)
7435 				PACKET_UNDO_NAT(m0, pd,
7436 				    (ip->ip_hl << 2) + (ip_off & IP_OFFMASK),
7437 				    s);
7438 
7439 			icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
7440 			    ifp->if_mtu);
7441 			goto done;
7442 		} else
7443 			goto bad;
7444 	}
7445 
7446 	error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
7447 	if (error)
7448 		goto bad;
7449 
7450 	for (; m0; m0 = m1) {
7451 		m1 = m0->m_nextpkt;
7452 		m0->m_nextpkt = NULL;
7453 		if (error == 0) {
7454 			m_clrprotoflags(m0);
7455 			md = m0;
7456 			pd->pf_mtag = pf_find_mtag(md);
7457 			error = pf_dummynet_route(pd, s, r, ifp,
7458 			    sintosa(&dst), &md);
7459 			if (md != NULL)
7460 				error = (*ifp->if_output)(ifp, md,
7461 				    sintosa(&dst), NULL);
7462 		} else
7463 			m_freem(m0);
7464 	}
7465 
7466 	if (error == 0)
7467 		KMOD_IPSTAT_INC(ips_fragmented);
7468 
7469 done:
7470 	if (r_rt != PF_DUPTO)
7471 		*m = NULL;
7472 	return;
7473 
7474 bad_locked:
7475 	if (s)
7476 		PF_STATE_UNLOCK(s);
7477 bad:
7478 	m_freem(m0);
7479 	goto done;
7480 }
7481 #endif /* INET */
7482 
7483 #ifdef INET6
7484 static void
7485 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp,
7486     struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp)
7487 {
7488 	struct mbuf		*m0, *md;
7489 	struct sockaddr_in6	dst;
7490 	struct ip6_hdr		*ip6;
7491 	struct pfi_kkif		*nkif = NULL;
7492 	struct ifnet		*ifp = NULL;
7493 	struct pf_addr		 naddr;
7494 	struct pf_ksrc_node	*sn = NULL;
7495 	int			 r_rt, r_dir;
7496 
7497 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
7498 
7499 	if (s) {
7500 		r_rt = s->rt;
7501 		r_dir = s->direction;
7502 	} else {
7503 		r_rt = r->rt;
7504 		r_dir = r->direction;
7505 	}
7506 
7507 	KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT ||
7508 	    r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction",
7509 	    __func__));
7510 
7511 	if ((pd->pf_mtag == NULL &&
7512 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
7513 	    pd->pf_mtag->routed++ > 3) {
7514 		m0 = *m;
7515 		*m = NULL;
7516 		goto bad_locked;
7517 	}
7518 
7519 	if (r_rt == PF_DUPTO) {
7520 		if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) {
7521 			if (s == NULL) {
7522 				ifp = r->rpool.cur->kif ?
7523 				    r->rpool.cur->kif->pfik_ifp : NULL;
7524 			} else {
7525 				ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7526 				/* If pfsync'd */
7527 				if (ifp == NULL && r->rpool.cur != NULL)
7528 					ifp = r->rpool.cur->kif ?
7529 					    r->rpool.cur->kif->pfik_ifp : NULL;
7530 				PF_STATE_UNLOCK(s);
7531 			}
7532 			if (ifp == oifp) {
7533 				/* When the 2nd interface is not skipped */
7534 				return;
7535 			} else {
7536 				m0 = *m;
7537 				*m = NULL;
7538 				goto bad;
7539 			}
7540 		} else {
7541 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED;
7542 			if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) {
7543 				if (s)
7544 					PF_STATE_UNLOCK(s);
7545 				return;
7546 			}
7547 		}
7548 	} else {
7549 		if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) {
7550 			pf_dummynet(pd, s, r, m);
7551 			if (s)
7552 				PF_STATE_UNLOCK(s);
7553 			return;
7554 		}
7555 		m0 = *m;
7556 	}
7557 
7558 	ip6 = mtod(m0, struct ip6_hdr *);
7559 
7560 	bzero(&dst, sizeof(dst));
7561 	dst.sin6_family = AF_INET6;
7562 	dst.sin6_len = sizeof(dst);
7563 	dst.sin6_addr = ip6->ip6_dst;
7564 
7565 	bzero(&naddr, sizeof(naddr));
7566 
7567 	if (s == NULL) {
7568 		if (TAILQ_EMPTY(&r->rpool.list)) {
7569 			DPFPRINTF(PF_DEBUG_URGENT,
7570 			    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
7571 			goto bad_locked;
7572 		}
7573 		pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
7574 		    &naddr, &nkif, NULL, &sn);
7575 		if (!PF_AZERO(&naddr, AF_INET6))
7576 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
7577 			    &naddr, AF_INET6);
7578 		ifp = nkif ? nkif->pfik_ifp : NULL;
7579 	} else {
7580 		struct pfi_kkif *kif;
7581 
7582 		if (!PF_AZERO(&s->rt_addr, AF_INET6))
7583 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
7584 			    &s->rt_addr, AF_INET6);
7585 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7586 		kif = s->rt_kif;
7587 		/* If pfsync'd */
7588 		if (ifp == NULL && r->rpool.cur != NULL) {
7589 			ifp = r->rpool.cur->kif ?
7590 			    r->rpool.cur->kif->pfik_ifp : NULL;
7591 			kif = r->rpool.cur->kif;
7592 		}
7593 		if (ifp != NULL && kif != NULL &&
7594 		    r->rule_flag & PFRULE_IFBOUND &&
7595 		    r->rt == PF_REPLYTO &&
7596 		    s->kif == V_pfi_all) {
7597 			s->kif = kif;
7598 			s->orig_kif = oifp->if_pf_kif;
7599 		}
7600 	}
7601 
7602 	if (s)
7603 		PF_STATE_UNLOCK(s);
7604 
7605 	if (ifp == NULL)
7606 		goto bad;
7607 
7608 	if (pd->dir == PF_IN) {
7609 		if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp, &pd->act) != PF_PASS)
7610 			goto bad;
7611 		else if (m0 == NULL)
7612 			goto done;
7613 		if (m0->m_len < sizeof(struct ip6_hdr)) {
7614 			DPFPRINTF(PF_DEBUG_URGENT,
7615 			    ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
7616 			    __func__));
7617 			goto bad;
7618 		}
7619 		ip6 = mtod(m0, struct ip6_hdr *);
7620 	}
7621 
7622 	if (ifp->if_flags & IFF_LOOPBACK)
7623 		m0->m_flags |= M_SKIP_FIREWALL;
7624 
7625 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
7626 	    ~ifp->if_hwassist) {
7627 		uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
7628 		in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
7629 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
7630 	}
7631 
7632 	/*
7633 	 * If the packet is too large for the outgoing interface,
7634 	 * send back an icmp6 error.
7635 	 */
7636 	if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
7637 		dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
7638 	if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) {
7639 		md = m0;
7640 		pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md);
7641 		if (md != NULL)
7642 			nd6_output_ifp(ifp, ifp, md, &dst, NULL);
7643 	}
7644 	else {
7645 		in6_ifstat_inc(ifp, ifs6_in_toobig);
7646 		if (r_rt != PF_DUPTO) {
7647 			if (s && pd->nat_rule != NULL)
7648 				PACKET_UNDO_NAT(m0, pd,
7649 				    ((caddr_t)ip6 - m0->m_data) +
7650 				    sizeof(struct ip6_hdr), s);
7651 
7652 			icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
7653 		} else
7654 			goto bad;
7655 	}
7656 
7657 done:
7658 	if (r_rt != PF_DUPTO)
7659 		*m = NULL;
7660 	return;
7661 
7662 bad_locked:
7663 	if (s)
7664 		PF_STATE_UNLOCK(s);
7665 bad:
7666 	m_freem(m0);
7667 	goto done;
7668 }
7669 #endif /* INET6 */
7670 
7671 /*
7672  * FreeBSD supports cksum offloads for the following drivers.
7673  *  em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4)
7674  *
7675  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
7676  *  network driver performed cksum including pseudo header, need to verify
7677  *   csum_data
7678  * CSUM_DATA_VALID :
7679  *  network driver performed cksum, needs to additional pseudo header
7680  *  cksum computation with partial csum_data(i.e. lack of H/W support for
7681  *  pseudo header, for instance sk(4) and possibly gem(4))
7682  *
7683  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
7684  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
7685  * TCP/UDP layer.
7686  * Also, set csum_data to 0xffff to force cksum validation.
7687  */
7688 static int
7689 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
7690 {
7691 	u_int16_t sum = 0;
7692 	int hw_assist = 0;
7693 	struct ip *ip;
7694 
7695 	if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
7696 		return (1);
7697 	if (m->m_pkthdr.len < off + len)
7698 		return (1);
7699 
7700 	switch (p) {
7701 	case IPPROTO_TCP:
7702 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
7703 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
7704 				sum = m->m_pkthdr.csum_data;
7705 			} else {
7706 				ip = mtod(m, struct ip *);
7707 				sum = in_pseudo(ip->ip_src.s_addr,
7708 				ip->ip_dst.s_addr, htonl((u_short)len +
7709 				m->m_pkthdr.csum_data + IPPROTO_TCP));
7710 			}
7711 			sum ^= 0xffff;
7712 			++hw_assist;
7713 		}
7714 		break;
7715 	case IPPROTO_UDP:
7716 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
7717 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
7718 				sum = m->m_pkthdr.csum_data;
7719 			} else {
7720 				ip = mtod(m, struct ip *);
7721 				sum = in_pseudo(ip->ip_src.s_addr,
7722 				ip->ip_dst.s_addr, htonl((u_short)len +
7723 				m->m_pkthdr.csum_data + IPPROTO_UDP));
7724 			}
7725 			sum ^= 0xffff;
7726 			++hw_assist;
7727 		}
7728 		break;
7729 	case IPPROTO_ICMP:
7730 #ifdef INET6
7731 	case IPPROTO_ICMPV6:
7732 #endif /* INET6 */
7733 		break;
7734 	default:
7735 		return (1);
7736 	}
7737 
7738 	if (!hw_assist) {
7739 		switch (af) {
7740 		case AF_INET:
7741 			if (p == IPPROTO_ICMP) {
7742 				if (m->m_len < off)
7743 					return (1);
7744 				m->m_data += off;
7745 				m->m_len -= off;
7746 				sum = in_cksum(m, len);
7747 				m->m_data -= off;
7748 				m->m_len += off;
7749 			} else {
7750 				if (m->m_len < sizeof(struct ip))
7751 					return (1);
7752 				sum = in4_cksum(m, p, off, len);
7753 			}
7754 			break;
7755 #ifdef INET6
7756 		case AF_INET6:
7757 			if (m->m_len < sizeof(struct ip6_hdr))
7758 				return (1);
7759 			sum = in6_cksum(m, p, off, len);
7760 			break;
7761 #endif /* INET6 */
7762 		default:
7763 			return (1);
7764 		}
7765 	}
7766 	if (sum) {
7767 		switch (p) {
7768 		case IPPROTO_TCP:
7769 		    {
7770 			KMOD_TCPSTAT_INC(tcps_rcvbadsum);
7771 			break;
7772 		    }
7773 		case IPPROTO_UDP:
7774 		    {
7775 			KMOD_UDPSTAT_INC(udps_badsum);
7776 			break;
7777 		    }
7778 #ifdef INET
7779 		case IPPROTO_ICMP:
7780 		    {
7781 			KMOD_ICMPSTAT_INC(icps_checksum);
7782 			break;
7783 		    }
7784 #endif
7785 #ifdef INET6
7786 		case IPPROTO_ICMPV6:
7787 		    {
7788 			KMOD_ICMP6STAT_INC(icp6s_checksum);
7789 			break;
7790 		    }
7791 #endif /* INET6 */
7792 		}
7793 		return (1);
7794 	} else {
7795 		if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
7796 			m->m_pkthdr.csum_flags |=
7797 			    (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
7798 			m->m_pkthdr.csum_data = 0xffff;
7799 		}
7800 	}
7801 	return (0);
7802 }
7803 
7804 static bool
7805 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r,
7806     const struct pf_kstate *s, struct ip_fw_args *dnflow)
7807 {
7808 	int dndir = r->direction;
7809 
7810 	if (s && dndir == PF_INOUT) {
7811 		dndir = s->direction;
7812 	} else if (dndir == PF_INOUT) {
7813 		/* Assume primary direction. Happens when we've set dnpipe in
7814 		 * the ethernet level code. */
7815 		dndir = pd->dir;
7816 	}
7817 
7818 	if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED)
7819 		return (false);
7820 
7821 	memset(dnflow, 0, sizeof(*dnflow));
7822 
7823 	if (pd->dport != NULL)
7824 		dnflow->f_id.dst_port = ntohs(*pd->dport);
7825 	if (pd->sport != NULL)
7826 		dnflow->f_id.src_port = ntohs(*pd->sport);
7827 
7828 	if (pd->dir == PF_IN)
7829 		dnflow->flags |= IPFW_ARGS_IN;
7830 	else
7831 		dnflow->flags |= IPFW_ARGS_OUT;
7832 
7833 	if (pd->dir != dndir && pd->act.dnrpipe) {
7834 		dnflow->rule.info = pd->act.dnrpipe;
7835 	}
7836 	else if (pd->dir == dndir && pd->act.dnpipe) {
7837 		dnflow->rule.info = pd->act.dnpipe;
7838 	}
7839 	else {
7840 		return (false);
7841 	}
7842 
7843 	dnflow->rule.info |= IPFW_IS_DUMMYNET;
7844 	if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE)
7845 		dnflow->rule.info |= IPFW_IS_PIPE;
7846 
7847 	dnflow->f_id.proto = pd->proto;
7848 	dnflow->f_id.extra = dnflow->rule.info;
7849 	switch (pd->af) {
7850 	case AF_INET:
7851 		dnflow->f_id.addr_type = 4;
7852 		dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr);
7853 		dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr);
7854 		break;
7855 	case AF_INET6:
7856 		dnflow->flags |= IPFW_ARGS_IP6;
7857 		dnflow->f_id.addr_type = 6;
7858 		dnflow->f_id.src_ip6 = pd->src->v6;
7859 		dnflow->f_id.dst_ip6 = pd->dst->v6;
7860 		break;
7861 	default:
7862 		panic("Invalid AF");
7863 		break;
7864 	}
7865 
7866 	return (true);
7867 }
7868 
7869 int
7870 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0,
7871     struct inpcb *inp)
7872 {
7873 	struct pfi_kkif		*kif;
7874 	struct mbuf		*m = *m0;
7875 
7876 	M_ASSERTPKTHDR(m);
7877 	MPASS(ifp->if_vnet == curvnet);
7878 	NET_EPOCH_ASSERT();
7879 
7880 	if (!V_pf_status.running)
7881 		return (PF_PASS);
7882 
7883 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
7884 
7885 	if (kif == NULL) {
7886 		DPFPRINTF(PF_DEBUG_URGENT,
7887 		    ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname));
7888 		return (PF_DROP);
7889 	}
7890 	if (kif->pfik_flags & PFI_IFLAG_SKIP)
7891 		return (PF_PASS);
7892 
7893 	if (m->m_flags & M_SKIP_FIREWALL)
7894 		return (PF_PASS);
7895 
7896 	/* Stateless! */
7897 	return (pf_test_eth_rule(dir, kif, m0));
7898 }
7899 
7900 static __inline void
7901 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag)
7902 {
7903 	struct m_tag *mtag;
7904 
7905 	pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET;
7906 
7907 	/* dummynet adds this tag, but pf does not need it,
7908 	 * and keeping it creates unexpected behavior,
7909 	 * e.g. in case of divert(4) usage right after dummynet. */
7910 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
7911 	if (mtag != NULL)
7912 		m_tag_delete(m, mtag);
7913 }
7914 
7915 static int
7916 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s,
7917     struct pf_krule *r, struct mbuf **m0)
7918 {
7919 	return (pf_dummynet_route(pd, s, r, NULL, NULL, m0));
7920 }
7921 
7922 static int
7923 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s,
7924     struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa,
7925     struct mbuf **m0)
7926 {
7927 	NET_EPOCH_ASSERT();
7928 
7929 	if (pd->act.dnpipe || pd->act.dnrpipe) {
7930 		struct ip_fw_args dnflow;
7931 		if (ip_dn_io_ptr == NULL) {
7932 			m_freem(*m0);
7933 			*m0 = NULL;
7934 			return (ENOMEM);
7935 		}
7936 
7937 		if (pd->pf_mtag == NULL &&
7938 		    ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) {
7939 			m_freem(*m0);
7940 			*m0 = NULL;
7941 			return (ENOMEM);
7942 		}
7943 
7944 		if (ifp != NULL) {
7945 			pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO;
7946 
7947 			pd->pf_mtag->if_index = ifp->if_index;
7948 			pd->pf_mtag->if_idxgen = ifp->if_idxgen;
7949 
7950 			MPASS(sa != NULL);
7951 
7952 			if (pd->af == AF_INET)
7953 				memcpy(&pd->pf_mtag->dst, sa,
7954 				    sizeof(struct sockaddr_in));
7955 			else
7956 				memcpy(&pd->pf_mtag->dst, sa,
7957 				    sizeof(struct sockaddr_in6));
7958 		}
7959 
7960 		if (s != NULL && s->nat_rule.ptr != NULL &&
7961 		    s->nat_rule.ptr->action == PF_RDR &&
7962 		    (
7963 #ifdef INET
7964 		    (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) ||
7965 #endif
7966 		    (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) {
7967 			/*
7968 			 * If we're redirecting to loopback mark this packet
7969 			 * as being local. Otherwise it might get dropped
7970 			 * if dummynet re-injects.
7971 			 */
7972 			(*m0)->m_pkthdr.rcvif = V_loif;
7973 		}
7974 
7975 		if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) {
7976 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET;
7977 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED;
7978 			ip_dn_io_ptr(m0, &dnflow);
7979 			if (*m0 != NULL) {
7980 				pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
7981 				pf_dummynet_flag_remove(*m0, pd->pf_mtag);
7982 			}
7983 		}
7984 	}
7985 
7986 	return (0);
7987 }
7988 
7989 #ifdef INET
7990 int
7991 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0,
7992     struct inpcb *inp, struct pf_rule_actions *default_actions)
7993 {
7994 	struct pfi_kkif		*kif;
7995 	u_short			 action, reason = 0;
7996 	struct mbuf		*m = *m0;
7997 	struct ip		*h = NULL;
7998 	struct m_tag		*mtag;
7999 	struct pf_krule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
8000 	struct pf_kstate	*s = NULL;
8001 	struct pf_kruleset	*ruleset = NULL;
8002 	struct pf_pdesc		 pd;
8003 	int			 off, dirndx, use_2nd_queue = 0;
8004 	uint16_t		 tag;
8005 	uint8_t			 rt;
8006 
8007 	PF_RULES_RLOCK_TRACKER;
8008 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir));
8009 	M_ASSERTPKTHDR(m);
8010 
8011 	if (!V_pf_status.running)
8012 		return (PF_PASS);
8013 
8014 	PF_RULES_RLOCK();
8015 
8016 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
8017 
8018 	if (__predict_false(kif == NULL)) {
8019 		DPFPRINTF(PF_DEBUG_URGENT,
8020 		    ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
8021 		PF_RULES_RUNLOCK();
8022 		return (PF_DROP);
8023 	}
8024 	if (kif->pfik_flags & PFI_IFLAG_SKIP) {
8025 		PF_RULES_RUNLOCK();
8026 		return (PF_PASS);
8027 	}
8028 
8029 	if (m->m_flags & M_SKIP_FIREWALL) {
8030 		PF_RULES_RUNLOCK();
8031 		return (PF_PASS);
8032 	}
8033 
8034 	memset(&pd, 0, sizeof(pd));
8035 	TAILQ_INIT(&pd.sctp_multihome_jobs);
8036 	if (default_actions != NULL)
8037 		memcpy(&pd.act, default_actions, sizeof(pd.act));
8038 	pd.pf_mtag = pf_find_mtag(m);
8039 
8040 	if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) {
8041 		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
8042 
8043 		ifp = ifnet_byindexgen(pd.pf_mtag->if_index,
8044 		    pd.pf_mtag->if_idxgen);
8045 		if (ifp == NULL || ifp->if_flags & IFF_DYING) {
8046 			PF_RULES_RUNLOCK();
8047 			m_freem(*m0);
8048 			*m0 = NULL;
8049 			return (PF_PASS);
8050 		}
8051 		PF_RULES_RUNLOCK();
8052 		(ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL);
8053 		*m0 = NULL;
8054 		return (PF_PASS);
8055 	}
8056 
8057 	if (pd.pf_mtag && pd.pf_mtag->dnpipe) {
8058 		pd.act.dnpipe = pd.pf_mtag->dnpipe;
8059 		pd.act.flags = pd.pf_mtag->dnflags;
8060 	}
8061 
8062 	if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL &&
8063 	    pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
8064 		/* Dummynet re-injects packets after they've
8065 		 * completed their delay. We've already
8066 		 * processed them, so pass unconditionally. */
8067 
8068 		/* But only once. We may see the packet multiple times (e.g.
8069 		 * PFIL_IN/PFIL_OUT). */
8070 		pf_dummynet_flag_remove(m, pd.pf_mtag);
8071 		PF_RULES_RUNLOCK();
8072 
8073 		return (PF_PASS);
8074 	}
8075 
8076 	pd.sport = pd.dport = NULL;
8077 	pd.proto_sum = NULL;
8078 	pd.dir = dir;
8079 	pd.sidx = (dir == PF_IN) ? 0 : 1;
8080 	pd.didx = (dir == PF_IN) ? 1 : 0;
8081 	pd.af = AF_INET;
8082 	pd.act.rtableid = -1;
8083 
8084 	h = mtod(m, struct ip *);
8085 	off = h->ip_hl << 2;
8086 
8087 	if (__predict_false(ip_divert_ptr != NULL) &&
8088 	    ((mtag = m_tag_locate(m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) {
8089 		struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1);
8090 		if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) ||
8091 		    (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) {
8092 			if (pd.pf_mtag == NULL &&
8093 			    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8094 				action = PF_DROP;
8095 				goto done;
8096 			}
8097 			pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED;
8098 		}
8099 		if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) {
8100 			m->m_flags |= M_FASTFWD_OURS;
8101 			pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT;
8102 		}
8103 		m_tag_delete(m, mtag);
8104 
8105 		mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
8106 		if (mtag != NULL)
8107 			m_tag_delete(m, mtag);
8108 	} else if (pf_normalize_ip(m0, kif, &reason, &pd) != PF_PASS) {
8109 		/* We do IP header normalization and packet reassembly here */
8110 		action = PF_DROP;
8111 		goto done;
8112 	}
8113 	m = *m0;	/* pf_normalize messes with m0 */
8114 	h = mtod(m, struct ip *);
8115 
8116 	off = h->ip_hl << 2;
8117 	if (off < (int)sizeof(struct ip)) {
8118 		action = PF_DROP;
8119 		REASON_SET(&reason, PFRES_SHORT);
8120 		pd.act.log = PF_LOG_FORCE;
8121 		goto done;
8122 	}
8123 
8124 	pd.src = (struct pf_addr *)&h->ip_src;
8125 	pd.dst = (struct pf_addr *)&h->ip_dst;
8126 	pd.ip_sum = &h->ip_sum;
8127 	pd.proto = h->ip_p;
8128 	pd.tos = h->ip_tos & ~IPTOS_ECN_MASK;
8129 	pd.tot_len = ntohs(h->ip_len);
8130 
8131 	/* handle fragments that didn't get reassembled by normalization */
8132 	if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
8133 		action = pf_test_fragment(&r, kif, m, h, &pd, &a, &ruleset);
8134 		goto done;
8135 	}
8136 
8137 	switch (h->ip_p) {
8138 	case IPPROTO_TCP: {
8139 		if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp),
8140 		    &action, &reason, AF_INET)) {
8141 			if (action != PF_PASS)
8142 				pd.act.log = PF_LOG_FORCE;
8143 			goto done;
8144 		}
8145 		pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2);
8146 
8147 		pd.sport = &pd.hdr.tcp.th_sport;
8148 		pd.dport = &pd.hdr.tcp.th_dport;
8149 
8150 		/* Respond to SYN with a syncookie. */
8151 		if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN &&
8152 		    pd.dir == PF_IN && pf_synflood_check(&pd)) {
8153 			pf_syncookie_send(m, off, &pd);
8154 			action = PF_DROP;
8155 			break;
8156 		}
8157 
8158 		if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0)
8159 			use_2nd_queue = 1;
8160 		action = pf_normalize_tcp(kif, m, 0, off, h, &pd);
8161 		if (action == PF_DROP)
8162 			goto done;
8163 		action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason);
8164 		if (action == PF_PASS) {
8165 			if (V_pfsync_update_state_ptr != NULL)
8166 				V_pfsync_update_state_ptr(s);
8167 			r = s->rule.ptr;
8168 			a = s->anchor.ptr;
8169 		} else if (s == NULL) {
8170 			/* Validate remote SYN|ACK, re-create original SYN if
8171 			 * valid. */
8172 			if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) ==
8173 			    TH_ACK && pf_syncookie_validate(&pd) &&
8174 			    pd.dir == PF_IN) {
8175 				struct mbuf *msyn;
8176 
8177 				msyn = pf_syncookie_recreate_syn(h->ip_ttl, off,
8178 				    &pd);
8179 				if (msyn == NULL) {
8180 					action = PF_DROP;
8181 					break;
8182 				}
8183 
8184 				action = pf_test(dir, pflags, ifp, &msyn, inp,
8185 				    &pd.act);
8186 				m_freem(msyn);
8187 				if (action != PF_PASS)
8188 					break;
8189 
8190 				action = pf_test_state_tcp(&s, kif, m, off, h,
8191 				    &pd, &reason);
8192 				if (action != PF_PASS || s == NULL) {
8193 					action = PF_DROP;
8194 					break;
8195 				}
8196 
8197 				s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1;
8198 				s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1;
8199 				pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST);
8200 				action = pf_synproxy(&pd, &s, &reason);
8201 				break;
8202 			} else {
8203 				action = pf_test_rule(&r, &s, kif, m, off, &pd,
8204 				    &a, &ruleset, inp);
8205 			}
8206 		}
8207 		break;
8208 	}
8209 
8210 	case IPPROTO_UDP: {
8211 		if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp),
8212 		    &action, &reason, AF_INET)) {
8213 			if (action != PF_PASS)
8214 				pd.act.log = PF_LOG_FORCE;
8215 			goto done;
8216 		}
8217 		pd.sport = &pd.hdr.udp.uh_sport;
8218 		pd.dport = &pd.hdr.udp.uh_dport;
8219 		if (pd.hdr.udp.uh_dport == 0 ||
8220 		    ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off ||
8221 		    ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) {
8222 			action = PF_DROP;
8223 			REASON_SET(&reason, PFRES_SHORT);
8224 			goto done;
8225 		}
8226 		action = pf_test_state_udp(&s, kif, m, off, h, &pd);
8227 		if (action == PF_PASS) {
8228 			if (V_pfsync_update_state_ptr != NULL)
8229 				V_pfsync_update_state_ptr(s);
8230 			r = s->rule.ptr;
8231 			a = s->anchor.ptr;
8232 		} else if (s == NULL)
8233 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8234 			    &a, &ruleset, inp);
8235 		break;
8236 	}
8237 
8238 	case IPPROTO_SCTP: {
8239 		if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp),
8240 		    &action, &reason, AF_INET)) {
8241 			if (action != PF_PASS)
8242 				pd.act.log |= PF_LOG_FORCE;
8243 			goto done;
8244 		}
8245 		pd.p_len = pd.tot_len - off;
8246 
8247 		pd.sport = &pd.hdr.sctp.src_port;
8248 		pd.dport = &pd.hdr.sctp.dest_port;
8249 		if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) {
8250 			action = PF_DROP;
8251 			REASON_SET(&reason, PFRES_SHORT);
8252 			goto done;
8253 		}
8254 		action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd);
8255 		if (action == PF_DROP)
8256 			goto done;
8257 		action = pf_test_state_sctp(&s, kif, m, off, h, &pd,
8258 		    &reason);
8259 		if (action == PF_PASS) {
8260 			if (V_pfsync_update_state_ptr != NULL)
8261 				V_pfsync_update_state_ptr(s);
8262 			r = s->rule.ptr;
8263 			a = s->anchor.ptr;
8264 		} else {
8265 			action = pf_test_rule(&r, &s, kif, m, off,
8266 			    &pd, &a, &ruleset, inp);
8267 		}
8268 		break;
8269 	}
8270 
8271 	case IPPROTO_ICMP: {
8272 		if (!pf_pull_hdr(m, off, &pd.hdr.icmp, ICMP_MINLEN,
8273 		    &action, &reason, AF_INET)) {
8274 			if (action != PF_PASS)
8275 				pd.act.log = PF_LOG_FORCE;
8276 			goto done;
8277 		}
8278 		action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason);
8279 		if (action == PF_PASS) {
8280 			if (V_pfsync_update_state_ptr != NULL)
8281 				V_pfsync_update_state_ptr(s);
8282 			r = s->rule.ptr;
8283 			a = s->anchor.ptr;
8284 		} else if (s == NULL)
8285 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8286 			    &a, &ruleset, inp);
8287 		break;
8288 	}
8289 
8290 #ifdef INET6
8291 	case IPPROTO_ICMPV6: {
8292 		action = PF_DROP;
8293 		DPFPRINTF(PF_DEBUG_MISC,
8294 		    ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
8295 		goto done;
8296 	}
8297 #endif
8298 
8299 	default:
8300 		action = pf_test_state_other(&s, kif, m, &pd);
8301 		if (action == PF_PASS) {
8302 			if (V_pfsync_update_state_ptr != NULL)
8303 				V_pfsync_update_state_ptr(s);
8304 			r = s->rule.ptr;
8305 			a = s->anchor.ptr;
8306 		} else if (s == NULL)
8307 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8308 			    &a, &ruleset, inp);
8309 		break;
8310 	}
8311 
8312 done:
8313 	PF_RULES_RUNLOCK();
8314 	if (action == PF_PASS && h->ip_hl > 5 &&
8315 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
8316 		action = PF_DROP;
8317 		REASON_SET(&reason, PFRES_IPOPTIONS);
8318 		pd.act.log = PF_LOG_FORCE;
8319 		DPFPRINTF(PF_DEBUG_MISC,
8320 		    ("pf: dropping packet with ip options\n"));
8321 	}
8322 
8323 	if (s) {
8324 		uint8_t log = pd.act.log;
8325 		memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions));
8326 		pd.act.log |= log;
8327 		tag = s->tag;
8328 		rt = s->rt;
8329 	} else {
8330 		tag = r->tag;
8331 		rt = r->rt;
8332 	}
8333 
8334 	if (tag > 0 && pf_tag_packet(m, &pd, tag)) {
8335 		action = PF_DROP;
8336 		REASON_SET(&reason, PFRES_MEMORY);
8337 	}
8338 
8339 	pf_scrub_ip(&m, &pd);
8340 	if (pd.proto == IPPROTO_TCP && pd.act.max_mss)
8341 		pf_normalize_mss(m, off, &pd);
8342 
8343 	if (pd.act.rtableid >= 0)
8344 		M_SETFIB(m, pd.act.rtableid);
8345 
8346 	if (pd.act.flags & PFSTATE_SETPRIO) {
8347 		if (pd.tos & IPTOS_LOWDELAY)
8348 			use_2nd_queue = 1;
8349 		if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) {
8350 			action = PF_DROP;
8351 			REASON_SET(&reason, PFRES_MEMORY);
8352 			pd.act.log = PF_LOG_FORCE;
8353 			DPFPRINTF(PF_DEBUG_MISC,
8354 			    ("pf: failed to allocate 802.1q mtag\n"));
8355 		}
8356 	}
8357 
8358 #ifdef ALTQ
8359 	if (action == PF_PASS && pd.act.qid) {
8360 		if (pd.pf_mtag == NULL &&
8361 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8362 			action = PF_DROP;
8363 			REASON_SET(&reason, PFRES_MEMORY);
8364 		} else {
8365 			if (s != NULL)
8366 				pd.pf_mtag->qid_hash = pf_state_hash(s);
8367 			if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY))
8368 				pd.pf_mtag->qid = pd.act.pqid;
8369 			else
8370 				pd.pf_mtag->qid = pd.act.qid;
8371 			/* Add hints for ecn. */
8372 			pd.pf_mtag->hdr = h;
8373 		}
8374 	}
8375 #endif /* ALTQ */
8376 
8377 	/*
8378 	 * connections redirected to loopback should not match sockets
8379 	 * bound specifically to loopback due to security implications,
8380 	 * see tcp_input() and in_pcblookup_listen().
8381 	 */
8382 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
8383 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
8384 	    (s->nat_rule.ptr->action == PF_RDR ||
8385 	    s->nat_rule.ptr->action == PF_BINAT) &&
8386 	    IN_LOOPBACK(ntohl(pd.dst->v4.s_addr)))
8387 		m->m_flags |= M_SKIP_FIREWALL;
8388 
8389 	if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS &&
8390 	    r->divert.port && !PACKET_LOOPED(&pd)) {
8391 		mtag = m_tag_alloc(MTAG_PF_DIVERT, 0,
8392 		    sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO);
8393 		if (mtag != NULL) {
8394 			((struct pf_divert_mtag *)(mtag+1))->port =
8395 			    ntohs(r->divert.port);
8396 			((struct pf_divert_mtag *)(mtag+1))->idir =
8397 			    (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN :
8398 			    PF_DIVERT_MTAG_DIR_OUT;
8399 
8400 			if (s)
8401 				PF_STATE_UNLOCK(s);
8402 
8403 			m_tag_prepend(m, mtag);
8404 			if (m->m_flags & M_FASTFWD_OURS) {
8405 				if (pd.pf_mtag == NULL &&
8406 				    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8407 					action = PF_DROP;
8408 					REASON_SET(&reason, PFRES_MEMORY);
8409 					pd.act.log = PF_LOG_FORCE;
8410 					DPFPRINTF(PF_DEBUG_MISC,
8411 					    ("pf: failed to allocate tag\n"));
8412 				} else {
8413 					pd.pf_mtag->flags |=
8414 					    PF_MTAG_FLAG_FASTFWD_OURS_PRESENT;
8415 					m->m_flags &= ~M_FASTFWD_OURS;
8416 				}
8417 			}
8418 			ip_divert_ptr(*m0, dir == PF_IN);
8419 			*m0 = NULL;
8420 
8421 			return (action);
8422 		} else {
8423 			/* XXX: ipfw has the same behaviour! */
8424 			action = PF_DROP;
8425 			REASON_SET(&reason, PFRES_MEMORY);
8426 			pd.act.log = PF_LOG_FORCE;
8427 			DPFPRINTF(PF_DEBUG_MISC,
8428 			    ("pf: failed to allocate divert tag\n"));
8429 		}
8430 	}
8431 	/* this flag will need revising if the pkt is forwarded */
8432 	if (pd.pf_mtag)
8433 		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED;
8434 
8435 	if (pd.act.log) {
8436 		struct pf_krule		*lr;
8437 		struct pf_krule_item	*ri;
8438 
8439 		if (s != NULL && s->nat_rule.ptr != NULL &&
8440 		    s->nat_rule.ptr->log & PF_LOG_ALL)
8441 			lr = s->nat_rule.ptr;
8442 		else
8443 			lr = r;
8444 
8445 		if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL)
8446 			PFLOG_PACKET(kif, m, AF_INET, action, reason, lr, a,
8447 			    ruleset, &pd, (s == NULL));
8448 		if (s) {
8449 			SLIST_FOREACH(ri, &s->match_rules, entry)
8450 				if (ri->r->log & PF_LOG_ALL)
8451 					PFLOG_PACKET(kif, m, AF_INET, action,
8452 					    reason, ri->r, a, ruleset, &pd, 0);
8453 		}
8454 	}
8455 
8456 	pf_counter_u64_critical_enter();
8457 	pf_counter_u64_add_protected(&kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS],
8458 	    pd.tot_len);
8459 	pf_counter_u64_add_protected(&kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS],
8460 	    1);
8461 
8462 	if (action == PF_PASS || r->action == PF_DROP) {
8463 		dirndx = (dir == PF_OUT);
8464 		pf_counter_u64_add_protected(&r->packets[dirndx], 1);
8465 		pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len);
8466 		pf_update_timestamp(r);
8467 
8468 		if (a != NULL) {
8469 			pf_counter_u64_add_protected(&a->packets[dirndx], 1);
8470 			pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len);
8471 		}
8472 		if (s != NULL) {
8473 			struct pf_krule_item	*ri;
8474 
8475 			if (s->nat_rule.ptr != NULL) {
8476 				pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx],
8477 				    1);
8478 				pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx],
8479 				    pd.tot_len);
8480 			}
8481 			if (s->src_node != NULL) {
8482 				counter_u64_add(s->src_node->packets[dirndx],
8483 				    1);
8484 				counter_u64_add(s->src_node->bytes[dirndx],
8485 				    pd.tot_len);
8486 			}
8487 			if (s->nat_src_node != NULL) {
8488 				counter_u64_add(s->nat_src_node->packets[dirndx],
8489 				    1);
8490 				counter_u64_add(s->nat_src_node->bytes[dirndx],
8491 				    pd.tot_len);
8492 			}
8493 			dirndx = (dir == s->direction) ? 0 : 1;
8494 			s->packets[dirndx]++;
8495 			s->bytes[dirndx] += pd.tot_len;
8496 			SLIST_FOREACH(ri, &s->match_rules, entry) {
8497 				pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1);
8498 				pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd.tot_len);
8499 			}
8500 		}
8501 		tr = r;
8502 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
8503 		if (nr != NULL && r == &V_pf_default_rule)
8504 			tr = nr;
8505 		if (tr->src.addr.type == PF_ADDR_TABLE)
8506 			pfr_update_stats(tr->src.addr.p.tbl,
8507 			    (s == NULL) ? pd.src :
8508 			    &s->key[(s->direction == PF_IN)]->
8509 				addr[(s->direction == PF_OUT)],
8510 			    pd.af, pd.tot_len, dir == PF_OUT,
8511 			    r->action == PF_PASS, tr->src.neg);
8512 		if (tr->dst.addr.type == PF_ADDR_TABLE)
8513 			pfr_update_stats(tr->dst.addr.p.tbl,
8514 			    (s == NULL) ? pd.dst :
8515 			    &s->key[(s->direction == PF_IN)]->
8516 				addr[(s->direction == PF_IN)],
8517 			    pd.af, pd.tot_len, dir == PF_OUT,
8518 			    r->action == PF_PASS, tr->dst.neg);
8519 	}
8520 	pf_counter_u64_critical_exit();
8521 
8522 	switch (action) {
8523 	case PF_SYNPROXY_DROP:
8524 		m_freem(*m0);
8525 	case PF_DEFER:
8526 		*m0 = NULL;
8527 		action = PF_PASS;
8528 		break;
8529 	case PF_DROP:
8530 		m_freem(*m0);
8531 		*m0 = NULL;
8532 		break;
8533 	default:
8534 		/* pf_route() returns unlocked. */
8535 		if (rt) {
8536 			pf_route(m0, r, kif->pfik_ifp, s, &pd, inp);
8537 			goto out;
8538 		}
8539 		if (pf_dummynet(&pd, s, r, m0) != 0) {
8540 			action = PF_DROP;
8541 			REASON_SET(&reason, PFRES_MEMORY);
8542 		}
8543 		break;
8544 	}
8545 
8546 	SDT_PROBE4(pf, ip, test, done, action, reason, r, s);
8547 
8548 	if (s && action != PF_DROP) {
8549 		if (!s->if_index_in && dir == PF_IN)
8550 			s->if_index_in = ifp->if_index;
8551 		else if (!s->if_index_out && dir == PF_OUT)
8552 			s->if_index_out = ifp->if_index;
8553 	}
8554 
8555 	if (s)
8556 		PF_STATE_UNLOCK(s);
8557 
8558 out:
8559 	pf_sctp_multihome_delayed(&pd, off, kif, s, action);
8560 
8561 	return (action);
8562 }
8563 #endif /* INET */
8564 
8565 #ifdef INET6
8566 int
8567 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp,
8568     struct pf_rule_actions *default_actions)
8569 {
8570 	struct pfi_kkif		*kif;
8571 	u_short			 action, reason = 0;
8572 	struct mbuf		*m = *m0, *n = NULL;
8573 	struct m_tag		*mtag;
8574 	struct ip6_hdr		*h = NULL;
8575 	struct pf_krule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
8576 	struct pf_kstate	*s = NULL;
8577 	struct pf_kruleset	*ruleset = NULL;
8578 	struct pf_pdesc		 pd;
8579 	int			 off, terminal = 0, dirndx, rh_cnt = 0, use_2nd_queue = 0;
8580 	uint16_t		 tag;
8581 	uint8_t			 rt;
8582 
8583 	PF_RULES_RLOCK_TRACKER;
8584 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir));
8585 	M_ASSERTPKTHDR(m);
8586 
8587 	if (!V_pf_status.running)
8588 		return (PF_PASS);
8589 
8590 	PF_RULES_RLOCK();
8591 
8592 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
8593 	if (__predict_false(kif == NULL)) {
8594 		DPFPRINTF(PF_DEBUG_URGENT,
8595 		    ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
8596 		PF_RULES_RUNLOCK();
8597 		return (PF_DROP);
8598 	}
8599 	if (kif->pfik_flags & PFI_IFLAG_SKIP) {
8600 		PF_RULES_RUNLOCK();
8601 		return (PF_PASS);
8602 	}
8603 
8604 	if (m->m_flags & M_SKIP_FIREWALL) {
8605 		PF_RULES_RUNLOCK();
8606 		return (PF_PASS);
8607 	}
8608 
8609 	/*
8610 	 * If we end up changing IP addresses (e.g. binat) the stack may get
8611 	 * confused and fail to send the icmp6 packet too big error. Just send
8612 	 * it here, before we do any NAT.
8613 	 */
8614 	if (dir == PF_OUT && pflags & PFIL_FWD && IN6_LINKMTU(ifp) < pf_max_frag_size(m)) {
8615 		PF_RULES_RUNLOCK();
8616 		*m0 = NULL;
8617 		icmp6_error(m, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp));
8618 		return (PF_DROP);
8619 	}
8620 
8621 	memset(&pd, 0, sizeof(pd));
8622 	TAILQ_INIT(&pd.sctp_multihome_jobs);
8623 	if (default_actions != NULL)
8624 		memcpy(&pd.act, default_actions, sizeof(pd.act));
8625 	pd.pf_mtag = pf_find_mtag(m);
8626 
8627 	if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) {
8628 		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
8629 
8630 		ifp = ifnet_byindexgen(pd.pf_mtag->if_index,
8631 		    pd.pf_mtag->if_idxgen);
8632 		if (ifp == NULL || ifp->if_flags & IFF_DYING) {
8633 			PF_RULES_RUNLOCK();
8634 			m_freem(*m0);
8635 			*m0 = NULL;
8636 			return (PF_PASS);
8637 		}
8638 		PF_RULES_RUNLOCK();
8639 		nd6_output_ifp(ifp, ifp, m,
8640                     (struct sockaddr_in6 *)&pd.pf_mtag->dst, NULL);
8641 		*m0 = NULL;
8642 		return (PF_PASS);
8643 	}
8644 
8645 	if (pd.pf_mtag && pd.pf_mtag->dnpipe) {
8646 		pd.act.dnpipe = pd.pf_mtag->dnpipe;
8647 		pd.act.flags = pd.pf_mtag->dnflags;
8648 	}
8649 
8650 	if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL &&
8651 	    pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
8652 		pf_dummynet_flag_remove(m, pd.pf_mtag);
8653 		/* Dummynet re-injects packets after they've
8654 		 * completed their delay. We've already
8655 		 * processed them, so pass unconditionally. */
8656 		PF_RULES_RUNLOCK();
8657 		return (PF_PASS);
8658 	}
8659 
8660 	pd.sport = pd.dport = NULL;
8661 	pd.ip_sum = NULL;
8662 	pd.proto_sum = NULL;
8663 	pd.dir = dir;
8664 	pd.sidx = (dir == PF_IN) ? 0 : 1;
8665 	pd.didx = (dir == PF_IN) ? 1 : 0;
8666 	pd.af = AF_INET6;
8667 	pd.act.rtableid = -1;
8668 
8669 	h = mtod(m, struct ip6_hdr *);
8670 	off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
8671 
8672 	/* We do IP header normalization and packet reassembly here */
8673 	if (pf_normalize_ip6(m0, kif, &reason, &pd) != PF_PASS) {
8674 		action = PF_DROP;
8675 		goto done;
8676 	}
8677 	m = *m0;	/* pf_normalize messes with m0 */
8678 	h = mtod(m, struct ip6_hdr *);
8679 	off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
8680 
8681 	/*
8682 	 * we do not support jumbogram.  if we keep going, zero ip6_plen
8683 	 * will do something bad, so drop the packet for now.
8684 	 */
8685 	if (htons(h->ip6_plen) == 0) {
8686 		action = PF_DROP;
8687 		REASON_SET(&reason, PFRES_NORM);	/*XXX*/
8688 		goto done;
8689 	}
8690 
8691 	pd.src = (struct pf_addr *)&h->ip6_src;
8692 	pd.dst = (struct pf_addr *)&h->ip6_dst;
8693 	pd.tos = IPV6_DSCP(h);
8694 	pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
8695 
8696 	pd.proto = h->ip6_nxt;
8697 	do {
8698 		switch (pd.proto) {
8699 		case IPPROTO_FRAGMENT:
8700 			action = pf_test_fragment(&r, kif, m, h, &pd, &a,
8701 			    &ruleset);
8702 			if (action == PF_DROP)
8703 				REASON_SET(&reason, PFRES_FRAG);
8704 			goto done;
8705 		case IPPROTO_ROUTING: {
8706 			struct ip6_rthdr rthdr;
8707 
8708 			if (rh_cnt++) {
8709 				DPFPRINTF(PF_DEBUG_MISC,
8710 				    ("pf: IPv6 more than one rthdr\n"));
8711 				action = PF_DROP;
8712 				REASON_SET(&reason, PFRES_IPOPTIONS);
8713 				pd.act.log = PF_LOG_FORCE;
8714 				goto done;
8715 			}
8716 			if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
8717 			    &reason, pd.af)) {
8718 				DPFPRINTF(PF_DEBUG_MISC,
8719 				    ("pf: IPv6 short rthdr\n"));
8720 				action = PF_DROP;
8721 				REASON_SET(&reason, PFRES_SHORT);
8722 				pd.act.log = PF_LOG_FORCE;
8723 				goto done;
8724 			}
8725 			if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
8726 				DPFPRINTF(PF_DEBUG_MISC,
8727 				    ("pf: IPv6 rthdr0\n"));
8728 				action = PF_DROP;
8729 				REASON_SET(&reason, PFRES_IPOPTIONS);
8730 				pd.act.log = PF_LOG_FORCE;
8731 				goto done;
8732 			}
8733 			/* FALLTHROUGH */
8734 		}
8735 		case IPPROTO_AH:
8736 		case IPPROTO_HOPOPTS:
8737 		case IPPROTO_DSTOPTS: {
8738 			/* get next header and header length */
8739 			struct ip6_ext	opt6;
8740 
8741 			if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
8742 			    NULL, &reason, pd.af)) {
8743 				DPFPRINTF(PF_DEBUG_MISC,
8744 				    ("pf: IPv6 short opt\n"));
8745 				action = PF_DROP;
8746 				pd.act.log = PF_LOG_FORCE;
8747 				goto done;
8748 			}
8749 			if (pd.proto == IPPROTO_AH)
8750 				off += (opt6.ip6e_len + 2) * 4;
8751 			else
8752 				off += (opt6.ip6e_len + 1) * 8;
8753 			pd.proto = opt6.ip6e_nxt;
8754 			/* goto the next header */
8755 			break;
8756 		}
8757 		default:
8758 			terminal++;
8759 			break;
8760 		}
8761 	} while (!terminal);
8762 
8763 	/* if there's no routing header, use unmodified mbuf for checksumming */
8764 	if (!n)
8765 		n = m;
8766 
8767 	switch (pd.proto) {
8768 	case IPPROTO_TCP: {
8769 		if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp),
8770 		    &action, &reason, AF_INET6)) {
8771 			if (action != PF_PASS)
8772 				pd.act.log |= PF_LOG_FORCE;
8773 			goto done;
8774 		}
8775 		pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2);
8776 		pd.sport = &pd.hdr.tcp.th_sport;
8777 		pd.dport = &pd.hdr.tcp.th_dport;
8778 
8779 		/* Respond to SYN with a syncookie. */
8780 		if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN &&
8781 		    pd.dir == PF_IN && pf_synflood_check(&pd)) {
8782 			pf_syncookie_send(m, off, &pd);
8783 			action = PF_DROP;
8784 			break;
8785 		}
8786 
8787 		action = pf_normalize_tcp(kif, m, 0, off, h, &pd);
8788 		if (action == PF_DROP)
8789 			goto done;
8790 		action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason);
8791 		if (action == PF_PASS) {
8792 			if (V_pfsync_update_state_ptr != NULL)
8793 				V_pfsync_update_state_ptr(s);
8794 			r = s->rule.ptr;
8795 			a = s->anchor.ptr;
8796 		} else if (s == NULL) {
8797 			/* Validate remote SYN|ACK, re-create original SYN if
8798 			 * valid. */
8799 			if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) ==
8800 			    TH_ACK && pf_syncookie_validate(&pd) &&
8801 			    pd.dir == PF_IN) {
8802 				struct mbuf *msyn;
8803 
8804 				msyn = pf_syncookie_recreate_syn(h->ip6_hlim,
8805 				    off, &pd);
8806 				if (msyn == NULL) {
8807 					action = PF_DROP;
8808 					break;
8809 				}
8810 
8811 				action = pf_test6(dir, pflags, ifp, &msyn, inp,
8812 				    &pd.act);
8813 				m_freem(msyn);
8814 				if (action != PF_PASS)
8815 					break;
8816 
8817 				action = pf_test_state_tcp(&s, kif, m, off, h,
8818 				    &pd, &reason);
8819 				if (action != PF_PASS || s == NULL) {
8820 					action = PF_DROP;
8821 					break;
8822 				}
8823 
8824 				s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1;
8825 				s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1;
8826 				pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST);
8827 
8828 				action = pf_synproxy(&pd, &s, &reason);
8829 				break;
8830 			} else {
8831 				action = pf_test_rule(&r, &s, kif, m, off, &pd,
8832 				    &a, &ruleset, inp);
8833 			}
8834 		}
8835 		break;
8836 	}
8837 
8838 	case IPPROTO_UDP: {
8839 		if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp),
8840 		    &action, &reason, AF_INET6)) {
8841 			if (action != PF_PASS)
8842 				pd.act.log |= PF_LOG_FORCE;
8843 			goto done;
8844 		}
8845 		pd.sport = &pd.hdr.udp.uh_sport;
8846 		pd.dport = &pd.hdr.udp.uh_dport;
8847 		if (pd.hdr.udp.uh_dport == 0 ||
8848 		    ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off ||
8849 		    ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) {
8850 			action = PF_DROP;
8851 			REASON_SET(&reason, PFRES_SHORT);
8852 			goto done;
8853 		}
8854 		action = pf_test_state_udp(&s, kif, m, off, h, &pd);
8855 		if (action == PF_PASS) {
8856 			if (V_pfsync_update_state_ptr != NULL)
8857 				V_pfsync_update_state_ptr(s);
8858 			r = s->rule.ptr;
8859 			a = s->anchor.ptr;
8860 		} else if (s == NULL)
8861 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8862 			    &a, &ruleset, inp);
8863 		break;
8864 	}
8865 
8866 	case IPPROTO_SCTP: {
8867 		if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp),
8868 		    &action, &reason, AF_INET6)) {
8869 			if (action != PF_PASS)
8870 				pd.act.log |= PF_LOG_FORCE;
8871 			goto done;
8872 		}
8873 		pd.sport = &pd.hdr.sctp.src_port;
8874 		pd.dport = &pd.hdr.sctp.dest_port;
8875 		if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) {
8876 			action = PF_DROP;
8877 			REASON_SET(&reason, PFRES_SHORT);
8878 			goto done;
8879 		}
8880 		action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd);
8881 		if (action == PF_DROP)
8882 			goto done;
8883 		action = pf_test_state_sctp(&s, kif, m, off, h, &pd,
8884 		    &reason);
8885 		if (action == PF_PASS) {
8886 			if (V_pfsync_update_state_ptr != NULL)
8887 				V_pfsync_update_state_ptr(s);
8888 			r = s->rule.ptr;
8889 			a = s->anchor.ptr;
8890 		} else {
8891 			action = pf_test_rule(&r, &s, kif, m, off,
8892 			    &pd, &a, &ruleset, inp);
8893 		}
8894 		break;
8895 	}
8896 
8897 	case IPPROTO_ICMP: {
8898 		action = PF_DROP;
8899 		DPFPRINTF(PF_DEBUG_MISC,
8900 		    ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
8901 		goto done;
8902 	}
8903 
8904 	case IPPROTO_ICMPV6: {
8905 		if (!pf_pull_hdr(m, off, &pd.hdr.icmp6, sizeof(pd.hdr.icmp6),
8906 		    &action, &reason, AF_INET6)) {
8907 			if (action != PF_PASS)
8908 				pd.act.log |= PF_LOG_FORCE;
8909 			goto done;
8910 		}
8911 		action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason);
8912 		if (action == PF_PASS) {
8913 			if (V_pfsync_update_state_ptr != NULL)
8914 				V_pfsync_update_state_ptr(s);
8915 			r = s->rule.ptr;
8916 			a = s->anchor.ptr;
8917 		} else if (s == NULL)
8918 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8919 			    &a, &ruleset, inp);
8920 		break;
8921 	}
8922 
8923 	default:
8924 		action = pf_test_state_other(&s, kif, m, &pd);
8925 		if (action == PF_PASS) {
8926 			if (V_pfsync_update_state_ptr != NULL)
8927 				V_pfsync_update_state_ptr(s);
8928 			r = s->rule.ptr;
8929 			a = s->anchor.ptr;
8930 		} else if (s == NULL)
8931 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8932 			    &a, &ruleset, inp);
8933 		break;
8934 	}
8935 
8936 done:
8937 	PF_RULES_RUNLOCK();
8938 	if (n != m) {
8939 		m_freem(n);
8940 		n = NULL;
8941 	}
8942 
8943 	/* handle dangerous IPv6 extension headers. */
8944 	if (action == PF_PASS && rh_cnt &&
8945 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
8946 		action = PF_DROP;
8947 		REASON_SET(&reason, PFRES_IPOPTIONS);
8948 		pd.act.log = r->log;
8949 		DPFPRINTF(PF_DEBUG_MISC,
8950 		    ("pf: dropping packet with dangerous v6 headers\n"));
8951 	}
8952 
8953 	if (s) {
8954 		uint8_t log = pd.act.log;
8955 		memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions));
8956 		pd.act.log |= log;
8957 		tag = s->tag;
8958 		rt = s->rt;
8959 	} else {
8960 		tag = r->tag;
8961 		rt = r->rt;
8962 	}
8963 
8964 	if (tag > 0 && pf_tag_packet(m, &pd, tag)) {
8965 		action = PF_DROP;
8966 		REASON_SET(&reason, PFRES_MEMORY);
8967 	}
8968 
8969 	pf_scrub_ip6(&m, &pd);
8970 	if (pd.proto == IPPROTO_TCP && pd.act.max_mss)
8971 		pf_normalize_mss(m, off, &pd);
8972 
8973 	if (pd.act.rtableid >= 0)
8974 		M_SETFIB(m, pd.act.rtableid);
8975 
8976 	if (pd.act.flags & PFSTATE_SETPRIO) {
8977 		if (pd.tos & IPTOS_LOWDELAY)
8978 			use_2nd_queue = 1;
8979 		if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) {
8980 			action = PF_DROP;
8981 			REASON_SET(&reason, PFRES_MEMORY);
8982 			pd.act.log = PF_LOG_FORCE;
8983 			DPFPRINTF(PF_DEBUG_MISC,
8984 			    ("pf: failed to allocate 802.1q mtag\n"));
8985 		}
8986 	}
8987 
8988 #ifdef ALTQ
8989 	if (action == PF_PASS && pd.act.qid) {
8990 		if (pd.pf_mtag == NULL &&
8991 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8992 			action = PF_DROP;
8993 			REASON_SET(&reason, PFRES_MEMORY);
8994 		} else {
8995 			if (s != NULL)
8996 				pd.pf_mtag->qid_hash = pf_state_hash(s);
8997 			if (pd.tos & IPTOS_LOWDELAY)
8998 				pd.pf_mtag->qid = pd.act.pqid;
8999 			else
9000 				pd.pf_mtag->qid = pd.act.qid;
9001 			/* Add hints for ecn. */
9002 			pd.pf_mtag->hdr = h;
9003 		}
9004 	}
9005 #endif /* ALTQ */
9006 
9007 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
9008 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
9009 	    (s->nat_rule.ptr->action == PF_RDR ||
9010 	    s->nat_rule.ptr->action == PF_BINAT) &&
9011 	    IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
9012 		m->m_flags |= M_SKIP_FIREWALL;
9013 
9014 	/* XXX: Anybody working on it?! */
9015 	if (r->divert.port)
9016 		printf("pf: divert(9) is not supported for IPv6\n");
9017 
9018 	if (pd.act.log) {
9019 		struct pf_krule		*lr;
9020 		struct pf_krule_item	*ri;
9021 
9022 		if (s != NULL && s->nat_rule.ptr != NULL &&
9023 		    s->nat_rule.ptr->log & PF_LOG_ALL)
9024 			lr = s->nat_rule.ptr;
9025 		else
9026 			lr = r;
9027 
9028 		if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL)
9029 			PFLOG_PACKET(kif, m, AF_INET6, action, reason, lr, a, ruleset,
9030 			    &pd, (s == NULL));
9031 		if (s) {
9032 			SLIST_FOREACH(ri, &s->match_rules, entry)
9033 				if (ri->r->log & PF_LOG_ALL)
9034 					PFLOG_PACKET(kif, m, AF_INET6, action, reason,
9035 					    ri->r, a, ruleset, &pd, 0);
9036 		}
9037 	}
9038 
9039 	pf_counter_u64_critical_enter();
9040 	pf_counter_u64_add_protected(&kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS],
9041 	    pd.tot_len);
9042 	pf_counter_u64_add_protected(&kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS],
9043 	    1);
9044 
9045 	if (action == PF_PASS || r->action == PF_DROP) {
9046 		dirndx = (dir == PF_OUT);
9047 		pf_counter_u64_add_protected(&r->packets[dirndx], 1);
9048 		pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len);
9049 		if (a != NULL) {
9050 			pf_counter_u64_add_protected(&a->packets[dirndx], 1);
9051 			pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len);
9052 		}
9053 		if (s != NULL) {
9054 			if (s->nat_rule.ptr != NULL) {
9055 				pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx],
9056 				    1);
9057 				pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx],
9058 				    pd.tot_len);
9059 			}
9060 			if (s->src_node != NULL) {
9061 				counter_u64_add(s->src_node->packets[dirndx],
9062 				    1);
9063 				counter_u64_add(s->src_node->bytes[dirndx],
9064 				    pd.tot_len);
9065 			}
9066 			if (s->nat_src_node != NULL) {
9067 				counter_u64_add(s->nat_src_node->packets[dirndx],
9068 				    1);
9069 				counter_u64_add(s->nat_src_node->bytes[dirndx],
9070 				    pd.tot_len);
9071 			}
9072 			dirndx = (dir == s->direction) ? 0 : 1;
9073 			s->packets[dirndx]++;
9074 			s->bytes[dirndx] += pd.tot_len;
9075 		}
9076 		tr = r;
9077 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
9078 		if (nr != NULL && r == &V_pf_default_rule)
9079 			tr = nr;
9080 		if (tr->src.addr.type == PF_ADDR_TABLE)
9081 			pfr_update_stats(tr->src.addr.p.tbl,
9082 			    (s == NULL) ? pd.src :
9083 			    &s->key[(s->direction == PF_IN)]->addr[0],
9084 			    pd.af, pd.tot_len, dir == PF_OUT,
9085 			    r->action == PF_PASS, tr->src.neg);
9086 		if (tr->dst.addr.type == PF_ADDR_TABLE)
9087 			pfr_update_stats(tr->dst.addr.p.tbl,
9088 			    (s == NULL) ? pd.dst :
9089 			    &s->key[(s->direction == PF_IN)]->addr[1],
9090 			    pd.af, pd.tot_len, dir == PF_OUT,
9091 			    r->action == PF_PASS, tr->dst.neg);
9092 	}
9093 	pf_counter_u64_critical_exit();
9094 
9095 	switch (action) {
9096 	case PF_SYNPROXY_DROP:
9097 		m_freem(*m0);
9098 	case PF_DEFER:
9099 		*m0 = NULL;
9100 		action = PF_PASS;
9101 		break;
9102 	case PF_DROP:
9103 		m_freem(*m0);
9104 		*m0 = NULL;
9105 		break;
9106 	default:
9107 		/* pf_route6() returns unlocked. */
9108 		if (rt) {
9109 			pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp);
9110 			goto out;
9111 		}
9112 		if (pf_dummynet(&pd, s, r, m0) != 0) {
9113 			action = PF_DROP;
9114 			REASON_SET(&reason, PFRES_MEMORY);
9115 		}
9116 		break;
9117 	}
9118 
9119 	if (s && action != PF_DROP) {
9120 		if (!s->if_index_in && dir == PF_IN)
9121 			s->if_index_in = ifp->if_index;
9122 		else if (!s->if_index_out && dir == PF_OUT)
9123 			s->if_index_out = ifp->if_index;
9124 	}
9125 
9126 	if (s)
9127 		PF_STATE_UNLOCK(s);
9128 
9129 	/* If reassembled packet passed, create new fragments. */
9130 	if (action == PF_PASS && *m0 && dir == PF_OUT &&
9131 	    (mtag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL)
9132 		action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD);
9133 
9134 out:
9135 	SDT_PROBE4(pf, ip, test6, done, action, reason, r, s);
9136 
9137 	pf_sctp_multihome_delayed(&pd, off, kif, s, action);
9138 
9139 	return (action);
9140 }
9141 #endif /* INET6 */
9142