xref: /freebsd/sys/netpfil/pf/pf.c (revision 3c91c65a2b999a8d27e99bcc6493de7cf812c948)
1 /*-
2  * Copyright (c) 2001 Daniel Hartmeier
3  * Copyright (c) 2002 - 2008 Henning Brauer
4  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  *    - Redistributions of source code must retain the above copyright
12  *      notice, this list of conditions and the following disclaimer.
13  *    - Redistributions in binary form must reproduce the above
14  *      copyright notice, this list of conditions and the following
15  *      disclaimer in the documentation and/or other materials provided
16  *      with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
21  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
22  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
24  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
25  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
26  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
28  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  *
31  * Effort sponsored in part by the Defense Advanced Research Projects
32  * Agency (DARPA) and Air Force Research Laboratory, Air Force
33  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
34  *
35  *	$OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include "opt_inet.h"
42 #include "opt_inet6.h"
43 #include "opt_bpf.h"
44 #include "opt_pf.h"
45 
46 #include <sys/param.h>
47 #include <sys/bus.h>
48 #include <sys/endian.h>
49 #include <sys/hash.h>
50 #include <sys/interrupt.h>
51 #include <sys/kernel.h>
52 #include <sys/kthread.h>
53 #include <sys/limits.h>
54 #include <sys/mbuf.h>
55 #include <sys/md5.h>
56 #include <sys/random.h>
57 #include <sys/refcount.h>
58 #include <sys/socket.h>
59 #include <sys/sysctl.h>
60 #include <sys/taskqueue.h>
61 #include <sys/ucred.h>
62 
63 #include <net/if.h>
64 #include <net/if_types.h>
65 #include <net/route.h>
66 #include <net/radix_mpath.h>
67 #include <net/vnet.h>
68 
69 #include <net/pfvar.h>
70 #include <net/pf_mtag.h>
71 #include <net/if_pflog.h>
72 #include <net/if_pfsync.h>
73 
74 #include <netinet/in_pcb.h>
75 #include <netinet/in_var.h>
76 #include <netinet/ip.h>
77 #include <netinet/ip_fw.h>
78 #include <netinet/ip_icmp.h>
79 #include <netinet/icmp_var.h>
80 #include <netinet/ip_var.h>
81 #include <netinet/tcp.h>
82 #include <netinet/tcp_fsm.h>
83 #include <netinet/tcp_seq.h>
84 #include <netinet/tcp_timer.h>
85 #include <netinet/tcp_var.h>
86 #include <netinet/udp.h>
87 #include <netinet/udp_var.h>
88 
89 #include <netpfil/ipfw/ip_fw_private.h> /* XXX: only for DIR_IN/DIR_OUT */
90 
91 #ifdef INET6
92 #include <netinet/ip6.h>
93 #include <netinet/icmp6.h>
94 #include <netinet6/nd6.h>
95 #include <netinet6/ip6_var.h>
96 #include <netinet6/in6_pcb.h>
97 #endif /* INET6 */
98 
99 #include <machine/in_cksum.h>
100 #include <security/mac/mac_framework.h>
101 
102 #define	DPFPRINTF(n, x)	if (V_pf_status.debug >= (n)) printf x
103 
104 /*
105  * Global variables
106  */
107 
108 /* state tables */
109 VNET_DEFINE(struct pf_altqqueue,	 pf_altqs[2]);
110 VNET_DEFINE(struct pf_palist,		 pf_pabuf);
111 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_active);
112 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_inactive);
113 VNET_DEFINE(struct pf_status,		 pf_status);
114 
115 VNET_DEFINE(u_int32_t,			 ticket_altqs_active);
116 VNET_DEFINE(u_int32_t,			 ticket_altqs_inactive);
117 VNET_DEFINE(int,			 altqs_inactive_open);
118 VNET_DEFINE(u_int32_t,			 ticket_pabuf);
119 
120 VNET_DEFINE(MD5_CTX,			 pf_tcp_secret_ctx);
121 #define	V_pf_tcp_secret_ctx		 VNET(pf_tcp_secret_ctx)
122 VNET_DEFINE(u_char,			 pf_tcp_secret[16]);
123 #define	V_pf_tcp_secret			 VNET(pf_tcp_secret)
124 VNET_DEFINE(int,			 pf_tcp_secret_init);
125 #define	V_pf_tcp_secret_init		 VNET(pf_tcp_secret_init)
126 VNET_DEFINE(int,			 pf_tcp_iss_off);
127 #define	V_pf_tcp_iss_off		 VNET(pf_tcp_iss_off)
128 
129 /*
130  * Queue for pf_intr() sends.
131  */
132 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
133 struct pf_send_entry {
134 	STAILQ_ENTRY(pf_send_entry)	pfse_next;
135 	struct mbuf			*pfse_m;
136 	enum {
137 		PFSE_IP,
138 		PFSE_IP6,
139 		PFSE_ICMP,
140 		PFSE_ICMP6,
141 	}				pfse_type;
142 	union {
143 		struct route		ro;
144 		struct {
145 			int		type;
146 			int		code;
147 			int		mtu;
148 		} icmpopts;
149 	} u;
150 #define	pfse_ro		u.ro
151 #define	pfse_icmp_type	u.icmpopts.type
152 #define	pfse_icmp_code	u.icmpopts.code
153 #define	pfse_icmp_mtu	u.icmpopts.mtu
154 };
155 
156 STAILQ_HEAD(pf_send_head, pf_send_entry);
157 static VNET_DEFINE(struct pf_send_head, pf_sendqueue);
158 #define	V_pf_sendqueue	VNET(pf_sendqueue)
159 
160 static struct mtx pf_sendqueue_mtx;
161 #define	PF_SENDQ_LOCK()		mtx_lock(&pf_sendqueue_mtx)
162 #define	PF_SENDQ_UNLOCK()	mtx_unlock(&pf_sendqueue_mtx)
163 
164 /*
165  * Queue for pf_overload_task() tasks.
166  */
167 struct pf_overload_entry {
168 	SLIST_ENTRY(pf_overload_entry)	next;
169 	struct pf_addr  		addr;
170 	sa_family_t			af;
171 	uint8_t				dir;
172 	struct pf_rule  		*rule;
173 };
174 
175 SLIST_HEAD(pf_overload_head, pf_overload_entry);
176 static VNET_DEFINE(struct pf_overload_head, pf_overloadqueue);
177 #define V_pf_overloadqueue	VNET(pf_overloadqueue)
178 static VNET_DEFINE(struct task, pf_overloadtask);
179 #define	V_pf_overloadtask	VNET(pf_overloadtask)
180 
181 static struct mtx pf_overloadqueue_mtx;
182 #define	PF_OVERLOADQ_LOCK()	mtx_lock(&pf_overloadqueue_mtx)
183 #define	PF_OVERLOADQ_UNLOCK()	mtx_unlock(&pf_overloadqueue_mtx)
184 
185 VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules);
186 struct mtx pf_unlnkdrules_mtx;
187 
188 static VNET_DEFINE(uma_zone_t,	pf_sources_z);
189 #define	V_pf_sources_z	VNET(pf_sources_z)
190 static VNET_DEFINE(uma_zone_t,	pf_mtag_z);
191 #define	V_pf_mtag_z	VNET(pf_mtag_z)
192 VNET_DEFINE(uma_zone_t,	 pf_state_z);
193 VNET_DEFINE(uma_zone_t,	 pf_state_key_z);
194 
195 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
196 #define	PFID_CPUBITS	8
197 #define	PFID_CPUSHIFT	(sizeof(uint64_t) * NBBY - PFID_CPUBITS)
198 #define	PFID_CPUMASK	((uint64_t)((1 << PFID_CPUBITS) - 1) <<	PFID_CPUSHIFT)
199 #define	PFID_MAXID	(~PFID_CPUMASK)
200 CTASSERT((1 << PFID_CPUBITS) > MAXCPU);
201 
202 static void		 pf_src_tree_remove_state(struct pf_state *);
203 static void		 pf_init_threshold(struct pf_threshold *, u_int32_t,
204 			    u_int32_t);
205 static void		 pf_add_threshold(struct pf_threshold *);
206 static int		 pf_check_threshold(struct pf_threshold *);
207 
208 static void		 pf_change_ap(struct pf_addr *, u_int16_t *,
209 			    u_int16_t *, u_int16_t *, struct pf_addr *,
210 			    u_int16_t, u_int8_t, sa_family_t);
211 static int		 pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
212 			    struct tcphdr *, struct pf_state_peer *);
213 static void		 pf_change_icmp(struct pf_addr *, u_int16_t *,
214 			    struct pf_addr *, struct pf_addr *, u_int16_t,
215 			    u_int16_t *, u_int16_t *, u_int16_t *,
216 			    u_int16_t *, u_int8_t, sa_family_t);
217 static void		 pf_send_tcp(struct mbuf *,
218 			    const struct pf_rule *, sa_family_t,
219 			    const struct pf_addr *, const struct pf_addr *,
220 			    u_int16_t, u_int16_t, u_int32_t, u_int32_t,
221 			    u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
222 			    u_int16_t, struct ifnet *);
223 static void		 pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
224 			    sa_family_t, struct pf_rule *);
225 static void		 pf_detach_state(struct pf_state *);
226 static int		 pf_state_key_attach(struct pf_state_key *,
227 			    struct pf_state_key *, struct pf_state *);
228 static void		 pf_state_key_detach(struct pf_state *, int);
229 static int		 pf_state_key_ctor(void *, int, void *, int);
230 static u_int32_t	 pf_tcp_iss(struct pf_pdesc *);
231 static int		 pf_test_rule(struct pf_rule **, struct pf_state **,
232 			    int, struct pfi_kif *, struct mbuf *, int,
233 			    struct pf_pdesc *, struct pf_rule **,
234 			    struct pf_ruleset **, struct inpcb *);
235 static int		 pf_create_state(struct pf_rule *, struct pf_rule *,
236 			    struct pf_rule *, struct pf_pdesc *,
237 			    struct pf_src_node *, struct pf_state_key *,
238 			    struct pf_state_key *, struct mbuf *, int,
239 			    u_int16_t, u_int16_t, int *, struct pfi_kif *,
240 			    struct pf_state **, int, u_int16_t, u_int16_t,
241 			    int);
242 static int		 pf_test_fragment(struct pf_rule **, int,
243 			    struct pfi_kif *, struct mbuf *, void *,
244 			    struct pf_pdesc *, struct pf_rule **,
245 			    struct pf_ruleset **);
246 static int		 pf_tcp_track_full(struct pf_state_peer *,
247 			    struct pf_state_peer *, struct pf_state **,
248 			    struct pfi_kif *, struct mbuf *, int,
249 			    struct pf_pdesc *, u_short *, int *);
250 static int		 pf_tcp_track_sloppy(struct pf_state_peer *,
251 			    struct pf_state_peer *, struct pf_state **,
252 			    struct pf_pdesc *, u_short *);
253 static int		 pf_test_state_tcp(struct pf_state **, int,
254 			    struct pfi_kif *, struct mbuf *, int,
255 			    void *, struct pf_pdesc *, u_short *);
256 static int		 pf_test_state_udp(struct pf_state **, int,
257 			    struct pfi_kif *, struct mbuf *, int,
258 			    void *, struct pf_pdesc *);
259 static int		 pf_test_state_icmp(struct pf_state **, int,
260 			    struct pfi_kif *, struct mbuf *, int,
261 			    void *, struct pf_pdesc *, u_short *);
262 static int		 pf_test_state_other(struct pf_state **, int,
263 			    struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
264 static u_int8_t		 pf_get_wscale(struct mbuf *, int, u_int16_t,
265 			    sa_family_t);
266 static u_int16_t	 pf_get_mss(struct mbuf *, int, u_int16_t,
267 			    sa_family_t);
268 static u_int16_t	 pf_calc_mss(struct pf_addr *, sa_family_t,
269 				int, u_int16_t);
270 static void		 pf_set_rt_ifp(struct pf_state *,
271 			    struct pf_addr *);
272 static int		 pf_check_proto_cksum(struct mbuf *, int, int,
273 			    u_int8_t, sa_family_t);
274 static void		 pf_print_state_parts(struct pf_state *,
275 			    struct pf_state_key *, struct pf_state_key *);
276 static int		 pf_addr_wrap_neq(struct pf_addr_wrap *,
277 			    struct pf_addr_wrap *);
278 static struct pf_state	*pf_find_state(struct pfi_kif *,
279 			    struct pf_state_key_cmp *, u_int);
280 static int		 pf_src_connlimit(struct pf_state **);
281 static void		 pf_overload_task(void *c, int pending);
282 static int		 pf_insert_src_node(struct pf_src_node **,
283 			    struct pf_rule *, struct pf_addr *, sa_family_t);
284 static u_int		 pf_purge_expired_states(u_int, int);
285 static void		 pf_purge_unlinked_rules(void);
286 static int		 pf_mtag_init(void *, int, int);
287 static void		 pf_mtag_free(struct m_tag *);
288 #ifdef INET
289 static void		 pf_route(struct mbuf **, struct pf_rule *, int,
290 			    struct ifnet *, struct pf_state *,
291 			    struct pf_pdesc *);
292 #endif /* INET */
293 #ifdef INET6
294 static void		 pf_change_a6(struct pf_addr *, u_int16_t *,
295 			    struct pf_addr *, u_int8_t);
296 static void		 pf_route6(struct mbuf **, struct pf_rule *, int,
297 			    struct ifnet *, struct pf_state *,
298 			    struct pf_pdesc *);
299 #endif /* INET6 */
300 
301 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
302 
303 VNET_DECLARE(int, pf_end_threads);
304 
305 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
306 
307 #define	PACKET_LOOPED(pd)	((pd)->pf_mtag &&			\
308 				 (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
309 
310 #define	STATE_LOOKUP(i, k, d, s, pd)					\
311 	do {								\
312 		(s) = pf_find_state((i), (k), (d));			\
313 		if ((s) == NULL || (s)->timeout == PFTM_PURGE)		\
314 			return (PF_DROP);				\
315 		if (PACKET_LOOPED(pd))					\
316 			return (PF_PASS);				\
317 		if ((d) == PF_OUT &&					\
318 		    (((s)->rule.ptr->rt == PF_ROUTETO &&		\
319 		    (s)->rule.ptr->direction == PF_OUT) ||		\
320 		    ((s)->rule.ptr->rt == PF_REPLYTO &&			\
321 		    (s)->rule.ptr->direction == PF_IN)) &&		\
322 		    (s)->rt_kif != NULL &&				\
323 		    (s)->rt_kif != (i))					\
324 			return (PF_PASS);				\
325 	} while (0)
326 
327 #define	BOUND_IFACE(r, k) \
328 	((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
329 
330 #define	STATE_INC_COUNTERS(s)				\
331 	do {						\
332 		s->rule.ptr->states_cur++;		\
333 		s->rule.ptr->states_tot++;		\
334 		if (s->anchor.ptr != NULL) {		\
335 			s->anchor.ptr->states_cur++;	\
336 			s->anchor.ptr->states_tot++;	\
337 		}					\
338 		if (s->nat_rule.ptr != NULL) {		\
339 			s->nat_rule.ptr->states_cur++;	\
340 			s->nat_rule.ptr->states_tot++;	\
341 		}					\
342 	} while (0)
343 
344 #define	STATE_DEC_COUNTERS(s)				\
345 	do {						\
346 		if (s->nat_rule.ptr != NULL)		\
347 			s->nat_rule.ptr->states_cur--;	\
348 		if (s->anchor.ptr != NULL)		\
349 			s->anchor.ptr->states_cur--;	\
350 		s->rule.ptr->states_cur--;		\
351 	} while (0)
352 
353 static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
354 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
355 VNET_DEFINE(struct pf_idhash *, pf_idhash);
356 VNET_DEFINE(u_long, pf_hashmask);
357 VNET_DEFINE(struct pf_srchash *, pf_srchash);
358 VNET_DEFINE(u_long, pf_srchashmask);
359 
360 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)");
361 
362 VNET_DEFINE(u_long, pf_hashsize);
363 #define	V_pf_hashsize	VNET(pf_hashsize)
364 SYSCTL_VNET_UINT(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
365     &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable");
366 
367 VNET_DEFINE(u_long, pf_srchashsize);
368 #define	V_pf_srchashsize	VNET(pf_srchashsize)
369 SYSCTL_VNET_UINT(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
370     &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable");
371 
372 VNET_DEFINE(void *, pf_swi_cookie);
373 
374 VNET_DEFINE(uint32_t, pf_hashseed);
375 #define	V_pf_hashseed	VNET(pf_hashseed)
376 
377 static __inline uint32_t
378 pf_hashkey(struct pf_state_key *sk)
379 {
380 	uint32_t h;
381 
382 	h = jenkins_hash32((uint32_t *)sk,
383 	    sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
384 	    V_pf_hashseed);
385 
386 	return (h & V_pf_hashmask);
387 }
388 
389 static __inline uint32_t
390 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
391 {
392 	uint32_t h;
393 
394 	switch (af) {
395 	case AF_INET:
396 		h = jenkins_hash32((uint32_t *)&addr->v4,
397 		    sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
398 		break;
399 	case AF_INET6:
400 		h = jenkins_hash32((uint32_t *)&addr->v6,
401 		    sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
402 		break;
403 	default:
404 		panic("%s: unknown address family %u", __func__, af);
405 	}
406 
407 	return (h & V_pf_srchashmask);
408 }
409 
410 #ifdef INET6
411 void
412 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
413 {
414 	switch (af) {
415 #ifdef INET
416 	case AF_INET:
417 		dst->addr32[0] = src->addr32[0];
418 		break;
419 #endif /* INET */
420 	case AF_INET6:
421 		dst->addr32[0] = src->addr32[0];
422 		dst->addr32[1] = src->addr32[1];
423 		dst->addr32[2] = src->addr32[2];
424 		dst->addr32[3] = src->addr32[3];
425 		break;
426 	}
427 }
428 #endif /* INET6 */
429 
430 static void
431 pf_init_threshold(struct pf_threshold *threshold,
432     u_int32_t limit, u_int32_t seconds)
433 {
434 	threshold->limit = limit * PF_THRESHOLD_MULT;
435 	threshold->seconds = seconds;
436 	threshold->count = 0;
437 	threshold->last = time_uptime;
438 }
439 
440 static void
441 pf_add_threshold(struct pf_threshold *threshold)
442 {
443 	u_int32_t t = time_uptime, diff = t - threshold->last;
444 
445 	if (diff >= threshold->seconds)
446 		threshold->count = 0;
447 	else
448 		threshold->count -= threshold->count * diff /
449 		    threshold->seconds;
450 	threshold->count += PF_THRESHOLD_MULT;
451 	threshold->last = t;
452 }
453 
454 static int
455 pf_check_threshold(struct pf_threshold *threshold)
456 {
457 	return (threshold->count > threshold->limit);
458 }
459 
460 static int
461 pf_src_connlimit(struct pf_state **state)
462 {
463 	struct pf_overload_entry *pfoe;
464 	int bad = 0;
465 
466 	PF_STATE_LOCK_ASSERT(*state);
467 
468 	(*state)->src_node->conn++;
469 	(*state)->src.tcp_est = 1;
470 	pf_add_threshold(&(*state)->src_node->conn_rate);
471 
472 	if ((*state)->rule.ptr->max_src_conn &&
473 	    (*state)->rule.ptr->max_src_conn <
474 	    (*state)->src_node->conn) {
475 		V_pf_status.lcounters[LCNT_SRCCONN]++;
476 		bad++;
477 	}
478 
479 	if ((*state)->rule.ptr->max_src_conn_rate.limit &&
480 	    pf_check_threshold(&(*state)->src_node->conn_rate)) {
481 		V_pf_status.lcounters[LCNT_SRCCONNRATE]++;
482 		bad++;
483 	}
484 
485 	if (!bad)
486 		return (0);
487 
488 	/* Kill this state. */
489 	(*state)->timeout = PFTM_PURGE;
490 	(*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
491 
492 	if ((*state)->rule.ptr->overload_tbl == NULL)
493 		return (1);
494 
495 	/* Schedule overloading and flushing task. */
496 	pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
497 	if (pfoe == NULL)
498 		return (1);	/* too bad :( */
499 
500 	bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
501 	pfoe->af = (*state)->key[PF_SK_WIRE]->af;
502 	pfoe->rule = (*state)->rule.ptr;
503 	pfoe->dir = (*state)->direction;
504 	PF_OVERLOADQ_LOCK();
505 	SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
506 	PF_OVERLOADQ_UNLOCK();
507 	taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
508 
509 	return (1);
510 }
511 
512 static void
513 pf_overload_task(void *c, int pending)
514 {
515 	struct pf_overload_head queue;
516 	struct pfr_addr p;
517 	struct pf_overload_entry *pfoe, *pfoe1;
518 	uint32_t killed = 0;
519 
520 	PF_OVERLOADQ_LOCK();
521 	queue = *(struct pf_overload_head *)c;
522 	SLIST_INIT((struct pf_overload_head *)c);
523 	PF_OVERLOADQ_UNLOCK();
524 
525 	bzero(&p, sizeof(p));
526 	SLIST_FOREACH(pfoe, &queue, next) {
527 		V_pf_status.lcounters[LCNT_OVERLOAD_TABLE]++;
528 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
529 			printf("%s: blocking address ", __func__);
530 			pf_print_host(&pfoe->addr, 0, pfoe->af);
531 			printf("\n");
532 		}
533 
534 		p.pfra_af = pfoe->af;
535 		switch (pfoe->af) {
536 #ifdef INET
537 		case AF_INET:
538 			p.pfra_net = 32;
539 			p.pfra_ip4addr = pfoe->addr.v4;
540 			break;
541 #endif
542 #ifdef INET6
543 		case AF_INET6:
544 			p.pfra_net = 128;
545 			p.pfra_ip6addr = pfoe->addr.v6;
546 			break;
547 #endif
548 		}
549 
550 		PF_RULES_WLOCK();
551 		pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
552 		PF_RULES_WUNLOCK();
553 	}
554 
555 	/*
556 	 * Remove those entries, that don't need flushing.
557 	 */
558 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
559 		if (pfoe->rule->flush == 0) {
560 			SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
561 			free(pfoe, M_PFTEMP);
562 		} else
563 			V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH]++;
564 
565 	/* If nothing to flush, return. */
566 	if (SLIST_EMPTY(&queue))
567 		return;
568 
569 	for (int i = 0; i <= V_pf_hashmask; i++) {
570 		struct pf_idhash *ih = &V_pf_idhash[i];
571 		struct pf_state_key *sk;
572 		struct pf_state *s;
573 
574 		PF_HASHROW_LOCK(ih);
575 		LIST_FOREACH(s, &ih->states, entry) {
576 		    sk = s->key[PF_SK_WIRE];
577 		    SLIST_FOREACH(pfoe, &queue, next)
578 			if (sk->af == pfoe->af &&
579 			    ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
580 			    pfoe->rule == s->rule.ptr) &&
581 			    ((pfoe->dir == PF_OUT &&
582 			    PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
583 			    (pfoe->dir == PF_IN &&
584 			    PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
585 				s->timeout = PFTM_PURGE;
586 				s->src.state = s->dst.state = TCPS_CLOSED;
587 				killed++;
588 			}
589 		}
590 		PF_HASHROW_UNLOCK(ih);
591 	}
592 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
593 		free(pfoe, M_PFTEMP);
594 	if (V_pf_status.debug >= PF_DEBUG_MISC)
595 		printf("%s: %u states killed", __func__, killed);
596 }
597 
598 /*
599  * Can return locked on failure, so that we can consistently
600  * allocate and insert a new one.
601  */
602 struct pf_src_node *
603 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
604 	int returnlocked)
605 {
606 	struct pf_srchash *sh;
607 	struct pf_src_node *n;
608 
609 	V_pf_status.scounters[SCNT_SRC_NODE_SEARCH]++;
610 
611 	sh = &V_pf_srchash[pf_hashsrc(src, af)];
612 	PF_HASHROW_LOCK(sh);
613 	LIST_FOREACH(n, &sh->nodes, entry)
614 		if (n->rule.ptr == rule && n->af == af &&
615 		    ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
616 		    (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
617 			break;
618 	if (n != NULL || returnlocked == 0)
619 		PF_HASHROW_UNLOCK(sh);
620 
621 	return (n);
622 }
623 
624 static int
625 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
626     struct pf_addr *src, sa_family_t af)
627 {
628 
629 	KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
630 	    rule->rpool.opts & PF_POOL_STICKYADDR),
631 	    ("%s for non-tracking rule %p", __func__, rule));
632 
633 	if (*sn == NULL)
634 		*sn = pf_find_src_node(src, rule, af, 1);
635 
636 	if (*sn == NULL) {
637 		struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
638 
639 		PF_HASHROW_ASSERT(sh);
640 
641 		if (!rule->max_src_nodes ||
642 		    rule->src_nodes < rule->max_src_nodes)
643 			(*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
644 		else
645 			V_pf_status.lcounters[LCNT_SRCNODES]++;
646 		if ((*sn) == NULL) {
647 			PF_HASHROW_UNLOCK(sh);
648 			return (-1);
649 		}
650 
651 		pf_init_threshold(&(*sn)->conn_rate,
652 		    rule->max_src_conn_rate.limit,
653 		    rule->max_src_conn_rate.seconds);
654 
655 		(*sn)->af = af;
656 		(*sn)->rule.ptr = rule;
657 		PF_ACPY(&(*sn)->addr, src, af);
658 		LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
659 		(*sn)->creation = time_uptime;
660 		(*sn)->ruletype = rule->action;
661 		if ((*sn)->rule.ptr != NULL)
662 			(*sn)->rule.ptr->src_nodes++;
663 		PF_HASHROW_UNLOCK(sh);
664 		V_pf_status.scounters[SCNT_SRC_NODE_INSERT]++;
665 		V_pf_status.src_nodes++;
666 	} else {
667 		if (rule->max_src_states &&
668 		    (*sn)->states >= rule->max_src_states) {
669 			V_pf_status.lcounters[LCNT_SRCSTATES]++;
670 			return (-1);
671 		}
672 	}
673 	return (0);
674 }
675 
676 static void
677 pf_remove_src_node(struct pf_src_node *src)
678 {
679 	struct pf_srchash *sh;
680 
681 	sh = &V_pf_srchash[pf_hashsrc(&src->addr, src->af)];
682 	PF_HASHROW_LOCK(sh);
683 	LIST_REMOVE(src, entry);
684 	PF_HASHROW_UNLOCK(sh);
685 
686 	V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS]++;
687 	V_pf_status.src_nodes--;
688 
689 	uma_zfree(V_pf_sources_z, src);
690 }
691 
692 /* Data storage structures initialization. */
693 void
694 pf_initialize()
695 {
696 	struct pf_keyhash	*kh;
697 	struct pf_idhash	*ih;
698 	struct pf_srchash	*sh;
699 	u_int i;
700 
701 	TUNABLE_ULONG_FETCH("net.pf.states_hashsize", &V_pf_hashsize);
702 	if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize))
703 		V_pf_hashsize = PF_HASHSIZ;
704 	TUNABLE_ULONG_FETCH("net.pf.source_nodes_hashsize", &V_pf_srchashsize);
705 	if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize))
706 		V_pf_srchashsize = PF_HASHSIZ / 4;
707 
708 	V_pf_hashseed = arc4random();
709 
710 	/* States and state keys storage. */
711 	V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
712 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
713 	V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
714 	uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
715 	uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
716 
717 	V_pf_state_key_z = uma_zcreate("pf state keys",
718 	    sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
719 	    UMA_ALIGN_PTR, 0);
720 	V_pf_keyhash = malloc(V_pf_hashsize * sizeof(struct pf_keyhash),
721 	    M_PFHASH, M_WAITOK | M_ZERO);
722 	V_pf_idhash = malloc(V_pf_hashsize * sizeof(struct pf_idhash),
723 	    M_PFHASH, M_WAITOK | M_ZERO);
724 	V_pf_hashmask = V_pf_hashsize - 1;
725 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
726 	    i++, kh++, ih++) {
727 		mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF);
728 		mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
729 	}
730 
731 	/* Source nodes. */
732 	V_pf_sources_z = uma_zcreate("pf source nodes",
733 	    sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
734 	    0);
735 	V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
736 	uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
737 	uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
738 	V_pf_srchash = malloc(V_pf_srchashsize * sizeof(struct pf_srchash),
739 	  M_PFHASH, M_WAITOK|M_ZERO);
740 	V_pf_srchashmask = V_pf_srchashsize - 1;
741 	for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++)
742 		mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
743 
744 	/* ALTQ */
745 	TAILQ_INIT(&V_pf_altqs[0]);
746 	TAILQ_INIT(&V_pf_altqs[1]);
747 	TAILQ_INIT(&V_pf_pabuf);
748 	V_pf_altqs_active = &V_pf_altqs[0];
749 	V_pf_altqs_inactive = &V_pf_altqs[1];
750 
751 	/* Mbuf tags */
752 	V_pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
753 	    sizeof(struct pf_mtag), NULL, NULL, pf_mtag_init, NULL,
754 	    UMA_ALIGN_PTR, 0);
755 
756 	/* Send & overload+flush queues. */
757 	STAILQ_INIT(&V_pf_sendqueue);
758 	SLIST_INIT(&V_pf_overloadqueue);
759 	TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, &V_pf_overloadqueue);
760 	mtx_init(&pf_sendqueue_mtx, "pf send queue", NULL, MTX_DEF);
761 	mtx_init(&pf_overloadqueue_mtx, "pf overload/flush queue", NULL,
762 	    MTX_DEF);
763 
764 	/* Unlinked, but may be referenced rules. */
765 	TAILQ_INIT(&V_pf_unlinked_rules);
766 	mtx_init(&pf_unlnkdrules_mtx, "pf unlinked rules", NULL, MTX_DEF);
767 }
768 
769 void
770 pf_cleanup()
771 {
772 	struct pf_keyhash	*kh;
773 	struct pf_idhash	*ih;
774 	struct pf_srchash	*sh;
775 	struct pf_send_entry	*pfse, *next;
776 	u_int i;
777 
778 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
779 	    i++, kh++, ih++) {
780 		KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
781 		    __func__));
782 		KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
783 		    __func__));
784 		mtx_destroy(&kh->lock);
785 		mtx_destroy(&ih->lock);
786 	}
787 	free(V_pf_keyhash, M_PFHASH);
788 	free(V_pf_idhash, M_PFHASH);
789 
790 	for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
791 		KASSERT(LIST_EMPTY(&sh->nodes),
792 		    ("%s: source node hash not empty", __func__));
793 		mtx_destroy(&sh->lock);
794 	}
795 	free(V_pf_srchash, M_PFHASH);
796 
797 	STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
798 		m_freem(pfse->pfse_m);
799 		free(pfse, M_PFTEMP);
800 	}
801 
802 	mtx_destroy(&pf_sendqueue_mtx);
803 	mtx_destroy(&pf_overloadqueue_mtx);
804 	mtx_destroy(&pf_unlnkdrules_mtx);
805 
806 	uma_zdestroy(V_pf_mtag_z);
807 	uma_zdestroy(V_pf_sources_z);
808 	uma_zdestroy(V_pf_state_z);
809 	uma_zdestroy(V_pf_state_key_z);
810 }
811 
812 static int
813 pf_mtag_init(void *mem, int size, int how)
814 {
815 	struct m_tag *t;
816 
817 	t = (struct m_tag *)mem;
818 	t->m_tag_cookie = MTAG_ABI_COMPAT;
819 	t->m_tag_id = PACKET_TAG_PF;
820 	t->m_tag_len = sizeof(struct pf_mtag);
821 	t->m_tag_free = pf_mtag_free;
822 
823 	return (0);
824 }
825 
826 static void
827 pf_mtag_free(struct m_tag *t)
828 {
829 
830 	uma_zfree(V_pf_mtag_z, t);
831 }
832 
833 struct pf_mtag *
834 pf_get_mtag(struct mbuf *m)
835 {
836 	struct m_tag *mtag;
837 
838 	if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
839 		return ((struct pf_mtag *)(mtag + 1));
840 
841 	mtag = uma_zalloc(V_pf_mtag_z, M_NOWAIT);
842 	if (mtag == NULL)
843 		return (NULL);
844 	bzero(mtag + 1, sizeof(struct pf_mtag));
845 	m_tag_prepend(m, mtag);
846 
847 	return ((struct pf_mtag *)(mtag + 1));
848 }
849 
850 static int
851 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
852     struct pf_state *s)
853 {
854 	struct pf_keyhash	*kh;
855 	struct pf_state_key	*sk, *cur;
856 	struct pf_state		*si, *olds = NULL;
857 	int idx;
858 
859 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
860 	KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
861 	KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
862 
863 	/*
864 	 * First run: start with wire key.
865 	 */
866 	sk = skw;
867 	idx = PF_SK_WIRE;
868 
869 keyattach:
870 	kh = &V_pf_keyhash[pf_hashkey(sk)];
871 
872 	PF_HASHROW_LOCK(kh);
873 	LIST_FOREACH(cur, &kh->keys, entry)
874 		if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
875 			break;
876 
877 	if (cur != NULL) {
878 		/* Key exists. Check for same kif, if none, add to key. */
879 		TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
880 			struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
881 
882 			PF_HASHROW_LOCK(ih);
883 			if (si->kif == s->kif &&
884 			    si->direction == s->direction) {
885 				if (sk->proto == IPPROTO_TCP &&
886 				    si->src.state >= TCPS_FIN_WAIT_2 &&
887 				    si->dst.state >= TCPS_FIN_WAIT_2) {
888 					si->src.state = si->dst.state =
889 					    TCPS_CLOSED;
890 					/* Unlink later or cur can go away. */
891 					pf_ref_state(si);
892 					olds = si;
893 				} else {
894 					if (V_pf_status.debug >= PF_DEBUG_MISC) {
895 						printf("pf: %s key attach "
896 						    "failed on %s: ",
897 						    (idx == PF_SK_WIRE) ?
898 						    "wire" : "stack",
899 						    s->kif->pfik_name);
900 						pf_print_state_parts(s,
901 						    (idx == PF_SK_WIRE) ?
902 						    sk : NULL,
903 						    (idx == PF_SK_STACK) ?
904 						    sk : NULL);
905 						printf(", existing: ");
906 						pf_print_state_parts(si,
907 						    (idx == PF_SK_WIRE) ?
908 						    sk : NULL,
909 						    (idx == PF_SK_STACK) ?
910 						    sk : NULL);
911 						printf("\n");
912 					}
913 					PF_HASHROW_UNLOCK(ih);
914 					PF_HASHROW_UNLOCK(kh);
915 					uma_zfree(V_pf_state_key_z, sk);
916 					if (idx == PF_SK_STACK)
917 						pf_detach_state(s);
918 					return (-1);	/* collision! */
919 				}
920 			}
921 			PF_HASHROW_UNLOCK(ih);
922 		}
923 		uma_zfree(V_pf_state_key_z, sk);
924 		s->key[idx] = cur;
925 	} else {
926 		LIST_INSERT_HEAD(&kh->keys, sk, entry);
927 		s->key[idx] = sk;
928 	}
929 
930 stateattach:
931 	/* List is sorted, if-bound states before floating. */
932 	if (s->kif == V_pfi_all)
933 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
934 	else
935 		TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
936 
937 	/*
938 	 * Attach done. See how should we (or should not?)
939 	 * attach a second key.
940 	 */
941 	if (sks == skw) {
942 		s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
943 		idx = PF_SK_STACK;
944 		sks = NULL;
945 		goto stateattach;
946 	} else if (sks != NULL) {
947 		PF_HASHROW_UNLOCK(kh);
948 		if (olds) {
949 			pf_unlink_state(olds, 0);
950 			pf_release_state(olds);
951 			olds = NULL;
952 		}
953 		/*
954 		 * Continue attaching with stack key.
955 		 */
956 		sk = sks;
957 		idx = PF_SK_STACK;
958 		sks = NULL;
959 		goto keyattach;
960 	} else
961 		PF_HASHROW_UNLOCK(kh);
962 
963 	if (olds) {
964 		pf_unlink_state(olds, 0);
965 		pf_release_state(olds);
966 	}
967 
968 	KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
969 	    ("%s failure", __func__));
970 
971 	return (0);
972 }
973 
974 static void
975 pf_detach_state(struct pf_state *s)
976 {
977 	struct pf_state_key *sks = s->key[PF_SK_STACK];
978 	struct pf_keyhash *kh;
979 
980 	if (sks != NULL) {
981 		kh = &V_pf_keyhash[pf_hashkey(sks)];
982 		PF_HASHROW_LOCK(kh);
983 		if (s->key[PF_SK_STACK] != NULL)
984 			pf_state_key_detach(s, PF_SK_STACK);
985 		/*
986 		 * If both point to same key, then we are done.
987 		 */
988 		if (sks == s->key[PF_SK_WIRE]) {
989 			pf_state_key_detach(s, PF_SK_WIRE);
990 			PF_HASHROW_UNLOCK(kh);
991 			return;
992 		}
993 		PF_HASHROW_UNLOCK(kh);
994 	}
995 
996 	if (s->key[PF_SK_WIRE] != NULL) {
997 		kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
998 		PF_HASHROW_LOCK(kh);
999 		if (s->key[PF_SK_WIRE] != NULL)
1000 			pf_state_key_detach(s, PF_SK_WIRE);
1001 		PF_HASHROW_UNLOCK(kh);
1002 	}
1003 }
1004 
1005 static void
1006 pf_state_key_detach(struct pf_state *s, int idx)
1007 {
1008 	struct pf_state_key *sk = s->key[idx];
1009 #ifdef INVARIANTS
1010 	struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1011 
1012 	PF_HASHROW_ASSERT(kh);
1013 #endif
1014 	TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1015 	s->key[idx] = NULL;
1016 
1017 	if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1018 		LIST_REMOVE(sk, entry);
1019 		uma_zfree(V_pf_state_key_z, sk);
1020 	}
1021 }
1022 
1023 static int
1024 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1025 {
1026 	struct pf_state_key *sk = mem;
1027 
1028 	bzero(sk, sizeof(struct pf_state_key_cmp));
1029 	TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1030 	TAILQ_INIT(&sk->states[PF_SK_STACK]);
1031 
1032 	return (0);
1033 }
1034 
1035 struct pf_state_key *
1036 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1037 	struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1038 {
1039 	struct pf_state_key *sk;
1040 
1041 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1042 	if (sk == NULL)
1043 		return (NULL);
1044 
1045 	PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1046 	PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1047 	sk->port[pd->sidx] = sport;
1048 	sk->port[pd->didx] = dport;
1049 	sk->proto = pd->proto;
1050 	sk->af = pd->af;
1051 
1052 	return (sk);
1053 }
1054 
1055 struct pf_state_key *
1056 pf_state_key_clone(struct pf_state_key *orig)
1057 {
1058 	struct pf_state_key *sk;
1059 
1060 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1061 	if (sk == NULL)
1062 		return (NULL);
1063 
1064 	bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1065 
1066 	return (sk);
1067 }
1068 
1069 int
1070 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1071     struct pf_state_key *sks, struct pf_state *s)
1072 {
1073 	struct pf_idhash *ih;
1074 	struct pf_state *cur;
1075 
1076 	KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1077 	    ("%s: sks not pristine", __func__));
1078 	KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1079 	    ("%s: skw not pristine", __func__));
1080 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1081 
1082 	s->kif = kif;
1083 
1084 	if (s->id == 0 && s->creatorid == 0) {
1085 		/* XXX: should be atomic, but probability of collision low */
1086 		if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1087 			V_pf_stateid[curcpu] = 1;
1088 		s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1089 		s->id = htobe64(s->id);
1090 		s->creatorid = V_pf_status.hostid;
1091 	}
1092 
1093 	if (pf_state_key_attach(skw, sks, s))
1094 		return (-1);
1095 
1096 	ih = &V_pf_idhash[PF_IDHASH(s)];
1097 	PF_HASHROW_LOCK(ih);
1098 	LIST_FOREACH(cur, &ih->states, entry)
1099 		if (cur->id == s->id && cur->creatorid == s->creatorid)
1100 			break;
1101 
1102 	if (cur != NULL) {
1103 		PF_HASHROW_UNLOCK(ih);
1104 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1105 			printf("pf: state insert failed: "
1106 			    "id: %016llx creatorid: %08x",
1107 			    (unsigned long long)be64toh(s->id),
1108 			    ntohl(s->creatorid));
1109 			printf("\n");
1110 		}
1111 		pf_detach_state(s);
1112 		return (-1);
1113 	}
1114 	LIST_INSERT_HEAD(&ih->states, s, entry);
1115 	/* One for keys, one for ID hash. */
1116 	refcount_init(&s->refs, 2);
1117 
1118 	V_pf_status.fcounters[FCNT_STATE_INSERT]++;
1119 	if (pfsync_insert_state_ptr != NULL)
1120 		pfsync_insert_state_ptr(s);
1121 
1122 	/* Returns locked. */
1123 	return (0);
1124 }
1125 
1126 /*
1127  * Find state by ID: returns with locked row on success.
1128  */
1129 struct pf_state *
1130 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1131 {
1132 	struct pf_idhash *ih;
1133 	struct pf_state *s;
1134 
1135 	V_pf_status.fcounters[FCNT_STATE_SEARCH]++;
1136 
1137 	ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))];
1138 
1139 	PF_HASHROW_LOCK(ih);
1140 	LIST_FOREACH(s, &ih->states, entry)
1141 		if (s->id == id && s->creatorid == creatorid)
1142 			break;
1143 
1144 	if (s == NULL)
1145 		PF_HASHROW_UNLOCK(ih);
1146 
1147 	return (s);
1148 }
1149 
1150 /*
1151  * Find state by key.
1152  * Returns with ID hash slot locked on success.
1153  */
1154 static struct pf_state *
1155 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1156 {
1157 	struct pf_keyhash	*kh;
1158 	struct pf_state_key	*sk;
1159 	struct pf_state		*s;
1160 	int idx;
1161 
1162 	V_pf_status.fcounters[FCNT_STATE_SEARCH]++;
1163 
1164 	kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1165 
1166 	PF_HASHROW_LOCK(kh);
1167 	LIST_FOREACH(sk, &kh->keys, entry)
1168 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1169 			break;
1170 	if (sk == NULL) {
1171 		PF_HASHROW_UNLOCK(kh);
1172 		return (NULL);
1173 	}
1174 
1175 	idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1176 
1177 	/* List is sorted, if-bound states before floating ones. */
1178 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1179 		if (s->kif == V_pfi_all || s->kif == kif) {
1180 			PF_STATE_LOCK(s);
1181 			PF_HASHROW_UNLOCK(kh);
1182 			if (s->timeout == PFTM_UNLINKED) {
1183 				/*
1184 				 * State is being processed
1185 				 * by pf_unlink_state() in
1186 				 * an other thread.
1187 				 */
1188 				PF_STATE_UNLOCK(s);
1189 				return (NULL);
1190 			}
1191 			return (s);
1192 		}
1193 	PF_HASHROW_UNLOCK(kh);
1194 
1195 	return (NULL);
1196 }
1197 
1198 struct pf_state *
1199 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1200 {
1201 	struct pf_keyhash	*kh;
1202 	struct pf_state_key	*sk;
1203 	struct pf_state		*s, *ret = NULL;
1204 	int			 idx, inout = 0;
1205 
1206 	V_pf_status.fcounters[FCNT_STATE_SEARCH]++;
1207 
1208 	kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1209 
1210 	PF_HASHROW_LOCK(kh);
1211 	LIST_FOREACH(sk, &kh->keys, entry)
1212 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1213 			break;
1214 	if (sk == NULL) {
1215 		PF_HASHROW_UNLOCK(kh);
1216 		return (NULL);
1217 	}
1218 	switch (dir) {
1219 	case PF_IN:
1220 		idx = PF_SK_WIRE;
1221 		break;
1222 	case PF_OUT:
1223 		idx = PF_SK_STACK;
1224 		break;
1225 	case PF_INOUT:
1226 		idx = PF_SK_WIRE;
1227 		inout = 1;
1228 		break;
1229 	default:
1230 		panic("%s: dir %u", __func__, dir);
1231 	}
1232 second_run:
1233 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1234 		if (more == NULL) {
1235 			PF_HASHROW_UNLOCK(kh);
1236 			return (s);
1237 		}
1238 
1239 		if (ret)
1240 			(*more)++;
1241 		else
1242 			ret = s;
1243 	}
1244 	if (inout == 1) {
1245 		inout = 0;
1246 		idx = PF_SK_STACK;
1247 		goto second_run;
1248 	}
1249 	PF_HASHROW_UNLOCK(kh);
1250 
1251 	return (ret);
1252 }
1253 
1254 /* END state table stuff */
1255 
1256 static void
1257 pf_send(struct pf_send_entry *pfse)
1258 {
1259 
1260 	PF_SENDQ_LOCK();
1261 	STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1262 	PF_SENDQ_UNLOCK();
1263 	swi_sched(V_pf_swi_cookie, 0);
1264 }
1265 
1266 void
1267 pf_intr(void *v)
1268 {
1269 	struct pf_send_head queue;
1270 	struct pf_send_entry *pfse, *next;
1271 
1272 	CURVNET_SET((struct vnet *)v);
1273 
1274 	PF_SENDQ_LOCK();
1275 	queue = V_pf_sendqueue;
1276 	STAILQ_INIT(&V_pf_sendqueue);
1277 	PF_SENDQ_UNLOCK();
1278 
1279 	STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1280 		switch (pfse->pfse_type) {
1281 #ifdef INET
1282 		case PFSE_IP:
1283 			ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1284 			break;
1285 		case PFSE_ICMP:
1286 			icmp_error(pfse->pfse_m, pfse->pfse_icmp_type,
1287 			    pfse->pfse_icmp_code, 0, pfse->pfse_icmp_mtu);
1288 			break;
1289 #endif /* INET */
1290 #ifdef INET6
1291 		case PFSE_IP6:
1292 			ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1293 			    NULL);
1294 			break;
1295 		case PFSE_ICMP6:
1296 			icmp6_error(pfse->pfse_m, pfse->pfse_icmp_type,
1297 			    pfse->pfse_icmp_code, pfse->pfse_icmp_mtu);
1298 			break;
1299 #endif /* INET6 */
1300 		default:
1301 			panic("%s: unknown type", __func__);
1302 		}
1303 		free(pfse, M_PFTEMP);
1304 	}
1305 	CURVNET_RESTORE();
1306 }
1307 
1308 void
1309 pf_purge_thread(void *v)
1310 {
1311 	u_int idx = 0;
1312 
1313 	CURVNET_SET((struct vnet *)v);
1314 
1315 	for (;;) {
1316 		PF_RULES_RLOCK();
1317 		rw_sleep(pf_purge_thread, &pf_rules_lock, 0, "pftm", hz / 10);
1318 
1319 		if (V_pf_end_threads) {
1320 			/*
1321 			 * To cleanse up all kifs and rules we need
1322 			 * two runs: first one clears reference flags,
1323 			 * then pf_purge_expired_states() doesn't
1324 			 * raise them, and then second run frees.
1325 			 */
1326 			PF_RULES_RUNLOCK();
1327 			pf_purge_unlinked_rules();
1328 			pfi_kif_purge();
1329 
1330 			/*
1331 			 * Now purge everything.
1332 			 */
1333 			pf_purge_expired_states(0, V_pf_hashmask);
1334 			pf_purge_expired_fragments();
1335 			pf_purge_expired_src_nodes();
1336 
1337 			/*
1338 			 * Now all kifs & rules should be unreferenced,
1339 			 * thus should be successfully freed.
1340 			 */
1341 			pf_purge_unlinked_rules();
1342 			pfi_kif_purge();
1343 
1344 			/*
1345 			 * Announce success and exit.
1346 			 */
1347 			PF_RULES_RLOCK();
1348 			V_pf_end_threads++;
1349 			PF_RULES_RUNLOCK();
1350 			wakeup(pf_purge_thread);
1351 			kproc_exit(0);
1352 		}
1353 		PF_RULES_RUNLOCK();
1354 
1355 		/* Process 1/interval fraction of the state table every run. */
1356 		idx = pf_purge_expired_states(idx, V_pf_hashmask /
1357 			    (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1358 
1359 		/* Purge other expired types every PFTM_INTERVAL seconds. */
1360 		if (idx == 0) {
1361 			/*
1362 			 * Order is important:
1363 			 * - states and src nodes reference rules
1364 			 * - states and rules reference kifs
1365 			 */
1366 			pf_purge_expired_fragments();
1367 			pf_purge_expired_src_nodes();
1368 			pf_purge_unlinked_rules();
1369 			pfi_kif_purge();
1370 		}
1371 	}
1372 	/* not reached */
1373 	CURVNET_RESTORE();
1374 }
1375 
1376 u_int32_t
1377 pf_state_expires(const struct pf_state *state)
1378 {
1379 	u_int32_t	timeout;
1380 	u_int32_t	start;
1381 	u_int32_t	end;
1382 	u_int32_t	states;
1383 
1384 	/* handle all PFTM_* > PFTM_MAX here */
1385 	if (state->timeout == PFTM_PURGE)
1386 		return (time_uptime);
1387 	if (state->timeout == PFTM_UNTIL_PACKET)
1388 		return (0);
1389 	KASSERT(state->timeout != PFTM_UNLINKED,
1390 	    ("pf_state_expires: timeout == PFTM_UNLINKED"));
1391 	KASSERT((state->timeout < PFTM_MAX),
1392 	    ("pf_state_expires: timeout > PFTM_MAX"));
1393 	timeout = state->rule.ptr->timeout[state->timeout];
1394 	if (!timeout)
1395 		timeout = V_pf_default_rule.timeout[state->timeout];
1396 	start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1397 	if (start) {
1398 		end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1399 		states = state->rule.ptr->states_cur;	/* XXXGL */
1400 	} else {
1401 		start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1402 		end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1403 		states = V_pf_status.states;
1404 	}
1405 	if (end && states > start && start < end) {
1406 		if (states < end)
1407 			return (state->expire + timeout * (end - states) /
1408 			    (end - start));
1409 		else
1410 			return (time_uptime);
1411 	}
1412 	return (state->expire + timeout);
1413 }
1414 
1415 void
1416 pf_purge_expired_src_nodes()
1417 {
1418 	struct pf_srchash	*sh;
1419 	struct pf_src_node	*cur, *next;
1420 	int i;
1421 
1422 	for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
1423 	    PF_HASHROW_LOCK(sh);
1424 	    LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1425 		if (cur->states <= 0 && cur->expire <= time_uptime) {
1426 			if (cur->rule.ptr != NULL)
1427 				cur->rule.ptr->src_nodes--;
1428 			LIST_REMOVE(cur, entry);
1429 			V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS]++;
1430 			V_pf_status.src_nodes--;
1431 			uma_zfree(V_pf_sources_z, cur);
1432 		} else if (cur->rule.ptr != NULL)
1433 			cur->rule.ptr->rule_flag |= PFRULE_REFS;
1434 	    PF_HASHROW_UNLOCK(sh);
1435 	}
1436 }
1437 
1438 static void
1439 pf_src_tree_remove_state(struct pf_state *s)
1440 {
1441 	u_int32_t timeout;
1442 
1443 	if (s->src_node != NULL) {
1444 		if (s->src.tcp_est)
1445 			--s->src_node->conn;
1446 		if (--s->src_node->states <= 0) {
1447 			timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1448 			if (!timeout)
1449 				timeout =
1450 				    V_pf_default_rule.timeout[PFTM_SRC_NODE];
1451 			s->src_node->expire = time_uptime + timeout;
1452 		}
1453 	}
1454 	if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1455 		if (--s->nat_src_node->states <= 0) {
1456 			timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1457 			if (!timeout)
1458 				timeout =
1459 				    V_pf_default_rule.timeout[PFTM_SRC_NODE];
1460 			s->nat_src_node->expire = time_uptime + timeout;
1461 		}
1462 	}
1463 	s->src_node = s->nat_src_node = NULL;
1464 }
1465 
1466 /*
1467  * Unlink and potentilly free a state. Function may be
1468  * called with ID hash row locked, but always returns
1469  * unlocked, since it needs to go through key hash locking.
1470  */
1471 int
1472 pf_unlink_state(struct pf_state *s, u_int flags)
1473 {
1474 	struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1475 
1476 	if ((flags & PF_ENTER_LOCKED) == 0)
1477 		PF_HASHROW_LOCK(ih);
1478 	else
1479 		PF_HASHROW_ASSERT(ih);
1480 
1481 	if (s->timeout == PFTM_UNLINKED) {
1482 		/*
1483 		 * State is being processed
1484 		 * by pf_unlink_state() in
1485 		 * an other thread.
1486 		 */
1487 		PF_HASHROW_UNLOCK(ih);
1488 		return (0);	/* XXXGL: undefined actually */
1489 	}
1490 
1491 	if (s->src.state == PF_TCPS_PROXY_DST) {
1492 		/* XXX wire key the right one? */
1493 		pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1494 		    &s->key[PF_SK_WIRE]->addr[1],
1495 		    &s->key[PF_SK_WIRE]->addr[0],
1496 		    s->key[PF_SK_WIRE]->port[1],
1497 		    s->key[PF_SK_WIRE]->port[0],
1498 		    s->src.seqhi, s->src.seqlo + 1,
1499 		    TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1500 	}
1501 
1502 	LIST_REMOVE(s, entry);
1503 	pf_src_tree_remove_state(s);
1504 
1505 	if (pfsync_delete_state_ptr != NULL)
1506 		pfsync_delete_state_ptr(s);
1507 
1508 	--s->rule.ptr->states_cur;
1509 	if (s->nat_rule.ptr != NULL)
1510 		--s->nat_rule.ptr->states_cur;
1511 	if (s->anchor.ptr != NULL)
1512 		--s->anchor.ptr->states_cur;
1513 
1514 	s->timeout = PFTM_UNLINKED;
1515 
1516 	PF_HASHROW_UNLOCK(ih);
1517 
1518 	pf_detach_state(s);
1519 	refcount_release(&s->refs);
1520 
1521 	return (pf_release_state(s));
1522 }
1523 
1524 void
1525 pf_free_state(struct pf_state *cur)
1526 {
1527 
1528 	KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1529 	KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1530 	    cur->timeout));
1531 
1532 	pf_normalize_tcp_cleanup(cur);
1533 	uma_zfree(V_pf_state_z, cur);
1534 	V_pf_status.fcounters[FCNT_STATE_REMOVALS]++;
1535 }
1536 
1537 /*
1538  * Called only from pf_purge_thread(), thus serialized.
1539  */
1540 static u_int
1541 pf_purge_expired_states(u_int i, int maxcheck)
1542 {
1543 	struct pf_idhash *ih;
1544 	struct pf_state *s;
1545 
1546 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1547 
1548 	/*
1549 	 * Go through hash and unlink states that expire now.
1550 	 */
1551 	while (maxcheck > 0) {
1552 
1553 		ih = &V_pf_idhash[i];
1554 relock:
1555 		PF_HASHROW_LOCK(ih);
1556 		LIST_FOREACH(s, &ih->states, entry) {
1557 			if (pf_state_expires(s) <= time_uptime) {
1558 				V_pf_status.states -=
1559 				    pf_unlink_state(s, PF_ENTER_LOCKED);
1560 				goto relock;
1561 			}
1562 			s->rule.ptr->rule_flag |= PFRULE_REFS;
1563 			if (s->nat_rule.ptr != NULL)
1564 				s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1565 			if (s->anchor.ptr != NULL)
1566 				s->anchor.ptr->rule_flag |= PFRULE_REFS;
1567 			s->kif->pfik_flags |= PFI_IFLAG_REFS;
1568 			if (s->rt_kif)
1569 				s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1570 		}
1571 		PF_HASHROW_UNLOCK(ih);
1572 
1573 		/* Return when we hit end of hash. */
1574 		if (++i > V_pf_hashmask) {
1575 			V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1576 			return (0);
1577 		}
1578 
1579 		maxcheck--;
1580 	}
1581 
1582 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1583 
1584 	return (i);
1585 }
1586 
1587 static void
1588 pf_purge_unlinked_rules()
1589 {
1590 	struct pf_rulequeue tmpq;
1591 	struct pf_rule *r, *r1;
1592 
1593 	/*
1594 	 * If we have overloading task pending, then we'd
1595 	 * better skip purging this time. There is a tiny
1596 	 * probability that overloading task references
1597 	 * an already unlinked rule.
1598 	 */
1599 	PF_OVERLOADQ_LOCK();
1600 	if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1601 		PF_OVERLOADQ_UNLOCK();
1602 		return;
1603 	}
1604 	PF_OVERLOADQ_UNLOCK();
1605 
1606 	/*
1607 	 * Do naive mark-and-sweep garbage collecting of old rules.
1608 	 * Reference flag is raised by pf_purge_expired_states()
1609 	 * and pf_purge_expired_src_nodes().
1610 	 *
1611 	 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1612 	 * use a temporary queue.
1613 	 */
1614 	TAILQ_INIT(&tmpq);
1615 	PF_UNLNKDRULES_LOCK();
1616 	TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1617 		if (!(r->rule_flag & PFRULE_REFS)) {
1618 			TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1619 			TAILQ_INSERT_TAIL(&tmpq, r, entries);
1620 		} else
1621 			r->rule_flag &= ~PFRULE_REFS;
1622 	}
1623 	PF_UNLNKDRULES_UNLOCK();
1624 
1625 	if (!TAILQ_EMPTY(&tmpq)) {
1626 		PF_RULES_WLOCK();
1627 		TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1628 			TAILQ_REMOVE(&tmpq, r, entries);
1629 			pf_free_rule(r);
1630 		}
1631 		PF_RULES_WUNLOCK();
1632 	}
1633 }
1634 
1635 void
1636 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1637 {
1638 	switch (af) {
1639 #ifdef INET
1640 	case AF_INET: {
1641 		u_int32_t a = ntohl(addr->addr32[0]);
1642 		printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1643 		    (a>>8)&255, a&255);
1644 		if (p) {
1645 			p = ntohs(p);
1646 			printf(":%u", p);
1647 		}
1648 		break;
1649 	}
1650 #endif /* INET */
1651 #ifdef INET6
1652 	case AF_INET6: {
1653 		u_int16_t b;
1654 		u_int8_t i, curstart, curend, maxstart, maxend;
1655 		curstart = curend = maxstart = maxend = 255;
1656 		for (i = 0; i < 8; i++) {
1657 			if (!addr->addr16[i]) {
1658 				if (curstart == 255)
1659 					curstart = i;
1660 				curend = i;
1661 			} else {
1662 				if ((curend - curstart) >
1663 				    (maxend - maxstart)) {
1664 					maxstart = curstart;
1665 					maxend = curend;
1666 				}
1667 				curstart = curend = 255;
1668 			}
1669 		}
1670 		if ((curend - curstart) >
1671 		    (maxend - maxstart)) {
1672 			maxstart = curstart;
1673 			maxend = curend;
1674 		}
1675 		for (i = 0; i < 8; i++) {
1676 			if (i >= maxstart && i <= maxend) {
1677 				if (i == 0)
1678 					printf(":");
1679 				if (i == maxend)
1680 					printf(":");
1681 			} else {
1682 				b = ntohs(addr->addr16[i]);
1683 				printf("%x", b);
1684 				if (i < 7)
1685 					printf(":");
1686 			}
1687 		}
1688 		if (p) {
1689 			p = ntohs(p);
1690 			printf("[%u]", p);
1691 		}
1692 		break;
1693 	}
1694 #endif /* INET6 */
1695 	}
1696 }
1697 
1698 void
1699 pf_print_state(struct pf_state *s)
1700 {
1701 	pf_print_state_parts(s, NULL, NULL);
1702 }
1703 
1704 static void
1705 pf_print_state_parts(struct pf_state *s,
1706     struct pf_state_key *skwp, struct pf_state_key *sksp)
1707 {
1708 	struct pf_state_key *skw, *sks;
1709 	u_int8_t proto, dir;
1710 
1711 	/* Do our best to fill these, but they're skipped if NULL */
1712 	skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1713 	sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1714 	proto = skw ? skw->proto : (sks ? sks->proto : 0);
1715 	dir = s ? s->direction : 0;
1716 
1717 	switch (proto) {
1718 	case IPPROTO_IPV4:
1719 		printf("IPv4");
1720 		break;
1721 	case IPPROTO_IPV6:
1722 		printf("IPv6");
1723 		break;
1724 	case IPPROTO_TCP:
1725 		printf("TCP");
1726 		break;
1727 	case IPPROTO_UDP:
1728 		printf("UDP");
1729 		break;
1730 	case IPPROTO_ICMP:
1731 		printf("ICMP");
1732 		break;
1733 	case IPPROTO_ICMPV6:
1734 		printf("ICMPv6");
1735 		break;
1736 	default:
1737 		printf("%u", skw->proto);
1738 		break;
1739 	}
1740 	switch (dir) {
1741 	case PF_IN:
1742 		printf(" in");
1743 		break;
1744 	case PF_OUT:
1745 		printf(" out");
1746 		break;
1747 	}
1748 	if (skw) {
1749 		printf(" wire: ");
1750 		pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1751 		printf(" ");
1752 		pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1753 	}
1754 	if (sks) {
1755 		printf(" stack: ");
1756 		if (sks != skw) {
1757 			pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1758 			printf(" ");
1759 			pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1760 		} else
1761 			printf("-");
1762 	}
1763 	if (s) {
1764 		if (proto == IPPROTO_TCP) {
1765 			printf(" [lo=%u high=%u win=%u modulator=%u",
1766 			    s->src.seqlo, s->src.seqhi,
1767 			    s->src.max_win, s->src.seqdiff);
1768 			if (s->src.wscale && s->dst.wscale)
1769 				printf(" wscale=%u",
1770 				    s->src.wscale & PF_WSCALE_MASK);
1771 			printf("]");
1772 			printf(" [lo=%u high=%u win=%u modulator=%u",
1773 			    s->dst.seqlo, s->dst.seqhi,
1774 			    s->dst.max_win, s->dst.seqdiff);
1775 			if (s->src.wscale && s->dst.wscale)
1776 				printf(" wscale=%u",
1777 				s->dst.wscale & PF_WSCALE_MASK);
1778 			printf("]");
1779 		}
1780 		printf(" %u:%u", s->src.state, s->dst.state);
1781 	}
1782 }
1783 
1784 void
1785 pf_print_flags(u_int8_t f)
1786 {
1787 	if (f)
1788 		printf(" ");
1789 	if (f & TH_FIN)
1790 		printf("F");
1791 	if (f & TH_SYN)
1792 		printf("S");
1793 	if (f & TH_RST)
1794 		printf("R");
1795 	if (f & TH_PUSH)
1796 		printf("P");
1797 	if (f & TH_ACK)
1798 		printf("A");
1799 	if (f & TH_URG)
1800 		printf("U");
1801 	if (f & TH_ECE)
1802 		printf("E");
1803 	if (f & TH_CWR)
1804 		printf("W");
1805 }
1806 
1807 #define	PF_SET_SKIP_STEPS(i)					\
1808 	do {							\
1809 		while (head[i] != cur) {			\
1810 			head[i]->skip[i].ptr = cur;		\
1811 			head[i] = TAILQ_NEXT(head[i], entries);	\
1812 		}						\
1813 	} while (0)
1814 
1815 void
1816 pf_calc_skip_steps(struct pf_rulequeue *rules)
1817 {
1818 	struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1819 	int i;
1820 
1821 	cur = TAILQ_FIRST(rules);
1822 	prev = cur;
1823 	for (i = 0; i < PF_SKIP_COUNT; ++i)
1824 		head[i] = cur;
1825 	while (cur != NULL) {
1826 
1827 		if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1828 			PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1829 		if (cur->direction != prev->direction)
1830 			PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1831 		if (cur->af != prev->af)
1832 			PF_SET_SKIP_STEPS(PF_SKIP_AF);
1833 		if (cur->proto != prev->proto)
1834 			PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
1835 		if (cur->src.neg != prev->src.neg ||
1836 		    pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
1837 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
1838 		if (cur->src.port[0] != prev->src.port[0] ||
1839 		    cur->src.port[1] != prev->src.port[1] ||
1840 		    cur->src.port_op != prev->src.port_op)
1841 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
1842 		if (cur->dst.neg != prev->dst.neg ||
1843 		    pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
1844 			PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
1845 		if (cur->dst.port[0] != prev->dst.port[0] ||
1846 		    cur->dst.port[1] != prev->dst.port[1] ||
1847 		    cur->dst.port_op != prev->dst.port_op)
1848 			PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
1849 
1850 		prev = cur;
1851 		cur = TAILQ_NEXT(cur, entries);
1852 	}
1853 	for (i = 0; i < PF_SKIP_COUNT; ++i)
1854 		PF_SET_SKIP_STEPS(i);
1855 }
1856 
1857 static int
1858 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
1859 {
1860 	if (aw1->type != aw2->type)
1861 		return (1);
1862 	switch (aw1->type) {
1863 	case PF_ADDR_ADDRMASK:
1864 	case PF_ADDR_RANGE:
1865 		if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0))
1866 			return (1);
1867 		if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0))
1868 			return (1);
1869 		return (0);
1870 	case PF_ADDR_DYNIFTL:
1871 		return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
1872 	case PF_ADDR_NOROUTE:
1873 	case PF_ADDR_URPFFAILED:
1874 		return (0);
1875 	case PF_ADDR_TABLE:
1876 		return (aw1->p.tbl != aw2->p.tbl);
1877 	default:
1878 		printf("invalid address type: %d\n", aw1->type);
1879 		return (1);
1880 	}
1881 }
1882 
1883 u_int16_t
1884 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
1885 {
1886 	u_int32_t	l;
1887 
1888 	if (udp && !cksum)
1889 		return (0x0000);
1890 	l = cksum + old - new;
1891 	l = (l >> 16) + (l & 65535);
1892 	l = l & 65535;
1893 	if (udp && !l)
1894 		return (0xFFFF);
1895 	return (l);
1896 }
1897 
1898 static void
1899 pf_change_ap(struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc,
1900     struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af)
1901 {
1902 	struct pf_addr	ao;
1903 	u_int16_t	po = *p;
1904 
1905 	PF_ACPY(&ao, a, af);
1906 	PF_ACPY(a, an, af);
1907 
1908 	*p = pn;
1909 
1910 	switch (af) {
1911 #ifdef INET
1912 	case AF_INET:
1913 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
1914 		    ao.addr16[0], an->addr16[0], 0),
1915 		    ao.addr16[1], an->addr16[1], 0);
1916 		*p = pn;
1917 		*pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
1918 		    ao.addr16[0], an->addr16[0], u),
1919 		    ao.addr16[1], an->addr16[1], u),
1920 		    po, pn, u);
1921 		break;
1922 #endif /* INET */
1923 #ifdef INET6
1924 	case AF_INET6:
1925 		*pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1926 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1927 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
1928 		    ao.addr16[0], an->addr16[0], u),
1929 		    ao.addr16[1], an->addr16[1], u),
1930 		    ao.addr16[2], an->addr16[2], u),
1931 		    ao.addr16[3], an->addr16[3], u),
1932 		    ao.addr16[4], an->addr16[4], u),
1933 		    ao.addr16[5], an->addr16[5], u),
1934 		    ao.addr16[6], an->addr16[6], u),
1935 		    ao.addr16[7], an->addr16[7], u),
1936 		    po, pn, u);
1937 		break;
1938 #endif /* INET6 */
1939 	}
1940 }
1941 
1942 
1943 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
1944 void
1945 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
1946 {
1947 	u_int32_t	ao;
1948 
1949 	memcpy(&ao, a, sizeof(ao));
1950 	memcpy(a, &an, sizeof(u_int32_t));
1951 	*c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
1952 	    ao % 65536, an % 65536, u);
1953 }
1954 
1955 #ifdef INET6
1956 static void
1957 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
1958 {
1959 	struct pf_addr	ao;
1960 
1961 	PF_ACPY(&ao, a, AF_INET6);
1962 	PF_ACPY(a, an, AF_INET6);
1963 
1964 	*c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1965 	    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1966 	    pf_cksum_fixup(pf_cksum_fixup(*c,
1967 	    ao.addr16[0], an->addr16[0], u),
1968 	    ao.addr16[1], an->addr16[1], u),
1969 	    ao.addr16[2], an->addr16[2], u),
1970 	    ao.addr16[3], an->addr16[3], u),
1971 	    ao.addr16[4], an->addr16[4], u),
1972 	    ao.addr16[5], an->addr16[5], u),
1973 	    ao.addr16[6], an->addr16[6], u),
1974 	    ao.addr16[7], an->addr16[7], u);
1975 }
1976 #endif /* INET6 */
1977 
1978 static void
1979 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
1980     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
1981     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
1982 {
1983 	struct pf_addr	oia, ooa;
1984 
1985 	PF_ACPY(&oia, ia, af);
1986 	if (oa)
1987 		PF_ACPY(&ooa, oa, af);
1988 
1989 	/* Change inner protocol port, fix inner protocol checksum. */
1990 	if (ip != NULL) {
1991 		u_int16_t	oip = *ip;
1992 		u_int32_t	opc;
1993 
1994 		if (pc != NULL)
1995 			opc = *pc;
1996 		*ip = np;
1997 		if (pc != NULL)
1998 			*pc = pf_cksum_fixup(*pc, oip, *ip, u);
1999 		*ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2000 		if (pc != NULL)
2001 			*ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2002 	}
2003 	/* Change inner ip address, fix inner ip and icmp checksums. */
2004 	PF_ACPY(ia, na, af);
2005 	switch (af) {
2006 #ifdef INET
2007 	case AF_INET: {
2008 		u_int32_t	 oh2c = *h2c;
2009 
2010 		*h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2011 		    oia.addr16[0], ia->addr16[0], 0),
2012 		    oia.addr16[1], ia->addr16[1], 0);
2013 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2014 		    oia.addr16[0], ia->addr16[0], 0),
2015 		    oia.addr16[1], ia->addr16[1], 0);
2016 		*ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2017 		break;
2018 	}
2019 #endif /* INET */
2020 #ifdef INET6
2021 	case AF_INET6:
2022 		*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2023 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2024 		    pf_cksum_fixup(pf_cksum_fixup(*ic,
2025 		    oia.addr16[0], ia->addr16[0], u),
2026 		    oia.addr16[1], ia->addr16[1], u),
2027 		    oia.addr16[2], ia->addr16[2], u),
2028 		    oia.addr16[3], ia->addr16[3], u),
2029 		    oia.addr16[4], ia->addr16[4], u),
2030 		    oia.addr16[5], ia->addr16[5], u),
2031 		    oia.addr16[6], ia->addr16[6], u),
2032 		    oia.addr16[7], ia->addr16[7], u);
2033 		break;
2034 #endif /* INET6 */
2035 	}
2036 	/* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2037 	if (oa) {
2038 		PF_ACPY(oa, na, af);
2039 		switch (af) {
2040 #ifdef INET
2041 		case AF_INET:
2042 			*hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2043 			    ooa.addr16[0], oa->addr16[0], 0),
2044 			    ooa.addr16[1], oa->addr16[1], 0);
2045 			break;
2046 #endif /* INET */
2047 #ifdef INET6
2048 		case AF_INET6:
2049 			*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2050 			    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2051 			    pf_cksum_fixup(pf_cksum_fixup(*ic,
2052 			    ooa.addr16[0], oa->addr16[0], u),
2053 			    ooa.addr16[1], oa->addr16[1], u),
2054 			    ooa.addr16[2], oa->addr16[2], u),
2055 			    ooa.addr16[3], oa->addr16[3], u),
2056 			    ooa.addr16[4], oa->addr16[4], u),
2057 			    ooa.addr16[5], oa->addr16[5], u),
2058 			    ooa.addr16[6], oa->addr16[6], u),
2059 			    ooa.addr16[7], oa->addr16[7], u);
2060 			break;
2061 #endif /* INET6 */
2062 		}
2063 	}
2064 }
2065 
2066 
2067 /*
2068  * Need to modulate the sequence numbers in the TCP SACK option
2069  * (credits to Krzysztof Pfaff for report and patch)
2070  */
2071 static int
2072 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2073     struct tcphdr *th, struct pf_state_peer *dst)
2074 {
2075 	int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2076 	u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2077 	int copyback = 0, i, olen;
2078 	struct sackblk sack;
2079 
2080 #define	TCPOLEN_SACKLEN	(TCPOLEN_SACK + 2)
2081 	if (hlen < TCPOLEN_SACKLEN ||
2082 	    !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2083 		return 0;
2084 
2085 	while (hlen >= TCPOLEN_SACKLEN) {
2086 		olen = opt[1];
2087 		switch (*opt) {
2088 		case TCPOPT_EOL:	/* FALLTHROUGH */
2089 		case TCPOPT_NOP:
2090 			opt++;
2091 			hlen--;
2092 			break;
2093 		case TCPOPT_SACK:
2094 			if (olen > hlen)
2095 				olen = hlen;
2096 			if (olen >= TCPOLEN_SACKLEN) {
2097 				for (i = 2; i + TCPOLEN_SACK <= olen;
2098 				    i += TCPOLEN_SACK) {
2099 					memcpy(&sack, &opt[i], sizeof(sack));
2100 					pf_change_a(&sack.start, &th->th_sum,
2101 					    htonl(ntohl(sack.start) -
2102 					    dst->seqdiff), 0);
2103 					pf_change_a(&sack.end, &th->th_sum,
2104 					    htonl(ntohl(sack.end) -
2105 					    dst->seqdiff), 0);
2106 					memcpy(&opt[i], &sack, sizeof(sack));
2107 				}
2108 				copyback = 1;
2109 			}
2110 			/* FALLTHROUGH */
2111 		default:
2112 			if (olen < 2)
2113 				olen = 2;
2114 			hlen -= olen;
2115 			opt += olen;
2116 		}
2117 	}
2118 
2119 	if (copyback)
2120 		m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2121 	return (copyback);
2122 }
2123 
2124 static void
2125 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2126     const struct pf_addr *saddr, const struct pf_addr *daddr,
2127     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2128     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2129     u_int16_t rtag, struct ifnet *ifp)
2130 {
2131 	struct pf_send_entry *pfse;
2132 	struct mbuf	*m;
2133 	int		 len, tlen;
2134 #ifdef INET
2135 	struct ip	*h = NULL;
2136 #endif /* INET */
2137 #ifdef INET6
2138 	struct ip6_hdr	*h6 = NULL;
2139 #endif /* INET6 */
2140 	struct tcphdr	*th;
2141 	char		*opt;
2142 	struct pf_mtag  *pf_mtag;
2143 
2144 	len = 0;
2145 	th = NULL;
2146 
2147 	/* maximum segment size tcp option */
2148 	tlen = sizeof(struct tcphdr);
2149 	if (mss)
2150 		tlen += 4;
2151 
2152 	switch (af) {
2153 #ifdef INET
2154 	case AF_INET:
2155 		len = sizeof(struct ip) + tlen;
2156 		break;
2157 #endif /* INET */
2158 #ifdef INET6
2159 	case AF_INET6:
2160 		len = sizeof(struct ip6_hdr) + tlen;
2161 		break;
2162 #endif /* INET6 */
2163 	default:
2164 		panic("%s: unsupported af %d", __func__, af);
2165 	}
2166 
2167 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
2168 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2169 	if (pfse == NULL)
2170 		return;
2171 	m = m_gethdr(M_NOWAIT, MT_HEADER);
2172 	if (m == NULL) {
2173 		free(pfse, M_PFTEMP);
2174 		return;
2175 	}
2176 #ifdef MAC
2177 	mac_netinet_firewall_send(m);
2178 #endif
2179 	if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2180 		free(pfse, M_PFTEMP);
2181 		m_freem(m);
2182 		return;
2183 	}
2184 	if (tag)
2185 		m->m_flags |= M_SKIP_FIREWALL;
2186 	pf_mtag->tag = rtag;
2187 
2188 	if (r != NULL && r->rtableid >= 0)
2189 		M_SETFIB(m, r->rtableid);
2190 
2191 #ifdef ALTQ
2192 	if (r != NULL && r->qid) {
2193 		pf_mtag->qid = r->qid;
2194 
2195 		/* add hints for ecn */
2196 		pf_mtag->hdr = mtod(m, struct ip *);
2197 	}
2198 #endif /* ALTQ */
2199 	m->m_data += max_linkhdr;
2200 	m->m_pkthdr.len = m->m_len = len;
2201 	m->m_pkthdr.rcvif = NULL;
2202 	bzero(m->m_data, len);
2203 	switch (af) {
2204 #ifdef INET
2205 	case AF_INET:
2206 		h = mtod(m, struct ip *);
2207 
2208 		/* IP header fields included in the TCP checksum */
2209 		h->ip_p = IPPROTO_TCP;
2210 		h->ip_len = htons(tlen);
2211 		h->ip_src.s_addr = saddr->v4.s_addr;
2212 		h->ip_dst.s_addr = daddr->v4.s_addr;
2213 
2214 		th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2215 		break;
2216 #endif /* INET */
2217 #ifdef INET6
2218 	case AF_INET6:
2219 		h6 = mtod(m, struct ip6_hdr *);
2220 
2221 		/* IP header fields included in the TCP checksum */
2222 		h6->ip6_nxt = IPPROTO_TCP;
2223 		h6->ip6_plen = htons(tlen);
2224 		memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2225 		memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2226 
2227 		th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2228 		break;
2229 #endif /* INET6 */
2230 	}
2231 
2232 	/* TCP header */
2233 	th->th_sport = sport;
2234 	th->th_dport = dport;
2235 	th->th_seq = htonl(seq);
2236 	th->th_ack = htonl(ack);
2237 	th->th_off = tlen >> 2;
2238 	th->th_flags = flags;
2239 	th->th_win = htons(win);
2240 
2241 	if (mss) {
2242 		opt = (char *)(th + 1);
2243 		opt[0] = TCPOPT_MAXSEG;
2244 		opt[1] = 4;
2245 		HTONS(mss);
2246 		bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2247 	}
2248 
2249 	switch (af) {
2250 #ifdef INET
2251 	case AF_INET:
2252 		/* TCP checksum */
2253 		th->th_sum = in_cksum(m, len);
2254 
2255 		/* Finish the IP header */
2256 		h->ip_v = 4;
2257 		h->ip_hl = sizeof(*h) >> 2;
2258 		h->ip_tos = IPTOS_LOWDELAY;
2259 		h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2260 		h->ip_len = htons(len);
2261 		h->ip_ttl = ttl ? ttl : V_ip_defttl;
2262 		h->ip_sum = 0;
2263 
2264 		pfse->pfse_type = PFSE_IP;
2265 		break;
2266 #endif /* INET */
2267 #ifdef INET6
2268 	case AF_INET6:
2269 		/* TCP checksum */
2270 		th->th_sum = in6_cksum(m, IPPROTO_TCP,
2271 		    sizeof(struct ip6_hdr), tlen);
2272 
2273 		h6->ip6_vfc |= IPV6_VERSION;
2274 		h6->ip6_hlim = IPV6_DEFHLIM;
2275 
2276 		pfse->pfse_type = PFSE_IP6;
2277 		break;
2278 #endif /* INET6 */
2279 	}
2280 	pfse->pfse_m = m;
2281 	pf_send(pfse);
2282 }
2283 
2284 static void
2285 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2286     struct pf_rule *r)
2287 {
2288 	struct pf_send_entry *pfse;
2289 	struct mbuf *m0;
2290 	struct pf_mtag *pf_mtag;
2291 
2292 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
2293 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2294 	if (pfse == NULL)
2295 		return;
2296 
2297 	if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2298 		free(pfse, M_PFTEMP);
2299 		return;
2300 	}
2301 
2302 	if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2303 		free(pfse, M_PFTEMP);
2304 		return;
2305 	}
2306 	/* XXX: revisit */
2307 	m0->m_flags |= M_SKIP_FIREWALL;
2308 
2309 	if (r->rtableid >= 0)
2310 		M_SETFIB(m0, r->rtableid);
2311 
2312 #ifdef ALTQ
2313 	if (r->qid) {
2314 		pf_mtag->qid = r->qid;
2315 		/* add hints for ecn */
2316 		pf_mtag->hdr = mtod(m0, struct ip *);
2317 	}
2318 #endif /* ALTQ */
2319 
2320 	switch (af) {
2321 #ifdef INET
2322 	case AF_INET:
2323 		pfse->pfse_type = PFSE_ICMP;
2324 		break;
2325 #endif /* INET */
2326 #ifdef INET6
2327 	case AF_INET6:
2328 		pfse->pfse_type = PFSE_ICMP6;
2329 		break;
2330 #endif /* INET6 */
2331 	}
2332 	pfse->pfse_m = m0;
2333 	pfse->pfse_icmp_type = type;
2334 	pfse->pfse_icmp_code = code;
2335 	pf_send(pfse);
2336 }
2337 
2338 /*
2339  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2340  * If n is 0, they match if they are equal. If n is != 0, they match if they
2341  * are different.
2342  */
2343 int
2344 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2345     struct pf_addr *b, sa_family_t af)
2346 {
2347 	int	match = 0;
2348 
2349 	switch (af) {
2350 #ifdef INET
2351 	case AF_INET:
2352 		if ((a->addr32[0] & m->addr32[0]) ==
2353 		    (b->addr32[0] & m->addr32[0]))
2354 			match++;
2355 		break;
2356 #endif /* INET */
2357 #ifdef INET6
2358 	case AF_INET6:
2359 		if (((a->addr32[0] & m->addr32[0]) ==
2360 		     (b->addr32[0] & m->addr32[0])) &&
2361 		    ((a->addr32[1] & m->addr32[1]) ==
2362 		     (b->addr32[1] & m->addr32[1])) &&
2363 		    ((a->addr32[2] & m->addr32[2]) ==
2364 		     (b->addr32[2] & m->addr32[2])) &&
2365 		    ((a->addr32[3] & m->addr32[3]) ==
2366 		     (b->addr32[3] & m->addr32[3])))
2367 			match++;
2368 		break;
2369 #endif /* INET6 */
2370 	}
2371 	if (match) {
2372 		if (n)
2373 			return (0);
2374 		else
2375 			return (1);
2376 	} else {
2377 		if (n)
2378 			return (1);
2379 		else
2380 			return (0);
2381 	}
2382 }
2383 
2384 /*
2385  * Return 1 if b <= a <= e, otherwise return 0.
2386  */
2387 int
2388 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2389     struct pf_addr *a, sa_family_t af)
2390 {
2391 	switch (af) {
2392 #ifdef INET
2393 	case AF_INET:
2394 		if ((a->addr32[0] < b->addr32[0]) ||
2395 		    (a->addr32[0] > e->addr32[0]))
2396 			return (0);
2397 		break;
2398 #endif /* INET */
2399 #ifdef INET6
2400 	case AF_INET6: {
2401 		int	i;
2402 
2403 		/* check a >= b */
2404 		for (i = 0; i < 4; ++i)
2405 			if (a->addr32[i] > b->addr32[i])
2406 				break;
2407 			else if (a->addr32[i] < b->addr32[i])
2408 				return (0);
2409 		/* check a <= e */
2410 		for (i = 0; i < 4; ++i)
2411 			if (a->addr32[i] < e->addr32[i])
2412 				break;
2413 			else if (a->addr32[i] > e->addr32[i])
2414 				return (0);
2415 		break;
2416 	}
2417 #endif /* INET6 */
2418 	}
2419 	return (1);
2420 }
2421 
2422 static int
2423 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2424 {
2425 	switch (op) {
2426 	case PF_OP_IRG:
2427 		return ((p > a1) && (p < a2));
2428 	case PF_OP_XRG:
2429 		return ((p < a1) || (p > a2));
2430 	case PF_OP_RRG:
2431 		return ((p >= a1) && (p <= a2));
2432 	case PF_OP_EQ:
2433 		return (p == a1);
2434 	case PF_OP_NE:
2435 		return (p != a1);
2436 	case PF_OP_LT:
2437 		return (p < a1);
2438 	case PF_OP_LE:
2439 		return (p <= a1);
2440 	case PF_OP_GT:
2441 		return (p > a1);
2442 	case PF_OP_GE:
2443 		return (p >= a1);
2444 	}
2445 	return (0); /* never reached */
2446 }
2447 
2448 int
2449 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2450 {
2451 	NTOHS(a1);
2452 	NTOHS(a2);
2453 	NTOHS(p);
2454 	return (pf_match(op, a1, a2, p));
2455 }
2456 
2457 static int
2458 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2459 {
2460 	if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2461 		return (0);
2462 	return (pf_match(op, a1, a2, u));
2463 }
2464 
2465 static int
2466 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2467 {
2468 	if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2469 		return (0);
2470 	return (pf_match(op, a1, a2, g));
2471 }
2472 
2473 int
2474 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2475 {
2476 	if (*tag == -1)
2477 		*tag = mtag;
2478 
2479 	return ((!r->match_tag_not && r->match_tag == *tag) ||
2480 	    (r->match_tag_not && r->match_tag != *tag));
2481 }
2482 
2483 int
2484 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2485 {
2486 
2487 	KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2488 
2489 	if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2490 		return (ENOMEM);
2491 
2492 	pd->pf_mtag->tag = tag;
2493 
2494 	return (0);
2495 }
2496 
2497 #define	PF_ANCHOR_STACKSIZE	32
2498 struct pf_anchor_stackframe {
2499 	struct pf_ruleset	*rs;
2500 	struct pf_rule		*r;	/* XXX: + match bit */
2501 	struct pf_anchor	*child;
2502 };
2503 
2504 /*
2505  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2506  */
2507 #define	PF_ANCHORSTACK_MATCH	0x00000001
2508 #define	PF_ANCHORSTACK_MASK	(PF_ANCHORSTACK_MATCH)
2509 
2510 #define	PF_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2511 #define	PF_ANCHOR_RULE(f)	(struct pf_rule *)			\
2512 				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2513 #define	PF_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 			\
2514 				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2515 } while (0)
2516 
2517 void
2518 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2519     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2520     int *match)
2521 {
2522 	struct pf_anchor_stackframe	*f;
2523 
2524 	PF_RULES_RASSERT();
2525 
2526 	if (match)
2527 		*match = 0;
2528 	if (*depth >= PF_ANCHOR_STACKSIZE) {
2529 		printf("%s: anchor stack overflow on %s\n",
2530 		    __func__, (*r)->anchor->name);
2531 		*r = TAILQ_NEXT(*r, entries);
2532 		return;
2533 	} else if (*depth == 0 && a != NULL)
2534 		*a = *r;
2535 	f = stack + (*depth)++;
2536 	f->rs = *rs;
2537 	f->r = *r;
2538 	if ((*r)->anchor_wildcard) {
2539 		struct pf_anchor_node *parent = &(*r)->anchor->children;
2540 
2541 		if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2542 			*r = NULL;
2543 			return;
2544 		}
2545 		*rs = &f->child->ruleset;
2546 	} else {
2547 		f->child = NULL;
2548 		*rs = &(*r)->anchor->ruleset;
2549 	}
2550 	*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2551 }
2552 
2553 int
2554 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2555     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2556     int *match)
2557 {
2558 	struct pf_anchor_stackframe	*f;
2559 	struct pf_rule *fr;
2560 	int quick = 0;
2561 
2562 	PF_RULES_RASSERT();
2563 
2564 	do {
2565 		if (*depth <= 0)
2566 			break;
2567 		f = stack + *depth - 1;
2568 		fr = PF_ANCHOR_RULE(f);
2569 		if (f->child != NULL) {
2570 			struct pf_anchor_node *parent;
2571 
2572 			/*
2573 			 * This block traverses through
2574 			 * a wildcard anchor.
2575 			 */
2576 			parent = &fr->anchor->children;
2577 			if (match != NULL && *match) {
2578 				/*
2579 				 * If any of "*" matched, then
2580 				 * "foo/ *" matched, mark frame
2581 				 * appropriately.
2582 				 */
2583 				PF_ANCHOR_SET_MATCH(f);
2584 				*match = 0;
2585 			}
2586 			f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2587 			if (f->child != NULL) {
2588 				*rs = &f->child->ruleset;
2589 				*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2590 				if (*r == NULL)
2591 					continue;
2592 				else
2593 					break;
2594 			}
2595 		}
2596 		(*depth)--;
2597 		if (*depth == 0 && a != NULL)
2598 			*a = NULL;
2599 		*rs = f->rs;
2600 		if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2601 			quick = fr->quick;
2602 		*r = TAILQ_NEXT(fr, entries);
2603 	} while (*r == NULL);
2604 
2605 	return (quick);
2606 }
2607 
2608 #ifdef INET6
2609 void
2610 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2611     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2612 {
2613 	switch (af) {
2614 #ifdef INET
2615 	case AF_INET:
2616 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2617 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2618 		break;
2619 #endif /* INET */
2620 	case AF_INET6:
2621 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2622 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2623 		naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2624 		((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2625 		naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2626 		((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2627 		naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2628 		((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2629 		break;
2630 	}
2631 }
2632 
2633 void
2634 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2635 {
2636 	switch (af) {
2637 #ifdef INET
2638 	case AF_INET:
2639 		addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2640 		break;
2641 #endif /* INET */
2642 	case AF_INET6:
2643 		if (addr->addr32[3] == 0xffffffff) {
2644 			addr->addr32[3] = 0;
2645 			if (addr->addr32[2] == 0xffffffff) {
2646 				addr->addr32[2] = 0;
2647 				if (addr->addr32[1] == 0xffffffff) {
2648 					addr->addr32[1] = 0;
2649 					addr->addr32[0] =
2650 					    htonl(ntohl(addr->addr32[0]) + 1);
2651 				} else
2652 					addr->addr32[1] =
2653 					    htonl(ntohl(addr->addr32[1]) + 1);
2654 			} else
2655 				addr->addr32[2] =
2656 				    htonl(ntohl(addr->addr32[2]) + 1);
2657 		} else
2658 			addr->addr32[3] =
2659 			    htonl(ntohl(addr->addr32[3]) + 1);
2660 		break;
2661 	}
2662 }
2663 #endif /* INET6 */
2664 
2665 int
2666 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
2667 {
2668 	struct pf_addr		*saddr, *daddr;
2669 	u_int16_t		 sport, dport;
2670 	struct inpcbinfo	*pi;
2671 	struct inpcb		*inp;
2672 
2673 	pd->lookup.uid = UID_MAX;
2674 	pd->lookup.gid = GID_MAX;
2675 
2676 	switch (pd->proto) {
2677 	case IPPROTO_TCP:
2678 		if (pd->hdr.tcp == NULL)
2679 			return (-1);
2680 		sport = pd->hdr.tcp->th_sport;
2681 		dport = pd->hdr.tcp->th_dport;
2682 		pi = &V_tcbinfo;
2683 		break;
2684 	case IPPROTO_UDP:
2685 		if (pd->hdr.udp == NULL)
2686 			return (-1);
2687 		sport = pd->hdr.udp->uh_sport;
2688 		dport = pd->hdr.udp->uh_dport;
2689 		pi = &V_udbinfo;
2690 		break;
2691 	default:
2692 		return (-1);
2693 	}
2694 	if (direction == PF_IN) {
2695 		saddr = pd->src;
2696 		daddr = pd->dst;
2697 	} else {
2698 		u_int16_t	p;
2699 
2700 		p = sport;
2701 		sport = dport;
2702 		dport = p;
2703 		saddr = pd->dst;
2704 		daddr = pd->src;
2705 	}
2706 	switch (pd->af) {
2707 #ifdef INET
2708 	case AF_INET:
2709 		inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
2710 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
2711 		if (inp == NULL) {
2712 			inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
2713 			   daddr->v4, dport, INPLOOKUP_WILDCARD |
2714 			   INPLOOKUP_RLOCKPCB, NULL, m);
2715 			if (inp == NULL)
2716 				return (-1);
2717 		}
2718 		break;
2719 #endif /* INET */
2720 #ifdef INET6
2721 	case AF_INET6:
2722 		inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
2723 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
2724 		if (inp == NULL) {
2725 			inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
2726 			    &daddr->v6, dport, INPLOOKUP_WILDCARD |
2727 			    INPLOOKUP_RLOCKPCB, NULL, m);
2728 			if (inp == NULL)
2729 				return (-1);
2730 		}
2731 		break;
2732 #endif /* INET6 */
2733 
2734 	default:
2735 		return (-1);
2736 	}
2737 	INP_RLOCK_ASSERT(inp);
2738 	pd->lookup.uid = inp->inp_cred->cr_uid;
2739 	pd->lookup.gid = inp->inp_cred->cr_groups[0];
2740 	INP_RUNLOCK(inp);
2741 
2742 	return (1);
2743 }
2744 
2745 static u_int8_t
2746 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2747 {
2748 	int		 hlen;
2749 	u_int8_t	 hdr[60];
2750 	u_int8_t	*opt, optlen;
2751 	u_int8_t	 wscale = 0;
2752 
2753 	hlen = th_off << 2;		/* hlen <= sizeof(hdr) */
2754 	if (hlen <= sizeof(struct tcphdr))
2755 		return (0);
2756 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2757 		return (0);
2758 	opt = hdr + sizeof(struct tcphdr);
2759 	hlen -= sizeof(struct tcphdr);
2760 	while (hlen >= 3) {
2761 		switch (*opt) {
2762 		case TCPOPT_EOL:
2763 		case TCPOPT_NOP:
2764 			++opt;
2765 			--hlen;
2766 			break;
2767 		case TCPOPT_WINDOW:
2768 			wscale = opt[2];
2769 			if (wscale > TCP_MAX_WINSHIFT)
2770 				wscale = TCP_MAX_WINSHIFT;
2771 			wscale |= PF_WSCALE_FLAG;
2772 			/* FALLTHROUGH */
2773 		default:
2774 			optlen = opt[1];
2775 			if (optlen < 2)
2776 				optlen = 2;
2777 			hlen -= optlen;
2778 			opt += optlen;
2779 			break;
2780 		}
2781 	}
2782 	return (wscale);
2783 }
2784 
2785 static u_int16_t
2786 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2787 {
2788 	int		 hlen;
2789 	u_int8_t	 hdr[60];
2790 	u_int8_t	*opt, optlen;
2791 	u_int16_t	 mss = V_tcp_mssdflt;
2792 
2793 	hlen = th_off << 2;	/* hlen <= sizeof(hdr) */
2794 	if (hlen <= sizeof(struct tcphdr))
2795 		return (0);
2796 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2797 		return (0);
2798 	opt = hdr + sizeof(struct tcphdr);
2799 	hlen -= sizeof(struct tcphdr);
2800 	while (hlen >= TCPOLEN_MAXSEG) {
2801 		switch (*opt) {
2802 		case TCPOPT_EOL:
2803 		case TCPOPT_NOP:
2804 			++opt;
2805 			--hlen;
2806 			break;
2807 		case TCPOPT_MAXSEG:
2808 			bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
2809 			NTOHS(mss);
2810 			/* FALLTHROUGH */
2811 		default:
2812 			optlen = opt[1];
2813 			if (optlen < 2)
2814 				optlen = 2;
2815 			hlen -= optlen;
2816 			opt += optlen;
2817 			break;
2818 		}
2819 	}
2820 	return (mss);
2821 }
2822 
2823 static u_int16_t
2824 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
2825 {
2826 #ifdef INET
2827 	struct sockaddr_in	*dst;
2828 	struct route		 ro;
2829 #endif /* INET */
2830 #ifdef INET6
2831 	struct sockaddr_in6	*dst6;
2832 	struct route_in6	 ro6;
2833 #endif /* INET6 */
2834 	struct rtentry		*rt = NULL;
2835 	int			 hlen = 0;
2836 	u_int16_t		 mss = V_tcp_mssdflt;
2837 
2838 	switch (af) {
2839 #ifdef INET
2840 	case AF_INET:
2841 		hlen = sizeof(struct ip);
2842 		bzero(&ro, sizeof(ro));
2843 		dst = (struct sockaddr_in *)&ro.ro_dst;
2844 		dst->sin_family = AF_INET;
2845 		dst->sin_len = sizeof(*dst);
2846 		dst->sin_addr = addr->v4;
2847 		in_rtalloc_ign(&ro, 0, rtableid);
2848 		rt = ro.ro_rt;
2849 		break;
2850 #endif /* INET */
2851 #ifdef INET6
2852 	case AF_INET6:
2853 		hlen = sizeof(struct ip6_hdr);
2854 		bzero(&ro6, sizeof(ro6));
2855 		dst6 = (struct sockaddr_in6 *)&ro6.ro_dst;
2856 		dst6->sin6_family = AF_INET6;
2857 		dst6->sin6_len = sizeof(*dst6);
2858 		dst6->sin6_addr = addr->v6;
2859 		in6_rtalloc_ign(&ro6, 0, rtableid);
2860 		rt = ro6.ro_rt;
2861 		break;
2862 #endif /* INET6 */
2863 	}
2864 
2865 	if (rt && rt->rt_ifp) {
2866 		mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr);
2867 		mss = max(V_tcp_mssdflt, mss);
2868 		RTFREE(rt);
2869 	}
2870 	mss = min(mss, offer);
2871 	mss = max(mss, 64);		/* sanity - at least max opt space */
2872 	return (mss);
2873 }
2874 
2875 static void
2876 pf_set_rt_ifp(struct pf_state *s, struct pf_addr *saddr)
2877 {
2878 	struct pf_rule *r = s->rule.ptr;
2879 	struct pf_src_node *sn = NULL;
2880 
2881 	s->rt_kif = NULL;
2882 	if (!r->rt || r->rt == PF_FASTROUTE)
2883 		return;
2884 	switch (s->key[PF_SK_WIRE]->af) {
2885 #ifdef INET
2886 	case AF_INET:
2887 		pf_map_addr(AF_INET, r, saddr, &s->rt_addr, NULL, &sn);
2888 		s->rt_kif = r->rpool.cur->kif;
2889 		break;
2890 #endif /* INET */
2891 #ifdef INET6
2892 	case AF_INET6:
2893 		pf_map_addr(AF_INET6, r, saddr, &s->rt_addr, NULL, &sn);
2894 		s->rt_kif = r->rpool.cur->kif;
2895 		break;
2896 #endif /* INET6 */
2897 	}
2898 }
2899 
2900 static u_int32_t
2901 pf_tcp_iss(struct pf_pdesc *pd)
2902 {
2903 	MD5_CTX ctx;
2904 	u_int32_t digest[4];
2905 
2906 	if (V_pf_tcp_secret_init == 0) {
2907 		read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
2908 		MD5Init(&V_pf_tcp_secret_ctx);
2909 		MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
2910 		    sizeof(V_pf_tcp_secret));
2911 		V_pf_tcp_secret_init = 1;
2912 	}
2913 
2914 	ctx = V_pf_tcp_secret_ctx;
2915 
2916 	MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
2917 	MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
2918 	if (pd->af == AF_INET6) {
2919 		MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
2920 		MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
2921 	} else {
2922 		MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
2923 		MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
2924 	}
2925 	MD5Final((u_char *)digest, &ctx);
2926 	V_pf_tcp_iss_off += 4096;
2927 #define	ISN_RANDOM_INCREMENT (4096 - 1)
2928 	return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
2929 	    V_pf_tcp_iss_off);
2930 #undef	ISN_RANDOM_INCREMENT
2931 }
2932 
2933 static int
2934 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
2935     struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
2936     struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
2937 {
2938 	struct pf_rule		*nr = NULL;
2939 	struct pf_addr		* const saddr = pd->src;
2940 	struct pf_addr		* const daddr = pd->dst;
2941 	sa_family_t		 af = pd->af;
2942 	struct pf_rule		*r, *a = NULL;
2943 	struct pf_ruleset	*ruleset = NULL;
2944 	struct pf_src_node	*nsn = NULL;
2945 	struct tcphdr		*th = pd->hdr.tcp;
2946 	struct pf_state_key	*sk = NULL, *nk = NULL;
2947 	u_short			 reason;
2948 	int			 rewrite = 0, hdrlen = 0;
2949 	int			 tag = -1, rtableid = -1;
2950 	int			 asd = 0;
2951 	int			 match = 0;
2952 	int			 state_icmp = 0;
2953 	u_int16_t		 sport = 0, dport = 0;
2954 	u_int16_t		 bproto_sum = 0, bip_sum = 0;
2955 	u_int8_t		 icmptype = 0, icmpcode = 0;
2956 	struct pf_anchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
2957 
2958 	PF_RULES_RASSERT();
2959 
2960 	if (inp != NULL) {
2961 		INP_LOCK_ASSERT(inp);
2962 		pd->lookup.uid = inp->inp_cred->cr_uid;
2963 		pd->lookup.gid = inp->inp_cred->cr_groups[0];
2964 		pd->lookup.done = 1;
2965 	}
2966 
2967 	switch (pd->proto) {
2968 	case IPPROTO_TCP:
2969 		sport = th->th_sport;
2970 		dport = th->th_dport;
2971 		hdrlen = sizeof(*th);
2972 		break;
2973 	case IPPROTO_UDP:
2974 		sport = pd->hdr.udp->uh_sport;
2975 		dport = pd->hdr.udp->uh_dport;
2976 		hdrlen = sizeof(*pd->hdr.udp);
2977 		break;
2978 #ifdef INET
2979 	case IPPROTO_ICMP:
2980 		if (pd->af != AF_INET)
2981 			break;
2982 		sport = dport = pd->hdr.icmp->icmp_id;
2983 		hdrlen = sizeof(*pd->hdr.icmp);
2984 		icmptype = pd->hdr.icmp->icmp_type;
2985 		icmpcode = pd->hdr.icmp->icmp_code;
2986 
2987 		if (icmptype == ICMP_UNREACH ||
2988 		    icmptype == ICMP_SOURCEQUENCH ||
2989 		    icmptype == ICMP_REDIRECT ||
2990 		    icmptype == ICMP_TIMXCEED ||
2991 		    icmptype == ICMP_PARAMPROB)
2992 			state_icmp++;
2993 		break;
2994 #endif /* INET */
2995 #ifdef INET6
2996 	case IPPROTO_ICMPV6:
2997 		if (af != AF_INET6)
2998 			break;
2999 		sport = dport = pd->hdr.icmp6->icmp6_id;
3000 		hdrlen = sizeof(*pd->hdr.icmp6);
3001 		icmptype = pd->hdr.icmp6->icmp6_type;
3002 		icmpcode = pd->hdr.icmp6->icmp6_code;
3003 
3004 		if (icmptype == ICMP6_DST_UNREACH ||
3005 		    icmptype == ICMP6_PACKET_TOO_BIG ||
3006 		    icmptype == ICMP6_TIME_EXCEEDED ||
3007 		    icmptype == ICMP6_PARAM_PROB)
3008 			state_icmp++;
3009 		break;
3010 #endif /* INET6 */
3011 	default:
3012 		sport = dport = hdrlen = 0;
3013 		break;
3014 	}
3015 
3016 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3017 
3018 	/* check packet for BINAT/NAT/RDR */
3019 	if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3020 	    &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3021 		KASSERT(sk != NULL, ("%s: null sk", __func__));
3022 		KASSERT(nk != NULL, ("%s: null nk", __func__));
3023 
3024 		if (pd->ip_sum)
3025 			bip_sum = *pd->ip_sum;
3026 
3027 		switch (pd->proto) {
3028 		case IPPROTO_TCP:
3029 			bproto_sum = th->th_sum;
3030 			pd->proto_sum = &th->th_sum;
3031 
3032 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3033 			    nk->port[pd->sidx] != sport) {
3034 				pf_change_ap(saddr, &th->th_sport, pd->ip_sum,
3035 				    &th->th_sum, &nk->addr[pd->sidx],
3036 				    nk->port[pd->sidx], 0, af);
3037 				pd->sport = &th->th_sport;
3038 				sport = th->th_sport;
3039 			}
3040 
3041 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3042 			    nk->port[pd->didx] != dport) {
3043 				pf_change_ap(daddr, &th->th_dport, pd->ip_sum,
3044 				    &th->th_sum, &nk->addr[pd->didx],
3045 				    nk->port[pd->didx], 0, af);
3046 				dport = th->th_dport;
3047 				pd->dport = &th->th_dport;
3048 			}
3049 			rewrite++;
3050 			break;
3051 		case IPPROTO_UDP:
3052 			bproto_sum = pd->hdr.udp->uh_sum;
3053 			pd->proto_sum = &pd->hdr.udp->uh_sum;
3054 
3055 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3056 			    nk->port[pd->sidx] != sport) {
3057 				pf_change_ap(saddr, &pd->hdr.udp->uh_sport,
3058 				    pd->ip_sum, &pd->hdr.udp->uh_sum,
3059 				    &nk->addr[pd->sidx],
3060 				    nk->port[pd->sidx], 1, af);
3061 				sport = pd->hdr.udp->uh_sport;
3062 				pd->sport = &pd->hdr.udp->uh_sport;
3063 			}
3064 
3065 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3066 			    nk->port[pd->didx] != dport) {
3067 				pf_change_ap(daddr, &pd->hdr.udp->uh_dport,
3068 				    pd->ip_sum, &pd->hdr.udp->uh_sum,
3069 				    &nk->addr[pd->didx],
3070 				    nk->port[pd->didx], 1, af);
3071 				dport = pd->hdr.udp->uh_dport;
3072 				pd->dport = &pd->hdr.udp->uh_dport;
3073 			}
3074 			rewrite++;
3075 			break;
3076 #ifdef INET
3077 		case IPPROTO_ICMP:
3078 			nk->port[0] = nk->port[1];
3079 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3080 				pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3081 				    nk->addr[pd->sidx].v4.s_addr, 0);
3082 
3083 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3084 				pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3085 				    nk->addr[pd->didx].v4.s_addr, 0);
3086 
3087 			if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3088 				pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3089 				    pd->hdr.icmp->icmp_cksum, sport,
3090 				    nk->port[1], 0);
3091 				pd->hdr.icmp->icmp_id = nk->port[1];
3092 				pd->sport = &pd->hdr.icmp->icmp_id;
3093 			}
3094 			m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3095 			break;
3096 #endif /* INET */
3097 #ifdef INET6
3098 		case IPPROTO_ICMPV6:
3099 			nk->port[0] = nk->port[1];
3100 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3101 				pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3102 				    &nk->addr[pd->sidx], 0);
3103 
3104 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3105 				pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3106 				    &nk->addr[pd->didx], 0);
3107 			rewrite++;
3108 			break;
3109 #endif /* INET */
3110 		default:
3111 			switch (af) {
3112 #ifdef INET
3113 			case AF_INET:
3114 				if (PF_ANEQ(saddr,
3115 				    &nk->addr[pd->sidx], AF_INET))
3116 					pf_change_a(&saddr->v4.s_addr,
3117 					    pd->ip_sum,
3118 					    nk->addr[pd->sidx].v4.s_addr, 0);
3119 
3120 				if (PF_ANEQ(daddr,
3121 				    &nk->addr[pd->didx], AF_INET))
3122 					pf_change_a(&daddr->v4.s_addr,
3123 					    pd->ip_sum,
3124 					    nk->addr[pd->didx].v4.s_addr, 0);
3125 				break;
3126 #endif /* INET */
3127 #ifdef INET6
3128 			case AF_INET6:
3129 				if (PF_ANEQ(saddr,
3130 				    &nk->addr[pd->sidx], AF_INET6))
3131 					PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3132 
3133 				if (PF_ANEQ(daddr,
3134 				    &nk->addr[pd->didx], AF_INET6))
3135 					PF_ACPY(saddr, &nk->addr[pd->didx], af);
3136 				break;
3137 #endif /* INET */
3138 			}
3139 			break;
3140 		}
3141 		if (nr->natpass)
3142 			r = NULL;
3143 		pd->nat_rule = nr;
3144 	}
3145 
3146 	while (r != NULL) {
3147 		r->evaluations++;
3148 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
3149 			r = r->skip[PF_SKIP_IFP].ptr;
3150 		else if (r->direction && r->direction != direction)
3151 			r = r->skip[PF_SKIP_DIR].ptr;
3152 		else if (r->af && r->af != af)
3153 			r = r->skip[PF_SKIP_AF].ptr;
3154 		else if (r->proto && r->proto != pd->proto)
3155 			r = r->skip[PF_SKIP_PROTO].ptr;
3156 		else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3157 		    r->src.neg, kif, M_GETFIB(m)))
3158 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3159 		/* tcp/udp only. port_op always 0 in other cases */
3160 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
3161 		    r->src.port[0], r->src.port[1], sport))
3162 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
3163 		else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3164 		    r->dst.neg, NULL, M_GETFIB(m)))
3165 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
3166 		/* tcp/udp only. port_op always 0 in other cases */
3167 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3168 		    r->dst.port[0], r->dst.port[1], dport))
3169 			r = r->skip[PF_SKIP_DST_PORT].ptr;
3170 		/* icmp only. type always 0 in other cases */
3171 		else if (r->type && r->type != icmptype + 1)
3172 			r = TAILQ_NEXT(r, entries);
3173 		/* icmp only. type always 0 in other cases */
3174 		else if (r->code && r->code != icmpcode + 1)
3175 			r = TAILQ_NEXT(r, entries);
3176 		else if (r->tos && !(r->tos == pd->tos))
3177 			r = TAILQ_NEXT(r, entries);
3178 		else if (r->rule_flag & PFRULE_FRAGMENT)
3179 			r = TAILQ_NEXT(r, entries);
3180 		else if (pd->proto == IPPROTO_TCP &&
3181 		    (r->flagset & th->th_flags) != r->flags)
3182 			r = TAILQ_NEXT(r, entries);
3183 		/* tcp/udp only. uid.op always 0 in other cases */
3184 		else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3185 		    pf_socket_lookup(direction, pd, m), 1)) &&
3186 		    !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3187 		    pd->lookup.uid))
3188 			r = TAILQ_NEXT(r, entries);
3189 		/* tcp/udp only. gid.op always 0 in other cases */
3190 		else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3191 		    pf_socket_lookup(direction, pd, m), 1)) &&
3192 		    !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3193 		    pd->lookup.gid))
3194 			r = TAILQ_NEXT(r, entries);
3195 		else if (r->prob &&
3196 		    r->prob <= arc4random())
3197 			r = TAILQ_NEXT(r, entries);
3198 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
3199 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
3200 			r = TAILQ_NEXT(r, entries);
3201 		else if (r->os_fingerprint != PF_OSFP_ANY &&
3202 		    (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3203 		    pf_osfp_fingerprint(pd, m, off, th),
3204 		    r->os_fingerprint)))
3205 			r = TAILQ_NEXT(r, entries);
3206 		else {
3207 			if (r->tag)
3208 				tag = r->tag;
3209 			if (r->rtableid >= 0)
3210 				rtableid = r->rtableid;
3211 			if (r->anchor == NULL) {
3212 				match = 1;
3213 				*rm = r;
3214 				*am = a;
3215 				*rsm = ruleset;
3216 				if ((*rm)->quick)
3217 					break;
3218 				r = TAILQ_NEXT(r, entries);
3219 			} else
3220 				pf_step_into_anchor(anchor_stack, &asd,
3221 				    &ruleset, PF_RULESET_FILTER, &r, &a,
3222 				    &match);
3223 		}
3224 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3225 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3226 			break;
3227 	}
3228 	r = *rm;
3229 	a = *am;
3230 	ruleset = *rsm;
3231 
3232 	REASON_SET(&reason, PFRES_MATCH);
3233 
3234 	if (r->log || (nr != NULL && nr->log)) {
3235 		if (rewrite)
3236 			m_copyback(m, off, hdrlen, pd->hdr.any);
3237 		PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3238 		    ruleset, pd, 1);
3239 	}
3240 
3241 	if ((r->action == PF_DROP) &&
3242 	    ((r->rule_flag & PFRULE_RETURNRST) ||
3243 	    (r->rule_flag & PFRULE_RETURNICMP) ||
3244 	    (r->rule_flag & PFRULE_RETURN))) {
3245 		/* undo NAT changes, if they have taken place */
3246 		if (nr != NULL) {
3247 			PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3248 			PF_ACPY(daddr, &sk->addr[pd->didx], af);
3249 			if (pd->sport)
3250 				*pd->sport = sk->port[pd->sidx];
3251 			if (pd->dport)
3252 				*pd->dport = sk->port[pd->didx];
3253 			if (pd->proto_sum)
3254 				*pd->proto_sum = bproto_sum;
3255 			if (pd->ip_sum)
3256 				*pd->ip_sum = bip_sum;
3257 			m_copyback(m, off, hdrlen, pd->hdr.any);
3258 		}
3259 		if (pd->proto == IPPROTO_TCP &&
3260 		    ((r->rule_flag & PFRULE_RETURNRST) ||
3261 		    (r->rule_flag & PFRULE_RETURN)) &&
3262 		    !(th->th_flags & TH_RST)) {
3263 			u_int32_t	 ack = ntohl(th->th_seq) + pd->p_len;
3264 			int		 len = 0;
3265 #ifdef INET
3266 			struct ip	*h4;
3267 #endif
3268 #ifdef INET6
3269 			struct ip6_hdr	*h6;
3270 #endif
3271 
3272 			switch (af) {
3273 #ifdef INET
3274 			case AF_INET:
3275 				h4 = mtod(m, struct ip *);
3276 				len = ntohs(h4->ip_len) - off;
3277 				break;
3278 #endif
3279 #ifdef INET6
3280 			case AF_INET6:
3281 				h6 = mtod(m, struct ip6_hdr *);
3282 				len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3283 				break;
3284 #endif
3285 			}
3286 
3287 			if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3288 				REASON_SET(&reason, PFRES_PROTCKSUM);
3289 			else {
3290 				if (th->th_flags & TH_SYN)
3291 					ack++;
3292 				if (th->th_flags & TH_FIN)
3293 					ack++;
3294 				pf_send_tcp(m, r, af, pd->dst,
3295 				    pd->src, th->th_dport, th->th_sport,
3296 				    ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3297 				    r->return_ttl, 1, 0, kif->pfik_ifp);
3298 			}
3299 		} else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3300 		    r->return_icmp)
3301 			pf_send_icmp(m, r->return_icmp >> 8,
3302 			    r->return_icmp & 255, af, r);
3303 		else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3304 		    r->return_icmp6)
3305 			pf_send_icmp(m, r->return_icmp6 >> 8,
3306 			    r->return_icmp6 & 255, af, r);
3307 	}
3308 
3309 	if (r->action == PF_DROP)
3310 		goto cleanup;
3311 
3312 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3313 		REASON_SET(&reason, PFRES_MEMORY);
3314 		goto cleanup;
3315 	}
3316 	if (rtableid >= 0)
3317 		M_SETFIB(m, rtableid);
3318 
3319 	if (!state_icmp && (r->keep_state || nr != NULL ||
3320 	    (pd->flags & PFDESC_TCP_NORM))) {
3321 		int action;
3322 		action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3323 		    sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3324 		    hdrlen);
3325 		if (action != PF_PASS)
3326 			return (action);
3327 	} else {
3328 		if (sk != NULL)
3329 			uma_zfree(V_pf_state_key_z, sk);
3330 		if (nk != NULL)
3331 			uma_zfree(V_pf_state_key_z, nk);
3332 	}
3333 
3334 	/* copy back packet headers if we performed NAT operations */
3335 	if (rewrite)
3336 		m_copyback(m, off, hdrlen, pd->hdr.any);
3337 
3338 	if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3339 	    direction == PF_OUT &&
3340 	    pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m))
3341 		/*
3342 		 * We want the state created, but we dont
3343 		 * want to send this in case a partner
3344 		 * firewall has to know about it to allow
3345 		 * replies through it.
3346 		 */
3347 		return (PF_DEFER);
3348 
3349 	return (PF_PASS);
3350 
3351 cleanup:
3352 	if (sk != NULL)
3353 		uma_zfree(V_pf_state_key_z, sk);
3354 	if (nk != NULL)
3355 		uma_zfree(V_pf_state_key_z, nk);
3356 	return (PF_DROP);
3357 }
3358 
3359 static int
3360 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3361     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3362     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3363     u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3364     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3365 {
3366 	struct pf_state		*s = NULL;
3367 	struct pf_src_node	*sn = NULL;
3368 	struct tcphdr		*th = pd->hdr.tcp;
3369 	u_int16_t		 mss = V_tcp_mssdflt;
3370 	u_short			 reason;
3371 
3372 	/* check maximums */
3373 	if (r->max_states && (r->states_cur >= r->max_states)) {
3374 		V_pf_status.lcounters[LCNT_STATES]++;
3375 		REASON_SET(&reason, PFRES_MAXSTATES);
3376 		return (PF_DROP);
3377 	}
3378 	/* src node for filter rule */
3379 	if ((r->rule_flag & PFRULE_SRCTRACK ||
3380 	    r->rpool.opts & PF_POOL_STICKYADDR) &&
3381 	    pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3382 		REASON_SET(&reason, PFRES_SRCLIMIT);
3383 		goto csfailed;
3384 	}
3385 	/* src node for translation rule */
3386 	if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3387 	    pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3388 		REASON_SET(&reason, PFRES_SRCLIMIT);
3389 		goto csfailed;
3390 	}
3391 	s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3392 	if (s == NULL) {
3393 		REASON_SET(&reason, PFRES_MEMORY);
3394 		goto csfailed;
3395 	}
3396 	s->rule.ptr = r;
3397 	s->nat_rule.ptr = nr;
3398 	s->anchor.ptr = a;
3399 	STATE_INC_COUNTERS(s);
3400 	if (r->allow_opts)
3401 		s->state_flags |= PFSTATE_ALLOWOPTS;
3402 	if (r->rule_flag & PFRULE_STATESLOPPY)
3403 		s->state_flags |= PFSTATE_SLOPPY;
3404 	s->log = r->log & PF_LOG_ALL;
3405 	s->sync_state = PFSYNC_S_NONE;
3406 	if (nr != NULL)
3407 		s->log |= nr->log & PF_LOG_ALL;
3408 	switch (pd->proto) {
3409 	case IPPROTO_TCP:
3410 		s->src.seqlo = ntohl(th->th_seq);
3411 		s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3412 		if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3413 		    r->keep_state == PF_STATE_MODULATE) {
3414 			/* Generate sequence number modulator */
3415 			if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3416 			    0)
3417 				s->src.seqdiff = 1;
3418 			pf_change_a(&th->th_seq, &th->th_sum,
3419 			    htonl(s->src.seqlo + s->src.seqdiff), 0);
3420 			*rewrite = 1;
3421 		} else
3422 			s->src.seqdiff = 0;
3423 		if (th->th_flags & TH_SYN) {
3424 			s->src.seqhi++;
3425 			s->src.wscale = pf_get_wscale(m, off,
3426 			    th->th_off, pd->af);
3427 		}
3428 		s->src.max_win = MAX(ntohs(th->th_win), 1);
3429 		if (s->src.wscale & PF_WSCALE_MASK) {
3430 			/* Remove scale factor from initial window */
3431 			int win = s->src.max_win;
3432 			win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3433 			s->src.max_win = (win - 1) >>
3434 			    (s->src.wscale & PF_WSCALE_MASK);
3435 		}
3436 		if (th->th_flags & TH_FIN)
3437 			s->src.seqhi++;
3438 		s->dst.seqhi = 1;
3439 		s->dst.max_win = 1;
3440 		s->src.state = TCPS_SYN_SENT;
3441 		s->dst.state = TCPS_CLOSED;
3442 		s->timeout = PFTM_TCP_FIRST_PACKET;
3443 		break;
3444 	case IPPROTO_UDP:
3445 		s->src.state = PFUDPS_SINGLE;
3446 		s->dst.state = PFUDPS_NO_TRAFFIC;
3447 		s->timeout = PFTM_UDP_FIRST_PACKET;
3448 		break;
3449 	case IPPROTO_ICMP:
3450 #ifdef INET6
3451 	case IPPROTO_ICMPV6:
3452 #endif
3453 		s->timeout = PFTM_ICMP_FIRST_PACKET;
3454 		break;
3455 	default:
3456 		s->src.state = PFOTHERS_SINGLE;
3457 		s->dst.state = PFOTHERS_NO_TRAFFIC;
3458 		s->timeout = PFTM_OTHER_FIRST_PACKET;
3459 	}
3460 
3461 	s->creation = time_uptime;
3462 	s->expire = time_uptime;
3463 
3464 	if (sn != NULL) {
3465 		s->src_node = sn;
3466 		s->src_node->states++;
3467 	}
3468 	if (nsn != NULL) {
3469 		/* XXX We only modify one side for now. */
3470 		PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3471 		s->nat_src_node = nsn;
3472 		s->nat_src_node->states++;
3473 	}
3474 	if (pd->proto == IPPROTO_TCP) {
3475 		if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3476 		    off, pd, th, &s->src, &s->dst)) {
3477 			REASON_SET(&reason, PFRES_MEMORY);
3478 			pf_src_tree_remove_state(s);
3479 			STATE_DEC_COUNTERS(s);
3480 			uma_zfree(V_pf_state_z, s);
3481 			return (PF_DROP);
3482 		}
3483 		if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3484 		    pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3485 		    &s->src, &s->dst, rewrite)) {
3486 			/* This really shouldn't happen!!! */
3487 			DPFPRINTF(PF_DEBUG_URGENT,
3488 			    ("pf_normalize_tcp_stateful failed on first pkt"));
3489 			pf_normalize_tcp_cleanup(s);
3490 			pf_src_tree_remove_state(s);
3491 			STATE_DEC_COUNTERS(s);
3492 			uma_zfree(V_pf_state_z, s);
3493 			return (PF_DROP);
3494 		}
3495 	}
3496 	s->direction = pd->dir;
3497 
3498 	/*
3499 	 * sk/nk could already been setup by pf_get_translation().
3500 	 */
3501 	if (nr == NULL) {
3502 		KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3503 		    __func__, nr, sk, nk));
3504 		sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3505 		if (sk == NULL)
3506 			goto csfailed;
3507 		nk = sk;
3508 	} else
3509 		KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3510 		    __func__, nr, sk, nk));
3511 
3512 	/* Swap sk/nk for PF_OUT. */
3513 	if (pf_state_insert(BOUND_IFACE(r, kif),
3514 	    (pd->dir == PF_IN) ? sk : nk,
3515 	    (pd->dir == PF_IN) ? nk : sk, s)) {
3516 		if (pd->proto == IPPROTO_TCP)
3517 			pf_normalize_tcp_cleanup(s);
3518 		REASON_SET(&reason, PFRES_STATEINS);
3519 		pf_src_tree_remove_state(s);
3520 		STATE_DEC_COUNTERS(s);
3521 		uma_zfree(V_pf_state_z, s);
3522 		return (PF_DROP);
3523 	} else
3524 		*sm = s;
3525 
3526 	pf_set_rt_ifp(s, pd->src);	/* needs s->state_key set */
3527 	if (tag > 0)
3528 		s->tag = tag;
3529 	if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3530 	    TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3531 		s->src.state = PF_TCPS_PROXY_SRC;
3532 		/* undo NAT changes, if they have taken place */
3533 		if (nr != NULL) {
3534 			struct pf_state_key *skt = s->key[PF_SK_WIRE];
3535 			if (pd->dir == PF_OUT)
3536 				skt = s->key[PF_SK_STACK];
3537 			PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3538 			PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3539 			if (pd->sport)
3540 				*pd->sport = skt->port[pd->sidx];
3541 			if (pd->dport)
3542 				*pd->dport = skt->port[pd->didx];
3543 			if (pd->proto_sum)
3544 				*pd->proto_sum = bproto_sum;
3545 			if (pd->ip_sum)
3546 				*pd->ip_sum = bip_sum;
3547 			m_copyback(m, off, hdrlen, pd->hdr.any);
3548 		}
3549 		s->src.seqhi = htonl(arc4random());
3550 		/* Find mss option */
3551 		int rtid = M_GETFIB(m);
3552 		mss = pf_get_mss(m, off, th->th_off, pd->af);
3553 		mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3554 		mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3555 		s->src.mss = mss;
3556 		pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3557 		    th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3558 		    TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3559 		REASON_SET(&reason, PFRES_SYNPROXY);
3560 		return (PF_SYNPROXY_DROP);
3561 	}
3562 
3563 	return (PF_PASS);
3564 
3565 csfailed:
3566 	if (sk != NULL)
3567 		uma_zfree(V_pf_state_key_z, sk);
3568 	if (nk != NULL)
3569 		uma_zfree(V_pf_state_key_z, nk);
3570 
3571 	if (sn != NULL && sn->states == 0 && sn->expire == 0)
3572 		pf_remove_src_node(sn);
3573 
3574 	if (nsn != sn && nsn != NULL && nsn->states == 0 && nsn->expire == 0)
3575 		pf_remove_src_node(nsn);
3576 
3577 	return (PF_DROP);
3578 }
3579 
3580 static int
3581 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
3582     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
3583     struct pf_ruleset **rsm)
3584 {
3585 	struct pf_rule		*r, *a = NULL;
3586 	struct pf_ruleset	*ruleset = NULL;
3587 	sa_family_t		 af = pd->af;
3588 	u_short			 reason;
3589 	int			 tag = -1;
3590 	int			 asd = 0;
3591 	int			 match = 0;
3592 	struct pf_anchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
3593 
3594 	PF_RULES_RASSERT();
3595 
3596 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3597 	while (r != NULL) {
3598 		r->evaluations++;
3599 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
3600 			r = r->skip[PF_SKIP_IFP].ptr;
3601 		else if (r->direction && r->direction != direction)
3602 			r = r->skip[PF_SKIP_DIR].ptr;
3603 		else if (r->af && r->af != af)
3604 			r = r->skip[PF_SKIP_AF].ptr;
3605 		else if (r->proto && r->proto != pd->proto)
3606 			r = r->skip[PF_SKIP_PROTO].ptr;
3607 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3608 		    r->src.neg, kif, M_GETFIB(m)))
3609 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3610 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3611 		    r->dst.neg, NULL, M_GETFIB(m)))
3612 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
3613 		else if (r->tos && !(r->tos == pd->tos))
3614 			r = TAILQ_NEXT(r, entries);
3615 		else if (r->os_fingerprint != PF_OSFP_ANY)
3616 			r = TAILQ_NEXT(r, entries);
3617 		else if (pd->proto == IPPROTO_UDP &&
3618 		    (r->src.port_op || r->dst.port_op))
3619 			r = TAILQ_NEXT(r, entries);
3620 		else if (pd->proto == IPPROTO_TCP &&
3621 		    (r->src.port_op || r->dst.port_op || r->flagset))
3622 			r = TAILQ_NEXT(r, entries);
3623 		else if ((pd->proto == IPPROTO_ICMP ||
3624 		    pd->proto == IPPROTO_ICMPV6) &&
3625 		    (r->type || r->code))
3626 			r = TAILQ_NEXT(r, entries);
3627 		else if (r->prob && r->prob <=
3628 		    (arc4random() % (UINT_MAX - 1) + 1))
3629 			r = TAILQ_NEXT(r, entries);
3630 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
3631 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
3632 			r = TAILQ_NEXT(r, entries);
3633 		else {
3634 			if (r->anchor == NULL) {
3635 				match = 1;
3636 				*rm = r;
3637 				*am = a;
3638 				*rsm = ruleset;
3639 				if ((*rm)->quick)
3640 					break;
3641 				r = TAILQ_NEXT(r, entries);
3642 			} else
3643 				pf_step_into_anchor(anchor_stack, &asd,
3644 				    &ruleset, PF_RULESET_FILTER, &r, &a,
3645 				    &match);
3646 		}
3647 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3648 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3649 			break;
3650 	}
3651 	r = *rm;
3652 	a = *am;
3653 	ruleset = *rsm;
3654 
3655 	REASON_SET(&reason, PFRES_MATCH);
3656 
3657 	if (r->log)
3658 		PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
3659 		    1);
3660 
3661 	if (r->action != PF_PASS)
3662 		return (PF_DROP);
3663 
3664 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3665 		REASON_SET(&reason, PFRES_MEMORY);
3666 		return (PF_DROP);
3667 	}
3668 
3669 	return (PF_PASS);
3670 }
3671 
3672 static int
3673 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
3674 	struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
3675 	struct pf_pdesc *pd, u_short *reason, int *copyback)
3676 {
3677 	struct tcphdr		*th = pd->hdr.tcp;
3678 	u_int16_t		 win = ntohs(th->th_win);
3679 	u_int32_t		 ack, end, seq, orig_seq;
3680 	u_int8_t		 sws, dws;
3681 	int			 ackskew;
3682 
3683 	if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
3684 		sws = src->wscale & PF_WSCALE_MASK;
3685 		dws = dst->wscale & PF_WSCALE_MASK;
3686 	} else
3687 		sws = dws = 0;
3688 
3689 	/*
3690 	 * Sequence tracking algorithm from Guido van Rooij's paper:
3691 	 *   http://www.madison-gurkha.com/publications/tcp_filtering/
3692 	 *	tcp_filtering.ps
3693 	 */
3694 
3695 	orig_seq = seq = ntohl(th->th_seq);
3696 	if (src->seqlo == 0) {
3697 		/* First packet from this end. Set its state */
3698 
3699 		if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
3700 		    src->scrub == NULL) {
3701 			if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
3702 				REASON_SET(reason, PFRES_MEMORY);
3703 				return (PF_DROP);
3704 			}
3705 		}
3706 
3707 		/* Deferred generation of sequence number modulator */
3708 		if (dst->seqdiff && !src->seqdiff) {
3709 			/* use random iss for the TCP server */
3710 			while ((src->seqdiff = arc4random() - seq) == 0)
3711 				;
3712 			ack = ntohl(th->th_ack) - dst->seqdiff;
3713 			pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
3714 			    src->seqdiff), 0);
3715 			pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
3716 			*copyback = 1;
3717 		} else {
3718 			ack = ntohl(th->th_ack);
3719 		}
3720 
3721 		end = seq + pd->p_len;
3722 		if (th->th_flags & TH_SYN) {
3723 			end++;
3724 			if (dst->wscale & PF_WSCALE_FLAG) {
3725 				src->wscale = pf_get_wscale(m, off, th->th_off,
3726 				    pd->af);
3727 				if (src->wscale & PF_WSCALE_FLAG) {
3728 					/* Remove scale factor from initial
3729 					 * window */
3730 					sws = src->wscale & PF_WSCALE_MASK;
3731 					win = ((u_int32_t)win + (1 << sws) - 1)
3732 					    >> sws;
3733 					dws = dst->wscale & PF_WSCALE_MASK;
3734 				} else {
3735 					/* fixup other window */
3736 					dst->max_win <<= dst->wscale &
3737 					    PF_WSCALE_MASK;
3738 					/* in case of a retrans SYN|ACK */
3739 					dst->wscale = 0;
3740 				}
3741 			}
3742 		}
3743 		if (th->th_flags & TH_FIN)
3744 			end++;
3745 
3746 		src->seqlo = seq;
3747 		if (src->state < TCPS_SYN_SENT)
3748 			src->state = TCPS_SYN_SENT;
3749 
3750 		/*
3751 		 * May need to slide the window (seqhi may have been set by
3752 		 * the crappy stack check or if we picked up the connection
3753 		 * after establishment)
3754 		 */
3755 		if (src->seqhi == 1 ||
3756 		    SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
3757 			src->seqhi = end + MAX(1, dst->max_win << dws);
3758 		if (win > src->max_win)
3759 			src->max_win = win;
3760 
3761 	} else {
3762 		ack = ntohl(th->th_ack) - dst->seqdiff;
3763 		if (src->seqdiff) {
3764 			/* Modulate sequence numbers */
3765 			pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
3766 			    src->seqdiff), 0);
3767 			pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
3768 			*copyback = 1;
3769 		}
3770 		end = seq + pd->p_len;
3771 		if (th->th_flags & TH_SYN)
3772 			end++;
3773 		if (th->th_flags & TH_FIN)
3774 			end++;
3775 	}
3776 
3777 	if ((th->th_flags & TH_ACK) == 0) {
3778 		/* Let it pass through the ack skew check */
3779 		ack = dst->seqlo;
3780 	} else if ((ack == 0 &&
3781 	    (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
3782 	    /* broken tcp stacks do not set ack */
3783 	    (dst->state < TCPS_SYN_SENT)) {
3784 		/*
3785 		 * Many stacks (ours included) will set the ACK number in an
3786 		 * FIN|ACK if the SYN times out -- no sequence to ACK.
3787 		 */
3788 		ack = dst->seqlo;
3789 	}
3790 
3791 	if (seq == end) {
3792 		/* Ease sequencing restrictions on no data packets */
3793 		seq = src->seqlo;
3794 		end = seq;
3795 	}
3796 
3797 	ackskew = dst->seqlo - ack;
3798 
3799 
3800 	/*
3801 	 * Need to demodulate the sequence numbers in any TCP SACK options
3802 	 * (Selective ACK). We could optionally validate the SACK values
3803 	 * against the current ACK window, either forwards or backwards, but
3804 	 * I'm not confident that SACK has been implemented properly
3805 	 * everywhere. It wouldn't surprise me if several stacks accidently
3806 	 * SACK too far backwards of previously ACKed data. There really aren't
3807 	 * any security implications of bad SACKing unless the target stack
3808 	 * doesn't validate the option length correctly. Someone trying to
3809 	 * spoof into a TCP connection won't bother blindly sending SACK
3810 	 * options anyway.
3811 	 */
3812 	if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
3813 		if (pf_modulate_sack(m, off, pd, th, dst))
3814 			*copyback = 1;
3815 	}
3816 
3817 
3818 #define	MAXACKWINDOW (0xffff + 1500)	/* 1500 is an arbitrary fudge factor */
3819 	if (SEQ_GEQ(src->seqhi, end) &&
3820 	    /* Last octet inside other's window space */
3821 	    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
3822 	    /* Retrans: not more than one window back */
3823 	    (ackskew >= -MAXACKWINDOW) &&
3824 	    /* Acking not more than one reassembled fragment backwards */
3825 	    (ackskew <= (MAXACKWINDOW << sws)) &&
3826 	    /* Acking not more than one window forward */
3827 	    ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
3828 	    (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
3829 	    (pd->flags & PFDESC_IP_REAS) == 0)) {
3830 	    /* Require an exact/+1 sequence match on resets when possible */
3831 
3832 		if (dst->scrub || src->scrub) {
3833 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
3834 			    *state, src, dst, copyback))
3835 				return (PF_DROP);
3836 		}
3837 
3838 		/* update max window */
3839 		if (src->max_win < win)
3840 			src->max_win = win;
3841 		/* synchronize sequencing */
3842 		if (SEQ_GT(end, src->seqlo))
3843 			src->seqlo = end;
3844 		/* slide the window of what the other end can send */
3845 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
3846 			dst->seqhi = ack + MAX((win << sws), 1);
3847 
3848 
3849 		/* update states */
3850 		if (th->th_flags & TH_SYN)
3851 			if (src->state < TCPS_SYN_SENT)
3852 				src->state = TCPS_SYN_SENT;
3853 		if (th->th_flags & TH_FIN)
3854 			if (src->state < TCPS_CLOSING)
3855 				src->state = TCPS_CLOSING;
3856 		if (th->th_flags & TH_ACK) {
3857 			if (dst->state == TCPS_SYN_SENT) {
3858 				dst->state = TCPS_ESTABLISHED;
3859 				if (src->state == TCPS_ESTABLISHED &&
3860 				    (*state)->src_node != NULL &&
3861 				    pf_src_connlimit(state)) {
3862 					REASON_SET(reason, PFRES_SRCLIMIT);
3863 					return (PF_DROP);
3864 				}
3865 			} else if (dst->state == TCPS_CLOSING)
3866 				dst->state = TCPS_FIN_WAIT_2;
3867 		}
3868 		if (th->th_flags & TH_RST)
3869 			src->state = dst->state = TCPS_TIME_WAIT;
3870 
3871 		/* update expire time */
3872 		(*state)->expire = time_uptime;
3873 		if (src->state >= TCPS_FIN_WAIT_2 &&
3874 		    dst->state >= TCPS_FIN_WAIT_2)
3875 			(*state)->timeout = PFTM_TCP_CLOSED;
3876 		else if (src->state >= TCPS_CLOSING &&
3877 		    dst->state >= TCPS_CLOSING)
3878 			(*state)->timeout = PFTM_TCP_FIN_WAIT;
3879 		else if (src->state < TCPS_ESTABLISHED ||
3880 		    dst->state < TCPS_ESTABLISHED)
3881 			(*state)->timeout = PFTM_TCP_OPENING;
3882 		else if (src->state >= TCPS_CLOSING ||
3883 		    dst->state >= TCPS_CLOSING)
3884 			(*state)->timeout = PFTM_TCP_CLOSING;
3885 		else
3886 			(*state)->timeout = PFTM_TCP_ESTABLISHED;
3887 
3888 		/* Fall through to PASS packet */
3889 
3890 	} else if ((dst->state < TCPS_SYN_SENT ||
3891 		dst->state >= TCPS_FIN_WAIT_2 ||
3892 		src->state >= TCPS_FIN_WAIT_2) &&
3893 	    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
3894 	    /* Within a window forward of the originating packet */
3895 	    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
3896 	    /* Within a window backward of the originating packet */
3897 
3898 		/*
3899 		 * This currently handles three situations:
3900 		 *  1) Stupid stacks will shotgun SYNs before their peer
3901 		 *     replies.
3902 		 *  2) When PF catches an already established stream (the
3903 		 *     firewall rebooted, the state table was flushed, routes
3904 		 *     changed...)
3905 		 *  3) Packets get funky immediately after the connection
3906 		 *     closes (this should catch Solaris spurious ACK|FINs
3907 		 *     that web servers like to spew after a close)
3908 		 *
3909 		 * This must be a little more careful than the above code
3910 		 * since packet floods will also be caught here. We don't
3911 		 * update the TTL here to mitigate the damage of a packet
3912 		 * flood and so the same code can handle awkward establishment
3913 		 * and a loosened connection close.
3914 		 * In the establishment case, a correct peer response will
3915 		 * validate the connection, go through the normal state code
3916 		 * and keep updating the state TTL.
3917 		 */
3918 
3919 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
3920 			printf("pf: loose state match: ");
3921 			pf_print_state(*state);
3922 			pf_print_flags(th->th_flags);
3923 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
3924 			    "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
3925 			    pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
3926 			    (unsigned long long)(*state)->packets[1],
3927 			    pd->dir == PF_IN ? "in" : "out",
3928 			    pd->dir == (*state)->direction ? "fwd" : "rev");
3929 		}
3930 
3931 		if (dst->scrub || src->scrub) {
3932 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
3933 			    *state, src, dst, copyback))
3934 				return (PF_DROP);
3935 		}
3936 
3937 		/* update max window */
3938 		if (src->max_win < win)
3939 			src->max_win = win;
3940 		/* synchronize sequencing */
3941 		if (SEQ_GT(end, src->seqlo))
3942 			src->seqlo = end;
3943 		/* slide the window of what the other end can send */
3944 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
3945 			dst->seqhi = ack + MAX((win << sws), 1);
3946 
3947 		/*
3948 		 * Cannot set dst->seqhi here since this could be a shotgunned
3949 		 * SYN and not an already established connection.
3950 		 */
3951 
3952 		if (th->th_flags & TH_FIN)
3953 			if (src->state < TCPS_CLOSING)
3954 				src->state = TCPS_CLOSING;
3955 		if (th->th_flags & TH_RST)
3956 			src->state = dst->state = TCPS_TIME_WAIT;
3957 
3958 		/* Fall through to PASS packet */
3959 
3960 	} else {
3961 		if ((*state)->dst.state == TCPS_SYN_SENT &&
3962 		    (*state)->src.state == TCPS_SYN_SENT) {
3963 			/* Send RST for state mismatches during handshake */
3964 			if (!(th->th_flags & TH_RST))
3965 				pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
3966 				    pd->dst, pd->src, th->th_dport,
3967 				    th->th_sport, ntohl(th->th_ack), 0,
3968 				    TH_RST, 0, 0,
3969 				    (*state)->rule.ptr->return_ttl, 1, 0,
3970 				    kif->pfik_ifp);
3971 			src->seqlo = 0;
3972 			src->seqhi = 1;
3973 			src->max_win = 1;
3974 		} else if (V_pf_status.debug >= PF_DEBUG_MISC) {
3975 			printf("pf: BAD state: ");
3976 			pf_print_state(*state);
3977 			pf_print_flags(th->th_flags);
3978 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
3979 			    "pkts=%llu:%llu dir=%s,%s\n",
3980 			    seq, orig_seq, ack, pd->p_len, ackskew,
3981 			    (unsigned long long)(*state)->packets[0],
3982 			    (unsigned long long)(*state)->packets[1],
3983 			    pd->dir == PF_IN ? "in" : "out",
3984 			    pd->dir == (*state)->direction ? "fwd" : "rev");
3985 			printf("pf: State failure on: %c %c %c %c | %c %c\n",
3986 			    SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
3987 			    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
3988 			    ' ': '2',
3989 			    (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
3990 			    (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
3991 			    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
3992 			    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
3993 		}
3994 		REASON_SET(reason, PFRES_BADSTATE);
3995 		return (PF_DROP);
3996 	}
3997 
3998 	return (PF_PASS);
3999 }
4000 
4001 static int
4002 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4003 	struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4004 {
4005 	struct tcphdr		*th = pd->hdr.tcp;
4006 
4007 	if (th->th_flags & TH_SYN)
4008 		if (src->state < TCPS_SYN_SENT)
4009 			src->state = TCPS_SYN_SENT;
4010 	if (th->th_flags & TH_FIN)
4011 		if (src->state < TCPS_CLOSING)
4012 			src->state = TCPS_CLOSING;
4013 	if (th->th_flags & TH_ACK) {
4014 		if (dst->state == TCPS_SYN_SENT) {
4015 			dst->state = TCPS_ESTABLISHED;
4016 			if (src->state == TCPS_ESTABLISHED &&
4017 			    (*state)->src_node != NULL &&
4018 			    pf_src_connlimit(state)) {
4019 				REASON_SET(reason, PFRES_SRCLIMIT);
4020 				return (PF_DROP);
4021 			}
4022 		} else if (dst->state == TCPS_CLOSING) {
4023 			dst->state = TCPS_FIN_WAIT_2;
4024 		} else if (src->state == TCPS_SYN_SENT &&
4025 		    dst->state < TCPS_SYN_SENT) {
4026 			/*
4027 			 * Handle a special sloppy case where we only see one
4028 			 * half of the connection. If there is a ACK after
4029 			 * the initial SYN without ever seeing a packet from
4030 			 * the destination, set the connection to established.
4031 			 */
4032 			dst->state = src->state = TCPS_ESTABLISHED;
4033 			if ((*state)->src_node != NULL &&
4034 			    pf_src_connlimit(state)) {
4035 				REASON_SET(reason, PFRES_SRCLIMIT);
4036 				return (PF_DROP);
4037 			}
4038 		} else if (src->state == TCPS_CLOSING &&
4039 		    dst->state == TCPS_ESTABLISHED &&
4040 		    dst->seqlo == 0) {
4041 			/*
4042 			 * Handle the closing of half connections where we
4043 			 * don't see the full bidirectional FIN/ACK+ACK
4044 			 * handshake.
4045 			 */
4046 			dst->state = TCPS_CLOSING;
4047 		}
4048 	}
4049 	if (th->th_flags & TH_RST)
4050 		src->state = dst->state = TCPS_TIME_WAIT;
4051 
4052 	/* update expire time */
4053 	(*state)->expire = time_uptime;
4054 	if (src->state >= TCPS_FIN_WAIT_2 &&
4055 	    dst->state >= TCPS_FIN_WAIT_2)
4056 		(*state)->timeout = PFTM_TCP_CLOSED;
4057 	else if (src->state >= TCPS_CLOSING &&
4058 	    dst->state >= TCPS_CLOSING)
4059 		(*state)->timeout = PFTM_TCP_FIN_WAIT;
4060 	else if (src->state < TCPS_ESTABLISHED ||
4061 	    dst->state < TCPS_ESTABLISHED)
4062 		(*state)->timeout = PFTM_TCP_OPENING;
4063 	else if (src->state >= TCPS_CLOSING ||
4064 	    dst->state >= TCPS_CLOSING)
4065 		(*state)->timeout = PFTM_TCP_CLOSING;
4066 	else
4067 		(*state)->timeout = PFTM_TCP_ESTABLISHED;
4068 
4069 	return (PF_PASS);
4070 }
4071 
4072 static int
4073 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4074     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4075     u_short *reason)
4076 {
4077 	struct pf_state_key_cmp	 key;
4078 	struct tcphdr		*th = pd->hdr.tcp;
4079 	int			 copyback = 0;
4080 	struct pf_state_peer	*src, *dst;
4081 	struct pf_state_key	*sk;
4082 
4083 	bzero(&key, sizeof(key));
4084 	key.af = pd->af;
4085 	key.proto = IPPROTO_TCP;
4086 	if (direction == PF_IN)	{	/* wire side, straight */
4087 		PF_ACPY(&key.addr[0], pd->src, key.af);
4088 		PF_ACPY(&key.addr[1], pd->dst, key.af);
4089 		key.port[0] = th->th_sport;
4090 		key.port[1] = th->th_dport;
4091 	} else {			/* stack side, reverse */
4092 		PF_ACPY(&key.addr[1], pd->src, key.af);
4093 		PF_ACPY(&key.addr[0], pd->dst, key.af);
4094 		key.port[1] = th->th_sport;
4095 		key.port[0] = th->th_dport;
4096 	}
4097 
4098 	STATE_LOOKUP(kif, &key, direction, *state, pd);
4099 
4100 	if (direction == (*state)->direction) {
4101 		src = &(*state)->src;
4102 		dst = &(*state)->dst;
4103 	} else {
4104 		src = &(*state)->dst;
4105 		dst = &(*state)->src;
4106 	}
4107 
4108 	sk = (*state)->key[pd->didx];
4109 
4110 	if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4111 		if (direction != (*state)->direction) {
4112 			REASON_SET(reason, PFRES_SYNPROXY);
4113 			return (PF_SYNPROXY_DROP);
4114 		}
4115 		if (th->th_flags & TH_SYN) {
4116 			if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4117 				REASON_SET(reason, PFRES_SYNPROXY);
4118 				return (PF_DROP);
4119 			}
4120 			pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4121 			    pd->src, th->th_dport, th->th_sport,
4122 			    (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4123 			    TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4124 			REASON_SET(reason, PFRES_SYNPROXY);
4125 			return (PF_SYNPROXY_DROP);
4126 		} else if (!(th->th_flags & TH_ACK) ||
4127 		    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4128 		    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4129 			REASON_SET(reason, PFRES_SYNPROXY);
4130 			return (PF_DROP);
4131 		} else if ((*state)->src_node != NULL &&
4132 		    pf_src_connlimit(state)) {
4133 			REASON_SET(reason, PFRES_SRCLIMIT);
4134 			return (PF_DROP);
4135 		} else
4136 			(*state)->src.state = PF_TCPS_PROXY_DST;
4137 	}
4138 	if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4139 		if (direction == (*state)->direction) {
4140 			if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4141 			    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4142 			    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4143 				REASON_SET(reason, PFRES_SYNPROXY);
4144 				return (PF_DROP);
4145 			}
4146 			(*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4147 			if ((*state)->dst.seqhi == 1)
4148 				(*state)->dst.seqhi = htonl(arc4random());
4149 			pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4150 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
4151 			    sk->port[pd->sidx], sk->port[pd->didx],
4152 			    (*state)->dst.seqhi, 0, TH_SYN, 0,
4153 			    (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4154 			REASON_SET(reason, PFRES_SYNPROXY);
4155 			return (PF_SYNPROXY_DROP);
4156 		} else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4157 		    (TH_SYN|TH_ACK)) ||
4158 		    (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4159 			REASON_SET(reason, PFRES_SYNPROXY);
4160 			return (PF_DROP);
4161 		} else {
4162 			(*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4163 			(*state)->dst.seqlo = ntohl(th->th_seq);
4164 			pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4165 			    pd->src, th->th_dport, th->th_sport,
4166 			    ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4167 			    TH_ACK, (*state)->src.max_win, 0, 0, 0,
4168 			    (*state)->tag, NULL);
4169 			pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4170 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
4171 			    sk->port[pd->sidx], sk->port[pd->didx],
4172 			    (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4173 			    TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4174 			(*state)->src.seqdiff = (*state)->dst.seqhi -
4175 			    (*state)->src.seqlo;
4176 			(*state)->dst.seqdiff = (*state)->src.seqhi -
4177 			    (*state)->dst.seqlo;
4178 			(*state)->src.seqhi = (*state)->src.seqlo +
4179 			    (*state)->dst.max_win;
4180 			(*state)->dst.seqhi = (*state)->dst.seqlo +
4181 			    (*state)->src.max_win;
4182 			(*state)->src.wscale = (*state)->dst.wscale = 0;
4183 			(*state)->src.state = (*state)->dst.state =
4184 			    TCPS_ESTABLISHED;
4185 			REASON_SET(reason, PFRES_SYNPROXY);
4186 			return (PF_SYNPROXY_DROP);
4187 		}
4188 	}
4189 
4190 	if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4191 	    dst->state >= TCPS_FIN_WAIT_2 &&
4192 	    src->state >= TCPS_FIN_WAIT_2) {
4193 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
4194 			printf("pf: state reuse ");
4195 			pf_print_state(*state);
4196 			pf_print_flags(th->th_flags);
4197 			printf("\n");
4198 		}
4199 		/* XXX make sure it's the same direction ?? */
4200 		(*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4201 		pf_unlink_state(*state, PF_ENTER_LOCKED);
4202 		*state = NULL;
4203 		return (PF_DROP);
4204 	}
4205 
4206 	if ((*state)->state_flags & PFSTATE_SLOPPY) {
4207 		if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4208 			return (PF_DROP);
4209 	} else {
4210 		if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4211 		    &copyback) == PF_DROP)
4212 			return (PF_DROP);
4213 	}
4214 
4215 	/* translate source/destination address, if necessary */
4216 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4217 		struct pf_state_key *nk = (*state)->key[pd->didx];
4218 
4219 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4220 		    nk->port[pd->sidx] != th->th_sport)
4221 			pf_change_ap(pd->src, &th->th_sport, pd->ip_sum,
4222 			    &th->th_sum, &nk->addr[pd->sidx],
4223 			    nk->port[pd->sidx], 0, pd->af);
4224 
4225 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4226 		    nk->port[pd->didx] != th->th_dport)
4227 			pf_change_ap(pd->dst, &th->th_dport, pd->ip_sum,
4228 			    &th->th_sum, &nk->addr[pd->didx],
4229 			    nk->port[pd->didx], 0, pd->af);
4230 		copyback = 1;
4231 	}
4232 
4233 	/* Copyback sequence modulation or stateful scrub changes if needed */
4234 	if (copyback)
4235 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
4236 
4237 	return (PF_PASS);
4238 }
4239 
4240 static int
4241 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
4242     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4243 {
4244 	struct pf_state_peer	*src, *dst;
4245 	struct pf_state_key_cmp	 key;
4246 	struct udphdr		*uh = pd->hdr.udp;
4247 
4248 	bzero(&key, sizeof(key));
4249 	key.af = pd->af;
4250 	key.proto = IPPROTO_UDP;
4251 	if (direction == PF_IN)	{	/* wire side, straight */
4252 		PF_ACPY(&key.addr[0], pd->src, key.af);
4253 		PF_ACPY(&key.addr[1], pd->dst, key.af);
4254 		key.port[0] = uh->uh_sport;
4255 		key.port[1] = uh->uh_dport;
4256 	} else {			/* stack side, reverse */
4257 		PF_ACPY(&key.addr[1], pd->src, key.af);
4258 		PF_ACPY(&key.addr[0], pd->dst, key.af);
4259 		key.port[1] = uh->uh_sport;
4260 		key.port[0] = uh->uh_dport;
4261 	}
4262 
4263 	STATE_LOOKUP(kif, &key, direction, *state, pd);
4264 
4265 	if (direction == (*state)->direction) {
4266 		src = &(*state)->src;
4267 		dst = &(*state)->dst;
4268 	} else {
4269 		src = &(*state)->dst;
4270 		dst = &(*state)->src;
4271 	}
4272 
4273 	/* update states */
4274 	if (src->state < PFUDPS_SINGLE)
4275 		src->state = PFUDPS_SINGLE;
4276 	if (dst->state == PFUDPS_SINGLE)
4277 		dst->state = PFUDPS_MULTIPLE;
4278 
4279 	/* update expire time */
4280 	(*state)->expire = time_uptime;
4281 	if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4282 		(*state)->timeout = PFTM_UDP_MULTIPLE;
4283 	else
4284 		(*state)->timeout = PFTM_UDP_SINGLE;
4285 
4286 	/* translate source/destination address, if necessary */
4287 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4288 		struct pf_state_key *nk = (*state)->key[pd->didx];
4289 
4290 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4291 		    nk->port[pd->sidx] != uh->uh_sport)
4292 			pf_change_ap(pd->src, &uh->uh_sport, pd->ip_sum,
4293 			    &uh->uh_sum, &nk->addr[pd->sidx],
4294 			    nk->port[pd->sidx], 1, pd->af);
4295 
4296 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4297 		    nk->port[pd->didx] != uh->uh_dport)
4298 			pf_change_ap(pd->dst, &uh->uh_dport, pd->ip_sum,
4299 			    &uh->uh_sum, &nk->addr[pd->didx],
4300 			    nk->port[pd->didx], 1, pd->af);
4301 		m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4302 	}
4303 
4304 	return (PF_PASS);
4305 }
4306 
4307 static int
4308 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
4309     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4310 {
4311 	struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
4312 	u_int16_t	 icmpid = 0, *icmpsum;
4313 	u_int8_t	 icmptype;
4314 	int		 state_icmp = 0;
4315 	struct pf_state_key_cmp key;
4316 
4317 	bzero(&key, sizeof(key));
4318 	switch (pd->proto) {
4319 #ifdef INET
4320 	case IPPROTO_ICMP:
4321 		icmptype = pd->hdr.icmp->icmp_type;
4322 		icmpid = pd->hdr.icmp->icmp_id;
4323 		icmpsum = &pd->hdr.icmp->icmp_cksum;
4324 
4325 		if (icmptype == ICMP_UNREACH ||
4326 		    icmptype == ICMP_SOURCEQUENCH ||
4327 		    icmptype == ICMP_REDIRECT ||
4328 		    icmptype == ICMP_TIMXCEED ||
4329 		    icmptype == ICMP_PARAMPROB)
4330 			state_icmp++;
4331 		break;
4332 #endif /* INET */
4333 #ifdef INET6
4334 	case IPPROTO_ICMPV6:
4335 		icmptype = pd->hdr.icmp6->icmp6_type;
4336 		icmpid = pd->hdr.icmp6->icmp6_id;
4337 		icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4338 
4339 		if (icmptype == ICMP6_DST_UNREACH ||
4340 		    icmptype == ICMP6_PACKET_TOO_BIG ||
4341 		    icmptype == ICMP6_TIME_EXCEEDED ||
4342 		    icmptype == ICMP6_PARAM_PROB)
4343 			state_icmp++;
4344 		break;
4345 #endif /* INET6 */
4346 	}
4347 
4348 	if (!state_icmp) {
4349 
4350 		/*
4351 		 * ICMP query/reply message not related to a TCP/UDP packet.
4352 		 * Search for an ICMP state.
4353 		 */
4354 		key.af = pd->af;
4355 		key.proto = pd->proto;
4356 		key.port[0] = key.port[1] = icmpid;
4357 		if (direction == PF_IN)	{	/* wire side, straight */
4358 			PF_ACPY(&key.addr[0], pd->src, key.af);
4359 			PF_ACPY(&key.addr[1], pd->dst, key.af);
4360 		} else {			/* stack side, reverse */
4361 			PF_ACPY(&key.addr[1], pd->src, key.af);
4362 			PF_ACPY(&key.addr[0], pd->dst, key.af);
4363 		}
4364 
4365 		STATE_LOOKUP(kif, &key, direction, *state, pd);
4366 
4367 		(*state)->expire = time_uptime;
4368 		(*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4369 
4370 		/* translate source/destination address, if necessary */
4371 		if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4372 			struct pf_state_key *nk = (*state)->key[pd->didx];
4373 
4374 			switch (pd->af) {
4375 #ifdef INET
4376 			case AF_INET:
4377 				if (PF_ANEQ(pd->src,
4378 				    &nk->addr[pd->sidx], AF_INET))
4379 					pf_change_a(&saddr->v4.s_addr,
4380 					    pd->ip_sum,
4381 					    nk->addr[pd->sidx].v4.s_addr, 0);
4382 
4383 				if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4384 				    AF_INET))
4385 					pf_change_a(&daddr->v4.s_addr,
4386 					    pd->ip_sum,
4387 					    nk->addr[pd->didx].v4.s_addr, 0);
4388 
4389 				if (nk->port[0] !=
4390 				    pd->hdr.icmp->icmp_id) {
4391 					pd->hdr.icmp->icmp_cksum =
4392 					    pf_cksum_fixup(
4393 					    pd->hdr.icmp->icmp_cksum, icmpid,
4394 					    nk->port[pd->sidx], 0);
4395 					pd->hdr.icmp->icmp_id =
4396 					    nk->port[pd->sidx];
4397 				}
4398 
4399 				m_copyback(m, off, ICMP_MINLEN,
4400 				    (caddr_t )pd->hdr.icmp);
4401 				break;
4402 #endif /* INET */
4403 #ifdef INET6
4404 			case AF_INET6:
4405 				if (PF_ANEQ(pd->src,
4406 				    &nk->addr[pd->sidx], AF_INET6))
4407 					pf_change_a6(saddr,
4408 					    &pd->hdr.icmp6->icmp6_cksum,
4409 					    &nk->addr[pd->sidx], 0);
4410 
4411 				if (PF_ANEQ(pd->dst,
4412 				    &nk->addr[pd->didx], AF_INET6))
4413 					pf_change_a6(daddr,
4414 					    &pd->hdr.icmp6->icmp6_cksum,
4415 					    &nk->addr[pd->didx], 0);
4416 
4417 				m_copyback(m, off, sizeof(struct icmp6_hdr),
4418 				    (caddr_t )pd->hdr.icmp6);
4419 				break;
4420 #endif /* INET6 */
4421 			}
4422 		}
4423 		return (PF_PASS);
4424 
4425 	} else {
4426 		/*
4427 		 * ICMP error message in response to a TCP/UDP packet.
4428 		 * Extract the inner TCP/UDP header and search for that state.
4429 		 */
4430 
4431 		struct pf_pdesc	pd2;
4432 		bzero(&pd2, sizeof pd2);
4433 #ifdef INET
4434 		struct ip	h2;
4435 #endif /* INET */
4436 #ifdef INET6
4437 		struct ip6_hdr	h2_6;
4438 		int		terminal = 0;
4439 #endif /* INET6 */
4440 		int		ipoff2 = 0;
4441 		int		off2 = 0;
4442 
4443 		pd2.af = pd->af;
4444 		/* Payload packet is from the opposite direction. */
4445 		pd2.sidx = (direction == PF_IN) ? 1 : 0;
4446 		pd2.didx = (direction == PF_IN) ? 0 : 1;
4447 		switch (pd->af) {
4448 #ifdef INET
4449 		case AF_INET:
4450 			/* offset of h2 in mbuf chain */
4451 			ipoff2 = off + ICMP_MINLEN;
4452 
4453 			if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4454 			    NULL, reason, pd2.af)) {
4455 				DPFPRINTF(PF_DEBUG_MISC,
4456 				    ("pf: ICMP error message too short "
4457 				    "(ip)\n"));
4458 				return (PF_DROP);
4459 			}
4460 			/*
4461 			 * ICMP error messages don't refer to non-first
4462 			 * fragments
4463 			 */
4464 			if (h2.ip_off & htons(IP_OFFMASK)) {
4465 				REASON_SET(reason, PFRES_FRAG);
4466 				return (PF_DROP);
4467 			}
4468 
4469 			/* offset of protocol header that follows h2 */
4470 			off2 = ipoff2 + (h2.ip_hl << 2);
4471 
4472 			pd2.proto = h2.ip_p;
4473 			pd2.src = (struct pf_addr *)&h2.ip_src;
4474 			pd2.dst = (struct pf_addr *)&h2.ip_dst;
4475 			pd2.ip_sum = &h2.ip_sum;
4476 			break;
4477 #endif /* INET */
4478 #ifdef INET6
4479 		case AF_INET6:
4480 			ipoff2 = off + sizeof(struct icmp6_hdr);
4481 
4482 			if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4483 			    NULL, reason, pd2.af)) {
4484 				DPFPRINTF(PF_DEBUG_MISC,
4485 				    ("pf: ICMP error message too short "
4486 				    "(ip6)\n"));
4487 				return (PF_DROP);
4488 			}
4489 			pd2.proto = h2_6.ip6_nxt;
4490 			pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4491 			pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4492 			pd2.ip_sum = NULL;
4493 			off2 = ipoff2 + sizeof(h2_6);
4494 			do {
4495 				switch (pd2.proto) {
4496 				case IPPROTO_FRAGMENT:
4497 					/*
4498 					 * ICMPv6 error messages for
4499 					 * non-first fragments
4500 					 */
4501 					REASON_SET(reason, PFRES_FRAG);
4502 					return (PF_DROP);
4503 				case IPPROTO_AH:
4504 				case IPPROTO_HOPOPTS:
4505 				case IPPROTO_ROUTING:
4506 				case IPPROTO_DSTOPTS: {
4507 					/* get next header and header length */
4508 					struct ip6_ext opt6;
4509 
4510 					if (!pf_pull_hdr(m, off2, &opt6,
4511 					    sizeof(opt6), NULL, reason,
4512 					    pd2.af)) {
4513 						DPFPRINTF(PF_DEBUG_MISC,
4514 						    ("pf: ICMPv6 short opt\n"));
4515 						return (PF_DROP);
4516 					}
4517 					if (pd2.proto == IPPROTO_AH)
4518 						off2 += (opt6.ip6e_len + 2) * 4;
4519 					else
4520 						off2 += (opt6.ip6e_len + 1) * 8;
4521 					pd2.proto = opt6.ip6e_nxt;
4522 					/* goto the next header */
4523 					break;
4524 				}
4525 				default:
4526 					terminal++;
4527 					break;
4528 				}
4529 			} while (!terminal);
4530 			break;
4531 #endif /* INET6 */
4532 		}
4533 
4534 		switch (pd2.proto) {
4535 		case IPPROTO_TCP: {
4536 			struct tcphdr		 th;
4537 			u_int32_t		 seq;
4538 			struct pf_state_peer	*src, *dst;
4539 			u_int8_t		 dws;
4540 			int			 copyback = 0;
4541 
4542 			/*
4543 			 * Only the first 8 bytes of the TCP header can be
4544 			 * expected. Don't access any TCP header fields after
4545 			 * th_seq, an ackskew test is not possible.
4546 			 */
4547 			if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4548 			    pd2.af)) {
4549 				DPFPRINTF(PF_DEBUG_MISC,
4550 				    ("pf: ICMP error message too short "
4551 				    "(tcp)\n"));
4552 				return (PF_DROP);
4553 			}
4554 
4555 			key.af = pd2.af;
4556 			key.proto = IPPROTO_TCP;
4557 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4558 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4559 			key.port[pd2.sidx] = th.th_sport;
4560 			key.port[pd2.didx] = th.th_dport;
4561 
4562 			STATE_LOOKUP(kif, &key, direction, *state, pd);
4563 
4564 			if (direction == (*state)->direction) {
4565 				src = &(*state)->dst;
4566 				dst = &(*state)->src;
4567 			} else {
4568 				src = &(*state)->src;
4569 				dst = &(*state)->dst;
4570 			}
4571 
4572 			if (src->wscale && dst->wscale)
4573 				dws = dst->wscale & PF_WSCALE_MASK;
4574 			else
4575 				dws = 0;
4576 
4577 			/* Demodulate sequence number */
4578 			seq = ntohl(th.th_seq) - src->seqdiff;
4579 			if (src->seqdiff) {
4580 				pf_change_a(&th.th_seq, icmpsum,
4581 				    htonl(seq), 0);
4582 				copyback = 1;
4583 			}
4584 
4585 			if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4586 			    (!SEQ_GEQ(src->seqhi, seq) ||
4587 			    !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4588 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
4589 					printf("pf: BAD ICMP %d:%d ",
4590 					    icmptype, pd->hdr.icmp->icmp_code);
4591 					pf_print_host(pd->src, 0, pd->af);
4592 					printf(" -> ");
4593 					pf_print_host(pd->dst, 0, pd->af);
4594 					printf(" state: ");
4595 					pf_print_state(*state);
4596 					printf(" seq=%u\n", seq);
4597 				}
4598 				REASON_SET(reason, PFRES_BADSTATE);
4599 				return (PF_DROP);
4600 			} else {
4601 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
4602 					printf("pf: OK ICMP %d:%d ",
4603 					    icmptype, pd->hdr.icmp->icmp_code);
4604 					pf_print_host(pd->src, 0, pd->af);
4605 					printf(" -> ");
4606 					pf_print_host(pd->dst, 0, pd->af);
4607 					printf(" state: ");
4608 					pf_print_state(*state);
4609 					printf(" seq=%u\n", seq);
4610 				}
4611 			}
4612 
4613 			/* translate source/destination address, if necessary */
4614 			if ((*state)->key[PF_SK_WIRE] !=
4615 			    (*state)->key[PF_SK_STACK]) {
4616 				struct pf_state_key *nk =
4617 				    (*state)->key[pd->didx];
4618 
4619 				if (PF_ANEQ(pd2.src,
4620 				    &nk->addr[pd2.sidx], pd2.af) ||
4621 				    nk->port[pd2.sidx] != th.th_sport)
4622 					pf_change_icmp(pd2.src, &th.th_sport,
4623 					    daddr, &nk->addr[pd2.sidx],
4624 					    nk->port[pd2.sidx], NULL,
4625 					    pd2.ip_sum, icmpsum,
4626 					    pd->ip_sum, 0, pd2.af);
4627 
4628 				if (PF_ANEQ(pd2.dst,
4629 				    &nk->addr[pd2.didx], pd2.af) ||
4630 				    nk->port[pd2.didx] != th.th_dport)
4631 					pf_change_icmp(pd2.dst, &th.th_dport,
4632 					    NULL, /* XXX Inbound NAT? */
4633 					    &nk->addr[pd2.didx],
4634 					    nk->port[pd2.didx], NULL,
4635 					    pd2.ip_sum, icmpsum,
4636 					    pd->ip_sum, 0, pd2.af);
4637 				copyback = 1;
4638 			}
4639 
4640 			if (copyback) {
4641 				switch (pd2.af) {
4642 #ifdef INET
4643 				case AF_INET:
4644 					m_copyback(m, off, ICMP_MINLEN,
4645 					    (caddr_t )pd->hdr.icmp);
4646 					m_copyback(m, ipoff2, sizeof(h2),
4647 					    (caddr_t )&h2);
4648 					break;
4649 #endif /* INET */
4650 #ifdef INET6
4651 				case AF_INET6:
4652 					m_copyback(m, off,
4653 					    sizeof(struct icmp6_hdr),
4654 					    (caddr_t )pd->hdr.icmp6);
4655 					m_copyback(m, ipoff2, sizeof(h2_6),
4656 					    (caddr_t )&h2_6);
4657 					break;
4658 #endif /* INET6 */
4659 				}
4660 				m_copyback(m, off2, 8, (caddr_t)&th);
4661 			}
4662 
4663 			return (PF_PASS);
4664 			break;
4665 		}
4666 		case IPPROTO_UDP: {
4667 			struct udphdr		uh;
4668 
4669 			if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
4670 			    NULL, reason, pd2.af)) {
4671 				DPFPRINTF(PF_DEBUG_MISC,
4672 				    ("pf: ICMP error message too short "
4673 				    "(udp)\n"));
4674 				return (PF_DROP);
4675 			}
4676 
4677 			key.af = pd2.af;
4678 			key.proto = IPPROTO_UDP;
4679 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4680 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4681 			key.port[pd2.sidx] = uh.uh_sport;
4682 			key.port[pd2.didx] = uh.uh_dport;
4683 
4684 			STATE_LOOKUP(kif, &key, direction, *state, pd);
4685 
4686 			/* translate source/destination address, if necessary */
4687 			if ((*state)->key[PF_SK_WIRE] !=
4688 			    (*state)->key[PF_SK_STACK]) {
4689 				struct pf_state_key *nk =
4690 				    (*state)->key[pd->didx];
4691 
4692 				if (PF_ANEQ(pd2.src,
4693 				    &nk->addr[pd2.sidx], pd2.af) ||
4694 				    nk->port[pd2.sidx] != uh.uh_sport)
4695 					pf_change_icmp(pd2.src, &uh.uh_sport,
4696 					    daddr, &nk->addr[pd2.sidx],
4697 					    nk->port[pd2.sidx], &uh.uh_sum,
4698 					    pd2.ip_sum, icmpsum,
4699 					    pd->ip_sum, 1, pd2.af);
4700 
4701 				if (PF_ANEQ(pd2.dst,
4702 				    &nk->addr[pd2.didx], pd2.af) ||
4703 				    nk->port[pd2.didx] != uh.uh_dport)
4704 					pf_change_icmp(pd2.dst, &uh.uh_dport,
4705 					    NULL, /* XXX Inbound NAT? */
4706 					    &nk->addr[pd2.didx],
4707 					    nk->port[pd2.didx], &uh.uh_sum,
4708 					    pd2.ip_sum, icmpsum,
4709 					    pd->ip_sum, 1, pd2.af);
4710 
4711 				switch (pd2.af) {
4712 #ifdef INET
4713 				case AF_INET:
4714 					m_copyback(m, off, ICMP_MINLEN,
4715 					    (caddr_t )pd->hdr.icmp);
4716 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4717 					break;
4718 #endif /* INET */
4719 #ifdef INET6
4720 				case AF_INET6:
4721 					m_copyback(m, off,
4722 					    sizeof(struct icmp6_hdr),
4723 					    (caddr_t )pd->hdr.icmp6);
4724 					m_copyback(m, ipoff2, sizeof(h2_6),
4725 					    (caddr_t )&h2_6);
4726 					break;
4727 #endif /* INET6 */
4728 				}
4729 				m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
4730 			}
4731 			return (PF_PASS);
4732 			break;
4733 		}
4734 #ifdef INET
4735 		case IPPROTO_ICMP: {
4736 			struct icmp		iih;
4737 
4738 			if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
4739 			    NULL, reason, pd2.af)) {
4740 				DPFPRINTF(PF_DEBUG_MISC,
4741 				    ("pf: ICMP error message too short i"
4742 				    "(icmp)\n"));
4743 				return (PF_DROP);
4744 			}
4745 
4746 			key.af = pd2.af;
4747 			key.proto = IPPROTO_ICMP;
4748 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4749 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4750 			key.port[0] = key.port[1] = iih.icmp_id;
4751 
4752 			STATE_LOOKUP(kif, &key, direction, *state, pd);
4753 
4754 			/* translate source/destination address, if necessary */
4755 			if ((*state)->key[PF_SK_WIRE] !=
4756 			    (*state)->key[PF_SK_STACK]) {
4757 				struct pf_state_key *nk =
4758 				    (*state)->key[pd->didx];
4759 
4760 				if (PF_ANEQ(pd2.src,
4761 				    &nk->addr[pd2.sidx], pd2.af) ||
4762 				    nk->port[pd2.sidx] != iih.icmp_id)
4763 					pf_change_icmp(pd2.src, &iih.icmp_id,
4764 					    daddr, &nk->addr[pd2.sidx],
4765 					    nk->port[pd2.sidx], NULL,
4766 					    pd2.ip_sum, icmpsum,
4767 					    pd->ip_sum, 0, AF_INET);
4768 
4769 				if (PF_ANEQ(pd2.dst,
4770 				    &nk->addr[pd2.didx], pd2.af) ||
4771 				    nk->port[pd2.didx] != iih.icmp_id)
4772 					pf_change_icmp(pd2.dst, &iih.icmp_id,
4773 					    NULL, /* XXX Inbound NAT? */
4774 					    &nk->addr[pd2.didx],
4775 					    nk->port[pd2.didx], NULL,
4776 					    pd2.ip_sum, icmpsum,
4777 					    pd->ip_sum, 0, AF_INET);
4778 
4779 				m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
4780 				m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4781 				m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
4782 			}
4783 			return (PF_PASS);
4784 			break;
4785 		}
4786 #endif /* INET */
4787 #ifdef INET6
4788 		case IPPROTO_ICMPV6: {
4789 			struct icmp6_hdr	iih;
4790 
4791 			if (!pf_pull_hdr(m, off2, &iih,
4792 			    sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
4793 				DPFPRINTF(PF_DEBUG_MISC,
4794 				    ("pf: ICMP error message too short "
4795 				    "(icmp6)\n"));
4796 				return (PF_DROP);
4797 			}
4798 
4799 			key.af = pd2.af;
4800 			key.proto = IPPROTO_ICMPV6;
4801 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4802 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4803 			key.port[0] = key.port[1] = iih.icmp6_id;
4804 
4805 			STATE_LOOKUP(kif, &key, direction, *state, pd);
4806 
4807 			/* translate source/destination address, if necessary */
4808 			if ((*state)->key[PF_SK_WIRE] !=
4809 			    (*state)->key[PF_SK_STACK]) {
4810 				struct pf_state_key *nk =
4811 				    (*state)->key[pd->didx];
4812 
4813 				if (PF_ANEQ(pd2.src,
4814 				    &nk->addr[pd2.sidx], pd2.af) ||
4815 				    nk->port[pd2.sidx] != iih.icmp6_id)
4816 					pf_change_icmp(pd2.src, &iih.icmp6_id,
4817 					    daddr, &nk->addr[pd2.sidx],
4818 					    nk->port[pd2.sidx], NULL,
4819 					    pd2.ip_sum, icmpsum,
4820 					    pd->ip_sum, 0, AF_INET6);
4821 
4822 				if (PF_ANEQ(pd2.dst,
4823 				    &nk->addr[pd2.didx], pd2.af) ||
4824 				    nk->port[pd2.didx] != iih.icmp6_id)
4825 					pf_change_icmp(pd2.dst, &iih.icmp6_id,
4826 					    NULL, /* XXX Inbound NAT? */
4827 					    &nk->addr[pd2.didx],
4828 					    nk->port[pd2.didx], NULL,
4829 					    pd2.ip_sum, icmpsum,
4830 					    pd->ip_sum, 0, AF_INET6);
4831 
4832 				m_copyback(m, off, sizeof(struct icmp6_hdr),
4833 				    (caddr_t)pd->hdr.icmp6);
4834 				m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
4835 				m_copyback(m, off2, sizeof(struct icmp6_hdr),
4836 				    (caddr_t)&iih);
4837 			}
4838 			return (PF_PASS);
4839 			break;
4840 		}
4841 #endif /* INET6 */
4842 		default: {
4843 			key.af = pd2.af;
4844 			key.proto = pd2.proto;
4845 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4846 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4847 			key.port[0] = key.port[1] = 0;
4848 
4849 			STATE_LOOKUP(kif, &key, direction, *state, pd);
4850 
4851 			/* translate source/destination address, if necessary */
4852 			if ((*state)->key[PF_SK_WIRE] !=
4853 			    (*state)->key[PF_SK_STACK]) {
4854 				struct pf_state_key *nk =
4855 				    (*state)->key[pd->didx];
4856 
4857 				if (PF_ANEQ(pd2.src,
4858 				    &nk->addr[pd2.sidx], pd2.af))
4859 					pf_change_icmp(pd2.src, NULL, daddr,
4860 					    &nk->addr[pd2.sidx], 0, NULL,
4861 					    pd2.ip_sum, icmpsum,
4862 					    pd->ip_sum, 0, pd2.af);
4863 
4864 				if (PF_ANEQ(pd2.dst,
4865 				    &nk->addr[pd2.didx], pd2.af))
4866 					pf_change_icmp(pd2.src, NULL,
4867 					    NULL, /* XXX Inbound NAT? */
4868 					    &nk->addr[pd2.didx], 0, NULL,
4869 					    pd2.ip_sum, icmpsum,
4870 					    pd->ip_sum, 0, pd2.af);
4871 
4872 				switch (pd2.af) {
4873 #ifdef INET
4874 				case AF_INET:
4875 					m_copyback(m, off, ICMP_MINLEN,
4876 					    (caddr_t)pd->hdr.icmp);
4877 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4878 					break;
4879 #endif /* INET */
4880 #ifdef INET6
4881 				case AF_INET6:
4882 					m_copyback(m, off,
4883 					    sizeof(struct icmp6_hdr),
4884 					    (caddr_t )pd->hdr.icmp6);
4885 					m_copyback(m, ipoff2, sizeof(h2_6),
4886 					    (caddr_t )&h2_6);
4887 					break;
4888 #endif /* INET6 */
4889 				}
4890 			}
4891 			return (PF_PASS);
4892 			break;
4893 		}
4894 		}
4895 	}
4896 }
4897 
4898 static int
4899 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
4900     struct mbuf *m, struct pf_pdesc *pd)
4901 {
4902 	struct pf_state_peer	*src, *dst;
4903 	struct pf_state_key_cmp	 key;
4904 
4905 	bzero(&key, sizeof(key));
4906 	key.af = pd->af;
4907 	key.proto = pd->proto;
4908 	if (direction == PF_IN)	{
4909 		PF_ACPY(&key.addr[0], pd->src, key.af);
4910 		PF_ACPY(&key.addr[1], pd->dst, key.af);
4911 		key.port[0] = key.port[1] = 0;
4912 	} else {
4913 		PF_ACPY(&key.addr[1], pd->src, key.af);
4914 		PF_ACPY(&key.addr[0], pd->dst, key.af);
4915 		key.port[1] = key.port[0] = 0;
4916 	}
4917 
4918 	STATE_LOOKUP(kif, &key, direction, *state, pd);
4919 
4920 	if (direction == (*state)->direction) {
4921 		src = &(*state)->src;
4922 		dst = &(*state)->dst;
4923 	} else {
4924 		src = &(*state)->dst;
4925 		dst = &(*state)->src;
4926 	}
4927 
4928 	/* update states */
4929 	if (src->state < PFOTHERS_SINGLE)
4930 		src->state = PFOTHERS_SINGLE;
4931 	if (dst->state == PFOTHERS_SINGLE)
4932 		dst->state = PFOTHERS_MULTIPLE;
4933 
4934 	/* update expire time */
4935 	(*state)->expire = time_uptime;
4936 	if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
4937 		(*state)->timeout = PFTM_OTHER_MULTIPLE;
4938 	else
4939 		(*state)->timeout = PFTM_OTHER_SINGLE;
4940 
4941 	/* translate source/destination address, if necessary */
4942 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4943 		struct pf_state_key *nk = (*state)->key[pd->didx];
4944 
4945 		KASSERT(nk, ("%s: nk is null", __func__));
4946 		KASSERT(pd, ("%s: pd is null", __func__));
4947 		KASSERT(pd->src, ("%s: pd->src is null", __func__));
4948 		KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
4949 		switch (pd->af) {
4950 #ifdef INET
4951 		case AF_INET:
4952 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
4953 				pf_change_a(&pd->src->v4.s_addr,
4954 				    pd->ip_sum,
4955 				    nk->addr[pd->sidx].v4.s_addr,
4956 				    0);
4957 
4958 
4959 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
4960 				pf_change_a(&pd->dst->v4.s_addr,
4961 				    pd->ip_sum,
4962 				    nk->addr[pd->didx].v4.s_addr,
4963 				    0);
4964 
4965 				break;
4966 #endif /* INET */
4967 #ifdef INET6
4968 		case AF_INET6:
4969 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
4970 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
4971 
4972 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
4973 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
4974 #endif /* INET6 */
4975 		}
4976 	}
4977 	return (PF_PASS);
4978 }
4979 
4980 /*
4981  * ipoff and off are measured from the start of the mbuf chain.
4982  * h must be at "ipoff" on the mbuf chain.
4983  */
4984 void *
4985 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
4986     u_short *actionp, u_short *reasonp, sa_family_t af)
4987 {
4988 	switch (af) {
4989 #ifdef INET
4990 	case AF_INET: {
4991 		struct ip	*h = mtod(m, struct ip *);
4992 		u_int16_t	 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
4993 
4994 		if (fragoff) {
4995 			if (fragoff >= len)
4996 				ACTION_SET(actionp, PF_PASS);
4997 			else {
4998 				ACTION_SET(actionp, PF_DROP);
4999 				REASON_SET(reasonp, PFRES_FRAG);
5000 			}
5001 			return (NULL);
5002 		}
5003 		if (m->m_pkthdr.len < off + len ||
5004 		    ntohs(h->ip_len) < off + len) {
5005 			ACTION_SET(actionp, PF_DROP);
5006 			REASON_SET(reasonp, PFRES_SHORT);
5007 			return (NULL);
5008 		}
5009 		break;
5010 	}
5011 #endif /* INET */
5012 #ifdef INET6
5013 	case AF_INET6: {
5014 		struct ip6_hdr	*h = mtod(m, struct ip6_hdr *);
5015 
5016 		if (m->m_pkthdr.len < off + len ||
5017 		    (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5018 		    (unsigned)(off + len)) {
5019 			ACTION_SET(actionp, PF_DROP);
5020 			REASON_SET(reasonp, PFRES_SHORT);
5021 			return (NULL);
5022 		}
5023 		break;
5024 	}
5025 #endif /* INET6 */
5026 	}
5027 	m_copydata(m, off, len, p);
5028 	return (p);
5029 }
5030 
5031 int
5032 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5033     int rtableid)
5034 {
5035 #ifdef RADIX_MPATH
5036 	struct radix_node_head	*rnh;
5037 #endif
5038 	struct sockaddr_in	*dst;
5039 	int			 ret = 1;
5040 	int			 check_mpath;
5041 #ifdef INET6
5042 	struct sockaddr_in6	*dst6;
5043 	struct route_in6	 ro;
5044 #else
5045 	struct route		 ro;
5046 #endif
5047 	struct radix_node	*rn;
5048 	struct rtentry		*rt;
5049 	struct ifnet		*ifp;
5050 
5051 	check_mpath = 0;
5052 #ifdef RADIX_MPATH
5053 	/* XXX: stick to table 0 for now */
5054 	rnh = rt_tables_get_rnh(0, af);
5055 	if (rnh != NULL && rn_mpath_capable(rnh))
5056 		check_mpath = 1;
5057 #endif
5058 	bzero(&ro, sizeof(ro));
5059 	switch (af) {
5060 	case AF_INET:
5061 		dst = satosin(&ro.ro_dst);
5062 		dst->sin_family = AF_INET;
5063 		dst->sin_len = sizeof(*dst);
5064 		dst->sin_addr = addr->v4;
5065 		break;
5066 #ifdef INET6
5067 	case AF_INET6:
5068 		/*
5069 		 * Skip check for addresses with embedded interface scope,
5070 		 * as they would always match anyway.
5071 		 */
5072 		if (IN6_IS_SCOPE_EMBED(&addr->v6))
5073 			goto out;
5074 		dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5075 		dst6->sin6_family = AF_INET6;
5076 		dst6->sin6_len = sizeof(*dst6);
5077 		dst6->sin6_addr = addr->v6;
5078 		break;
5079 #endif /* INET6 */
5080 	default:
5081 		return (0);
5082 	}
5083 
5084 	/* Skip checks for ipsec interfaces */
5085 	if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5086 		goto out;
5087 
5088 	switch (af) {
5089 #ifdef INET6
5090 	case AF_INET6:
5091 		in6_rtalloc_ign(&ro, 0, rtableid);
5092 		break;
5093 #endif
5094 #ifdef INET
5095 	case AF_INET:
5096 		in_rtalloc_ign((struct route *)&ro, 0, rtableid);
5097 		break;
5098 #endif
5099 	default:
5100 		rtalloc_ign((struct route *)&ro, 0);	/* No/default FIB. */
5101 		break;
5102 	}
5103 
5104 	if (ro.ro_rt != NULL) {
5105 		/* No interface given, this is a no-route check */
5106 		if (kif == NULL)
5107 			goto out;
5108 
5109 		if (kif->pfik_ifp == NULL) {
5110 			ret = 0;
5111 			goto out;
5112 		}
5113 
5114 		/* Perform uRPF check if passed input interface */
5115 		ret = 0;
5116 		rn = (struct radix_node *)ro.ro_rt;
5117 		do {
5118 			rt = (struct rtentry *)rn;
5119 			ifp = rt->rt_ifp;
5120 
5121 			if (kif->pfik_ifp == ifp)
5122 				ret = 1;
5123 #ifdef RADIX_MPATH
5124 			rn = rn_mpath_next(rn);
5125 #endif
5126 		} while (check_mpath == 1 && rn != NULL && ret == 0);
5127 	} else
5128 		ret = 0;
5129 out:
5130 	if (ro.ro_rt != NULL)
5131 		RTFREE(ro.ro_rt);
5132 	return (ret);
5133 }
5134 
5135 #ifdef INET
5136 static void
5137 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5138     struct pf_state *s, struct pf_pdesc *pd)
5139 {
5140 	struct mbuf		*m0, *m1;
5141 	struct sockaddr_in	dst;
5142 	struct ip		*ip;
5143 	struct ifnet		*ifp = NULL;
5144 	struct pf_addr		 naddr;
5145 	struct pf_src_node	*sn = NULL;
5146 	int			 error = 0;
5147 	uint16_t		 ip_len, ip_off;
5148 
5149 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5150 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5151 	    __func__));
5152 
5153 	if ((pd->pf_mtag == NULL &&
5154 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5155 	    pd->pf_mtag->routed++ > 3) {
5156 		m0 = *m;
5157 		*m = NULL;
5158 		goto bad_locked;
5159 	}
5160 
5161 	if (r->rt == PF_DUPTO) {
5162 		if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5163 			if (s)
5164 				PF_STATE_UNLOCK(s);
5165 			return;
5166 		}
5167 	} else {
5168 		if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5169 			if (s)
5170 				PF_STATE_UNLOCK(s);
5171 			return;
5172 		}
5173 		m0 = *m;
5174 	}
5175 
5176 	ip = mtod(m0, struct ip *);
5177 
5178 	bzero(&dst, sizeof(dst));
5179 	dst.sin_family = AF_INET;
5180 	dst.sin_len = sizeof(dst);
5181 	dst.sin_addr = ip->ip_dst;
5182 
5183 	if (r->rt == PF_FASTROUTE) {
5184 		struct rtentry *rt;
5185 
5186 		if (s)
5187 			PF_STATE_UNLOCK(s);
5188 		rt = rtalloc1_fib(sintosa(&dst), 0, 0, M_GETFIB(m0));
5189 		if (rt == NULL) {
5190 			RTFREE_LOCKED(rt);
5191 			KMOD_IPSTAT_INC(ips_noroute);
5192 			error = EHOSTUNREACH;
5193 			goto bad;
5194 		}
5195 
5196 		ifp = rt->rt_ifp;
5197 		rt->rt_rmx.rmx_pksent++;
5198 
5199 		if (rt->rt_flags & RTF_GATEWAY)
5200 			bcopy(satosin(rt->rt_gateway), &dst, sizeof(dst));
5201 		RTFREE_LOCKED(rt);
5202 	} else {
5203 		if (TAILQ_EMPTY(&r->rpool.list)) {
5204 			DPFPRINTF(PF_DEBUG_URGENT,
5205 			    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5206 			goto bad_locked;
5207 		}
5208 		if (s == NULL) {
5209 			pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5210 			    &naddr, NULL, &sn);
5211 			if (!PF_AZERO(&naddr, AF_INET))
5212 				dst.sin_addr.s_addr = naddr.v4.s_addr;
5213 			ifp = r->rpool.cur->kif ?
5214 			    r->rpool.cur->kif->pfik_ifp : NULL;
5215 		} else {
5216 			if (!PF_AZERO(&s->rt_addr, AF_INET))
5217 				dst.sin_addr.s_addr =
5218 				    s->rt_addr.v4.s_addr;
5219 			ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5220 			PF_STATE_UNLOCK(s);
5221 		}
5222 	}
5223 	if (ifp == NULL)
5224 		goto bad;
5225 
5226 	if (oifp != ifp) {
5227 		if (pf_test(PF_OUT, ifp, &m0, NULL) != PF_PASS)
5228 			goto bad;
5229 		else if (m0 == NULL)
5230 			goto done;
5231 		if (m0->m_len < sizeof(struct ip)) {
5232 			DPFPRINTF(PF_DEBUG_URGENT,
5233 			    ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5234 			goto bad;
5235 		}
5236 		ip = mtod(m0, struct ip *);
5237 	}
5238 
5239 	if (ifp->if_flags & IFF_LOOPBACK)
5240 		m0->m_flags |= M_SKIP_FIREWALL;
5241 
5242 	ip_len = ntohs(ip->ip_len);
5243 	ip_off = ntohs(ip->ip_off);
5244 
5245 	/* Copied from FreeBSD 10.0-CURRENT ip_output. */
5246 	m0->m_pkthdr.csum_flags |= CSUM_IP;
5247 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5248 		in_delayed_cksum(m0);
5249 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5250 	}
5251 #ifdef SCTP
5252 	if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5253 		sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
5254 		m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5255 	}
5256 #endif
5257 
5258 	/*
5259 	 * If small enough for interface, or the interface will take
5260 	 * care of the fragmentation for us, we can just send directly.
5261 	 */
5262 	if (ip_len <= ifp->if_mtu ||
5263 	    (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 ||
5264 	    ((ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) {
5265 		ip->ip_sum = 0;
5266 		if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5267 			ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5268 			m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5269 		}
5270 		m0->m_flags &= ~(M_PROTOFLAGS);
5271 		error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5272 		goto done;
5273 	}
5274 
5275 	/* Balk when DF bit is set or the interface didn't support TSO. */
5276 	if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5277 		error = EMSGSIZE;
5278 		KMOD_IPSTAT_INC(ips_cantfrag);
5279 		if (r->rt != PF_DUPTO) {
5280 			icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5281 			    ifp->if_mtu);
5282 			goto done;
5283 		} else
5284 			goto bad;
5285 	}
5286 
5287 	error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5288 	if (error)
5289 		goto bad;
5290 
5291 	for (; m0; m0 = m1) {
5292 		m1 = m0->m_nextpkt;
5293 		m0->m_nextpkt = NULL;
5294 		if (error == 0) {
5295 			m0->m_flags &= ~(M_PROTOFLAGS);
5296 			error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5297 		} else
5298 			m_freem(m0);
5299 	}
5300 
5301 	if (error == 0)
5302 		KMOD_IPSTAT_INC(ips_fragmented);
5303 
5304 done:
5305 	if (r->rt != PF_DUPTO)
5306 		*m = NULL;
5307 	return;
5308 
5309 bad_locked:
5310 	if (s)
5311 		PF_STATE_UNLOCK(s);
5312 bad:
5313 	m_freem(m0);
5314 	goto done;
5315 }
5316 #endif /* INET */
5317 
5318 #ifdef INET6
5319 static void
5320 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5321     struct pf_state *s, struct pf_pdesc *pd)
5322 {
5323 	struct mbuf		*m0;
5324 	struct sockaddr_in6	dst;
5325 	struct ip6_hdr		*ip6;
5326 	struct ifnet		*ifp = NULL;
5327 	struct pf_addr		 naddr;
5328 	struct pf_src_node	*sn = NULL;
5329 
5330 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5331 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5332 	    __func__));
5333 
5334 	if ((pd->pf_mtag == NULL &&
5335 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5336 	    pd->pf_mtag->routed++ > 3) {
5337 		m0 = *m;
5338 		*m = NULL;
5339 		goto bad_locked;
5340 	}
5341 
5342 	if (r->rt == PF_DUPTO) {
5343 		if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5344 			if (s)
5345 				PF_STATE_UNLOCK(s);
5346 			return;
5347 		}
5348 	} else {
5349 		if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5350 			if (s)
5351 				PF_STATE_UNLOCK(s);
5352 			return;
5353 		}
5354 		m0 = *m;
5355 	}
5356 
5357 	ip6 = mtod(m0, struct ip6_hdr *);
5358 
5359 	bzero(&dst, sizeof(dst));
5360 	dst.sin6_family = AF_INET6;
5361 	dst.sin6_len = sizeof(dst);
5362 	dst.sin6_addr = ip6->ip6_dst;
5363 
5364 	/* Cheat. XXX why only in the v6 case??? */
5365 	if (r->rt == PF_FASTROUTE) {
5366 		if (s)
5367 			PF_STATE_UNLOCK(s);
5368 		m0->m_flags |= M_SKIP_FIREWALL;
5369 		ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL);
5370 		return;
5371 	}
5372 
5373 	if (TAILQ_EMPTY(&r->rpool.list)) {
5374 		DPFPRINTF(PF_DEBUG_URGENT,
5375 		    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5376 		goto bad_locked;
5377 	}
5378 	if (s == NULL) {
5379 		pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5380 		    &naddr, NULL, &sn);
5381 		if (!PF_AZERO(&naddr, AF_INET6))
5382 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5383 			    &naddr, AF_INET6);
5384 		ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5385 	} else {
5386 		if (!PF_AZERO(&s->rt_addr, AF_INET6))
5387 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5388 			    &s->rt_addr, AF_INET6);
5389 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5390 	}
5391 
5392 	if (s)
5393 		PF_STATE_UNLOCK(s);
5394 
5395 	if (ifp == NULL)
5396 		goto bad;
5397 
5398 	if (oifp != ifp) {
5399 		if (pf_test6(PF_OUT, ifp, &m0, NULL) != PF_PASS)
5400 			goto bad;
5401 		else if (m0 == NULL)
5402 			goto done;
5403 		if (m0->m_len < sizeof(struct ip6_hdr)) {
5404 			DPFPRINTF(PF_DEBUG_URGENT,
5405 			    ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5406 			    __func__));
5407 			goto bad;
5408 		}
5409 		ip6 = mtod(m0, struct ip6_hdr *);
5410 	}
5411 
5412 	if (ifp->if_flags & IFF_LOOPBACK)
5413 		m0->m_flags |= M_SKIP_FIREWALL;
5414 
5415 	/*
5416 	 * If the packet is too large for the outgoing interface,
5417 	 * send back an icmp6 error.
5418 	 */
5419 	if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5420 		dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5421 	if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5422 		nd6_output(ifp, ifp, m0, &dst, NULL);
5423 	else {
5424 		in6_ifstat_inc(ifp, ifs6_in_toobig);
5425 		if (r->rt != PF_DUPTO)
5426 			icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5427 		else
5428 			goto bad;
5429 	}
5430 
5431 done:
5432 	if (r->rt != PF_DUPTO)
5433 		*m = NULL;
5434 	return;
5435 
5436 bad_locked:
5437 	if (s)
5438 		PF_STATE_UNLOCK(s);
5439 bad:
5440 	m_freem(m0);
5441 	goto done;
5442 }
5443 #endif /* INET6 */
5444 
5445 /*
5446  * FreeBSD supports cksum offloads for the following drivers.
5447  *  em(4), fxp(4), ixgb(4), lge(4), ndis(4), nge(4), re(4),
5448  *   ti(4), txp(4), xl(4)
5449  *
5450  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5451  *  network driver performed cksum including pseudo header, need to verify
5452  *   csum_data
5453  * CSUM_DATA_VALID :
5454  *  network driver performed cksum, needs to additional pseudo header
5455  *  cksum computation with partial csum_data(i.e. lack of H/W support for
5456  *  pseudo header, for instance hme(4), sk(4) and possibly gem(4))
5457  *
5458  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5459  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5460  * TCP/UDP layer.
5461  * Also, set csum_data to 0xffff to force cksum validation.
5462  */
5463 static int
5464 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5465 {
5466 	u_int16_t sum = 0;
5467 	int hw_assist = 0;
5468 	struct ip *ip;
5469 
5470 	if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5471 		return (1);
5472 	if (m->m_pkthdr.len < off + len)
5473 		return (1);
5474 
5475 	switch (p) {
5476 	case IPPROTO_TCP:
5477 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5478 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5479 				sum = m->m_pkthdr.csum_data;
5480 			} else {
5481 				ip = mtod(m, struct ip *);
5482 				sum = in_pseudo(ip->ip_src.s_addr,
5483 				ip->ip_dst.s_addr, htonl((u_short)len +
5484 				m->m_pkthdr.csum_data + IPPROTO_TCP));
5485 			}
5486 			sum ^= 0xffff;
5487 			++hw_assist;
5488 		}
5489 		break;
5490 	case IPPROTO_UDP:
5491 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5492 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5493 				sum = m->m_pkthdr.csum_data;
5494 			} else {
5495 				ip = mtod(m, struct ip *);
5496 				sum = in_pseudo(ip->ip_src.s_addr,
5497 				ip->ip_dst.s_addr, htonl((u_short)len +
5498 				m->m_pkthdr.csum_data + IPPROTO_UDP));
5499 			}
5500 			sum ^= 0xffff;
5501 			++hw_assist;
5502 		}
5503 		break;
5504 	case IPPROTO_ICMP:
5505 #ifdef INET6
5506 	case IPPROTO_ICMPV6:
5507 #endif /* INET6 */
5508 		break;
5509 	default:
5510 		return (1);
5511 	}
5512 
5513 	if (!hw_assist) {
5514 		switch (af) {
5515 		case AF_INET:
5516 			if (p == IPPROTO_ICMP) {
5517 				if (m->m_len < off)
5518 					return (1);
5519 				m->m_data += off;
5520 				m->m_len -= off;
5521 				sum = in_cksum(m, len);
5522 				m->m_data -= off;
5523 				m->m_len += off;
5524 			} else {
5525 				if (m->m_len < sizeof(struct ip))
5526 					return (1);
5527 				sum = in4_cksum(m, p, off, len);
5528 			}
5529 			break;
5530 #ifdef INET6
5531 		case AF_INET6:
5532 			if (m->m_len < sizeof(struct ip6_hdr))
5533 				return (1);
5534 			sum = in6_cksum(m, p, off, len);
5535 			break;
5536 #endif /* INET6 */
5537 		default:
5538 			return (1);
5539 		}
5540 	}
5541 	if (sum) {
5542 		switch (p) {
5543 		case IPPROTO_TCP:
5544 		    {
5545 			KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5546 			break;
5547 		    }
5548 		case IPPROTO_UDP:
5549 		    {
5550 			KMOD_UDPSTAT_INC(udps_badsum);
5551 			break;
5552 		    }
5553 #ifdef INET
5554 		case IPPROTO_ICMP:
5555 		    {
5556 			KMOD_ICMPSTAT_INC(icps_checksum);
5557 			break;
5558 		    }
5559 #endif
5560 #ifdef INET6
5561 		case IPPROTO_ICMPV6:
5562 		    {
5563 			KMOD_ICMP6STAT_INC(icp6s_checksum);
5564 			break;
5565 		    }
5566 #endif /* INET6 */
5567 		}
5568 		return (1);
5569 	} else {
5570 		if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5571 			m->m_pkthdr.csum_flags |=
5572 			    (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5573 			m->m_pkthdr.csum_data = 0xffff;
5574 		}
5575 	}
5576 	return (0);
5577 }
5578 
5579 
5580 #ifdef INET
5581 int
5582 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5583 {
5584 	struct pfi_kif		*kif;
5585 	u_short			 action, reason = 0, log = 0;
5586 	struct mbuf		*m = *m0;
5587 	struct ip		*h = NULL;
5588 	struct m_tag		*ipfwtag;
5589 	struct pf_rule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5590 	struct pf_state		*s = NULL;
5591 	struct pf_ruleset	*ruleset = NULL;
5592 	struct pf_pdesc		 pd;
5593 	int			 off, dirndx, pqid = 0;
5594 
5595 	M_ASSERTPKTHDR(m);
5596 
5597 	if (!V_pf_status.running)
5598 		return (PF_PASS);
5599 
5600 	memset(&pd, 0, sizeof(pd));
5601 
5602 	kif = (struct pfi_kif *)ifp->if_pf_kif;
5603 
5604 	if (kif == NULL) {
5605 		DPFPRINTF(PF_DEBUG_URGENT,
5606 		    ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5607 		return (PF_DROP);
5608 	}
5609 	if (kif->pfik_flags & PFI_IFLAG_SKIP)
5610 		return (PF_PASS);
5611 
5612 	if (m->m_flags & M_SKIP_FIREWALL)
5613 		return (PF_PASS);
5614 
5615 	pd.pf_mtag = pf_find_mtag(m);
5616 
5617 	PF_RULES_RLOCK();
5618 
5619 	if (ip_divert_ptr != NULL &&
5620 	    ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5621 		struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5622 		if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5623 			if (pd.pf_mtag == NULL &&
5624 			    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5625 				action = PF_DROP;
5626 				goto done;
5627 			}
5628 			pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5629 			m_tag_delete(m, ipfwtag);
5630 		}
5631 		if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5632 			m->m_flags |= M_FASTFWD_OURS;
5633 			pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5634 		}
5635 	} else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5636 		/* We do IP header normalization and packet reassembly here */
5637 		action = PF_DROP;
5638 		goto done;
5639 	}
5640 	m = *m0;	/* pf_normalize messes with m0 */
5641 	h = mtod(m, struct ip *);
5642 
5643 	off = h->ip_hl << 2;
5644 	if (off < (int)sizeof(struct ip)) {
5645 		action = PF_DROP;
5646 		REASON_SET(&reason, PFRES_SHORT);
5647 		log = 1;
5648 		goto done;
5649 	}
5650 
5651 	pd.src = (struct pf_addr *)&h->ip_src;
5652 	pd.dst = (struct pf_addr *)&h->ip_dst;
5653 	pd.sport = pd.dport = NULL;
5654 	pd.ip_sum = &h->ip_sum;
5655 	pd.proto_sum = NULL;
5656 	pd.proto = h->ip_p;
5657 	pd.dir = dir;
5658 	pd.sidx = (dir == PF_IN) ? 0 : 1;
5659 	pd.didx = (dir == PF_IN) ? 1 : 0;
5660 	pd.af = AF_INET;
5661 	pd.tos = h->ip_tos;
5662 	pd.tot_len = ntohs(h->ip_len);
5663 
5664 	/* handle fragments that didn't get reassembled by normalization */
5665 	if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
5666 		action = pf_test_fragment(&r, dir, kif, m, h,
5667 		    &pd, &a, &ruleset);
5668 		goto done;
5669 	}
5670 
5671 	switch (h->ip_p) {
5672 
5673 	case IPPROTO_TCP: {
5674 		struct tcphdr	th;
5675 
5676 		pd.hdr.tcp = &th;
5677 		if (!pf_pull_hdr(m, off, &th, sizeof(th),
5678 		    &action, &reason, AF_INET)) {
5679 			log = action != PF_PASS;
5680 			goto done;
5681 		}
5682 		pd.p_len = pd.tot_len - off - (th.th_off << 2);
5683 		if ((th.th_flags & TH_ACK) && pd.p_len == 0)
5684 			pqid = 1;
5685 		action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
5686 		if (action == PF_DROP)
5687 			goto done;
5688 		action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
5689 		    &reason);
5690 		if (action == PF_PASS) {
5691 			if (pfsync_update_state_ptr != NULL)
5692 				pfsync_update_state_ptr(s);
5693 			r = s->rule.ptr;
5694 			a = s->anchor.ptr;
5695 			log = s->log;
5696 		} else if (s == NULL)
5697 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5698 			    &a, &ruleset, inp);
5699 		break;
5700 	}
5701 
5702 	case IPPROTO_UDP: {
5703 		struct udphdr	uh;
5704 
5705 		pd.hdr.udp = &uh;
5706 		if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
5707 		    &action, &reason, AF_INET)) {
5708 			log = action != PF_PASS;
5709 			goto done;
5710 		}
5711 		if (uh.uh_dport == 0 ||
5712 		    ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
5713 		    ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
5714 			action = PF_DROP;
5715 			REASON_SET(&reason, PFRES_SHORT);
5716 			goto done;
5717 		}
5718 		action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
5719 		if (action == PF_PASS) {
5720 			if (pfsync_update_state_ptr != NULL)
5721 				pfsync_update_state_ptr(s);
5722 			r = s->rule.ptr;
5723 			a = s->anchor.ptr;
5724 			log = s->log;
5725 		} else if (s == NULL)
5726 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5727 			    &a, &ruleset, inp);
5728 		break;
5729 	}
5730 
5731 	case IPPROTO_ICMP: {
5732 		struct icmp	ih;
5733 
5734 		pd.hdr.icmp = &ih;
5735 		if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
5736 		    &action, &reason, AF_INET)) {
5737 			log = action != PF_PASS;
5738 			goto done;
5739 		}
5740 		action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
5741 		    &reason);
5742 		if (action == PF_PASS) {
5743 			if (pfsync_update_state_ptr != NULL)
5744 				pfsync_update_state_ptr(s);
5745 			r = s->rule.ptr;
5746 			a = s->anchor.ptr;
5747 			log = s->log;
5748 		} else if (s == NULL)
5749 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5750 			    &a, &ruleset, inp);
5751 		break;
5752 	}
5753 
5754 #ifdef INET6
5755 	case IPPROTO_ICMPV6: {
5756 		action = PF_DROP;
5757 		DPFPRINTF(PF_DEBUG_MISC,
5758 		    ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
5759 		goto done;
5760 	}
5761 #endif
5762 
5763 	default:
5764 		action = pf_test_state_other(&s, dir, kif, m, &pd);
5765 		if (action == PF_PASS) {
5766 			if (pfsync_update_state_ptr != NULL)
5767 				pfsync_update_state_ptr(s);
5768 			r = s->rule.ptr;
5769 			a = s->anchor.ptr;
5770 			log = s->log;
5771 		} else if (s == NULL)
5772 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5773 			    &a, &ruleset, inp);
5774 		break;
5775 	}
5776 
5777 done:
5778 	PF_RULES_RUNLOCK();
5779 	if (action == PF_PASS && h->ip_hl > 5 &&
5780 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
5781 		action = PF_DROP;
5782 		REASON_SET(&reason, PFRES_IPOPTIONS);
5783 		log = 1;
5784 		DPFPRINTF(PF_DEBUG_MISC,
5785 		    ("pf: dropping packet with ip options\n"));
5786 	}
5787 
5788 	if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
5789 		action = PF_DROP;
5790 		REASON_SET(&reason, PFRES_MEMORY);
5791 	}
5792 	if (r->rtableid >= 0)
5793 		M_SETFIB(m, r->rtableid);
5794 
5795 #ifdef ALTQ
5796 	if (action == PF_PASS && r->qid) {
5797 		if (pd.pf_mtag == NULL &&
5798 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5799 			action = PF_DROP;
5800 			REASON_SET(&reason, PFRES_MEMORY);
5801 		}
5802 		if (pqid || (pd.tos & IPTOS_LOWDELAY))
5803 			pd.pf_mtag->qid = r->pqid;
5804 		else
5805 			pd.pf_mtag->qid = r->qid;
5806 		/* add hints for ecn */
5807 		pd.pf_mtag->hdr = h;
5808 
5809 	}
5810 #endif /* ALTQ */
5811 
5812 	/*
5813 	 * connections redirected to loopback should not match sockets
5814 	 * bound specifically to loopback due to security implications,
5815 	 * see tcp_input() and in_pcblookup_listen().
5816 	 */
5817 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
5818 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
5819 	    (s->nat_rule.ptr->action == PF_RDR ||
5820 	    s->nat_rule.ptr->action == PF_BINAT) &&
5821 	    (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
5822 		m->m_flags |= M_SKIP_FIREWALL;
5823 
5824 	if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
5825 	    !PACKET_LOOPED(&pd)) {
5826 
5827 		ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
5828 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
5829 		if (ipfwtag != NULL) {
5830 			((struct ipfw_rule_ref *)(ipfwtag+1))->info =
5831 			    ntohs(r->divert.port);
5832 			((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
5833 
5834 			if (s)
5835 				PF_STATE_UNLOCK(s);
5836 
5837 			m_tag_prepend(m, ipfwtag);
5838 			if (m->m_flags & M_FASTFWD_OURS) {
5839 				if (pd.pf_mtag == NULL &&
5840 				    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5841 					action = PF_DROP;
5842 					REASON_SET(&reason, PFRES_MEMORY);
5843 					log = 1;
5844 					DPFPRINTF(PF_DEBUG_MISC,
5845 					    ("pf: failed to allocate tag\n"));
5846 				}
5847 				pd.pf_mtag->flags |= PF_FASTFWD_OURS_PRESENT;
5848 				m->m_flags &= ~M_FASTFWD_OURS;
5849 			}
5850 			ip_divert_ptr(*m0, dir ==  PF_IN ? DIR_IN : DIR_OUT);
5851 			*m0 = NULL;
5852 
5853 			return (action);
5854 		} else {
5855 			/* XXX: ipfw has the same behaviour! */
5856 			action = PF_DROP;
5857 			REASON_SET(&reason, PFRES_MEMORY);
5858 			log = 1;
5859 			DPFPRINTF(PF_DEBUG_MISC,
5860 			    ("pf: failed to allocate divert tag\n"));
5861 		}
5862 	}
5863 
5864 	if (log) {
5865 		struct pf_rule *lr;
5866 
5867 		if (s != NULL && s->nat_rule.ptr != NULL &&
5868 		    s->nat_rule.ptr->log & PF_LOG_ALL)
5869 			lr = s->nat_rule.ptr;
5870 		else
5871 			lr = r;
5872 		PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
5873 		    (s == NULL));
5874 	}
5875 
5876 	kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
5877 	kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
5878 
5879 	if (action == PF_PASS || r->action == PF_DROP) {
5880 		dirndx = (dir == PF_OUT);
5881 		r->packets[dirndx]++;
5882 		r->bytes[dirndx] += pd.tot_len;
5883 		if (a != NULL) {
5884 			a->packets[dirndx]++;
5885 			a->bytes[dirndx] += pd.tot_len;
5886 		}
5887 		if (s != NULL) {
5888 			if (s->nat_rule.ptr != NULL) {
5889 				s->nat_rule.ptr->packets[dirndx]++;
5890 				s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
5891 			}
5892 			if (s->src_node != NULL) {
5893 				s->src_node->packets[dirndx]++;
5894 				s->src_node->bytes[dirndx] += pd.tot_len;
5895 			}
5896 			if (s->nat_src_node != NULL) {
5897 				s->nat_src_node->packets[dirndx]++;
5898 				s->nat_src_node->bytes[dirndx] += pd.tot_len;
5899 			}
5900 			dirndx = (dir == s->direction) ? 0 : 1;
5901 			s->packets[dirndx]++;
5902 			s->bytes[dirndx] += pd.tot_len;
5903 		}
5904 		tr = r;
5905 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
5906 		if (nr != NULL && r == &V_pf_default_rule)
5907 			tr = nr;
5908 		if (tr->src.addr.type == PF_ADDR_TABLE)
5909 			pfr_update_stats(tr->src.addr.p.tbl,
5910 			    (s == NULL) ? pd.src :
5911 			    &s->key[(s->direction == PF_IN)]->
5912 				addr[(s->direction == PF_OUT)],
5913 			    pd.af, pd.tot_len, dir == PF_OUT,
5914 			    r->action == PF_PASS, tr->src.neg);
5915 		if (tr->dst.addr.type == PF_ADDR_TABLE)
5916 			pfr_update_stats(tr->dst.addr.p.tbl,
5917 			    (s == NULL) ? pd.dst :
5918 			    &s->key[(s->direction == PF_IN)]->
5919 				addr[(s->direction == PF_IN)],
5920 			    pd.af, pd.tot_len, dir == PF_OUT,
5921 			    r->action == PF_PASS, tr->dst.neg);
5922 	}
5923 
5924 	switch (action) {
5925 	case PF_SYNPROXY_DROP:
5926 		m_freem(*m0);
5927 	case PF_DEFER:
5928 		*m0 = NULL;
5929 		action = PF_PASS;
5930 		break;
5931 	default:
5932 		/* pf_route() returns unlocked. */
5933 		if (r->rt) {
5934 			pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
5935 			return (action);
5936 		}
5937 		break;
5938 	}
5939 	if (s)
5940 		PF_STATE_UNLOCK(s);
5941 
5942 	return (action);
5943 }
5944 #endif /* INET */
5945 
5946 #ifdef INET6
5947 int
5948 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5949 {
5950 	struct pfi_kif		*kif;
5951 	u_short			 action, reason = 0, log = 0;
5952 	struct mbuf		*m = *m0, *n = NULL;
5953 	struct ip6_hdr		*h = NULL;
5954 	struct pf_rule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5955 	struct pf_state		*s = NULL;
5956 	struct pf_ruleset	*ruleset = NULL;
5957 	struct pf_pdesc		 pd;
5958 	int			 off, terminal = 0, dirndx, rh_cnt = 0;
5959 
5960 	M_ASSERTPKTHDR(m);
5961 
5962 	if (!V_pf_status.running)
5963 		return (PF_PASS);
5964 
5965 	memset(&pd, 0, sizeof(pd));
5966 	pd.pf_mtag = pf_find_mtag(m);
5967 
5968 	if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
5969 		return (PF_PASS);
5970 
5971 	kif = (struct pfi_kif *)ifp->if_pf_kif;
5972 	if (kif == NULL) {
5973 		DPFPRINTF(PF_DEBUG_URGENT,
5974 		    ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
5975 		return (PF_DROP);
5976 	}
5977 	if (kif->pfik_flags & PFI_IFLAG_SKIP)
5978 		return (PF_PASS);
5979 
5980 	PF_RULES_RLOCK();
5981 
5982 	/* We do IP header normalization and packet reassembly here */
5983 	if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
5984 		action = PF_DROP;
5985 		goto done;
5986 	}
5987 	m = *m0;	/* pf_normalize messes with m0 */
5988 	h = mtod(m, struct ip6_hdr *);
5989 
5990 #if 1
5991 	/*
5992 	 * we do not support jumbogram yet.  if we keep going, zero ip6_plen
5993 	 * will do something bad, so drop the packet for now.
5994 	 */
5995 	if (htons(h->ip6_plen) == 0) {
5996 		action = PF_DROP;
5997 		REASON_SET(&reason, PFRES_NORM);	/*XXX*/
5998 		goto done;
5999 	}
6000 #endif
6001 
6002 	pd.src = (struct pf_addr *)&h->ip6_src;
6003 	pd.dst = (struct pf_addr *)&h->ip6_dst;
6004 	pd.sport = pd.dport = NULL;
6005 	pd.ip_sum = NULL;
6006 	pd.proto_sum = NULL;
6007 	pd.dir = dir;
6008 	pd.sidx = (dir == PF_IN) ? 0 : 1;
6009 	pd.didx = (dir == PF_IN) ? 1 : 0;
6010 	pd.af = AF_INET6;
6011 	pd.tos = 0;
6012 	pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6013 
6014 	off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6015 	pd.proto = h->ip6_nxt;
6016 	do {
6017 		switch (pd.proto) {
6018 		case IPPROTO_FRAGMENT:
6019 			action = pf_test_fragment(&r, dir, kif, m, h,
6020 			    &pd, &a, &ruleset);
6021 			if (action == PF_DROP)
6022 				REASON_SET(&reason, PFRES_FRAG);
6023 			goto done;
6024 		case IPPROTO_ROUTING: {
6025 			struct ip6_rthdr rthdr;
6026 
6027 			if (rh_cnt++) {
6028 				DPFPRINTF(PF_DEBUG_MISC,
6029 				    ("pf: IPv6 more than one rthdr\n"));
6030 				action = PF_DROP;
6031 				REASON_SET(&reason, PFRES_IPOPTIONS);
6032 				log = 1;
6033 				goto done;
6034 			}
6035 			if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6036 			    &reason, pd.af)) {
6037 				DPFPRINTF(PF_DEBUG_MISC,
6038 				    ("pf: IPv6 short rthdr\n"));
6039 				action = PF_DROP;
6040 				REASON_SET(&reason, PFRES_SHORT);
6041 				log = 1;
6042 				goto done;
6043 			}
6044 			if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6045 				DPFPRINTF(PF_DEBUG_MISC,
6046 				    ("pf: IPv6 rthdr0\n"));
6047 				action = PF_DROP;
6048 				REASON_SET(&reason, PFRES_IPOPTIONS);
6049 				log = 1;
6050 				goto done;
6051 			}
6052 			/* FALLTHROUGH */
6053 		}
6054 		case IPPROTO_AH:
6055 		case IPPROTO_HOPOPTS:
6056 		case IPPROTO_DSTOPTS: {
6057 			/* get next header and header length */
6058 			struct ip6_ext	opt6;
6059 
6060 			if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6061 			    NULL, &reason, pd.af)) {
6062 				DPFPRINTF(PF_DEBUG_MISC,
6063 				    ("pf: IPv6 short opt\n"));
6064 				action = PF_DROP;
6065 				log = 1;
6066 				goto done;
6067 			}
6068 			if (pd.proto == IPPROTO_AH)
6069 				off += (opt6.ip6e_len + 2) * 4;
6070 			else
6071 				off += (opt6.ip6e_len + 1) * 8;
6072 			pd.proto = opt6.ip6e_nxt;
6073 			/* goto the next header */
6074 			break;
6075 		}
6076 		default:
6077 			terminal++;
6078 			break;
6079 		}
6080 	} while (!terminal);
6081 
6082 	/* if there's no routing header, use unmodified mbuf for checksumming */
6083 	if (!n)
6084 		n = m;
6085 
6086 	switch (pd.proto) {
6087 
6088 	case IPPROTO_TCP: {
6089 		struct tcphdr	th;
6090 
6091 		pd.hdr.tcp = &th;
6092 		if (!pf_pull_hdr(m, off, &th, sizeof(th),
6093 		    &action, &reason, AF_INET6)) {
6094 			log = action != PF_PASS;
6095 			goto done;
6096 		}
6097 		pd.p_len = pd.tot_len - off - (th.th_off << 2);
6098 		action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6099 		if (action == PF_DROP)
6100 			goto done;
6101 		action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6102 		    &reason);
6103 		if (action == PF_PASS) {
6104 			if (pfsync_update_state_ptr != NULL)
6105 				pfsync_update_state_ptr(s);
6106 			r = s->rule.ptr;
6107 			a = s->anchor.ptr;
6108 			log = s->log;
6109 		} else if (s == NULL)
6110 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6111 			    &a, &ruleset, inp);
6112 		break;
6113 	}
6114 
6115 	case IPPROTO_UDP: {
6116 		struct udphdr	uh;
6117 
6118 		pd.hdr.udp = &uh;
6119 		if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6120 		    &action, &reason, AF_INET6)) {
6121 			log = action != PF_PASS;
6122 			goto done;
6123 		}
6124 		if (uh.uh_dport == 0 ||
6125 		    ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6126 		    ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6127 			action = PF_DROP;
6128 			REASON_SET(&reason, PFRES_SHORT);
6129 			goto done;
6130 		}
6131 		action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6132 		if (action == PF_PASS) {
6133 			if (pfsync_update_state_ptr != NULL)
6134 				pfsync_update_state_ptr(s);
6135 			r = s->rule.ptr;
6136 			a = s->anchor.ptr;
6137 			log = s->log;
6138 		} else if (s == NULL)
6139 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6140 			    &a, &ruleset, inp);
6141 		break;
6142 	}
6143 
6144 	case IPPROTO_ICMP: {
6145 		action = PF_DROP;
6146 		DPFPRINTF(PF_DEBUG_MISC,
6147 		    ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6148 		goto done;
6149 	}
6150 
6151 	case IPPROTO_ICMPV6: {
6152 		struct icmp6_hdr	ih;
6153 
6154 		pd.hdr.icmp6 = &ih;
6155 		if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6156 		    &action, &reason, AF_INET6)) {
6157 			log = action != PF_PASS;
6158 			goto done;
6159 		}
6160 		action = pf_test_state_icmp(&s, dir, kif,
6161 		    m, off, h, &pd, &reason);
6162 		if (action == PF_PASS) {
6163 			if (pfsync_update_state_ptr != NULL)
6164 				pfsync_update_state_ptr(s);
6165 			r = s->rule.ptr;
6166 			a = s->anchor.ptr;
6167 			log = s->log;
6168 		} else if (s == NULL)
6169 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6170 			    &a, &ruleset, inp);
6171 		break;
6172 	}
6173 
6174 	default:
6175 		action = pf_test_state_other(&s, dir, kif, m, &pd);
6176 		if (action == PF_PASS) {
6177 			if (pfsync_update_state_ptr != NULL)
6178 				pfsync_update_state_ptr(s);
6179 			r = s->rule.ptr;
6180 			a = s->anchor.ptr;
6181 			log = s->log;
6182 		} else if (s == NULL)
6183 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6184 			    &a, &ruleset, inp);
6185 		break;
6186 	}
6187 
6188 done:
6189 	PF_RULES_RUNLOCK();
6190 	if (n != m) {
6191 		m_freem(n);
6192 		n = NULL;
6193 	}
6194 
6195 	/* handle dangerous IPv6 extension headers. */
6196 	if (action == PF_PASS && rh_cnt &&
6197 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6198 		action = PF_DROP;
6199 		REASON_SET(&reason, PFRES_IPOPTIONS);
6200 		log = 1;
6201 		DPFPRINTF(PF_DEBUG_MISC,
6202 		    ("pf: dropping packet with dangerous v6 headers\n"));
6203 	}
6204 
6205 	if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6206 		action = PF_DROP;
6207 		REASON_SET(&reason, PFRES_MEMORY);
6208 	}
6209 	if (r->rtableid >= 0)
6210 		M_SETFIB(m, r->rtableid);
6211 
6212 #ifdef ALTQ
6213 	if (action == PF_PASS && r->qid) {
6214 		if (pd.pf_mtag == NULL &&
6215 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6216 			action = PF_DROP;
6217 			REASON_SET(&reason, PFRES_MEMORY);
6218 		}
6219 		if (pd.tos & IPTOS_LOWDELAY)
6220 			pd.pf_mtag->qid = r->pqid;
6221 		else
6222 			pd.pf_mtag->qid = r->qid;
6223 		/* add hints for ecn */
6224 		pd.pf_mtag->hdr = h;
6225 	}
6226 #endif /* ALTQ */
6227 
6228 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6229 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6230 	    (s->nat_rule.ptr->action == PF_RDR ||
6231 	    s->nat_rule.ptr->action == PF_BINAT) &&
6232 	    IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6233 		m->m_flags |= M_SKIP_FIREWALL;
6234 
6235 	/* XXX: Anybody working on it?! */
6236 	if (r->divert.port)
6237 		printf("pf: divert(9) is not supported for IPv6\n");
6238 
6239 	if (log) {
6240 		struct pf_rule *lr;
6241 
6242 		if (s != NULL && s->nat_rule.ptr != NULL &&
6243 		    s->nat_rule.ptr->log & PF_LOG_ALL)
6244 			lr = s->nat_rule.ptr;
6245 		else
6246 			lr = r;
6247 		PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6248 		    &pd, (s == NULL));
6249 	}
6250 
6251 	kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6252 	kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
6253 
6254 	if (action == PF_PASS || r->action == PF_DROP) {
6255 		dirndx = (dir == PF_OUT);
6256 		r->packets[dirndx]++;
6257 		r->bytes[dirndx] += pd.tot_len;
6258 		if (a != NULL) {
6259 			a->packets[dirndx]++;
6260 			a->bytes[dirndx] += pd.tot_len;
6261 		}
6262 		if (s != NULL) {
6263 			if (s->nat_rule.ptr != NULL) {
6264 				s->nat_rule.ptr->packets[dirndx]++;
6265 				s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6266 			}
6267 			if (s->src_node != NULL) {
6268 				s->src_node->packets[dirndx]++;
6269 				s->src_node->bytes[dirndx] += pd.tot_len;
6270 			}
6271 			if (s->nat_src_node != NULL) {
6272 				s->nat_src_node->packets[dirndx]++;
6273 				s->nat_src_node->bytes[dirndx] += pd.tot_len;
6274 			}
6275 			dirndx = (dir == s->direction) ? 0 : 1;
6276 			s->packets[dirndx]++;
6277 			s->bytes[dirndx] += pd.tot_len;
6278 		}
6279 		tr = r;
6280 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6281 		if (nr != NULL && r == &V_pf_default_rule)
6282 			tr = nr;
6283 		if (tr->src.addr.type == PF_ADDR_TABLE)
6284 			pfr_update_stats(tr->src.addr.p.tbl,
6285 			    (s == NULL) ? pd.src :
6286 			    &s->key[(s->direction == PF_IN)]->addr[0],
6287 			    pd.af, pd.tot_len, dir == PF_OUT,
6288 			    r->action == PF_PASS, tr->src.neg);
6289 		if (tr->dst.addr.type == PF_ADDR_TABLE)
6290 			pfr_update_stats(tr->dst.addr.p.tbl,
6291 			    (s == NULL) ? pd.dst :
6292 			    &s->key[(s->direction == PF_IN)]->addr[1],
6293 			    pd.af, pd.tot_len, dir == PF_OUT,
6294 			    r->action == PF_PASS, tr->dst.neg);
6295 	}
6296 
6297 	switch (action) {
6298 	case PF_SYNPROXY_DROP:
6299 		m_freem(*m0);
6300 	case PF_DEFER:
6301 		*m0 = NULL;
6302 		action = PF_PASS;
6303 		break;
6304 	default:
6305 		/* pf_route6() returns unlocked. */
6306 		if (r->rt) {
6307 			pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
6308 			return (action);
6309 		}
6310 		break;
6311 	}
6312 
6313 	if (s)
6314 		PF_STATE_UNLOCK(s);
6315 
6316 	return (action);
6317 }
6318 #endif /* INET6 */
6319