xref: /freebsd/sys/netpfil/pf/pf.c (revision ce6a89e27cd190313be39bb479880aeda4778436)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2001 Daniel Hartmeier
5  * Copyright (c) 2002 - 2008 Henning Brauer
6  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  *
13  *    - Redistributions of source code must retain the above copyright
14  *      notice, this list of conditions and the following disclaimer.
15  *    - Redistributions in binary form must reproduce the above
16  *      copyright notice, this list of conditions and the following
17  *      disclaimer in the documentation and/or other materials provided
18  *      with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
28  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
30  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31  * POSSIBILITY OF SUCH DAMAGE.
32  *
33  * Effort sponsored in part by the Defense Advanced Research Projects
34  * Agency (DARPA) and Air Force Research Laboratory, Air Force
35  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
36  *
37  *	$OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42 
43 #include "opt_inet.h"
44 #include "opt_inet6.h"
45 #include "opt_bpf.h"
46 #include "opt_pf.h"
47 #include "opt_sctp.h"
48 
49 #include <sys/param.h>
50 #include <sys/bus.h>
51 #include <sys/endian.h>
52 #include <sys/gsb_crc32.h>
53 #include <sys/hash.h>
54 #include <sys/interrupt.h>
55 #include <sys/kernel.h>
56 #include <sys/kthread.h>
57 #include <sys/limits.h>
58 #include <sys/mbuf.h>
59 #include <sys/md5.h>
60 #include <sys/random.h>
61 #include <sys/refcount.h>
62 #include <sys/socket.h>
63 #include <sys/sysctl.h>
64 #include <sys/taskqueue.h>
65 #include <sys/ucred.h>
66 
67 #include <net/if.h>
68 #include <net/if_var.h>
69 #include <net/if_types.h>
70 #include <net/if_vlan_var.h>
71 #include <net/route.h>
72 #include <net/radix_mpath.h>
73 #include <net/vnet.h>
74 
75 #include <net/pfil.h>
76 #include <net/pfvar.h>
77 #include <net/if_pflog.h>
78 #include <net/if_pfsync.h>
79 
80 #include <netinet/in_pcb.h>
81 #include <netinet/in_var.h>
82 #include <netinet/in_fib.h>
83 #include <netinet/ip.h>
84 #include <netinet/ip_fw.h>
85 #include <netinet/ip_icmp.h>
86 #include <netinet/icmp_var.h>
87 #include <netinet/ip_var.h>
88 #include <netinet/tcp.h>
89 #include <netinet/tcp_fsm.h>
90 #include <netinet/tcp_seq.h>
91 #include <netinet/tcp_timer.h>
92 #include <netinet/tcp_var.h>
93 #include <netinet/udp.h>
94 #include <netinet/udp_var.h>
95 
96 #ifdef INET6
97 #include <netinet/ip6.h>
98 #include <netinet/icmp6.h>
99 #include <netinet6/nd6.h>
100 #include <netinet6/ip6_var.h>
101 #include <netinet6/in6_pcb.h>
102 #include <netinet6/in6_fib.h>
103 #include <netinet6/scope6_var.h>
104 #endif /* INET6 */
105 
106 #ifdef SCTP
107 #include <netinet/sctp_crc32.h>
108 #endif
109 
110 #include <machine/in_cksum.h>
111 #include <security/mac/mac_framework.h>
112 
113 #define	DPFPRINTF(n, x)	if (V_pf_status.debug >= (n)) printf x
114 
115 /*
116  * Global variables
117  */
118 
119 /* state tables */
120 VNET_DEFINE(struct pf_altqqueue,	 pf_altqs[4]);
121 VNET_DEFINE(struct pf_palist,		 pf_pabuf);
122 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_active);
123 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_active);
124 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_inactive);
125 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_inactive);
126 VNET_DEFINE(struct pf_kstatus,		 pf_status);
127 
128 VNET_DEFINE(u_int32_t,			 ticket_altqs_active);
129 VNET_DEFINE(u_int32_t,			 ticket_altqs_inactive);
130 VNET_DEFINE(int,			 altqs_inactive_open);
131 VNET_DEFINE(u_int32_t,			 ticket_pabuf);
132 
133 VNET_DEFINE(MD5_CTX,			 pf_tcp_secret_ctx);
134 #define	V_pf_tcp_secret_ctx		 VNET(pf_tcp_secret_ctx)
135 VNET_DEFINE(u_char,			 pf_tcp_secret[16]);
136 #define	V_pf_tcp_secret			 VNET(pf_tcp_secret)
137 VNET_DEFINE(int,			 pf_tcp_secret_init);
138 #define	V_pf_tcp_secret_init		 VNET(pf_tcp_secret_init)
139 VNET_DEFINE(int,			 pf_tcp_iss_off);
140 #define	V_pf_tcp_iss_off		 VNET(pf_tcp_iss_off)
141 VNET_DECLARE(int,			 pf_vnet_active);
142 #define	V_pf_vnet_active		 VNET(pf_vnet_active)
143 
144 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx);
145 #define V_pf_purge_idx	VNET(pf_purge_idx)
146 
147 /*
148  * Queue for pf_intr() sends.
149  */
150 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
151 struct pf_send_entry {
152 	STAILQ_ENTRY(pf_send_entry)	pfse_next;
153 	struct mbuf			*pfse_m;
154 	enum {
155 		PFSE_IP,
156 		PFSE_IP6,
157 		PFSE_ICMP,
158 		PFSE_ICMP6,
159 	}				pfse_type;
160 	struct {
161 		int		type;
162 		int		code;
163 		int		mtu;
164 	} icmpopts;
165 };
166 
167 STAILQ_HEAD(pf_send_head, pf_send_entry);
168 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue);
169 #define	V_pf_sendqueue	VNET(pf_sendqueue)
170 
171 static struct mtx pf_sendqueue_mtx;
172 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
173 #define	PF_SENDQ_LOCK()		mtx_lock(&pf_sendqueue_mtx)
174 #define	PF_SENDQ_UNLOCK()	mtx_unlock(&pf_sendqueue_mtx)
175 
176 /*
177  * Queue for pf_overload_task() tasks.
178  */
179 struct pf_overload_entry {
180 	SLIST_ENTRY(pf_overload_entry)	next;
181 	struct pf_addr  		addr;
182 	sa_family_t			af;
183 	uint8_t				dir;
184 	struct pf_rule  		*rule;
185 };
186 
187 SLIST_HEAD(pf_overload_head, pf_overload_entry);
188 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue);
189 #define V_pf_overloadqueue	VNET(pf_overloadqueue)
190 VNET_DEFINE_STATIC(struct task, pf_overloadtask);
191 #define	V_pf_overloadtask	VNET(pf_overloadtask)
192 
193 static struct mtx pf_overloadqueue_mtx;
194 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
195     "pf overload/flush queue", MTX_DEF);
196 #define	PF_OVERLOADQ_LOCK()	mtx_lock(&pf_overloadqueue_mtx)
197 #define	PF_OVERLOADQ_UNLOCK()	mtx_unlock(&pf_overloadqueue_mtx)
198 
199 VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules);
200 struct mtx pf_unlnkdrules_mtx;
201 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
202     MTX_DEF);
203 
204 VNET_DEFINE_STATIC(uma_zone_t,	pf_sources_z);
205 #define	V_pf_sources_z	VNET(pf_sources_z)
206 uma_zone_t		pf_mtag_z;
207 VNET_DEFINE(uma_zone_t,	 pf_state_z);
208 VNET_DEFINE(uma_zone_t,	 pf_state_key_z);
209 
210 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
211 #define	PFID_CPUBITS	8
212 #define	PFID_CPUSHIFT	(sizeof(uint64_t) * NBBY - PFID_CPUBITS)
213 #define	PFID_CPUMASK	((uint64_t)((1 << PFID_CPUBITS) - 1) <<	PFID_CPUSHIFT)
214 #define	PFID_MAXID	(~PFID_CPUMASK)
215 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU);
216 
217 static void		 pf_src_tree_remove_state(struct pf_state *);
218 static void		 pf_init_threshold(struct pf_threshold *, u_int32_t,
219 			    u_int32_t);
220 static void		 pf_add_threshold(struct pf_threshold *);
221 static int		 pf_check_threshold(struct pf_threshold *);
222 
223 static void		 pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
224 			    u_int16_t *, u_int16_t *, struct pf_addr *,
225 			    u_int16_t, u_int8_t, sa_family_t);
226 static int		 pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
227 			    struct tcphdr *, struct pf_state_peer *);
228 static void		 pf_change_icmp(struct pf_addr *, u_int16_t *,
229 			    struct pf_addr *, struct pf_addr *, u_int16_t,
230 			    u_int16_t *, u_int16_t *, u_int16_t *,
231 			    u_int16_t *, u_int8_t, sa_family_t);
232 static void		 pf_send_tcp(struct mbuf *,
233 			    const struct pf_rule *, sa_family_t,
234 			    const struct pf_addr *, const struct pf_addr *,
235 			    u_int16_t, u_int16_t, u_int32_t, u_int32_t,
236 			    u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
237 			    u_int16_t, struct ifnet *);
238 static void		 pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
239 			    sa_family_t, struct pf_rule *);
240 static void		 pf_detach_state(struct pf_state *);
241 static int		 pf_state_key_attach(struct pf_state_key *,
242 			    struct pf_state_key *, struct pf_state *);
243 static void		 pf_state_key_detach(struct pf_state *, int);
244 static int		 pf_state_key_ctor(void *, int, void *, int);
245 static u_int32_t	 pf_tcp_iss(struct pf_pdesc *);
246 static int		 pf_test_rule(struct pf_rule **, struct pf_state **,
247 			    int, struct pfi_kif *, struct mbuf *, int,
248 			    struct pf_pdesc *, struct pf_rule **,
249 			    struct pf_ruleset **, struct inpcb *);
250 static int		 pf_create_state(struct pf_rule *, struct pf_rule *,
251 			    struct pf_rule *, struct pf_pdesc *,
252 			    struct pf_src_node *, struct pf_state_key *,
253 			    struct pf_state_key *, struct mbuf *, int,
254 			    u_int16_t, u_int16_t, int *, struct pfi_kif *,
255 			    struct pf_state **, int, u_int16_t, u_int16_t,
256 			    int);
257 static int		 pf_test_fragment(struct pf_rule **, int,
258 			    struct pfi_kif *, struct mbuf *, void *,
259 			    struct pf_pdesc *, struct pf_rule **,
260 			    struct pf_ruleset **);
261 static int		 pf_tcp_track_full(struct pf_state_peer *,
262 			    struct pf_state_peer *, struct pf_state **,
263 			    struct pfi_kif *, struct mbuf *, int,
264 			    struct pf_pdesc *, u_short *, int *);
265 static int		 pf_tcp_track_sloppy(struct pf_state_peer *,
266 			    struct pf_state_peer *, struct pf_state **,
267 			    struct pf_pdesc *, u_short *);
268 static int		 pf_test_state_tcp(struct pf_state **, int,
269 			    struct pfi_kif *, struct mbuf *, int,
270 			    void *, struct pf_pdesc *, u_short *);
271 static int		 pf_test_state_udp(struct pf_state **, int,
272 			    struct pfi_kif *, struct mbuf *, int,
273 			    void *, struct pf_pdesc *);
274 static int		 pf_test_state_icmp(struct pf_state **, int,
275 			    struct pfi_kif *, struct mbuf *, int,
276 			    void *, struct pf_pdesc *, u_short *);
277 static int		 pf_test_state_other(struct pf_state **, int,
278 			    struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
279 static u_int8_t		 pf_get_wscale(struct mbuf *, int, u_int16_t,
280 			    sa_family_t);
281 static u_int16_t	 pf_get_mss(struct mbuf *, int, u_int16_t,
282 			    sa_family_t);
283 static u_int16_t	 pf_calc_mss(struct pf_addr *, sa_family_t,
284 				int, u_int16_t);
285 static int		 pf_check_proto_cksum(struct mbuf *, int, int,
286 			    u_int8_t, sa_family_t);
287 static void		 pf_print_state_parts(struct pf_state *,
288 			    struct pf_state_key *, struct pf_state_key *);
289 static int		 pf_addr_wrap_neq(struct pf_addr_wrap *,
290 			    struct pf_addr_wrap *);
291 static struct pf_state	*pf_find_state(struct pfi_kif *,
292 			    struct pf_state_key_cmp *, u_int);
293 static int		 pf_src_connlimit(struct pf_state **);
294 static void		 pf_overload_task(void *v, int pending);
295 static int		 pf_insert_src_node(struct pf_src_node **,
296 			    struct pf_rule *, struct pf_addr *, sa_family_t);
297 static u_int		 pf_purge_expired_states(u_int, int);
298 static void		 pf_purge_unlinked_rules(void);
299 static int		 pf_mtag_uminit(void *, int, int);
300 static void		 pf_mtag_free(struct m_tag *);
301 #ifdef INET
302 static void		 pf_route(struct mbuf **, struct pf_rule *, int,
303 			    struct ifnet *, struct pf_state *,
304 			    struct pf_pdesc *, struct inpcb *);
305 #endif /* INET */
306 #ifdef INET6
307 static void		 pf_change_a6(struct pf_addr *, u_int16_t *,
308 			    struct pf_addr *, u_int8_t);
309 static void		 pf_route6(struct mbuf **, struct pf_rule *, int,
310 			    struct ifnet *, struct pf_state *,
311 			    struct pf_pdesc *, struct inpcb *);
312 #endif /* INET6 */
313 
314 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
315 
316 extern int pf_end_threads;
317 extern struct proc *pf_purge_proc;
318 
319 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
320 
321 #define	PACKET_LOOPED(pd)	((pd)->pf_mtag &&			\
322 				 (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
323 
324 #define	STATE_LOOKUP(i, k, d, s, pd)					\
325 	do {								\
326 		(s) = pf_find_state((i), (k), (d));			\
327 		if ((s) == NULL)					\
328 			return (PF_DROP);				\
329 		if (PACKET_LOOPED(pd))					\
330 			return (PF_PASS);				\
331 		if ((d) == PF_OUT &&					\
332 		    (((s)->rule.ptr->rt == PF_ROUTETO &&		\
333 		    (s)->rule.ptr->direction == PF_OUT) ||		\
334 		    ((s)->rule.ptr->rt == PF_REPLYTO &&			\
335 		    (s)->rule.ptr->direction == PF_IN)) &&		\
336 		    (s)->rt_kif != NULL &&				\
337 		    (s)->rt_kif != (i))					\
338 			return (PF_PASS);				\
339 	} while (0)
340 
341 #define	BOUND_IFACE(r, k) \
342 	((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
343 
344 #define	STATE_INC_COUNTERS(s)						\
345 	do {								\
346 		counter_u64_add(s->rule.ptr->states_cur, 1);		\
347 		counter_u64_add(s->rule.ptr->states_tot, 1);		\
348 		if (s->anchor.ptr != NULL) {				\
349 			counter_u64_add(s->anchor.ptr->states_cur, 1);	\
350 			counter_u64_add(s->anchor.ptr->states_tot, 1);	\
351 		}							\
352 		if (s->nat_rule.ptr != NULL) {				\
353 			counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
354 			counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
355 		}							\
356 	} while (0)
357 
358 #define	STATE_DEC_COUNTERS(s)						\
359 	do {								\
360 		if (s->nat_rule.ptr != NULL)				\
361 			counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
362 		if (s->anchor.ptr != NULL)				\
363 			counter_u64_add(s->anchor.ptr->states_cur, -1);	\
364 		counter_u64_add(s->rule.ptr->states_cur, -1);		\
365 	} while (0)
366 
367 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
368 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
369 VNET_DEFINE(struct pf_idhash *, pf_idhash);
370 VNET_DEFINE(struct pf_srchash *, pf_srchash);
371 
372 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
373     "pf(4)");
374 
375 u_long	pf_hashmask;
376 u_long	pf_srchashmask;
377 static u_long	pf_hashsize;
378 static u_long	pf_srchashsize;
379 u_long	pf_ioctl_maxcount = 65535;
380 
381 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
382     &pf_hashsize, 0, "Size of pf(4) states hashtable");
383 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
384     &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
385 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RW,
386     &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call");
387 
388 VNET_DEFINE(void *, pf_swi_cookie);
389 
390 VNET_DEFINE(uint32_t, pf_hashseed);
391 #define	V_pf_hashseed	VNET(pf_hashseed)
392 
393 int
394 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
395 {
396 
397 	switch (af) {
398 #ifdef INET
399 	case AF_INET:
400 		if (a->addr32[0] > b->addr32[0])
401 			return (1);
402 		if (a->addr32[0] < b->addr32[0])
403 			return (-1);
404 		break;
405 #endif /* INET */
406 #ifdef INET6
407 	case AF_INET6:
408 		if (a->addr32[3] > b->addr32[3])
409 			return (1);
410 		if (a->addr32[3] < b->addr32[3])
411 			return (-1);
412 		if (a->addr32[2] > b->addr32[2])
413 			return (1);
414 		if (a->addr32[2] < b->addr32[2])
415 			return (-1);
416 		if (a->addr32[1] > b->addr32[1])
417 			return (1);
418 		if (a->addr32[1] < b->addr32[1])
419 			return (-1);
420 		if (a->addr32[0] > b->addr32[0])
421 			return (1);
422 		if (a->addr32[0] < b->addr32[0])
423 			return (-1);
424 		break;
425 #endif /* INET6 */
426 	default:
427 		panic("%s: unknown address family %u", __func__, af);
428 	}
429 	return (0);
430 }
431 
432 static __inline uint32_t
433 pf_hashkey(struct pf_state_key *sk)
434 {
435 	uint32_t h;
436 
437 	h = murmur3_32_hash32((uint32_t *)sk,
438 	    sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
439 	    V_pf_hashseed);
440 
441 	return (h & pf_hashmask);
442 }
443 
444 static __inline uint32_t
445 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
446 {
447 	uint32_t h;
448 
449 	switch (af) {
450 	case AF_INET:
451 		h = murmur3_32_hash32((uint32_t *)&addr->v4,
452 		    sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
453 		break;
454 	case AF_INET6:
455 		h = murmur3_32_hash32((uint32_t *)&addr->v6,
456 		    sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
457 		break;
458 	default:
459 		panic("%s: unknown address family %u", __func__, af);
460 	}
461 
462 	return (h & pf_srchashmask);
463 }
464 
465 #ifdef ALTQ
466 static int
467 pf_state_hash(struct pf_state *s)
468 {
469 	u_int32_t hv = (intptr_t)s / sizeof(*s);
470 
471 	hv ^= crc32(&s->src, sizeof(s->src));
472 	hv ^= crc32(&s->dst, sizeof(s->dst));
473 	if (hv == 0)
474 		hv = 1;
475 	return (hv);
476 }
477 #endif
478 
479 #ifdef INET6
480 void
481 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
482 {
483 	switch (af) {
484 #ifdef INET
485 	case AF_INET:
486 		dst->addr32[0] = src->addr32[0];
487 		break;
488 #endif /* INET */
489 	case AF_INET6:
490 		dst->addr32[0] = src->addr32[0];
491 		dst->addr32[1] = src->addr32[1];
492 		dst->addr32[2] = src->addr32[2];
493 		dst->addr32[3] = src->addr32[3];
494 		break;
495 	}
496 }
497 #endif /* INET6 */
498 
499 static void
500 pf_init_threshold(struct pf_threshold *threshold,
501     u_int32_t limit, u_int32_t seconds)
502 {
503 	threshold->limit = limit * PF_THRESHOLD_MULT;
504 	threshold->seconds = seconds;
505 	threshold->count = 0;
506 	threshold->last = time_uptime;
507 }
508 
509 static void
510 pf_add_threshold(struct pf_threshold *threshold)
511 {
512 	u_int32_t t = time_uptime, diff = t - threshold->last;
513 
514 	if (diff >= threshold->seconds)
515 		threshold->count = 0;
516 	else
517 		threshold->count -= threshold->count * diff /
518 		    threshold->seconds;
519 	threshold->count += PF_THRESHOLD_MULT;
520 	threshold->last = t;
521 }
522 
523 static int
524 pf_check_threshold(struct pf_threshold *threshold)
525 {
526 	return (threshold->count > threshold->limit);
527 }
528 
529 static int
530 pf_src_connlimit(struct pf_state **state)
531 {
532 	struct pf_overload_entry *pfoe;
533 	int bad = 0;
534 
535 	PF_STATE_LOCK_ASSERT(*state);
536 
537 	(*state)->src_node->conn++;
538 	(*state)->src.tcp_est = 1;
539 	pf_add_threshold(&(*state)->src_node->conn_rate);
540 
541 	if ((*state)->rule.ptr->max_src_conn &&
542 	    (*state)->rule.ptr->max_src_conn <
543 	    (*state)->src_node->conn) {
544 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
545 		bad++;
546 	}
547 
548 	if ((*state)->rule.ptr->max_src_conn_rate.limit &&
549 	    pf_check_threshold(&(*state)->src_node->conn_rate)) {
550 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
551 		bad++;
552 	}
553 
554 	if (!bad)
555 		return (0);
556 
557 	/* Kill this state. */
558 	(*state)->timeout = PFTM_PURGE;
559 	(*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
560 
561 	if ((*state)->rule.ptr->overload_tbl == NULL)
562 		return (1);
563 
564 	/* Schedule overloading and flushing task. */
565 	pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
566 	if (pfoe == NULL)
567 		return (1);	/* too bad :( */
568 
569 	bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
570 	pfoe->af = (*state)->key[PF_SK_WIRE]->af;
571 	pfoe->rule = (*state)->rule.ptr;
572 	pfoe->dir = (*state)->direction;
573 	PF_OVERLOADQ_LOCK();
574 	SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
575 	PF_OVERLOADQ_UNLOCK();
576 	taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
577 
578 	return (1);
579 }
580 
581 static void
582 pf_overload_task(void *v, int pending)
583 {
584 	struct pf_overload_head queue;
585 	struct pfr_addr p;
586 	struct pf_overload_entry *pfoe, *pfoe1;
587 	uint32_t killed = 0;
588 
589 	CURVNET_SET((struct vnet *)v);
590 
591 	PF_OVERLOADQ_LOCK();
592 	queue = V_pf_overloadqueue;
593 	SLIST_INIT(&V_pf_overloadqueue);
594 	PF_OVERLOADQ_UNLOCK();
595 
596 	bzero(&p, sizeof(p));
597 	SLIST_FOREACH(pfoe, &queue, next) {
598 		counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
599 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
600 			printf("%s: blocking address ", __func__);
601 			pf_print_host(&pfoe->addr, 0, pfoe->af);
602 			printf("\n");
603 		}
604 
605 		p.pfra_af = pfoe->af;
606 		switch (pfoe->af) {
607 #ifdef INET
608 		case AF_INET:
609 			p.pfra_net = 32;
610 			p.pfra_ip4addr = pfoe->addr.v4;
611 			break;
612 #endif
613 #ifdef INET6
614 		case AF_INET6:
615 			p.pfra_net = 128;
616 			p.pfra_ip6addr = pfoe->addr.v6;
617 			break;
618 #endif
619 		}
620 
621 		PF_RULES_WLOCK();
622 		pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
623 		PF_RULES_WUNLOCK();
624 	}
625 
626 	/*
627 	 * Remove those entries, that don't need flushing.
628 	 */
629 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
630 		if (pfoe->rule->flush == 0) {
631 			SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
632 			free(pfoe, M_PFTEMP);
633 		} else
634 			counter_u64_add(
635 			    V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
636 
637 	/* If nothing to flush, return. */
638 	if (SLIST_EMPTY(&queue)) {
639 		CURVNET_RESTORE();
640 		return;
641 	}
642 
643 	for (int i = 0; i <= pf_hashmask; i++) {
644 		struct pf_idhash *ih = &V_pf_idhash[i];
645 		struct pf_state_key *sk;
646 		struct pf_state *s;
647 
648 		PF_HASHROW_LOCK(ih);
649 		LIST_FOREACH(s, &ih->states, entry) {
650 		    sk = s->key[PF_SK_WIRE];
651 		    SLIST_FOREACH(pfoe, &queue, next)
652 			if (sk->af == pfoe->af &&
653 			    ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
654 			    pfoe->rule == s->rule.ptr) &&
655 			    ((pfoe->dir == PF_OUT &&
656 			    PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
657 			    (pfoe->dir == PF_IN &&
658 			    PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
659 				s->timeout = PFTM_PURGE;
660 				s->src.state = s->dst.state = TCPS_CLOSED;
661 				killed++;
662 			}
663 		}
664 		PF_HASHROW_UNLOCK(ih);
665 	}
666 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
667 		free(pfoe, M_PFTEMP);
668 	if (V_pf_status.debug >= PF_DEBUG_MISC)
669 		printf("%s: %u states killed", __func__, killed);
670 
671 	CURVNET_RESTORE();
672 }
673 
674 /*
675  * Can return locked on failure, so that we can consistently
676  * allocate and insert a new one.
677  */
678 struct pf_src_node *
679 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
680 	int returnlocked)
681 {
682 	struct pf_srchash *sh;
683 	struct pf_src_node *n;
684 
685 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
686 
687 	sh = &V_pf_srchash[pf_hashsrc(src, af)];
688 	PF_HASHROW_LOCK(sh);
689 	LIST_FOREACH(n, &sh->nodes, entry)
690 		if (n->rule.ptr == rule && n->af == af &&
691 		    ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
692 		    (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
693 			break;
694 	if (n != NULL) {
695 		n->states++;
696 		PF_HASHROW_UNLOCK(sh);
697 	} else if (returnlocked == 0)
698 		PF_HASHROW_UNLOCK(sh);
699 
700 	return (n);
701 }
702 
703 static int
704 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
705     struct pf_addr *src, sa_family_t af)
706 {
707 
708 	KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
709 	    rule->rpool.opts & PF_POOL_STICKYADDR),
710 	    ("%s for non-tracking rule %p", __func__, rule));
711 
712 	if (*sn == NULL)
713 		*sn = pf_find_src_node(src, rule, af, 1);
714 
715 	if (*sn == NULL) {
716 		struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
717 
718 		PF_HASHROW_ASSERT(sh);
719 
720 		if (!rule->max_src_nodes ||
721 		    counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
722 			(*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
723 		else
724 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
725 			    1);
726 		if ((*sn) == NULL) {
727 			PF_HASHROW_UNLOCK(sh);
728 			return (-1);
729 		}
730 
731 		pf_init_threshold(&(*sn)->conn_rate,
732 		    rule->max_src_conn_rate.limit,
733 		    rule->max_src_conn_rate.seconds);
734 
735 		(*sn)->af = af;
736 		(*sn)->rule.ptr = rule;
737 		PF_ACPY(&(*sn)->addr, src, af);
738 		LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
739 		(*sn)->creation = time_uptime;
740 		(*sn)->ruletype = rule->action;
741 		(*sn)->states = 1;
742 		if ((*sn)->rule.ptr != NULL)
743 			counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
744 		PF_HASHROW_UNLOCK(sh);
745 		counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
746 	} else {
747 		if (rule->max_src_states &&
748 		    (*sn)->states >= rule->max_src_states) {
749 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
750 			    1);
751 			return (-1);
752 		}
753 	}
754 	return (0);
755 }
756 
757 void
758 pf_unlink_src_node(struct pf_src_node *src)
759 {
760 
761 	PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]);
762 	LIST_REMOVE(src, entry);
763 	if (src->rule.ptr)
764 		counter_u64_add(src->rule.ptr->src_nodes, -1);
765 }
766 
767 u_int
768 pf_free_src_nodes(struct pf_src_node_list *head)
769 {
770 	struct pf_src_node *sn, *tmp;
771 	u_int count = 0;
772 
773 	LIST_FOREACH_SAFE(sn, head, entry, tmp) {
774 		uma_zfree(V_pf_sources_z, sn);
775 		count++;
776 	}
777 
778 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
779 
780 	return (count);
781 }
782 
783 void
784 pf_mtag_initialize()
785 {
786 
787 	pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
788 	    sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
789 	    UMA_ALIGN_PTR, 0);
790 }
791 
792 /* Per-vnet data storage structures initialization. */
793 void
794 pf_initialize()
795 {
796 	struct pf_keyhash	*kh;
797 	struct pf_idhash	*ih;
798 	struct pf_srchash	*sh;
799 	u_int i;
800 
801 	if (pf_hashsize == 0 || !powerof2(pf_hashsize))
802 		pf_hashsize = PF_HASHSIZ;
803 	if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
804 		pf_srchashsize = PF_SRCHASHSIZ;
805 
806 	V_pf_hashseed = arc4random();
807 
808 	/* States and state keys storage. */
809 	V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
810 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
811 	V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
812 	uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
813 	uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
814 
815 	V_pf_state_key_z = uma_zcreate("pf state keys",
816 	    sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
817 	    UMA_ALIGN_PTR, 0);
818 
819 	V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash),
820 	    M_PFHASH, M_NOWAIT | M_ZERO);
821 	V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash),
822 	    M_PFHASH, M_NOWAIT | M_ZERO);
823 	if (V_pf_keyhash == NULL || V_pf_idhash == NULL) {
824 		printf("pf: Unable to allocate memory for "
825 		    "state_hashsize %lu.\n", pf_hashsize);
826 
827 		free(V_pf_keyhash, M_PFHASH);
828 		free(V_pf_idhash, M_PFHASH);
829 
830 		pf_hashsize = PF_HASHSIZ;
831 		V_pf_keyhash = mallocarray(pf_hashsize,
832 		    sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO);
833 		V_pf_idhash = mallocarray(pf_hashsize,
834 		    sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO);
835 	}
836 
837 	pf_hashmask = pf_hashsize - 1;
838 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
839 	    i++, kh++, ih++) {
840 		mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
841 		mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
842 	}
843 
844 	/* Source nodes. */
845 	V_pf_sources_z = uma_zcreate("pf source nodes",
846 	    sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
847 	    0);
848 	V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
849 	uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
850 	uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
851 
852 	V_pf_srchash = mallocarray(pf_srchashsize,
853 	    sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO);
854 	if (V_pf_srchash == NULL) {
855 		printf("pf: Unable to allocate memory for "
856 		    "source_hashsize %lu.\n", pf_srchashsize);
857 
858 		pf_srchashsize = PF_SRCHASHSIZ;
859 		V_pf_srchash = mallocarray(pf_srchashsize,
860 		    sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO);
861 	}
862 
863 	pf_srchashmask = pf_srchashsize - 1;
864 	for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
865 		mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
866 
867 	/* ALTQ */
868 	TAILQ_INIT(&V_pf_altqs[0]);
869 	TAILQ_INIT(&V_pf_altqs[1]);
870 	TAILQ_INIT(&V_pf_altqs[2]);
871 	TAILQ_INIT(&V_pf_altqs[3]);
872 	TAILQ_INIT(&V_pf_pabuf);
873 	V_pf_altqs_active = &V_pf_altqs[0];
874 	V_pf_altq_ifs_active = &V_pf_altqs[1];
875 	V_pf_altqs_inactive = &V_pf_altqs[2];
876 	V_pf_altq_ifs_inactive = &V_pf_altqs[3];
877 
878 	/* Send & overload+flush queues. */
879 	STAILQ_INIT(&V_pf_sendqueue);
880 	SLIST_INIT(&V_pf_overloadqueue);
881 	TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
882 
883 	/* Unlinked, but may be referenced rules. */
884 	TAILQ_INIT(&V_pf_unlinked_rules);
885 }
886 
887 void
888 pf_mtag_cleanup()
889 {
890 
891 	uma_zdestroy(pf_mtag_z);
892 }
893 
894 void
895 pf_cleanup()
896 {
897 	struct pf_keyhash	*kh;
898 	struct pf_idhash	*ih;
899 	struct pf_srchash	*sh;
900 	struct pf_send_entry	*pfse, *next;
901 	u_int i;
902 
903 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
904 	    i++, kh++, ih++) {
905 		KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
906 		    __func__));
907 		KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
908 		    __func__));
909 		mtx_destroy(&kh->lock);
910 		mtx_destroy(&ih->lock);
911 	}
912 	free(V_pf_keyhash, M_PFHASH);
913 	free(V_pf_idhash, M_PFHASH);
914 
915 	for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
916 		KASSERT(LIST_EMPTY(&sh->nodes),
917 		    ("%s: source node hash not empty", __func__));
918 		mtx_destroy(&sh->lock);
919 	}
920 	free(V_pf_srchash, M_PFHASH);
921 
922 	STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
923 		m_freem(pfse->pfse_m);
924 		free(pfse, M_PFTEMP);
925 	}
926 
927 	uma_zdestroy(V_pf_sources_z);
928 	uma_zdestroy(V_pf_state_z);
929 	uma_zdestroy(V_pf_state_key_z);
930 }
931 
932 static int
933 pf_mtag_uminit(void *mem, int size, int how)
934 {
935 	struct m_tag *t;
936 
937 	t = (struct m_tag *)mem;
938 	t->m_tag_cookie = MTAG_ABI_COMPAT;
939 	t->m_tag_id = PACKET_TAG_PF;
940 	t->m_tag_len = sizeof(struct pf_mtag);
941 	t->m_tag_free = pf_mtag_free;
942 
943 	return (0);
944 }
945 
946 static void
947 pf_mtag_free(struct m_tag *t)
948 {
949 
950 	uma_zfree(pf_mtag_z, t);
951 }
952 
953 struct pf_mtag *
954 pf_get_mtag(struct mbuf *m)
955 {
956 	struct m_tag *mtag;
957 
958 	if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
959 		return ((struct pf_mtag *)(mtag + 1));
960 
961 	mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
962 	if (mtag == NULL)
963 		return (NULL);
964 	bzero(mtag + 1, sizeof(struct pf_mtag));
965 	m_tag_prepend(m, mtag);
966 
967 	return ((struct pf_mtag *)(mtag + 1));
968 }
969 
970 static int
971 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
972     struct pf_state *s)
973 {
974 	struct pf_keyhash	*khs, *khw, *kh;
975 	struct pf_state_key	*sk, *cur;
976 	struct pf_state		*si, *olds = NULL;
977 	int idx;
978 
979 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
980 	KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
981 	KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
982 
983 	/*
984 	 * We need to lock hash slots of both keys. To avoid deadlock
985 	 * we always lock the slot with lower address first. Unlock order
986 	 * isn't important.
987 	 *
988 	 * We also need to lock ID hash slot before dropping key
989 	 * locks. On success we return with ID hash slot locked.
990 	 */
991 
992 	if (skw == sks) {
993 		khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
994 		PF_HASHROW_LOCK(khs);
995 	} else {
996 		khs = &V_pf_keyhash[pf_hashkey(sks)];
997 		khw = &V_pf_keyhash[pf_hashkey(skw)];
998 		if (khs == khw) {
999 			PF_HASHROW_LOCK(khs);
1000 		} else if (khs < khw) {
1001 			PF_HASHROW_LOCK(khs);
1002 			PF_HASHROW_LOCK(khw);
1003 		} else {
1004 			PF_HASHROW_LOCK(khw);
1005 			PF_HASHROW_LOCK(khs);
1006 		}
1007 	}
1008 
1009 #define	KEYS_UNLOCK()	do {			\
1010 	if (khs != khw) {			\
1011 		PF_HASHROW_UNLOCK(khs);		\
1012 		PF_HASHROW_UNLOCK(khw);		\
1013 	} else					\
1014 		PF_HASHROW_UNLOCK(khs);		\
1015 } while (0)
1016 
1017 	/*
1018 	 * First run: start with wire key.
1019 	 */
1020 	sk = skw;
1021 	kh = khw;
1022 	idx = PF_SK_WIRE;
1023 
1024 keyattach:
1025 	LIST_FOREACH(cur, &kh->keys, entry)
1026 		if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
1027 			break;
1028 
1029 	if (cur != NULL) {
1030 		/* Key exists. Check for same kif, if none, add to key. */
1031 		TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
1032 			struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
1033 
1034 			PF_HASHROW_LOCK(ih);
1035 			if (si->kif == s->kif &&
1036 			    si->direction == s->direction) {
1037 				if (sk->proto == IPPROTO_TCP &&
1038 				    si->src.state >= TCPS_FIN_WAIT_2 &&
1039 				    si->dst.state >= TCPS_FIN_WAIT_2) {
1040 					/*
1041 					 * New state matches an old >FIN_WAIT_2
1042 					 * state. We can't drop key hash locks,
1043 					 * thus we can't unlink it properly.
1044 					 *
1045 					 * As a workaround we drop it into
1046 					 * TCPS_CLOSED state, schedule purge
1047 					 * ASAP and push it into the very end
1048 					 * of the slot TAILQ, so that it won't
1049 					 * conflict with our new state.
1050 					 */
1051 					si->src.state = si->dst.state =
1052 					    TCPS_CLOSED;
1053 					si->timeout = PFTM_PURGE;
1054 					olds = si;
1055 				} else {
1056 					if (V_pf_status.debug >= PF_DEBUG_MISC) {
1057 						printf("pf: %s key attach "
1058 						    "failed on %s: ",
1059 						    (idx == PF_SK_WIRE) ?
1060 						    "wire" : "stack",
1061 						    s->kif->pfik_name);
1062 						pf_print_state_parts(s,
1063 						    (idx == PF_SK_WIRE) ?
1064 						    sk : NULL,
1065 						    (idx == PF_SK_STACK) ?
1066 						    sk : NULL);
1067 						printf(", existing: ");
1068 						pf_print_state_parts(si,
1069 						    (idx == PF_SK_WIRE) ?
1070 						    sk : NULL,
1071 						    (idx == PF_SK_STACK) ?
1072 						    sk : NULL);
1073 						printf("\n");
1074 					}
1075 					PF_HASHROW_UNLOCK(ih);
1076 					KEYS_UNLOCK();
1077 					uma_zfree(V_pf_state_key_z, sk);
1078 					if (idx == PF_SK_STACK)
1079 						pf_detach_state(s);
1080 					return (EEXIST); /* collision! */
1081 				}
1082 			}
1083 			PF_HASHROW_UNLOCK(ih);
1084 		}
1085 		uma_zfree(V_pf_state_key_z, sk);
1086 		s->key[idx] = cur;
1087 	} else {
1088 		LIST_INSERT_HEAD(&kh->keys, sk, entry);
1089 		s->key[idx] = sk;
1090 	}
1091 
1092 stateattach:
1093 	/* List is sorted, if-bound states before floating. */
1094 	if (s->kif == V_pfi_all)
1095 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1096 	else
1097 		TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1098 
1099 	if (olds) {
1100 		TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1101 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1102 		    key_list[idx]);
1103 		olds = NULL;
1104 	}
1105 
1106 	/*
1107 	 * Attach done. See how should we (or should not?)
1108 	 * attach a second key.
1109 	 */
1110 	if (sks == skw) {
1111 		s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1112 		idx = PF_SK_STACK;
1113 		sks = NULL;
1114 		goto stateattach;
1115 	} else if (sks != NULL) {
1116 		/*
1117 		 * Continue attaching with stack key.
1118 		 */
1119 		sk = sks;
1120 		kh = khs;
1121 		idx = PF_SK_STACK;
1122 		sks = NULL;
1123 		goto keyattach;
1124 	}
1125 
1126 	PF_STATE_LOCK(s);
1127 	KEYS_UNLOCK();
1128 
1129 	KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1130 	    ("%s failure", __func__));
1131 
1132 	return (0);
1133 #undef	KEYS_UNLOCK
1134 }
1135 
1136 static void
1137 pf_detach_state(struct pf_state *s)
1138 {
1139 	struct pf_state_key *sks = s->key[PF_SK_STACK];
1140 	struct pf_keyhash *kh;
1141 
1142 	if (sks != NULL) {
1143 		kh = &V_pf_keyhash[pf_hashkey(sks)];
1144 		PF_HASHROW_LOCK(kh);
1145 		if (s->key[PF_SK_STACK] != NULL)
1146 			pf_state_key_detach(s, PF_SK_STACK);
1147 		/*
1148 		 * If both point to same key, then we are done.
1149 		 */
1150 		if (sks == s->key[PF_SK_WIRE]) {
1151 			pf_state_key_detach(s, PF_SK_WIRE);
1152 			PF_HASHROW_UNLOCK(kh);
1153 			return;
1154 		}
1155 		PF_HASHROW_UNLOCK(kh);
1156 	}
1157 
1158 	if (s->key[PF_SK_WIRE] != NULL) {
1159 		kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1160 		PF_HASHROW_LOCK(kh);
1161 		if (s->key[PF_SK_WIRE] != NULL)
1162 			pf_state_key_detach(s, PF_SK_WIRE);
1163 		PF_HASHROW_UNLOCK(kh);
1164 	}
1165 }
1166 
1167 static void
1168 pf_state_key_detach(struct pf_state *s, int idx)
1169 {
1170 	struct pf_state_key *sk = s->key[idx];
1171 #ifdef INVARIANTS
1172 	struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1173 
1174 	PF_HASHROW_ASSERT(kh);
1175 #endif
1176 	TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1177 	s->key[idx] = NULL;
1178 
1179 	if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1180 		LIST_REMOVE(sk, entry);
1181 		uma_zfree(V_pf_state_key_z, sk);
1182 	}
1183 }
1184 
1185 static int
1186 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1187 {
1188 	struct pf_state_key *sk = mem;
1189 
1190 	bzero(sk, sizeof(struct pf_state_key_cmp));
1191 	TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1192 	TAILQ_INIT(&sk->states[PF_SK_STACK]);
1193 
1194 	return (0);
1195 }
1196 
1197 struct pf_state_key *
1198 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1199 	struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1200 {
1201 	struct pf_state_key *sk;
1202 
1203 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1204 	if (sk == NULL)
1205 		return (NULL);
1206 
1207 	PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1208 	PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1209 	sk->port[pd->sidx] = sport;
1210 	sk->port[pd->didx] = dport;
1211 	sk->proto = pd->proto;
1212 	sk->af = pd->af;
1213 
1214 	return (sk);
1215 }
1216 
1217 struct pf_state_key *
1218 pf_state_key_clone(struct pf_state_key *orig)
1219 {
1220 	struct pf_state_key *sk;
1221 
1222 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1223 	if (sk == NULL)
1224 		return (NULL);
1225 
1226 	bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1227 
1228 	return (sk);
1229 }
1230 
1231 int
1232 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1233     struct pf_state_key *sks, struct pf_state *s)
1234 {
1235 	struct pf_idhash *ih;
1236 	struct pf_state *cur;
1237 	int error;
1238 
1239 	KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1240 	    ("%s: sks not pristine", __func__));
1241 	KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1242 	    ("%s: skw not pristine", __func__));
1243 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1244 
1245 	s->kif = kif;
1246 
1247 	if (s->id == 0 && s->creatorid == 0) {
1248 		/* XXX: should be atomic, but probability of collision low */
1249 		if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1250 			V_pf_stateid[curcpu] = 1;
1251 		s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1252 		s->id = htobe64(s->id);
1253 		s->creatorid = V_pf_status.hostid;
1254 	}
1255 
1256 	/* Returns with ID locked on success. */
1257 	if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1258 		return (error);
1259 
1260 	ih = &V_pf_idhash[PF_IDHASH(s)];
1261 	PF_HASHROW_ASSERT(ih);
1262 	LIST_FOREACH(cur, &ih->states, entry)
1263 		if (cur->id == s->id && cur->creatorid == s->creatorid)
1264 			break;
1265 
1266 	if (cur != NULL) {
1267 		PF_HASHROW_UNLOCK(ih);
1268 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1269 			printf("pf: state ID collision: "
1270 			    "id: %016llx creatorid: %08x\n",
1271 			    (unsigned long long)be64toh(s->id),
1272 			    ntohl(s->creatorid));
1273 		}
1274 		pf_detach_state(s);
1275 		return (EEXIST);
1276 	}
1277 	LIST_INSERT_HEAD(&ih->states, s, entry);
1278 	/* One for keys, one for ID hash. */
1279 	refcount_init(&s->refs, 2);
1280 
1281 	counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1282 	if (V_pfsync_insert_state_ptr != NULL)
1283 		V_pfsync_insert_state_ptr(s);
1284 
1285 	/* Returns locked. */
1286 	return (0);
1287 }
1288 
1289 /*
1290  * Find state by ID: returns with locked row on success.
1291  */
1292 struct pf_state *
1293 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1294 {
1295 	struct pf_idhash *ih;
1296 	struct pf_state *s;
1297 
1298 	counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1299 
1300 	ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1301 
1302 	PF_HASHROW_LOCK(ih);
1303 	LIST_FOREACH(s, &ih->states, entry)
1304 		if (s->id == id && s->creatorid == creatorid)
1305 			break;
1306 
1307 	if (s == NULL)
1308 		PF_HASHROW_UNLOCK(ih);
1309 
1310 	return (s);
1311 }
1312 
1313 /*
1314  * Find state by key.
1315  * Returns with ID hash slot locked on success.
1316  */
1317 static struct pf_state *
1318 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1319 {
1320 	struct pf_keyhash	*kh;
1321 	struct pf_state_key	*sk;
1322 	struct pf_state		*s;
1323 	int idx;
1324 
1325 	counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1326 
1327 	kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1328 
1329 	PF_HASHROW_LOCK(kh);
1330 	LIST_FOREACH(sk, &kh->keys, entry)
1331 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1332 			break;
1333 	if (sk == NULL) {
1334 		PF_HASHROW_UNLOCK(kh);
1335 		return (NULL);
1336 	}
1337 
1338 	idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1339 
1340 	/* List is sorted, if-bound states before floating ones. */
1341 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1342 		if (s->kif == V_pfi_all || s->kif == kif) {
1343 			PF_STATE_LOCK(s);
1344 			PF_HASHROW_UNLOCK(kh);
1345 			if (s->timeout >= PFTM_MAX) {
1346 				/*
1347 				 * State is either being processed by
1348 				 * pf_unlink_state() in an other thread, or
1349 				 * is scheduled for immediate expiry.
1350 				 */
1351 				PF_STATE_UNLOCK(s);
1352 				return (NULL);
1353 			}
1354 			return (s);
1355 		}
1356 	PF_HASHROW_UNLOCK(kh);
1357 
1358 	return (NULL);
1359 }
1360 
1361 struct pf_state *
1362 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1363 {
1364 	struct pf_keyhash	*kh;
1365 	struct pf_state_key	*sk;
1366 	struct pf_state		*s, *ret = NULL;
1367 	int			 idx, inout = 0;
1368 
1369 	counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1370 
1371 	kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1372 
1373 	PF_HASHROW_LOCK(kh);
1374 	LIST_FOREACH(sk, &kh->keys, entry)
1375 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1376 			break;
1377 	if (sk == NULL) {
1378 		PF_HASHROW_UNLOCK(kh);
1379 		return (NULL);
1380 	}
1381 	switch (dir) {
1382 	case PF_IN:
1383 		idx = PF_SK_WIRE;
1384 		break;
1385 	case PF_OUT:
1386 		idx = PF_SK_STACK;
1387 		break;
1388 	case PF_INOUT:
1389 		idx = PF_SK_WIRE;
1390 		inout = 1;
1391 		break;
1392 	default:
1393 		panic("%s: dir %u", __func__, dir);
1394 	}
1395 second_run:
1396 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1397 		if (more == NULL) {
1398 			PF_HASHROW_UNLOCK(kh);
1399 			return (s);
1400 		}
1401 
1402 		if (ret)
1403 			(*more)++;
1404 		else
1405 			ret = s;
1406 	}
1407 	if (inout == 1) {
1408 		inout = 0;
1409 		idx = PF_SK_STACK;
1410 		goto second_run;
1411 	}
1412 	PF_HASHROW_UNLOCK(kh);
1413 
1414 	return (ret);
1415 }
1416 
1417 /* END state table stuff */
1418 
1419 static void
1420 pf_send(struct pf_send_entry *pfse)
1421 {
1422 
1423 	PF_SENDQ_LOCK();
1424 	STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1425 	PF_SENDQ_UNLOCK();
1426 	swi_sched(V_pf_swi_cookie, 0);
1427 }
1428 
1429 void
1430 pf_intr(void *v)
1431 {
1432 	struct epoch_tracker et;
1433 	struct pf_send_head queue;
1434 	struct pf_send_entry *pfse, *next;
1435 
1436 	CURVNET_SET((struct vnet *)v);
1437 
1438 	PF_SENDQ_LOCK();
1439 	queue = V_pf_sendqueue;
1440 	STAILQ_INIT(&V_pf_sendqueue);
1441 	PF_SENDQ_UNLOCK();
1442 
1443 	NET_EPOCH_ENTER(et);
1444 
1445 	STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1446 		switch (pfse->pfse_type) {
1447 #ifdef INET
1448 		case PFSE_IP:
1449 			ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1450 			break;
1451 		case PFSE_ICMP:
1452 			icmp_error(pfse->pfse_m, pfse->icmpopts.type,
1453 			    pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
1454 			break;
1455 #endif /* INET */
1456 #ifdef INET6
1457 		case PFSE_IP6:
1458 			ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1459 			    NULL);
1460 			break;
1461 		case PFSE_ICMP6:
1462 			icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
1463 			    pfse->icmpopts.code, pfse->icmpopts.mtu);
1464 			break;
1465 #endif /* INET6 */
1466 		default:
1467 			panic("%s: unknown type", __func__);
1468 		}
1469 		free(pfse, M_PFTEMP);
1470 	}
1471 	NET_EPOCH_EXIT(et);
1472 	CURVNET_RESTORE();
1473 }
1474 
1475 void
1476 pf_purge_thread(void *unused __unused)
1477 {
1478 	VNET_ITERATOR_DECL(vnet_iter);
1479 
1480 	sx_xlock(&pf_end_lock);
1481 	while (pf_end_threads == 0) {
1482 		sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", hz / 10);
1483 
1484 		VNET_LIST_RLOCK();
1485 		VNET_FOREACH(vnet_iter) {
1486 			CURVNET_SET(vnet_iter);
1487 
1488 
1489 			/* Wait until V_pf_default_rule is initialized. */
1490 			if (V_pf_vnet_active == 0) {
1491 				CURVNET_RESTORE();
1492 				continue;
1493 			}
1494 
1495 			/*
1496 			 *  Process 1/interval fraction of the state
1497 			 * table every run.
1498 			 */
1499 			V_pf_purge_idx =
1500 			    pf_purge_expired_states(V_pf_purge_idx, pf_hashmask /
1501 			    (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1502 
1503 			/*
1504 			 * Purge other expired types every
1505 			 * PFTM_INTERVAL seconds.
1506 			 */
1507 			if (V_pf_purge_idx == 0) {
1508 				/*
1509 				 * Order is important:
1510 				 * - states and src nodes reference rules
1511 				 * - states and rules reference kifs
1512 				 */
1513 				pf_purge_expired_fragments();
1514 				pf_purge_expired_src_nodes();
1515 				pf_purge_unlinked_rules();
1516 				pfi_kif_purge();
1517 			}
1518 			CURVNET_RESTORE();
1519 		}
1520 		VNET_LIST_RUNLOCK();
1521 	}
1522 
1523 	pf_end_threads++;
1524 	sx_xunlock(&pf_end_lock);
1525 	kproc_exit(0);
1526 }
1527 
1528 void
1529 pf_unload_vnet_purge(void)
1530 {
1531 
1532 	/*
1533 	 * To cleanse up all kifs and rules we need
1534 	 * two runs: first one clears reference flags,
1535 	 * then pf_purge_expired_states() doesn't
1536 	 * raise them, and then second run frees.
1537 	 */
1538 	pf_purge_unlinked_rules();
1539 	pfi_kif_purge();
1540 
1541 	/*
1542 	 * Now purge everything.
1543 	 */
1544 	pf_purge_expired_states(0, pf_hashmask);
1545 	pf_purge_fragments(UINT_MAX);
1546 	pf_purge_expired_src_nodes();
1547 
1548 	/*
1549 	 * Now all kifs & rules should be unreferenced,
1550 	 * thus should be successfully freed.
1551 	 */
1552 	pf_purge_unlinked_rules();
1553 	pfi_kif_purge();
1554 }
1555 
1556 
1557 u_int32_t
1558 pf_state_expires(const struct pf_state *state)
1559 {
1560 	u_int32_t	timeout;
1561 	u_int32_t	start;
1562 	u_int32_t	end;
1563 	u_int32_t	states;
1564 
1565 	/* handle all PFTM_* > PFTM_MAX here */
1566 	if (state->timeout == PFTM_PURGE)
1567 		return (time_uptime);
1568 	KASSERT(state->timeout != PFTM_UNLINKED,
1569 	    ("pf_state_expires: timeout == PFTM_UNLINKED"));
1570 	KASSERT((state->timeout < PFTM_MAX),
1571 	    ("pf_state_expires: timeout > PFTM_MAX"));
1572 	timeout = state->rule.ptr->timeout[state->timeout];
1573 	if (!timeout)
1574 		timeout = V_pf_default_rule.timeout[state->timeout];
1575 	start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1576 	if (start && state->rule.ptr != &V_pf_default_rule) {
1577 		end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1578 		states = counter_u64_fetch(state->rule.ptr->states_cur);
1579 	} else {
1580 		start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1581 		end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1582 		states = V_pf_status.states;
1583 	}
1584 	if (end && states > start && start < end) {
1585 		if (states < end) {
1586 			timeout = (u_int64_t)timeout * (end - states) /
1587 			    (end - start);
1588 			return (state->expire + timeout);
1589 		}
1590 		else
1591 			return (time_uptime);
1592 	}
1593 	return (state->expire + timeout);
1594 }
1595 
1596 void
1597 pf_purge_expired_src_nodes()
1598 {
1599 	struct pf_src_node_list	 freelist;
1600 	struct pf_srchash	*sh;
1601 	struct pf_src_node	*cur, *next;
1602 	int i;
1603 
1604 	LIST_INIT(&freelist);
1605 	for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1606 	    PF_HASHROW_LOCK(sh);
1607 	    LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1608 		if (cur->states == 0 && cur->expire <= time_uptime) {
1609 			pf_unlink_src_node(cur);
1610 			LIST_INSERT_HEAD(&freelist, cur, entry);
1611 		} else if (cur->rule.ptr != NULL)
1612 			cur->rule.ptr->rule_flag |= PFRULE_REFS;
1613 	    PF_HASHROW_UNLOCK(sh);
1614 	}
1615 
1616 	pf_free_src_nodes(&freelist);
1617 
1618 	V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1619 }
1620 
1621 static void
1622 pf_src_tree_remove_state(struct pf_state *s)
1623 {
1624 	struct pf_src_node *sn;
1625 	struct pf_srchash *sh;
1626 	uint32_t timeout;
1627 
1628 	timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
1629 	    s->rule.ptr->timeout[PFTM_SRC_NODE] :
1630 	    V_pf_default_rule.timeout[PFTM_SRC_NODE];
1631 
1632 	if (s->src_node != NULL) {
1633 		sn = s->src_node;
1634 		sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1635 	    	PF_HASHROW_LOCK(sh);
1636 		if (s->src.tcp_est)
1637 			--sn->conn;
1638 		if (--sn->states == 0)
1639 			sn->expire = time_uptime + timeout;
1640 	    	PF_HASHROW_UNLOCK(sh);
1641 	}
1642 	if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1643 		sn = s->nat_src_node;
1644 		sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1645 	    	PF_HASHROW_LOCK(sh);
1646 		if (--sn->states == 0)
1647 			sn->expire = time_uptime + timeout;
1648 	    	PF_HASHROW_UNLOCK(sh);
1649 	}
1650 	s->src_node = s->nat_src_node = NULL;
1651 }
1652 
1653 /*
1654  * Unlink and potentilly free a state. Function may be
1655  * called with ID hash row locked, but always returns
1656  * unlocked, since it needs to go through key hash locking.
1657  */
1658 int
1659 pf_unlink_state(struct pf_state *s, u_int flags)
1660 {
1661 	struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1662 
1663 	if ((flags & PF_ENTER_LOCKED) == 0)
1664 		PF_HASHROW_LOCK(ih);
1665 	else
1666 		PF_HASHROW_ASSERT(ih);
1667 
1668 	if (s->timeout == PFTM_UNLINKED) {
1669 		/*
1670 		 * State is being processed
1671 		 * by pf_unlink_state() in
1672 		 * an other thread.
1673 		 */
1674 		PF_HASHROW_UNLOCK(ih);
1675 		return (0);	/* XXXGL: undefined actually */
1676 	}
1677 
1678 	if (s->src.state == PF_TCPS_PROXY_DST) {
1679 		/* XXX wire key the right one? */
1680 		pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1681 		    &s->key[PF_SK_WIRE]->addr[1],
1682 		    &s->key[PF_SK_WIRE]->addr[0],
1683 		    s->key[PF_SK_WIRE]->port[1],
1684 		    s->key[PF_SK_WIRE]->port[0],
1685 		    s->src.seqhi, s->src.seqlo + 1,
1686 		    TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1687 	}
1688 
1689 	LIST_REMOVE(s, entry);
1690 	pf_src_tree_remove_state(s);
1691 
1692 	if (V_pfsync_delete_state_ptr != NULL)
1693 		V_pfsync_delete_state_ptr(s);
1694 
1695 	STATE_DEC_COUNTERS(s);
1696 
1697 	s->timeout = PFTM_UNLINKED;
1698 
1699 	PF_HASHROW_UNLOCK(ih);
1700 
1701 	pf_detach_state(s);
1702 	/* pf_state_insert() initialises refs to 2, so we can never release the
1703 	 * last reference here, only in pf_release_state(). */
1704 	(void)refcount_release(&s->refs);
1705 
1706 	return (pf_release_state(s));
1707 }
1708 
1709 void
1710 pf_free_state(struct pf_state *cur)
1711 {
1712 
1713 	KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1714 	KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1715 	    cur->timeout));
1716 
1717 	pf_normalize_tcp_cleanup(cur);
1718 	uma_zfree(V_pf_state_z, cur);
1719 	counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
1720 }
1721 
1722 /*
1723  * Called only from pf_purge_thread(), thus serialized.
1724  */
1725 static u_int
1726 pf_purge_expired_states(u_int i, int maxcheck)
1727 {
1728 	struct pf_idhash *ih;
1729 	struct pf_state *s;
1730 
1731 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1732 
1733 	/*
1734 	 * Go through hash and unlink states that expire now.
1735 	 */
1736 	while (maxcheck > 0) {
1737 
1738 		ih = &V_pf_idhash[i];
1739 
1740 		/* only take the lock if we expect to do work */
1741 		if (!LIST_EMPTY(&ih->states)) {
1742 relock:
1743 			PF_HASHROW_LOCK(ih);
1744 			LIST_FOREACH(s, &ih->states, entry) {
1745 				if (pf_state_expires(s) <= time_uptime) {
1746 					V_pf_status.states -=
1747 					    pf_unlink_state(s, PF_ENTER_LOCKED);
1748 					goto relock;
1749 				}
1750 				s->rule.ptr->rule_flag |= PFRULE_REFS;
1751 				if (s->nat_rule.ptr != NULL)
1752 					s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1753 				if (s->anchor.ptr != NULL)
1754 					s->anchor.ptr->rule_flag |= PFRULE_REFS;
1755 				s->kif->pfik_flags |= PFI_IFLAG_REFS;
1756 				if (s->rt_kif)
1757 					s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1758 			}
1759 			PF_HASHROW_UNLOCK(ih);
1760 		}
1761 
1762 		/* Return when we hit end of hash. */
1763 		if (++i > pf_hashmask) {
1764 			V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1765 			return (0);
1766 		}
1767 
1768 		maxcheck--;
1769 	}
1770 
1771 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1772 
1773 	return (i);
1774 }
1775 
1776 static void
1777 pf_purge_unlinked_rules()
1778 {
1779 	struct pf_rulequeue tmpq;
1780 	struct pf_rule *r, *r1;
1781 
1782 	/*
1783 	 * If we have overloading task pending, then we'd
1784 	 * better skip purging this time. There is a tiny
1785 	 * probability that overloading task references
1786 	 * an already unlinked rule.
1787 	 */
1788 	PF_OVERLOADQ_LOCK();
1789 	if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1790 		PF_OVERLOADQ_UNLOCK();
1791 		return;
1792 	}
1793 	PF_OVERLOADQ_UNLOCK();
1794 
1795 	/*
1796 	 * Do naive mark-and-sweep garbage collecting of old rules.
1797 	 * Reference flag is raised by pf_purge_expired_states()
1798 	 * and pf_purge_expired_src_nodes().
1799 	 *
1800 	 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1801 	 * use a temporary queue.
1802 	 */
1803 	TAILQ_INIT(&tmpq);
1804 	PF_UNLNKDRULES_LOCK();
1805 	TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1806 		if (!(r->rule_flag & PFRULE_REFS)) {
1807 			TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1808 			TAILQ_INSERT_TAIL(&tmpq, r, entries);
1809 		} else
1810 			r->rule_flag &= ~PFRULE_REFS;
1811 	}
1812 	PF_UNLNKDRULES_UNLOCK();
1813 
1814 	if (!TAILQ_EMPTY(&tmpq)) {
1815 		PF_RULES_WLOCK();
1816 		TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1817 			TAILQ_REMOVE(&tmpq, r, entries);
1818 			pf_free_rule(r);
1819 		}
1820 		PF_RULES_WUNLOCK();
1821 	}
1822 }
1823 
1824 void
1825 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1826 {
1827 	switch (af) {
1828 #ifdef INET
1829 	case AF_INET: {
1830 		u_int32_t a = ntohl(addr->addr32[0]);
1831 		printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1832 		    (a>>8)&255, a&255);
1833 		if (p) {
1834 			p = ntohs(p);
1835 			printf(":%u", p);
1836 		}
1837 		break;
1838 	}
1839 #endif /* INET */
1840 #ifdef INET6
1841 	case AF_INET6: {
1842 		u_int16_t b;
1843 		u_int8_t i, curstart, curend, maxstart, maxend;
1844 		curstart = curend = maxstart = maxend = 255;
1845 		for (i = 0; i < 8; i++) {
1846 			if (!addr->addr16[i]) {
1847 				if (curstart == 255)
1848 					curstart = i;
1849 				curend = i;
1850 			} else {
1851 				if ((curend - curstart) >
1852 				    (maxend - maxstart)) {
1853 					maxstart = curstart;
1854 					maxend = curend;
1855 				}
1856 				curstart = curend = 255;
1857 			}
1858 		}
1859 		if ((curend - curstart) >
1860 		    (maxend - maxstart)) {
1861 			maxstart = curstart;
1862 			maxend = curend;
1863 		}
1864 		for (i = 0; i < 8; i++) {
1865 			if (i >= maxstart && i <= maxend) {
1866 				if (i == 0)
1867 					printf(":");
1868 				if (i == maxend)
1869 					printf(":");
1870 			} else {
1871 				b = ntohs(addr->addr16[i]);
1872 				printf("%x", b);
1873 				if (i < 7)
1874 					printf(":");
1875 			}
1876 		}
1877 		if (p) {
1878 			p = ntohs(p);
1879 			printf("[%u]", p);
1880 		}
1881 		break;
1882 	}
1883 #endif /* INET6 */
1884 	}
1885 }
1886 
1887 void
1888 pf_print_state(struct pf_state *s)
1889 {
1890 	pf_print_state_parts(s, NULL, NULL);
1891 }
1892 
1893 static void
1894 pf_print_state_parts(struct pf_state *s,
1895     struct pf_state_key *skwp, struct pf_state_key *sksp)
1896 {
1897 	struct pf_state_key *skw, *sks;
1898 	u_int8_t proto, dir;
1899 
1900 	/* Do our best to fill these, but they're skipped if NULL */
1901 	skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1902 	sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1903 	proto = skw ? skw->proto : (sks ? sks->proto : 0);
1904 	dir = s ? s->direction : 0;
1905 
1906 	switch (proto) {
1907 	case IPPROTO_IPV4:
1908 		printf("IPv4");
1909 		break;
1910 	case IPPROTO_IPV6:
1911 		printf("IPv6");
1912 		break;
1913 	case IPPROTO_TCP:
1914 		printf("TCP");
1915 		break;
1916 	case IPPROTO_UDP:
1917 		printf("UDP");
1918 		break;
1919 	case IPPROTO_ICMP:
1920 		printf("ICMP");
1921 		break;
1922 	case IPPROTO_ICMPV6:
1923 		printf("ICMPv6");
1924 		break;
1925 	default:
1926 		printf("%u", proto);
1927 		break;
1928 	}
1929 	switch (dir) {
1930 	case PF_IN:
1931 		printf(" in");
1932 		break;
1933 	case PF_OUT:
1934 		printf(" out");
1935 		break;
1936 	}
1937 	if (skw) {
1938 		printf(" wire: ");
1939 		pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1940 		printf(" ");
1941 		pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1942 	}
1943 	if (sks) {
1944 		printf(" stack: ");
1945 		if (sks != skw) {
1946 			pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1947 			printf(" ");
1948 			pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1949 		} else
1950 			printf("-");
1951 	}
1952 	if (s) {
1953 		if (proto == IPPROTO_TCP) {
1954 			printf(" [lo=%u high=%u win=%u modulator=%u",
1955 			    s->src.seqlo, s->src.seqhi,
1956 			    s->src.max_win, s->src.seqdiff);
1957 			if (s->src.wscale && s->dst.wscale)
1958 				printf(" wscale=%u",
1959 				    s->src.wscale & PF_WSCALE_MASK);
1960 			printf("]");
1961 			printf(" [lo=%u high=%u win=%u modulator=%u",
1962 			    s->dst.seqlo, s->dst.seqhi,
1963 			    s->dst.max_win, s->dst.seqdiff);
1964 			if (s->src.wscale && s->dst.wscale)
1965 				printf(" wscale=%u",
1966 				s->dst.wscale & PF_WSCALE_MASK);
1967 			printf("]");
1968 		}
1969 		printf(" %u:%u", s->src.state, s->dst.state);
1970 	}
1971 }
1972 
1973 void
1974 pf_print_flags(u_int8_t f)
1975 {
1976 	if (f)
1977 		printf(" ");
1978 	if (f & TH_FIN)
1979 		printf("F");
1980 	if (f & TH_SYN)
1981 		printf("S");
1982 	if (f & TH_RST)
1983 		printf("R");
1984 	if (f & TH_PUSH)
1985 		printf("P");
1986 	if (f & TH_ACK)
1987 		printf("A");
1988 	if (f & TH_URG)
1989 		printf("U");
1990 	if (f & TH_ECE)
1991 		printf("E");
1992 	if (f & TH_CWR)
1993 		printf("W");
1994 }
1995 
1996 #define	PF_SET_SKIP_STEPS(i)					\
1997 	do {							\
1998 		while (head[i] != cur) {			\
1999 			head[i]->skip[i].ptr = cur;		\
2000 			head[i] = TAILQ_NEXT(head[i], entries);	\
2001 		}						\
2002 	} while (0)
2003 
2004 void
2005 pf_calc_skip_steps(struct pf_rulequeue *rules)
2006 {
2007 	struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
2008 	int i;
2009 
2010 	cur = TAILQ_FIRST(rules);
2011 	prev = cur;
2012 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2013 		head[i] = cur;
2014 	while (cur != NULL) {
2015 
2016 		if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
2017 			PF_SET_SKIP_STEPS(PF_SKIP_IFP);
2018 		if (cur->direction != prev->direction)
2019 			PF_SET_SKIP_STEPS(PF_SKIP_DIR);
2020 		if (cur->af != prev->af)
2021 			PF_SET_SKIP_STEPS(PF_SKIP_AF);
2022 		if (cur->proto != prev->proto)
2023 			PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
2024 		if (cur->src.neg != prev->src.neg ||
2025 		    pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
2026 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
2027 		if (cur->src.port[0] != prev->src.port[0] ||
2028 		    cur->src.port[1] != prev->src.port[1] ||
2029 		    cur->src.port_op != prev->src.port_op)
2030 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
2031 		if (cur->dst.neg != prev->dst.neg ||
2032 		    pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
2033 			PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
2034 		if (cur->dst.port[0] != prev->dst.port[0] ||
2035 		    cur->dst.port[1] != prev->dst.port[1] ||
2036 		    cur->dst.port_op != prev->dst.port_op)
2037 			PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
2038 
2039 		prev = cur;
2040 		cur = TAILQ_NEXT(cur, entries);
2041 	}
2042 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2043 		PF_SET_SKIP_STEPS(i);
2044 }
2045 
2046 static int
2047 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
2048 {
2049 	if (aw1->type != aw2->type)
2050 		return (1);
2051 	switch (aw1->type) {
2052 	case PF_ADDR_ADDRMASK:
2053 	case PF_ADDR_RANGE:
2054 		if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
2055 			return (1);
2056 		if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
2057 			return (1);
2058 		return (0);
2059 	case PF_ADDR_DYNIFTL:
2060 		return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
2061 	case PF_ADDR_NOROUTE:
2062 	case PF_ADDR_URPFFAILED:
2063 		return (0);
2064 	case PF_ADDR_TABLE:
2065 		return (aw1->p.tbl != aw2->p.tbl);
2066 	default:
2067 		printf("invalid address type: %d\n", aw1->type);
2068 		return (1);
2069 	}
2070 }
2071 
2072 /**
2073  * Checksum updates are a little complicated because the checksum in the TCP/UDP
2074  * header isn't always a full checksum. In some cases (i.e. output) it's a
2075  * pseudo-header checksum, which is a partial checksum over src/dst IP
2076  * addresses, protocol number and length.
2077  *
2078  * That means we have the following cases:
2079  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2080  *  	checksums, we need to update the checksum whenever we change anything.
2081  *  * Output (i.e. the checksum is a pseudo-header checksum):
2082  *  	x The field being updated is src/dst address or affects the length of
2083  *  	the packet. We need to update the pseudo-header checksum (note that this
2084  *  	checksum is not ones' complement).
2085  *  	x Some other field is being modified (e.g. src/dst port numbers): We
2086  *  	don't have to update anything.
2087  **/
2088 u_int16_t
2089 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2090 {
2091 	u_int32_t	l;
2092 
2093 	if (udp && !cksum)
2094 		return (0x0000);
2095 	l = cksum + old - new;
2096 	l = (l >> 16) + (l & 65535);
2097 	l = l & 65535;
2098 	if (udp && !l)
2099 		return (0xFFFF);
2100 	return (l);
2101 }
2102 
2103 u_int16_t
2104 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2105         u_int16_t new, u_int8_t udp)
2106 {
2107 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2108 		return (cksum);
2109 
2110 	return (pf_cksum_fixup(cksum, old, new, udp));
2111 }
2112 
2113 static void
2114 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2115         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2116         sa_family_t af)
2117 {
2118 	struct pf_addr	ao;
2119 	u_int16_t	po = *p;
2120 
2121 	PF_ACPY(&ao, a, af);
2122 	PF_ACPY(a, an, af);
2123 
2124 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2125 		*pc = ~*pc;
2126 
2127 	*p = pn;
2128 
2129 	switch (af) {
2130 #ifdef INET
2131 	case AF_INET:
2132 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2133 		    ao.addr16[0], an->addr16[0], 0),
2134 		    ao.addr16[1], an->addr16[1], 0);
2135 		*p = pn;
2136 
2137 		*pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2138 		    ao.addr16[0], an->addr16[0], u),
2139 		    ao.addr16[1], an->addr16[1], u);
2140 
2141 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2142 		break;
2143 #endif /* INET */
2144 #ifdef INET6
2145 	case AF_INET6:
2146 		*pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2147 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2148 		    pf_cksum_fixup(pf_cksum_fixup(*pc,
2149 		    ao.addr16[0], an->addr16[0], u),
2150 		    ao.addr16[1], an->addr16[1], u),
2151 		    ao.addr16[2], an->addr16[2], u),
2152 		    ao.addr16[3], an->addr16[3], u),
2153 		    ao.addr16[4], an->addr16[4], u),
2154 		    ao.addr16[5], an->addr16[5], u),
2155 		    ao.addr16[6], an->addr16[6], u),
2156 		    ao.addr16[7], an->addr16[7], u);
2157 
2158 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2159 		break;
2160 #endif /* INET6 */
2161 	}
2162 
2163 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2164 	    CSUM_DELAY_DATA_IPV6)) {
2165 		*pc = ~*pc;
2166 		if (! *pc)
2167 			*pc = 0xffff;
2168 	}
2169 }
2170 
2171 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2172 void
2173 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2174 {
2175 	u_int32_t	ao;
2176 
2177 	memcpy(&ao, a, sizeof(ao));
2178 	memcpy(a, &an, sizeof(u_int32_t));
2179 	*c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2180 	    ao % 65536, an % 65536, u);
2181 }
2182 
2183 void
2184 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2185 {
2186 	u_int32_t	ao;
2187 
2188 	memcpy(&ao, a, sizeof(ao));
2189 	memcpy(a, &an, sizeof(u_int32_t));
2190 
2191 	*c = pf_proto_cksum_fixup(m,
2192 	    pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2193 	    ao % 65536, an % 65536, udp);
2194 }
2195 
2196 #ifdef INET6
2197 static void
2198 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2199 {
2200 	struct pf_addr	ao;
2201 
2202 	PF_ACPY(&ao, a, AF_INET6);
2203 	PF_ACPY(a, an, AF_INET6);
2204 
2205 	*c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2206 	    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2207 	    pf_cksum_fixup(pf_cksum_fixup(*c,
2208 	    ao.addr16[0], an->addr16[0], u),
2209 	    ao.addr16[1], an->addr16[1], u),
2210 	    ao.addr16[2], an->addr16[2], u),
2211 	    ao.addr16[3], an->addr16[3], u),
2212 	    ao.addr16[4], an->addr16[4], u),
2213 	    ao.addr16[5], an->addr16[5], u),
2214 	    ao.addr16[6], an->addr16[6], u),
2215 	    ao.addr16[7], an->addr16[7], u);
2216 }
2217 #endif /* INET6 */
2218 
2219 static void
2220 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2221     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2222     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2223 {
2224 	struct pf_addr	oia, ooa;
2225 
2226 	PF_ACPY(&oia, ia, af);
2227 	if (oa)
2228 		PF_ACPY(&ooa, oa, af);
2229 
2230 	/* Change inner protocol port, fix inner protocol checksum. */
2231 	if (ip != NULL) {
2232 		u_int16_t	oip = *ip;
2233 		u_int32_t	opc;
2234 
2235 		if (pc != NULL)
2236 			opc = *pc;
2237 		*ip = np;
2238 		if (pc != NULL)
2239 			*pc = pf_cksum_fixup(*pc, oip, *ip, u);
2240 		*ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2241 		if (pc != NULL)
2242 			*ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2243 	}
2244 	/* Change inner ip address, fix inner ip and icmp checksums. */
2245 	PF_ACPY(ia, na, af);
2246 	switch (af) {
2247 #ifdef INET
2248 	case AF_INET: {
2249 		u_int32_t	 oh2c = *h2c;
2250 
2251 		*h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2252 		    oia.addr16[0], ia->addr16[0], 0),
2253 		    oia.addr16[1], ia->addr16[1], 0);
2254 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2255 		    oia.addr16[0], ia->addr16[0], 0),
2256 		    oia.addr16[1], ia->addr16[1], 0);
2257 		*ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2258 		break;
2259 	}
2260 #endif /* INET */
2261 #ifdef INET6
2262 	case AF_INET6:
2263 		*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2264 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2265 		    pf_cksum_fixup(pf_cksum_fixup(*ic,
2266 		    oia.addr16[0], ia->addr16[0], u),
2267 		    oia.addr16[1], ia->addr16[1], u),
2268 		    oia.addr16[2], ia->addr16[2], u),
2269 		    oia.addr16[3], ia->addr16[3], u),
2270 		    oia.addr16[4], ia->addr16[4], u),
2271 		    oia.addr16[5], ia->addr16[5], u),
2272 		    oia.addr16[6], ia->addr16[6], u),
2273 		    oia.addr16[7], ia->addr16[7], u);
2274 		break;
2275 #endif /* INET6 */
2276 	}
2277 	/* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2278 	if (oa) {
2279 		PF_ACPY(oa, na, af);
2280 		switch (af) {
2281 #ifdef INET
2282 		case AF_INET:
2283 			*hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2284 			    ooa.addr16[0], oa->addr16[0], 0),
2285 			    ooa.addr16[1], oa->addr16[1], 0);
2286 			break;
2287 #endif /* INET */
2288 #ifdef INET6
2289 		case AF_INET6:
2290 			*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2291 			    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2292 			    pf_cksum_fixup(pf_cksum_fixup(*ic,
2293 			    ooa.addr16[0], oa->addr16[0], u),
2294 			    ooa.addr16[1], oa->addr16[1], u),
2295 			    ooa.addr16[2], oa->addr16[2], u),
2296 			    ooa.addr16[3], oa->addr16[3], u),
2297 			    ooa.addr16[4], oa->addr16[4], u),
2298 			    ooa.addr16[5], oa->addr16[5], u),
2299 			    ooa.addr16[6], oa->addr16[6], u),
2300 			    ooa.addr16[7], oa->addr16[7], u);
2301 			break;
2302 #endif /* INET6 */
2303 		}
2304 	}
2305 }
2306 
2307 
2308 /*
2309  * Need to modulate the sequence numbers in the TCP SACK option
2310  * (credits to Krzysztof Pfaff for report and patch)
2311  */
2312 static int
2313 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2314     struct tcphdr *th, struct pf_state_peer *dst)
2315 {
2316 	int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2317 	u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2318 	int copyback = 0, i, olen;
2319 	struct sackblk sack;
2320 
2321 #define	TCPOLEN_SACKLEN	(TCPOLEN_SACK + 2)
2322 	if (hlen < TCPOLEN_SACKLEN ||
2323 	    !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2324 		return 0;
2325 
2326 	while (hlen >= TCPOLEN_SACKLEN) {
2327 		olen = opt[1];
2328 		switch (*opt) {
2329 		case TCPOPT_EOL:	/* FALLTHROUGH */
2330 		case TCPOPT_NOP:
2331 			opt++;
2332 			hlen--;
2333 			break;
2334 		case TCPOPT_SACK:
2335 			if (olen > hlen)
2336 				olen = hlen;
2337 			if (olen >= TCPOLEN_SACKLEN) {
2338 				for (i = 2; i + TCPOLEN_SACK <= olen;
2339 				    i += TCPOLEN_SACK) {
2340 					memcpy(&sack, &opt[i], sizeof(sack));
2341 					pf_change_proto_a(m, &sack.start, &th->th_sum,
2342 					    htonl(ntohl(sack.start) - dst->seqdiff), 0);
2343 					pf_change_proto_a(m, &sack.end, &th->th_sum,
2344 					    htonl(ntohl(sack.end) - dst->seqdiff), 0);
2345 					memcpy(&opt[i], &sack, sizeof(sack));
2346 				}
2347 				copyback = 1;
2348 			}
2349 			/* FALLTHROUGH */
2350 		default:
2351 			if (olen < 2)
2352 				olen = 2;
2353 			hlen -= olen;
2354 			opt += olen;
2355 		}
2356 	}
2357 
2358 	if (copyback)
2359 		m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2360 	return (copyback);
2361 }
2362 
2363 static void
2364 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2365     const struct pf_addr *saddr, const struct pf_addr *daddr,
2366     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2367     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2368     u_int16_t rtag, struct ifnet *ifp)
2369 {
2370 	struct pf_send_entry *pfse;
2371 	struct mbuf	*m;
2372 	int		 len, tlen;
2373 #ifdef INET
2374 	struct ip	*h = NULL;
2375 #endif /* INET */
2376 #ifdef INET6
2377 	struct ip6_hdr	*h6 = NULL;
2378 #endif /* INET6 */
2379 	struct tcphdr	*th;
2380 	char		*opt;
2381 	struct pf_mtag  *pf_mtag;
2382 
2383 	len = 0;
2384 	th = NULL;
2385 
2386 	/* maximum segment size tcp option */
2387 	tlen = sizeof(struct tcphdr);
2388 	if (mss)
2389 		tlen += 4;
2390 
2391 	switch (af) {
2392 #ifdef INET
2393 	case AF_INET:
2394 		len = sizeof(struct ip) + tlen;
2395 		break;
2396 #endif /* INET */
2397 #ifdef INET6
2398 	case AF_INET6:
2399 		len = sizeof(struct ip6_hdr) + tlen;
2400 		break;
2401 #endif /* INET6 */
2402 	default:
2403 		panic("%s: unsupported af %d", __func__, af);
2404 	}
2405 
2406 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
2407 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2408 	if (pfse == NULL)
2409 		return;
2410 	m = m_gethdr(M_NOWAIT, MT_DATA);
2411 	if (m == NULL) {
2412 		free(pfse, M_PFTEMP);
2413 		return;
2414 	}
2415 #ifdef MAC
2416 	mac_netinet_firewall_send(m);
2417 #endif
2418 	if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2419 		free(pfse, M_PFTEMP);
2420 		m_freem(m);
2421 		return;
2422 	}
2423 	if (tag)
2424 		m->m_flags |= M_SKIP_FIREWALL;
2425 	pf_mtag->tag = rtag;
2426 
2427 	if (r != NULL && r->rtableid >= 0)
2428 		M_SETFIB(m, r->rtableid);
2429 
2430 #ifdef ALTQ
2431 	if (r != NULL && r->qid) {
2432 		pf_mtag->qid = r->qid;
2433 
2434 		/* add hints for ecn */
2435 		pf_mtag->hdr = mtod(m, struct ip *);
2436 	}
2437 #endif /* ALTQ */
2438 	m->m_data += max_linkhdr;
2439 	m->m_pkthdr.len = m->m_len = len;
2440 	m->m_pkthdr.rcvif = NULL;
2441 	bzero(m->m_data, len);
2442 	switch (af) {
2443 #ifdef INET
2444 	case AF_INET:
2445 		h = mtod(m, struct ip *);
2446 
2447 		/* IP header fields included in the TCP checksum */
2448 		h->ip_p = IPPROTO_TCP;
2449 		h->ip_len = htons(tlen);
2450 		h->ip_src.s_addr = saddr->v4.s_addr;
2451 		h->ip_dst.s_addr = daddr->v4.s_addr;
2452 
2453 		th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2454 		break;
2455 #endif /* INET */
2456 #ifdef INET6
2457 	case AF_INET6:
2458 		h6 = mtod(m, struct ip6_hdr *);
2459 
2460 		/* IP header fields included in the TCP checksum */
2461 		h6->ip6_nxt = IPPROTO_TCP;
2462 		h6->ip6_plen = htons(tlen);
2463 		memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2464 		memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2465 
2466 		th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2467 		break;
2468 #endif /* INET6 */
2469 	}
2470 
2471 	/* TCP header */
2472 	th->th_sport = sport;
2473 	th->th_dport = dport;
2474 	th->th_seq = htonl(seq);
2475 	th->th_ack = htonl(ack);
2476 	th->th_off = tlen >> 2;
2477 	th->th_flags = flags;
2478 	th->th_win = htons(win);
2479 
2480 	if (mss) {
2481 		opt = (char *)(th + 1);
2482 		opt[0] = TCPOPT_MAXSEG;
2483 		opt[1] = 4;
2484 		HTONS(mss);
2485 		bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2486 	}
2487 
2488 	switch (af) {
2489 #ifdef INET
2490 	case AF_INET:
2491 		/* TCP checksum */
2492 		th->th_sum = in_cksum(m, len);
2493 
2494 		/* Finish the IP header */
2495 		h->ip_v = 4;
2496 		h->ip_hl = sizeof(*h) >> 2;
2497 		h->ip_tos = IPTOS_LOWDELAY;
2498 		h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2499 		h->ip_len = htons(len);
2500 		h->ip_ttl = ttl ? ttl : V_ip_defttl;
2501 		h->ip_sum = 0;
2502 
2503 		pfse->pfse_type = PFSE_IP;
2504 		break;
2505 #endif /* INET */
2506 #ifdef INET6
2507 	case AF_INET6:
2508 		/* TCP checksum */
2509 		th->th_sum = in6_cksum(m, IPPROTO_TCP,
2510 		    sizeof(struct ip6_hdr), tlen);
2511 
2512 		h6->ip6_vfc |= IPV6_VERSION;
2513 		h6->ip6_hlim = IPV6_DEFHLIM;
2514 
2515 		pfse->pfse_type = PFSE_IP6;
2516 		break;
2517 #endif /* INET6 */
2518 	}
2519 	pfse->pfse_m = m;
2520 	pf_send(pfse);
2521 }
2522 
2523 static void
2524 pf_return(struct pf_rule *r, struct pf_rule *nr, struct pf_pdesc *pd,
2525     struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th,
2526     struct pfi_kif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
2527     u_short *reason)
2528 {
2529 	struct pf_addr	* const saddr = pd->src;
2530 	struct pf_addr	* const daddr = pd->dst;
2531 	sa_family_t	 af = pd->af;
2532 
2533 	/* undo NAT changes, if they have taken place */
2534 	if (nr != NULL) {
2535 		PF_ACPY(saddr, &sk->addr[pd->sidx], af);
2536 		PF_ACPY(daddr, &sk->addr[pd->didx], af);
2537 		if (pd->sport)
2538 			*pd->sport = sk->port[pd->sidx];
2539 		if (pd->dport)
2540 			*pd->dport = sk->port[pd->didx];
2541 		if (pd->proto_sum)
2542 			*pd->proto_sum = bproto_sum;
2543 		if (pd->ip_sum)
2544 			*pd->ip_sum = bip_sum;
2545 		m_copyback(m, off, hdrlen, pd->hdr.any);
2546 	}
2547 	if (pd->proto == IPPROTO_TCP &&
2548 	    ((r->rule_flag & PFRULE_RETURNRST) ||
2549 	    (r->rule_flag & PFRULE_RETURN)) &&
2550 	    !(th->th_flags & TH_RST)) {
2551 		u_int32_t	 ack = ntohl(th->th_seq) + pd->p_len;
2552 		int		 len = 0;
2553 #ifdef INET
2554 		struct ip	*h4;
2555 #endif
2556 #ifdef INET6
2557 		struct ip6_hdr	*h6;
2558 #endif
2559 
2560 		switch (af) {
2561 #ifdef INET
2562 		case AF_INET:
2563 			h4 = mtod(m, struct ip *);
2564 			len = ntohs(h4->ip_len) - off;
2565 			break;
2566 #endif
2567 #ifdef INET6
2568 		case AF_INET6:
2569 			h6 = mtod(m, struct ip6_hdr *);
2570 			len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
2571 			break;
2572 #endif
2573 		}
2574 
2575 		if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
2576 			REASON_SET(reason, PFRES_PROTCKSUM);
2577 		else {
2578 			if (th->th_flags & TH_SYN)
2579 				ack++;
2580 			if (th->th_flags & TH_FIN)
2581 				ack++;
2582 			pf_send_tcp(m, r, af, pd->dst,
2583 				pd->src, th->th_dport, th->th_sport,
2584 				ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
2585 				r->return_ttl, 1, 0, kif->pfik_ifp);
2586 		}
2587 	} else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
2588 		r->return_icmp)
2589 		pf_send_icmp(m, r->return_icmp >> 8,
2590 			r->return_icmp & 255, af, r);
2591 	else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
2592 		r->return_icmp6)
2593 		pf_send_icmp(m, r->return_icmp6 >> 8,
2594 			r->return_icmp6 & 255, af, r);
2595 }
2596 
2597 
2598 static int
2599 pf_ieee8021q_setpcp(struct mbuf *m, u_int8_t prio)
2600 {
2601 	struct m_tag *mtag;
2602 
2603 	KASSERT(prio <= PF_PRIO_MAX,
2604 	    ("%s with invalid pcp", __func__));
2605 
2606 	mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_OUT, NULL);
2607 	if (mtag == NULL) {
2608 		mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_OUT,
2609 		    sizeof(uint8_t), M_NOWAIT);
2610 		if (mtag == NULL)
2611 			return (ENOMEM);
2612 		m_tag_prepend(m, mtag);
2613 	}
2614 
2615 	*(uint8_t *)(mtag + 1) = prio;
2616 	return (0);
2617 }
2618 
2619 static int
2620 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
2621 {
2622 	struct m_tag *mtag;
2623 	u_int8_t mpcp;
2624 
2625 	mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
2626 	if (mtag == NULL)
2627 		return (0);
2628 
2629 	if (prio == PF_PRIO_ZERO)
2630 		prio = 0;
2631 
2632 	mpcp = *(uint8_t *)(mtag + 1);
2633 
2634 	return (mpcp == prio);
2635 }
2636 
2637 static void
2638 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2639     struct pf_rule *r)
2640 {
2641 	struct pf_send_entry *pfse;
2642 	struct mbuf *m0;
2643 	struct pf_mtag *pf_mtag;
2644 
2645 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
2646 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2647 	if (pfse == NULL)
2648 		return;
2649 
2650 	if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2651 		free(pfse, M_PFTEMP);
2652 		return;
2653 	}
2654 
2655 	if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2656 		free(pfse, M_PFTEMP);
2657 		return;
2658 	}
2659 	/* XXX: revisit */
2660 	m0->m_flags |= M_SKIP_FIREWALL;
2661 
2662 	if (r->rtableid >= 0)
2663 		M_SETFIB(m0, r->rtableid);
2664 
2665 #ifdef ALTQ
2666 	if (r->qid) {
2667 		pf_mtag->qid = r->qid;
2668 		/* add hints for ecn */
2669 		pf_mtag->hdr = mtod(m0, struct ip *);
2670 	}
2671 #endif /* ALTQ */
2672 
2673 	switch (af) {
2674 #ifdef INET
2675 	case AF_INET:
2676 		pfse->pfse_type = PFSE_ICMP;
2677 		break;
2678 #endif /* INET */
2679 #ifdef INET6
2680 	case AF_INET6:
2681 		pfse->pfse_type = PFSE_ICMP6;
2682 		break;
2683 #endif /* INET6 */
2684 	}
2685 	pfse->pfse_m = m0;
2686 	pfse->icmpopts.type = type;
2687 	pfse->icmpopts.code = code;
2688 	pf_send(pfse);
2689 }
2690 
2691 /*
2692  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2693  * If n is 0, they match if they are equal. If n is != 0, they match if they
2694  * are different.
2695  */
2696 int
2697 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2698     struct pf_addr *b, sa_family_t af)
2699 {
2700 	int	match = 0;
2701 
2702 	switch (af) {
2703 #ifdef INET
2704 	case AF_INET:
2705 		if ((a->addr32[0] & m->addr32[0]) ==
2706 		    (b->addr32[0] & m->addr32[0]))
2707 			match++;
2708 		break;
2709 #endif /* INET */
2710 #ifdef INET6
2711 	case AF_INET6:
2712 		if (((a->addr32[0] & m->addr32[0]) ==
2713 		     (b->addr32[0] & m->addr32[0])) &&
2714 		    ((a->addr32[1] & m->addr32[1]) ==
2715 		     (b->addr32[1] & m->addr32[1])) &&
2716 		    ((a->addr32[2] & m->addr32[2]) ==
2717 		     (b->addr32[2] & m->addr32[2])) &&
2718 		    ((a->addr32[3] & m->addr32[3]) ==
2719 		     (b->addr32[3] & m->addr32[3])))
2720 			match++;
2721 		break;
2722 #endif /* INET6 */
2723 	}
2724 	if (match) {
2725 		if (n)
2726 			return (0);
2727 		else
2728 			return (1);
2729 	} else {
2730 		if (n)
2731 			return (1);
2732 		else
2733 			return (0);
2734 	}
2735 }
2736 
2737 /*
2738  * Return 1 if b <= a <= e, otherwise return 0.
2739  */
2740 int
2741 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2742     struct pf_addr *a, sa_family_t af)
2743 {
2744 	switch (af) {
2745 #ifdef INET
2746 	case AF_INET:
2747 		if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
2748 		    (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
2749 			return (0);
2750 		break;
2751 #endif /* INET */
2752 #ifdef INET6
2753 	case AF_INET6: {
2754 		int	i;
2755 
2756 		/* check a >= b */
2757 		for (i = 0; i < 4; ++i)
2758 			if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
2759 				break;
2760 			else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
2761 				return (0);
2762 		/* check a <= e */
2763 		for (i = 0; i < 4; ++i)
2764 			if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
2765 				break;
2766 			else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
2767 				return (0);
2768 		break;
2769 	}
2770 #endif /* INET6 */
2771 	}
2772 	return (1);
2773 }
2774 
2775 static int
2776 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2777 {
2778 	switch (op) {
2779 	case PF_OP_IRG:
2780 		return ((p > a1) && (p < a2));
2781 	case PF_OP_XRG:
2782 		return ((p < a1) || (p > a2));
2783 	case PF_OP_RRG:
2784 		return ((p >= a1) && (p <= a2));
2785 	case PF_OP_EQ:
2786 		return (p == a1);
2787 	case PF_OP_NE:
2788 		return (p != a1);
2789 	case PF_OP_LT:
2790 		return (p < a1);
2791 	case PF_OP_LE:
2792 		return (p <= a1);
2793 	case PF_OP_GT:
2794 		return (p > a1);
2795 	case PF_OP_GE:
2796 		return (p >= a1);
2797 	}
2798 	return (0); /* never reached */
2799 }
2800 
2801 int
2802 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2803 {
2804 	NTOHS(a1);
2805 	NTOHS(a2);
2806 	NTOHS(p);
2807 	return (pf_match(op, a1, a2, p));
2808 }
2809 
2810 static int
2811 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2812 {
2813 	if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2814 		return (0);
2815 	return (pf_match(op, a1, a2, u));
2816 }
2817 
2818 static int
2819 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2820 {
2821 	if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2822 		return (0);
2823 	return (pf_match(op, a1, a2, g));
2824 }
2825 
2826 int
2827 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2828 {
2829 	if (*tag == -1)
2830 		*tag = mtag;
2831 
2832 	return ((!r->match_tag_not && r->match_tag == *tag) ||
2833 	    (r->match_tag_not && r->match_tag != *tag));
2834 }
2835 
2836 int
2837 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2838 {
2839 
2840 	KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2841 
2842 	if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2843 		return (ENOMEM);
2844 
2845 	pd->pf_mtag->tag = tag;
2846 
2847 	return (0);
2848 }
2849 
2850 #define	PF_ANCHOR_STACKSIZE	32
2851 struct pf_anchor_stackframe {
2852 	struct pf_ruleset	*rs;
2853 	struct pf_rule		*r;	/* XXX: + match bit */
2854 	struct pf_anchor	*child;
2855 };
2856 
2857 /*
2858  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2859  */
2860 #define	PF_ANCHORSTACK_MATCH	0x00000001
2861 #define	PF_ANCHORSTACK_MASK	(PF_ANCHORSTACK_MATCH)
2862 
2863 #define	PF_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2864 #define	PF_ANCHOR_RULE(f)	(struct pf_rule *)			\
2865 				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2866 #define	PF_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 			\
2867 				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2868 } while (0)
2869 
2870 void
2871 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2872     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2873     int *match)
2874 {
2875 	struct pf_anchor_stackframe	*f;
2876 
2877 	PF_RULES_RASSERT();
2878 
2879 	if (match)
2880 		*match = 0;
2881 	if (*depth >= PF_ANCHOR_STACKSIZE) {
2882 		printf("%s: anchor stack overflow on %s\n",
2883 		    __func__, (*r)->anchor->name);
2884 		*r = TAILQ_NEXT(*r, entries);
2885 		return;
2886 	} else if (*depth == 0 && a != NULL)
2887 		*a = *r;
2888 	f = stack + (*depth)++;
2889 	f->rs = *rs;
2890 	f->r = *r;
2891 	if ((*r)->anchor_wildcard) {
2892 		struct pf_anchor_node *parent = &(*r)->anchor->children;
2893 
2894 		if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2895 			*r = NULL;
2896 			return;
2897 		}
2898 		*rs = &f->child->ruleset;
2899 	} else {
2900 		f->child = NULL;
2901 		*rs = &(*r)->anchor->ruleset;
2902 	}
2903 	*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2904 }
2905 
2906 int
2907 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2908     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2909     int *match)
2910 {
2911 	struct pf_anchor_stackframe	*f;
2912 	struct pf_rule *fr;
2913 	int quick = 0;
2914 
2915 	PF_RULES_RASSERT();
2916 
2917 	do {
2918 		if (*depth <= 0)
2919 			break;
2920 		f = stack + *depth - 1;
2921 		fr = PF_ANCHOR_RULE(f);
2922 		if (f->child != NULL) {
2923 			struct pf_anchor_node *parent;
2924 
2925 			/*
2926 			 * This block traverses through
2927 			 * a wildcard anchor.
2928 			 */
2929 			parent = &fr->anchor->children;
2930 			if (match != NULL && *match) {
2931 				/*
2932 				 * If any of "*" matched, then
2933 				 * "foo/ *" matched, mark frame
2934 				 * appropriately.
2935 				 */
2936 				PF_ANCHOR_SET_MATCH(f);
2937 				*match = 0;
2938 			}
2939 			f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2940 			if (f->child != NULL) {
2941 				*rs = &f->child->ruleset;
2942 				*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2943 				if (*r == NULL)
2944 					continue;
2945 				else
2946 					break;
2947 			}
2948 		}
2949 		(*depth)--;
2950 		if (*depth == 0 && a != NULL)
2951 			*a = NULL;
2952 		*rs = f->rs;
2953 		if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2954 			quick = fr->quick;
2955 		*r = TAILQ_NEXT(fr, entries);
2956 	} while (*r == NULL);
2957 
2958 	return (quick);
2959 }
2960 
2961 #ifdef INET6
2962 void
2963 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2964     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2965 {
2966 	switch (af) {
2967 #ifdef INET
2968 	case AF_INET:
2969 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2970 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2971 		break;
2972 #endif /* INET */
2973 	case AF_INET6:
2974 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2975 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2976 		naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2977 		((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2978 		naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2979 		((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2980 		naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2981 		((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2982 		break;
2983 	}
2984 }
2985 
2986 void
2987 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2988 {
2989 	switch (af) {
2990 #ifdef INET
2991 	case AF_INET:
2992 		addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2993 		break;
2994 #endif /* INET */
2995 	case AF_INET6:
2996 		if (addr->addr32[3] == 0xffffffff) {
2997 			addr->addr32[3] = 0;
2998 			if (addr->addr32[2] == 0xffffffff) {
2999 				addr->addr32[2] = 0;
3000 				if (addr->addr32[1] == 0xffffffff) {
3001 					addr->addr32[1] = 0;
3002 					addr->addr32[0] =
3003 					    htonl(ntohl(addr->addr32[0]) + 1);
3004 				} else
3005 					addr->addr32[1] =
3006 					    htonl(ntohl(addr->addr32[1]) + 1);
3007 			} else
3008 				addr->addr32[2] =
3009 				    htonl(ntohl(addr->addr32[2]) + 1);
3010 		} else
3011 			addr->addr32[3] =
3012 			    htonl(ntohl(addr->addr32[3]) + 1);
3013 		break;
3014 	}
3015 }
3016 #endif /* INET6 */
3017 
3018 int
3019 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
3020 {
3021 	struct pf_addr		*saddr, *daddr;
3022 	u_int16_t		 sport, dport;
3023 	struct inpcbinfo	*pi;
3024 	struct inpcb		*inp;
3025 
3026 	pd->lookup.uid = UID_MAX;
3027 	pd->lookup.gid = GID_MAX;
3028 
3029 	switch (pd->proto) {
3030 	case IPPROTO_TCP:
3031 		if (pd->hdr.tcp == NULL)
3032 			return (-1);
3033 		sport = pd->hdr.tcp->th_sport;
3034 		dport = pd->hdr.tcp->th_dport;
3035 		pi = &V_tcbinfo;
3036 		break;
3037 	case IPPROTO_UDP:
3038 		if (pd->hdr.udp == NULL)
3039 			return (-1);
3040 		sport = pd->hdr.udp->uh_sport;
3041 		dport = pd->hdr.udp->uh_dport;
3042 		pi = &V_udbinfo;
3043 		break;
3044 	default:
3045 		return (-1);
3046 	}
3047 	if (direction == PF_IN) {
3048 		saddr = pd->src;
3049 		daddr = pd->dst;
3050 	} else {
3051 		u_int16_t	p;
3052 
3053 		p = sport;
3054 		sport = dport;
3055 		dport = p;
3056 		saddr = pd->dst;
3057 		daddr = pd->src;
3058 	}
3059 	switch (pd->af) {
3060 #ifdef INET
3061 	case AF_INET:
3062 		inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
3063 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
3064 		if (inp == NULL) {
3065 			inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
3066 			   daddr->v4, dport, INPLOOKUP_WILDCARD |
3067 			   INPLOOKUP_RLOCKPCB, NULL, m);
3068 			if (inp == NULL)
3069 				return (-1);
3070 		}
3071 		break;
3072 #endif /* INET */
3073 #ifdef INET6
3074 	case AF_INET6:
3075 		inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
3076 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
3077 		if (inp == NULL) {
3078 			inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
3079 			    &daddr->v6, dport, INPLOOKUP_WILDCARD |
3080 			    INPLOOKUP_RLOCKPCB, NULL, m);
3081 			if (inp == NULL)
3082 				return (-1);
3083 		}
3084 		break;
3085 #endif /* INET6 */
3086 
3087 	default:
3088 		return (-1);
3089 	}
3090 	INP_RLOCK_ASSERT(inp);
3091 	pd->lookup.uid = inp->inp_cred->cr_uid;
3092 	pd->lookup.gid = inp->inp_cred->cr_groups[0];
3093 	INP_RUNLOCK(inp);
3094 
3095 	return (1);
3096 }
3097 
3098 static u_int8_t
3099 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3100 {
3101 	int		 hlen;
3102 	u_int8_t	 hdr[60];
3103 	u_int8_t	*opt, optlen;
3104 	u_int8_t	 wscale = 0;
3105 
3106 	hlen = th_off << 2;		/* hlen <= sizeof(hdr) */
3107 	if (hlen <= sizeof(struct tcphdr))
3108 		return (0);
3109 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3110 		return (0);
3111 	opt = hdr + sizeof(struct tcphdr);
3112 	hlen -= sizeof(struct tcphdr);
3113 	while (hlen >= 3) {
3114 		switch (*opt) {
3115 		case TCPOPT_EOL:
3116 		case TCPOPT_NOP:
3117 			++opt;
3118 			--hlen;
3119 			break;
3120 		case TCPOPT_WINDOW:
3121 			wscale = opt[2];
3122 			if (wscale > TCP_MAX_WINSHIFT)
3123 				wscale = TCP_MAX_WINSHIFT;
3124 			wscale |= PF_WSCALE_FLAG;
3125 			/* FALLTHROUGH */
3126 		default:
3127 			optlen = opt[1];
3128 			if (optlen < 2)
3129 				optlen = 2;
3130 			hlen -= optlen;
3131 			opt += optlen;
3132 			break;
3133 		}
3134 	}
3135 	return (wscale);
3136 }
3137 
3138 static u_int16_t
3139 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3140 {
3141 	int		 hlen;
3142 	u_int8_t	 hdr[60];
3143 	u_int8_t	*opt, optlen;
3144 	u_int16_t	 mss = V_tcp_mssdflt;
3145 
3146 	hlen = th_off << 2;	/* hlen <= sizeof(hdr) */
3147 	if (hlen <= sizeof(struct tcphdr))
3148 		return (0);
3149 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3150 		return (0);
3151 	opt = hdr + sizeof(struct tcphdr);
3152 	hlen -= sizeof(struct tcphdr);
3153 	while (hlen >= TCPOLEN_MAXSEG) {
3154 		switch (*opt) {
3155 		case TCPOPT_EOL:
3156 		case TCPOPT_NOP:
3157 			++opt;
3158 			--hlen;
3159 			break;
3160 		case TCPOPT_MAXSEG:
3161 			bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
3162 			NTOHS(mss);
3163 			/* FALLTHROUGH */
3164 		default:
3165 			optlen = opt[1];
3166 			if (optlen < 2)
3167 				optlen = 2;
3168 			hlen -= optlen;
3169 			opt += optlen;
3170 			break;
3171 		}
3172 	}
3173 	return (mss);
3174 }
3175 
3176 static u_int16_t
3177 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
3178 {
3179 #ifdef INET
3180 	struct nhop4_basic	nh4;
3181 #endif /* INET */
3182 #ifdef INET6
3183 	struct nhop6_basic	nh6;
3184 	struct in6_addr		dst6;
3185 	uint32_t		scopeid;
3186 #endif /* INET6 */
3187 	int			 hlen = 0;
3188 	uint16_t		 mss = 0;
3189 
3190 	switch (af) {
3191 #ifdef INET
3192 	case AF_INET:
3193 		hlen = sizeof(struct ip);
3194 		if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) == 0)
3195 			mss = nh4.nh_mtu - hlen - sizeof(struct tcphdr);
3196 		break;
3197 #endif /* INET */
3198 #ifdef INET6
3199 	case AF_INET6:
3200 		hlen = sizeof(struct ip6_hdr);
3201 		in6_splitscope(&addr->v6, &dst6, &scopeid);
3202 		if (fib6_lookup_nh_basic(rtableid, &dst6, scopeid, 0,0,&nh6)==0)
3203 			mss = nh6.nh_mtu - hlen - sizeof(struct tcphdr);
3204 		break;
3205 #endif /* INET6 */
3206 	}
3207 
3208 	mss = max(V_tcp_mssdflt, mss);
3209 	mss = min(mss, offer);
3210 	mss = max(mss, 64);		/* sanity - at least max opt space */
3211 	return (mss);
3212 }
3213 
3214 static u_int32_t
3215 pf_tcp_iss(struct pf_pdesc *pd)
3216 {
3217 	MD5_CTX ctx;
3218 	u_int32_t digest[4];
3219 
3220 	if (V_pf_tcp_secret_init == 0) {
3221 		arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
3222 		MD5Init(&V_pf_tcp_secret_ctx);
3223 		MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
3224 		    sizeof(V_pf_tcp_secret));
3225 		V_pf_tcp_secret_init = 1;
3226 	}
3227 
3228 	ctx = V_pf_tcp_secret_ctx;
3229 
3230 	MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
3231 	MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
3232 	if (pd->af == AF_INET6) {
3233 		MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3234 		MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3235 	} else {
3236 		MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3237 		MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3238 	}
3239 	MD5Final((u_char *)digest, &ctx);
3240 	V_pf_tcp_iss_off += 4096;
3241 #define	ISN_RANDOM_INCREMENT (4096 - 1)
3242 	return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
3243 	    V_pf_tcp_iss_off);
3244 #undef	ISN_RANDOM_INCREMENT
3245 }
3246 
3247 static int
3248 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
3249     struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
3250     struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
3251 {
3252 	struct pf_rule		*nr = NULL;
3253 	struct pf_addr		* const saddr = pd->src;
3254 	struct pf_addr		* const daddr = pd->dst;
3255 	sa_family_t		 af = pd->af;
3256 	struct pf_rule		*r, *a = NULL;
3257 	struct pf_ruleset	*ruleset = NULL;
3258 	struct pf_src_node	*nsn = NULL;
3259 	struct tcphdr		*th = pd->hdr.tcp;
3260 	struct pf_state_key	*sk = NULL, *nk = NULL;
3261 	u_short			 reason;
3262 	int			 rewrite = 0, hdrlen = 0;
3263 	int			 tag = -1, rtableid = -1;
3264 	int			 asd = 0;
3265 	int			 match = 0;
3266 	int			 state_icmp = 0;
3267 	u_int16_t		 sport = 0, dport = 0;
3268 	u_int16_t		 bproto_sum = 0, bip_sum = 0;
3269 	u_int8_t		 icmptype = 0, icmpcode = 0;
3270 	struct pf_anchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
3271 
3272 	PF_RULES_RASSERT();
3273 
3274 	if (inp != NULL) {
3275 		INP_LOCK_ASSERT(inp);
3276 		pd->lookup.uid = inp->inp_cred->cr_uid;
3277 		pd->lookup.gid = inp->inp_cred->cr_groups[0];
3278 		pd->lookup.done = 1;
3279 	}
3280 
3281 	switch (pd->proto) {
3282 	case IPPROTO_TCP:
3283 		sport = th->th_sport;
3284 		dport = th->th_dport;
3285 		hdrlen = sizeof(*th);
3286 		break;
3287 	case IPPROTO_UDP:
3288 		sport = pd->hdr.udp->uh_sport;
3289 		dport = pd->hdr.udp->uh_dport;
3290 		hdrlen = sizeof(*pd->hdr.udp);
3291 		break;
3292 #ifdef INET
3293 	case IPPROTO_ICMP:
3294 		if (pd->af != AF_INET)
3295 			break;
3296 		sport = dport = pd->hdr.icmp->icmp_id;
3297 		hdrlen = sizeof(*pd->hdr.icmp);
3298 		icmptype = pd->hdr.icmp->icmp_type;
3299 		icmpcode = pd->hdr.icmp->icmp_code;
3300 
3301 		if (icmptype == ICMP_UNREACH ||
3302 		    icmptype == ICMP_SOURCEQUENCH ||
3303 		    icmptype == ICMP_REDIRECT ||
3304 		    icmptype == ICMP_TIMXCEED ||
3305 		    icmptype == ICMP_PARAMPROB)
3306 			state_icmp++;
3307 		break;
3308 #endif /* INET */
3309 #ifdef INET6
3310 	case IPPROTO_ICMPV6:
3311 		if (af != AF_INET6)
3312 			break;
3313 		sport = dport = pd->hdr.icmp6->icmp6_id;
3314 		hdrlen = sizeof(*pd->hdr.icmp6);
3315 		icmptype = pd->hdr.icmp6->icmp6_type;
3316 		icmpcode = pd->hdr.icmp6->icmp6_code;
3317 
3318 		if (icmptype == ICMP6_DST_UNREACH ||
3319 		    icmptype == ICMP6_PACKET_TOO_BIG ||
3320 		    icmptype == ICMP6_TIME_EXCEEDED ||
3321 		    icmptype == ICMP6_PARAM_PROB)
3322 			state_icmp++;
3323 		break;
3324 #endif /* INET6 */
3325 	default:
3326 		sport = dport = hdrlen = 0;
3327 		break;
3328 	}
3329 
3330 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3331 
3332 	/* check packet for BINAT/NAT/RDR */
3333 	if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3334 	    &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3335 		KASSERT(sk != NULL, ("%s: null sk", __func__));
3336 		KASSERT(nk != NULL, ("%s: null nk", __func__));
3337 
3338 		if (pd->ip_sum)
3339 			bip_sum = *pd->ip_sum;
3340 
3341 		switch (pd->proto) {
3342 		case IPPROTO_TCP:
3343 			bproto_sum = th->th_sum;
3344 			pd->proto_sum = &th->th_sum;
3345 
3346 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3347 			    nk->port[pd->sidx] != sport) {
3348 				pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
3349 				    &th->th_sum, &nk->addr[pd->sidx],
3350 				    nk->port[pd->sidx], 0, af);
3351 				pd->sport = &th->th_sport;
3352 				sport = th->th_sport;
3353 			}
3354 
3355 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3356 			    nk->port[pd->didx] != dport) {
3357 				pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
3358 				    &th->th_sum, &nk->addr[pd->didx],
3359 				    nk->port[pd->didx], 0, af);
3360 				dport = th->th_dport;
3361 				pd->dport = &th->th_dport;
3362 			}
3363 			rewrite++;
3364 			break;
3365 		case IPPROTO_UDP:
3366 			bproto_sum = pd->hdr.udp->uh_sum;
3367 			pd->proto_sum = &pd->hdr.udp->uh_sum;
3368 
3369 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3370 			    nk->port[pd->sidx] != sport) {
3371 				pf_change_ap(m, saddr, &pd->hdr.udp->uh_sport,
3372 				    pd->ip_sum, &pd->hdr.udp->uh_sum,
3373 				    &nk->addr[pd->sidx],
3374 				    nk->port[pd->sidx], 1, af);
3375 				sport = pd->hdr.udp->uh_sport;
3376 				pd->sport = &pd->hdr.udp->uh_sport;
3377 			}
3378 
3379 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3380 			    nk->port[pd->didx] != dport) {
3381 				pf_change_ap(m, daddr, &pd->hdr.udp->uh_dport,
3382 				    pd->ip_sum, &pd->hdr.udp->uh_sum,
3383 				    &nk->addr[pd->didx],
3384 				    nk->port[pd->didx], 1, af);
3385 				dport = pd->hdr.udp->uh_dport;
3386 				pd->dport = &pd->hdr.udp->uh_dport;
3387 			}
3388 			rewrite++;
3389 			break;
3390 #ifdef INET
3391 		case IPPROTO_ICMP:
3392 			nk->port[0] = nk->port[1];
3393 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3394 				pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3395 				    nk->addr[pd->sidx].v4.s_addr, 0);
3396 
3397 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3398 				pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3399 				    nk->addr[pd->didx].v4.s_addr, 0);
3400 
3401 			if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3402 				pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3403 				    pd->hdr.icmp->icmp_cksum, sport,
3404 				    nk->port[1], 0);
3405 				pd->hdr.icmp->icmp_id = nk->port[1];
3406 				pd->sport = &pd->hdr.icmp->icmp_id;
3407 			}
3408 			m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3409 			break;
3410 #endif /* INET */
3411 #ifdef INET6
3412 		case IPPROTO_ICMPV6:
3413 			nk->port[0] = nk->port[1];
3414 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3415 				pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3416 				    &nk->addr[pd->sidx], 0);
3417 
3418 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3419 				pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3420 				    &nk->addr[pd->didx], 0);
3421 			rewrite++;
3422 			break;
3423 #endif /* INET */
3424 		default:
3425 			switch (af) {
3426 #ifdef INET
3427 			case AF_INET:
3428 				if (PF_ANEQ(saddr,
3429 				    &nk->addr[pd->sidx], AF_INET))
3430 					pf_change_a(&saddr->v4.s_addr,
3431 					    pd->ip_sum,
3432 					    nk->addr[pd->sidx].v4.s_addr, 0);
3433 
3434 				if (PF_ANEQ(daddr,
3435 				    &nk->addr[pd->didx], AF_INET))
3436 					pf_change_a(&daddr->v4.s_addr,
3437 					    pd->ip_sum,
3438 					    nk->addr[pd->didx].v4.s_addr, 0);
3439 				break;
3440 #endif /* INET */
3441 #ifdef INET6
3442 			case AF_INET6:
3443 				if (PF_ANEQ(saddr,
3444 				    &nk->addr[pd->sidx], AF_INET6))
3445 					PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3446 
3447 				if (PF_ANEQ(daddr,
3448 				    &nk->addr[pd->didx], AF_INET6))
3449 					PF_ACPY(daddr, &nk->addr[pd->didx], af);
3450 				break;
3451 #endif /* INET */
3452 			}
3453 			break;
3454 		}
3455 		if (nr->natpass)
3456 			r = NULL;
3457 		pd->nat_rule = nr;
3458 	}
3459 
3460 	while (r != NULL) {
3461 		r->evaluations++;
3462 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
3463 			r = r->skip[PF_SKIP_IFP].ptr;
3464 		else if (r->direction && r->direction != direction)
3465 			r = r->skip[PF_SKIP_DIR].ptr;
3466 		else if (r->af && r->af != af)
3467 			r = r->skip[PF_SKIP_AF].ptr;
3468 		else if (r->proto && r->proto != pd->proto)
3469 			r = r->skip[PF_SKIP_PROTO].ptr;
3470 		else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3471 		    r->src.neg, kif, M_GETFIB(m)))
3472 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3473 		/* tcp/udp only. port_op always 0 in other cases */
3474 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
3475 		    r->src.port[0], r->src.port[1], sport))
3476 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
3477 		else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3478 		    r->dst.neg, NULL, M_GETFIB(m)))
3479 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
3480 		/* tcp/udp only. port_op always 0 in other cases */
3481 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3482 		    r->dst.port[0], r->dst.port[1], dport))
3483 			r = r->skip[PF_SKIP_DST_PORT].ptr;
3484 		/* icmp only. type always 0 in other cases */
3485 		else if (r->type && r->type != icmptype + 1)
3486 			r = TAILQ_NEXT(r, entries);
3487 		/* icmp only. type always 0 in other cases */
3488 		else if (r->code && r->code != icmpcode + 1)
3489 			r = TAILQ_NEXT(r, entries);
3490 		else if (r->tos && !(r->tos == pd->tos))
3491 			r = TAILQ_NEXT(r, entries);
3492 		else if (r->rule_flag & PFRULE_FRAGMENT)
3493 			r = TAILQ_NEXT(r, entries);
3494 		else if (pd->proto == IPPROTO_TCP &&
3495 		    (r->flagset & th->th_flags) != r->flags)
3496 			r = TAILQ_NEXT(r, entries);
3497 		/* tcp/udp only. uid.op always 0 in other cases */
3498 		else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3499 		    pf_socket_lookup(direction, pd, m), 1)) &&
3500 		    !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3501 		    pd->lookup.uid))
3502 			r = TAILQ_NEXT(r, entries);
3503 		/* tcp/udp only. gid.op always 0 in other cases */
3504 		else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3505 		    pf_socket_lookup(direction, pd, m), 1)) &&
3506 		    !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3507 		    pd->lookup.gid))
3508 			r = TAILQ_NEXT(r, entries);
3509 		else if (r->prio &&
3510 		    !pf_match_ieee8021q_pcp(r->prio, m))
3511 			r = TAILQ_NEXT(r, entries);
3512 		else if (r->prob &&
3513 		    r->prob <= arc4random())
3514 			r = TAILQ_NEXT(r, entries);
3515 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
3516 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
3517 			r = TAILQ_NEXT(r, entries);
3518 		else if (r->os_fingerprint != PF_OSFP_ANY &&
3519 		    (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3520 		    pf_osfp_fingerprint(pd, m, off, th),
3521 		    r->os_fingerprint)))
3522 			r = TAILQ_NEXT(r, entries);
3523 		else {
3524 			if (r->tag)
3525 				tag = r->tag;
3526 			if (r->rtableid >= 0)
3527 				rtableid = r->rtableid;
3528 			if (r->anchor == NULL) {
3529 				match = 1;
3530 				*rm = r;
3531 				*am = a;
3532 				*rsm = ruleset;
3533 				if ((*rm)->quick)
3534 					break;
3535 				r = TAILQ_NEXT(r, entries);
3536 			} else
3537 				pf_step_into_anchor(anchor_stack, &asd,
3538 				    &ruleset, PF_RULESET_FILTER, &r, &a,
3539 				    &match);
3540 		}
3541 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3542 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3543 			break;
3544 	}
3545 	r = *rm;
3546 	a = *am;
3547 	ruleset = *rsm;
3548 
3549 	REASON_SET(&reason, PFRES_MATCH);
3550 
3551 	if (r->log || (nr != NULL && nr->log)) {
3552 		if (rewrite)
3553 			m_copyback(m, off, hdrlen, pd->hdr.any);
3554 		PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3555 		    ruleset, pd, 1);
3556 	}
3557 
3558 	if ((r->action == PF_DROP) &&
3559 	    ((r->rule_flag & PFRULE_RETURNRST) ||
3560 	    (r->rule_flag & PFRULE_RETURNICMP) ||
3561 	    (r->rule_flag & PFRULE_RETURN))) {
3562 		pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum,
3563 		    bip_sum, hdrlen, &reason);
3564 	}
3565 
3566 	if (r->action == PF_DROP)
3567 		goto cleanup;
3568 
3569 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3570 		REASON_SET(&reason, PFRES_MEMORY);
3571 		goto cleanup;
3572 	}
3573 	if (rtableid >= 0)
3574 		M_SETFIB(m, rtableid);
3575 
3576 	if (!state_icmp && (r->keep_state || nr != NULL ||
3577 	    (pd->flags & PFDESC_TCP_NORM))) {
3578 		int action;
3579 		action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3580 		    sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3581 		    hdrlen);
3582 		if (action != PF_PASS) {
3583 			if (action == PF_DROP &&
3584 			    (r->rule_flag & PFRULE_RETURN))
3585 				pf_return(r, nr, pd, sk, off, m, th, kif,
3586 				    bproto_sum, bip_sum, hdrlen, &reason);
3587 			return (action);
3588 		}
3589 	} else {
3590 		if (sk != NULL)
3591 			uma_zfree(V_pf_state_key_z, sk);
3592 		if (nk != NULL)
3593 			uma_zfree(V_pf_state_key_z, nk);
3594 	}
3595 
3596 	/* copy back packet headers if we performed NAT operations */
3597 	if (rewrite)
3598 		m_copyback(m, off, hdrlen, pd->hdr.any);
3599 
3600 	if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3601 	    direction == PF_OUT &&
3602 	    V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m))
3603 		/*
3604 		 * We want the state created, but we dont
3605 		 * want to send this in case a partner
3606 		 * firewall has to know about it to allow
3607 		 * replies through it.
3608 		 */
3609 		return (PF_DEFER);
3610 
3611 	return (PF_PASS);
3612 
3613 cleanup:
3614 	if (sk != NULL)
3615 		uma_zfree(V_pf_state_key_z, sk);
3616 	if (nk != NULL)
3617 		uma_zfree(V_pf_state_key_z, nk);
3618 	return (PF_DROP);
3619 }
3620 
3621 static int
3622 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3623     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3624     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3625     u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3626     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3627 {
3628 	struct pf_state		*s = NULL;
3629 	struct pf_src_node	*sn = NULL;
3630 	struct tcphdr		*th = pd->hdr.tcp;
3631 	u_int16_t		 mss = V_tcp_mssdflt;
3632 	u_short			 reason;
3633 
3634 	/* check maximums */
3635 	if (r->max_states &&
3636 	    (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3637 		counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
3638 		REASON_SET(&reason, PFRES_MAXSTATES);
3639 		goto csfailed;
3640 	}
3641 	/* src node for filter rule */
3642 	if ((r->rule_flag & PFRULE_SRCTRACK ||
3643 	    r->rpool.opts & PF_POOL_STICKYADDR) &&
3644 	    pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3645 		REASON_SET(&reason, PFRES_SRCLIMIT);
3646 		goto csfailed;
3647 	}
3648 	/* src node for translation rule */
3649 	if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3650 	    pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3651 		REASON_SET(&reason, PFRES_SRCLIMIT);
3652 		goto csfailed;
3653 	}
3654 	s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3655 	if (s == NULL) {
3656 		REASON_SET(&reason, PFRES_MEMORY);
3657 		goto csfailed;
3658 	}
3659 	s->rule.ptr = r;
3660 	s->nat_rule.ptr = nr;
3661 	s->anchor.ptr = a;
3662 	STATE_INC_COUNTERS(s);
3663 	if (r->allow_opts)
3664 		s->state_flags |= PFSTATE_ALLOWOPTS;
3665 	if (r->rule_flag & PFRULE_STATESLOPPY)
3666 		s->state_flags |= PFSTATE_SLOPPY;
3667 	s->log = r->log & PF_LOG_ALL;
3668 	s->sync_state = PFSYNC_S_NONE;
3669 	if (nr != NULL)
3670 		s->log |= nr->log & PF_LOG_ALL;
3671 	switch (pd->proto) {
3672 	case IPPROTO_TCP:
3673 		s->src.seqlo = ntohl(th->th_seq);
3674 		s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3675 		if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3676 		    r->keep_state == PF_STATE_MODULATE) {
3677 			/* Generate sequence number modulator */
3678 			if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3679 			    0)
3680 				s->src.seqdiff = 1;
3681 			pf_change_proto_a(m, &th->th_seq, &th->th_sum,
3682 			    htonl(s->src.seqlo + s->src.seqdiff), 0);
3683 			*rewrite = 1;
3684 		} else
3685 			s->src.seqdiff = 0;
3686 		if (th->th_flags & TH_SYN) {
3687 			s->src.seqhi++;
3688 			s->src.wscale = pf_get_wscale(m, off,
3689 			    th->th_off, pd->af);
3690 		}
3691 		s->src.max_win = MAX(ntohs(th->th_win), 1);
3692 		if (s->src.wscale & PF_WSCALE_MASK) {
3693 			/* Remove scale factor from initial window */
3694 			int win = s->src.max_win;
3695 			win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3696 			s->src.max_win = (win - 1) >>
3697 			    (s->src.wscale & PF_WSCALE_MASK);
3698 		}
3699 		if (th->th_flags & TH_FIN)
3700 			s->src.seqhi++;
3701 		s->dst.seqhi = 1;
3702 		s->dst.max_win = 1;
3703 		s->src.state = TCPS_SYN_SENT;
3704 		s->dst.state = TCPS_CLOSED;
3705 		s->timeout = PFTM_TCP_FIRST_PACKET;
3706 		break;
3707 	case IPPROTO_UDP:
3708 		s->src.state = PFUDPS_SINGLE;
3709 		s->dst.state = PFUDPS_NO_TRAFFIC;
3710 		s->timeout = PFTM_UDP_FIRST_PACKET;
3711 		break;
3712 	case IPPROTO_ICMP:
3713 #ifdef INET6
3714 	case IPPROTO_ICMPV6:
3715 #endif
3716 		s->timeout = PFTM_ICMP_FIRST_PACKET;
3717 		break;
3718 	default:
3719 		s->src.state = PFOTHERS_SINGLE;
3720 		s->dst.state = PFOTHERS_NO_TRAFFIC;
3721 		s->timeout = PFTM_OTHER_FIRST_PACKET;
3722 	}
3723 
3724 	if (r->rt) {
3725 		if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) {
3726 			REASON_SET(&reason, PFRES_MAPFAILED);
3727 			pf_src_tree_remove_state(s);
3728 			STATE_DEC_COUNTERS(s);
3729 			uma_zfree(V_pf_state_z, s);
3730 			goto csfailed;
3731 		}
3732 		s->rt_kif = r->rpool.cur->kif;
3733 	}
3734 
3735 	s->creation = time_uptime;
3736 	s->expire = time_uptime;
3737 
3738 	if (sn != NULL)
3739 		s->src_node = sn;
3740 	if (nsn != NULL) {
3741 		/* XXX We only modify one side for now. */
3742 		PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3743 		s->nat_src_node = nsn;
3744 	}
3745 	if (pd->proto == IPPROTO_TCP) {
3746 		if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3747 		    off, pd, th, &s->src, &s->dst)) {
3748 			REASON_SET(&reason, PFRES_MEMORY);
3749 			pf_src_tree_remove_state(s);
3750 			STATE_DEC_COUNTERS(s);
3751 			uma_zfree(V_pf_state_z, s);
3752 			return (PF_DROP);
3753 		}
3754 		if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3755 		    pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3756 		    &s->src, &s->dst, rewrite)) {
3757 			/* This really shouldn't happen!!! */
3758 			DPFPRINTF(PF_DEBUG_URGENT,
3759 			    ("pf_normalize_tcp_stateful failed on first "
3760 			     "pkt\n"));
3761 			pf_normalize_tcp_cleanup(s);
3762 			pf_src_tree_remove_state(s);
3763 			STATE_DEC_COUNTERS(s);
3764 			uma_zfree(V_pf_state_z, s);
3765 			return (PF_DROP);
3766 		}
3767 	}
3768 	s->direction = pd->dir;
3769 
3770 	/*
3771 	 * sk/nk could already been setup by pf_get_translation().
3772 	 */
3773 	if (nr == NULL) {
3774 		KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3775 		    __func__, nr, sk, nk));
3776 		sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3777 		if (sk == NULL)
3778 			goto csfailed;
3779 		nk = sk;
3780 	} else
3781 		KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3782 		    __func__, nr, sk, nk));
3783 
3784 	/* Swap sk/nk for PF_OUT. */
3785 	if (pf_state_insert(BOUND_IFACE(r, kif),
3786 	    (pd->dir == PF_IN) ? sk : nk,
3787 	    (pd->dir == PF_IN) ? nk : sk, s)) {
3788 		if (pd->proto == IPPROTO_TCP)
3789 			pf_normalize_tcp_cleanup(s);
3790 		REASON_SET(&reason, PFRES_STATEINS);
3791 		pf_src_tree_remove_state(s);
3792 		STATE_DEC_COUNTERS(s);
3793 		uma_zfree(V_pf_state_z, s);
3794 		return (PF_DROP);
3795 	} else
3796 		*sm = s;
3797 
3798 	if (tag > 0)
3799 		s->tag = tag;
3800 	if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3801 	    TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3802 		s->src.state = PF_TCPS_PROXY_SRC;
3803 		/* undo NAT changes, if they have taken place */
3804 		if (nr != NULL) {
3805 			struct pf_state_key *skt = s->key[PF_SK_WIRE];
3806 			if (pd->dir == PF_OUT)
3807 				skt = s->key[PF_SK_STACK];
3808 			PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3809 			PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3810 			if (pd->sport)
3811 				*pd->sport = skt->port[pd->sidx];
3812 			if (pd->dport)
3813 				*pd->dport = skt->port[pd->didx];
3814 			if (pd->proto_sum)
3815 				*pd->proto_sum = bproto_sum;
3816 			if (pd->ip_sum)
3817 				*pd->ip_sum = bip_sum;
3818 			m_copyback(m, off, hdrlen, pd->hdr.any);
3819 		}
3820 		s->src.seqhi = htonl(arc4random());
3821 		/* Find mss option */
3822 		int rtid = M_GETFIB(m);
3823 		mss = pf_get_mss(m, off, th->th_off, pd->af);
3824 		mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3825 		mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3826 		s->src.mss = mss;
3827 		pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3828 		    th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3829 		    TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3830 		REASON_SET(&reason, PFRES_SYNPROXY);
3831 		return (PF_SYNPROXY_DROP);
3832 	}
3833 
3834 	return (PF_PASS);
3835 
3836 csfailed:
3837 	if (sk != NULL)
3838 		uma_zfree(V_pf_state_key_z, sk);
3839 	if (nk != NULL)
3840 		uma_zfree(V_pf_state_key_z, nk);
3841 
3842 	if (sn != NULL) {
3843 		struct pf_srchash *sh;
3844 
3845 		sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
3846 		PF_HASHROW_LOCK(sh);
3847 		if (--sn->states == 0 && sn->expire == 0) {
3848 			pf_unlink_src_node(sn);
3849 			uma_zfree(V_pf_sources_z, sn);
3850 			counter_u64_add(
3851 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3852 		}
3853 		PF_HASHROW_UNLOCK(sh);
3854 	}
3855 
3856 	if (nsn != sn && nsn != NULL) {
3857 		struct pf_srchash *sh;
3858 
3859 		sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)];
3860 		PF_HASHROW_LOCK(sh);
3861 		if (--nsn->states == 0 && nsn->expire == 0) {
3862 			pf_unlink_src_node(nsn);
3863 			uma_zfree(V_pf_sources_z, nsn);
3864 			counter_u64_add(
3865 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3866 		}
3867 		PF_HASHROW_UNLOCK(sh);
3868 	}
3869 
3870 	return (PF_DROP);
3871 }
3872 
3873 static int
3874 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
3875     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
3876     struct pf_ruleset **rsm)
3877 {
3878 	struct pf_rule		*r, *a = NULL;
3879 	struct pf_ruleset	*ruleset = NULL;
3880 	sa_family_t		 af = pd->af;
3881 	u_short			 reason;
3882 	int			 tag = -1;
3883 	int			 asd = 0;
3884 	int			 match = 0;
3885 	struct pf_anchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
3886 
3887 	PF_RULES_RASSERT();
3888 
3889 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3890 	while (r != NULL) {
3891 		r->evaluations++;
3892 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
3893 			r = r->skip[PF_SKIP_IFP].ptr;
3894 		else if (r->direction && r->direction != direction)
3895 			r = r->skip[PF_SKIP_DIR].ptr;
3896 		else if (r->af && r->af != af)
3897 			r = r->skip[PF_SKIP_AF].ptr;
3898 		else if (r->proto && r->proto != pd->proto)
3899 			r = r->skip[PF_SKIP_PROTO].ptr;
3900 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3901 		    r->src.neg, kif, M_GETFIB(m)))
3902 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3903 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3904 		    r->dst.neg, NULL, M_GETFIB(m)))
3905 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
3906 		else if (r->tos && !(r->tos == pd->tos))
3907 			r = TAILQ_NEXT(r, entries);
3908 		else if (r->os_fingerprint != PF_OSFP_ANY)
3909 			r = TAILQ_NEXT(r, entries);
3910 		else if (pd->proto == IPPROTO_UDP &&
3911 		    (r->src.port_op || r->dst.port_op))
3912 			r = TAILQ_NEXT(r, entries);
3913 		else if (pd->proto == IPPROTO_TCP &&
3914 		    (r->src.port_op || r->dst.port_op || r->flagset))
3915 			r = TAILQ_NEXT(r, entries);
3916 		else if ((pd->proto == IPPROTO_ICMP ||
3917 		    pd->proto == IPPROTO_ICMPV6) &&
3918 		    (r->type || r->code))
3919 			r = TAILQ_NEXT(r, entries);
3920 		else if (r->prio &&
3921 		    !pf_match_ieee8021q_pcp(r->prio, m))
3922 			r = TAILQ_NEXT(r, entries);
3923 		else if (r->prob && r->prob <=
3924 		    (arc4random() % (UINT_MAX - 1) + 1))
3925 			r = TAILQ_NEXT(r, entries);
3926 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
3927 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
3928 			r = TAILQ_NEXT(r, entries);
3929 		else {
3930 			if (r->anchor == NULL) {
3931 				match = 1;
3932 				*rm = r;
3933 				*am = a;
3934 				*rsm = ruleset;
3935 				if ((*rm)->quick)
3936 					break;
3937 				r = TAILQ_NEXT(r, entries);
3938 			} else
3939 				pf_step_into_anchor(anchor_stack, &asd,
3940 				    &ruleset, PF_RULESET_FILTER, &r, &a,
3941 				    &match);
3942 		}
3943 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3944 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3945 			break;
3946 	}
3947 	r = *rm;
3948 	a = *am;
3949 	ruleset = *rsm;
3950 
3951 	REASON_SET(&reason, PFRES_MATCH);
3952 
3953 	if (r->log)
3954 		PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
3955 		    1);
3956 
3957 	if (r->action != PF_PASS)
3958 		return (PF_DROP);
3959 
3960 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3961 		REASON_SET(&reason, PFRES_MEMORY);
3962 		return (PF_DROP);
3963 	}
3964 
3965 	return (PF_PASS);
3966 }
3967 
3968 static int
3969 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
3970 	struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
3971 	struct pf_pdesc *pd, u_short *reason, int *copyback)
3972 {
3973 	struct tcphdr		*th = pd->hdr.tcp;
3974 	u_int16_t		 win = ntohs(th->th_win);
3975 	u_int32_t		 ack, end, seq, orig_seq;
3976 	u_int8_t		 sws, dws;
3977 	int			 ackskew;
3978 
3979 	if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
3980 		sws = src->wscale & PF_WSCALE_MASK;
3981 		dws = dst->wscale & PF_WSCALE_MASK;
3982 	} else
3983 		sws = dws = 0;
3984 
3985 	/*
3986 	 * Sequence tracking algorithm from Guido van Rooij's paper:
3987 	 *   http://www.madison-gurkha.com/publications/tcp_filtering/
3988 	 *	tcp_filtering.ps
3989 	 */
3990 
3991 	orig_seq = seq = ntohl(th->th_seq);
3992 	if (src->seqlo == 0) {
3993 		/* First packet from this end. Set its state */
3994 
3995 		if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
3996 		    src->scrub == NULL) {
3997 			if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
3998 				REASON_SET(reason, PFRES_MEMORY);
3999 				return (PF_DROP);
4000 			}
4001 		}
4002 
4003 		/* Deferred generation of sequence number modulator */
4004 		if (dst->seqdiff && !src->seqdiff) {
4005 			/* use random iss for the TCP server */
4006 			while ((src->seqdiff = arc4random() - seq) == 0)
4007 				;
4008 			ack = ntohl(th->th_ack) - dst->seqdiff;
4009 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
4010 			    src->seqdiff), 0);
4011 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
4012 			*copyback = 1;
4013 		} else {
4014 			ack = ntohl(th->th_ack);
4015 		}
4016 
4017 		end = seq + pd->p_len;
4018 		if (th->th_flags & TH_SYN) {
4019 			end++;
4020 			if (dst->wscale & PF_WSCALE_FLAG) {
4021 				src->wscale = pf_get_wscale(m, off, th->th_off,
4022 				    pd->af);
4023 				if (src->wscale & PF_WSCALE_FLAG) {
4024 					/* Remove scale factor from initial
4025 					 * window */
4026 					sws = src->wscale & PF_WSCALE_MASK;
4027 					win = ((u_int32_t)win + (1 << sws) - 1)
4028 					    >> sws;
4029 					dws = dst->wscale & PF_WSCALE_MASK;
4030 				} else {
4031 					/* fixup other window */
4032 					dst->max_win <<= dst->wscale &
4033 					    PF_WSCALE_MASK;
4034 					/* in case of a retrans SYN|ACK */
4035 					dst->wscale = 0;
4036 				}
4037 			}
4038 		}
4039 		if (th->th_flags & TH_FIN)
4040 			end++;
4041 
4042 		src->seqlo = seq;
4043 		if (src->state < TCPS_SYN_SENT)
4044 			src->state = TCPS_SYN_SENT;
4045 
4046 		/*
4047 		 * May need to slide the window (seqhi may have been set by
4048 		 * the crappy stack check or if we picked up the connection
4049 		 * after establishment)
4050 		 */
4051 		if (src->seqhi == 1 ||
4052 		    SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
4053 			src->seqhi = end + MAX(1, dst->max_win << dws);
4054 		if (win > src->max_win)
4055 			src->max_win = win;
4056 
4057 	} else {
4058 		ack = ntohl(th->th_ack) - dst->seqdiff;
4059 		if (src->seqdiff) {
4060 			/* Modulate sequence numbers */
4061 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
4062 			    src->seqdiff), 0);
4063 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
4064 			*copyback = 1;
4065 		}
4066 		end = seq + pd->p_len;
4067 		if (th->th_flags & TH_SYN)
4068 			end++;
4069 		if (th->th_flags & TH_FIN)
4070 			end++;
4071 	}
4072 
4073 	if ((th->th_flags & TH_ACK) == 0) {
4074 		/* Let it pass through the ack skew check */
4075 		ack = dst->seqlo;
4076 	} else if ((ack == 0 &&
4077 	    (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
4078 	    /* broken tcp stacks do not set ack */
4079 	    (dst->state < TCPS_SYN_SENT)) {
4080 		/*
4081 		 * Many stacks (ours included) will set the ACK number in an
4082 		 * FIN|ACK if the SYN times out -- no sequence to ACK.
4083 		 */
4084 		ack = dst->seqlo;
4085 	}
4086 
4087 	if (seq == end) {
4088 		/* Ease sequencing restrictions on no data packets */
4089 		seq = src->seqlo;
4090 		end = seq;
4091 	}
4092 
4093 	ackskew = dst->seqlo - ack;
4094 
4095 
4096 	/*
4097 	 * Need to demodulate the sequence numbers in any TCP SACK options
4098 	 * (Selective ACK). We could optionally validate the SACK values
4099 	 * against the current ACK window, either forwards or backwards, but
4100 	 * I'm not confident that SACK has been implemented properly
4101 	 * everywhere. It wouldn't surprise me if several stacks accidentally
4102 	 * SACK too far backwards of previously ACKed data. There really aren't
4103 	 * any security implications of bad SACKing unless the target stack
4104 	 * doesn't validate the option length correctly. Someone trying to
4105 	 * spoof into a TCP connection won't bother blindly sending SACK
4106 	 * options anyway.
4107 	 */
4108 	if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
4109 		if (pf_modulate_sack(m, off, pd, th, dst))
4110 			*copyback = 1;
4111 	}
4112 
4113 
4114 #define	MAXACKWINDOW (0xffff + 1500)	/* 1500 is an arbitrary fudge factor */
4115 	if (SEQ_GEQ(src->seqhi, end) &&
4116 	    /* Last octet inside other's window space */
4117 	    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
4118 	    /* Retrans: not more than one window back */
4119 	    (ackskew >= -MAXACKWINDOW) &&
4120 	    /* Acking not more than one reassembled fragment backwards */
4121 	    (ackskew <= (MAXACKWINDOW << sws)) &&
4122 	    /* Acking not more than one window forward */
4123 	    ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
4124 	    (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
4125 	    (pd->flags & PFDESC_IP_REAS) == 0)) {
4126 	    /* Require an exact/+1 sequence match on resets when possible */
4127 
4128 		if (dst->scrub || src->scrub) {
4129 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4130 			    *state, src, dst, copyback))
4131 				return (PF_DROP);
4132 		}
4133 
4134 		/* update max window */
4135 		if (src->max_win < win)
4136 			src->max_win = win;
4137 		/* synchronize sequencing */
4138 		if (SEQ_GT(end, src->seqlo))
4139 			src->seqlo = end;
4140 		/* slide the window of what the other end can send */
4141 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4142 			dst->seqhi = ack + MAX((win << sws), 1);
4143 
4144 
4145 		/* update states */
4146 		if (th->th_flags & TH_SYN)
4147 			if (src->state < TCPS_SYN_SENT)
4148 				src->state = TCPS_SYN_SENT;
4149 		if (th->th_flags & TH_FIN)
4150 			if (src->state < TCPS_CLOSING)
4151 				src->state = TCPS_CLOSING;
4152 		if (th->th_flags & TH_ACK) {
4153 			if (dst->state == TCPS_SYN_SENT) {
4154 				dst->state = TCPS_ESTABLISHED;
4155 				if (src->state == TCPS_ESTABLISHED &&
4156 				    (*state)->src_node != NULL &&
4157 				    pf_src_connlimit(state)) {
4158 					REASON_SET(reason, PFRES_SRCLIMIT);
4159 					return (PF_DROP);
4160 				}
4161 			} else if (dst->state == TCPS_CLOSING)
4162 				dst->state = TCPS_FIN_WAIT_2;
4163 		}
4164 		if (th->th_flags & TH_RST)
4165 			src->state = dst->state = TCPS_TIME_WAIT;
4166 
4167 		/* update expire time */
4168 		(*state)->expire = time_uptime;
4169 		if (src->state >= TCPS_FIN_WAIT_2 &&
4170 		    dst->state >= TCPS_FIN_WAIT_2)
4171 			(*state)->timeout = PFTM_TCP_CLOSED;
4172 		else if (src->state >= TCPS_CLOSING &&
4173 		    dst->state >= TCPS_CLOSING)
4174 			(*state)->timeout = PFTM_TCP_FIN_WAIT;
4175 		else if (src->state < TCPS_ESTABLISHED ||
4176 		    dst->state < TCPS_ESTABLISHED)
4177 			(*state)->timeout = PFTM_TCP_OPENING;
4178 		else if (src->state >= TCPS_CLOSING ||
4179 		    dst->state >= TCPS_CLOSING)
4180 			(*state)->timeout = PFTM_TCP_CLOSING;
4181 		else
4182 			(*state)->timeout = PFTM_TCP_ESTABLISHED;
4183 
4184 		/* Fall through to PASS packet */
4185 
4186 	} else if ((dst->state < TCPS_SYN_SENT ||
4187 		dst->state >= TCPS_FIN_WAIT_2 ||
4188 		src->state >= TCPS_FIN_WAIT_2) &&
4189 	    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4190 	    /* Within a window forward of the originating packet */
4191 	    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4192 	    /* Within a window backward of the originating packet */
4193 
4194 		/*
4195 		 * This currently handles three situations:
4196 		 *  1) Stupid stacks will shotgun SYNs before their peer
4197 		 *     replies.
4198 		 *  2) When PF catches an already established stream (the
4199 		 *     firewall rebooted, the state table was flushed, routes
4200 		 *     changed...)
4201 		 *  3) Packets get funky immediately after the connection
4202 		 *     closes (this should catch Solaris spurious ACK|FINs
4203 		 *     that web servers like to spew after a close)
4204 		 *
4205 		 * This must be a little more careful than the above code
4206 		 * since packet floods will also be caught here. We don't
4207 		 * update the TTL here to mitigate the damage of a packet
4208 		 * flood and so the same code can handle awkward establishment
4209 		 * and a loosened connection close.
4210 		 * In the establishment case, a correct peer response will
4211 		 * validate the connection, go through the normal state code
4212 		 * and keep updating the state TTL.
4213 		 */
4214 
4215 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
4216 			printf("pf: loose state match: ");
4217 			pf_print_state(*state);
4218 			pf_print_flags(th->th_flags);
4219 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4220 			    "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
4221 			    pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
4222 			    (unsigned long long)(*state)->packets[1],
4223 			    pd->dir == PF_IN ? "in" : "out",
4224 			    pd->dir == (*state)->direction ? "fwd" : "rev");
4225 		}
4226 
4227 		if (dst->scrub || src->scrub) {
4228 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4229 			    *state, src, dst, copyback))
4230 				return (PF_DROP);
4231 		}
4232 
4233 		/* update max window */
4234 		if (src->max_win < win)
4235 			src->max_win = win;
4236 		/* synchronize sequencing */
4237 		if (SEQ_GT(end, src->seqlo))
4238 			src->seqlo = end;
4239 		/* slide the window of what the other end can send */
4240 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4241 			dst->seqhi = ack + MAX((win << sws), 1);
4242 
4243 		/*
4244 		 * Cannot set dst->seqhi here since this could be a shotgunned
4245 		 * SYN and not an already established connection.
4246 		 */
4247 
4248 		if (th->th_flags & TH_FIN)
4249 			if (src->state < TCPS_CLOSING)
4250 				src->state = TCPS_CLOSING;
4251 		if (th->th_flags & TH_RST)
4252 			src->state = dst->state = TCPS_TIME_WAIT;
4253 
4254 		/* Fall through to PASS packet */
4255 
4256 	} else {
4257 		if ((*state)->dst.state == TCPS_SYN_SENT &&
4258 		    (*state)->src.state == TCPS_SYN_SENT) {
4259 			/* Send RST for state mismatches during handshake */
4260 			if (!(th->th_flags & TH_RST))
4261 				pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4262 				    pd->dst, pd->src, th->th_dport,
4263 				    th->th_sport, ntohl(th->th_ack), 0,
4264 				    TH_RST, 0, 0,
4265 				    (*state)->rule.ptr->return_ttl, 1, 0,
4266 				    kif->pfik_ifp);
4267 			src->seqlo = 0;
4268 			src->seqhi = 1;
4269 			src->max_win = 1;
4270 		} else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4271 			printf("pf: BAD state: ");
4272 			pf_print_state(*state);
4273 			pf_print_flags(th->th_flags);
4274 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4275 			    "pkts=%llu:%llu dir=%s,%s\n",
4276 			    seq, orig_seq, ack, pd->p_len, ackskew,
4277 			    (unsigned long long)(*state)->packets[0],
4278 			    (unsigned long long)(*state)->packets[1],
4279 			    pd->dir == PF_IN ? "in" : "out",
4280 			    pd->dir == (*state)->direction ? "fwd" : "rev");
4281 			printf("pf: State failure on: %c %c %c %c | %c %c\n",
4282 			    SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4283 			    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4284 			    ' ': '2',
4285 			    (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4286 			    (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4287 			    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4288 			    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4289 		}
4290 		REASON_SET(reason, PFRES_BADSTATE);
4291 		return (PF_DROP);
4292 	}
4293 
4294 	return (PF_PASS);
4295 }
4296 
4297 static int
4298 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4299 	struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4300 {
4301 	struct tcphdr		*th = pd->hdr.tcp;
4302 
4303 	if (th->th_flags & TH_SYN)
4304 		if (src->state < TCPS_SYN_SENT)
4305 			src->state = TCPS_SYN_SENT;
4306 	if (th->th_flags & TH_FIN)
4307 		if (src->state < TCPS_CLOSING)
4308 			src->state = TCPS_CLOSING;
4309 	if (th->th_flags & TH_ACK) {
4310 		if (dst->state == TCPS_SYN_SENT) {
4311 			dst->state = TCPS_ESTABLISHED;
4312 			if (src->state == TCPS_ESTABLISHED &&
4313 			    (*state)->src_node != NULL &&
4314 			    pf_src_connlimit(state)) {
4315 				REASON_SET(reason, PFRES_SRCLIMIT);
4316 				return (PF_DROP);
4317 			}
4318 		} else if (dst->state == TCPS_CLOSING) {
4319 			dst->state = TCPS_FIN_WAIT_2;
4320 		} else if (src->state == TCPS_SYN_SENT &&
4321 		    dst->state < TCPS_SYN_SENT) {
4322 			/*
4323 			 * Handle a special sloppy case where we only see one
4324 			 * half of the connection. If there is a ACK after
4325 			 * the initial SYN without ever seeing a packet from
4326 			 * the destination, set the connection to established.
4327 			 */
4328 			dst->state = src->state = TCPS_ESTABLISHED;
4329 			if ((*state)->src_node != NULL &&
4330 			    pf_src_connlimit(state)) {
4331 				REASON_SET(reason, PFRES_SRCLIMIT);
4332 				return (PF_DROP);
4333 			}
4334 		} else if (src->state == TCPS_CLOSING &&
4335 		    dst->state == TCPS_ESTABLISHED &&
4336 		    dst->seqlo == 0) {
4337 			/*
4338 			 * Handle the closing of half connections where we
4339 			 * don't see the full bidirectional FIN/ACK+ACK
4340 			 * handshake.
4341 			 */
4342 			dst->state = TCPS_CLOSING;
4343 		}
4344 	}
4345 	if (th->th_flags & TH_RST)
4346 		src->state = dst->state = TCPS_TIME_WAIT;
4347 
4348 	/* update expire time */
4349 	(*state)->expire = time_uptime;
4350 	if (src->state >= TCPS_FIN_WAIT_2 &&
4351 	    dst->state >= TCPS_FIN_WAIT_2)
4352 		(*state)->timeout = PFTM_TCP_CLOSED;
4353 	else if (src->state >= TCPS_CLOSING &&
4354 	    dst->state >= TCPS_CLOSING)
4355 		(*state)->timeout = PFTM_TCP_FIN_WAIT;
4356 	else if (src->state < TCPS_ESTABLISHED ||
4357 	    dst->state < TCPS_ESTABLISHED)
4358 		(*state)->timeout = PFTM_TCP_OPENING;
4359 	else if (src->state >= TCPS_CLOSING ||
4360 	    dst->state >= TCPS_CLOSING)
4361 		(*state)->timeout = PFTM_TCP_CLOSING;
4362 	else
4363 		(*state)->timeout = PFTM_TCP_ESTABLISHED;
4364 
4365 	return (PF_PASS);
4366 }
4367 
4368 static int
4369 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4370     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4371     u_short *reason)
4372 {
4373 	struct pf_state_key_cmp	 key;
4374 	struct tcphdr		*th = pd->hdr.tcp;
4375 	int			 copyback = 0;
4376 	struct pf_state_peer	*src, *dst;
4377 	struct pf_state_key	*sk;
4378 
4379 	bzero(&key, sizeof(key));
4380 	key.af = pd->af;
4381 	key.proto = IPPROTO_TCP;
4382 	if (direction == PF_IN)	{	/* wire side, straight */
4383 		PF_ACPY(&key.addr[0], pd->src, key.af);
4384 		PF_ACPY(&key.addr[1], pd->dst, key.af);
4385 		key.port[0] = th->th_sport;
4386 		key.port[1] = th->th_dport;
4387 	} else {			/* stack side, reverse */
4388 		PF_ACPY(&key.addr[1], pd->src, key.af);
4389 		PF_ACPY(&key.addr[0], pd->dst, key.af);
4390 		key.port[1] = th->th_sport;
4391 		key.port[0] = th->th_dport;
4392 	}
4393 
4394 	STATE_LOOKUP(kif, &key, direction, *state, pd);
4395 
4396 	if (direction == (*state)->direction) {
4397 		src = &(*state)->src;
4398 		dst = &(*state)->dst;
4399 	} else {
4400 		src = &(*state)->dst;
4401 		dst = &(*state)->src;
4402 	}
4403 
4404 	sk = (*state)->key[pd->didx];
4405 
4406 	if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4407 		if (direction != (*state)->direction) {
4408 			REASON_SET(reason, PFRES_SYNPROXY);
4409 			return (PF_SYNPROXY_DROP);
4410 		}
4411 		if (th->th_flags & TH_SYN) {
4412 			if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4413 				REASON_SET(reason, PFRES_SYNPROXY);
4414 				return (PF_DROP);
4415 			}
4416 			pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4417 			    pd->src, th->th_dport, th->th_sport,
4418 			    (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4419 			    TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4420 			REASON_SET(reason, PFRES_SYNPROXY);
4421 			return (PF_SYNPROXY_DROP);
4422 		} else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK ||
4423 		    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4424 		    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4425 			REASON_SET(reason, PFRES_SYNPROXY);
4426 			return (PF_DROP);
4427 		} else if ((*state)->src_node != NULL &&
4428 		    pf_src_connlimit(state)) {
4429 			REASON_SET(reason, PFRES_SRCLIMIT);
4430 			return (PF_DROP);
4431 		} else
4432 			(*state)->src.state = PF_TCPS_PROXY_DST;
4433 	}
4434 	if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4435 		if (direction == (*state)->direction) {
4436 			if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4437 			    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4438 			    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4439 				REASON_SET(reason, PFRES_SYNPROXY);
4440 				return (PF_DROP);
4441 			}
4442 			(*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4443 			if ((*state)->dst.seqhi == 1)
4444 				(*state)->dst.seqhi = htonl(arc4random());
4445 			pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4446 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
4447 			    sk->port[pd->sidx], sk->port[pd->didx],
4448 			    (*state)->dst.seqhi, 0, TH_SYN, 0,
4449 			    (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4450 			REASON_SET(reason, PFRES_SYNPROXY);
4451 			return (PF_SYNPROXY_DROP);
4452 		} else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4453 		    (TH_SYN|TH_ACK)) ||
4454 		    (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4455 			REASON_SET(reason, PFRES_SYNPROXY);
4456 			return (PF_DROP);
4457 		} else {
4458 			(*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4459 			(*state)->dst.seqlo = ntohl(th->th_seq);
4460 			pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4461 			    pd->src, th->th_dport, th->th_sport,
4462 			    ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4463 			    TH_ACK, (*state)->src.max_win, 0, 0, 0,
4464 			    (*state)->tag, NULL);
4465 			pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4466 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
4467 			    sk->port[pd->sidx], sk->port[pd->didx],
4468 			    (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4469 			    TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4470 			(*state)->src.seqdiff = (*state)->dst.seqhi -
4471 			    (*state)->src.seqlo;
4472 			(*state)->dst.seqdiff = (*state)->src.seqhi -
4473 			    (*state)->dst.seqlo;
4474 			(*state)->src.seqhi = (*state)->src.seqlo +
4475 			    (*state)->dst.max_win;
4476 			(*state)->dst.seqhi = (*state)->dst.seqlo +
4477 			    (*state)->src.max_win;
4478 			(*state)->src.wscale = (*state)->dst.wscale = 0;
4479 			(*state)->src.state = (*state)->dst.state =
4480 			    TCPS_ESTABLISHED;
4481 			REASON_SET(reason, PFRES_SYNPROXY);
4482 			return (PF_SYNPROXY_DROP);
4483 		}
4484 	}
4485 
4486 	if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4487 	    dst->state >= TCPS_FIN_WAIT_2 &&
4488 	    src->state >= TCPS_FIN_WAIT_2) {
4489 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
4490 			printf("pf: state reuse ");
4491 			pf_print_state(*state);
4492 			pf_print_flags(th->th_flags);
4493 			printf("\n");
4494 		}
4495 		/* XXX make sure it's the same direction ?? */
4496 		(*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4497 		pf_unlink_state(*state, PF_ENTER_LOCKED);
4498 		*state = NULL;
4499 		return (PF_DROP);
4500 	}
4501 
4502 	if ((*state)->state_flags & PFSTATE_SLOPPY) {
4503 		if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4504 			return (PF_DROP);
4505 	} else {
4506 		if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4507 		    &copyback) == PF_DROP)
4508 			return (PF_DROP);
4509 	}
4510 
4511 	/* translate source/destination address, if necessary */
4512 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4513 		struct pf_state_key *nk = (*state)->key[pd->didx];
4514 
4515 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4516 		    nk->port[pd->sidx] != th->th_sport)
4517 			pf_change_ap(m, pd->src, &th->th_sport,
4518 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
4519 			    nk->port[pd->sidx], 0, pd->af);
4520 
4521 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4522 		    nk->port[pd->didx] != th->th_dport)
4523 			pf_change_ap(m, pd->dst, &th->th_dport,
4524 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
4525 			    nk->port[pd->didx], 0, pd->af);
4526 		copyback = 1;
4527 	}
4528 
4529 	/* Copyback sequence modulation or stateful scrub changes if needed */
4530 	if (copyback)
4531 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
4532 
4533 	return (PF_PASS);
4534 }
4535 
4536 static int
4537 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
4538     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4539 {
4540 	struct pf_state_peer	*src, *dst;
4541 	struct pf_state_key_cmp	 key;
4542 	struct udphdr		*uh = pd->hdr.udp;
4543 
4544 	bzero(&key, sizeof(key));
4545 	key.af = pd->af;
4546 	key.proto = IPPROTO_UDP;
4547 	if (direction == PF_IN)	{	/* wire side, straight */
4548 		PF_ACPY(&key.addr[0], pd->src, key.af);
4549 		PF_ACPY(&key.addr[1], pd->dst, key.af);
4550 		key.port[0] = uh->uh_sport;
4551 		key.port[1] = uh->uh_dport;
4552 	} else {			/* stack side, reverse */
4553 		PF_ACPY(&key.addr[1], pd->src, key.af);
4554 		PF_ACPY(&key.addr[0], pd->dst, key.af);
4555 		key.port[1] = uh->uh_sport;
4556 		key.port[0] = uh->uh_dport;
4557 	}
4558 
4559 	STATE_LOOKUP(kif, &key, direction, *state, pd);
4560 
4561 	if (direction == (*state)->direction) {
4562 		src = &(*state)->src;
4563 		dst = &(*state)->dst;
4564 	} else {
4565 		src = &(*state)->dst;
4566 		dst = &(*state)->src;
4567 	}
4568 
4569 	/* update states */
4570 	if (src->state < PFUDPS_SINGLE)
4571 		src->state = PFUDPS_SINGLE;
4572 	if (dst->state == PFUDPS_SINGLE)
4573 		dst->state = PFUDPS_MULTIPLE;
4574 
4575 	/* update expire time */
4576 	(*state)->expire = time_uptime;
4577 	if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4578 		(*state)->timeout = PFTM_UDP_MULTIPLE;
4579 	else
4580 		(*state)->timeout = PFTM_UDP_SINGLE;
4581 
4582 	/* translate source/destination address, if necessary */
4583 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4584 		struct pf_state_key *nk = (*state)->key[pd->didx];
4585 
4586 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4587 		    nk->port[pd->sidx] != uh->uh_sport)
4588 			pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
4589 			    &uh->uh_sum, &nk->addr[pd->sidx],
4590 			    nk->port[pd->sidx], 1, pd->af);
4591 
4592 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4593 		    nk->port[pd->didx] != uh->uh_dport)
4594 			pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
4595 			    &uh->uh_sum, &nk->addr[pd->didx],
4596 			    nk->port[pd->didx], 1, pd->af);
4597 		m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4598 	}
4599 
4600 	return (PF_PASS);
4601 }
4602 
4603 static int
4604 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
4605     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4606 {
4607 	struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
4608 	u_int16_t	 icmpid = 0, *icmpsum;
4609 	u_int8_t	 icmptype, icmpcode;
4610 	int		 state_icmp = 0;
4611 	struct pf_state_key_cmp key;
4612 
4613 	bzero(&key, sizeof(key));
4614 	switch (pd->proto) {
4615 #ifdef INET
4616 	case IPPROTO_ICMP:
4617 		icmptype = pd->hdr.icmp->icmp_type;
4618 		icmpcode = pd->hdr.icmp->icmp_code;
4619 		icmpid = pd->hdr.icmp->icmp_id;
4620 		icmpsum = &pd->hdr.icmp->icmp_cksum;
4621 
4622 		if (icmptype == ICMP_UNREACH ||
4623 		    icmptype == ICMP_SOURCEQUENCH ||
4624 		    icmptype == ICMP_REDIRECT ||
4625 		    icmptype == ICMP_TIMXCEED ||
4626 		    icmptype == ICMP_PARAMPROB)
4627 			state_icmp++;
4628 		break;
4629 #endif /* INET */
4630 #ifdef INET6
4631 	case IPPROTO_ICMPV6:
4632 		icmptype = pd->hdr.icmp6->icmp6_type;
4633 		icmpcode = pd->hdr.icmp6->icmp6_code;
4634 		icmpid = pd->hdr.icmp6->icmp6_id;
4635 		icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4636 
4637 		if (icmptype == ICMP6_DST_UNREACH ||
4638 		    icmptype == ICMP6_PACKET_TOO_BIG ||
4639 		    icmptype == ICMP6_TIME_EXCEEDED ||
4640 		    icmptype == ICMP6_PARAM_PROB)
4641 			state_icmp++;
4642 		break;
4643 #endif /* INET6 */
4644 	}
4645 
4646 	if (!state_icmp) {
4647 
4648 		/*
4649 		 * ICMP query/reply message not related to a TCP/UDP packet.
4650 		 * Search for an ICMP state.
4651 		 */
4652 		key.af = pd->af;
4653 		key.proto = pd->proto;
4654 		key.port[0] = key.port[1] = icmpid;
4655 		if (direction == PF_IN)	{	/* wire side, straight */
4656 			PF_ACPY(&key.addr[0], pd->src, key.af);
4657 			PF_ACPY(&key.addr[1], pd->dst, key.af);
4658 		} else {			/* stack side, reverse */
4659 			PF_ACPY(&key.addr[1], pd->src, key.af);
4660 			PF_ACPY(&key.addr[0], pd->dst, key.af);
4661 		}
4662 
4663 		STATE_LOOKUP(kif, &key, direction, *state, pd);
4664 
4665 		(*state)->expire = time_uptime;
4666 		(*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4667 
4668 		/* translate source/destination address, if necessary */
4669 		if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4670 			struct pf_state_key *nk = (*state)->key[pd->didx];
4671 
4672 			switch (pd->af) {
4673 #ifdef INET
4674 			case AF_INET:
4675 				if (PF_ANEQ(pd->src,
4676 				    &nk->addr[pd->sidx], AF_INET))
4677 					pf_change_a(&saddr->v4.s_addr,
4678 					    pd->ip_sum,
4679 					    nk->addr[pd->sidx].v4.s_addr, 0);
4680 
4681 				if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4682 				    AF_INET))
4683 					pf_change_a(&daddr->v4.s_addr,
4684 					    pd->ip_sum,
4685 					    nk->addr[pd->didx].v4.s_addr, 0);
4686 
4687 				if (nk->port[0] !=
4688 				    pd->hdr.icmp->icmp_id) {
4689 					pd->hdr.icmp->icmp_cksum =
4690 					    pf_cksum_fixup(
4691 					    pd->hdr.icmp->icmp_cksum, icmpid,
4692 					    nk->port[pd->sidx], 0);
4693 					pd->hdr.icmp->icmp_id =
4694 					    nk->port[pd->sidx];
4695 				}
4696 
4697 				m_copyback(m, off, ICMP_MINLEN,
4698 				    (caddr_t )pd->hdr.icmp);
4699 				break;
4700 #endif /* INET */
4701 #ifdef INET6
4702 			case AF_INET6:
4703 				if (PF_ANEQ(pd->src,
4704 				    &nk->addr[pd->sidx], AF_INET6))
4705 					pf_change_a6(saddr,
4706 					    &pd->hdr.icmp6->icmp6_cksum,
4707 					    &nk->addr[pd->sidx], 0);
4708 
4709 				if (PF_ANEQ(pd->dst,
4710 				    &nk->addr[pd->didx], AF_INET6))
4711 					pf_change_a6(daddr,
4712 					    &pd->hdr.icmp6->icmp6_cksum,
4713 					    &nk->addr[pd->didx], 0);
4714 
4715 				m_copyback(m, off, sizeof(struct icmp6_hdr),
4716 				    (caddr_t )pd->hdr.icmp6);
4717 				break;
4718 #endif /* INET6 */
4719 			}
4720 		}
4721 		return (PF_PASS);
4722 
4723 	} else {
4724 		/*
4725 		 * ICMP error message in response to a TCP/UDP packet.
4726 		 * Extract the inner TCP/UDP header and search for that state.
4727 		 */
4728 
4729 		struct pf_pdesc	pd2;
4730 		bzero(&pd2, sizeof pd2);
4731 #ifdef INET
4732 		struct ip	h2;
4733 #endif /* INET */
4734 #ifdef INET6
4735 		struct ip6_hdr	h2_6;
4736 		int		terminal = 0;
4737 #endif /* INET6 */
4738 		int		ipoff2 = 0;
4739 		int		off2 = 0;
4740 
4741 		pd2.af = pd->af;
4742 		/* Payload packet is from the opposite direction. */
4743 		pd2.sidx = (direction == PF_IN) ? 1 : 0;
4744 		pd2.didx = (direction == PF_IN) ? 0 : 1;
4745 		switch (pd->af) {
4746 #ifdef INET
4747 		case AF_INET:
4748 			/* offset of h2 in mbuf chain */
4749 			ipoff2 = off + ICMP_MINLEN;
4750 
4751 			if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4752 			    NULL, reason, pd2.af)) {
4753 				DPFPRINTF(PF_DEBUG_MISC,
4754 				    ("pf: ICMP error message too short "
4755 				    "(ip)\n"));
4756 				return (PF_DROP);
4757 			}
4758 			/*
4759 			 * ICMP error messages don't refer to non-first
4760 			 * fragments
4761 			 */
4762 			if (h2.ip_off & htons(IP_OFFMASK)) {
4763 				REASON_SET(reason, PFRES_FRAG);
4764 				return (PF_DROP);
4765 			}
4766 
4767 			/* offset of protocol header that follows h2 */
4768 			off2 = ipoff2 + (h2.ip_hl << 2);
4769 
4770 			pd2.proto = h2.ip_p;
4771 			pd2.src = (struct pf_addr *)&h2.ip_src;
4772 			pd2.dst = (struct pf_addr *)&h2.ip_dst;
4773 			pd2.ip_sum = &h2.ip_sum;
4774 			break;
4775 #endif /* INET */
4776 #ifdef INET6
4777 		case AF_INET6:
4778 			ipoff2 = off + sizeof(struct icmp6_hdr);
4779 
4780 			if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4781 			    NULL, reason, pd2.af)) {
4782 				DPFPRINTF(PF_DEBUG_MISC,
4783 				    ("pf: ICMP error message too short "
4784 				    "(ip6)\n"));
4785 				return (PF_DROP);
4786 			}
4787 			pd2.proto = h2_6.ip6_nxt;
4788 			pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4789 			pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4790 			pd2.ip_sum = NULL;
4791 			off2 = ipoff2 + sizeof(h2_6);
4792 			do {
4793 				switch (pd2.proto) {
4794 				case IPPROTO_FRAGMENT:
4795 					/*
4796 					 * ICMPv6 error messages for
4797 					 * non-first fragments
4798 					 */
4799 					REASON_SET(reason, PFRES_FRAG);
4800 					return (PF_DROP);
4801 				case IPPROTO_AH:
4802 				case IPPROTO_HOPOPTS:
4803 				case IPPROTO_ROUTING:
4804 				case IPPROTO_DSTOPTS: {
4805 					/* get next header and header length */
4806 					struct ip6_ext opt6;
4807 
4808 					if (!pf_pull_hdr(m, off2, &opt6,
4809 					    sizeof(opt6), NULL, reason,
4810 					    pd2.af)) {
4811 						DPFPRINTF(PF_DEBUG_MISC,
4812 						    ("pf: ICMPv6 short opt\n"));
4813 						return (PF_DROP);
4814 					}
4815 					if (pd2.proto == IPPROTO_AH)
4816 						off2 += (opt6.ip6e_len + 2) * 4;
4817 					else
4818 						off2 += (opt6.ip6e_len + 1) * 8;
4819 					pd2.proto = opt6.ip6e_nxt;
4820 					/* goto the next header */
4821 					break;
4822 				}
4823 				default:
4824 					terminal++;
4825 					break;
4826 				}
4827 			} while (!terminal);
4828 			break;
4829 #endif /* INET6 */
4830 		}
4831 
4832 		if (PF_ANEQ(pd->dst, pd2.src, pd->af)) {
4833 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
4834 				printf("pf: BAD ICMP %d:%d outer dst: ",
4835 				    icmptype, icmpcode);
4836 				pf_print_host(pd->src, 0, pd->af);
4837 				printf(" -> ");
4838 				pf_print_host(pd->dst, 0, pd->af);
4839 				printf(" inner src: ");
4840 				pf_print_host(pd2.src, 0, pd2.af);
4841 				printf(" -> ");
4842 				pf_print_host(pd2.dst, 0, pd2.af);
4843 				printf("\n");
4844 			}
4845 			REASON_SET(reason, PFRES_BADSTATE);
4846 			return (PF_DROP);
4847 		}
4848 
4849 		switch (pd2.proto) {
4850 		case IPPROTO_TCP: {
4851 			struct tcphdr		 th;
4852 			u_int32_t		 seq;
4853 			struct pf_state_peer	*src, *dst;
4854 			u_int8_t		 dws;
4855 			int			 copyback = 0;
4856 
4857 			/*
4858 			 * Only the first 8 bytes of the TCP header can be
4859 			 * expected. Don't access any TCP header fields after
4860 			 * th_seq, an ackskew test is not possible.
4861 			 */
4862 			if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4863 			    pd2.af)) {
4864 				DPFPRINTF(PF_DEBUG_MISC,
4865 				    ("pf: ICMP error message too short "
4866 				    "(tcp)\n"));
4867 				return (PF_DROP);
4868 			}
4869 
4870 			key.af = pd2.af;
4871 			key.proto = IPPROTO_TCP;
4872 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4873 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4874 			key.port[pd2.sidx] = th.th_sport;
4875 			key.port[pd2.didx] = th.th_dport;
4876 
4877 			STATE_LOOKUP(kif, &key, direction, *state, pd);
4878 
4879 			if (direction == (*state)->direction) {
4880 				src = &(*state)->dst;
4881 				dst = &(*state)->src;
4882 			} else {
4883 				src = &(*state)->src;
4884 				dst = &(*state)->dst;
4885 			}
4886 
4887 			if (src->wscale && dst->wscale)
4888 				dws = dst->wscale & PF_WSCALE_MASK;
4889 			else
4890 				dws = 0;
4891 
4892 			/* Demodulate sequence number */
4893 			seq = ntohl(th.th_seq) - src->seqdiff;
4894 			if (src->seqdiff) {
4895 				pf_change_a(&th.th_seq, icmpsum,
4896 				    htonl(seq), 0);
4897 				copyback = 1;
4898 			}
4899 
4900 			if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4901 			    (!SEQ_GEQ(src->seqhi, seq) ||
4902 			    !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4903 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
4904 					printf("pf: BAD ICMP %d:%d ",
4905 					    icmptype, icmpcode);
4906 					pf_print_host(pd->src, 0, pd->af);
4907 					printf(" -> ");
4908 					pf_print_host(pd->dst, 0, pd->af);
4909 					printf(" state: ");
4910 					pf_print_state(*state);
4911 					printf(" seq=%u\n", seq);
4912 				}
4913 				REASON_SET(reason, PFRES_BADSTATE);
4914 				return (PF_DROP);
4915 			} else {
4916 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
4917 					printf("pf: OK ICMP %d:%d ",
4918 					    icmptype, icmpcode);
4919 					pf_print_host(pd->src, 0, pd->af);
4920 					printf(" -> ");
4921 					pf_print_host(pd->dst, 0, pd->af);
4922 					printf(" state: ");
4923 					pf_print_state(*state);
4924 					printf(" seq=%u\n", seq);
4925 				}
4926 			}
4927 
4928 			/* translate source/destination address, if necessary */
4929 			if ((*state)->key[PF_SK_WIRE] !=
4930 			    (*state)->key[PF_SK_STACK]) {
4931 				struct pf_state_key *nk =
4932 				    (*state)->key[pd->didx];
4933 
4934 				if (PF_ANEQ(pd2.src,
4935 				    &nk->addr[pd2.sidx], pd2.af) ||
4936 				    nk->port[pd2.sidx] != th.th_sport)
4937 					pf_change_icmp(pd2.src, &th.th_sport,
4938 					    daddr, &nk->addr[pd2.sidx],
4939 					    nk->port[pd2.sidx], NULL,
4940 					    pd2.ip_sum, icmpsum,
4941 					    pd->ip_sum, 0, pd2.af);
4942 
4943 				if (PF_ANEQ(pd2.dst,
4944 				    &nk->addr[pd2.didx], pd2.af) ||
4945 				    nk->port[pd2.didx] != th.th_dport)
4946 					pf_change_icmp(pd2.dst, &th.th_dport,
4947 					    saddr, &nk->addr[pd2.didx],
4948 					    nk->port[pd2.didx], NULL,
4949 					    pd2.ip_sum, icmpsum,
4950 					    pd->ip_sum, 0, pd2.af);
4951 				copyback = 1;
4952 			}
4953 
4954 			if (copyback) {
4955 				switch (pd2.af) {
4956 #ifdef INET
4957 				case AF_INET:
4958 					m_copyback(m, off, ICMP_MINLEN,
4959 					    (caddr_t )pd->hdr.icmp);
4960 					m_copyback(m, ipoff2, sizeof(h2),
4961 					    (caddr_t )&h2);
4962 					break;
4963 #endif /* INET */
4964 #ifdef INET6
4965 				case AF_INET6:
4966 					m_copyback(m, off,
4967 					    sizeof(struct icmp6_hdr),
4968 					    (caddr_t )pd->hdr.icmp6);
4969 					m_copyback(m, ipoff2, sizeof(h2_6),
4970 					    (caddr_t )&h2_6);
4971 					break;
4972 #endif /* INET6 */
4973 				}
4974 				m_copyback(m, off2, 8, (caddr_t)&th);
4975 			}
4976 
4977 			return (PF_PASS);
4978 			break;
4979 		}
4980 		case IPPROTO_UDP: {
4981 			struct udphdr		uh;
4982 
4983 			if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
4984 			    NULL, reason, pd2.af)) {
4985 				DPFPRINTF(PF_DEBUG_MISC,
4986 				    ("pf: ICMP error message too short "
4987 				    "(udp)\n"));
4988 				return (PF_DROP);
4989 			}
4990 
4991 			key.af = pd2.af;
4992 			key.proto = IPPROTO_UDP;
4993 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4994 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4995 			key.port[pd2.sidx] = uh.uh_sport;
4996 			key.port[pd2.didx] = uh.uh_dport;
4997 
4998 			STATE_LOOKUP(kif, &key, direction, *state, pd);
4999 
5000 			/* translate source/destination address, if necessary */
5001 			if ((*state)->key[PF_SK_WIRE] !=
5002 			    (*state)->key[PF_SK_STACK]) {
5003 				struct pf_state_key *nk =
5004 				    (*state)->key[pd->didx];
5005 
5006 				if (PF_ANEQ(pd2.src,
5007 				    &nk->addr[pd2.sidx], pd2.af) ||
5008 				    nk->port[pd2.sidx] != uh.uh_sport)
5009 					pf_change_icmp(pd2.src, &uh.uh_sport,
5010 					    daddr, &nk->addr[pd2.sidx],
5011 					    nk->port[pd2.sidx], &uh.uh_sum,
5012 					    pd2.ip_sum, icmpsum,
5013 					    pd->ip_sum, 1, pd2.af);
5014 
5015 				if (PF_ANEQ(pd2.dst,
5016 				    &nk->addr[pd2.didx], pd2.af) ||
5017 				    nk->port[pd2.didx] != uh.uh_dport)
5018 					pf_change_icmp(pd2.dst, &uh.uh_dport,
5019 					    saddr, &nk->addr[pd2.didx],
5020 					    nk->port[pd2.didx], &uh.uh_sum,
5021 					    pd2.ip_sum, icmpsum,
5022 					    pd->ip_sum, 1, pd2.af);
5023 
5024 				switch (pd2.af) {
5025 #ifdef INET
5026 				case AF_INET:
5027 					m_copyback(m, off, ICMP_MINLEN,
5028 					    (caddr_t )pd->hdr.icmp);
5029 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5030 					break;
5031 #endif /* INET */
5032 #ifdef INET6
5033 				case AF_INET6:
5034 					m_copyback(m, off,
5035 					    sizeof(struct icmp6_hdr),
5036 					    (caddr_t )pd->hdr.icmp6);
5037 					m_copyback(m, ipoff2, sizeof(h2_6),
5038 					    (caddr_t )&h2_6);
5039 					break;
5040 #endif /* INET6 */
5041 				}
5042 				m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
5043 			}
5044 			return (PF_PASS);
5045 			break;
5046 		}
5047 #ifdef INET
5048 		case IPPROTO_ICMP: {
5049 			struct icmp		iih;
5050 
5051 			if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
5052 			    NULL, reason, pd2.af)) {
5053 				DPFPRINTF(PF_DEBUG_MISC,
5054 				    ("pf: ICMP error message too short i"
5055 				    "(icmp)\n"));
5056 				return (PF_DROP);
5057 			}
5058 
5059 			key.af = pd2.af;
5060 			key.proto = IPPROTO_ICMP;
5061 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5062 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5063 			key.port[0] = key.port[1] = iih.icmp_id;
5064 
5065 			STATE_LOOKUP(kif, &key, direction, *state, pd);
5066 
5067 			/* translate source/destination address, if necessary */
5068 			if ((*state)->key[PF_SK_WIRE] !=
5069 			    (*state)->key[PF_SK_STACK]) {
5070 				struct pf_state_key *nk =
5071 				    (*state)->key[pd->didx];
5072 
5073 				if (PF_ANEQ(pd2.src,
5074 				    &nk->addr[pd2.sidx], pd2.af) ||
5075 				    nk->port[pd2.sidx] != iih.icmp_id)
5076 					pf_change_icmp(pd2.src, &iih.icmp_id,
5077 					    daddr, &nk->addr[pd2.sidx],
5078 					    nk->port[pd2.sidx], NULL,
5079 					    pd2.ip_sum, icmpsum,
5080 					    pd->ip_sum, 0, AF_INET);
5081 
5082 				if (PF_ANEQ(pd2.dst,
5083 				    &nk->addr[pd2.didx], pd2.af) ||
5084 				    nk->port[pd2.didx] != iih.icmp_id)
5085 					pf_change_icmp(pd2.dst, &iih.icmp_id,
5086 					    saddr, &nk->addr[pd2.didx],
5087 					    nk->port[pd2.didx], NULL,
5088 					    pd2.ip_sum, icmpsum,
5089 					    pd->ip_sum, 0, AF_INET);
5090 
5091 				m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
5092 				m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5093 				m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
5094 			}
5095 			return (PF_PASS);
5096 			break;
5097 		}
5098 #endif /* INET */
5099 #ifdef INET6
5100 		case IPPROTO_ICMPV6: {
5101 			struct icmp6_hdr	iih;
5102 
5103 			if (!pf_pull_hdr(m, off2, &iih,
5104 			    sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
5105 				DPFPRINTF(PF_DEBUG_MISC,
5106 				    ("pf: ICMP error message too short "
5107 				    "(icmp6)\n"));
5108 				return (PF_DROP);
5109 			}
5110 
5111 			key.af = pd2.af;
5112 			key.proto = IPPROTO_ICMPV6;
5113 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5114 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5115 			key.port[0] = key.port[1] = iih.icmp6_id;
5116 
5117 			STATE_LOOKUP(kif, &key, direction, *state, pd);
5118 
5119 			/* translate source/destination address, if necessary */
5120 			if ((*state)->key[PF_SK_WIRE] !=
5121 			    (*state)->key[PF_SK_STACK]) {
5122 				struct pf_state_key *nk =
5123 				    (*state)->key[pd->didx];
5124 
5125 				if (PF_ANEQ(pd2.src,
5126 				    &nk->addr[pd2.sidx], pd2.af) ||
5127 				    nk->port[pd2.sidx] != iih.icmp6_id)
5128 					pf_change_icmp(pd2.src, &iih.icmp6_id,
5129 					    daddr, &nk->addr[pd2.sidx],
5130 					    nk->port[pd2.sidx], NULL,
5131 					    pd2.ip_sum, icmpsum,
5132 					    pd->ip_sum, 0, AF_INET6);
5133 
5134 				if (PF_ANEQ(pd2.dst,
5135 				    &nk->addr[pd2.didx], pd2.af) ||
5136 				    nk->port[pd2.didx] != iih.icmp6_id)
5137 					pf_change_icmp(pd2.dst, &iih.icmp6_id,
5138 					    saddr, &nk->addr[pd2.didx],
5139 					    nk->port[pd2.didx], NULL,
5140 					    pd2.ip_sum, icmpsum,
5141 					    pd->ip_sum, 0, AF_INET6);
5142 
5143 				m_copyback(m, off, sizeof(struct icmp6_hdr),
5144 				    (caddr_t)pd->hdr.icmp6);
5145 				m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
5146 				m_copyback(m, off2, sizeof(struct icmp6_hdr),
5147 				    (caddr_t)&iih);
5148 			}
5149 			return (PF_PASS);
5150 			break;
5151 		}
5152 #endif /* INET6 */
5153 		default: {
5154 			key.af = pd2.af;
5155 			key.proto = pd2.proto;
5156 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5157 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5158 			key.port[0] = key.port[1] = 0;
5159 
5160 			STATE_LOOKUP(kif, &key, direction, *state, pd);
5161 
5162 			/* translate source/destination address, if necessary */
5163 			if ((*state)->key[PF_SK_WIRE] !=
5164 			    (*state)->key[PF_SK_STACK]) {
5165 				struct pf_state_key *nk =
5166 				    (*state)->key[pd->didx];
5167 
5168 				if (PF_ANEQ(pd2.src,
5169 				    &nk->addr[pd2.sidx], pd2.af))
5170 					pf_change_icmp(pd2.src, NULL, daddr,
5171 					    &nk->addr[pd2.sidx], 0, NULL,
5172 					    pd2.ip_sum, icmpsum,
5173 					    pd->ip_sum, 0, pd2.af);
5174 
5175 				if (PF_ANEQ(pd2.dst,
5176 				    &nk->addr[pd2.didx], pd2.af))
5177 					pf_change_icmp(pd2.dst, NULL, saddr,
5178 					    &nk->addr[pd2.didx], 0, NULL,
5179 					    pd2.ip_sum, icmpsum,
5180 					    pd->ip_sum, 0, pd2.af);
5181 
5182 				switch (pd2.af) {
5183 #ifdef INET
5184 				case AF_INET:
5185 					m_copyback(m, off, ICMP_MINLEN,
5186 					    (caddr_t)pd->hdr.icmp);
5187 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5188 					break;
5189 #endif /* INET */
5190 #ifdef INET6
5191 				case AF_INET6:
5192 					m_copyback(m, off,
5193 					    sizeof(struct icmp6_hdr),
5194 					    (caddr_t )pd->hdr.icmp6);
5195 					m_copyback(m, ipoff2, sizeof(h2_6),
5196 					    (caddr_t )&h2_6);
5197 					break;
5198 #endif /* INET6 */
5199 				}
5200 			}
5201 			return (PF_PASS);
5202 			break;
5203 		}
5204 		}
5205 	}
5206 }
5207 
5208 static int
5209 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
5210     struct mbuf *m, struct pf_pdesc *pd)
5211 {
5212 	struct pf_state_peer	*src, *dst;
5213 	struct pf_state_key_cmp	 key;
5214 
5215 	bzero(&key, sizeof(key));
5216 	key.af = pd->af;
5217 	key.proto = pd->proto;
5218 	if (direction == PF_IN)	{
5219 		PF_ACPY(&key.addr[0], pd->src, key.af);
5220 		PF_ACPY(&key.addr[1], pd->dst, key.af);
5221 		key.port[0] = key.port[1] = 0;
5222 	} else {
5223 		PF_ACPY(&key.addr[1], pd->src, key.af);
5224 		PF_ACPY(&key.addr[0], pd->dst, key.af);
5225 		key.port[1] = key.port[0] = 0;
5226 	}
5227 
5228 	STATE_LOOKUP(kif, &key, direction, *state, pd);
5229 
5230 	if (direction == (*state)->direction) {
5231 		src = &(*state)->src;
5232 		dst = &(*state)->dst;
5233 	} else {
5234 		src = &(*state)->dst;
5235 		dst = &(*state)->src;
5236 	}
5237 
5238 	/* update states */
5239 	if (src->state < PFOTHERS_SINGLE)
5240 		src->state = PFOTHERS_SINGLE;
5241 	if (dst->state == PFOTHERS_SINGLE)
5242 		dst->state = PFOTHERS_MULTIPLE;
5243 
5244 	/* update expire time */
5245 	(*state)->expire = time_uptime;
5246 	if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5247 		(*state)->timeout = PFTM_OTHER_MULTIPLE;
5248 	else
5249 		(*state)->timeout = PFTM_OTHER_SINGLE;
5250 
5251 	/* translate source/destination address, if necessary */
5252 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5253 		struct pf_state_key *nk = (*state)->key[pd->didx];
5254 
5255 		KASSERT(nk, ("%s: nk is null", __func__));
5256 		KASSERT(pd, ("%s: pd is null", __func__));
5257 		KASSERT(pd->src, ("%s: pd->src is null", __func__));
5258 		KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5259 		switch (pd->af) {
5260 #ifdef INET
5261 		case AF_INET:
5262 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5263 				pf_change_a(&pd->src->v4.s_addr,
5264 				    pd->ip_sum,
5265 				    nk->addr[pd->sidx].v4.s_addr,
5266 				    0);
5267 
5268 
5269 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5270 				pf_change_a(&pd->dst->v4.s_addr,
5271 				    pd->ip_sum,
5272 				    nk->addr[pd->didx].v4.s_addr,
5273 				    0);
5274 
5275 			break;
5276 #endif /* INET */
5277 #ifdef INET6
5278 		case AF_INET6:
5279 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5280 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5281 
5282 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5283 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5284 #endif /* INET6 */
5285 		}
5286 	}
5287 	return (PF_PASS);
5288 }
5289 
5290 /*
5291  * ipoff and off are measured from the start of the mbuf chain.
5292  * h must be at "ipoff" on the mbuf chain.
5293  */
5294 void *
5295 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5296     u_short *actionp, u_short *reasonp, sa_family_t af)
5297 {
5298 	switch (af) {
5299 #ifdef INET
5300 	case AF_INET: {
5301 		struct ip	*h = mtod(m, struct ip *);
5302 		u_int16_t	 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5303 
5304 		if (fragoff) {
5305 			if (fragoff >= len)
5306 				ACTION_SET(actionp, PF_PASS);
5307 			else {
5308 				ACTION_SET(actionp, PF_DROP);
5309 				REASON_SET(reasonp, PFRES_FRAG);
5310 			}
5311 			return (NULL);
5312 		}
5313 		if (m->m_pkthdr.len < off + len ||
5314 		    ntohs(h->ip_len) < off + len) {
5315 			ACTION_SET(actionp, PF_DROP);
5316 			REASON_SET(reasonp, PFRES_SHORT);
5317 			return (NULL);
5318 		}
5319 		break;
5320 	}
5321 #endif /* INET */
5322 #ifdef INET6
5323 	case AF_INET6: {
5324 		struct ip6_hdr	*h = mtod(m, struct ip6_hdr *);
5325 
5326 		if (m->m_pkthdr.len < off + len ||
5327 		    (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5328 		    (unsigned)(off + len)) {
5329 			ACTION_SET(actionp, PF_DROP);
5330 			REASON_SET(reasonp, PFRES_SHORT);
5331 			return (NULL);
5332 		}
5333 		break;
5334 	}
5335 #endif /* INET6 */
5336 	}
5337 	m_copydata(m, off, len, p);
5338 	return (p);
5339 }
5340 
5341 #ifdef RADIX_MPATH
5342 static int
5343 pf_routable_oldmpath(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5344     int rtableid)
5345 {
5346 	struct radix_node_head	*rnh;
5347 	struct sockaddr_in	*dst;
5348 	int			 ret = 1;
5349 	int			 check_mpath;
5350 #ifdef INET6
5351 	struct sockaddr_in6	*dst6;
5352 	struct route_in6	 ro;
5353 #else
5354 	struct route		 ro;
5355 #endif
5356 	struct radix_node	*rn;
5357 	struct rtentry		*rt;
5358 	struct ifnet		*ifp;
5359 
5360 	check_mpath = 0;
5361 	/* XXX: stick to table 0 for now */
5362 	rnh = rt_tables_get_rnh(0, af);
5363 	if (rnh != NULL && rn_mpath_capable(rnh))
5364 		check_mpath = 1;
5365 	bzero(&ro, sizeof(ro));
5366 	switch (af) {
5367 	case AF_INET:
5368 		dst = satosin(&ro.ro_dst);
5369 		dst->sin_family = AF_INET;
5370 		dst->sin_len = sizeof(*dst);
5371 		dst->sin_addr = addr->v4;
5372 		break;
5373 #ifdef INET6
5374 	case AF_INET6:
5375 		/*
5376 		 * Skip check for addresses with embedded interface scope,
5377 		 * as they would always match anyway.
5378 		 */
5379 		if (IN6_IS_SCOPE_EMBED(&addr->v6))
5380 			goto out;
5381 		dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5382 		dst6->sin6_family = AF_INET6;
5383 		dst6->sin6_len = sizeof(*dst6);
5384 		dst6->sin6_addr = addr->v6;
5385 		break;
5386 #endif /* INET6 */
5387 	default:
5388 		return (0);
5389 	}
5390 
5391 	/* Skip checks for ipsec interfaces */
5392 	if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5393 		goto out;
5394 
5395 	switch (af) {
5396 #ifdef INET6
5397 	case AF_INET6:
5398 		in6_rtalloc_ign(&ro, 0, rtableid);
5399 		break;
5400 #endif
5401 #ifdef INET
5402 	case AF_INET:
5403 		in_rtalloc_ign((struct route *)&ro, 0, rtableid);
5404 		break;
5405 #endif
5406 	}
5407 
5408 	if (ro.ro_rt != NULL) {
5409 		/* No interface given, this is a no-route check */
5410 		if (kif == NULL)
5411 			goto out;
5412 
5413 		if (kif->pfik_ifp == NULL) {
5414 			ret = 0;
5415 			goto out;
5416 		}
5417 
5418 		/* Perform uRPF check if passed input interface */
5419 		ret = 0;
5420 		rn = (struct radix_node *)ro.ro_rt;
5421 		do {
5422 			rt = (struct rtentry *)rn;
5423 			ifp = rt->rt_ifp;
5424 
5425 			if (kif->pfik_ifp == ifp)
5426 				ret = 1;
5427 			rn = rn_mpath_next(rn);
5428 		} while (check_mpath == 1 && rn != NULL && ret == 0);
5429 	} else
5430 		ret = 0;
5431 out:
5432 	if (ro.ro_rt != NULL)
5433 		RTFREE(ro.ro_rt);
5434 	return (ret);
5435 }
5436 #endif
5437 
5438 int
5439 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5440     int rtableid)
5441 {
5442 #ifdef INET
5443 	struct nhop4_basic	nh4;
5444 #endif
5445 #ifdef INET6
5446 	struct nhop6_basic	nh6;
5447 #endif
5448 	struct ifnet		*ifp;
5449 #ifdef RADIX_MPATH
5450 	struct radix_node_head	*rnh;
5451 
5452 	/* XXX: stick to table 0 for now */
5453 	rnh = rt_tables_get_rnh(0, af);
5454 	if (rnh != NULL && rn_mpath_capable(rnh))
5455 		return (pf_routable_oldmpath(addr, af, kif, rtableid));
5456 #endif
5457 	/*
5458 	 * Skip check for addresses with embedded interface scope,
5459 	 * as they would always match anyway.
5460 	 */
5461 	if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
5462 		return (1);
5463 
5464 	if (af != AF_INET && af != AF_INET6)
5465 		return (0);
5466 
5467 	/* Skip checks for ipsec interfaces */
5468 	if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5469 		return (1);
5470 
5471 	ifp = NULL;
5472 
5473 	switch (af) {
5474 #ifdef INET6
5475 	case AF_INET6:
5476 		if (fib6_lookup_nh_basic(rtableid, &addr->v6, 0, 0, 0, &nh6)!=0)
5477 			return (0);
5478 		ifp = nh6.nh_ifp;
5479 		break;
5480 #endif
5481 #ifdef INET
5482 	case AF_INET:
5483 		if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) != 0)
5484 			return (0);
5485 		ifp = nh4.nh_ifp;
5486 		break;
5487 #endif
5488 	}
5489 
5490 	/* No interface given, this is a no-route check */
5491 	if (kif == NULL)
5492 		return (1);
5493 
5494 	if (kif->pfik_ifp == NULL)
5495 		return (0);
5496 
5497 	/* Perform uRPF check if passed input interface */
5498 	if (kif->pfik_ifp == ifp)
5499 		return (1);
5500 	return (0);
5501 }
5502 
5503 #ifdef INET
5504 static void
5505 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5506     struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp)
5507 {
5508 	struct mbuf		*m0, *m1;
5509 	struct sockaddr_in	dst;
5510 	struct ip		*ip;
5511 	struct ifnet		*ifp = NULL;
5512 	struct pf_addr		 naddr;
5513 	struct pf_src_node	*sn = NULL;
5514 	int			 error = 0;
5515 	uint16_t		 ip_len, ip_off;
5516 
5517 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5518 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5519 	    __func__));
5520 
5521 	if ((pd->pf_mtag == NULL &&
5522 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5523 	    pd->pf_mtag->routed++ > 3) {
5524 		m0 = *m;
5525 		*m = NULL;
5526 		goto bad_locked;
5527 	}
5528 
5529 	if (r->rt == PF_DUPTO) {
5530 		if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5531 			if (s)
5532 				PF_STATE_UNLOCK(s);
5533 			return;
5534 		}
5535 	} else {
5536 		if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5537 			if (s)
5538 				PF_STATE_UNLOCK(s);
5539 			return;
5540 		}
5541 		m0 = *m;
5542 	}
5543 
5544 	ip = mtod(m0, struct ip *);
5545 
5546 	bzero(&dst, sizeof(dst));
5547 	dst.sin_family = AF_INET;
5548 	dst.sin_len = sizeof(dst);
5549 	dst.sin_addr = ip->ip_dst;
5550 
5551 	bzero(&naddr, sizeof(naddr));
5552 
5553 	if (TAILQ_EMPTY(&r->rpool.list)) {
5554 		DPFPRINTF(PF_DEBUG_URGENT,
5555 		    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5556 		goto bad_locked;
5557 	}
5558 	if (s == NULL) {
5559 		pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5560 		    &naddr, NULL, &sn);
5561 		if (!PF_AZERO(&naddr, AF_INET))
5562 			dst.sin_addr.s_addr = naddr.v4.s_addr;
5563 		ifp = r->rpool.cur->kif ?
5564 		    r->rpool.cur->kif->pfik_ifp : NULL;
5565 	} else {
5566 		if (!PF_AZERO(&s->rt_addr, AF_INET))
5567 			dst.sin_addr.s_addr =
5568 			    s->rt_addr.v4.s_addr;
5569 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5570 		PF_STATE_UNLOCK(s);
5571 	}
5572 	if (ifp == NULL)
5573 		goto bad;
5574 
5575 	if (oifp != ifp) {
5576 		if (pf_test(PF_OUT, 0, ifp, &m0, inp) != PF_PASS)
5577 			goto bad;
5578 		else if (m0 == NULL)
5579 			goto done;
5580 		if (m0->m_len < sizeof(struct ip)) {
5581 			DPFPRINTF(PF_DEBUG_URGENT,
5582 			    ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5583 			goto bad;
5584 		}
5585 		ip = mtod(m0, struct ip *);
5586 	}
5587 
5588 	if (ifp->if_flags & IFF_LOOPBACK)
5589 		m0->m_flags |= M_SKIP_FIREWALL;
5590 
5591 	ip_len = ntohs(ip->ip_len);
5592 	ip_off = ntohs(ip->ip_off);
5593 
5594 	/* Copied from FreeBSD 10.0-CURRENT ip_output. */
5595 	m0->m_pkthdr.csum_flags |= CSUM_IP;
5596 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5597 		in_delayed_cksum(m0);
5598 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5599 	}
5600 #ifdef SCTP
5601 	if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5602 		sctp_delayed_cksum(m0, (uint32_t)(ip->ip_hl << 2));
5603 		m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5604 	}
5605 #endif
5606 
5607 	/*
5608 	 * If small enough for interface, or the interface will take
5609 	 * care of the fragmentation for us, we can just send directly.
5610 	 */
5611 	if (ip_len <= ifp->if_mtu ||
5612 	    (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
5613 		ip->ip_sum = 0;
5614 		if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5615 			ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5616 			m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5617 		}
5618 		m_clrprotoflags(m0);	/* Avoid confusing lower layers. */
5619 		error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5620 		goto done;
5621 	}
5622 
5623 	/* Balk when DF bit is set or the interface didn't support TSO. */
5624 	if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5625 		error = EMSGSIZE;
5626 		KMOD_IPSTAT_INC(ips_cantfrag);
5627 		if (r->rt != PF_DUPTO) {
5628 			icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5629 			    ifp->if_mtu);
5630 			goto done;
5631 		} else
5632 			goto bad;
5633 	}
5634 
5635 	error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5636 	if (error)
5637 		goto bad;
5638 
5639 	for (; m0; m0 = m1) {
5640 		m1 = m0->m_nextpkt;
5641 		m0->m_nextpkt = NULL;
5642 		if (error == 0) {
5643 			m_clrprotoflags(m0);
5644 			error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5645 		} else
5646 			m_freem(m0);
5647 	}
5648 
5649 	if (error == 0)
5650 		KMOD_IPSTAT_INC(ips_fragmented);
5651 
5652 done:
5653 	if (r->rt != PF_DUPTO)
5654 		*m = NULL;
5655 	return;
5656 
5657 bad_locked:
5658 	if (s)
5659 		PF_STATE_UNLOCK(s);
5660 bad:
5661 	m_freem(m0);
5662 	goto done;
5663 }
5664 #endif /* INET */
5665 
5666 #ifdef INET6
5667 static void
5668 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5669     struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp)
5670 {
5671 	struct mbuf		*m0;
5672 	struct sockaddr_in6	dst;
5673 	struct ip6_hdr		*ip6;
5674 	struct ifnet		*ifp = NULL;
5675 	struct pf_addr		 naddr;
5676 	struct pf_src_node	*sn = NULL;
5677 
5678 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5679 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5680 	    __func__));
5681 
5682 	if ((pd->pf_mtag == NULL &&
5683 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5684 	    pd->pf_mtag->routed++ > 3) {
5685 		m0 = *m;
5686 		*m = NULL;
5687 		goto bad_locked;
5688 	}
5689 
5690 	if (r->rt == PF_DUPTO) {
5691 		if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5692 			if (s)
5693 				PF_STATE_UNLOCK(s);
5694 			return;
5695 		}
5696 	} else {
5697 		if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5698 			if (s)
5699 				PF_STATE_UNLOCK(s);
5700 			return;
5701 		}
5702 		m0 = *m;
5703 	}
5704 
5705 	ip6 = mtod(m0, struct ip6_hdr *);
5706 
5707 	bzero(&dst, sizeof(dst));
5708 	dst.sin6_family = AF_INET6;
5709 	dst.sin6_len = sizeof(dst);
5710 	dst.sin6_addr = ip6->ip6_dst;
5711 
5712 	bzero(&naddr, sizeof(naddr));
5713 
5714 	if (TAILQ_EMPTY(&r->rpool.list)) {
5715 		DPFPRINTF(PF_DEBUG_URGENT,
5716 		    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5717 		goto bad_locked;
5718 	}
5719 	if (s == NULL) {
5720 		pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5721 		    &naddr, NULL, &sn);
5722 		if (!PF_AZERO(&naddr, AF_INET6))
5723 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5724 			    &naddr, AF_INET6);
5725 		ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5726 	} else {
5727 		if (!PF_AZERO(&s->rt_addr, AF_INET6))
5728 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5729 			    &s->rt_addr, AF_INET6);
5730 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5731 	}
5732 
5733 	if (s)
5734 		PF_STATE_UNLOCK(s);
5735 
5736 	if (ifp == NULL)
5737 		goto bad;
5738 
5739 	if (oifp != ifp) {
5740 		if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp) != PF_PASS)
5741 			goto bad;
5742 		else if (m0 == NULL)
5743 			goto done;
5744 		if (m0->m_len < sizeof(struct ip6_hdr)) {
5745 			DPFPRINTF(PF_DEBUG_URGENT,
5746 			    ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5747 			    __func__));
5748 			goto bad;
5749 		}
5750 		ip6 = mtod(m0, struct ip6_hdr *);
5751 	}
5752 
5753 	if (ifp->if_flags & IFF_LOOPBACK)
5754 		m0->m_flags |= M_SKIP_FIREWALL;
5755 
5756 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
5757 	    ~ifp->if_hwassist) {
5758 		uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
5759 		in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
5760 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
5761 	}
5762 
5763 	/*
5764 	 * If the packet is too large for the outgoing interface,
5765 	 * send back an icmp6 error.
5766 	 */
5767 	if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5768 		dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5769 	if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5770 		nd6_output_ifp(ifp, ifp, m0, &dst, NULL);
5771 	else {
5772 		in6_ifstat_inc(ifp, ifs6_in_toobig);
5773 		if (r->rt != PF_DUPTO)
5774 			icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5775 		else
5776 			goto bad;
5777 	}
5778 
5779 done:
5780 	if (r->rt != PF_DUPTO)
5781 		*m = NULL;
5782 	return;
5783 
5784 bad_locked:
5785 	if (s)
5786 		PF_STATE_UNLOCK(s);
5787 bad:
5788 	m_freem(m0);
5789 	goto done;
5790 }
5791 #endif /* INET6 */
5792 
5793 /*
5794  * FreeBSD supports cksum offloads for the following drivers.
5795  *  em(4), fxp(4), lge(4), ndis(4), nge(4), re(4), ti(4), txp(4), xl(4)
5796  *
5797  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5798  *  network driver performed cksum including pseudo header, need to verify
5799  *   csum_data
5800  * CSUM_DATA_VALID :
5801  *  network driver performed cksum, needs to additional pseudo header
5802  *  cksum computation with partial csum_data(i.e. lack of H/W support for
5803  *  pseudo header, for instance hme(4), sk(4) and possibly gem(4))
5804  *
5805  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5806  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5807  * TCP/UDP layer.
5808  * Also, set csum_data to 0xffff to force cksum validation.
5809  */
5810 static int
5811 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5812 {
5813 	u_int16_t sum = 0;
5814 	int hw_assist = 0;
5815 	struct ip *ip;
5816 
5817 	if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5818 		return (1);
5819 	if (m->m_pkthdr.len < off + len)
5820 		return (1);
5821 
5822 	switch (p) {
5823 	case IPPROTO_TCP:
5824 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5825 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5826 				sum = m->m_pkthdr.csum_data;
5827 			} else {
5828 				ip = mtod(m, struct ip *);
5829 				sum = in_pseudo(ip->ip_src.s_addr,
5830 				ip->ip_dst.s_addr, htonl((u_short)len +
5831 				m->m_pkthdr.csum_data + IPPROTO_TCP));
5832 			}
5833 			sum ^= 0xffff;
5834 			++hw_assist;
5835 		}
5836 		break;
5837 	case IPPROTO_UDP:
5838 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5839 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5840 				sum = m->m_pkthdr.csum_data;
5841 			} else {
5842 				ip = mtod(m, struct ip *);
5843 				sum = in_pseudo(ip->ip_src.s_addr,
5844 				ip->ip_dst.s_addr, htonl((u_short)len +
5845 				m->m_pkthdr.csum_data + IPPROTO_UDP));
5846 			}
5847 			sum ^= 0xffff;
5848 			++hw_assist;
5849 		}
5850 		break;
5851 	case IPPROTO_ICMP:
5852 #ifdef INET6
5853 	case IPPROTO_ICMPV6:
5854 #endif /* INET6 */
5855 		break;
5856 	default:
5857 		return (1);
5858 	}
5859 
5860 	if (!hw_assist) {
5861 		switch (af) {
5862 		case AF_INET:
5863 			if (p == IPPROTO_ICMP) {
5864 				if (m->m_len < off)
5865 					return (1);
5866 				m->m_data += off;
5867 				m->m_len -= off;
5868 				sum = in_cksum(m, len);
5869 				m->m_data -= off;
5870 				m->m_len += off;
5871 			} else {
5872 				if (m->m_len < sizeof(struct ip))
5873 					return (1);
5874 				sum = in4_cksum(m, p, off, len);
5875 			}
5876 			break;
5877 #ifdef INET6
5878 		case AF_INET6:
5879 			if (m->m_len < sizeof(struct ip6_hdr))
5880 				return (1);
5881 			sum = in6_cksum(m, p, off, len);
5882 			break;
5883 #endif /* INET6 */
5884 		default:
5885 			return (1);
5886 		}
5887 	}
5888 	if (sum) {
5889 		switch (p) {
5890 		case IPPROTO_TCP:
5891 		    {
5892 			KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5893 			break;
5894 		    }
5895 		case IPPROTO_UDP:
5896 		    {
5897 			KMOD_UDPSTAT_INC(udps_badsum);
5898 			break;
5899 		    }
5900 #ifdef INET
5901 		case IPPROTO_ICMP:
5902 		    {
5903 			KMOD_ICMPSTAT_INC(icps_checksum);
5904 			break;
5905 		    }
5906 #endif
5907 #ifdef INET6
5908 		case IPPROTO_ICMPV6:
5909 		    {
5910 			KMOD_ICMP6STAT_INC(icp6s_checksum);
5911 			break;
5912 		    }
5913 #endif /* INET6 */
5914 		}
5915 		return (1);
5916 	} else {
5917 		if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5918 			m->m_pkthdr.csum_flags |=
5919 			    (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5920 			m->m_pkthdr.csum_data = 0xffff;
5921 		}
5922 	}
5923 	return (0);
5924 }
5925 
5926 
5927 #ifdef INET
5928 int
5929 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5930 {
5931 	struct pfi_kif		*kif;
5932 	u_short			 action, reason = 0, log = 0;
5933 	struct mbuf		*m = *m0;
5934 	struct ip		*h = NULL;
5935 	struct m_tag		*ipfwtag;
5936 	struct pf_rule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5937 	struct pf_state		*s = NULL;
5938 	struct pf_ruleset	*ruleset = NULL;
5939 	struct pf_pdesc		 pd;
5940 	int			 off, dirndx, pqid = 0;
5941 
5942 	PF_RULES_RLOCK_TRACKER;
5943 
5944 	M_ASSERTPKTHDR(m);
5945 
5946 	if (!V_pf_status.running)
5947 		return (PF_PASS);
5948 
5949 	memset(&pd, 0, sizeof(pd));
5950 
5951 	kif = (struct pfi_kif *)ifp->if_pf_kif;
5952 
5953 	if (kif == NULL) {
5954 		DPFPRINTF(PF_DEBUG_URGENT,
5955 		    ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5956 		return (PF_DROP);
5957 	}
5958 	if (kif->pfik_flags & PFI_IFLAG_SKIP)
5959 		return (PF_PASS);
5960 
5961 	if (m->m_flags & M_SKIP_FIREWALL)
5962 		return (PF_PASS);
5963 
5964 	pd.pf_mtag = pf_find_mtag(m);
5965 
5966 	PF_RULES_RLOCK();
5967 
5968 	if (ip_divert_ptr != NULL &&
5969 	    ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5970 		struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5971 		if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5972 			if (pd.pf_mtag == NULL &&
5973 			    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5974 				action = PF_DROP;
5975 				goto done;
5976 			}
5977 			pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5978 			m_tag_delete(m, ipfwtag);
5979 		}
5980 		if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5981 			m->m_flags |= M_FASTFWD_OURS;
5982 			pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5983 		}
5984 	} else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5985 		/* We do IP header normalization and packet reassembly here */
5986 		action = PF_DROP;
5987 		goto done;
5988 	}
5989 	m = *m0;	/* pf_normalize messes with m0 */
5990 	h = mtod(m, struct ip *);
5991 
5992 	off = h->ip_hl << 2;
5993 	if (off < (int)sizeof(struct ip)) {
5994 		action = PF_DROP;
5995 		REASON_SET(&reason, PFRES_SHORT);
5996 		log = 1;
5997 		goto done;
5998 	}
5999 
6000 	pd.src = (struct pf_addr *)&h->ip_src;
6001 	pd.dst = (struct pf_addr *)&h->ip_dst;
6002 	pd.sport = pd.dport = NULL;
6003 	pd.ip_sum = &h->ip_sum;
6004 	pd.proto_sum = NULL;
6005 	pd.proto = h->ip_p;
6006 	pd.dir = dir;
6007 	pd.sidx = (dir == PF_IN) ? 0 : 1;
6008 	pd.didx = (dir == PF_IN) ? 1 : 0;
6009 	pd.af = AF_INET;
6010 	pd.tos = h->ip_tos & ~IPTOS_ECN_MASK;
6011 	pd.tot_len = ntohs(h->ip_len);
6012 
6013 	/* handle fragments that didn't get reassembled by normalization */
6014 	if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
6015 		action = pf_test_fragment(&r, dir, kif, m, h,
6016 		    &pd, &a, &ruleset);
6017 		goto done;
6018 	}
6019 
6020 	switch (h->ip_p) {
6021 
6022 	case IPPROTO_TCP: {
6023 		struct tcphdr	th;
6024 
6025 		pd.hdr.tcp = &th;
6026 		if (!pf_pull_hdr(m, off, &th, sizeof(th),
6027 		    &action, &reason, AF_INET)) {
6028 			log = action != PF_PASS;
6029 			goto done;
6030 		}
6031 		pd.p_len = pd.tot_len - off - (th.th_off << 2);
6032 		if ((th.th_flags & TH_ACK) && pd.p_len == 0)
6033 			pqid = 1;
6034 		action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6035 		if (action == PF_DROP)
6036 			goto done;
6037 		action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6038 		    &reason);
6039 		if (action == PF_PASS) {
6040 			if (V_pfsync_update_state_ptr != NULL)
6041 				V_pfsync_update_state_ptr(s);
6042 			r = s->rule.ptr;
6043 			a = s->anchor.ptr;
6044 			log = s->log;
6045 		} else if (s == NULL)
6046 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6047 			    &a, &ruleset, inp);
6048 		break;
6049 	}
6050 
6051 	case IPPROTO_UDP: {
6052 		struct udphdr	uh;
6053 
6054 		pd.hdr.udp = &uh;
6055 		if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6056 		    &action, &reason, AF_INET)) {
6057 			log = action != PF_PASS;
6058 			goto done;
6059 		}
6060 		if (uh.uh_dport == 0 ||
6061 		    ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6062 		    ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6063 			action = PF_DROP;
6064 			REASON_SET(&reason, PFRES_SHORT);
6065 			goto done;
6066 		}
6067 		action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6068 		if (action == PF_PASS) {
6069 			if (V_pfsync_update_state_ptr != NULL)
6070 				V_pfsync_update_state_ptr(s);
6071 			r = s->rule.ptr;
6072 			a = s->anchor.ptr;
6073 			log = s->log;
6074 		} else if (s == NULL)
6075 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6076 			    &a, &ruleset, inp);
6077 		break;
6078 	}
6079 
6080 	case IPPROTO_ICMP: {
6081 		struct icmp	ih;
6082 
6083 		pd.hdr.icmp = &ih;
6084 		if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
6085 		    &action, &reason, AF_INET)) {
6086 			log = action != PF_PASS;
6087 			goto done;
6088 		}
6089 		action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
6090 		    &reason);
6091 		if (action == PF_PASS) {
6092 			if (V_pfsync_update_state_ptr != NULL)
6093 				V_pfsync_update_state_ptr(s);
6094 			r = s->rule.ptr;
6095 			a = s->anchor.ptr;
6096 			log = s->log;
6097 		} else if (s == NULL)
6098 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6099 			    &a, &ruleset, inp);
6100 		break;
6101 	}
6102 
6103 #ifdef INET6
6104 	case IPPROTO_ICMPV6: {
6105 		action = PF_DROP;
6106 		DPFPRINTF(PF_DEBUG_MISC,
6107 		    ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
6108 		goto done;
6109 	}
6110 #endif
6111 
6112 	default:
6113 		action = pf_test_state_other(&s, dir, kif, m, &pd);
6114 		if (action == PF_PASS) {
6115 			if (V_pfsync_update_state_ptr != NULL)
6116 				V_pfsync_update_state_ptr(s);
6117 			r = s->rule.ptr;
6118 			a = s->anchor.ptr;
6119 			log = s->log;
6120 		} else if (s == NULL)
6121 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6122 			    &a, &ruleset, inp);
6123 		break;
6124 	}
6125 
6126 done:
6127 	PF_RULES_RUNLOCK();
6128 	if (action == PF_PASS && h->ip_hl > 5 &&
6129 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6130 		action = PF_DROP;
6131 		REASON_SET(&reason, PFRES_IPOPTIONS);
6132 		log = r->log;
6133 		DPFPRINTF(PF_DEBUG_MISC,
6134 		    ("pf: dropping packet with ip options\n"));
6135 	}
6136 
6137 	if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6138 		action = PF_DROP;
6139 		REASON_SET(&reason, PFRES_MEMORY);
6140 	}
6141 	if (r->rtableid >= 0)
6142 		M_SETFIB(m, r->rtableid);
6143 
6144 	if (r->scrub_flags & PFSTATE_SETPRIO) {
6145 		if (pd.tos & IPTOS_LOWDELAY)
6146 			pqid = 1;
6147 		if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6148 			action = PF_DROP;
6149 			REASON_SET(&reason, PFRES_MEMORY);
6150 			log = 1;
6151 			DPFPRINTF(PF_DEBUG_MISC,
6152 			    ("pf: failed to allocate 802.1q mtag\n"));
6153 		}
6154 	}
6155 
6156 #ifdef ALTQ
6157 	if (action == PF_PASS && r->qid) {
6158 		if (pd.pf_mtag == NULL &&
6159 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6160 			action = PF_DROP;
6161 			REASON_SET(&reason, PFRES_MEMORY);
6162 		} else {
6163 			if (s != NULL)
6164 				pd.pf_mtag->qid_hash = pf_state_hash(s);
6165 			if (pqid || (pd.tos & IPTOS_LOWDELAY))
6166 				pd.pf_mtag->qid = r->pqid;
6167 			else
6168 				pd.pf_mtag->qid = r->qid;
6169 			/* Add hints for ecn. */
6170 			pd.pf_mtag->hdr = h;
6171 		}
6172 
6173 	}
6174 #endif /* ALTQ */
6175 
6176 	/*
6177 	 * connections redirected to loopback should not match sockets
6178 	 * bound specifically to loopback due to security implications,
6179 	 * see tcp_input() and in_pcblookup_listen().
6180 	 */
6181 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6182 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6183 	    (s->nat_rule.ptr->action == PF_RDR ||
6184 	    s->nat_rule.ptr->action == PF_BINAT) &&
6185 	    IN_LOOPBACK(ntohl(pd.dst->v4.s_addr)))
6186 		m->m_flags |= M_SKIP_FIREWALL;
6187 
6188 	if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
6189 	    !PACKET_LOOPED(&pd)) {
6190 
6191 		ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
6192 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
6193 		if (ipfwtag != NULL) {
6194 			((struct ipfw_rule_ref *)(ipfwtag+1))->info =
6195 			    ntohs(r->divert.port);
6196 			((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
6197 
6198 			if (s)
6199 				PF_STATE_UNLOCK(s);
6200 
6201 			m_tag_prepend(m, ipfwtag);
6202 			if (m->m_flags & M_FASTFWD_OURS) {
6203 				if (pd.pf_mtag == NULL &&
6204 				    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6205 					action = PF_DROP;
6206 					REASON_SET(&reason, PFRES_MEMORY);
6207 					log = 1;
6208 					DPFPRINTF(PF_DEBUG_MISC,
6209 					    ("pf: failed to allocate tag\n"));
6210 				} else {
6211 					pd.pf_mtag->flags |=
6212 					    PF_FASTFWD_OURS_PRESENT;
6213 					m->m_flags &= ~M_FASTFWD_OURS;
6214 				}
6215 			}
6216 			ip_divert_ptr(*m0, dir == PF_IN);
6217 			*m0 = NULL;
6218 
6219 			return (action);
6220 		} else {
6221 			/* XXX: ipfw has the same behaviour! */
6222 			action = PF_DROP;
6223 			REASON_SET(&reason, PFRES_MEMORY);
6224 			log = 1;
6225 			DPFPRINTF(PF_DEBUG_MISC,
6226 			    ("pf: failed to allocate divert tag\n"));
6227 		}
6228 	}
6229 
6230 	if (log) {
6231 		struct pf_rule *lr;
6232 
6233 		if (s != NULL && s->nat_rule.ptr != NULL &&
6234 		    s->nat_rule.ptr->log & PF_LOG_ALL)
6235 			lr = s->nat_rule.ptr;
6236 		else
6237 			lr = r;
6238 		PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
6239 		    (s == NULL));
6240 	}
6241 
6242 	kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6243 	kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
6244 
6245 	if (action == PF_PASS || r->action == PF_DROP) {
6246 		dirndx = (dir == PF_OUT);
6247 		r->packets[dirndx]++;
6248 		r->bytes[dirndx] += pd.tot_len;
6249 		if (a != NULL) {
6250 			a->packets[dirndx]++;
6251 			a->bytes[dirndx] += pd.tot_len;
6252 		}
6253 		if (s != NULL) {
6254 			if (s->nat_rule.ptr != NULL) {
6255 				s->nat_rule.ptr->packets[dirndx]++;
6256 				s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6257 			}
6258 			if (s->src_node != NULL) {
6259 				s->src_node->packets[dirndx]++;
6260 				s->src_node->bytes[dirndx] += pd.tot_len;
6261 			}
6262 			if (s->nat_src_node != NULL) {
6263 				s->nat_src_node->packets[dirndx]++;
6264 				s->nat_src_node->bytes[dirndx] += pd.tot_len;
6265 			}
6266 			dirndx = (dir == s->direction) ? 0 : 1;
6267 			s->packets[dirndx]++;
6268 			s->bytes[dirndx] += pd.tot_len;
6269 		}
6270 		tr = r;
6271 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6272 		if (nr != NULL && r == &V_pf_default_rule)
6273 			tr = nr;
6274 		if (tr->src.addr.type == PF_ADDR_TABLE)
6275 			pfr_update_stats(tr->src.addr.p.tbl,
6276 			    (s == NULL) ? pd.src :
6277 			    &s->key[(s->direction == PF_IN)]->
6278 				addr[(s->direction == PF_OUT)],
6279 			    pd.af, pd.tot_len, dir == PF_OUT,
6280 			    r->action == PF_PASS, tr->src.neg);
6281 		if (tr->dst.addr.type == PF_ADDR_TABLE)
6282 			pfr_update_stats(tr->dst.addr.p.tbl,
6283 			    (s == NULL) ? pd.dst :
6284 			    &s->key[(s->direction == PF_IN)]->
6285 				addr[(s->direction == PF_IN)],
6286 			    pd.af, pd.tot_len, dir == PF_OUT,
6287 			    r->action == PF_PASS, tr->dst.neg);
6288 	}
6289 
6290 	switch (action) {
6291 	case PF_SYNPROXY_DROP:
6292 		m_freem(*m0);
6293 	case PF_DEFER:
6294 		*m0 = NULL;
6295 		action = PF_PASS;
6296 		break;
6297 	case PF_DROP:
6298 		m_freem(*m0);
6299 		*m0 = NULL;
6300 		break;
6301 	default:
6302 		/* pf_route() returns unlocked. */
6303 		if (r->rt) {
6304 			pf_route(m0, r, dir, kif->pfik_ifp, s, &pd, inp);
6305 			return (action);
6306 		}
6307 		break;
6308 	}
6309 	if (s)
6310 		PF_STATE_UNLOCK(s);
6311 
6312 	return (action);
6313 }
6314 #endif /* INET */
6315 
6316 #ifdef INET6
6317 int
6318 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6319 {
6320 	struct pfi_kif		*kif;
6321 	u_short			 action, reason = 0, log = 0;
6322 	struct mbuf		*m = *m0, *n = NULL;
6323 	struct m_tag		*mtag;
6324 	struct ip6_hdr		*h = NULL;
6325 	struct pf_rule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6326 	struct pf_state		*s = NULL;
6327 	struct pf_ruleset	*ruleset = NULL;
6328 	struct pf_pdesc		 pd;
6329 	int			 off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0;
6330 
6331 	PF_RULES_RLOCK_TRACKER;
6332 	M_ASSERTPKTHDR(m);
6333 
6334 	if (!V_pf_status.running)
6335 		return (PF_PASS);
6336 
6337 	memset(&pd, 0, sizeof(pd));
6338 	pd.pf_mtag = pf_find_mtag(m);
6339 
6340 	if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6341 		return (PF_PASS);
6342 
6343 	kif = (struct pfi_kif *)ifp->if_pf_kif;
6344 	if (kif == NULL) {
6345 		DPFPRINTF(PF_DEBUG_URGENT,
6346 		    ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6347 		return (PF_DROP);
6348 	}
6349 	if (kif->pfik_flags & PFI_IFLAG_SKIP)
6350 		return (PF_PASS);
6351 
6352 	if (m->m_flags & M_SKIP_FIREWALL)
6353 		return (PF_PASS);
6354 
6355 	PF_RULES_RLOCK();
6356 
6357 	/* We do IP header normalization and packet reassembly here */
6358 	if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6359 		action = PF_DROP;
6360 		goto done;
6361 	}
6362 	m = *m0;	/* pf_normalize messes with m0 */
6363 	h = mtod(m, struct ip6_hdr *);
6364 
6365 	/*
6366 	 * we do not support jumbogram.  if we keep going, zero ip6_plen
6367 	 * will do something bad, so drop the packet for now.
6368 	 */
6369 	if (htons(h->ip6_plen) == 0) {
6370 		action = PF_DROP;
6371 		REASON_SET(&reason, PFRES_NORM);	/*XXX*/
6372 		goto done;
6373 	}
6374 
6375 	pd.src = (struct pf_addr *)&h->ip6_src;
6376 	pd.dst = (struct pf_addr *)&h->ip6_dst;
6377 	pd.sport = pd.dport = NULL;
6378 	pd.ip_sum = NULL;
6379 	pd.proto_sum = NULL;
6380 	pd.dir = dir;
6381 	pd.sidx = (dir == PF_IN) ? 0 : 1;
6382 	pd.didx = (dir == PF_IN) ? 1 : 0;
6383 	pd.af = AF_INET6;
6384 	pd.tos = 0;
6385 	pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6386 
6387 	off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6388 	pd.proto = h->ip6_nxt;
6389 	do {
6390 		switch (pd.proto) {
6391 		case IPPROTO_FRAGMENT:
6392 			action = pf_test_fragment(&r, dir, kif, m, h,
6393 			    &pd, &a, &ruleset);
6394 			if (action == PF_DROP)
6395 				REASON_SET(&reason, PFRES_FRAG);
6396 			goto done;
6397 		case IPPROTO_ROUTING: {
6398 			struct ip6_rthdr rthdr;
6399 
6400 			if (rh_cnt++) {
6401 				DPFPRINTF(PF_DEBUG_MISC,
6402 				    ("pf: IPv6 more than one rthdr\n"));
6403 				action = PF_DROP;
6404 				REASON_SET(&reason, PFRES_IPOPTIONS);
6405 				log = 1;
6406 				goto done;
6407 			}
6408 			if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6409 			    &reason, pd.af)) {
6410 				DPFPRINTF(PF_DEBUG_MISC,
6411 				    ("pf: IPv6 short rthdr\n"));
6412 				action = PF_DROP;
6413 				REASON_SET(&reason, PFRES_SHORT);
6414 				log = 1;
6415 				goto done;
6416 			}
6417 			if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6418 				DPFPRINTF(PF_DEBUG_MISC,
6419 				    ("pf: IPv6 rthdr0\n"));
6420 				action = PF_DROP;
6421 				REASON_SET(&reason, PFRES_IPOPTIONS);
6422 				log = 1;
6423 				goto done;
6424 			}
6425 			/* FALLTHROUGH */
6426 		}
6427 		case IPPROTO_AH:
6428 		case IPPROTO_HOPOPTS:
6429 		case IPPROTO_DSTOPTS: {
6430 			/* get next header and header length */
6431 			struct ip6_ext	opt6;
6432 
6433 			if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6434 			    NULL, &reason, pd.af)) {
6435 				DPFPRINTF(PF_DEBUG_MISC,
6436 				    ("pf: IPv6 short opt\n"));
6437 				action = PF_DROP;
6438 				log = 1;
6439 				goto done;
6440 			}
6441 			if (pd.proto == IPPROTO_AH)
6442 				off += (opt6.ip6e_len + 2) * 4;
6443 			else
6444 				off += (opt6.ip6e_len + 1) * 8;
6445 			pd.proto = opt6.ip6e_nxt;
6446 			/* goto the next header */
6447 			break;
6448 		}
6449 		default:
6450 			terminal++;
6451 			break;
6452 		}
6453 	} while (!terminal);
6454 
6455 	/* if there's no routing header, use unmodified mbuf for checksumming */
6456 	if (!n)
6457 		n = m;
6458 
6459 	switch (pd.proto) {
6460 
6461 	case IPPROTO_TCP: {
6462 		struct tcphdr	th;
6463 
6464 		pd.hdr.tcp = &th;
6465 		if (!pf_pull_hdr(m, off, &th, sizeof(th),
6466 		    &action, &reason, AF_INET6)) {
6467 			log = action != PF_PASS;
6468 			goto done;
6469 		}
6470 		pd.p_len = pd.tot_len - off - (th.th_off << 2);
6471 		action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6472 		if (action == PF_DROP)
6473 			goto done;
6474 		action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6475 		    &reason);
6476 		if (action == PF_PASS) {
6477 			if (V_pfsync_update_state_ptr != NULL)
6478 				V_pfsync_update_state_ptr(s);
6479 			r = s->rule.ptr;
6480 			a = s->anchor.ptr;
6481 			log = s->log;
6482 		} else if (s == NULL)
6483 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6484 			    &a, &ruleset, inp);
6485 		break;
6486 	}
6487 
6488 	case IPPROTO_UDP: {
6489 		struct udphdr	uh;
6490 
6491 		pd.hdr.udp = &uh;
6492 		if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6493 		    &action, &reason, AF_INET6)) {
6494 			log = action != PF_PASS;
6495 			goto done;
6496 		}
6497 		if (uh.uh_dport == 0 ||
6498 		    ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6499 		    ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6500 			action = PF_DROP;
6501 			REASON_SET(&reason, PFRES_SHORT);
6502 			goto done;
6503 		}
6504 		action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6505 		if (action == PF_PASS) {
6506 			if (V_pfsync_update_state_ptr != NULL)
6507 				V_pfsync_update_state_ptr(s);
6508 			r = s->rule.ptr;
6509 			a = s->anchor.ptr;
6510 			log = s->log;
6511 		} else if (s == NULL)
6512 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6513 			    &a, &ruleset, inp);
6514 		break;
6515 	}
6516 
6517 	case IPPROTO_ICMP: {
6518 		action = PF_DROP;
6519 		DPFPRINTF(PF_DEBUG_MISC,
6520 		    ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6521 		goto done;
6522 	}
6523 
6524 	case IPPROTO_ICMPV6: {
6525 		struct icmp6_hdr	ih;
6526 
6527 		pd.hdr.icmp6 = &ih;
6528 		if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6529 		    &action, &reason, AF_INET6)) {
6530 			log = action != PF_PASS;
6531 			goto done;
6532 		}
6533 		action = pf_test_state_icmp(&s, dir, kif,
6534 		    m, off, h, &pd, &reason);
6535 		if (action == PF_PASS) {
6536 			if (V_pfsync_update_state_ptr != NULL)
6537 				V_pfsync_update_state_ptr(s);
6538 			r = s->rule.ptr;
6539 			a = s->anchor.ptr;
6540 			log = s->log;
6541 		} else if (s == NULL)
6542 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6543 			    &a, &ruleset, inp);
6544 		break;
6545 	}
6546 
6547 	default:
6548 		action = pf_test_state_other(&s, dir, kif, m, &pd);
6549 		if (action == PF_PASS) {
6550 			if (V_pfsync_update_state_ptr != NULL)
6551 				V_pfsync_update_state_ptr(s);
6552 			r = s->rule.ptr;
6553 			a = s->anchor.ptr;
6554 			log = s->log;
6555 		} else if (s == NULL)
6556 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6557 			    &a, &ruleset, inp);
6558 		break;
6559 	}
6560 
6561 done:
6562 	PF_RULES_RUNLOCK();
6563 	if (n != m) {
6564 		m_freem(n);
6565 		n = NULL;
6566 	}
6567 
6568 	/* handle dangerous IPv6 extension headers. */
6569 	if (action == PF_PASS && rh_cnt &&
6570 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6571 		action = PF_DROP;
6572 		REASON_SET(&reason, PFRES_IPOPTIONS);
6573 		log = r->log;
6574 		DPFPRINTF(PF_DEBUG_MISC,
6575 		    ("pf: dropping packet with dangerous v6 headers\n"));
6576 	}
6577 
6578 	if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6579 		action = PF_DROP;
6580 		REASON_SET(&reason, PFRES_MEMORY);
6581 	}
6582 	if (r->rtableid >= 0)
6583 		M_SETFIB(m, r->rtableid);
6584 
6585 	if (r->scrub_flags & PFSTATE_SETPRIO) {
6586 		if (pd.tos & IPTOS_LOWDELAY)
6587 			pqid = 1;
6588 		if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6589 			action = PF_DROP;
6590 			REASON_SET(&reason, PFRES_MEMORY);
6591 			log = 1;
6592 			DPFPRINTF(PF_DEBUG_MISC,
6593 			    ("pf: failed to allocate 802.1q mtag\n"));
6594 		}
6595 	}
6596 
6597 #ifdef ALTQ
6598 	if (action == PF_PASS && r->qid) {
6599 		if (pd.pf_mtag == NULL &&
6600 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6601 			action = PF_DROP;
6602 			REASON_SET(&reason, PFRES_MEMORY);
6603 		} else {
6604 			if (s != NULL)
6605 				pd.pf_mtag->qid_hash = pf_state_hash(s);
6606 			if (pd.tos & IPTOS_LOWDELAY)
6607 				pd.pf_mtag->qid = r->pqid;
6608 			else
6609 				pd.pf_mtag->qid = r->qid;
6610 			/* Add hints for ecn. */
6611 			pd.pf_mtag->hdr = h;
6612 		}
6613 	}
6614 #endif /* ALTQ */
6615 
6616 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6617 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6618 	    (s->nat_rule.ptr->action == PF_RDR ||
6619 	    s->nat_rule.ptr->action == PF_BINAT) &&
6620 	    IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6621 		m->m_flags |= M_SKIP_FIREWALL;
6622 
6623 	/* XXX: Anybody working on it?! */
6624 	if (r->divert.port)
6625 		printf("pf: divert(9) is not supported for IPv6\n");
6626 
6627 	if (log) {
6628 		struct pf_rule *lr;
6629 
6630 		if (s != NULL && s->nat_rule.ptr != NULL &&
6631 		    s->nat_rule.ptr->log & PF_LOG_ALL)
6632 			lr = s->nat_rule.ptr;
6633 		else
6634 			lr = r;
6635 		PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6636 		    &pd, (s == NULL));
6637 	}
6638 
6639 	kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6640 	kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
6641 
6642 	if (action == PF_PASS || r->action == PF_DROP) {
6643 		dirndx = (dir == PF_OUT);
6644 		r->packets[dirndx]++;
6645 		r->bytes[dirndx] += pd.tot_len;
6646 		if (a != NULL) {
6647 			a->packets[dirndx]++;
6648 			a->bytes[dirndx] += pd.tot_len;
6649 		}
6650 		if (s != NULL) {
6651 			if (s->nat_rule.ptr != NULL) {
6652 				s->nat_rule.ptr->packets[dirndx]++;
6653 				s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6654 			}
6655 			if (s->src_node != NULL) {
6656 				s->src_node->packets[dirndx]++;
6657 				s->src_node->bytes[dirndx] += pd.tot_len;
6658 			}
6659 			if (s->nat_src_node != NULL) {
6660 				s->nat_src_node->packets[dirndx]++;
6661 				s->nat_src_node->bytes[dirndx] += pd.tot_len;
6662 			}
6663 			dirndx = (dir == s->direction) ? 0 : 1;
6664 			s->packets[dirndx]++;
6665 			s->bytes[dirndx] += pd.tot_len;
6666 		}
6667 		tr = r;
6668 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6669 		if (nr != NULL && r == &V_pf_default_rule)
6670 			tr = nr;
6671 		if (tr->src.addr.type == PF_ADDR_TABLE)
6672 			pfr_update_stats(tr->src.addr.p.tbl,
6673 			    (s == NULL) ? pd.src :
6674 			    &s->key[(s->direction == PF_IN)]->addr[0],
6675 			    pd.af, pd.tot_len, dir == PF_OUT,
6676 			    r->action == PF_PASS, tr->src.neg);
6677 		if (tr->dst.addr.type == PF_ADDR_TABLE)
6678 			pfr_update_stats(tr->dst.addr.p.tbl,
6679 			    (s == NULL) ? pd.dst :
6680 			    &s->key[(s->direction == PF_IN)]->addr[1],
6681 			    pd.af, pd.tot_len, dir == PF_OUT,
6682 			    r->action == PF_PASS, tr->dst.neg);
6683 	}
6684 
6685 	switch (action) {
6686 	case PF_SYNPROXY_DROP:
6687 		m_freem(*m0);
6688 	case PF_DEFER:
6689 		*m0 = NULL;
6690 		action = PF_PASS;
6691 		break;
6692 	case PF_DROP:
6693 		m_freem(*m0);
6694 		*m0 = NULL;
6695 		break;
6696 	default:
6697 		/* pf_route6() returns unlocked. */
6698 		if (r->rt) {
6699 			pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd, inp);
6700 			return (action);
6701 		}
6702 		break;
6703 	}
6704 
6705 	if (s)
6706 		PF_STATE_UNLOCK(s);
6707 
6708 	/* If reassembled packet passed, create new fragments. */
6709 	if (action == PF_PASS && *m0 && (pflags & PFIL_FWD) &&
6710 	    (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL)
6711 		action = pf_refragment6(ifp, m0, mtag);
6712 
6713 	return (action);
6714 }
6715 #endif /* INET6 */
6716