xref: /freebsd/sys/netpfil/pf/pf.c (revision 82ea1a07b4894bea20c61afbf702cc46f0711ead)
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_bpf.h"
44 #include "opt_inet.h"
45 #include "opt_inet6.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/sdt.h>
63 #include <sys/socket.h>
64 #include <sys/sysctl.h>
65 #include <sys/taskqueue.h>
66 #include <sys/ucred.h>
67 
68 #include <net/if.h>
69 #include <net/if_var.h>
70 #include <net/if_types.h>
71 #include <net/if_vlan_var.h>
72 #include <net/route.h>
73 #include <net/route/nhop.h>
74 #include <net/vnet.h>
75 
76 #include <net/pfil.h>
77 #include <net/pfvar.h>
78 #include <net/if_pflog.h>
79 #include <net/if_pfsync.h>
80 
81 #include <netinet/in_pcb.h>
82 #include <netinet/in_var.h>
83 #include <netinet/in_fib.h>
84 #include <netinet/ip.h>
85 #include <netinet/ip_fw.h>
86 #include <netinet/ip_icmp.h>
87 #include <netinet/icmp_var.h>
88 #include <netinet/ip_var.h>
89 #include <netinet/tcp.h>
90 #include <netinet/tcp_fsm.h>
91 #include <netinet/tcp_seq.h>
92 #include <netinet/tcp_timer.h>
93 #include <netinet/tcp_var.h>
94 #include <netinet/udp.h>
95 #include <netinet/udp_var.h>
96 
97 #ifdef INET6
98 #include <netinet/ip6.h>
99 #include <netinet/icmp6.h>
100 #include <netinet6/nd6.h>
101 #include <netinet6/ip6_var.h>
102 #include <netinet6/in6_pcb.h>
103 #include <netinet6/in6_fib.h>
104 #include <netinet6/scope6_var.h>
105 #endif /* INET6 */
106 
107 #if defined(SCTP) || defined(SCTP_SUPPORT)
108 #include <netinet/sctp_crc32.h>
109 #endif
110 
111 #include <machine/in_cksum.h>
112 #include <security/mac/mac_framework.h>
113 
114 #define	DPFPRINTF(n, x)	if (V_pf_status.debug >= (n)) printf x
115 
116 SDT_PROVIDER_DEFINE(pf);
117 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *",
118     "struct pf_state *");
119 SDT_PROBE_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *",
120     "struct pf_state *");
121 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *",
122     "struct pf_state_key_cmp *", "int", "struct pf_pdesc *",
123     "struct pf_state *");
124 
125 /*
126  * Global variables
127  */
128 
129 /* state tables */
130 VNET_DEFINE(struct pf_altqqueue,	 pf_altqs[4]);
131 VNET_DEFINE(struct pf_kpalist,		 pf_pabuf);
132 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_active);
133 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_active);
134 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_inactive);
135 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_inactive);
136 VNET_DEFINE(struct pf_kstatus,		 pf_status);
137 
138 VNET_DEFINE(u_int32_t,			 ticket_altqs_active);
139 VNET_DEFINE(u_int32_t,			 ticket_altqs_inactive);
140 VNET_DEFINE(int,			 altqs_inactive_open);
141 VNET_DEFINE(u_int32_t,			 ticket_pabuf);
142 
143 VNET_DEFINE(MD5_CTX,			 pf_tcp_secret_ctx);
144 #define	V_pf_tcp_secret_ctx		 VNET(pf_tcp_secret_ctx)
145 VNET_DEFINE(u_char,			 pf_tcp_secret[16]);
146 #define	V_pf_tcp_secret			 VNET(pf_tcp_secret)
147 VNET_DEFINE(int,			 pf_tcp_secret_init);
148 #define	V_pf_tcp_secret_init		 VNET(pf_tcp_secret_init)
149 VNET_DEFINE(int,			 pf_tcp_iss_off);
150 #define	V_pf_tcp_iss_off		 VNET(pf_tcp_iss_off)
151 VNET_DECLARE(int,			 pf_vnet_active);
152 #define	V_pf_vnet_active		 VNET(pf_vnet_active)
153 
154 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx);
155 #define V_pf_purge_idx	VNET(pf_purge_idx)
156 
157 /*
158  * Queue for pf_intr() sends.
159  */
160 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
161 struct pf_send_entry {
162 	STAILQ_ENTRY(pf_send_entry)	pfse_next;
163 	struct mbuf			*pfse_m;
164 	enum {
165 		PFSE_IP,
166 		PFSE_IP6,
167 		PFSE_ICMP,
168 		PFSE_ICMP6,
169 	}				pfse_type;
170 	struct {
171 		int		type;
172 		int		code;
173 		int		mtu;
174 	} icmpopts;
175 };
176 
177 STAILQ_HEAD(pf_send_head, pf_send_entry);
178 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue);
179 #define	V_pf_sendqueue	VNET(pf_sendqueue)
180 
181 static struct mtx pf_sendqueue_mtx;
182 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
183 #define	PF_SENDQ_LOCK()		mtx_lock(&pf_sendqueue_mtx)
184 #define	PF_SENDQ_UNLOCK()	mtx_unlock(&pf_sendqueue_mtx)
185 
186 /*
187  * Queue for pf_overload_task() tasks.
188  */
189 struct pf_overload_entry {
190 	SLIST_ENTRY(pf_overload_entry)	next;
191 	struct pf_addr  		addr;
192 	sa_family_t			af;
193 	uint8_t				dir;
194 	struct pf_krule  		*rule;
195 };
196 
197 SLIST_HEAD(pf_overload_head, pf_overload_entry);
198 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue);
199 #define V_pf_overloadqueue	VNET(pf_overloadqueue)
200 VNET_DEFINE_STATIC(struct task, pf_overloadtask);
201 #define	V_pf_overloadtask	VNET(pf_overloadtask)
202 
203 static struct mtx pf_overloadqueue_mtx;
204 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
205     "pf overload/flush queue", MTX_DEF);
206 #define	PF_OVERLOADQ_LOCK()	mtx_lock(&pf_overloadqueue_mtx)
207 #define	PF_OVERLOADQ_UNLOCK()	mtx_unlock(&pf_overloadqueue_mtx)
208 
209 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules);
210 struct mtx pf_unlnkdrules_mtx;
211 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
212     MTX_DEF);
213 
214 VNET_DEFINE_STATIC(uma_zone_t,	pf_sources_z);
215 #define	V_pf_sources_z	VNET(pf_sources_z)
216 uma_zone_t		pf_mtag_z;
217 VNET_DEFINE(uma_zone_t,	 pf_state_z);
218 VNET_DEFINE(uma_zone_t,	 pf_state_key_z);
219 
220 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
221 #define	PFID_CPUBITS	8
222 #define	PFID_CPUSHIFT	(sizeof(uint64_t) * NBBY - PFID_CPUBITS)
223 #define	PFID_CPUMASK	((uint64_t)((1 << PFID_CPUBITS) - 1) <<	PFID_CPUSHIFT)
224 #define	PFID_MAXID	(~PFID_CPUMASK)
225 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU);
226 
227 static void		 pf_src_tree_remove_state(struct pf_state *);
228 static void		 pf_init_threshold(struct pf_threshold *, u_int32_t,
229 			    u_int32_t);
230 static void		 pf_add_threshold(struct pf_threshold *);
231 static int		 pf_check_threshold(struct pf_threshold *);
232 
233 static void		 pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
234 			    u_int16_t *, u_int16_t *, struct pf_addr *,
235 			    u_int16_t, u_int8_t, sa_family_t);
236 static int		 pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
237 			    struct tcphdr *, struct pf_state_peer *);
238 static void		 pf_change_icmp(struct pf_addr *, u_int16_t *,
239 			    struct pf_addr *, struct pf_addr *, u_int16_t,
240 			    u_int16_t *, u_int16_t *, u_int16_t *,
241 			    u_int16_t *, u_int8_t, sa_family_t);
242 static void		 pf_send_tcp(struct mbuf *,
243 			    const struct pf_krule *, sa_family_t,
244 			    const struct pf_addr *, const struct pf_addr *,
245 			    u_int16_t, u_int16_t, u_int32_t, u_int32_t,
246 			    u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
247 			    u_int16_t, struct ifnet *);
248 static void		 pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
249 			    sa_family_t, struct pf_krule *);
250 static void		 pf_detach_state(struct pf_state *);
251 static int		 pf_state_key_attach(struct pf_state_key *,
252 			    struct pf_state_key *, struct pf_state *);
253 static void		 pf_state_key_detach(struct pf_state *, int);
254 static int		 pf_state_key_ctor(void *, int, void *, int);
255 static u_int32_t	 pf_tcp_iss(struct pf_pdesc *);
256 static int		 pf_test_rule(struct pf_krule **, struct pf_state **,
257 			    int, struct pfi_kkif *, struct mbuf *, int,
258 			    struct pf_pdesc *, struct pf_krule **,
259 			    struct pf_kruleset **, struct inpcb *);
260 static int		 pf_create_state(struct pf_krule *, struct pf_krule *,
261 			    struct pf_krule *, struct pf_pdesc *,
262 			    struct pf_ksrc_node *, struct pf_state_key *,
263 			    struct pf_state_key *, struct mbuf *, int,
264 			    u_int16_t, u_int16_t, int *, struct pfi_kkif *,
265 			    struct pf_state **, int, u_int16_t, u_int16_t,
266 			    int);
267 static int		 pf_test_fragment(struct pf_krule **, int,
268 			    struct pfi_kkif *, struct mbuf *, void *,
269 			    struct pf_pdesc *, struct pf_krule **,
270 			    struct pf_kruleset **);
271 static int		 pf_tcp_track_full(struct pf_state_peer *,
272 			    struct pf_state_peer *, struct pf_state **,
273 			    struct pfi_kkif *, struct mbuf *, int,
274 			    struct pf_pdesc *, u_short *, int *);
275 static int		 pf_tcp_track_sloppy(struct pf_state_peer *,
276 			    struct pf_state_peer *, struct pf_state **,
277 			    struct pf_pdesc *, u_short *);
278 static int		 pf_test_state_tcp(struct pf_state **, int,
279 			    struct pfi_kkif *, struct mbuf *, int,
280 			    void *, struct pf_pdesc *, u_short *);
281 static int		 pf_test_state_udp(struct pf_state **, int,
282 			    struct pfi_kkif *, struct mbuf *, int,
283 			    void *, struct pf_pdesc *);
284 static int		 pf_test_state_icmp(struct pf_state **, int,
285 			    struct pfi_kkif *, struct mbuf *, int,
286 			    void *, struct pf_pdesc *, u_short *);
287 static int		 pf_test_state_other(struct pf_state **, int,
288 			    struct pfi_kkif *, struct mbuf *, struct pf_pdesc *);
289 static u_int8_t		 pf_get_wscale(struct mbuf *, int, u_int16_t,
290 			    sa_family_t);
291 static u_int16_t	 pf_get_mss(struct mbuf *, int, u_int16_t,
292 			    sa_family_t);
293 static u_int16_t	 pf_calc_mss(struct pf_addr *, sa_family_t,
294 				int, u_int16_t);
295 static int		 pf_check_proto_cksum(struct mbuf *, int, int,
296 			    u_int8_t, sa_family_t);
297 static void		 pf_print_state_parts(struct pf_state *,
298 			    struct pf_state_key *, struct pf_state_key *);
299 static int		 pf_addr_wrap_neq(struct pf_addr_wrap *,
300 			    struct pf_addr_wrap *);
301 static void		 pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t,
302 			    bool, u_int8_t);
303 static struct pf_state	*pf_find_state(struct pfi_kkif *,
304 			    struct pf_state_key_cmp *, u_int);
305 static int		 pf_src_connlimit(struct pf_state **);
306 static void		 pf_overload_task(void *v, int pending);
307 static int		 pf_insert_src_node(struct pf_ksrc_node **,
308 			    struct pf_krule *, struct pf_addr *, sa_family_t);
309 static u_int		 pf_purge_expired_states(u_int, int);
310 static void		 pf_purge_unlinked_rules(void);
311 static int		 pf_mtag_uminit(void *, int, int);
312 static void		 pf_mtag_free(struct m_tag *);
313 #ifdef INET
314 static void		 pf_route(struct mbuf **, struct pf_krule *, int,
315 			    struct ifnet *, struct pf_state *,
316 			    struct pf_pdesc *, struct inpcb *);
317 #endif /* INET */
318 #ifdef INET6
319 static void		 pf_change_a6(struct pf_addr *, u_int16_t *,
320 			    struct pf_addr *, u_int8_t);
321 static void		 pf_route6(struct mbuf **, struct pf_krule *, int,
322 			    struct ifnet *, struct pf_state *,
323 			    struct pf_pdesc *, struct inpcb *);
324 #endif /* INET6 */
325 
326 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
327 
328 extern int pf_end_threads;
329 extern struct proc *pf_purge_proc;
330 
331 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
332 
333 #define	PACKET_LOOPED(pd)	((pd)->pf_mtag &&			\
334 				 (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
335 
336 #define	STATE_LOOKUP(i, k, d, s, pd)					\
337 	do {								\
338 		(s) = pf_find_state((i), (k), (d));			\
339 		SDT_PROBE5(pf, ip, state, lookup, i, k, d, pd, (s));	\
340 		if ((s) == NULL)					\
341 			return (PF_DROP);				\
342 		if (PACKET_LOOPED(pd))					\
343 			return (PF_PASS);				\
344 		if ((d) == PF_OUT &&					\
345 		    (s)->rule.ptr->rt == PF_ROUTETO &&			\
346 		    (s)->rule.ptr->direction == PF_OUT &&		\
347 		    (s)->rt_kif != NULL &&				\
348 		    (s)->rt_kif != (i))					\
349 			return (PF_PASS);				\
350 	} while (0)
351 
352 #define	BOUND_IFACE(r, k) \
353 	((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
354 
355 #define	STATE_INC_COUNTERS(s)						\
356 	do {								\
357 		counter_u64_add(s->rule.ptr->states_cur, 1);		\
358 		counter_u64_add(s->rule.ptr->states_tot, 1);		\
359 		if (s->anchor.ptr != NULL) {				\
360 			counter_u64_add(s->anchor.ptr->states_cur, 1);	\
361 			counter_u64_add(s->anchor.ptr->states_tot, 1);	\
362 		}							\
363 		if (s->nat_rule.ptr != NULL) {				\
364 			counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
365 			counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
366 		}							\
367 	} while (0)
368 
369 #define	STATE_DEC_COUNTERS(s)						\
370 	do {								\
371 		if (s->nat_rule.ptr != NULL)				\
372 			counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
373 		if (s->anchor.ptr != NULL)				\
374 			counter_u64_add(s->anchor.ptr->states_cur, -1);	\
375 		counter_u64_add(s->rule.ptr->states_cur, -1);		\
376 	} while (0)
377 
378 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
379 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
380 VNET_DEFINE(struct pf_idhash *, pf_idhash);
381 VNET_DEFINE(struct pf_srchash *, pf_srchash);
382 
383 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
384     "pf(4)");
385 
386 u_long	pf_hashmask;
387 u_long	pf_srchashmask;
388 static u_long	pf_hashsize;
389 static u_long	pf_srchashsize;
390 u_long	pf_ioctl_maxcount = 65535;
391 
392 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
393     &pf_hashsize, 0, "Size of pf(4) states hashtable");
394 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
395     &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
396 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN,
397     &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call");
398 
399 VNET_DEFINE(void *, pf_swi_cookie);
400 VNET_DEFINE(struct intr_event *, pf_swi_ie);
401 
402 VNET_DEFINE(uint32_t, pf_hashseed);
403 #define	V_pf_hashseed	VNET(pf_hashseed)
404 
405 int
406 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
407 {
408 
409 	switch (af) {
410 #ifdef INET
411 	case AF_INET:
412 		if (a->addr32[0] > b->addr32[0])
413 			return (1);
414 		if (a->addr32[0] < b->addr32[0])
415 			return (-1);
416 		break;
417 #endif /* INET */
418 #ifdef INET6
419 	case AF_INET6:
420 		if (a->addr32[3] > b->addr32[3])
421 			return (1);
422 		if (a->addr32[3] < b->addr32[3])
423 			return (-1);
424 		if (a->addr32[2] > b->addr32[2])
425 			return (1);
426 		if (a->addr32[2] < b->addr32[2])
427 			return (-1);
428 		if (a->addr32[1] > b->addr32[1])
429 			return (1);
430 		if (a->addr32[1] < b->addr32[1])
431 			return (-1);
432 		if (a->addr32[0] > b->addr32[0])
433 			return (1);
434 		if (a->addr32[0] < b->addr32[0])
435 			return (-1);
436 		break;
437 #endif /* INET6 */
438 	default:
439 		panic("%s: unknown address family %u", __func__, af);
440 	}
441 	return (0);
442 }
443 
444 static __inline uint32_t
445 pf_hashkey(struct pf_state_key *sk)
446 {
447 	uint32_t h;
448 
449 	h = murmur3_32_hash32((uint32_t *)sk,
450 	    sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
451 	    V_pf_hashseed);
452 
453 	return (h & pf_hashmask);
454 }
455 
456 static __inline uint32_t
457 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
458 {
459 	uint32_t h;
460 
461 	switch (af) {
462 	case AF_INET:
463 		h = murmur3_32_hash32((uint32_t *)&addr->v4,
464 		    sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
465 		break;
466 	case AF_INET6:
467 		h = murmur3_32_hash32((uint32_t *)&addr->v6,
468 		    sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
469 		break;
470 	default:
471 		panic("%s: unknown address family %u", __func__, af);
472 	}
473 
474 	return (h & pf_srchashmask);
475 }
476 
477 #ifdef ALTQ
478 static int
479 pf_state_hash(struct pf_state *s)
480 {
481 	u_int32_t hv = (intptr_t)s / sizeof(*s);
482 
483 	hv ^= crc32(&s->src, sizeof(s->src));
484 	hv ^= crc32(&s->dst, sizeof(s->dst));
485 	if (hv == 0)
486 		hv = 1;
487 	return (hv);
488 }
489 #endif
490 
491 #ifdef INET6
492 void
493 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
494 {
495 	switch (af) {
496 #ifdef INET
497 	case AF_INET:
498 		dst->addr32[0] = src->addr32[0];
499 		break;
500 #endif /* INET */
501 	case AF_INET6:
502 		dst->addr32[0] = src->addr32[0];
503 		dst->addr32[1] = src->addr32[1];
504 		dst->addr32[2] = src->addr32[2];
505 		dst->addr32[3] = src->addr32[3];
506 		break;
507 	}
508 }
509 #endif /* INET6 */
510 
511 static void
512 pf_init_threshold(struct pf_threshold *threshold,
513     u_int32_t limit, u_int32_t seconds)
514 {
515 	threshold->limit = limit * PF_THRESHOLD_MULT;
516 	threshold->seconds = seconds;
517 	threshold->count = 0;
518 	threshold->last = time_uptime;
519 }
520 
521 static void
522 pf_add_threshold(struct pf_threshold *threshold)
523 {
524 	u_int32_t t = time_uptime, diff = t - threshold->last;
525 
526 	if (diff >= threshold->seconds)
527 		threshold->count = 0;
528 	else
529 		threshold->count -= threshold->count * diff /
530 		    threshold->seconds;
531 	threshold->count += PF_THRESHOLD_MULT;
532 	threshold->last = t;
533 }
534 
535 static int
536 pf_check_threshold(struct pf_threshold *threshold)
537 {
538 	return (threshold->count > threshold->limit);
539 }
540 
541 static int
542 pf_src_connlimit(struct pf_state **state)
543 {
544 	struct pf_overload_entry *pfoe;
545 	int bad = 0;
546 
547 	PF_STATE_LOCK_ASSERT(*state);
548 
549 	(*state)->src_node->conn++;
550 	(*state)->src.tcp_est = 1;
551 	pf_add_threshold(&(*state)->src_node->conn_rate);
552 
553 	if ((*state)->rule.ptr->max_src_conn &&
554 	    (*state)->rule.ptr->max_src_conn <
555 	    (*state)->src_node->conn) {
556 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
557 		bad++;
558 	}
559 
560 	if ((*state)->rule.ptr->max_src_conn_rate.limit &&
561 	    pf_check_threshold(&(*state)->src_node->conn_rate)) {
562 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
563 		bad++;
564 	}
565 
566 	if (!bad)
567 		return (0);
568 
569 	/* Kill this state. */
570 	(*state)->timeout = PFTM_PURGE;
571 	(*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
572 
573 	if ((*state)->rule.ptr->overload_tbl == NULL)
574 		return (1);
575 
576 	/* Schedule overloading and flushing task. */
577 	pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
578 	if (pfoe == NULL)
579 		return (1);	/* too bad :( */
580 
581 	bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
582 	pfoe->af = (*state)->key[PF_SK_WIRE]->af;
583 	pfoe->rule = (*state)->rule.ptr;
584 	pfoe->dir = (*state)->direction;
585 	PF_OVERLOADQ_LOCK();
586 	SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
587 	PF_OVERLOADQ_UNLOCK();
588 	taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
589 
590 	return (1);
591 }
592 
593 static void
594 pf_overload_task(void *v, int pending)
595 {
596 	struct pf_overload_head queue;
597 	struct pfr_addr p;
598 	struct pf_overload_entry *pfoe, *pfoe1;
599 	uint32_t killed = 0;
600 
601 	CURVNET_SET((struct vnet *)v);
602 
603 	PF_OVERLOADQ_LOCK();
604 	queue = V_pf_overloadqueue;
605 	SLIST_INIT(&V_pf_overloadqueue);
606 	PF_OVERLOADQ_UNLOCK();
607 
608 	bzero(&p, sizeof(p));
609 	SLIST_FOREACH(pfoe, &queue, next) {
610 		counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
611 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
612 			printf("%s: blocking address ", __func__);
613 			pf_print_host(&pfoe->addr, 0, pfoe->af);
614 			printf("\n");
615 		}
616 
617 		p.pfra_af = pfoe->af;
618 		switch (pfoe->af) {
619 #ifdef INET
620 		case AF_INET:
621 			p.pfra_net = 32;
622 			p.pfra_ip4addr = pfoe->addr.v4;
623 			break;
624 #endif
625 #ifdef INET6
626 		case AF_INET6:
627 			p.pfra_net = 128;
628 			p.pfra_ip6addr = pfoe->addr.v6;
629 			break;
630 #endif
631 		}
632 
633 		PF_RULES_WLOCK();
634 		pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
635 		PF_RULES_WUNLOCK();
636 	}
637 
638 	/*
639 	 * Remove those entries, that don't need flushing.
640 	 */
641 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
642 		if (pfoe->rule->flush == 0) {
643 			SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
644 			free(pfoe, M_PFTEMP);
645 		} else
646 			counter_u64_add(
647 			    V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
648 
649 	/* If nothing to flush, return. */
650 	if (SLIST_EMPTY(&queue)) {
651 		CURVNET_RESTORE();
652 		return;
653 	}
654 
655 	for (int i = 0; i <= pf_hashmask; i++) {
656 		struct pf_idhash *ih = &V_pf_idhash[i];
657 		struct pf_state_key *sk;
658 		struct pf_state *s;
659 
660 		PF_HASHROW_LOCK(ih);
661 		LIST_FOREACH(s, &ih->states, entry) {
662 		    sk = s->key[PF_SK_WIRE];
663 		    SLIST_FOREACH(pfoe, &queue, next)
664 			if (sk->af == pfoe->af &&
665 			    ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
666 			    pfoe->rule == s->rule.ptr) &&
667 			    ((pfoe->dir == PF_OUT &&
668 			    PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
669 			    (pfoe->dir == PF_IN &&
670 			    PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
671 				s->timeout = PFTM_PURGE;
672 				s->src.state = s->dst.state = TCPS_CLOSED;
673 				killed++;
674 			}
675 		}
676 		PF_HASHROW_UNLOCK(ih);
677 	}
678 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
679 		free(pfoe, M_PFTEMP);
680 	if (V_pf_status.debug >= PF_DEBUG_MISC)
681 		printf("%s: %u states killed", __func__, killed);
682 
683 	CURVNET_RESTORE();
684 }
685 
686 /*
687  * Can return locked on failure, so that we can consistently
688  * allocate and insert a new one.
689  */
690 struct pf_ksrc_node *
691 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af,
692 	int returnlocked)
693 {
694 	struct pf_srchash *sh;
695 	struct pf_ksrc_node *n;
696 
697 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
698 
699 	sh = &V_pf_srchash[pf_hashsrc(src, af)];
700 	PF_HASHROW_LOCK(sh);
701 	LIST_FOREACH(n, &sh->nodes, entry)
702 		if (n->rule.ptr == rule && n->af == af &&
703 		    ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
704 		    (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
705 			break;
706 	if (n != NULL) {
707 		n->states++;
708 		PF_HASHROW_UNLOCK(sh);
709 	} else if (returnlocked == 0)
710 		PF_HASHROW_UNLOCK(sh);
711 
712 	return (n);
713 }
714 
715 static void
716 pf_free_src_node(struct pf_ksrc_node *sn)
717 {
718 
719 	for (int i = 0; i < 2; i++) {
720 		counter_u64_free(sn->bytes[i]);
721 		counter_u64_free(sn->packets[i]);
722 	}
723 	uma_zfree(V_pf_sources_z, sn);
724 }
725 
726 static int
727 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule,
728     struct pf_addr *src, sa_family_t af)
729 {
730 
731 	KASSERT((rule->rule_flag & PFRULE_SRCTRACK ||
732 	    rule->rpool.opts & PF_POOL_STICKYADDR),
733 	    ("%s for non-tracking rule %p", __func__, rule));
734 
735 	if (*sn == NULL)
736 		*sn = pf_find_src_node(src, rule, af, 1);
737 
738 	if (*sn == NULL) {
739 		struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
740 
741 		PF_HASHROW_ASSERT(sh);
742 
743 		if (!rule->max_src_nodes ||
744 		    counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
745 			(*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
746 		else
747 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
748 			    1);
749 		if ((*sn) == NULL) {
750 			PF_HASHROW_UNLOCK(sh);
751 			return (-1);
752 		}
753 
754 		for (int i = 0; i < 2; i++) {
755 			(*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT);
756 			(*sn)->packets[i] = counter_u64_alloc(M_NOWAIT);
757 
758 			if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) {
759 				pf_free_src_node(*sn);
760 				PF_HASHROW_UNLOCK(sh);
761 				return (-1);
762 			}
763 		}
764 
765 		pf_init_threshold(&(*sn)->conn_rate,
766 		    rule->max_src_conn_rate.limit,
767 		    rule->max_src_conn_rate.seconds);
768 
769 		(*sn)->af = af;
770 		(*sn)->rule.ptr = rule;
771 		PF_ACPY(&(*sn)->addr, src, af);
772 		LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
773 		(*sn)->creation = time_uptime;
774 		(*sn)->ruletype = rule->action;
775 		(*sn)->states = 1;
776 		if ((*sn)->rule.ptr != NULL)
777 			counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
778 		PF_HASHROW_UNLOCK(sh);
779 		counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
780 	} else {
781 		if (rule->max_src_states &&
782 		    (*sn)->states >= rule->max_src_states) {
783 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
784 			    1);
785 			return (-1);
786 		}
787 	}
788 	return (0);
789 }
790 
791 void
792 pf_unlink_src_node(struct pf_ksrc_node *src)
793 {
794 
795 	PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]);
796 	LIST_REMOVE(src, entry);
797 	if (src->rule.ptr)
798 		counter_u64_add(src->rule.ptr->src_nodes, -1);
799 }
800 
801 u_int
802 pf_free_src_nodes(struct pf_ksrc_node_list *head)
803 {
804 	struct pf_ksrc_node *sn, *tmp;
805 	u_int count = 0;
806 
807 	LIST_FOREACH_SAFE(sn, head, entry, tmp) {
808 		pf_free_src_node(sn);
809 		count++;
810 	}
811 
812 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
813 
814 	return (count);
815 }
816 
817 void
818 pf_mtag_initialize()
819 {
820 
821 	pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
822 	    sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
823 	    UMA_ALIGN_PTR, 0);
824 }
825 
826 /* Per-vnet data storage structures initialization. */
827 void
828 pf_initialize()
829 {
830 	struct pf_keyhash	*kh;
831 	struct pf_idhash	*ih;
832 	struct pf_srchash	*sh;
833 	u_int i;
834 
835 	if (pf_hashsize == 0 || !powerof2(pf_hashsize))
836 		pf_hashsize = PF_HASHSIZ;
837 	if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
838 		pf_srchashsize = PF_SRCHASHSIZ;
839 
840 	V_pf_hashseed = arc4random();
841 
842 	/* States and state keys storage. */
843 	V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
844 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
845 	V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
846 	uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
847 	uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
848 
849 	V_pf_state_key_z = uma_zcreate("pf state keys",
850 	    sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
851 	    UMA_ALIGN_PTR, 0);
852 
853 	V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash),
854 	    M_PFHASH, M_NOWAIT | M_ZERO);
855 	V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash),
856 	    M_PFHASH, M_NOWAIT | M_ZERO);
857 	if (V_pf_keyhash == NULL || V_pf_idhash == NULL) {
858 		printf("pf: Unable to allocate memory for "
859 		    "state_hashsize %lu.\n", pf_hashsize);
860 
861 		free(V_pf_keyhash, M_PFHASH);
862 		free(V_pf_idhash, M_PFHASH);
863 
864 		pf_hashsize = PF_HASHSIZ;
865 		V_pf_keyhash = mallocarray(pf_hashsize,
866 		    sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO);
867 		V_pf_idhash = mallocarray(pf_hashsize,
868 		    sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO);
869 	}
870 
871 	pf_hashmask = pf_hashsize - 1;
872 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
873 	    i++, kh++, ih++) {
874 		mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
875 		mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
876 	}
877 
878 	/* Source nodes. */
879 	V_pf_sources_z = uma_zcreate("pf source nodes",
880 	    sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
881 	    0);
882 	V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
883 	uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
884 	uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
885 
886 	V_pf_srchash = mallocarray(pf_srchashsize,
887 	    sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO);
888 	if (V_pf_srchash == NULL) {
889 		printf("pf: Unable to allocate memory for "
890 		    "source_hashsize %lu.\n", pf_srchashsize);
891 
892 		pf_srchashsize = PF_SRCHASHSIZ;
893 		V_pf_srchash = mallocarray(pf_srchashsize,
894 		    sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO);
895 	}
896 
897 	pf_srchashmask = pf_srchashsize - 1;
898 	for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
899 		mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
900 
901 	/* ALTQ */
902 	TAILQ_INIT(&V_pf_altqs[0]);
903 	TAILQ_INIT(&V_pf_altqs[1]);
904 	TAILQ_INIT(&V_pf_altqs[2]);
905 	TAILQ_INIT(&V_pf_altqs[3]);
906 	TAILQ_INIT(&V_pf_pabuf);
907 	V_pf_altqs_active = &V_pf_altqs[0];
908 	V_pf_altq_ifs_active = &V_pf_altqs[1];
909 	V_pf_altqs_inactive = &V_pf_altqs[2];
910 	V_pf_altq_ifs_inactive = &V_pf_altqs[3];
911 
912 	/* Send & overload+flush queues. */
913 	STAILQ_INIT(&V_pf_sendqueue);
914 	SLIST_INIT(&V_pf_overloadqueue);
915 	TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
916 
917 	/* Unlinked, but may be referenced rules. */
918 	TAILQ_INIT(&V_pf_unlinked_rules);
919 }
920 
921 void
922 pf_mtag_cleanup()
923 {
924 
925 	uma_zdestroy(pf_mtag_z);
926 }
927 
928 void
929 pf_cleanup()
930 {
931 	struct pf_keyhash	*kh;
932 	struct pf_idhash	*ih;
933 	struct pf_srchash	*sh;
934 	struct pf_send_entry	*pfse, *next;
935 	u_int i;
936 
937 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
938 	    i++, kh++, ih++) {
939 		KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
940 		    __func__));
941 		KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
942 		    __func__));
943 		mtx_destroy(&kh->lock);
944 		mtx_destroy(&ih->lock);
945 	}
946 	free(V_pf_keyhash, M_PFHASH);
947 	free(V_pf_idhash, M_PFHASH);
948 
949 	for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
950 		KASSERT(LIST_EMPTY(&sh->nodes),
951 		    ("%s: source node hash not empty", __func__));
952 		mtx_destroy(&sh->lock);
953 	}
954 	free(V_pf_srchash, M_PFHASH);
955 
956 	STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
957 		m_freem(pfse->pfse_m);
958 		free(pfse, M_PFTEMP);
959 	}
960 
961 	uma_zdestroy(V_pf_sources_z);
962 	uma_zdestroy(V_pf_state_z);
963 	uma_zdestroy(V_pf_state_key_z);
964 }
965 
966 static int
967 pf_mtag_uminit(void *mem, int size, int how)
968 {
969 	struct m_tag *t;
970 
971 	t = (struct m_tag *)mem;
972 	t->m_tag_cookie = MTAG_ABI_COMPAT;
973 	t->m_tag_id = PACKET_TAG_PF;
974 	t->m_tag_len = sizeof(struct pf_mtag);
975 	t->m_tag_free = pf_mtag_free;
976 
977 	return (0);
978 }
979 
980 static void
981 pf_mtag_free(struct m_tag *t)
982 {
983 
984 	uma_zfree(pf_mtag_z, t);
985 }
986 
987 struct pf_mtag *
988 pf_get_mtag(struct mbuf *m)
989 {
990 	struct m_tag *mtag;
991 
992 	if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
993 		return ((struct pf_mtag *)(mtag + 1));
994 
995 	mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
996 	if (mtag == NULL)
997 		return (NULL);
998 	bzero(mtag + 1, sizeof(struct pf_mtag));
999 	m_tag_prepend(m, mtag);
1000 
1001 	return ((struct pf_mtag *)(mtag + 1));
1002 }
1003 
1004 static int
1005 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
1006     struct pf_state *s)
1007 {
1008 	struct pf_keyhash	*khs, *khw, *kh;
1009 	struct pf_state_key	*sk, *cur;
1010 	struct pf_state		*si, *olds = NULL;
1011 	int idx;
1012 
1013 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1014 	KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
1015 	KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
1016 
1017 	/*
1018 	 * We need to lock hash slots of both keys. To avoid deadlock
1019 	 * we always lock the slot with lower address first. Unlock order
1020 	 * isn't important.
1021 	 *
1022 	 * We also need to lock ID hash slot before dropping key
1023 	 * locks. On success we return with ID hash slot locked.
1024 	 */
1025 
1026 	if (skw == sks) {
1027 		khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
1028 		PF_HASHROW_LOCK(khs);
1029 	} else {
1030 		khs = &V_pf_keyhash[pf_hashkey(sks)];
1031 		khw = &V_pf_keyhash[pf_hashkey(skw)];
1032 		if (khs == khw) {
1033 			PF_HASHROW_LOCK(khs);
1034 		} else if (khs < khw) {
1035 			PF_HASHROW_LOCK(khs);
1036 			PF_HASHROW_LOCK(khw);
1037 		} else {
1038 			PF_HASHROW_LOCK(khw);
1039 			PF_HASHROW_LOCK(khs);
1040 		}
1041 	}
1042 
1043 #define	KEYS_UNLOCK()	do {			\
1044 	if (khs != khw) {			\
1045 		PF_HASHROW_UNLOCK(khs);		\
1046 		PF_HASHROW_UNLOCK(khw);		\
1047 	} else					\
1048 		PF_HASHROW_UNLOCK(khs);		\
1049 } while (0)
1050 
1051 	/*
1052 	 * First run: start with wire key.
1053 	 */
1054 	sk = skw;
1055 	kh = khw;
1056 	idx = PF_SK_WIRE;
1057 
1058 keyattach:
1059 	LIST_FOREACH(cur, &kh->keys, entry)
1060 		if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
1061 			break;
1062 
1063 	if (cur != NULL) {
1064 		/* Key exists. Check for same kif, if none, add to key. */
1065 		TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
1066 			struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
1067 
1068 			PF_HASHROW_LOCK(ih);
1069 			if (si->kif == s->kif &&
1070 			    si->direction == s->direction) {
1071 				if (sk->proto == IPPROTO_TCP &&
1072 				    si->src.state >= TCPS_FIN_WAIT_2 &&
1073 				    si->dst.state >= TCPS_FIN_WAIT_2) {
1074 					/*
1075 					 * New state matches an old >FIN_WAIT_2
1076 					 * state. We can't drop key hash locks,
1077 					 * thus we can't unlink it properly.
1078 					 *
1079 					 * As a workaround we drop it into
1080 					 * TCPS_CLOSED state, schedule purge
1081 					 * ASAP and push it into the very end
1082 					 * of the slot TAILQ, so that it won't
1083 					 * conflict with our new state.
1084 					 */
1085 					si->src.state = si->dst.state =
1086 					    TCPS_CLOSED;
1087 					si->timeout = PFTM_PURGE;
1088 					olds = si;
1089 				} else {
1090 					if (V_pf_status.debug >= PF_DEBUG_MISC) {
1091 						printf("pf: %s key attach "
1092 						    "failed on %s: ",
1093 						    (idx == PF_SK_WIRE) ?
1094 						    "wire" : "stack",
1095 						    s->kif->pfik_name);
1096 						pf_print_state_parts(s,
1097 						    (idx == PF_SK_WIRE) ?
1098 						    sk : NULL,
1099 						    (idx == PF_SK_STACK) ?
1100 						    sk : NULL);
1101 						printf(", existing: ");
1102 						pf_print_state_parts(si,
1103 						    (idx == PF_SK_WIRE) ?
1104 						    sk : NULL,
1105 						    (idx == PF_SK_STACK) ?
1106 						    sk : NULL);
1107 						printf("\n");
1108 					}
1109 					PF_HASHROW_UNLOCK(ih);
1110 					KEYS_UNLOCK();
1111 					uma_zfree(V_pf_state_key_z, sk);
1112 					if (idx == PF_SK_STACK)
1113 						pf_detach_state(s);
1114 					return (EEXIST); /* collision! */
1115 				}
1116 			}
1117 			PF_HASHROW_UNLOCK(ih);
1118 		}
1119 		uma_zfree(V_pf_state_key_z, sk);
1120 		s->key[idx] = cur;
1121 	} else {
1122 		LIST_INSERT_HEAD(&kh->keys, sk, entry);
1123 		s->key[idx] = sk;
1124 	}
1125 
1126 stateattach:
1127 	/* List is sorted, if-bound states before floating. */
1128 	if (s->kif == V_pfi_all)
1129 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1130 	else
1131 		TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1132 
1133 	if (olds) {
1134 		TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1135 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1136 		    key_list[idx]);
1137 		olds = NULL;
1138 	}
1139 
1140 	/*
1141 	 * Attach done. See how should we (or should not?)
1142 	 * attach a second key.
1143 	 */
1144 	if (sks == skw) {
1145 		s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1146 		idx = PF_SK_STACK;
1147 		sks = NULL;
1148 		goto stateattach;
1149 	} else if (sks != NULL) {
1150 		/*
1151 		 * Continue attaching with stack key.
1152 		 */
1153 		sk = sks;
1154 		kh = khs;
1155 		idx = PF_SK_STACK;
1156 		sks = NULL;
1157 		goto keyattach;
1158 	}
1159 
1160 	PF_STATE_LOCK(s);
1161 	KEYS_UNLOCK();
1162 
1163 	KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1164 	    ("%s failure", __func__));
1165 
1166 	return (0);
1167 #undef	KEYS_UNLOCK
1168 }
1169 
1170 static void
1171 pf_detach_state(struct pf_state *s)
1172 {
1173 	struct pf_state_key *sks = s->key[PF_SK_STACK];
1174 	struct pf_keyhash *kh;
1175 
1176 	if (sks != NULL) {
1177 		kh = &V_pf_keyhash[pf_hashkey(sks)];
1178 		PF_HASHROW_LOCK(kh);
1179 		if (s->key[PF_SK_STACK] != NULL)
1180 			pf_state_key_detach(s, PF_SK_STACK);
1181 		/*
1182 		 * If both point to same key, then we are done.
1183 		 */
1184 		if (sks == s->key[PF_SK_WIRE]) {
1185 			pf_state_key_detach(s, PF_SK_WIRE);
1186 			PF_HASHROW_UNLOCK(kh);
1187 			return;
1188 		}
1189 		PF_HASHROW_UNLOCK(kh);
1190 	}
1191 
1192 	if (s->key[PF_SK_WIRE] != NULL) {
1193 		kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1194 		PF_HASHROW_LOCK(kh);
1195 		if (s->key[PF_SK_WIRE] != NULL)
1196 			pf_state_key_detach(s, PF_SK_WIRE);
1197 		PF_HASHROW_UNLOCK(kh);
1198 	}
1199 }
1200 
1201 static void
1202 pf_state_key_detach(struct pf_state *s, int idx)
1203 {
1204 	struct pf_state_key *sk = s->key[idx];
1205 #ifdef INVARIANTS
1206 	struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1207 
1208 	PF_HASHROW_ASSERT(kh);
1209 #endif
1210 	TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1211 	s->key[idx] = NULL;
1212 
1213 	if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1214 		LIST_REMOVE(sk, entry);
1215 		uma_zfree(V_pf_state_key_z, sk);
1216 	}
1217 }
1218 
1219 static int
1220 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1221 {
1222 	struct pf_state_key *sk = mem;
1223 
1224 	bzero(sk, sizeof(struct pf_state_key_cmp));
1225 	TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1226 	TAILQ_INIT(&sk->states[PF_SK_STACK]);
1227 
1228 	return (0);
1229 }
1230 
1231 struct pf_state_key *
1232 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1233 	struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1234 {
1235 	struct pf_state_key *sk;
1236 
1237 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1238 	if (sk == NULL)
1239 		return (NULL);
1240 
1241 	PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1242 	PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1243 	sk->port[pd->sidx] = sport;
1244 	sk->port[pd->didx] = dport;
1245 	sk->proto = pd->proto;
1246 	sk->af = pd->af;
1247 
1248 	return (sk);
1249 }
1250 
1251 struct pf_state_key *
1252 pf_state_key_clone(struct pf_state_key *orig)
1253 {
1254 	struct pf_state_key *sk;
1255 
1256 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1257 	if (sk == NULL)
1258 		return (NULL);
1259 
1260 	bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1261 
1262 	return (sk);
1263 }
1264 
1265 int
1266 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif,
1267     struct pf_state_key *skw, struct pf_state_key *sks, struct pf_state *s)
1268 {
1269 	struct pf_idhash *ih;
1270 	struct pf_state *cur;
1271 	int error;
1272 
1273 	KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1274 	    ("%s: sks not pristine", __func__));
1275 	KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1276 	    ("%s: skw not pristine", __func__));
1277 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1278 
1279 	s->kif = kif;
1280 	s->orig_kif = orig_kif;
1281 
1282 	if (s->id == 0 && s->creatorid == 0) {
1283 		/* XXX: should be atomic, but probability of collision low */
1284 		if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1285 			V_pf_stateid[curcpu] = 1;
1286 		s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1287 		s->id = htobe64(s->id);
1288 		s->creatorid = V_pf_status.hostid;
1289 	}
1290 
1291 	/* Returns with ID locked on success. */
1292 	if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1293 		return (error);
1294 
1295 	ih = &V_pf_idhash[PF_IDHASH(s)];
1296 	PF_HASHROW_ASSERT(ih);
1297 	LIST_FOREACH(cur, &ih->states, entry)
1298 		if (cur->id == s->id && cur->creatorid == s->creatorid)
1299 			break;
1300 
1301 	if (cur != NULL) {
1302 		PF_HASHROW_UNLOCK(ih);
1303 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1304 			printf("pf: state ID collision: "
1305 			    "id: %016llx creatorid: %08x\n",
1306 			    (unsigned long long)be64toh(s->id),
1307 			    ntohl(s->creatorid));
1308 		}
1309 		pf_detach_state(s);
1310 		return (EEXIST);
1311 	}
1312 	LIST_INSERT_HEAD(&ih->states, s, entry);
1313 	/* One for keys, one for ID hash. */
1314 	refcount_init(&s->refs, 2);
1315 
1316 	counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1317 	if (V_pfsync_insert_state_ptr != NULL)
1318 		V_pfsync_insert_state_ptr(s);
1319 
1320 	/* Returns locked. */
1321 	return (0);
1322 }
1323 
1324 /*
1325  * Find state by ID: returns with locked row on success.
1326  */
1327 struct pf_state *
1328 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1329 {
1330 	struct pf_idhash *ih;
1331 	struct pf_state *s;
1332 
1333 	counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1334 
1335 	ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1336 
1337 	PF_HASHROW_LOCK(ih);
1338 	LIST_FOREACH(s, &ih->states, entry)
1339 		if (s->id == id && s->creatorid == creatorid)
1340 			break;
1341 
1342 	if (s == NULL)
1343 		PF_HASHROW_UNLOCK(ih);
1344 
1345 	return (s);
1346 }
1347 
1348 /*
1349  * Find state by key.
1350  * Returns with ID hash slot locked on success.
1351  */
1352 static struct pf_state *
1353 pf_find_state(struct pfi_kkif *kif, struct pf_state_key_cmp *key, u_int dir)
1354 {
1355 	struct pf_keyhash	*kh;
1356 	struct pf_state_key	*sk;
1357 	struct pf_state		*s;
1358 	int idx;
1359 
1360 	counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1361 
1362 	kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1363 
1364 	PF_HASHROW_LOCK(kh);
1365 	LIST_FOREACH(sk, &kh->keys, entry)
1366 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1367 			break;
1368 	if (sk == NULL) {
1369 		PF_HASHROW_UNLOCK(kh);
1370 		return (NULL);
1371 	}
1372 
1373 	idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1374 
1375 	/* List is sorted, if-bound states before floating ones. */
1376 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1377 		if (s->kif == V_pfi_all || s->kif == kif) {
1378 			PF_STATE_LOCK(s);
1379 			PF_HASHROW_UNLOCK(kh);
1380 			if (s->timeout >= PFTM_MAX) {
1381 				/*
1382 				 * State is either being processed by
1383 				 * pf_unlink_state() in an other thread, or
1384 				 * is scheduled for immediate expiry.
1385 				 */
1386 				PF_STATE_UNLOCK(s);
1387 				return (NULL);
1388 			}
1389 			return (s);
1390 		}
1391 	PF_HASHROW_UNLOCK(kh);
1392 
1393 	return (NULL);
1394 }
1395 
1396 struct pf_state *
1397 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1398 {
1399 	struct pf_keyhash	*kh;
1400 	struct pf_state_key	*sk;
1401 	struct pf_state		*s, *ret = NULL;
1402 	int			 idx, inout = 0;
1403 
1404 	counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1405 
1406 	kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1407 
1408 	PF_HASHROW_LOCK(kh);
1409 	LIST_FOREACH(sk, &kh->keys, entry)
1410 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1411 			break;
1412 	if (sk == NULL) {
1413 		PF_HASHROW_UNLOCK(kh);
1414 		return (NULL);
1415 	}
1416 	switch (dir) {
1417 	case PF_IN:
1418 		idx = PF_SK_WIRE;
1419 		break;
1420 	case PF_OUT:
1421 		idx = PF_SK_STACK;
1422 		break;
1423 	case PF_INOUT:
1424 		idx = PF_SK_WIRE;
1425 		inout = 1;
1426 		break;
1427 	default:
1428 		panic("%s: dir %u", __func__, dir);
1429 	}
1430 second_run:
1431 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1432 		if (more == NULL) {
1433 			PF_HASHROW_UNLOCK(kh);
1434 			return (s);
1435 		}
1436 
1437 		if (ret)
1438 			(*more)++;
1439 		else
1440 			ret = s;
1441 	}
1442 	if (inout == 1) {
1443 		inout = 0;
1444 		idx = PF_SK_STACK;
1445 		goto second_run;
1446 	}
1447 	PF_HASHROW_UNLOCK(kh);
1448 
1449 	return (ret);
1450 }
1451 
1452 /* END state table stuff */
1453 
1454 static void
1455 pf_send(struct pf_send_entry *pfse)
1456 {
1457 
1458 	PF_SENDQ_LOCK();
1459 	STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1460 	PF_SENDQ_UNLOCK();
1461 	swi_sched(V_pf_swi_cookie, 0);
1462 }
1463 
1464 void
1465 pf_intr(void *v)
1466 {
1467 	struct epoch_tracker et;
1468 	struct pf_send_head queue;
1469 	struct pf_send_entry *pfse, *next;
1470 
1471 	CURVNET_SET((struct vnet *)v);
1472 
1473 	PF_SENDQ_LOCK();
1474 	queue = V_pf_sendqueue;
1475 	STAILQ_INIT(&V_pf_sendqueue);
1476 	PF_SENDQ_UNLOCK();
1477 
1478 	NET_EPOCH_ENTER(et);
1479 
1480 	STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1481 		switch (pfse->pfse_type) {
1482 #ifdef INET
1483 		case PFSE_IP:
1484 			ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1485 			break;
1486 		case PFSE_ICMP:
1487 			icmp_error(pfse->pfse_m, pfse->icmpopts.type,
1488 			    pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
1489 			break;
1490 #endif /* INET */
1491 #ifdef INET6
1492 		case PFSE_IP6:
1493 			ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1494 			    NULL);
1495 			break;
1496 		case PFSE_ICMP6:
1497 			icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
1498 			    pfse->icmpopts.code, pfse->icmpopts.mtu);
1499 			break;
1500 #endif /* INET6 */
1501 		default:
1502 			panic("%s: unknown type", __func__);
1503 		}
1504 		free(pfse, M_PFTEMP);
1505 	}
1506 	NET_EPOCH_EXIT(et);
1507 	CURVNET_RESTORE();
1508 }
1509 
1510 void
1511 pf_purge_thread(void *unused __unused)
1512 {
1513 	VNET_ITERATOR_DECL(vnet_iter);
1514 
1515 	sx_xlock(&pf_end_lock);
1516 	while (pf_end_threads == 0) {
1517 		sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", hz / 10);
1518 
1519 		VNET_LIST_RLOCK();
1520 		VNET_FOREACH(vnet_iter) {
1521 			CURVNET_SET(vnet_iter);
1522 
1523 			/* Wait until V_pf_default_rule is initialized. */
1524 			if (V_pf_vnet_active == 0) {
1525 				CURVNET_RESTORE();
1526 				continue;
1527 			}
1528 
1529 			/*
1530 			 *  Process 1/interval fraction of the state
1531 			 * table every run.
1532 			 */
1533 			V_pf_purge_idx =
1534 			    pf_purge_expired_states(V_pf_purge_idx, pf_hashmask /
1535 			    (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1536 
1537 			/*
1538 			 * Purge other expired types every
1539 			 * PFTM_INTERVAL seconds.
1540 			 */
1541 			if (V_pf_purge_idx == 0) {
1542 				/*
1543 				 * Order is important:
1544 				 * - states and src nodes reference rules
1545 				 * - states and rules reference kifs
1546 				 */
1547 				pf_purge_expired_fragments();
1548 				pf_purge_expired_src_nodes();
1549 				pf_purge_unlinked_rules();
1550 				pfi_kkif_purge();
1551 			}
1552 			CURVNET_RESTORE();
1553 		}
1554 		VNET_LIST_RUNLOCK();
1555 	}
1556 
1557 	pf_end_threads++;
1558 	sx_xunlock(&pf_end_lock);
1559 	kproc_exit(0);
1560 }
1561 
1562 void
1563 pf_unload_vnet_purge(void)
1564 {
1565 
1566 	/*
1567 	 * To cleanse up all kifs and rules we need
1568 	 * two runs: first one clears reference flags,
1569 	 * then pf_purge_expired_states() doesn't
1570 	 * raise them, and then second run frees.
1571 	 */
1572 	pf_purge_unlinked_rules();
1573 	pfi_kkif_purge();
1574 
1575 	/*
1576 	 * Now purge everything.
1577 	 */
1578 	pf_purge_expired_states(0, pf_hashmask);
1579 	pf_purge_fragments(UINT_MAX);
1580 	pf_purge_expired_src_nodes();
1581 
1582 	/*
1583 	 * Now all kifs & rules should be unreferenced,
1584 	 * thus should be successfully freed.
1585 	 */
1586 	pf_purge_unlinked_rules();
1587 	pfi_kkif_purge();
1588 }
1589 
1590 u_int32_t
1591 pf_state_expires(const struct pf_state *state)
1592 {
1593 	u_int32_t	timeout;
1594 	u_int32_t	start;
1595 	u_int32_t	end;
1596 	u_int32_t	states;
1597 
1598 	/* handle all PFTM_* > PFTM_MAX here */
1599 	if (state->timeout == PFTM_PURGE)
1600 		return (time_uptime);
1601 	KASSERT(state->timeout != PFTM_UNLINKED,
1602 	    ("pf_state_expires: timeout == PFTM_UNLINKED"));
1603 	KASSERT((state->timeout < PFTM_MAX),
1604 	    ("pf_state_expires: timeout > PFTM_MAX"));
1605 	timeout = state->rule.ptr->timeout[state->timeout];
1606 	if (!timeout)
1607 		timeout = V_pf_default_rule.timeout[state->timeout];
1608 	start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1609 	if (start && state->rule.ptr != &V_pf_default_rule) {
1610 		end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1611 		states = counter_u64_fetch(state->rule.ptr->states_cur);
1612 	} else {
1613 		start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1614 		end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1615 		states = V_pf_status.states;
1616 	}
1617 	if (end && states > start && start < end) {
1618 		if (states < end) {
1619 			timeout = (u_int64_t)timeout * (end - states) /
1620 			    (end - start);
1621 			return (state->expire + timeout);
1622 		}
1623 		else
1624 			return (time_uptime);
1625 	}
1626 	return (state->expire + timeout);
1627 }
1628 
1629 void
1630 pf_purge_expired_src_nodes()
1631 {
1632 	struct pf_ksrc_node_list	 freelist;
1633 	struct pf_srchash	*sh;
1634 	struct pf_ksrc_node	*cur, *next;
1635 	int i;
1636 
1637 	LIST_INIT(&freelist);
1638 	for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1639 	    PF_HASHROW_LOCK(sh);
1640 	    LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1641 		if (cur->states == 0 && cur->expire <= time_uptime) {
1642 			pf_unlink_src_node(cur);
1643 			LIST_INSERT_HEAD(&freelist, cur, entry);
1644 		} else if (cur->rule.ptr != NULL)
1645 			cur->rule.ptr->rule_ref |= PFRULE_REFS;
1646 	    PF_HASHROW_UNLOCK(sh);
1647 	}
1648 
1649 	pf_free_src_nodes(&freelist);
1650 
1651 	V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1652 }
1653 
1654 static void
1655 pf_src_tree_remove_state(struct pf_state *s)
1656 {
1657 	struct pf_ksrc_node *sn;
1658 	struct pf_srchash *sh;
1659 	uint32_t timeout;
1660 
1661 	timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
1662 	    s->rule.ptr->timeout[PFTM_SRC_NODE] :
1663 	    V_pf_default_rule.timeout[PFTM_SRC_NODE];
1664 
1665 	if (s->src_node != NULL) {
1666 		sn = s->src_node;
1667 		sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1668 	    	PF_HASHROW_LOCK(sh);
1669 		if (s->src.tcp_est)
1670 			--sn->conn;
1671 		if (--sn->states == 0)
1672 			sn->expire = time_uptime + timeout;
1673 	    	PF_HASHROW_UNLOCK(sh);
1674 	}
1675 	if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1676 		sn = s->nat_src_node;
1677 		sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1678 	    	PF_HASHROW_LOCK(sh);
1679 		if (--sn->states == 0)
1680 			sn->expire = time_uptime + timeout;
1681 	    	PF_HASHROW_UNLOCK(sh);
1682 	}
1683 	s->src_node = s->nat_src_node = NULL;
1684 }
1685 
1686 /*
1687  * Unlink and potentilly free a state. Function may be
1688  * called with ID hash row locked, but always returns
1689  * unlocked, since it needs to go through key hash locking.
1690  */
1691 int
1692 pf_unlink_state(struct pf_state *s, u_int flags)
1693 {
1694 	struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1695 
1696 	if ((flags & PF_ENTER_LOCKED) == 0)
1697 		PF_HASHROW_LOCK(ih);
1698 	else
1699 		PF_HASHROW_ASSERT(ih);
1700 
1701 	if (s->timeout == PFTM_UNLINKED) {
1702 		/*
1703 		 * State is being processed
1704 		 * by pf_unlink_state() in
1705 		 * an other thread.
1706 		 */
1707 		PF_HASHROW_UNLOCK(ih);
1708 		return (0);	/* XXXGL: undefined actually */
1709 	}
1710 
1711 	if (s->src.state == PF_TCPS_PROXY_DST) {
1712 		/* XXX wire key the right one? */
1713 		pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1714 		    &s->key[PF_SK_WIRE]->addr[1],
1715 		    &s->key[PF_SK_WIRE]->addr[0],
1716 		    s->key[PF_SK_WIRE]->port[1],
1717 		    s->key[PF_SK_WIRE]->port[0],
1718 		    s->src.seqhi, s->src.seqlo + 1,
1719 		    TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1720 	}
1721 
1722 	LIST_REMOVE(s, entry);
1723 	pf_src_tree_remove_state(s);
1724 
1725 	if (V_pfsync_delete_state_ptr != NULL)
1726 		V_pfsync_delete_state_ptr(s);
1727 
1728 	STATE_DEC_COUNTERS(s);
1729 
1730 	s->timeout = PFTM_UNLINKED;
1731 
1732 	PF_HASHROW_UNLOCK(ih);
1733 
1734 	pf_detach_state(s);
1735 	/* pf_state_insert() initialises refs to 2 */
1736 	return (pf_release_staten(s, 2));
1737 }
1738 
1739 struct pf_state *
1740 pf_alloc_state(int flags)
1741 {
1742 
1743 	return (uma_zalloc(V_pf_state_z, flags | M_ZERO));
1744 }
1745 
1746 void
1747 pf_free_state(struct pf_state *cur)
1748 {
1749 
1750 	KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1751 	KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1752 	    cur->timeout));
1753 
1754 	pf_normalize_tcp_cleanup(cur);
1755 	uma_zfree(V_pf_state_z, cur);
1756 	counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
1757 }
1758 
1759 /*
1760  * Called only from pf_purge_thread(), thus serialized.
1761  */
1762 static u_int
1763 pf_purge_expired_states(u_int i, int maxcheck)
1764 {
1765 	struct pf_idhash *ih;
1766 	struct pf_state *s;
1767 
1768 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1769 
1770 	/*
1771 	 * Go through hash and unlink states that expire now.
1772 	 */
1773 	while (maxcheck > 0) {
1774 		ih = &V_pf_idhash[i];
1775 
1776 		/* only take the lock if we expect to do work */
1777 		if (!LIST_EMPTY(&ih->states)) {
1778 relock:
1779 			PF_HASHROW_LOCK(ih);
1780 			LIST_FOREACH(s, &ih->states, entry) {
1781 				if (pf_state_expires(s) <= time_uptime) {
1782 					V_pf_status.states -=
1783 					    pf_unlink_state(s, PF_ENTER_LOCKED);
1784 					goto relock;
1785 				}
1786 				s->rule.ptr->rule_ref |= PFRULE_REFS;
1787 				if (s->nat_rule.ptr != NULL)
1788 					s->nat_rule.ptr->rule_ref |= PFRULE_REFS;
1789 				if (s->anchor.ptr != NULL)
1790 					s->anchor.ptr->rule_ref |= PFRULE_REFS;
1791 				s->kif->pfik_flags |= PFI_IFLAG_REFS;
1792 				if (s->rt_kif)
1793 					s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1794 			}
1795 			PF_HASHROW_UNLOCK(ih);
1796 		}
1797 
1798 		/* Return when we hit end of hash. */
1799 		if (++i > pf_hashmask) {
1800 			V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1801 			return (0);
1802 		}
1803 
1804 		maxcheck--;
1805 	}
1806 
1807 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1808 
1809 	return (i);
1810 }
1811 
1812 static void
1813 pf_purge_unlinked_rules()
1814 {
1815 	struct pf_krulequeue tmpq;
1816 	struct pf_krule *r, *r1;
1817 
1818 	/*
1819 	 * If we have overloading task pending, then we'd
1820 	 * better skip purging this time. There is a tiny
1821 	 * probability that overloading task references
1822 	 * an already unlinked rule.
1823 	 */
1824 	PF_OVERLOADQ_LOCK();
1825 	if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1826 		PF_OVERLOADQ_UNLOCK();
1827 		return;
1828 	}
1829 	PF_OVERLOADQ_UNLOCK();
1830 
1831 	/*
1832 	 * Do naive mark-and-sweep garbage collecting of old rules.
1833 	 * Reference flag is raised by pf_purge_expired_states()
1834 	 * and pf_purge_expired_src_nodes().
1835 	 *
1836 	 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1837 	 * use a temporary queue.
1838 	 */
1839 	TAILQ_INIT(&tmpq);
1840 	PF_UNLNKDRULES_LOCK();
1841 	TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1842 		if (!(r->rule_ref & PFRULE_REFS)) {
1843 			TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1844 			TAILQ_INSERT_TAIL(&tmpq, r, entries);
1845 		} else
1846 			r->rule_ref &= ~PFRULE_REFS;
1847 	}
1848 	PF_UNLNKDRULES_UNLOCK();
1849 
1850 	if (!TAILQ_EMPTY(&tmpq)) {
1851 		PF_RULES_WLOCK();
1852 		TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1853 			TAILQ_REMOVE(&tmpq, r, entries);
1854 			pf_free_rule(r);
1855 		}
1856 		PF_RULES_WUNLOCK();
1857 	}
1858 }
1859 
1860 void
1861 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1862 {
1863 	switch (af) {
1864 #ifdef INET
1865 	case AF_INET: {
1866 		u_int32_t a = ntohl(addr->addr32[0]);
1867 		printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1868 		    (a>>8)&255, a&255);
1869 		if (p) {
1870 			p = ntohs(p);
1871 			printf(":%u", p);
1872 		}
1873 		break;
1874 	}
1875 #endif /* INET */
1876 #ifdef INET6
1877 	case AF_INET6: {
1878 		u_int16_t b;
1879 		u_int8_t i, curstart, curend, maxstart, maxend;
1880 		curstart = curend = maxstart = maxend = 255;
1881 		for (i = 0; i < 8; i++) {
1882 			if (!addr->addr16[i]) {
1883 				if (curstart == 255)
1884 					curstart = i;
1885 				curend = i;
1886 			} else {
1887 				if ((curend - curstart) >
1888 				    (maxend - maxstart)) {
1889 					maxstart = curstart;
1890 					maxend = curend;
1891 				}
1892 				curstart = curend = 255;
1893 			}
1894 		}
1895 		if ((curend - curstart) >
1896 		    (maxend - maxstart)) {
1897 			maxstart = curstart;
1898 			maxend = curend;
1899 		}
1900 		for (i = 0; i < 8; i++) {
1901 			if (i >= maxstart && i <= maxend) {
1902 				if (i == 0)
1903 					printf(":");
1904 				if (i == maxend)
1905 					printf(":");
1906 			} else {
1907 				b = ntohs(addr->addr16[i]);
1908 				printf("%x", b);
1909 				if (i < 7)
1910 					printf(":");
1911 			}
1912 		}
1913 		if (p) {
1914 			p = ntohs(p);
1915 			printf("[%u]", p);
1916 		}
1917 		break;
1918 	}
1919 #endif /* INET6 */
1920 	}
1921 }
1922 
1923 void
1924 pf_print_state(struct pf_state *s)
1925 {
1926 	pf_print_state_parts(s, NULL, NULL);
1927 }
1928 
1929 static void
1930 pf_print_state_parts(struct pf_state *s,
1931     struct pf_state_key *skwp, struct pf_state_key *sksp)
1932 {
1933 	struct pf_state_key *skw, *sks;
1934 	u_int8_t proto, dir;
1935 
1936 	/* Do our best to fill these, but they're skipped if NULL */
1937 	skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1938 	sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1939 	proto = skw ? skw->proto : (sks ? sks->proto : 0);
1940 	dir = s ? s->direction : 0;
1941 
1942 	switch (proto) {
1943 	case IPPROTO_IPV4:
1944 		printf("IPv4");
1945 		break;
1946 	case IPPROTO_IPV6:
1947 		printf("IPv6");
1948 		break;
1949 	case IPPROTO_TCP:
1950 		printf("TCP");
1951 		break;
1952 	case IPPROTO_UDP:
1953 		printf("UDP");
1954 		break;
1955 	case IPPROTO_ICMP:
1956 		printf("ICMP");
1957 		break;
1958 	case IPPROTO_ICMPV6:
1959 		printf("ICMPv6");
1960 		break;
1961 	default:
1962 		printf("%u", proto);
1963 		break;
1964 	}
1965 	switch (dir) {
1966 	case PF_IN:
1967 		printf(" in");
1968 		break;
1969 	case PF_OUT:
1970 		printf(" out");
1971 		break;
1972 	}
1973 	if (skw) {
1974 		printf(" wire: ");
1975 		pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1976 		printf(" ");
1977 		pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1978 	}
1979 	if (sks) {
1980 		printf(" stack: ");
1981 		if (sks != skw) {
1982 			pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1983 			printf(" ");
1984 			pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1985 		} else
1986 			printf("-");
1987 	}
1988 	if (s) {
1989 		if (proto == IPPROTO_TCP) {
1990 			printf(" [lo=%u high=%u win=%u modulator=%u",
1991 			    s->src.seqlo, s->src.seqhi,
1992 			    s->src.max_win, s->src.seqdiff);
1993 			if (s->src.wscale && s->dst.wscale)
1994 				printf(" wscale=%u",
1995 				    s->src.wscale & PF_WSCALE_MASK);
1996 			printf("]");
1997 			printf(" [lo=%u high=%u win=%u modulator=%u",
1998 			    s->dst.seqlo, s->dst.seqhi,
1999 			    s->dst.max_win, s->dst.seqdiff);
2000 			if (s->src.wscale && s->dst.wscale)
2001 				printf(" wscale=%u",
2002 				s->dst.wscale & PF_WSCALE_MASK);
2003 			printf("]");
2004 		}
2005 		printf(" %u:%u", s->src.state, s->dst.state);
2006 	}
2007 }
2008 
2009 void
2010 pf_print_flags(u_int8_t f)
2011 {
2012 	if (f)
2013 		printf(" ");
2014 	if (f & TH_FIN)
2015 		printf("F");
2016 	if (f & TH_SYN)
2017 		printf("S");
2018 	if (f & TH_RST)
2019 		printf("R");
2020 	if (f & TH_PUSH)
2021 		printf("P");
2022 	if (f & TH_ACK)
2023 		printf("A");
2024 	if (f & TH_URG)
2025 		printf("U");
2026 	if (f & TH_ECE)
2027 		printf("E");
2028 	if (f & TH_CWR)
2029 		printf("W");
2030 }
2031 
2032 #define	PF_SET_SKIP_STEPS(i)					\
2033 	do {							\
2034 		while (head[i] != cur) {			\
2035 			head[i]->skip[i].ptr = cur;		\
2036 			head[i] = TAILQ_NEXT(head[i], entries);	\
2037 		}						\
2038 	} while (0)
2039 
2040 void
2041 pf_calc_skip_steps(struct pf_krulequeue *rules)
2042 {
2043 	struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT];
2044 	int i;
2045 
2046 	cur = TAILQ_FIRST(rules);
2047 	prev = cur;
2048 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2049 		head[i] = cur;
2050 	while (cur != NULL) {
2051 		if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
2052 			PF_SET_SKIP_STEPS(PF_SKIP_IFP);
2053 		if (cur->direction != prev->direction)
2054 			PF_SET_SKIP_STEPS(PF_SKIP_DIR);
2055 		if (cur->af != prev->af)
2056 			PF_SET_SKIP_STEPS(PF_SKIP_AF);
2057 		if (cur->proto != prev->proto)
2058 			PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
2059 		if (cur->src.neg != prev->src.neg ||
2060 		    pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
2061 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
2062 		if (cur->src.port[0] != prev->src.port[0] ||
2063 		    cur->src.port[1] != prev->src.port[1] ||
2064 		    cur->src.port_op != prev->src.port_op)
2065 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
2066 		if (cur->dst.neg != prev->dst.neg ||
2067 		    pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
2068 			PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
2069 		if (cur->dst.port[0] != prev->dst.port[0] ||
2070 		    cur->dst.port[1] != prev->dst.port[1] ||
2071 		    cur->dst.port_op != prev->dst.port_op)
2072 			PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
2073 
2074 		prev = cur;
2075 		cur = TAILQ_NEXT(cur, entries);
2076 	}
2077 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2078 		PF_SET_SKIP_STEPS(i);
2079 }
2080 
2081 static int
2082 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
2083 {
2084 	if (aw1->type != aw2->type)
2085 		return (1);
2086 	switch (aw1->type) {
2087 	case PF_ADDR_ADDRMASK:
2088 	case PF_ADDR_RANGE:
2089 		if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
2090 			return (1);
2091 		if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
2092 			return (1);
2093 		return (0);
2094 	case PF_ADDR_DYNIFTL:
2095 		return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
2096 	case PF_ADDR_NOROUTE:
2097 	case PF_ADDR_URPFFAILED:
2098 		return (0);
2099 	case PF_ADDR_TABLE:
2100 		return (aw1->p.tbl != aw2->p.tbl);
2101 	default:
2102 		printf("invalid address type: %d\n", aw1->type);
2103 		return (1);
2104 	}
2105 }
2106 
2107 /**
2108  * Checksum updates are a little complicated because the checksum in the TCP/UDP
2109  * header isn't always a full checksum. In some cases (i.e. output) it's a
2110  * pseudo-header checksum, which is a partial checksum over src/dst IP
2111  * addresses, protocol number and length.
2112  *
2113  * That means we have the following cases:
2114  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2115  *  	checksums, we need to update the checksum whenever we change anything.
2116  *  * Output (i.e. the checksum is a pseudo-header checksum):
2117  *  	x The field being updated is src/dst address or affects the length of
2118  *  	the packet. We need to update the pseudo-header checksum (note that this
2119  *  	checksum is not ones' complement).
2120  *  	x Some other field is being modified (e.g. src/dst port numbers): We
2121  *  	don't have to update anything.
2122  **/
2123 u_int16_t
2124 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2125 {
2126 	u_int32_t x;
2127 
2128 	x = cksum + old - new;
2129 	x = (x + (x >> 16)) & 0xffff;
2130 
2131 	/* optimise: eliminate a branch when not udp */
2132 	if (udp && cksum == 0x0000)
2133 		return cksum;
2134 	if (udp && x == 0x0000)
2135 		x = 0xffff;
2136 
2137 	return (u_int16_t)(x);
2138 }
2139 
2140 static void
2141 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi,
2142     u_int8_t udp)
2143 {
2144 	u_int16_t old = htons(hi ? (*f << 8) : *f);
2145 	u_int16_t new = htons(hi ? ( v << 8) :  v);
2146 
2147 	if (*f == v)
2148 		return;
2149 
2150 	*f = v;
2151 
2152 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2153 		return;
2154 
2155 	*cksum = pf_cksum_fixup(*cksum, old, new, udp);
2156 }
2157 
2158 void
2159 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v,
2160     bool hi, u_int8_t udp)
2161 {
2162 	u_int8_t *fb = (u_int8_t *)f;
2163 	u_int8_t *vb = (u_int8_t *)&v;
2164 
2165 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2166 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2167 }
2168 
2169 void
2170 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v,
2171     bool hi, u_int8_t udp)
2172 {
2173 	u_int8_t *fb = (u_int8_t *)f;
2174 	u_int8_t *vb = (u_int8_t *)&v;
2175 
2176 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2177 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2178 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2179 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2180 }
2181 
2182 u_int16_t
2183 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2184         u_int16_t new, u_int8_t udp)
2185 {
2186 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2187 		return (cksum);
2188 
2189 	return (pf_cksum_fixup(cksum, old, new, udp));
2190 }
2191 
2192 static void
2193 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2194         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2195         sa_family_t af)
2196 {
2197 	struct pf_addr	ao;
2198 	u_int16_t	po = *p;
2199 
2200 	PF_ACPY(&ao, a, af);
2201 	PF_ACPY(a, an, af);
2202 
2203 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2204 		*pc = ~*pc;
2205 
2206 	*p = pn;
2207 
2208 	switch (af) {
2209 #ifdef INET
2210 	case AF_INET:
2211 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2212 		    ao.addr16[0], an->addr16[0], 0),
2213 		    ao.addr16[1], an->addr16[1], 0);
2214 		*p = pn;
2215 
2216 		*pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2217 		    ao.addr16[0], an->addr16[0], u),
2218 		    ao.addr16[1], an->addr16[1], u);
2219 
2220 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2221 		break;
2222 #endif /* INET */
2223 #ifdef INET6
2224 	case AF_INET6:
2225 		*pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2226 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2227 		    pf_cksum_fixup(pf_cksum_fixup(*pc,
2228 		    ao.addr16[0], an->addr16[0], u),
2229 		    ao.addr16[1], an->addr16[1], u),
2230 		    ao.addr16[2], an->addr16[2], u),
2231 		    ao.addr16[3], an->addr16[3], u),
2232 		    ao.addr16[4], an->addr16[4], u),
2233 		    ao.addr16[5], an->addr16[5], u),
2234 		    ao.addr16[6], an->addr16[6], u),
2235 		    ao.addr16[7], an->addr16[7], u);
2236 
2237 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2238 		break;
2239 #endif /* INET6 */
2240 	}
2241 
2242 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2243 	    CSUM_DELAY_DATA_IPV6)) {
2244 		*pc = ~*pc;
2245 		if (! *pc)
2246 			*pc = 0xffff;
2247 	}
2248 }
2249 
2250 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2251 void
2252 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2253 {
2254 	u_int32_t	ao;
2255 
2256 	memcpy(&ao, a, sizeof(ao));
2257 	memcpy(a, &an, sizeof(u_int32_t));
2258 	*c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2259 	    ao % 65536, an % 65536, u);
2260 }
2261 
2262 void
2263 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2264 {
2265 	u_int32_t	ao;
2266 
2267 	memcpy(&ao, a, sizeof(ao));
2268 	memcpy(a, &an, sizeof(u_int32_t));
2269 
2270 	*c = pf_proto_cksum_fixup(m,
2271 	    pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2272 	    ao % 65536, an % 65536, udp);
2273 }
2274 
2275 #ifdef INET6
2276 static void
2277 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2278 {
2279 	struct pf_addr	ao;
2280 
2281 	PF_ACPY(&ao, a, AF_INET6);
2282 	PF_ACPY(a, an, AF_INET6);
2283 
2284 	*c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2285 	    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2286 	    pf_cksum_fixup(pf_cksum_fixup(*c,
2287 	    ao.addr16[0], an->addr16[0], u),
2288 	    ao.addr16[1], an->addr16[1], u),
2289 	    ao.addr16[2], an->addr16[2], u),
2290 	    ao.addr16[3], an->addr16[3], u),
2291 	    ao.addr16[4], an->addr16[4], u),
2292 	    ao.addr16[5], an->addr16[5], u),
2293 	    ao.addr16[6], an->addr16[6], u),
2294 	    ao.addr16[7], an->addr16[7], u);
2295 }
2296 #endif /* INET6 */
2297 
2298 static void
2299 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2300     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2301     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2302 {
2303 	struct pf_addr	oia, ooa;
2304 
2305 	PF_ACPY(&oia, ia, af);
2306 	if (oa)
2307 		PF_ACPY(&ooa, oa, af);
2308 
2309 	/* Change inner protocol port, fix inner protocol checksum. */
2310 	if (ip != NULL) {
2311 		u_int16_t	oip = *ip;
2312 		u_int32_t	opc;
2313 
2314 		if (pc != NULL)
2315 			opc = *pc;
2316 		*ip = np;
2317 		if (pc != NULL)
2318 			*pc = pf_cksum_fixup(*pc, oip, *ip, u);
2319 		*ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2320 		if (pc != NULL)
2321 			*ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2322 	}
2323 	/* Change inner ip address, fix inner ip and icmp checksums. */
2324 	PF_ACPY(ia, na, af);
2325 	switch (af) {
2326 #ifdef INET
2327 	case AF_INET: {
2328 		u_int32_t	 oh2c = *h2c;
2329 
2330 		*h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2331 		    oia.addr16[0], ia->addr16[0], 0),
2332 		    oia.addr16[1], ia->addr16[1], 0);
2333 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2334 		    oia.addr16[0], ia->addr16[0], 0),
2335 		    oia.addr16[1], ia->addr16[1], 0);
2336 		*ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2337 		break;
2338 	}
2339 #endif /* INET */
2340 #ifdef INET6
2341 	case AF_INET6:
2342 		*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2343 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2344 		    pf_cksum_fixup(pf_cksum_fixup(*ic,
2345 		    oia.addr16[0], ia->addr16[0], u),
2346 		    oia.addr16[1], ia->addr16[1], u),
2347 		    oia.addr16[2], ia->addr16[2], u),
2348 		    oia.addr16[3], ia->addr16[3], u),
2349 		    oia.addr16[4], ia->addr16[4], u),
2350 		    oia.addr16[5], ia->addr16[5], u),
2351 		    oia.addr16[6], ia->addr16[6], u),
2352 		    oia.addr16[7], ia->addr16[7], u);
2353 		break;
2354 #endif /* INET6 */
2355 	}
2356 	/* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2357 	if (oa) {
2358 		PF_ACPY(oa, na, af);
2359 		switch (af) {
2360 #ifdef INET
2361 		case AF_INET:
2362 			*hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2363 			    ooa.addr16[0], oa->addr16[0], 0),
2364 			    ooa.addr16[1], oa->addr16[1], 0);
2365 			break;
2366 #endif /* INET */
2367 #ifdef INET6
2368 		case AF_INET6:
2369 			*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2370 			    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2371 			    pf_cksum_fixup(pf_cksum_fixup(*ic,
2372 			    ooa.addr16[0], oa->addr16[0], u),
2373 			    ooa.addr16[1], oa->addr16[1], u),
2374 			    ooa.addr16[2], oa->addr16[2], u),
2375 			    ooa.addr16[3], oa->addr16[3], u),
2376 			    ooa.addr16[4], oa->addr16[4], u),
2377 			    ooa.addr16[5], oa->addr16[5], u),
2378 			    ooa.addr16[6], oa->addr16[6], u),
2379 			    ooa.addr16[7], oa->addr16[7], u);
2380 			break;
2381 #endif /* INET6 */
2382 		}
2383 	}
2384 }
2385 
2386 /*
2387  * Need to modulate the sequence numbers in the TCP SACK option
2388  * (credits to Krzysztof Pfaff for report and patch)
2389  */
2390 static int
2391 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2392     struct tcphdr *th, struct pf_state_peer *dst)
2393 {
2394 	int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2395 	u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2396 	int copyback = 0, i, olen;
2397 	struct sackblk sack;
2398 
2399 #define	TCPOLEN_SACKLEN	(TCPOLEN_SACK + 2)
2400 	if (hlen < TCPOLEN_SACKLEN ||
2401 	    !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2402 		return 0;
2403 
2404 	while (hlen >= TCPOLEN_SACKLEN) {
2405 		size_t startoff = opt - opts;
2406 		olen = opt[1];
2407 		switch (*opt) {
2408 		case TCPOPT_EOL:	/* FALLTHROUGH */
2409 		case TCPOPT_NOP:
2410 			opt++;
2411 			hlen--;
2412 			break;
2413 		case TCPOPT_SACK:
2414 			if (olen > hlen)
2415 				olen = hlen;
2416 			if (olen >= TCPOLEN_SACKLEN) {
2417 				for (i = 2; i + TCPOLEN_SACK <= olen;
2418 				    i += TCPOLEN_SACK) {
2419 					memcpy(&sack, &opt[i], sizeof(sack));
2420 					pf_patch_32_unaligned(m,
2421 					    &th->th_sum, &sack.start,
2422 					    htonl(ntohl(sack.start) - dst->seqdiff),
2423 					    PF_ALGNMNT(startoff),
2424 					    0);
2425 					pf_patch_32_unaligned(m, &th->th_sum,
2426 					    &sack.end,
2427 					    htonl(ntohl(sack.end) - dst->seqdiff),
2428 					    PF_ALGNMNT(startoff),
2429 					    0);
2430 					memcpy(&opt[i], &sack, sizeof(sack));
2431 				}
2432 				copyback = 1;
2433 			}
2434 			/* FALLTHROUGH */
2435 		default:
2436 			if (olen < 2)
2437 				olen = 2;
2438 			hlen -= olen;
2439 			opt += olen;
2440 		}
2441 	}
2442 
2443 	if (copyback)
2444 		m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2445 	return (copyback);
2446 }
2447 
2448 static void
2449 pf_send_tcp(struct mbuf *replyto, const struct pf_krule *r, sa_family_t af,
2450     const struct pf_addr *saddr, const struct pf_addr *daddr,
2451     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2452     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2453     u_int16_t rtag, struct ifnet *ifp)
2454 {
2455 	struct pf_send_entry *pfse;
2456 	struct mbuf	*m;
2457 	int		 len, tlen;
2458 #ifdef INET
2459 	struct ip	*h = NULL;
2460 #endif /* INET */
2461 #ifdef INET6
2462 	struct ip6_hdr	*h6 = NULL;
2463 #endif /* INET6 */
2464 	struct tcphdr	*th;
2465 	char		*opt;
2466 	struct pf_mtag  *pf_mtag;
2467 
2468 	len = 0;
2469 	th = NULL;
2470 
2471 	/* maximum segment size tcp option */
2472 	tlen = sizeof(struct tcphdr);
2473 	if (mss)
2474 		tlen += 4;
2475 
2476 	switch (af) {
2477 #ifdef INET
2478 	case AF_INET:
2479 		len = sizeof(struct ip) + tlen;
2480 		break;
2481 #endif /* INET */
2482 #ifdef INET6
2483 	case AF_INET6:
2484 		len = sizeof(struct ip6_hdr) + tlen;
2485 		break;
2486 #endif /* INET6 */
2487 	default:
2488 		panic("%s: unsupported af %d", __func__, af);
2489 	}
2490 
2491 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
2492 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2493 	if (pfse == NULL)
2494 		return;
2495 	m = m_gethdr(M_NOWAIT, MT_DATA);
2496 	if (m == NULL) {
2497 		free(pfse, M_PFTEMP);
2498 		return;
2499 	}
2500 #ifdef MAC
2501 	mac_netinet_firewall_send(m);
2502 #endif
2503 	if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2504 		free(pfse, M_PFTEMP);
2505 		m_freem(m);
2506 		return;
2507 	}
2508 	if (tag)
2509 		m->m_flags |= M_SKIP_FIREWALL;
2510 	pf_mtag->tag = rtag;
2511 
2512 	if (r != NULL && r->rtableid >= 0)
2513 		M_SETFIB(m, r->rtableid);
2514 
2515 #ifdef ALTQ
2516 	if (r != NULL && r->qid) {
2517 		pf_mtag->qid = r->qid;
2518 
2519 		/* add hints for ecn */
2520 		pf_mtag->hdr = mtod(m, struct ip *);
2521 	}
2522 #endif /* ALTQ */
2523 	m->m_data += max_linkhdr;
2524 	m->m_pkthdr.len = m->m_len = len;
2525 	m->m_pkthdr.rcvif = NULL;
2526 	bzero(m->m_data, len);
2527 	switch (af) {
2528 #ifdef INET
2529 	case AF_INET:
2530 		h = mtod(m, struct ip *);
2531 
2532 		/* IP header fields included in the TCP checksum */
2533 		h->ip_p = IPPROTO_TCP;
2534 		h->ip_len = htons(tlen);
2535 		h->ip_src.s_addr = saddr->v4.s_addr;
2536 		h->ip_dst.s_addr = daddr->v4.s_addr;
2537 
2538 		th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2539 		break;
2540 #endif /* INET */
2541 #ifdef INET6
2542 	case AF_INET6:
2543 		h6 = mtod(m, struct ip6_hdr *);
2544 
2545 		/* IP header fields included in the TCP checksum */
2546 		h6->ip6_nxt = IPPROTO_TCP;
2547 		h6->ip6_plen = htons(tlen);
2548 		memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2549 		memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2550 
2551 		th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2552 		break;
2553 #endif /* INET6 */
2554 	}
2555 
2556 	/* TCP header */
2557 	th->th_sport = sport;
2558 	th->th_dport = dport;
2559 	th->th_seq = htonl(seq);
2560 	th->th_ack = htonl(ack);
2561 	th->th_off = tlen >> 2;
2562 	th->th_flags = flags;
2563 	th->th_win = htons(win);
2564 
2565 	if (mss) {
2566 		opt = (char *)(th + 1);
2567 		opt[0] = TCPOPT_MAXSEG;
2568 		opt[1] = 4;
2569 		HTONS(mss);
2570 		bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2571 	}
2572 
2573 	switch (af) {
2574 #ifdef INET
2575 	case AF_INET:
2576 		/* TCP checksum */
2577 		th->th_sum = in_cksum(m, len);
2578 
2579 		/* Finish the IP header */
2580 		h->ip_v = 4;
2581 		h->ip_hl = sizeof(*h) >> 2;
2582 		h->ip_tos = IPTOS_LOWDELAY;
2583 		h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2584 		h->ip_len = htons(len);
2585 		h->ip_ttl = ttl ? ttl : V_ip_defttl;
2586 		h->ip_sum = 0;
2587 
2588 		pfse->pfse_type = PFSE_IP;
2589 		break;
2590 #endif /* INET */
2591 #ifdef INET6
2592 	case AF_INET6:
2593 		/* TCP checksum */
2594 		th->th_sum = in6_cksum(m, IPPROTO_TCP,
2595 		    sizeof(struct ip6_hdr), tlen);
2596 
2597 		h6->ip6_vfc |= IPV6_VERSION;
2598 		h6->ip6_hlim = IPV6_DEFHLIM;
2599 
2600 		pfse->pfse_type = PFSE_IP6;
2601 		break;
2602 #endif /* INET6 */
2603 	}
2604 	pfse->pfse_m = m;
2605 	pf_send(pfse);
2606 }
2607 
2608 static void
2609 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd,
2610     struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th,
2611     struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
2612     u_short *reason)
2613 {
2614 	struct pf_addr	* const saddr = pd->src;
2615 	struct pf_addr	* const daddr = pd->dst;
2616 	sa_family_t	 af = pd->af;
2617 
2618 	/* undo NAT changes, if they have taken place */
2619 	if (nr != NULL) {
2620 		PF_ACPY(saddr, &sk->addr[pd->sidx], af);
2621 		PF_ACPY(daddr, &sk->addr[pd->didx], af);
2622 		if (pd->sport)
2623 			*pd->sport = sk->port[pd->sidx];
2624 		if (pd->dport)
2625 			*pd->dport = sk->port[pd->didx];
2626 		if (pd->proto_sum)
2627 			*pd->proto_sum = bproto_sum;
2628 		if (pd->ip_sum)
2629 			*pd->ip_sum = bip_sum;
2630 		m_copyback(m, off, hdrlen, pd->hdr.any);
2631 	}
2632 	if (pd->proto == IPPROTO_TCP &&
2633 	    ((r->rule_flag & PFRULE_RETURNRST) ||
2634 	    (r->rule_flag & PFRULE_RETURN)) &&
2635 	    !(th->th_flags & TH_RST)) {
2636 		u_int32_t	 ack = ntohl(th->th_seq) + pd->p_len;
2637 		int		 len = 0;
2638 #ifdef INET
2639 		struct ip	*h4;
2640 #endif
2641 #ifdef INET6
2642 		struct ip6_hdr	*h6;
2643 #endif
2644 
2645 		switch (af) {
2646 #ifdef INET
2647 		case AF_INET:
2648 			h4 = mtod(m, struct ip *);
2649 			len = ntohs(h4->ip_len) - off;
2650 			break;
2651 #endif
2652 #ifdef INET6
2653 		case AF_INET6:
2654 			h6 = mtod(m, struct ip6_hdr *);
2655 			len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
2656 			break;
2657 #endif
2658 		}
2659 
2660 		if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
2661 			REASON_SET(reason, PFRES_PROTCKSUM);
2662 		else {
2663 			if (th->th_flags & TH_SYN)
2664 				ack++;
2665 			if (th->th_flags & TH_FIN)
2666 				ack++;
2667 			pf_send_tcp(m, r, af, pd->dst,
2668 				pd->src, th->th_dport, th->th_sport,
2669 				ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
2670 				r->return_ttl, 1, 0, kif->pfik_ifp);
2671 		}
2672 	} else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
2673 		r->return_icmp)
2674 		pf_send_icmp(m, r->return_icmp >> 8,
2675 			r->return_icmp & 255, af, r);
2676 	else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
2677 		r->return_icmp6)
2678 		pf_send_icmp(m, r->return_icmp6 >> 8,
2679 			r->return_icmp6 & 255, af, r);
2680 }
2681 
2682 static int
2683 pf_ieee8021q_setpcp(struct mbuf *m, u_int8_t prio)
2684 {
2685 	struct m_tag *mtag;
2686 
2687 	KASSERT(prio <= PF_PRIO_MAX,
2688 	    ("%s with invalid pcp", __func__));
2689 
2690 	mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_OUT, NULL);
2691 	if (mtag == NULL) {
2692 		mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_OUT,
2693 		    sizeof(uint8_t), M_NOWAIT);
2694 		if (mtag == NULL)
2695 			return (ENOMEM);
2696 		m_tag_prepend(m, mtag);
2697 	}
2698 
2699 	*(uint8_t *)(mtag + 1) = prio;
2700 	return (0);
2701 }
2702 
2703 static int
2704 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
2705 {
2706 	struct m_tag *mtag;
2707 	u_int8_t mpcp;
2708 
2709 	mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
2710 	if (mtag == NULL)
2711 		return (0);
2712 
2713 	if (prio == PF_PRIO_ZERO)
2714 		prio = 0;
2715 
2716 	mpcp = *(uint8_t *)(mtag + 1);
2717 
2718 	return (mpcp == prio);
2719 }
2720 
2721 static void
2722 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2723     struct pf_krule *r)
2724 {
2725 	struct pf_send_entry *pfse;
2726 	struct mbuf *m0;
2727 	struct pf_mtag *pf_mtag;
2728 
2729 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
2730 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2731 	if (pfse == NULL)
2732 		return;
2733 
2734 	if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2735 		free(pfse, M_PFTEMP);
2736 		return;
2737 	}
2738 
2739 	if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2740 		free(pfse, M_PFTEMP);
2741 		return;
2742 	}
2743 	/* XXX: revisit */
2744 	m0->m_flags |= M_SKIP_FIREWALL;
2745 
2746 	if (r->rtableid >= 0)
2747 		M_SETFIB(m0, r->rtableid);
2748 
2749 #ifdef ALTQ
2750 	if (r->qid) {
2751 		pf_mtag->qid = r->qid;
2752 		/* add hints for ecn */
2753 		pf_mtag->hdr = mtod(m0, struct ip *);
2754 	}
2755 #endif /* ALTQ */
2756 
2757 	switch (af) {
2758 #ifdef INET
2759 	case AF_INET:
2760 		pfse->pfse_type = PFSE_ICMP;
2761 		break;
2762 #endif /* INET */
2763 #ifdef INET6
2764 	case AF_INET6:
2765 		pfse->pfse_type = PFSE_ICMP6;
2766 		break;
2767 #endif /* INET6 */
2768 	}
2769 	pfse->pfse_m = m0;
2770 	pfse->icmpopts.type = type;
2771 	pfse->icmpopts.code = code;
2772 	pf_send(pfse);
2773 }
2774 
2775 /*
2776  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2777  * If n is 0, they match if they are equal. If n is != 0, they match if they
2778  * are different.
2779  */
2780 int
2781 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2782     struct pf_addr *b, sa_family_t af)
2783 {
2784 	int	match = 0;
2785 
2786 	switch (af) {
2787 #ifdef INET
2788 	case AF_INET:
2789 		if ((a->addr32[0] & m->addr32[0]) ==
2790 		    (b->addr32[0] & m->addr32[0]))
2791 			match++;
2792 		break;
2793 #endif /* INET */
2794 #ifdef INET6
2795 	case AF_INET6:
2796 		if (((a->addr32[0] & m->addr32[0]) ==
2797 		     (b->addr32[0] & m->addr32[0])) &&
2798 		    ((a->addr32[1] & m->addr32[1]) ==
2799 		     (b->addr32[1] & m->addr32[1])) &&
2800 		    ((a->addr32[2] & m->addr32[2]) ==
2801 		     (b->addr32[2] & m->addr32[2])) &&
2802 		    ((a->addr32[3] & m->addr32[3]) ==
2803 		     (b->addr32[3] & m->addr32[3])))
2804 			match++;
2805 		break;
2806 #endif /* INET6 */
2807 	}
2808 	if (match) {
2809 		if (n)
2810 			return (0);
2811 		else
2812 			return (1);
2813 	} else {
2814 		if (n)
2815 			return (1);
2816 		else
2817 			return (0);
2818 	}
2819 }
2820 
2821 /*
2822  * Return 1 if b <= a <= e, otherwise return 0.
2823  */
2824 int
2825 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2826     struct pf_addr *a, sa_family_t af)
2827 {
2828 	switch (af) {
2829 #ifdef INET
2830 	case AF_INET:
2831 		if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
2832 		    (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
2833 			return (0);
2834 		break;
2835 #endif /* INET */
2836 #ifdef INET6
2837 	case AF_INET6: {
2838 		int	i;
2839 
2840 		/* check a >= b */
2841 		for (i = 0; i < 4; ++i)
2842 			if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
2843 				break;
2844 			else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
2845 				return (0);
2846 		/* check a <= e */
2847 		for (i = 0; i < 4; ++i)
2848 			if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
2849 				break;
2850 			else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
2851 				return (0);
2852 		break;
2853 	}
2854 #endif /* INET6 */
2855 	}
2856 	return (1);
2857 }
2858 
2859 static int
2860 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2861 {
2862 	switch (op) {
2863 	case PF_OP_IRG:
2864 		return ((p > a1) && (p < a2));
2865 	case PF_OP_XRG:
2866 		return ((p < a1) || (p > a2));
2867 	case PF_OP_RRG:
2868 		return ((p >= a1) && (p <= a2));
2869 	case PF_OP_EQ:
2870 		return (p == a1);
2871 	case PF_OP_NE:
2872 		return (p != a1);
2873 	case PF_OP_LT:
2874 		return (p < a1);
2875 	case PF_OP_LE:
2876 		return (p <= a1);
2877 	case PF_OP_GT:
2878 		return (p > a1);
2879 	case PF_OP_GE:
2880 		return (p >= a1);
2881 	}
2882 	return (0); /* never reached */
2883 }
2884 
2885 int
2886 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2887 {
2888 	NTOHS(a1);
2889 	NTOHS(a2);
2890 	NTOHS(p);
2891 	return (pf_match(op, a1, a2, p));
2892 }
2893 
2894 static int
2895 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2896 {
2897 	if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2898 		return (0);
2899 	return (pf_match(op, a1, a2, u));
2900 }
2901 
2902 static int
2903 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2904 {
2905 	if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2906 		return (0);
2907 	return (pf_match(op, a1, a2, g));
2908 }
2909 
2910 int
2911 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag)
2912 {
2913 	if (*tag == -1)
2914 		*tag = mtag;
2915 
2916 	return ((!r->match_tag_not && r->match_tag == *tag) ||
2917 	    (r->match_tag_not && r->match_tag != *tag));
2918 }
2919 
2920 int
2921 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2922 {
2923 
2924 	KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2925 
2926 	if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2927 		return (ENOMEM);
2928 
2929 	pd->pf_mtag->tag = tag;
2930 
2931 	return (0);
2932 }
2933 
2934 #define	PF_ANCHOR_STACKSIZE	32
2935 struct pf_kanchor_stackframe {
2936 	struct pf_kruleset	*rs;
2937 	struct pf_krule		*r;	/* XXX: + match bit */
2938 	struct pf_kanchor	*child;
2939 };
2940 
2941 /*
2942  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2943  */
2944 #define	PF_ANCHORSTACK_MATCH	0x00000001
2945 #define	PF_ANCHORSTACK_MASK	(PF_ANCHORSTACK_MATCH)
2946 
2947 #define	PF_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2948 #define	PF_ANCHOR_RULE(f)	(struct pf_krule *)			\
2949 				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2950 #define	PF_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 			\
2951 				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2952 } while (0)
2953 
2954 void
2955 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth,
2956     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
2957     int *match)
2958 {
2959 	struct pf_kanchor_stackframe	*f;
2960 
2961 	PF_RULES_RASSERT();
2962 
2963 	if (match)
2964 		*match = 0;
2965 	if (*depth >= PF_ANCHOR_STACKSIZE) {
2966 		printf("%s: anchor stack overflow on %s\n",
2967 		    __func__, (*r)->anchor->name);
2968 		*r = TAILQ_NEXT(*r, entries);
2969 		return;
2970 	} else if (*depth == 0 && a != NULL)
2971 		*a = *r;
2972 	f = stack + (*depth)++;
2973 	f->rs = *rs;
2974 	f->r = *r;
2975 	if ((*r)->anchor_wildcard) {
2976 		struct pf_kanchor_node *parent = &(*r)->anchor->children;
2977 
2978 		if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) {
2979 			*r = NULL;
2980 			return;
2981 		}
2982 		*rs = &f->child->ruleset;
2983 	} else {
2984 		f->child = NULL;
2985 		*rs = &(*r)->anchor->ruleset;
2986 	}
2987 	*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2988 }
2989 
2990 int
2991 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth,
2992     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
2993     int *match)
2994 {
2995 	struct pf_kanchor_stackframe	*f;
2996 	struct pf_krule *fr;
2997 	int quick = 0;
2998 
2999 	PF_RULES_RASSERT();
3000 
3001 	do {
3002 		if (*depth <= 0)
3003 			break;
3004 		f = stack + *depth - 1;
3005 		fr = PF_ANCHOR_RULE(f);
3006 		if (f->child != NULL) {
3007 			struct pf_kanchor_node *parent;
3008 
3009 			/*
3010 			 * This block traverses through
3011 			 * a wildcard anchor.
3012 			 */
3013 			parent = &fr->anchor->children;
3014 			if (match != NULL && *match) {
3015 				/*
3016 				 * If any of "*" matched, then
3017 				 * "foo/ *" matched, mark frame
3018 				 * appropriately.
3019 				 */
3020 				PF_ANCHOR_SET_MATCH(f);
3021 				*match = 0;
3022 			}
3023 			f->child = RB_NEXT(pf_kanchor_node, parent, f->child);
3024 			if (f->child != NULL) {
3025 				*rs = &f->child->ruleset;
3026 				*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
3027 				if (*r == NULL)
3028 					continue;
3029 				else
3030 					break;
3031 			}
3032 		}
3033 		(*depth)--;
3034 		if (*depth == 0 && a != NULL)
3035 			*a = NULL;
3036 		*rs = f->rs;
3037 		if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
3038 			quick = fr->quick;
3039 		*r = TAILQ_NEXT(fr, entries);
3040 	} while (*r == NULL);
3041 
3042 	return (quick);
3043 }
3044 
3045 #ifdef INET6
3046 void
3047 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
3048     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
3049 {
3050 	switch (af) {
3051 #ifdef INET
3052 	case AF_INET:
3053 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
3054 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
3055 		break;
3056 #endif /* INET */
3057 	case AF_INET6:
3058 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
3059 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
3060 		naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
3061 		((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
3062 		naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
3063 		((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
3064 		naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
3065 		((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
3066 		break;
3067 	}
3068 }
3069 
3070 void
3071 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
3072 {
3073 	switch (af) {
3074 #ifdef INET
3075 	case AF_INET:
3076 		addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
3077 		break;
3078 #endif /* INET */
3079 	case AF_INET6:
3080 		if (addr->addr32[3] == 0xffffffff) {
3081 			addr->addr32[3] = 0;
3082 			if (addr->addr32[2] == 0xffffffff) {
3083 				addr->addr32[2] = 0;
3084 				if (addr->addr32[1] == 0xffffffff) {
3085 					addr->addr32[1] = 0;
3086 					addr->addr32[0] =
3087 					    htonl(ntohl(addr->addr32[0]) + 1);
3088 				} else
3089 					addr->addr32[1] =
3090 					    htonl(ntohl(addr->addr32[1]) + 1);
3091 			} else
3092 				addr->addr32[2] =
3093 				    htonl(ntohl(addr->addr32[2]) + 1);
3094 		} else
3095 			addr->addr32[3] =
3096 			    htonl(ntohl(addr->addr32[3]) + 1);
3097 		break;
3098 	}
3099 }
3100 #endif /* INET6 */
3101 
3102 int
3103 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
3104 {
3105 	struct pf_addr		*saddr, *daddr;
3106 	u_int16_t		 sport, dport;
3107 	struct inpcbinfo	*pi;
3108 	struct inpcb		*inp;
3109 
3110 	pd->lookup.uid = UID_MAX;
3111 	pd->lookup.gid = GID_MAX;
3112 
3113 	switch (pd->proto) {
3114 	case IPPROTO_TCP:
3115 		sport = pd->hdr.tcp.th_sport;
3116 		dport = pd->hdr.tcp.th_dport;
3117 		pi = &V_tcbinfo;
3118 		break;
3119 	case IPPROTO_UDP:
3120 		sport = pd->hdr.udp.uh_sport;
3121 		dport = pd->hdr.udp.uh_dport;
3122 		pi = &V_udbinfo;
3123 		break;
3124 	default:
3125 		return (-1);
3126 	}
3127 	if (direction == PF_IN) {
3128 		saddr = pd->src;
3129 		daddr = pd->dst;
3130 	} else {
3131 		u_int16_t	p;
3132 
3133 		p = sport;
3134 		sport = dport;
3135 		dport = p;
3136 		saddr = pd->dst;
3137 		daddr = pd->src;
3138 	}
3139 	switch (pd->af) {
3140 #ifdef INET
3141 	case AF_INET:
3142 		inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
3143 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
3144 		if (inp == NULL) {
3145 			inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
3146 			   daddr->v4, dport, INPLOOKUP_WILDCARD |
3147 			   INPLOOKUP_RLOCKPCB, NULL, m);
3148 			if (inp == NULL)
3149 				return (-1);
3150 		}
3151 		break;
3152 #endif /* INET */
3153 #ifdef INET6
3154 	case AF_INET6:
3155 		inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
3156 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
3157 		if (inp == NULL) {
3158 			inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
3159 			    &daddr->v6, dport, INPLOOKUP_WILDCARD |
3160 			    INPLOOKUP_RLOCKPCB, NULL, m);
3161 			if (inp == NULL)
3162 				return (-1);
3163 		}
3164 		break;
3165 #endif /* INET6 */
3166 
3167 	default:
3168 		return (-1);
3169 	}
3170 	INP_RLOCK_ASSERT(inp);
3171 	pd->lookup.uid = inp->inp_cred->cr_uid;
3172 	pd->lookup.gid = inp->inp_cred->cr_groups[0];
3173 	INP_RUNLOCK(inp);
3174 
3175 	return (1);
3176 }
3177 
3178 static u_int8_t
3179 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3180 {
3181 	int		 hlen;
3182 	u_int8_t	 hdr[60];
3183 	u_int8_t	*opt, optlen;
3184 	u_int8_t	 wscale = 0;
3185 
3186 	hlen = th_off << 2;		/* hlen <= sizeof(hdr) */
3187 	if (hlen <= sizeof(struct tcphdr))
3188 		return (0);
3189 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3190 		return (0);
3191 	opt = hdr + sizeof(struct tcphdr);
3192 	hlen -= sizeof(struct tcphdr);
3193 	while (hlen >= 3) {
3194 		switch (*opt) {
3195 		case TCPOPT_EOL:
3196 		case TCPOPT_NOP:
3197 			++opt;
3198 			--hlen;
3199 			break;
3200 		case TCPOPT_WINDOW:
3201 			wscale = opt[2];
3202 			if (wscale > TCP_MAX_WINSHIFT)
3203 				wscale = TCP_MAX_WINSHIFT;
3204 			wscale |= PF_WSCALE_FLAG;
3205 			/* FALLTHROUGH */
3206 		default:
3207 			optlen = opt[1];
3208 			if (optlen < 2)
3209 				optlen = 2;
3210 			hlen -= optlen;
3211 			opt += optlen;
3212 			break;
3213 		}
3214 	}
3215 	return (wscale);
3216 }
3217 
3218 static u_int16_t
3219 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3220 {
3221 	int		 hlen;
3222 	u_int8_t	 hdr[60];
3223 	u_int8_t	*opt, optlen;
3224 	u_int16_t	 mss = V_tcp_mssdflt;
3225 
3226 	hlen = th_off << 2;	/* hlen <= sizeof(hdr) */
3227 	if (hlen <= sizeof(struct tcphdr))
3228 		return (0);
3229 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3230 		return (0);
3231 	opt = hdr + sizeof(struct tcphdr);
3232 	hlen -= sizeof(struct tcphdr);
3233 	while (hlen >= TCPOLEN_MAXSEG) {
3234 		switch (*opt) {
3235 		case TCPOPT_EOL:
3236 		case TCPOPT_NOP:
3237 			++opt;
3238 			--hlen;
3239 			break;
3240 		case TCPOPT_MAXSEG:
3241 			bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
3242 			NTOHS(mss);
3243 			/* FALLTHROUGH */
3244 		default:
3245 			optlen = opt[1];
3246 			if (optlen < 2)
3247 				optlen = 2;
3248 			hlen -= optlen;
3249 			opt += optlen;
3250 			break;
3251 		}
3252 	}
3253 	return (mss);
3254 }
3255 
3256 static u_int16_t
3257 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
3258 {
3259 	struct nhop_object *nh;
3260 #ifdef INET6
3261 	struct in6_addr		dst6;
3262 	uint32_t		scopeid;
3263 #endif /* INET6 */
3264 	int			 hlen = 0;
3265 	uint16_t		 mss = 0;
3266 
3267 	NET_EPOCH_ASSERT();
3268 
3269 	switch (af) {
3270 #ifdef INET
3271 	case AF_INET:
3272 		hlen = sizeof(struct ip);
3273 		nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0);
3274 		if (nh != NULL)
3275 			mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
3276 		break;
3277 #endif /* INET */
3278 #ifdef INET6
3279 	case AF_INET6:
3280 		hlen = sizeof(struct ip6_hdr);
3281 		in6_splitscope(&addr->v6, &dst6, &scopeid);
3282 		nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0);
3283 		if (nh != NULL)
3284 			mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
3285 		break;
3286 #endif /* INET6 */
3287 	}
3288 
3289 	mss = max(V_tcp_mssdflt, mss);
3290 	mss = min(mss, offer);
3291 	mss = max(mss, 64);		/* sanity - at least max opt space */
3292 	return (mss);
3293 }
3294 
3295 static u_int32_t
3296 pf_tcp_iss(struct pf_pdesc *pd)
3297 {
3298 	MD5_CTX ctx;
3299 	u_int32_t digest[4];
3300 
3301 	if (V_pf_tcp_secret_init == 0) {
3302 		arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
3303 		MD5Init(&V_pf_tcp_secret_ctx);
3304 		MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
3305 		    sizeof(V_pf_tcp_secret));
3306 		V_pf_tcp_secret_init = 1;
3307 	}
3308 
3309 	ctx = V_pf_tcp_secret_ctx;
3310 
3311 	MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short));
3312 	MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short));
3313 	if (pd->af == AF_INET6) {
3314 		MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3315 		MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3316 	} else {
3317 		MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3318 		MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3319 	}
3320 	MD5Final((u_char *)digest, &ctx);
3321 	V_pf_tcp_iss_off += 4096;
3322 #define	ISN_RANDOM_INCREMENT (4096 - 1)
3323 	return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
3324 	    V_pf_tcp_iss_off);
3325 #undef	ISN_RANDOM_INCREMENT
3326 }
3327 
3328 static int
3329 pf_test_rule(struct pf_krule **rm, struct pf_state **sm, int direction,
3330     struct pfi_kkif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
3331     struct pf_krule **am, struct pf_kruleset **rsm, struct inpcb *inp)
3332 {
3333 	struct pf_krule		*nr = NULL;
3334 	struct pf_addr		* const saddr = pd->src;
3335 	struct pf_addr		* const daddr = pd->dst;
3336 	sa_family_t		 af = pd->af;
3337 	struct pf_krule		*r, *a = NULL;
3338 	struct pf_kruleset	*ruleset = NULL;
3339 	struct pf_ksrc_node	*nsn = NULL;
3340 	struct tcphdr		*th = &pd->hdr.tcp;
3341 	struct pf_state_key	*sk = NULL, *nk = NULL;
3342 	u_short			 reason;
3343 	int			 rewrite = 0, hdrlen = 0;
3344 	int			 tag = -1, rtableid = -1;
3345 	int			 asd = 0;
3346 	int			 match = 0;
3347 	int			 state_icmp = 0;
3348 	u_int16_t		 sport = 0, dport = 0;
3349 	u_int16_t		 bproto_sum = 0, bip_sum = 0;
3350 	u_int8_t		 icmptype = 0, icmpcode = 0;
3351 	struct pf_kanchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
3352 
3353 	PF_RULES_RASSERT();
3354 
3355 	if (inp != NULL) {
3356 		INP_LOCK_ASSERT(inp);
3357 		pd->lookup.uid = inp->inp_cred->cr_uid;
3358 		pd->lookup.gid = inp->inp_cred->cr_groups[0];
3359 		pd->lookup.done = 1;
3360 	}
3361 
3362 	switch (pd->proto) {
3363 	case IPPROTO_TCP:
3364 		sport = th->th_sport;
3365 		dport = th->th_dport;
3366 		hdrlen = sizeof(*th);
3367 		break;
3368 	case IPPROTO_UDP:
3369 		sport = pd->hdr.udp.uh_sport;
3370 		dport = pd->hdr.udp.uh_dport;
3371 		hdrlen = sizeof(pd->hdr.udp);
3372 		break;
3373 #ifdef INET
3374 	case IPPROTO_ICMP:
3375 		if (pd->af != AF_INET)
3376 			break;
3377 		sport = dport = pd->hdr.icmp.icmp_id;
3378 		hdrlen = sizeof(pd->hdr.icmp);
3379 		icmptype = pd->hdr.icmp.icmp_type;
3380 		icmpcode = pd->hdr.icmp.icmp_code;
3381 
3382 		if (icmptype == ICMP_UNREACH ||
3383 		    icmptype == ICMP_SOURCEQUENCH ||
3384 		    icmptype == ICMP_REDIRECT ||
3385 		    icmptype == ICMP_TIMXCEED ||
3386 		    icmptype == ICMP_PARAMPROB)
3387 			state_icmp++;
3388 		break;
3389 #endif /* INET */
3390 #ifdef INET6
3391 	case IPPROTO_ICMPV6:
3392 		if (af != AF_INET6)
3393 			break;
3394 		sport = dport = pd->hdr.icmp6.icmp6_id;
3395 		hdrlen = sizeof(pd->hdr.icmp6);
3396 		icmptype = pd->hdr.icmp6.icmp6_type;
3397 		icmpcode = pd->hdr.icmp6.icmp6_code;
3398 
3399 		if (icmptype == ICMP6_DST_UNREACH ||
3400 		    icmptype == ICMP6_PACKET_TOO_BIG ||
3401 		    icmptype == ICMP6_TIME_EXCEEDED ||
3402 		    icmptype == ICMP6_PARAM_PROB)
3403 			state_icmp++;
3404 		break;
3405 #endif /* INET6 */
3406 	default:
3407 		sport = dport = hdrlen = 0;
3408 		break;
3409 	}
3410 
3411 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3412 
3413 	/* check packet for BINAT/NAT/RDR */
3414 	if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3415 	    &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3416 		KASSERT(sk != NULL, ("%s: null sk", __func__));
3417 		KASSERT(nk != NULL, ("%s: null nk", __func__));
3418 
3419 		if (pd->ip_sum)
3420 			bip_sum = *pd->ip_sum;
3421 
3422 		switch (pd->proto) {
3423 		case IPPROTO_TCP:
3424 			bproto_sum = th->th_sum;
3425 			pd->proto_sum = &th->th_sum;
3426 
3427 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3428 			    nk->port[pd->sidx] != sport) {
3429 				pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
3430 				    &th->th_sum, &nk->addr[pd->sidx],
3431 				    nk->port[pd->sidx], 0, af);
3432 				pd->sport = &th->th_sport;
3433 				sport = th->th_sport;
3434 			}
3435 
3436 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3437 			    nk->port[pd->didx] != dport) {
3438 				pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
3439 				    &th->th_sum, &nk->addr[pd->didx],
3440 				    nk->port[pd->didx], 0, af);
3441 				dport = th->th_dport;
3442 				pd->dport = &th->th_dport;
3443 			}
3444 			rewrite++;
3445 			break;
3446 		case IPPROTO_UDP:
3447 			bproto_sum = pd->hdr.udp.uh_sum;
3448 			pd->proto_sum = &pd->hdr.udp.uh_sum;
3449 
3450 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3451 			    nk->port[pd->sidx] != sport) {
3452 				pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport,
3453 				    pd->ip_sum, &pd->hdr.udp.uh_sum,
3454 				    &nk->addr[pd->sidx],
3455 				    nk->port[pd->sidx], 1, af);
3456 				sport = pd->hdr.udp.uh_sport;
3457 				pd->sport = &pd->hdr.udp.uh_sport;
3458 			}
3459 
3460 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3461 			    nk->port[pd->didx] != dport) {
3462 				pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport,
3463 				    pd->ip_sum, &pd->hdr.udp.uh_sum,
3464 				    &nk->addr[pd->didx],
3465 				    nk->port[pd->didx], 1, af);
3466 				dport = pd->hdr.udp.uh_dport;
3467 				pd->dport = &pd->hdr.udp.uh_dport;
3468 			}
3469 			rewrite++;
3470 			break;
3471 #ifdef INET
3472 		case IPPROTO_ICMP:
3473 			nk->port[0] = nk->port[1];
3474 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3475 				pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3476 				    nk->addr[pd->sidx].v4.s_addr, 0);
3477 
3478 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3479 				pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3480 				    nk->addr[pd->didx].v4.s_addr, 0);
3481 
3482 			if (nk->port[1] != pd->hdr.icmp.icmp_id) {
3483 				pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
3484 				    pd->hdr.icmp.icmp_cksum, sport,
3485 				    nk->port[1], 0);
3486 				pd->hdr.icmp.icmp_id = nk->port[1];
3487 				pd->sport = &pd->hdr.icmp.icmp_id;
3488 			}
3489 			m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
3490 			break;
3491 #endif /* INET */
3492 #ifdef INET6
3493 		case IPPROTO_ICMPV6:
3494 			nk->port[0] = nk->port[1];
3495 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3496 				pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum,
3497 				    &nk->addr[pd->sidx], 0);
3498 
3499 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3500 				pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum,
3501 				    &nk->addr[pd->didx], 0);
3502 			rewrite++;
3503 			break;
3504 #endif /* INET */
3505 		default:
3506 			switch (af) {
3507 #ifdef INET
3508 			case AF_INET:
3509 				if (PF_ANEQ(saddr,
3510 				    &nk->addr[pd->sidx], AF_INET))
3511 					pf_change_a(&saddr->v4.s_addr,
3512 					    pd->ip_sum,
3513 					    nk->addr[pd->sidx].v4.s_addr, 0);
3514 
3515 				if (PF_ANEQ(daddr,
3516 				    &nk->addr[pd->didx], AF_INET))
3517 					pf_change_a(&daddr->v4.s_addr,
3518 					    pd->ip_sum,
3519 					    nk->addr[pd->didx].v4.s_addr, 0);
3520 				break;
3521 #endif /* INET */
3522 #ifdef INET6
3523 			case AF_INET6:
3524 				if (PF_ANEQ(saddr,
3525 				    &nk->addr[pd->sidx], AF_INET6))
3526 					PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3527 
3528 				if (PF_ANEQ(daddr,
3529 				    &nk->addr[pd->didx], AF_INET6))
3530 					PF_ACPY(daddr, &nk->addr[pd->didx], af);
3531 				break;
3532 #endif /* INET */
3533 			}
3534 			break;
3535 		}
3536 		if (nr->natpass)
3537 			r = NULL;
3538 		pd->nat_rule = nr;
3539 	}
3540 
3541 	while (r != NULL) {
3542 		counter_u64_add(r->evaluations, 1);
3543 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
3544 			r = r->skip[PF_SKIP_IFP].ptr;
3545 		else if (r->direction && r->direction != direction)
3546 			r = r->skip[PF_SKIP_DIR].ptr;
3547 		else if (r->af && r->af != af)
3548 			r = r->skip[PF_SKIP_AF].ptr;
3549 		else if (r->proto && r->proto != pd->proto)
3550 			r = r->skip[PF_SKIP_PROTO].ptr;
3551 		else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3552 		    r->src.neg, kif, M_GETFIB(m)))
3553 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3554 		/* tcp/udp only. port_op always 0 in other cases */
3555 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
3556 		    r->src.port[0], r->src.port[1], sport))
3557 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
3558 		else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3559 		    r->dst.neg, NULL, M_GETFIB(m)))
3560 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
3561 		/* tcp/udp only. port_op always 0 in other cases */
3562 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3563 		    r->dst.port[0], r->dst.port[1], dport))
3564 			r = r->skip[PF_SKIP_DST_PORT].ptr;
3565 		/* icmp only. type always 0 in other cases */
3566 		else if (r->type && r->type != icmptype + 1)
3567 			r = TAILQ_NEXT(r, entries);
3568 		/* icmp only. type always 0 in other cases */
3569 		else if (r->code && r->code != icmpcode + 1)
3570 			r = TAILQ_NEXT(r, entries);
3571 		else if (r->tos && !(r->tos == pd->tos))
3572 			r = TAILQ_NEXT(r, entries);
3573 		else if (r->rule_flag & PFRULE_FRAGMENT)
3574 			r = TAILQ_NEXT(r, entries);
3575 		else if (pd->proto == IPPROTO_TCP &&
3576 		    (r->flagset & th->th_flags) != r->flags)
3577 			r = TAILQ_NEXT(r, entries);
3578 		/* tcp/udp only. uid.op always 0 in other cases */
3579 		else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3580 		    pf_socket_lookup(direction, pd, m), 1)) &&
3581 		    !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3582 		    pd->lookup.uid))
3583 			r = TAILQ_NEXT(r, entries);
3584 		/* tcp/udp only. gid.op always 0 in other cases */
3585 		else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3586 		    pf_socket_lookup(direction, pd, m), 1)) &&
3587 		    !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3588 		    pd->lookup.gid))
3589 			r = TAILQ_NEXT(r, entries);
3590 		else if (r->prio &&
3591 		    !pf_match_ieee8021q_pcp(r->prio, m))
3592 			r = TAILQ_NEXT(r, entries);
3593 		else if (r->prob &&
3594 		    r->prob <= arc4random())
3595 			r = TAILQ_NEXT(r, entries);
3596 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
3597 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
3598 			r = TAILQ_NEXT(r, entries);
3599 		else if (r->os_fingerprint != PF_OSFP_ANY &&
3600 		    (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3601 		    pf_osfp_fingerprint(pd, m, off, th),
3602 		    r->os_fingerprint)))
3603 			r = TAILQ_NEXT(r, entries);
3604 		else {
3605 			if (r->tag)
3606 				tag = r->tag;
3607 			if (r->rtableid >= 0)
3608 				rtableid = r->rtableid;
3609 			if (r->anchor == NULL) {
3610 				match = 1;
3611 				*rm = r;
3612 				*am = a;
3613 				*rsm = ruleset;
3614 				if ((*rm)->quick)
3615 					break;
3616 				r = TAILQ_NEXT(r, entries);
3617 			} else
3618 				pf_step_into_anchor(anchor_stack, &asd,
3619 				    &ruleset, PF_RULESET_FILTER, &r, &a,
3620 				    &match);
3621 		}
3622 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3623 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3624 			break;
3625 	}
3626 	r = *rm;
3627 	a = *am;
3628 	ruleset = *rsm;
3629 
3630 	REASON_SET(&reason, PFRES_MATCH);
3631 
3632 	if (r->log || (nr != NULL && nr->log)) {
3633 		if (rewrite)
3634 			m_copyback(m, off, hdrlen, pd->hdr.any);
3635 		PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3636 		    ruleset, pd, 1);
3637 	}
3638 
3639 	if ((r->action == PF_DROP) &&
3640 	    ((r->rule_flag & PFRULE_RETURNRST) ||
3641 	    (r->rule_flag & PFRULE_RETURNICMP) ||
3642 	    (r->rule_flag & PFRULE_RETURN))) {
3643 		pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum,
3644 		    bip_sum, hdrlen, &reason);
3645 	}
3646 
3647 	if (r->action == PF_DROP)
3648 		goto cleanup;
3649 
3650 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3651 		REASON_SET(&reason, PFRES_MEMORY);
3652 		goto cleanup;
3653 	}
3654 	if (rtableid >= 0)
3655 		M_SETFIB(m, rtableid);
3656 
3657 	if (!state_icmp && (r->keep_state || nr != NULL ||
3658 	    (pd->flags & PFDESC_TCP_NORM))) {
3659 		int action;
3660 		action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3661 		    sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3662 		    hdrlen);
3663 		if (action != PF_PASS) {
3664 			if (action == PF_DROP &&
3665 			    (r->rule_flag & PFRULE_RETURN))
3666 				pf_return(r, nr, pd, sk, off, m, th, kif,
3667 				    bproto_sum, bip_sum, hdrlen, &reason);
3668 			return (action);
3669 		}
3670 	} else {
3671 		if (sk != NULL)
3672 			uma_zfree(V_pf_state_key_z, sk);
3673 		if (nk != NULL)
3674 			uma_zfree(V_pf_state_key_z, nk);
3675 	}
3676 
3677 	/* copy back packet headers if we performed NAT operations */
3678 	if (rewrite)
3679 		m_copyback(m, off, hdrlen, pd->hdr.any);
3680 
3681 	if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3682 	    direction == PF_OUT &&
3683 	    V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m))
3684 		/*
3685 		 * We want the state created, but we dont
3686 		 * want to send this in case a partner
3687 		 * firewall has to know about it to allow
3688 		 * replies through it.
3689 		 */
3690 		return (PF_DEFER);
3691 
3692 	return (PF_PASS);
3693 
3694 cleanup:
3695 	if (sk != NULL)
3696 		uma_zfree(V_pf_state_key_z, sk);
3697 	if (nk != NULL)
3698 		uma_zfree(V_pf_state_key_z, nk);
3699 	return (PF_DROP);
3700 }
3701 
3702 static int
3703 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a,
3704     struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk,
3705     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3706     u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_state **sm,
3707     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3708 {
3709 	struct pf_state		*s = NULL;
3710 	struct pf_ksrc_node	*sn = NULL;
3711 	struct tcphdr		*th = &pd->hdr.tcp;
3712 	u_int16_t		 mss = V_tcp_mssdflt;
3713 	u_short			 reason;
3714 
3715 	/* check maximums */
3716 	if (r->max_states &&
3717 	    (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3718 		counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
3719 		REASON_SET(&reason, PFRES_MAXSTATES);
3720 		goto csfailed;
3721 	}
3722 	/* src node for filter rule */
3723 	if ((r->rule_flag & PFRULE_SRCTRACK ||
3724 	    r->rpool.opts & PF_POOL_STICKYADDR) &&
3725 	    pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3726 		REASON_SET(&reason, PFRES_SRCLIMIT);
3727 		goto csfailed;
3728 	}
3729 	/* src node for translation rule */
3730 	if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3731 	    pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3732 		REASON_SET(&reason, PFRES_SRCLIMIT);
3733 		goto csfailed;
3734 	}
3735 	s = pf_alloc_state(M_NOWAIT);
3736 	if (s == NULL) {
3737 		REASON_SET(&reason, PFRES_MEMORY);
3738 		goto csfailed;
3739 	}
3740 	s->rule.ptr = r;
3741 	s->nat_rule.ptr = nr;
3742 	s->anchor.ptr = a;
3743 	STATE_INC_COUNTERS(s);
3744 	if (r->allow_opts)
3745 		s->state_flags |= PFSTATE_ALLOWOPTS;
3746 	if (r->rule_flag & PFRULE_STATESLOPPY)
3747 		s->state_flags |= PFSTATE_SLOPPY;
3748 	s->log = r->log & PF_LOG_ALL;
3749 	s->sync_state = PFSYNC_S_NONE;
3750 	if (nr != NULL)
3751 		s->log |= nr->log & PF_LOG_ALL;
3752 	switch (pd->proto) {
3753 	case IPPROTO_TCP:
3754 		s->src.seqlo = ntohl(th->th_seq);
3755 		s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3756 		if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3757 		    r->keep_state == PF_STATE_MODULATE) {
3758 			/* Generate sequence number modulator */
3759 			if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3760 			    0)
3761 				s->src.seqdiff = 1;
3762 			pf_change_proto_a(m, &th->th_seq, &th->th_sum,
3763 			    htonl(s->src.seqlo + s->src.seqdiff), 0);
3764 			*rewrite = 1;
3765 		} else
3766 			s->src.seqdiff = 0;
3767 		if (th->th_flags & TH_SYN) {
3768 			s->src.seqhi++;
3769 			s->src.wscale = pf_get_wscale(m, off,
3770 			    th->th_off, pd->af);
3771 		}
3772 		s->src.max_win = MAX(ntohs(th->th_win), 1);
3773 		if (s->src.wscale & PF_WSCALE_MASK) {
3774 			/* Remove scale factor from initial window */
3775 			int win = s->src.max_win;
3776 			win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3777 			s->src.max_win = (win - 1) >>
3778 			    (s->src.wscale & PF_WSCALE_MASK);
3779 		}
3780 		if (th->th_flags & TH_FIN)
3781 			s->src.seqhi++;
3782 		s->dst.seqhi = 1;
3783 		s->dst.max_win = 1;
3784 		s->src.state = TCPS_SYN_SENT;
3785 		s->dst.state = TCPS_CLOSED;
3786 		s->timeout = PFTM_TCP_FIRST_PACKET;
3787 		break;
3788 	case IPPROTO_UDP:
3789 		s->src.state = PFUDPS_SINGLE;
3790 		s->dst.state = PFUDPS_NO_TRAFFIC;
3791 		s->timeout = PFTM_UDP_FIRST_PACKET;
3792 		break;
3793 	case IPPROTO_ICMP:
3794 #ifdef INET6
3795 	case IPPROTO_ICMPV6:
3796 #endif
3797 		s->timeout = PFTM_ICMP_FIRST_PACKET;
3798 		break;
3799 	default:
3800 		s->src.state = PFOTHERS_SINGLE;
3801 		s->dst.state = PFOTHERS_NO_TRAFFIC;
3802 		s->timeout = PFTM_OTHER_FIRST_PACKET;
3803 	}
3804 
3805 	if (r->rt) {
3806 		if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) {
3807 			REASON_SET(&reason, PFRES_MAPFAILED);
3808 			pf_src_tree_remove_state(s);
3809 			STATE_DEC_COUNTERS(s);
3810 			pf_free_state(s);
3811 			goto csfailed;
3812 		}
3813 		s->rt_kif = r->rpool.cur->kif;
3814 	}
3815 
3816 	s->creation = time_uptime;
3817 	s->expire = time_uptime;
3818 
3819 	if (sn != NULL)
3820 		s->src_node = sn;
3821 	if (nsn != NULL) {
3822 		/* XXX We only modify one side for now. */
3823 		PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3824 		s->nat_src_node = nsn;
3825 	}
3826 	if (pd->proto == IPPROTO_TCP) {
3827 		if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3828 		    off, pd, th, &s->src, &s->dst)) {
3829 			REASON_SET(&reason, PFRES_MEMORY);
3830 			pf_src_tree_remove_state(s);
3831 			STATE_DEC_COUNTERS(s);
3832 			pf_free_state(s);
3833 			return (PF_DROP);
3834 		}
3835 		if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3836 		    pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3837 		    &s->src, &s->dst, rewrite)) {
3838 			/* This really shouldn't happen!!! */
3839 			DPFPRINTF(PF_DEBUG_URGENT,
3840 			    ("pf_normalize_tcp_stateful failed on first "
3841 			     "pkt\n"));
3842 			pf_src_tree_remove_state(s);
3843 			STATE_DEC_COUNTERS(s);
3844 			pf_free_state(s);
3845 			return (PF_DROP);
3846 		}
3847 	}
3848 	s->direction = pd->dir;
3849 
3850 	/*
3851 	 * sk/nk could already been setup by pf_get_translation().
3852 	 */
3853 	if (nr == NULL) {
3854 		KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3855 		    __func__, nr, sk, nk));
3856 		sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3857 		if (sk == NULL)
3858 			goto csfailed;
3859 		nk = sk;
3860 	} else
3861 		KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3862 		    __func__, nr, sk, nk));
3863 
3864 	/* Swap sk/nk for PF_OUT. */
3865 	if (pf_state_insert(BOUND_IFACE(r, kif), kif,
3866 	    (pd->dir == PF_IN) ? sk : nk,
3867 	    (pd->dir == PF_IN) ? nk : sk, s)) {
3868 		REASON_SET(&reason, PFRES_STATEINS);
3869 		pf_src_tree_remove_state(s);
3870 		STATE_DEC_COUNTERS(s);
3871 		pf_free_state(s);
3872 		return (PF_DROP);
3873 	} else
3874 		*sm = s;
3875 
3876 	if (tag > 0)
3877 		s->tag = tag;
3878 	if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3879 	    TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3880 		s->src.state = PF_TCPS_PROXY_SRC;
3881 		/* undo NAT changes, if they have taken place */
3882 		if (nr != NULL) {
3883 			struct pf_state_key *skt = s->key[PF_SK_WIRE];
3884 			if (pd->dir == PF_OUT)
3885 				skt = s->key[PF_SK_STACK];
3886 			PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3887 			PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3888 			if (pd->sport)
3889 				*pd->sport = skt->port[pd->sidx];
3890 			if (pd->dport)
3891 				*pd->dport = skt->port[pd->didx];
3892 			if (pd->proto_sum)
3893 				*pd->proto_sum = bproto_sum;
3894 			if (pd->ip_sum)
3895 				*pd->ip_sum = bip_sum;
3896 			m_copyback(m, off, hdrlen, pd->hdr.any);
3897 		}
3898 		s->src.seqhi = htonl(arc4random());
3899 		/* Find mss option */
3900 		int rtid = M_GETFIB(m);
3901 		mss = pf_get_mss(m, off, th->th_off, pd->af);
3902 		mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3903 		mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3904 		s->src.mss = mss;
3905 		pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3906 		    th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3907 		    TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3908 		REASON_SET(&reason, PFRES_SYNPROXY);
3909 		return (PF_SYNPROXY_DROP);
3910 	}
3911 
3912 	return (PF_PASS);
3913 
3914 csfailed:
3915 	if (sk != NULL)
3916 		uma_zfree(V_pf_state_key_z, sk);
3917 	if (nk != NULL)
3918 		uma_zfree(V_pf_state_key_z, nk);
3919 
3920 	if (sn != NULL) {
3921 		struct pf_srchash *sh;
3922 
3923 		sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
3924 		PF_HASHROW_LOCK(sh);
3925 		if (--sn->states == 0 && sn->expire == 0) {
3926 			pf_unlink_src_node(sn);
3927 			uma_zfree(V_pf_sources_z, sn);
3928 			counter_u64_add(
3929 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3930 		}
3931 		PF_HASHROW_UNLOCK(sh);
3932 	}
3933 
3934 	if (nsn != sn && nsn != NULL) {
3935 		struct pf_srchash *sh;
3936 
3937 		sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)];
3938 		PF_HASHROW_LOCK(sh);
3939 		if (--nsn->states == 0 && nsn->expire == 0) {
3940 			pf_unlink_src_node(nsn);
3941 			uma_zfree(V_pf_sources_z, nsn);
3942 			counter_u64_add(
3943 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3944 		}
3945 		PF_HASHROW_UNLOCK(sh);
3946 	}
3947 
3948 	return (PF_DROP);
3949 }
3950 
3951 static int
3952 pf_test_fragment(struct pf_krule **rm, int direction, struct pfi_kkif *kif,
3953     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am,
3954     struct pf_kruleset **rsm)
3955 {
3956 	struct pf_krule		*r, *a = NULL;
3957 	struct pf_kruleset	*ruleset = NULL;
3958 	sa_family_t		 af = pd->af;
3959 	u_short			 reason;
3960 	int			 tag = -1;
3961 	int			 asd = 0;
3962 	int			 match = 0;
3963 	struct pf_kanchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
3964 
3965 	PF_RULES_RASSERT();
3966 
3967 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3968 	while (r != NULL) {
3969 		counter_u64_add(r->evaluations, 1);
3970 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
3971 			r = r->skip[PF_SKIP_IFP].ptr;
3972 		else if (r->direction && r->direction != direction)
3973 			r = r->skip[PF_SKIP_DIR].ptr;
3974 		else if (r->af && r->af != af)
3975 			r = r->skip[PF_SKIP_AF].ptr;
3976 		else if (r->proto && r->proto != pd->proto)
3977 			r = r->skip[PF_SKIP_PROTO].ptr;
3978 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3979 		    r->src.neg, kif, M_GETFIB(m)))
3980 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3981 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3982 		    r->dst.neg, NULL, M_GETFIB(m)))
3983 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
3984 		else if (r->tos && !(r->tos == pd->tos))
3985 			r = TAILQ_NEXT(r, entries);
3986 		else if (r->os_fingerprint != PF_OSFP_ANY)
3987 			r = TAILQ_NEXT(r, entries);
3988 		else if (pd->proto == IPPROTO_UDP &&
3989 		    (r->src.port_op || r->dst.port_op))
3990 			r = TAILQ_NEXT(r, entries);
3991 		else if (pd->proto == IPPROTO_TCP &&
3992 		    (r->src.port_op || r->dst.port_op || r->flagset))
3993 			r = TAILQ_NEXT(r, entries);
3994 		else if ((pd->proto == IPPROTO_ICMP ||
3995 		    pd->proto == IPPROTO_ICMPV6) &&
3996 		    (r->type || r->code))
3997 			r = TAILQ_NEXT(r, entries);
3998 		else if (r->prio &&
3999 		    !pf_match_ieee8021q_pcp(r->prio, m))
4000 			r = TAILQ_NEXT(r, entries);
4001 		else if (r->prob && r->prob <=
4002 		    (arc4random() % (UINT_MAX - 1) + 1))
4003 			r = TAILQ_NEXT(r, entries);
4004 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
4005 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
4006 			r = TAILQ_NEXT(r, entries);
4007 		else {
4008 			if (r->anchor == NULL) {
4009 				match = 1;
4010 				*rm = r;
4011 				*am = a;
4012 				*rsm = ruleset;
4013 				if ((*rm)->quick)
4014 					break;
4015 				r = TAILQ_NEXT(r, entries);
4016 			} else
4017 				pf_step_into_anchor(anchor_stack, &asd,
4018 				    &ruleset, PF_RULESET_FILTER, &r, &a,
4019 				    &match);
4020 		}
4021 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
4022 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
4023 			break;
4024 	}
4025 	r = *rm;
4026 	a = *am;
4027 	ruleset = *rsm;
4028 
4029 	REASON_SET(&reason, PFRES_MATCH);
4030 
4031 	if (r->log)
4032 		PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
4033 		    1);
4034 
4035 	if (r->action != PF_PASS)
4036 		return (PF_DROP);
4037 
4038 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
4039 		REASON_SET(&reason, PFRES_MEMORY);
4040 		return (PF_DROP);
4041 	}
4042 
4043 	return (PF_PASS);
4044 }
4045 
4046 static int
4047 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
4048 	struct pf_state **state, struct pfi_kkif *kif, struct mbuf *m, int off,
4049 	struct pf_pdesc *pd, u_short *reason, int *copyback)
4050 {
4051 	struct tcphdr		*th = &pd->hdr.tcp;
4052 	u_int16_t		 win = ntohs(th->th_win);
4053 	u_int32_t		 ack, end, seq, orig_seq;
4054 	u_int8_t		 sws, dws;
4055 	int			 ackskew;
4056 
4057 	if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
4058 		sws = src->wscale & PF_WSCALE_MASK;
4059 		dws = dst->wscale & PF_WSCALE_MASK;
4060 	} else
4061 		sws = dws = 0;
4062 
4063 	/*
4064 	 * Sequence tracking algorithm from Guido van Rooij's paper:
4065 	 *   http://www.madison-gurkha.com/publications/tcp_filtering/
4066 	 *	tcp_filtering.ps
4067 	 */
4068 
4069 	orig_seq = seq = ntohl(th->th_seq);
4070 	if (src->seqlo == 0) {
4071 		/* First packet from this end. Set its state */
4072 
4073 		if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
4074 		    src->scrub == NULL) {
4075 			if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
4076 				REASON_SET(reason, PFRES_MEMORY);
4077 				return (PF_DROP);
4078 			}
4079 		}
4080 
4081 		/* Deferred generation of sequence number modulator */
4082 		if (dst->seqdiff && !src->seqdiff) {
4083 			/* use random iss for the TCP server */
4084 			while ((src->seqdiff = arc4random() - seq) == 0)
4085 				;
4086 			ack = ntohl(th->th_ack) - dst->seqdiff;
4087 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
4088 			    src->seqdiff), 0);
4089 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
4090 			*copyback = 1;
4091 		} else {
4092 			ack = ntohl(th->th_ack);
4093 		}
4094 
4095 		end = seq + pd->p_len;
4096 		if (th->th_flags & TH_SYN) {
4097 			end++;
4098 			if (dst->wscale & PF_WSCALE_FLAG) {
4099 				src->wscale = pf_get_wscale(m, off, th->th_off,
4100 				    pd->af);
4101 				if (src->wscale & PF_WSCALE_FLAG) {
4102 					/* Remove scale factor from initial
4103 					 * window */
4104 					sws = src->wscale & PF_WSCALE_MASK;
4105 					win = ((u_int32_t)win + (1 << sws) - 1)
4106 					    >> sws;
4107 					dws = dst->wscale & PF_WSCALE_MASK;
4108 				} else {
4109 					/* fixup other window */
4110 					dst->max_win <<= dst->wscale &
4111 					    PF_WSCALE_MASK;
4112 					/* in case of a retrans SYN|ACK */
4113 					dst->wscale = 0;
4114 				}
4115 			}
4116 		}
4117 		if (th->th_flags & TH_FIN)
4118 			end++;
4119 
4120 		src->seqlo = seq;
4121 		if (src->state < TCPS_SYN_SENT)
4122 			src->state = TCPS_SYN_SENT;
4123 
4124 		/*
4125 		 * May need to slide the window (seqhi may have been set by
4126 		 * the crappy stack check or if we picked up the connection
4127 		 * after establishment)
4128 		 */
4129 		if (src->seqhi == 1 ||
4130 		    SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
4131 			src->seqhi = end + MAX(1, dst->max_win << dws);
4132 		if (win > src->max_win)
4133 			src->max_win = win;
4134 
4135 	} else {
4136 		ack = ntohl(th->th_ack) - dst->seqdiff;
4137 		if (src->seqdiff) {
4138 			/* Modulate sequence numbers */
4139 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
4140 			    src->seqdiff), 0);
4141 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
4142 			*copyback = 1;
4143 		}
4144 		end = seq + pd->p_len;
4145 		if (th->th_flags & TH_SYN)
4146 			end++;
4147 		if (th->th_flags & TH_FIN)
4148 			end++;
4149 	}
4150 
4151 	if ((th->th_flags & TH_ACK) == 0) {
4152 		/* Let it pass through the ack skew check */
4153 		ack = dst->seqlo;
4154 	} else if ((ack == 0 &&
4155 	    (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
4156 	    /* broken tcp stacks do not set ack */
4157 	    (dst->state < TCPS_SYN_SENT)) {
4158 		/*
4159 		 * Many stacks (ours included) will set the ACK number in an
4160 		 * FIN|ACK if the SYN times out -- no sequence to ACK.
4161 		 */
4162 		ack = dst->seqlo;
4163 	}
4164 
4165 	if (seq == end) {
4166 		/* Ease sequencing restrictions on no data packets */
4167 		seq = src->seqlo;
4168 		end = seq;
4169 	}
4170 
4171 	ackskew = dst->seqlo - ack;
4172 
4173 	/*
4174 	 * Need to demodulate the sequence numbers in any TCP SACK options
4175 	 * (Selective ACK). We could optionally validate the SACK values
4176 	 * against the current ACK window, either forwards or backwards, but
4177 	 * I'm not confident that SACK has been implemented properly
4178 	 * everywhere. It wouldn't surprise me if several stacks accidentally
4179 	 * SACK too far backwards of previously ACKed data. There really aren't
4180 	 * any security implications of bad SACKing unless the target stack
4181 	 * doesn't validate the option length correctly. Someone trying to
4182 	 * spoof into a TCP connection won't bother blindly sending SACK
4183 	 * options anyway.
4184 	 */
4185 	if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
4186 		if (pf_modulate_sack(m, off, pd, th, dst))
4187 			*copyback = 1;
4188 	}
4189 
4190 #define	MAXACKWINDOW (0xffff + 1500)	/* 1500 is an arbitrary fudge factor */
4191 	if (SEQ_GEQ(src->seqhi, end) &&
4192 	    /* Last octet inside other's window space */
4193 	    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
4194 	    /* Retrans: not more than one window back */
4195 	    (ackskew >= -MAXACKWINDOW) &&
4196 	    /* Acking not more than one reassembled fragment backwards */
4197 	    (ackskew <= (MAXACKWINDOW << sws)) &&
4198 	    /* Acking not more than one window forward */
4199 	    ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
4200 	    (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
4201 	    (pd->flags & PFDESC_IP_REAS) == 0)) {
4202 	    /* Require an exact/+1 sequence match on resets when possible */
4203 
4204 		if (dst->scrub || src->scrub) {
4205 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4206 			    *state, src, dst, copyback))
4207 				return (PF_DROP);
4208 		}
4209 
4210 		/* update max window */
4211 		if (src->max_win < win)
4212 			src->max_win = win;
4213 		/* synchronize sequencing */
4214 		if (SEQ_GT(end, src->seqlo))
4215 			src->seqlo = end;
4216 		/* slide the window of what the other end can send */
4217 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4218 			dst->seqhi = ack + MAX((win << sws), 1);
4219 
4220 		/* update states */
4221 		if (th->th_flags & TH_SYN)
4222 			if (src->state < TCPS_SYN_SENT)
4223 				src->state = TCPS_SYN_SENT;
4224 		if (th->th_flags & TH_FIN)
4225 			if (src->state < TCPS_CLOSING)
4226 				src->state = TCPS_CLOSING;
4227 		if (th->th_flags & TH_ACK) {
4228 			if (dst->state == TCPS_SYN_SENT) {
4229 				dst->state = TCPS_ESTABLISHED;
4230 				if (src->state == TCPS_ESTABLISHED &&
4231 				    (*state)->src_node != NULL &&
4232 				    pf_src_connlimit(state)) {
4233 					REASON_SET(reason, PFRES_SRCLIMIT);
4234 					return (PF_DROP);
4235 				}
4236 			} else if (dst->state == TCPS_CLOSING)
4237 				dst->state = TCPS_FIN_WAIT_2;
4238 		}
4239 		if (th->th_flags & TH_RST)
4240 			src->state = dst->state = TCPS_TIME_WAIT;
4241 
4242 		/* update expire time */
4243 		(*state)->expire = time_uptime;
4244 		if (src->state >= TCPS_FIN_WAIT_2 &&
4245 		    dst->state >= TCPS_FIN_WAIT_2)
4246 			(*state)->timeout = PFTM_TCP_CLOSED;
4247 		else if (src->state >= TCPS_CLOSING &&
4248 		    dst->state >= TCPS_CLOSING)
4249 			(*state)->timeout = PFTM_TCP_FIN_WAIT;
4250 		else if (src->state < TCPS_ESTABLISHED ||
4251 		    dst->state < TCPS_ESTABLISHED)
4252 			(*state)->timeout = PFTM_TCP_OPENING;
4253 		else if (src->state >= TCPS_CLOSING ||
4254 		    dst->state >= TCPS_CLOSING)
4255 			(*state)->timeout = PFTM_TCP_CLOSING;
4256 		else
4257 			(*state)->timeout = PFTM_TCP_ESTABLISHED;
4258 
4259 		/* Fall through to PASS packet */
4260 
4261 	} else if ((dst->state < TCPS_SYN_SENT ||
4262 		dst->state >= TCPS_FIN_WAIT_2 ||
4263 		src->state >= TCPS_FIN_WAIT_2) &&
4264 	    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4265 	    /* Within a window forward of the originating packet */
4266 	    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4267 	    /* Within a window backward of the originating packet */
4268 
4269 		/*
4270 		 * This currently handles three situations:
4271 		 *  1) Stupid stacks will shotgun SYNs before their peer
4272 		 *     replies.
4273 		 *  2) When PF catches an already established stream (the
4274 		 *     firewall rebooted, the state table was flushed, routes
4275 		 *     changed...)
4276 		 *  3) Packets get funky immediately after the connection
4277 		 *     closes (this should catch Solaris spurious ACK|FINs
4278 		 *     that web servers like to spew after a close)
4279 		 *
4280 		 * This must be a little more careful than the above code
4281 		 * since packet floods will also be caught here. We don't
4282 		 * update the TTL here to mitigate the damage of a packet
4283 		 * flood and so the same code can handle awkward establishment
4284 		 * and a loosened connection close.
4285 		 * In the establishment case, a correct peer response will
4286 		 * validate the connection, go through the normal state code
4287 		 * and keep updating the state TTL.
4288 		 */
4289 
4290 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
4291 			printf("pf: loose state match: ");
4292 			pf_print_state(*state);
4293 			pf_print_flags(th->th_flags);
4294 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4295 			    "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
4296 			    pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
4297 			    (unsigned long long)(*state)->packets[1],
4298 			    pd->dir == PF_IN ? "in" : "out",
4299 			    pd->dir == (*state)->direction ? "fwd" : "rev");
4300 		}
4301 
4302 		if (dst->scrub || src->scrub) {
4303 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4304 			    *state, src, dst, copyback))
4305 				return (PF_DROP);
4306 		}
4307 
4308 		/* update max window */
4309 		if (src->max_win < win)
4310 			src->max_win = win;
4311 		/* synchronize sequencing */
4312 		if (SEQ_GT(end, src->seqlo))
4313 			src->seqlo = end;
4314 		/* slide the window of what the other end can send */
4315 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4316 			dst->seqhi = ack + MAX((win << sws), 1);
4317 
4318 		/*
4319 		 * Cannot set dst->seqhi here since this could be a shotgunned
4320 		 * SYN and not an already established connection.
4321 		 */
4322 
4323 		if (th->th_flags & TH_FIN)
4324 			if (src->state < TCPS_CLOSING)
4325 				src->state = TCPS_CLOSING;
4326 		if (th->th_flags & TH_RST)
4327 			src->state = dst->state = TCPS_TIME_WAIT;
4328 
4329 		/* Fall through to PASS packet */
4330 
4331 	} else {
4332 		if ((*state)->dst.state == TCPS_SYN_SENT &&
4333 		    (*state)->src.state == TCPS_SYN_SENT) {
4334 			/* Send RST for state mismatches during handshake */
4335 			if (!(th->th_flags & TH_RST))
4336 				pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4337 				    pd->dst, pd->src, th->th_dport,
4338 				    th->th_sport, ntohl(th->th_ack), 0,
4339 				    TH_RST, 0, 0,
4340 				    (*state)->rule.ptr->return_ttl, 1, 0,
4341 				    kif->pfik_ifp);
4342 			src->seqlo = 0;
4343 			src->seqhi = 1;
4344 			src->max_win = 1;
4345 		} else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4346 			printf("pf: BAD state: ");
4347 			pf_print_state(*state);
4348 			pf_print_flags(th->th_flags);
4349 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4350 			    "pkts=%llu:%llu dir=%s,%s\n",
4351 			    seq, orig_seq, ack, pd->p_len, ackskew,
4352 			    (unsigned long long)(*state)->packets[0],
4353 			    (unsigned long long)(*state)->packets[1],
4354 			    pd->dir == PF_IN ? "in" : "out",
4355 			    pd->dir == (*state)->direction ? "fwd" : "rev");
4356 			printf("pf: State failure on: %c %c %c %c | %c %c\n",
4357 			    SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4358 			    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4359 			    ' ': '2',
4360 			    (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4361 			    (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4362 			    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4363 			    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4364 		}
4365 		REASON_SET(reason, PFRES_BADSTATE);
4366 		return (PF_DROP);
4367 	}
4368 
4369 	return (PF_PASS);
4370 }
4371 
4372 static int
4373 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4374 	struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4375 {
4376 	struct tcphdr		*th = &pd->hdr.tcp;
4377 
4378 	if (th->th_flags & TH_SYN)
4379 		if (src->state < TCPS_SYN_SENT)
4380 			src->state = TCPS_SYN_SENT;
4381 	if (th->th_flags & TH_FIN)
4382 		if (src->state < TCPS_CLOSING)
4383 			src->state = TCPS_CLOSING;
4384 	if (th->th_flags & TH_ACK) {
4385 		if (dst->state == TCPS_SYN_SENT) {
4386 			dst->state = TCPS_ESTABLISHED;
4387 			if (src->state == TCPS_ESTABLISHED &&
4388 			    (*state)->src_node != NULL &&
4389 			    pf_src_connlimit(state)) {
4390 				REASON_SET(reason, PFRES_SRCLIMIT);
4391 				return (PF_DROP);
4392 			}
4393 		} else if (dst->state == TCPS_CLOSING) {
4394 			dst->state = TCPS_FIN_WAIT_2;
4395 		} else if (src->state == TCPS_SYN_SENT &&
4396 		    dst->state < TCPS_SYN_SENT) {
4397 			/*
4398 			 * Handle a special sloppy case where we only see one
4399 			 * half of the connection. If there is a ACK after
4400 			 * the initial SYN without ever seeing a packet from
4401 			 * the destination, set the connection to established.
4402 			 */
4403 			dst->state = src->state = TCPS_ESTABLISHED;
4404 			if ((*state)->src_node != NULL &&
4405 			    pf_src_connlimit(state)) {
4406 				REASON_SET(reason, PFRES_SRCLIMIT);
4407 				return (PF_DROP);
4408 			}
4409 		} else if (src->state == TCPS_CLOSING &&
4410 		    dst->state == TCPS_ESTABLISHED &&
4411 		    dst->seqlo == 0) {
4412 			/*
4413 			 * Handle the closing of half connections where we
4414 			 * don't see the full bidirectional FIN/ACK+ACK
4415 			 * handshake.
4416 			 */
4417 			dst->state = TCPS_CLOSING;
4418 		}
4419 	}
4420 	if (th->th_flags & TH_RST)
4421 		src->state = dst->state = TCPS_TIME_WAIT;
4422 
4423 	/* update expire time */
4424 	(*state)->expire = time_uptime;
4425 	if (src->state >= TCPS_FIN_WAIT_2 &&
4426 	    dst->state >= TCPS_FIN_WAIT_2)
4427 		(*state)->timeout = PFTM_TCP_CLOSED;
4428 	else if (src->state >= TCPS_CLOSING &&
4429 	    dst->state >= TCPS_CLOSING)
4430 		(*state)->timeout = PFTM_TCP_FIN_WAIT;
4431 	else if (src->state < TCPS_ESTABLISHED ||
4432 	    dst->state < TCPS_ESTABLISHED)
4433 		(*state)->timeout = PFTM_TCP_OPENING;
4434 	else if (src->state >= TCPS_CLOSING ||
4435 	    dst->state >= TCPS_CLOSING)
4436 		(*state)->timeout = PFTM_TCP_CLOSING;
4437 	else
4438 		(*state)->timeout = PFTM_TCP_ESTABLISHED;
4439 
4440 	return (PF_PASS);
4441 }
4442 
4443 static int
4444 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kkif *kif,
4445     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4446     u_short *reason)
4447 {
4448 	struct pf_state_key_cmp	 key;
4449 	struct tcphdr		*th = &pd->hdr.tcp;
4450 	int			 copyback = 0;
4451 	struct pf_state_peer	*src, *dst;
4452 	struct pf_state_key	*sk;
4453 
4454 	bzero(&key, sizeof(key));
4455 	key.af = pd->af;
4456 	key.proto = IPPROTO_TCP;
4457 	if (direction == PF_IN)	{	/* wire side, straight */
4458 		PF_ACPY(&key.addr[0], pd->src, key.af);
4459 		PF_ACPY(&key.addr[1], pd->dst, key.af);
4460 		key.port[0] = th->th_sport;
4461 		key.port[1] = th->th_dport;
4462 	} else {			/* stack side, reverse */
4463 		PF_ACPY(&key.addr[1], pd->src, key.af);
4464 		PF_ACPY(&key.addr[0], pd->dst, key.af);
4465 		key.port[1] = th->th_sport;
4466 		key.port[0] = th->th_dport;
4467 	}
4468 
4469 	STATE_LOOKUP(kif, &key, direction, *state, pd);
4470 
4471 	if (direction == (*state)->direction) {
4472 		src = &(*state)->src;
4473 		dst = &(*state)->dst;
4474 	} else {
4475 		src = &(*state)->dst;
4476 		dst = &(*state)->src;
4477 	}
4478 
4479 	sk = (*state)->key[pd->didx];
4480 
4481 	if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4482 		if (direction != (*state)->direction) {
4483 			REASON_SET(reason, PFRES_SYNPROXY);
4484 			return (PF_SYNPROXY_DROP);
4485 		}
4486 		if (th->th_flags & TH_SYN) {
4487 			if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4488 				REASON_SET(reason, PFRES_SYNPROXY);
4489 				return (PF_DROP);
4490 			}
4491 			pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4492 			    pd->src, th->th_dport, th->th_sport,
4493 			    (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4494 			    TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4495 			REASON_SET(reason, PFRES_SYNPROXY);
4496 			return (PF_SYNPROXY_DROP);
4497 		} else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK ||
4498 		    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4499 		    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4500 			REASON_SET(reason, PFRES_SYNPROXY);
4501 			return (PF_DROP);
4502 		} else if ((*state)->src_node != NULL &&
4503 		    pf_src_connlimit(state)) {
4504 			REASON_SET(reason, PFRES_SRCLIMIT);
4505 			return (PF_DROP);
4506 		} else
4507 			(*state)->src.state = PF_TCPS_PROXY_DST;
4508 	}
4509 	if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4510 		if (direction == (*state)->direction) {
4511 			if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4512 			    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4513 			    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4514 				REASON_SET(reason, PFRES_SYNPROXY);
4515 				return (PF_DROP);
4516 			}
4517 			(*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4518 			if ((*state)->dst.seqhi == 1)
4519 				(*state)->dst.seqhi = htonl(arc4random());
4520 			pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4521 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
4522 			    sk->port[pd->sidx], sk->port[pd->didx],
4523 			    (*state)->dst.seqhi, 0, TH_SYN, 0,
4524 			    (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4525 			REASON_SET(reason, PFRES_SYNPROXY);
4526 			return (PF_SYNPROXY_DROP);
4527 		} else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4528 		    (TH_SYN|TH_ACK)) ||
4529 		    (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4530 			REASON_SET(reason, PFRES_SYNPROXY);
4531 			return (PF_DROP);
4532 		} else {
4533 			(*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4534 			(*state)->dst.seqlo = ntohl(th->th_seq);
4535 			pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4536 			    pd->src, th->th_dport, th->th_sport,
4537 			    ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4538 			    TH_ACK, (*state)->src.max_win, 0, 0, 0,
4539 			    (*state)->tag, NULL);
4540 			pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4541 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
4542 			    sk->port[pd->sidx], sk->port[pd->didx],
4543 			    (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4544 			    TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4545 			(*state)->src.seqdiff = (*state)->dst.seqhi -
4546 			    (*state)->src.seqlo;
4547 			(*state)->dst.seqdiff = (*state)->src.seqhi -
4548 			    (*state)->dst.seqlo;
4549 			(*state)->src.seqhi = (*state)->src.seqlo +
4550 			    (*state)->dst.max_win;
4551 			(*state)->dst.seqhi = (*state)->dst.seqlo +
4552 			    (*state)->src.max_win;
4553 			(*state)->src.wscale = (*state)->dst.wscale = 0;
4554 			(*state)->src.state = (*state)->dst.state =
4555 			    TCPS_ESTABLISHED;
4556 			REASON_SET(reason, PFRES_SYNPROXY);
4557 			return (PF_SYNPROXY_DROP);
4558 		}
4559 	}
4560 
4561 	if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4562 	    dst->state >= TCPS_FIN_WAIT_2 &&
4563 	    src->state >= TCPS_FIN_WAIT_2) {
4564 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
4565 			printf("pf: state reuse ");
4566 			pf_print_state(*state);
4567 			pf_print_flags(th->th_flags);
4568 			printf("\n");
4569 		}
4570 		/* XXX make sure it's the same direction ?? */
4571 		(*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4572 		pf_unlink_state(*state, PF_ENTER_LOCKED);
4573 		*state = NULL;
4574 		return (PF_DROP);
4575 	}
4576 
4577 	if ((*state)->state_flags & PFSTATE_SLOPPY) {
4578 		if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4579 			return (PF_DROP);
4580 	} else {
4581 		if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4582 		    &copyback) == PF_DROP)
4583 			return (PF_DROP);
4584 	}
4585 
4586 	/* translate source/destination address, if necessary */
4587 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4588 		struct pf_state_key *nk = (*state)->key[pd->didx];
4589 
4590 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4591 		    nk->port[pd->sidx] != th->th_sport)
4592 			pf_change_ap(m, pd->src, &th->th_sport,
4593 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
4594 			    nk->port[pd->sidx], 0, pd->af);
4595 
4596 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4597 		    nk->port[pd->didx] != th->th_dport)
4598 			pf_change_ap(m, pd->dst, &th->th_dport,
4599 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
4600 			    nk->port[pd->didx], 0, pd->af);
4601 		copyback = 1;
4602 	}
4603 
4604 	/* Copyback sequence modulation or stateful scrub changes if needed */
4605 	if (copyback)
4606 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
4607 
4608 	return (PF_PASS);
4609 }
4610 
4611 static int
4612 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kkif *kif,
4613     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4614 {
4615 	struct pf_state_peer	*src, *dst;
4616 	struct pf_state_key_cmp	 key;
4617 	struct udphdr		*uh = &pd->hdr.udp;
4618 
4619 	bzero(&key, sizeof(key));
4620 	key.af = pd->af;
4621 	key.proto = IPPROTO_UDP;
4622 	if (direction == PF_IN)	{	/* wire side, straight */
4623 		PF_ACPY(&key.addr[0], pd->src, key.af);
4624 		PF_ACPY(&key.addr[1], pd->dst, key.af);
4625 		key.port[0] = uh->uh_sport;
4626 		key.port[1] = uh->uh_dport;
4627 	} else {			/* stack side, reverse */
4628 		PF_ACPY(&key.addr[1], pd->src, key.af);
4629 		PF_ACPY(&key.addr[0], pd->dst, key.af);
4630 		key.port[1] = uh->uh_sport;
4631 		key.port[0] = uh->uh_dport;
4632 	}
4633 
4634 	STATE_LOOKUP(kif, &key, direction, *state, pd);
4635 
4636 	if (direction == (*state)->direction) {
4637 		src = &(*state)->src;
4638 		dst = &(*state)->dst;
4639 	} else {
4640 		src = &(*state)->dst;
4641 		dst = &(*state)->src;
4642 	}
4643 
4644 	/* update states */
4645 	if (src->state < PFUDPS_SINGLE)
4646 		src->state = PFUDPS_SINGLE;
4647 	if (dst->state == PFUDPS_SINGLE)
4648 		dst->state = PFUDPS_MULTIPLE;
4649 
4650 	/* update expire time */
4651 	(*state)->expire = time_uptime;
4652 	if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4653 		(*state)->timeout = PFTM_UDP_MULTIPLE;
4654 	else
4655 		(*state)->timeout = PFTM_UDP_SINGLE;
4656 
4657 	/* translate source/destination address, if necessary */
4658 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4659 		struct pf_state_key *nk = (*state)->key[pd->didx];
4660 
4661 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4662 		    nk->port[pd->sidx] != uh->uh_sport)
4663 			pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
4664 			    &uh->uh_sum, &nk->addr[pd->sidx],
4665 			    nk->port[pd->sidx], 1, pd->af);
4666 
4667 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4668 		    nk->port[pd->didx] != uh->uh_dport)
4669 			pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
4670 			    &uh->uh_sum, &nk->addr[pd->didx],
4671 			    nk->port[pd->didx], 1, pd->af);
4672 		m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4673 	}
4674 
4675 	return (PF_PASS);
4676 }
4677 
4678 static int
4679 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kkif *kif,
4680     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4681 {
4682 	struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
4683 	u_int16_t	 icmpid = 0, *icmpsum;
4684 	u_int8_t	 icmptype, icmpcode;
4685 	int		 state_icmp = 0;
4686 	struct pf_state_key_cmp key;
4687 
4688 	bzero(&key, sizeof(key));
4689 	switch (pd->proto) {
4690 #ifdef INET
4691 	case IPPROTO_ICMP:
4692 		icmptype = pd->hdr.icmp.icmp_type;
4693 		icmpcode = pd->hdr.icmp.icmp_code;
4694 		icmpid = pd->hdr.icmp.icmp_id;
4695 		icmpsum = &pd->hdr.icmp.icmp_cksum;
4696 
4697 		if (icmptype == ICMP_UNREACH ||
4698 		    icmptype == ICMP_SOURCEQUENCH ||
4699 		    icmptype == ICMP_REDIRECT ||
4700 		    icmptype == ICMP_TIMXCEED ||
4701 		    icmptype == ICMP_PARAMPROB)
4702 			state_icmp++;
4703 		break;
4704 #endif /* INET */
4705 #ifdef INET6
4706 	case IPPROTO_ICMPV6:
4707 		icmptype = pd->hdr.icmp6.icmp6_type;
4708 		icmpcode = pd->hdr.icmp6.icmp6_code;
4709 		icmpid = pd->hdr.icmp6.icmp6_id;
4710 		icmpsum = &pd->hdr.icmp6.icmp6_cksum;
4711 
4712 		if (icmptype == ICMP6_DST_UNREACH ||
4713 		    icmptype == ICMP6_PACKET_TOO_BIG ||
4714 		    icmptype == ICMP6_TIME_EXCEEDED ||
4715 		    icmptype == ICMP6_PARAM_PROB)
4716 			state_icmp++;
4717 		break;
4718 #endif /* INET6 */
4719 	}
4720 
4721 	if (!state_icmp) {
4722 		/*
4723 		 * ICMP query/reply message not related to a TCP/UDP packet.
4724 		 * Search for an ICMP state.
4725 		 */
4726 		key.af = pd->af;
4727 		key.proto = pd->proto;
4728 		key.port[0] = key.port[1] = icmpid;
4729 		if (direction == PF_IN)	{	/* wire side, straight */
4730 			PF_ACPY(&key.addr[0], pd->src, key.af);
4731 			PF_ACPY(&key.addr[1], pd->dst, key.af);
4732 		} else {			/* stack side, reverse */
4733 			PF_ACPY(&key.addr[1], pd->src, key.af);
4734 			PF_ACPY(&key.addr[0], pd->dst, key.af);
4735 		}
4736 
4737 		STATE_LOOKUP(kif, &key, direction, *state, pd);
4738 
4739 		(*state)->expire = time_uptime;
4740 		(*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4741 
4742 		/* translate source/destination address, if necessary */
4743 		if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4744 			struct pf_state_key *nk = (*state)->key[pd->didx];
4745 
4746 			switch (pd->af) {
4747 #ifdef INET
4748 			case AF_INET:
4749 				if (PF_ANEQ(pd->src,
4750 				    &nk->addr[pd->sidx], AF_INET))
4751 					pf_change_a(&saddr->v4.s_addr,
4752 					    pd->ip_sum,
4753 					    nk->addr[pd->sidx].v4.s_addr, 0);
4754 
4755 				if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4756 				    AF_INET))
4757 					pf_change_a(&daddr->v4.s_addr,
4758 					    pd->ip_sum,
4759 					    nk->addr[pd->didx].v4.s_addr, 0);
4760 
4761 				if (nk->port[0] !=
4762 				    pd->hdr.icmp.icmp_id) {
4763 					pd->hdr.icmp.icmp_cksum =
4764 					    pf_cksum_fixup(
4765 					    pd->hdr.icmp.icmp_cksum, icmpid,
4766 					    nk->port[pd->sidx], 0);
4767 					pd->hdr.icmp.icmp_id =
4768 					    nk->port[pd->sidx];
4769 				}
4770 
4771 				m_copyback(m, off, ICMP_MINLEN,
4772 				    (caddr_t )&pd->hdr.icmp);
4773 				break;
4774 #endif /* INET */
4775 #ifdef INET6
4776 			case AF_INET6:
4777 				if (PF_ANEQ(pd->src,
4778 				    &nk->addr[pd->sidx], AF_INET6))
4779 					pf_change_a6(saddr,
4780 					    &pd->hdr.icmp6.icmp6_cksum,
4781 					    &nk->addr[pd->sidx], 0);
4782 
4783 				if (PF_ANEQ(pd->dst,
4784 				    &nk->addr[pd->didx], AF_INET6))
4785 					pf_change_a6(daddr,
4786 					    &pd->hdr.icmp6.icmp6_cksum,
4787 					    &nk->addr[pd->didx], 0);
4788 
4789 				m_copyback(m, off, sizeof(struct icmp6_hdr),
4790 				    (caddr_t )&pd->hdr.icmp6);
4791 				break;
4792 #endif /* INET6 */
4793 			}
4794 		}
4795 		return (PF_PASS);
4796 
4797 	} else {
4798 		/*
4799 		 * ICMP error message in response to a TCP/UDP packet.
4800 		 * Extract the inner TCP/UDP header and search for that state.
4801 		 */
4802 
4803 		struct pf_pdesc	pd2;
4804 		bzero(&pd2, sizeof pd2);
4805 #ifdef INET
4806 		struct ip	h2;
4807 #endif /* INET */
4808 #ifdef INET6
4809 		struct ip6_hdr	h2_6;
4810 		int		terminal = 0;
4811 #endif /* INET6 */
4812 		int		ipoff2 = 0;
4813 		int		off2 = 0;
4814 
4815 		pd2.af = pd->af;
4816 		/* Payload packet is from the opposite direction. */
4817 		pd2.sidx = (direction == PF_IN) ? 1 : 0;
4818 		pd2.didx = (direction == PF_IN) ? 0 : 1;
4819 		switch (pd->af) {
4820 #ifdef INET
4821 		case AF_INET:
4822 			/* offset of h2 in mbuf chain */
4823 			ipoff2 = off + ICMP_MINLEN;
4824 
4825 			if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4826 			    NULL, reason, pd2.af)) {
4827 				DPFPRINTF(PF_DEBUG_MISC,
4828 				    ("pf: ICMP error message too short "
4829 				    "(ip)\n"));
4830 				return (PF_DROP);
4831 			}
4832 			/*
4833 			 * ICMP error messages don't refer to non-first
4834 			 * fragments
4835 			 */
4836 			if (h2.ip_off & htons(IP_OFFMASK)) {
4837 				REASON_SET(reason, PFRES_FRAG);
4838 				return (PF_DROP);
4839 			}
4840 
4841 			/* offset of protocol header that follows h2 */
4842 			off2 = ipoff2 + (h2.ip_hl << 2);
4843 
4844 			pd2.proto = h2.ip_p;
4845 			pd2.src = (struct pf_addr *)&h2.ip_src;
4846 			pd2.dst = (struct pf_addr *)&h2.ip_dst;
4847 			pd2.ip_sum = &h2.ip_sum;
4848 			break;
4849 #endif /* INET */
4850 #ifdef INET6
4851 		case AF_INET6:
4852 			ipoff2 = off + sizeof(struct icmp6_hdr);
4853 
4854 			if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4855 			    NULL, reason, pd2.af)) {
4856 				DPFPRINTF(PF_DEBUG_MISC,
4857 				    ("pf: ICMP error message too short "
4858 				    "(ip6)\n"));
4859 				return (PF_DROP);
4860 			}
4861 			pd2.proto = h2_6.ip6_nxt;
4862 			pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4863 			pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4864 			pd2.ip_sum = NULL;
4865 			off2 = ipoff2 + sizeof(h2_6);
4866 			do {
4867 				switch (pd2.proto) {
4868 				case IPPROTO_FRAGMENT:
4869 					/*
4870 					 * ICMPv6 error messages for
4871 					 * non-first fragments
4872 					 */
4873 					REASON_SET(reason, PFRES_FRAG);
4874 					return (PF_DROP);
4875 				case IPPROTO_AH:
4876 				case IPPROTO_HOPOPTS:
4877 				case IPPROTO_ROUTING:
4878 				case IPPROTO_DSTOPTS: {
4879 					/* get next header and header length */
4880 					struct ip6_ext opt6;
4881 
4882 					if (!pf_pull_hdr(m, off2, &opt6,
4883 					    sizeof(opt6), NULL, reason,
4884 					    pd2.af)) {
4885 						DPFPRINTF(PF_DEBUG_MISC,
4886 						    ("pf: ICMPv6 short opt\n"));
4887 						return (PF_DROP);
4888 					}
4889 					if (pd2.proto == IPPROTO_AH)
4890 						off2 += (opt6.ip6e_len + 2) * 4;
4891 					else
4892 						off2 += (opt6.ip6e_len + 1) * 8;
4893 					pd2.proto = opt6.ip6e_nxt;
4894 					/* goto the next header */
4895 					break;
4896 				}
4897 				default:
4898 					terminal++;
4899 					break;
4900 				}
4901 			} while (!terminal);
4902 			break;
4903 #endif /* INET6 */
4904 		}
4905 
4906 		if (PF_ANEQ(pd->dst, pd2.src, pd->af)) {
4907 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
4908 				printf("pf: BAD ICMP %d:%d outer dst: ",
4909 				    icmptype, icmpcode);
4910 				pf_print_host(pd->src, 0, pd->af);
4911 				printf(" -> ");
4912 				pf_print_host(pd->dst, 0, pd->af);
4913 				printf(" inner src: ");
4914 				pf_print_host(pd2.src, 0, pd2.af);
4915 				printf(" -> ");
4916 				pf_print_host(pd2.dst, 0, pd2.af);
4917 				printf("\n");
4918 			}
4919 			REASON_SET(reason, PFRES_BADSTATE);
4920 			return (PF_DROP);
4921 		}
4922 
4923 		switch (pd2.proto) {
4924 		case IPPROTO_TCP: {
4925 			struct tcphdr		 th;
4926 			u_int32_t		 seq;
4927 			struct pf_state_peer	*src, *dst;
4928 			u_int8_t		 dws;
4929 			int			 copyback = 0;
4930 
4931 			/*
4932 			 * Only the first 8 bytes of the TCP header can be
4933 			 * expected. Don't access any TCP header fields after
4934 			 * th_seq, an ackskew test is not possible.
4935 			 */
4936 			if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4937 			    pd2.af)) {
4938 				DPFPRINTF(PF_DEBUG_MISC,
4939 				    ("pf: ICMP error message too short "
4940 				    "(tcp)\n"));
4941 				return (PF_DROP);
4942 			}
4943 
4944 			key.af = pd2.af;
4945 			key.proto = IPPROTO_TCP;
4946 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4947 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4948 			key.port[pd2.sidx] = th.th_sport;
4949 			key.port[pd2.didx] = th.th_dport;
4950 
4951 			STATE_LOOKUP(kif, &key, direction, *state, pd);
4952 
4953 			if (direction == (*state)->direction) {
4954 				src = &(*state)->dst;
4955 				dst = &(*state)->src;
4956 			} else {
4957 				src = &(*state)->src;
4958 				dst = &(*state)->dst;
4959 			}
4960 
4961 			if (src->wscale && dst->wscale)
4962 				dws = dst->wscale & PF_WSCALE_MASK;
4963 			else
4964 				dws = 0;
4965 
4966 			/* Demodulate sequence number */
4967 			seq = ntohl(th.th_seq) - src->seqdiff;
4968 			if (src->seqdiff) {
4969 				pf_change_a(&th.th_seq, icmpsum,
4970 				    htonl(seq), 0);
4971 				copyback = 1;
4972 			}
4973 
4974 			if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4975 			    (!SEQ_GEQ(src->seqhi, seq) ||
4976 			    !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4977 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
4978 					printf("pf: BAD ICMP %d:%d ",
4979 					    icmptype, icmpcode);
4980 					pf_print_host(pd->src, 0, pd->af);
4981 					printf(" -> ");
4982 					pf_print_host(pd->dst, 0, pd->af);
4983 					printf(" state: ");
4984 					pf_print_state(*state);
4985 					printf(" seq=%u\n", seq);
4986 				}
4987 				REASON_SET(reason, PFRES_BADSTATE);
4988 				return (PF_DROP);
4989 			} else {
4990 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
4991 					printf("pf: OK ICMP %d:%d ",
4992 					    icmptype, icmpcode);
4993 					pf_print_host(pd->src, 0, pd->af);
4994 					printf(" -> ");
4995 					pf_print_host(pd->dst, 0, pd->af);
4996 					printf(" state: ");
4997 					pf_print_state(*state);
4998 					printf(" seq=%u\n", seq);
4999 				}
5000 			}
5001 
5002 			/* translate source/destination address, if necessary */
5003 			if ((*state)->key[PF_SK_WIRE] !=
5004 			    (*state)->key[PF_SK_STACK]) {
5005 				struct pf_state_key *nk =
5006 				    (*state)->key[pd->didx];
5007 
5008 				if (PF_ANEQ(pd2.src,
5009 				    &nk->addr[pd2.sidx], pd2.af) ||
5010 				    nk->port[pd2.sidx] != th.th_sport)
5011 					pf_change_icmp(pd2.src, &th.th_sport,
5012 					    daddr, &nk->addr[pd2.sidx],
5013 					    nk->port[pd2.sidx], NULL,
5014 					    pd2.ip_sum, icmpsum,
5015 					    pd->ip_sum, 0, pd2.af);
5016 
5017 				if (PF_ANEQ(pd2.dst,
5018 				    &nk->addr[pd2.didx], pd2.af) ||
5019 				    nk->port[pd2.didx] != th.th_dport)
5020 					pf_change_icmp(pd2.dst, &th.th_dport,
5021 					    saddr, &nk->addr[pd2.didx],
5022 					    nk->port[pd2.didx], NULL,
5023 					    pd2.ip_sum, icmpsum,
5024 					    pd->ip_sum, 0, pd2.af);
5025 				copyback = 1;
5026 			}
5027 
5028 			if (copyback) {
5029 				switch (pd2.af) {
5030 #ifdef INET
5031 				case AF_INET:
5032 					m_copyback(m, off, ICMP_MINLEN,
5033 					    (caddr_t )&pd->hdr.icmp);
5034 					m_copyback(m, ipoff2, sizeof(h2),
5035 					    (caddr_t )&h2);
5036 					break;
5037 #endif /* INET */
5038 #ifdef INET6
5039 				case AF_INET6:
5040 					m_copyback(m, off,
5041 					    sizeof(struct icmp6_hdr),
5042 					    (caddr_t )&pd->hdr.icmp6);
5043 					m_copyback(m, ipoff2, sizeof(h2_6),
5044 					    (caddr_t )&h2_6);
5045 					break;
5046 #endif /* INET6 */
5047 				}
5048 				m_copyback(m, off2, 8, (caddr_t)&th);
5049 			}
5050 
5051 			return (PF_PASS);
5052 			break;
5053 		}
5054 		case IPPROTO_UDP: {
5055 			struct udphdr		uh;
5056 
5057 			if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
5058 			    NULL, reason, pd2.af)) {
5059 				DPFPRINTF(PF_DEBUG_MISC,
5060 				    ("pf: ICMP error message too short "
5061 				    "(udp)\n"));
5062 				return (PF_DROP);
5063 			}
5064 
5065 			key.af = pd2.af;
5066 			key.proto = IPPROTO_UDP;
5067 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5068 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5069 			key.port[pd2.sidx] = uh.uh_sport;
5070 			key.port[pd2.didx] = uh.uh_dport;
5071 
5072 			STATE_LOOKUP(kif, &key, direction, *state, pd);
5073 
5074 			/* translate source/destination address, if necessary */
5075 			if ((*state)->key[PF_SK_WIRE] !=
5076 			    (*state)->key[PF_SK_STACK]) {
5077 				struct pf_state_key *nk =
5078 				    (*state)->key[pd->didx];
5079 
5080 				if (PF_ANEQ(pd2.src,
5081 				    &nk->addr[pd2.sidx], pd2.af) ||
5082 				    nk->port[pd2.sidx] != uh.uh_sport)
5083 					pf_change_icmp(pd2.src, &uh.uh_sport,
5084 					    daddr, &nk->addr[pd2.sidx],
5085 					    nk->port[pd2.sidx], &uh.uh_sum,
5086 					    pd2.ip_sum, icmpsum,
5087 					    pd->ip_sum, 1, pd2.af);
5088 
5089 				if (PF_ANEQ(pd2.dst,
5090 				    &nk->addr[pd2.didx], pd2.af) ||
5091 				    nk->port[pd2.didx] != uh.uh_dport)
5092 					pf_change_icmp(pd2.dst, &uh.uh_dport,
5093 					    saddr, &nk->addr[pd2.didx],
5094 					    nk->port[pd2.didx], &uh.uh_sum,
5095 					    pd2.ip_sum, icmpsum,
5096 					    pd->ip_sum, 1, pd2.af);
5097 
5098 				switch (pd2.af) {
5099 #ifdef INET
5100 				case AF_INET:
5101 					m_copyback(m, off, ICMP_MINLEN,
5102 					    (caddr_t )&pd->hdr.icmp);
5103 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5104 					break;
5105 #endif /* INET */
5106 #ifdef INET6
5107 				case AF_INET6:
5108 					m_copyback(m, off,
5109 					    sizeof(struct icmp6_hdr),
5110 					    (caddr_t )&pd->hdr.icmp6);
5111 					m_copyback(m, ipoff2, sizeof(h2_6),
5112 					    (caddr_t )&h2_6);
5113 					break;
5114 #endif /* INET6 */
5115 				}
5116 				m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
5117 			}
5118 			return (PF_PASS);
5119 			break;
5120 		}
5121 #ifdef INET
5122 		case IPPROTO_ICMP: {
5123 			struct icmp		iih;
5124 
5125 			if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
5126 			    NULL, reason, pd2.af)) {
5127 				DPFPRINTF(PF_DEBUG_MISC,
5128 				    ("pf: ICMP error message too short i"
5129 				    "(icmp)\n"));
5130 				return (PF_DROP);
5131 			}
5132 
5133 			key.af = pd2.af;
5134 			key.proto = IPPROTO_ICMP;
5135 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5136 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5137 			key.port[0] = key.port[1] = iih.icmp_id;
5138 
5139 			STATE_LOOKUP(kif, &key, direction, *state, pd);
5140 
5141 			/* translate source/destination address, if necessary */
5142 			if ((*state)->key[PF_SK_WIRE] !=
5143 			    (*state)->key[PF_SK_STACK]) {
5144 				struct pf_state_key *nk =
5145 				    (*state)->key[pd->didx];
5146 
5147 				if (PF_ANEQ(pd2.src,
5148 				    &nk->addr[pd2.sidx], pd2.af) ||
5149 				    nk->port[pd2.sidx] != iih.icmp_id)
5150 					pf_change_icmp(pd2.src, &iih.icmp_id,
5151 					    daddr, &nk->addr[pd2.sidx],
5152 					    nk->port[pd2.sidx], NULL,
5153 					    pd2.ip_sum, icmpsum,
5154 					    pd->ip_sum, 0, AF_INET);
5155 
5156 				if (PF_ANEQ(pd2.dst,
5157 				    &nk->addr[pd2.didx], pd2.af) ||
5158 				    nk->port[pd2.didx] != iih.icmp_id)
5159 					pf_change_icmp(pd2.dst, &iih.icmp_id,
5160 					    saddr, &nk->addr[pd2.didx],
5161 					    nk->port[pd2.didx], NULL,
5162 					    pd2.ip_sum, icmpsum,
5163 					    pd->ip_sum, 0, AF_INET);
5164 
5165 				m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
5166 				m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5167 				m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
5168 			}
5169 			return (PF_PASS);
5170 			break;
5171 		}
5172 #endif /* INET */
5173 #ifdef INET6
5174 		case IPPROTO_ICMPV6: {
5175 			struct icmp6_hdr	iih;
5176 
5177 			if (!pf_pull_hdr(m, off2, &iih,
5178 			    sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
5179 				DPFPRINTF(PF_DEBUG_MISC,
5180 				    ("pf: ICMP error message too short "
5181 				    "(icmp6)\n"));
5182 				return (PF_DROP);
5183 			}
5184 
5185 			key.af = pd2.af;
5186 			key.proto = IPPROTO_ICMPV6;
5187 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5188 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5189 			key.port[0] = key.port[1] = iih.icmp6_id;
5190 
5191 			STATE_LOOKUP(kif, &key, direction, *state, pd);
5192 
5193 			/* translate source/destination address, if necessary */
5194 			if ((*state)->key[PF_SK_WIRE] !=
5195 			    (*state)->key[PF_SK_STACK]) {
5196 				struct pf_state_key *nk =
5197 				    (*state)->key[pd->didx];
5198 
5199 				if (PF_ANEQ(pd2.src,
5200 				    &nk->addr[pd2.sidx], pd2.af) ||
5201 				    nk->port[pd2.sidx] != iih.icmp6_id)
5202 					pf_change_icmp(pd2.src, &iih.icmp6_id,
5203 					    daddr, &nk->addr[pd2.sidx],
5204 					    nk->port[pd2.sidx], NULL,
5205 					    pd2.ip_sum, icmpsum,
5206 					    pd->ip_sum, 0, AF_INET6);
5207 
5208 				if (PF_ANEQ(pd2.dst,
5209 				    &nk->addr[pd2.didx], pd2.af) ||
5210 				    nk->port[pd2.didx] != iih.icmp6_id)
5211 					pf_change_icmp(pd2.dst, &iih.icmp6_id,
5212 					    saddr, &nk->addr[pd2.didx],
5213 					    nk->port[pd2.didx], NULL,
5214 					    pd2.ip_sum, icmpsum,
5215 					    pd->ip_sum, 0, AF_INET6);
5216 
5217 				m_copyback(m, off, sizeof(struct icmp6_hdr),
5218 				    (caddr_t)&pd->hdr.icmp6);
5219 				m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
5220 				m_copyback(m, off2, sizeof(struct icmp6_hdr),
5221 				    (caddr_t)&iih);
5222 			}
5223 			return (PF_PASS);
5224 			break;
5225 		}
5226 #endif /* INET6 */
5227 		default: {
5228 			key.af = pd2.af;
5229 			key.proto = pd2.proto;
5230 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5231 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5232 			key.port[0] = key.port[1] = 0;
5233 
5234 			STATE_LOOKUP(kif, &key, direction, *state, pd);
5235 
5236 			/* translate source/destination address, if necessary */
5237 			if ((*state)->key[PF_SK_WIRE] !=
5238 			    (*state)->key[PF_SK_STACK]) {
5239 				struct pf_state_key *nk =
5240 				    (*state)->key[pd->didx];
5241 
5242 				if (PF_ANEQ(pd2.src,
5243 				    &nk->addr[pd2.sidx], pd2.af))
5244 					pf_change_icmp(pd2.src, NULL, daddr,
5245 					    &nk->addr[pd2.sidx], 0, NULL,
5246 					    pd2.ip_sum, icmpsum,
5247 					    pd->ip_sum, 0, pd2.af);
5248 
5249 				if (PF_ANEQ(pd2.dst,
5250 				    &nk->addr[pd2.didx], pd2.af))
5251 					pf_change_icmp(pd2.dst, NULL, saddr,
5252 					    &nk->addr[pd2.didx], 0, NULL,
5253 					    pd2.ip_sum, icmpsum,
5254 					    pd->ip_sum, 0, pd2.af);
5255 
5256 				switch (pd2.af) {
5257 #ifdef INET
5258 				case AF_INET:
5259 					m_copyback(m, off, ICMP_MINLEN,
5260 					    (caddr_t)&pd->hdr.icmp);
5261 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5262 					break;
5263 #endif /* INET */
5264 #ifdef INET6
5265 				case AF_INET6:
5266 					m_copyback(m, off,
5267 					    sizeof(struct icmp6_hdr),
5268 					    (caddr_t )&pd->hdr.icmp6);
5269 					m_copyback(m, ipoff2, sizeof(h2_6),
5270 					    (caddr_t )&h2_6);
5271 					break;
5272 #endif /* INET6 */
5273 				}
5274 			}
5275 			return (PF_PASS);
5276 			break;
5277 		}
5278 		}
5279 	}
5280 }
5281 
5282 static int
5283 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kkif *kif,
5284     struct mbuf *m, struct pf_pdesc *pd)
5285 {
5286 	struct pf_state_peer	*src, *dst;
5287 	struct pf_state_key_cmp	 key;
5288 
5289 	bzero(&key, sizeof(key));
5290 	key.af = pd->af;
5291 	key.proto = pd->proto;
5292 	if (direction == PF_IN)	{
5293 		PF_ACPY(&key.addr[0], pd->src, key.af);
5294 		PF_ACPY(&key.addr[1], pd->dst, key.af);
5295 		key.port[0] = key.port[1] = 0;
5296 	} else {
5297 		PF_ACPY(&key.addr[1], pd->src, key.af);
5298 		PF_ACPY(&key.addr[0], pd->dst, key.af);
5299 		key.port[1] = key.port[0] = 0;
5300 	}
5301 
5302 	STATE_LOOKUP(kif, &key, direction, *state, pd);
5303 
5304 	if (direction == (*state)->direction) {
5305 		src = &(*state)->src;
5306 		dst = &(*state)->dst;
5307 	} else {
5308 		src = &(*state)->dst;
5309 		dst = &(*state)->src;
5310 	}
5311 
5312 	/* update states */
5313 	if (src->state < PFOTHERS_SINGLE)
5314 		src->state = PFOTHERS_SINGLE;
5315 	if (dst->state == PFOTHERS_SINGLE)
5316 		dst->state = PFOTHERS_MULTIPLE;
5317 
5318 	/* update expire time */
5319 	(*state)->expire = time_uptime;
5320 	if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5321 		(*state)->timeout = PFTM_OTHER_MULTIPLE;
5322 	else
5323 		(*state)->timeout = PFTM_OTHER_SINGLE;
5324 
5325 	/* translate source/destination address, if necessary */
5326 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5327 		struct pf_state_key *nk = (*state)->key[pd->didx];
5328 
5329 		KASSERT(nk, ("%s: nk is null", __func__));
5330 		KASSERT(pd, ("%s: pd is null", __func__));
5331 		KASSERT(pd->src, ("%s: pd->src is null", __func__));
5332 		KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5333 		switch (pd->af) {
5334 #ifdef INET
5335 		case AF_INET:
5336 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5337 				pf_change_a(&pd->src->v4.s_addr,
5338 				    pd->ip_sum,
5339 				    nk->addr[pd->sidx].v4.s_addr,
5340 				    0);
5341 
5342 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5343 				pf_change_a(&pd->dst->v4.s_addr,
5344 				    pd->ip_sum,
5345 				    nk->addr[pd->didx].v4.s_addr,
5346 				    0);
5347 
5348 			break;
5349 #endif /* INET */
5350 #ifdef INET6
5351 		case AF_INET6:
5352 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5353 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5354 
5355 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5356 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5357 #endif /* INET6 */
5358 		}
5359 	}
5360 	return (PF_PASS);
5361 }
5362 
5363 /*
5364  * ipoff and off are measured from the start of the mbuf chain.
5365  * h must be at "ipoff" on the mbuf chain.
5366  */
5367 void *
5368 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5369     u_short *actionp, u_short *reasonp, sa_family_t af)
5370 {
5371 	switch (af) {
5372 #ifdef INET
5373 	case AF_INET: {
5374 		struct ip	*h = mtod(m, struct ip *);
5375 		u_int16_t	 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5376 
5377 		if (fragoff) {
5378 			if (fragoff >= len)
5379 				ACTION_SET(actionp, PF_PASS);
5380 			else {
5381 				ACTION_SET(actionp, PF_DROP);
5382 				REASON_SET(reasonp, PFRES_FRAG);
5383 			}
5384 			return (NULL);
5385 		}
5386 		if (m->m_pkthdr.len < off + len ||
5387 		    ntohs(h->ip_len) < off + len) {
5388 			ACTION_SET(actionp, PF_DROP);
5389 			REASON_SET(reasonp, PFRES_SHORT);
5390 			return (NULL);
5391 		}
5392 		break;
5393 	}
5394 #endif /* INET */
5395 #ifdef INET6
5396 	case AF_INET6: {
5397 		struct ip6_hdr	*h = mtod(m, struct ip6_hdr *);
5398 
5399 		if (m->m_pkthdr.len < off + len ||
5400 		    (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5401 		    (unsigned)(off + len)) {
5402 			ACTION_SET(actionp, PF_DROP);
5403 			REASON_SET(reasonp, PFRES_SHORT);
5404 			return (NULL);
5405 		}
5406 		break;
5407 	}
5408 #endif /* INET6 */
5409 	}
5410 	m_copydata(m, off, len, p);
5411 	return (p);
5412 }
5413 
5414 int
5415 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif,
5416     int rtableid)
5417 {
5418 	struct ifnet		*ifp;
5419 
5420 	/*
5421 	 * Skip check for addresses with embedded interface scope,
5422 	 * as they would always match anyway.
5423 	 */
5424 	if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
5425 		return (1);
5426 
5427 	if (af != AF_INET && af != AF_INET6)
5428 		return (0);
5429 
5430 	/* Skip checks for ipsec interfaces */
5431 	if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5432 		return (1);
5433 
5434 	ifp = (kif != NULL) ? kif->pfik_ifp : NULL;
5435 
5436 	switch (af) {
5437 #ifdef INET6
5438 	case AF_INET6:
5439 		return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE,
5440 		    ifp));
5441 #endif
5442 #ifdef INET
5443 	case AF_INET:
5444 		return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE,
5445 		    ifp));
5446 #endif
5447 	}
5448 
5449 	return (0);
5450 }
5451 
5452 #ifdef INET
5453 static void
5454 pf_route(struct mbuf **m, struct pf_krule *r, int dir, struct ifnet *oifp,
5455     struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp)
5456 {
5457 	struct mbuf		*m0, *m1;
5458 	struct sockaddr_in	dst;
5459 	struct ip		*ip;
5460 	struct ifnet		*ifp = NULL;
5461 	struct pf_addr		 naddr;
5462 	struct pf_ksrc_node	*sn = NULL;
5463 	int			 error = 0;
5464 	uint16_t		 ip_len, ip_off;
5465 
5466 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5467 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5468 	    __func__));
5469 
5470 	if ((pd->pf_mtag == NULL &&
5471 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5472 	    pd->pf_mtag->routed++ > 3) {
5473 		m0 = *m;
5474 		*m = NULL;
5475 		goto bad_locked;
5476 	}
5477 
5478 	if (r->rt == PF_DUPTO) {
5479 		if ((pd->pf_mtag->flags & PF_DUPLICATED)) {
5480 			if (s == NULL) {
5481 				ifp = r->rpool.cur->kif ?
5482 				    r->rpool.cur->kif->pfik_ifp : NULL;
5483 			} else {
5484 				ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5485 				PF_STATE_UNLOCK(s);
5486 			}
5487 			if (ifp == oifp) {
5488 				/* When the 2nd interface is not skipped */
5489 				return;
5490 			} else {
5491 				m0 = *m;
5492 				*m = NULL;
5493 				goto bad;
5494 			}
5495 		} else {
5496 			pd->pf_mtag->flags |= PF_DUPLICATED;
5497 			if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) {
5498 				if (s)
5499 					PF_STATE_UNLOCK(s);
5500 				return;
5501 			}
5502 		}
5503 	} else {
5504 		if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5505 			if (s)
5506 				PF_STATE_UNLOCK(s);
5507 			return;
5508 		}
5509 		m0 = *m;
5510 	}
5511 
5512 	ip = mtod(m0, struct ip *);
5513 
5514 	bzero(&dst, sizeof(dst));
5515 	dst.sin_family = AF_INET;
5516 	dst.sin_len = sizeof(dst);
5517 	dst.sin_addr = ip->ip_dst;
5518 
5519 	bzero(&naddr, sizeof(naddr));
5520 
5521 	if (TAILQ_EMPTY(&r->rpool.list)) {
5522 		DPFPRINTF(PF_DEBUG_URGENT,
5523 		    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5524 		goto bad_locked;
5525 	}
5526 	if (s == NULL) {
5527 		pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5528 		    &naddr, NULL, &sn);
5529 		if (!PF_AZERO(&naddr, AF_INET))
5530 			dst.sin_addr.s_addr = naddr.v4.s_addr;
5531 		ifp = r->rpool.cur->kif ?
5532 		    r->rpool.cur->kif->pfik_ifp : NULL;
5533 	} else {
5534 		if (!PF_AZERO(&s->rt_addr, AF_INET))
5535 			dst.sin_addr.s_addr =
5536 			    s->rt_addr.v4.s_addr;
5537 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5538 		PF_STATE_UNLOCK(s);
5539 	}
5540 	if (ifp == NULL)
5541 		goto bad;
5542 
5543 	if (dir == PF_IN) {
5544 		if (pf_test(PF_OUT, 0, ifp, &m0, inp) != PF_PASS)
5545 			goto bad;
5546 		else if (m0 == NULL)
5547 			goto done;
5548 		if (m0->m_len < sizeof(struct ip)) {
5549 			DPFPRINTF(PF_DEBUG_URGENT,
5550 			    ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5551 			goto bad;
5552 		}
5553 		ip = mtod(m0, struct ip *);
5554 	}
5555 
5556 	if (ifp->if_flags & IFF_LOOPBACK)
5557 		m0->m_flags |= M_SKIP_FIREWALL;
5558 
5559 	ip_len = ntohs(ip->ip_len);
5560 	ip_off = ntohs(ip->ip_off);
5561 
5562 	/* Copied from FreeBSD 10.0-CURRENT ip_output. */
5563 	m0->m_pkthdr.csum_flags |= CSUM_IP;
5564 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5565 		m0 = mb_unmapped_to_ext(m0);
5566 		if (m0 == NULL)
5567 			goto done;
5568 		in_delayed_cksum(m0);
5569 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5570 	}
5571 #if defined(SCTP) || defined(SCTP_SUPPORT)
5572 	if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5573 		m0 = mb_unmapped_to_ext(m0);
5574 		if (m0 == NULL)
5575 			goto done;
5576 		sctp_delayed_cksum(m0, (uint32_t)(ip->ip_hl << 2));
5577 		m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5578 	}
5579 #endif
5580 
5581 	/*
5582 	 * If small enough for interface, or the interface will take
5583 	 * care of the fragmentation for us, we can just send directly.
5584 	 */
5585 	if (ip_len <= ifp->if_mtu ||
5586 	    (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
5587 		ip->ip_sum = 0;
5588 		if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5589 			ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5590 			m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5591 		}
5592 		m_clrprotoflags(m0);	/* Avoid confusing lower layers. */
5593 		error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5594 		goto done;
5595 	}
5596 
5597 	/* Balk when DF bit is set or the interface didn't support TSO. */
5598 	if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5599 		error = EMSGSIZE;
5600 		KMOD_IPSTAT_INC(ips_cantfrag);
5601 		if (r->rt != PF_DUPTO) {
5602 			icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5603 			    ifp->if_mtu);
5604 			goto done;
5605 		} else
5606 			goto bad;
5607 	}
5608 
5609 	error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5610 	if (error)
5611 		goto bad;
5612 
5613 	for (; m0; m0 = m1) {
5614 		m1 = m0->m_nextpkt;
5615 		m0->m_nextpkt = NULL;
5616 		if (error == 0) {
5617 			m_clrprotoflags(m0);
5618 			error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5619 		} else
5620 			m_freem(m0);
5621 	}
5622 
5623 	if (error == 0)
5624 		KMOD_IPSTAT_INC(ips_fragmented);
5625 
5626 done:
5627 	if (r->rt != PF_DUPTO)
5628 		*m = NULL;
5629 	return;
5630 
5631 bad_locked:
5632 	if (s)
5633 		PF_STATE_UNLOCK(s);
5634 bad:
5635 	m_freem(m0);
5636 	goto done;
5637 }
5638 #endif /* INET */
5639 
5640 #ifdef INET6
5641 static void
5642 pf_route6(struct mbuf **m, struct pf_krule *r, int dir, struct ifnet *oifp,
5643     struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp)
5644 {
5645 	struct mbuf		*m0;
5646 	struct sockaddr_in6	dst;
5647 	struct ip6_hdr		*ip6;
5648 	struct ifnet		*ifp = NULL;
5649 	struct pf_addr		 naddr;
5650 	struct pf_ksrc_node	*sn = NULL;
5651 
5652 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5653 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5654 	    __func__));
5655 
5656 	if ((pd->pf_mtag == NULL &&
5657 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5658 	    pd->pf_mtag->routed++ > 3) {
5659 		m0 = *m;
5660 		*m = NULL;
5661 		goto bad_locked;
5662 	}
5663 
5664 	if (r->rt == PF_DUPTO) {
5665 		if ((pd->pf_mtag->flags & PF_DUPLICATED)) {
5666 			if (s == NULL) {
5667 				ifp = r->rpool.cur->kif ?
5668 				    r->rpool.cur->kif->pfik_ifp : NULL;
5669 			} else {
5670 				ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5671 				PF_STATE_UNLOCK(s);
5672 			}
5673 			if (ifp == oifp) {
5674 				/* When the 2nd interface is not skipped */
5675 				return;
5676 			} else {
5677 				m0 = *m;
5678 				*m = NULL;
5679 				goto bad;
5680 			}
5681 		} else {
5682 			pd->pf_mtag->flags |= PF_DUPLICATED;
5683 			if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) {
5684 				if (s)
5685 					PF_STATE_UNLOCK(s);
5686 				return;
5687 			}
5688 		}
5689 	} else {
5690 		if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5691 			if (s)
5692 				PF_STATE_UNLOCK(s);
5693 			return;
5694 		}
5695 		m0 = *m;
5696 	}
5697 
5698 	ip6 = mtod(m0, struct ip6_hdr *);
5699 
5700 	bzero(&dst, sizeof(dst));
5701 	dst.sin6_family = AF_INET6;
5702 	dst.sin6_len = sizeof(dst);
5703 	dst.sin6_addr = ip6->ip6_dst;
5704 
5705 	bzero(&naddr, sizeof(naddr));
5706 
5707 	if (TAILQ_EMPTY(&r->rpool.list)) {
5708 		DPFPRINTF(PF_DEBUG_URGENT,
5709 		    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5710 		goto bad_locked;
5711 	}
5712 	if (s == NULL) {
5713 		pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5714 		    &naddr, NULL, &sn);
5715 		if (!PF_AZERO(&naddr, AF_INET6))
5716 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5717 			    &naddr, AF_INET6);
5718 		ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5719 	} else {
5720 		if (!PF_AZERO(&s->rt_addr, AF_INET6))
5721 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5722 			    &s->rt_addr, AF_INET6);
5723 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5724 	}
5725 
5726 	if (s)
5727 		PF_STATE_UNLOCK(s);
5728 
5729 	if (ifp == NULL)
5730 		goto bad;
5731 
5732 	if (dir == PF_IN) {
5733 		if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp) != PF_PASS)
5734 			goto bad;
5735 		else if (m0 == NULL)
5736 			goto done;
5737 		if (m0->m_len < sizeof(struct ip6_hdr)) {
5738 			DPFPRINTF(PF_DEBUG_URGENT,
5739 			    ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5740 			    __func__));
5741 			goto bad;
5742 		}
5743 		ip6 = mtod(m0, struct ip6_hdr *);
5744 	}
5745 
5746 	if (ifp->if_flags & IFF_LOOPBACK)
5747 		m0->m_flags |= M_SKIP_FIREWALL;
5748 
5749 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
5750 	    ~ifp->if_hwassist) {
5751 		uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
5752 		m0 = mb_unmapped_to_ext(m0);
5753 		if (m0 == NULL)
5754 			goto done;
5755 		in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
5756 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
5757 	}
5758 
5759 	/*
5760 	 * If the packet is too large for the outgoing interface,
5761 	 * send back an icmp6 error.
5762 	 */
5763 	if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5764 		dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5765 	if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5766 		nd6_output_ifp(ifp, ifp, m0, &dst, NULL);
5767 	else {
5768 		in6_ifstat_inc(ifp, ifs6_in_toobig);
5769 		if (r->rt != PF_DUPTO)
5770 			icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5771 		else
5772 			goto bad;
5773 	}
5774 
5775 done:
5776 	if (r->rt != PF_DUPTO)
5777 		*m = NULL;
5778 	return;
5779 
5780 bad_locked:
5781 	if (s)
5782 		PF_STATE_UNLOCK(s);
5783 bad:
5784 	m_freem(m0);
5785 	goto done;
5786 }
5787 #endif /* INET6 */
5788 
5789 /*
5790  * FreeBSD supports cksum offloads for the following drivers.
5791  *  em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4)
5792  *
5793  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5794  *  network driver performed cksum including pseudo header, need to verify
5795  *   csum_data
5796  * CSUM_DATA_VALID :
5797  *  network driver performed cksum, needs to additional pseudo header
5798  *  cksum computation with partial csum_data(i.e. lack of H/W support for
5799  *  pseudo header, for instance sk(4) and possibly gem(4))
5800  *
5801  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5802  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5803  * TCP/UDP layer.
5804  * Also, set csum_data to 0xffff to force cksum validation.
5805  */
5806 static int
5807 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5808 {
5809 	u_int16_t sum = 0;
5810 	int hw_assist = 0;
5811 	struct ip *ip;
5812 
5813 	if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5814 		return (1);
5815 	if (m->m_pkthdr.len < off + len)
5816 		return (1);
5817 
5818 	switch (p) {
5819 	case IPPROTO_TCP:
5820 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5821 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5822 				sum = m->m_pkthdr.csum_data;
5823 			} else {
5824 				ip = mtod(m, struct ip *);
5825 				sum = in_pseudo(ip->ip_src.s_addr,
5826 				ip->ip_dst.s_addr, htonl((u_short)len +
5827 				m->m_pkthdr.csum_data + IPPROTO_TCP));
5828 			}
5829 			sum ^= 0xffff;
5830 			++hw_assist;
5831 		}
5832 		break;
5833 	case IPPROTO_UDP:
5834 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5835 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5836 				sum = m->m_pkthdr.csum_data;
5837 			} else {
5838 				ip = mtod(m, struct ip *);
5839 				sum = in_pseudo(ip->ip_src.s_addr,
5840 				ip->ip_dst.s_addr, htonl((u_short)len +
5841 				m->m_pkthdr.csum_data + IPPROTO_UDP));
5842 			}
5843 			sum ^= 0xffff;
5844 			++hw_assist;
5845 		}
5846 		break;
5847 	case IPPROTO_ICMP:
5848 #ifdef INET6
5849 	case IPPROTO_ICMPV6:
5850 #endif /* INET6 */
5851 		break;
5852 	default:
5853 		return (1);
5854 	}
5855 
5856 	if (!hw_assist) {
5857 		switch (af) {
5858 		case AF_INET:
5859 			if (p == IPPROTO_ICMP) {
5860 				if (m->m_len < off)
5861 					return (1);
5862 				m->m_data += off;
5863 				m->m_len -= off;
5864 				sum = in_cksum(m, len);
5865 				m->m_data -= off;
5866 				m->m_len += off;
5867 			} else {
5868 				if (m->m_len < sizeof(struct ip))
5869 					return (1);
5870 				sum = in4_cksum(m, p, off, len);
5871 			}
5872 			break;
5873 #ifdef INET6
5874 		case AF_INET6:
5875 			if (m->m_len < sizeof(struct ip6_hdr))
5876 				return (1);
5877 			sum = in6_cksum(m, p, off, len);
5878 			break;
5879 #endif /* INET6 */
5880 		default:
5881 			return (1);
5882 		}
5883 	}
5884 	if (sum) {
5885 		switch (p) {
5886 		case IPPROTO_TCP:
5887 		    {
5888 			KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5889 			break;
5890 		    }
5891 		case IPPROTO_UDP:
5892 		    {
5893 			KMOD_UDPSTAT_INC(udps_badsum);
5894 			break;
5895 		    }
5896 #ifdef INET
5897 		case IPPROTO_ICMP:
5898 		    {
5899 			KMOD_ICMPSTAT_INC(icps_checksum);
5900 			break;
5901 		    }
5902 #endif
5903 #ifdef INET6
5904 		case IPPROTO_ICMPV6:
5905 		    {
5906 			KMOD_ICMP6STAT_INC(icp6s_checksum);
5907 			break;
5908 		    }
5909 #endif /* INET6 */
5910 		}
5911 		return (1);
5912 	} else {
5913 		if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5914 			m->m_pkthdr.csum_flags |=
5915 			    (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5916 			m->m_pkthdr.csum_data = 0xffff;
5917 		}
5918 	}
5919 	return (0);
5920 }
5921 
5922 #ifdef INET
5923 int
5924 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5925 {
5926 	struct pfi_kkif		*kif;
5927 	u_short			 action, reason = 0, log = 0;
5928 	struct mbuf		*m = *m0;
5929 	struct ip		*h = NULL;
5930 	struct m_tag		*ipfwtag;
5931 	struct pf_krule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5932 	struct pf_state		*s = NULL;
5933 	struct pf_kruleset	*ruleset = NULL;
5934 	struct pf_pdesc		 pd;
5935 	int			 off, dirndx, pqid = 0;
5936 
5937 	PF_RULES_RLOCK_TRACKER;
5938 
5939 	M_ASSERTPKTHDR(m);
5940 
5941 	if (!V_pf_status.running)
5942 		return (PF_PASS);
5943 
5944 	memset(&pd, 0, sizeof(pd));
5945 
5946 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
5947 
5948 	if (kif == NULL) {
5949 		DPFPRINTF(PF_DEBUG_URGENT,
5950 		    ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5951 		return (PF_DROP);
5952 	}
5953 	if (kif->pfik_flags & PFI_IFLAG_SKIP)
5954 		return (PF_PASS);
5955 
5956 	if (m->m_flags & M_SKIP_FIREWALL)
5957 		return (PF_PASS);
5958 
5959 	pd.pf_mtag = pf_find_mtag(m);
5960 
5961 	PF_RULES_RLOCK();
5962 
5963 	if (ip_divert_ptr != NULL &&
5964 	    ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5965 		struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5966 		if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5967 			if (pd.pf_mtag == NULL &&
5968 			    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5969 				action = PF_DROP;
5970 				goto done;
5971 			}
5972 			pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5973 			m_tag_delete(m, ipfwtag);
5974 		}
5975 		if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5976 			m->m_flags |= M_FASTFWD_OURS;
5977 			pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5978 		}
5979 	} else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5980 		/* We do IP header normalization and packet reassembly here */
5981 		action = PF_DROP;
5982 		goto done;
5983 	}
5984 	m = *m0;	/* pf_normalize messes with m0 */
5985 	h = mtod(m, struct ip *);
5986 
5987 	off = h->ip_hl << 2;
5988 	if (off < (int)sizeof(struct ip)) {
5989 		action = PF_DROP;
5990 		REASON_SET(&reason, PFRES_SHORT);
5991 		log = 1;
5992 		goto done;
5993 	}
5994 
5995 	pd.src = (struct pf_addr *)&h->ip_src;
5996 	pd.dst = (struct pf_addr *)&h->ip_dst;
5997 	pd.sport = pd.dport = NULL;
5998 	pd.ip_sum = &h->ip_sum;
5999 	pd.proto_sum = NULL;
6000 	pd.proto = h->ip_p;
6001 	pd.dir = dir;
6002 	pd.sidx = (dir == PF_IN) ? 0 : 1;
6003 	pd.didx = (dir == PF_IN) ? 1 : 0;
6004 	pd.af = AF_INET;
6005 	pd.tos = h->ip_tos & ~IPTOS_ECN_MASK;
6006 	pd.tot_len = ntohs(h->ip_len);
6007 
6008 	/* handle fragments that didn't get reassembled by normalization */
6009 	if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
6010 		action = pf_test_fragment(&r, dir, kif, m, h,
6011 		    &pd, &a, &ruleset);
6012 		goto done;
6013 	}
6014 
6015 	switch (h->ip_p) {
6016 	case IPPROTO_TCP: {
6017 		if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp),
6018 		    &action, &reason, AF_INET)) {
6019 			log = action != PF_PASS;
6020 			goto done;
6021 		}
6022 		pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2);
6023 		if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0)
6024 			pqid = 1;
6025 		action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6026 		if (action == PF_DROP)
6027 			goto done;
6028 		action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6029 		    &reason);
6030 		if (action == PF_PASS) {
6031 			if (V_pfsync_update_state_ptr != NULL)
6032 				V_pfsync_update_state_ptr(s);
6033 			r = s->rule.ptr;
6034 			a = s->anchor.ptr;
6035 			log = s->log;
6036 		} else if (s == NULL)
6037 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6038 			    &a, &ruleset, inp);
6039 		break;
6040 	}
6041 
6042 	case IPPROTO_UDP: {
6043 		if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp),
6044 		    &action, &reason, AF_INET)) {
6045 			log = action != PF_PASS;
6046 			goto done;
6047 		}
6048 		if (pd.hdr.udp.uh_dport == 0 ||
6049 		    ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off ||
6050 		    ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) {
6051 			action = PF_DROP;
6052 			REASON_SET(&reason, PFRES_SHORT);
6053 			goto done;
6054 		}
6055 		action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6056 		if (action == PF_PASS) {
6057 			if (V_pfsync_update_state_ptr != NULL)
6058 				V_pfsync_update_state_ptr(s);
6059 			r = s->rule.ptr;
6060 			a = s->anchor.ptr;
6061 			log = s->log;
6062 		} else if (s == NULL)
6063 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6064 			    &a, &ruleset, inp);
6065 		break;
6066 	}
6067 
6068 	case IPPROTO_ICMP: {
6069 		if (!pf_pull_hdr(m, off, &pd.hdr.icmp, ICMP_MINLEN,
6070 		    &action, &reason, AF_INET)) {
6071 			log = action != PF_PASS;
6072 			goto done;
6073 		}
6074 		action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
6075 		    &reason);
6076 		if (action == PF_PASS) {
6077 			if (V_pfsync_update_state_ptr != NULL)
6078 				V_pfsync_update_state_ptr(s);
6079 			r = s->rule.ptr;
6080 			a = s->anchor.ptr;
6081 			log = s->log;
6082 		} else if (s == NULL)
6083 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6084 			    &a, &ruleset, inp);
6085 		break;
6086 	}
6087 
6088 #ifdef INET6
6089 	case IPPROTO_ICMPV6: {
6090 		action = PF_DROP;
6091 		DPFPRINTF(PF_DEBUG_MISC,
6092 		    ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
6093 		goto done;
6094 	}
6095 #endif
6096 
6097 	default:
6098 		action = pf_test_state_other(&s, dir, kif, m, &pd);
6099 		if (action == PF_PASS) {
6100 			if (V_pfsync_update_state_ptr != NULL)
6101 				V_pfsync_update_state_ptr(s);
6102 			r = s->rule.ptr;
6103 			a = s->anchor.ptr;
6104 			log = s->log;
6105 		} else if (s == NULL)
6106 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6107 			    &a, &ruleset, inp);
6108 		break;
6109 	}
6110 
6111 done:
6112 	PF_RULES_RUNLOCK();
6113 	if (action == PF_PASS && h->ip_hl > 5 &&
6114 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6115 		action = PF_DROP;
6116 		REASON_SET(&reason, PFRES_IPOPTIONS);
6117 		log = r->log;
6118 		DPFPRINTF(PF_DEBUG_MISC,
6119 		    ("pf: dropping packet with ip options\n"));
6120 	}
6121 
6122 	if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6123 		action = PF_DROP;
6124 		REASON_SET(&reason, PFRES_MEMORY);
6125 	}
6126 	if (r->rtableid >= 0)
6127 		M_SETFIB(m, r->rtableid);
6128 
6129 	if (r->scrub_flags & PFSTATE_SETPRIO) {
6130 		if (pd.tos & IPTOS_LOWDELAY)
6131 			pqid = 1;
6132 		if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6133 			action = PF_DROP;
6134 			REASON_SET(&reason, PFRES_MEMORY);
6135 			log = 1;
6136 			DPFPRINTF(PF_DEBUG_MISC,
6137 			    ("pf: failed to allocate 802.1q mtag\n"));
6138 		}
6139 	}
6140 
6141 #ifdef ALTQ
6142 	if (action == PF_PASS && r->qid) {
6143 		if (pd.pf_mtag == NULL &&
6144 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6145 			action = PF_DROP;
6146 			REASON_SET(&reason, PFRES_MEMORY);
6147 		} else {
6148 			if (s != NULL)
6149 				pd.pf_mtag->qid_hash = pf_state_hash(s);
6150 			if (pqid || (pd.tos & IPTOS_LOWDELAY))
6151 				pd.pf_mtag->qid = r->pqid;
6152 			else
6153 				pd.pf_mtag->qid = r->qid;
6154 			/* Add hints for ecn. */
6155 			pd.pf_mtag->hdr = h;
6156 		}
6157 	}
6158 #endif /* ALTQ */
6159 
6160 	/*
6161 	 * connections redirected to loopback should not match sockets
6162 	 * bound specifically to loopback due to security implications,
6163 	 * see tcp_input() and in_pcblookup_listen().
6164 	 */
6165 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6166 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6167 	    (s->nat_rule.ptr->action == PF_RDR ||
6168 	    s->nat_rule.ptr->action == PF_BINAT) &&
6169 	    IN_LOOPBACK(ntohl(pd.dst->v4.s_addr)))
6170 		m->m_flags |= M_SKIP_FIREWALL;
6171 
6172 	if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
6173 	    !PACKET_LOOPED(&pd)) {
6174 		ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
6175 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
6176 		if (ipfwtag != NULL) {
6177 			((struct ipfw_rule_ref *)(ipfwtag+1))->info =
6178 			    ntohs(r->divert.port);
6179 			((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
6180 
6181 			if (s)
6182 				PF_STATE_UNLOCK(s);
6183 
6184 			m_tag_prepend(m, ipfwtag);
6185 			if (m->m_flags & M_FASTFWD_OURS) {
6186 				if (pd.pf_mtag == NULL &&
6187 				    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6188 					action = PF_DROP;
6189 					REASON_SET(&reason, PFRES_MEMORY);
6190 					log = 1;
6191 					DPFPRINTF(PF_DEBUG_MISC,
6192 					    ("pf: failed to allocate tag\n"));
6193 				} else {
6194 					pd.pf_mtag->flags |=
6195 					    PF_FASTFWD_OURS_PRESENT;
6196 					m->m_flags &= ~M_FASTFWD_OURS;
6197 				}
6198 			}
6199 			ip_divert_ptr(*m0, dir == PF_IN);
6200 			*m0 = NULL;
6201 
6202 			return (action);
6203 		} else {
6204 			/* XXX: ipfw has the same behaviour! */
6205 			action = PF_DROP;
6206 			REASON_SET(&reason, PFRES_MEMORY);
6207 			log = 1;
6208 			DPFPRINTF(PF_DEBUG_MISC,
6209 			    ("pf: failed to allocate divert tag\n"));
6210 		}
6211 	}
6212 
6213 	if (log) {
6214 		struct pf_krule *lr;
6215 
6216 		if (s != NULL && s->nat_rule.ptr != NULL &&
6217 		    s->nat_rule.ptr->log & PF_LOG_ALL)
6218 			lr = s->nat_rule.ptr;
6219 		else
6220 			lr = r;
6221 		PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
6222 		    (s == NULL));
6223 	}
6224 
6225 	counter_u64_add(kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS],
6226 	    pd.tot_len);
6227 	counter_u64_add(kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS],
6228 	    1);
6229 
6230 	if (action == PF_PASS || r->action == PF_DROP) {
6231 		dirndx = (dir == PF_OUT);
6232 		counter_u64_add(r->packets[dirndx], 1);
6233 		counter_u64_add(r->bytes[dirndx], pd.tot_len);
6234 		if (a != NULL) {
6235 			counter_u64_add(a->packets[dirndx], 1);
6236 			counter_u64_add(a->bytes[dirndx], pd.tot_len);
6237 		}
6238 		if (s != NULL) {
6239 			if (s->nat_rule.ptr != NULL) {
6240 				counter_u64_add(s->nat_rule.ptr->packets[dirndx],
6241 				    1);
6242 				counter_u64_add(s->nat_rule.ptr->bytes[dirndx],
6243 				    pd.tot_len);
6244 			}
6245 			if (s->src_node != NULL) {
6246 				counter_u64_add(s->src_node->packets[dirndx],
6247 				    1);
6248 				counter_u64_add(s->src_node->bytes[dirndx],
6249 				    pd.tot_len);
6250 			}
6251 			if (s->nat_src_node != NULL) {
6252 				counter_u64_add(s->nat_src_node->packets[dirndx],
6253 				    1);
6254 				counter_u64_add(s->nat_src_node->bytes[dirndx],
6255 				    pd.tot_len);
6256 			}
6257 			dirndx = (dir == s->direction) ? 0 : 1;
6258 			s->packets[dirndx]++;
6259 			s->bytes[dirndx] += pd.tot_len;
6260 		}
6261 		tr = r;
6262 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6263 		if (nr != NULL && r == &V_pf_default_rule)
6264 			tr = nr;
6265 		if (tr->src.addr.type == PF_ADDR_TABLE)
6266 			pfr_update_stats(tr->src.addr.p.tbl,
6267 			    (s == NULL) ? pd.src :
6268 			    &s->key[(s->direction == PF_IN)]->
6269 				addr[(s->direction == PF_OUT)],
6270 			    pd.af, pd.tot_len, dir == PF_OUT,
6271 			    r->action == PF_PASS, tr->src.neg);
6272 		if (tr->dst.addr.type == PF_ADDR_TABLE)
6273 			pfr_update_stats(tr->dst.addr.p.tbl,
6274 			    (s == NULL) ? pd.dst :
6275 			    &s->key[(s->direction == PF_IN)]->
6276 				addr[(s->direction == PF_IN)],
6277 			    pd.af, pd.tot_len, dir == PF_OUT,
6278 			    r->action == PF_PASS, tr->dst.neg);
6279 	}
6280 
6281 	switch (action) {
6282 	case PF_SYNPROXY_DROP:
6283 		m_freem(*m0);
6284 	case PF_DEFER:
6285 		*m0 = NULL;
6286 		action = PF_PASS;
6287 		break;
6288 	case PF_DROP:
6289 		m_freem(*m0);
6290 		*m0 = NULL;
6291 		break;
6292 	default:
6293 		/* pf_route() returns unlocked. */
6294 		if (r->rt) {
6295 			pf_route(m0, r, dir, kif->pfik_ifp, s, &pd, inp);
6296 			return (action);
6297 		}
6298 		break;
6299 	}
6300 
6301 	SDT_PROBE4(pf, ip, test, done, action, reason, r, s);
6302 
6303 	if (s)
6304 		PF_STATE_UNLOCK(s);
6305 
6306 	return (action);
6307 }
6308 #endif /* INET */
6309 
6310 #ifdef INET6
6311 int
6312 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6313 {
6314 	struct pfi_kkif		*kif;
6315 	u_short			 action, reason = 0, log = 0;
6316 	struct mbuf		*m = *m0, *n = NULL;
6317 	struct m_tag		*mtag;
6318 	struct ip6_hdr		*h = NULL;
6319 	struct pf_krule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6320 	struct pf_state		*s = NULL;
6321 	struct pf_kruleset	*ruleset = NULL;
6322 	struct pf_pdesc		 pd;
6323 	int			 off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0;
6324 
6325 	PF_RULES_RLOCK_TRACKER;
6326 	M_ASSERTPKTHDR(m);
6327 
6328 	if (!V_pf_status.running)
6329 		return (PF_PASS);
6330 
6331 	memset(&pd, 0, sizeof(pd));
6332 	pd.pf_mtag = pf_find_mtag(m);
6333 
6334 	if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6335 		return (PF_PASS);
6336 
6337 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
6338 	if (kif == NULL) {
6339 		DPFPRINTF(PF_DEBUG_URGENT,
6340 		    ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6341 		return (PF_DROP);
6342 	}
6343 	if (kif->pfik_flags & PFI_IFLAG_SKIP)
6344 		return (PF_PASS);
6345 
6346 	if (m->m_flags & M_SKIP_FIREWALL)
6347 		return (PF_PASS);
6348 
6349 	PF_RULES_RLOCK();
6350 
6351 	/* We do IP header normalization and packet reassembly here */
6352 	if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6353 		action = PF_DROP;
6354 		goto done;
6355 	}
6356 	m = *m0;	/* pf_normalize messes with m0 */
6357 	h = mtod(m, struct ip6_hdr *);
6358 
6359 	/*
6360 	 * we do not support jumbogram.  if we keep going, zero ip6_plen
6361 	 * will do something bad, so drop the packet for now.
6362 	 */
6363 	if (htons(h->ip6_plen) == 0) {
6364 		action = PF_DROP;
6365 		REASON_SET(&reason, PFRES_NORM);	/*XXX*/
6366 		goto done;
6367 	}
6368 
6369 	pd.src = (struct pf_addr *)&h->ip6_src;
6370 	pd.dst = (struct pf_addr *)&h->ip6_dst;
6371 	pd.sport = pd.dport = NULL;
6372 	pd.ip_sum = NULL;
6373 	pd.proto_sum = NULL;
6374 	pd.dir = dir;
6375 	pd.sidx = (dir == PF_IN) ? 0 : 1;
6376 	pd.didx = (dir == PF_IN) ? 1 : 0;
6377 	pd.af = AF_INET6;
6378 	pd.tos = IPV6_DSCP(h);
6379 	pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6380 
6381 	off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6382 	pd.proto = h->ip6_nxt;
6383 	do {
6384 		switch (pd.proto) {
6385 		case IPPROTO_FRAGMENT:
6386 			action = pf_test_fragment(&r, dir, kif, m, h,
6387 			    &pd, &a, &ruleset);
6388 			if (action == PF_DROP)
6389 				REASON_SET(&reason, PFRES_FRAG);
6390 			goto done;
6391 		case IPPROTO_ROUTING: {
6392 			struct ip6_rthdr rthdr;
6393 
6394 			if (rh_cnt++) {
6395 				DPFPRINTF(PF_DEBUG_MISC,
6396 				    ("pf: IPv6 more than one rthdr\n"));
6397 				action = PF_DROP;
6398 				REASON_SET(&reason, PFRES_IPOPTIONS);
6399 				log = 1;
6400 				goto done;
6401 			}
6402 			if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6403 			    &reason, pd.af)) {
6404 				DPFPRINTF(PF_DEBUG_MISC,
6405 				    ("pf: IPv6 short rthdr\n"));
6406 				action = PF_DROP;
6407 				REASON_SET(&reason, PFRES_SHORT);
6408 				log = 1;
6409 				goto done;
6410 			}
6411 			if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6412 				DPFPRINTF(PF_DEBUG_MISC,
6413 				    ("pf: IPv6 rthdr0\n"));
6414 				action = PF_DROP;
6415 				REASON_SET(&reason, PFRES_IPOPTIONS);
6416 				log = 1;
6417 				goto done;
6418 			}
6419 			/* FALLTHROUGH */
6420 		}
6421 		case IPPROTO_AH:
6422 		case IPPROTO_HOPOPTS:
6423 		case IPPROTO_DSTOPTS: {
6424 			/* get next header and header length */
6425 			struct ip6_ext	opt6;
6426 
6427 			if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6428 			    NULL, &reason, pd.af)) {
6429 				DPFPRINTF(PF_DEBUG_MISC,
6430 				    ("pf: IPv6 short opt\n"));
6431 				action = PF_DROP;
6432 				log = 1;
6433 				goto done;
6434 			}
6435 			if (pd.proto == IPPROTO_AH)
6436 				off += (opt6.ip6e_len + 2) * 4;
6437 			else
6438 				off += (opt6.ip6e_len + 1) * 8;
6439 			pd.proto = opt6.ip6e_nxt;
6440 			/* goto the next header */
6441 			break;
6442 		}
6443 		default:
6444 			terminal++;
6445 			break;
6446 		}
6447 	} while (!terminal);
6448 
6449 	/* if there's no routing header, use unmodified mbuf for checksumming */
6450 	if (!n)
6451 		n = m;
6452 
6453 	switch (pd.proto) {
6454 	case IPPROTO_TCP: {
6455 		if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp),
6456 		    &action, &reason, AF_INET6)) {
6457 			log = action != PF_PASS;
6458 			goto done;
6459 		}
6460 		pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2);
6461 		action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6462 		if (action == PF_DROP)
6463 			goto done;
6464 		action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6465 		    &reason);
6466 		if (action == PF_PASS) {
6467 			if (V_pfsync_update_state_ptr != NULL)
6468 				V_pfsync_update_state_ptr(s);
6469 			r = s->rule.ptr;
6470 			a = s->anchor.ptr;
6471 			log = s->log;
6472 		} else if (s == NULL)
6473 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6474 			    &a, &ruleset, inp);
6475 		break;
6476 	}
6477 
6478 	case IPPROTO_UDP: {
6479 		if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp),
6480 		    &action, &reason, AF_INET6)) {
6481 			log = action != PF_PASS;
6482 			goto done;
6483 		}
6484 		if (pd.hdr.udp.uh_dport == 0 ||
6485 		    ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off ||
6486 		    ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) {
6487 			action = PF_DROP;
6488 			REASON_SET(&reason, PFRES_SHORT);
6489 			goto done;
6490 		}
6491 		action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6492 		if (action == PF_PASS) {
6493 			if (V_pfsync_update_state_ptr != NULL)
6494 				V_pfsync_update_state_ptr(s);
6495 			r = s->rule.ptr;
6496 			a = s->anchor.ptr;
6497 			log = s->log;
6498 		} else if (s == NULL)
6499 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6500 			    &a, &ruleset, inp);
6501 		break;
6502 	}
6503 
6504 	case IPPROTO_ICMP: {
6505 		action = PF_DROP;
6506 		DPFPRINTF(PF_DEBUG_MISC,
6507 		    ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6508 		goto done;
6509 	}
6510 
6511 	case IPPROTO_ICMPV6: {
6512 		if (!pf_pull_hdr(m, off, &pd.hdr.icmp6, sizeof(pd.hdr.icmp6),
6513 		    &action, &reason, AF_INET6)) {
6514 			log = action != PF_PASS;
6515 			goto done;
6516 		}
6517 		action = pf_test_state_icmp(&s, dir, kif,
6518 		    m, off, h, &pd, &reason);
6519 		if (action == PF_PASS) {
6520 			if (V_pfsync_update_state_ptr != NULL)
6521 				V_pfsync_update_state_ptr(s);
6522 			r = s->rule.ptr;
6523 			a = s->anchor.ptr;
6524 			log = s->log;
6525 		} else if (s == NULL)
6526 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6527 			    &a, &ruleset, inp);
6528 		break;
6529 	}
6530 
6531 	default:
6532 		action = pf_test_state_other(&s, dir, kif, m, &pd);
6533 		if (action == PF_PASS) {
6534 			if (V_pfsync_update_state_ptr != NULL)
6535 				V_pfsync_update_state_ptr(s);
6536 			r = s->rule.ptr;
6537 			a = s->anchor.ptr;
6538 			log = s->log;
6539 		} else if (s == NULL)
6540 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6541 			    &a, &ruleset, inp);
6542 		break;
6543 	}
6544 
6545 done:
6546 	PF_RULES_RUNLOCK();
6547 	if (n != m) {
6548 		m_freem(n);
6549 		n = NULL;
6550 	}
6551 
6552 	/* handle dangerous IPv6 extension headers. */
6553 	if (action == PF_PASS && rh_cnt &&
6554 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6555 		action = PF_DROP;
6556 		REASON_SET(&reason, PFRES_IPOPTIONS);
6557 		log = r->log;
6558 		DPFPRINTF(PF_DEBUG_MISC,
6559 		    ("pf: dropping packet with dangerous v6 headers\n"));
6560 	}
6561 
6562 	if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6563 		action = PF_DROP;
6564 		REASON_SET(&reason, PFRES_MEMORY);
6565 	}
6566 	if (r->rtableid >= 0)
6567 		M_SETFIB(m, r->rtableid);
6568 
6569 	if (r->scrub_flags & PFSTATE_SETPRIO) {
6570 		if (pd.tos & IPTOS_LOWDELAY)
6571 			pqid = 1;
6572 		if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6573 			action = PF_DROP;
6574 			REASON_SET(&reason, PFRES_MEMORY);
6575 			log = 1;
6576 			DPFPRINTF(PF_DEBUG_MISC,
6577 			    ("pf: failed to allocate 802.1q mtag\n"));
6578 		}
6579 	}
6580 
6581 #ifdef ALTQ
6582 	if (action == PF_PASS && r->qid) {
6583 		if (pd.pf_mtag == NULL &&
6584 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6585 			action = PF_DROP;
6586 			REASON_SET(&reason, PFRES_MEMORY);
6587 		} else {
6588 			if (s != NULL)
6589 				pd.pf_mtag->qid_hash = pf_state_hash(s);
6590 			if (pd.tos & IPTOS_LOWDELAY)
6591 				pd.pf_mtag->qid = r->pqid;
6592 			else
6593 				pd.pf_mtag->qid = r->qid;
6594 			/* Add hints for ecn. */
6595 			pd.pf_mtag->hdr = h;
6596 		}
6597 	}
6598 #endif /* ALTQ */
6599 
6600 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6601 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6602 	    (s->nat_rule.ptr->action == PF_RDR ||
6603 	    s->nat_rule.ptr->action == PF_BINAT) &&
6604 	    IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6605 		m->m_flags |= M_SKIP_FIREWALL;
6606 
6607 	/* XXX: Anybody working on it?! */
6608 	if (r->divert.port)
6609 		printf("pf: divert(9) is not supported for IPv6\n");
6610 
6611 	if (log) {
6612 		struct pf_krule *lr;
6613 
6614 		if (s != NULL && s->nat_rule.ptr != NULL &&
6615 		    s->nat_rule.ptr->log & PF_LOG_ALL)
6616 			lr = s->nat_rule.ptr;
6617 		else
6618 			lr = r;
6619 		PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6620 		    &pd, (s == NULL));
6621 	}
6622 
6623 	counter_u64_add(kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS],
6624 	    pd.tot_len);
6625 	counter_u64_add(kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS],
6626 	    1);
6627 
6628 	if (action == PF_PASS || r->action == PF_DROP) {
6629 		dirndx = (dir == PF_OUT);
6630 		counter_u64_add(r->packets[dirndx], 1);
6631 		counter_u64_add(r->bytes[dirndx], pd.tot_len);
6632 		if (a != NULL) {
6633 			counter_u64_add(a->packets[dirndx], 1);
6634 			counter_u64_add(a->bytes[dirndx], pd.tot_len);
6635 		}
6636 		if (s != NULL) {
6637 			if (s->nat_rule.ptr != NULL) {
6638 				counter_u64_add(s->nat_rule.ptr->packets[dirndx],
6639 				    1);
6640 				counter_u64_add(s->nat_rule.ptr->bytes[dirndx],
6641 				    pd.tot_len);
6642 			}
6643 			if (s->src_node != NULL) {
6644 				counter_u64_add(s->src_node->packets[dirndx],
6645 				    1);
6646 				counter_u64_add(s->src_node->bytes[dirndx],
6647 				    pd.tot_len);
6648 			}
6649 			if (s->nat_src_node != NULL) {
6650 				counter_u64_add(s->nat_src_node->packets[dirndx],
6651 				    1);
6652 				counter_u64_add(s->nat_src_node->bytes[dirndx],
6653 				    pd.tot_len);
6654 			}
6655 			dirndx = (dir == s->direction) ? 0 : 1;
6656 			s->packets[dirndx]++;
6657 			s->bytes[dirndx] += pd.tot_len;
6658 		}
6659 		tr = r;
6660 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6661 		if (nr != NULL && r == &V_pf_default_rule)
6662 			tr = nr;
6663 		if (tr->src.addr.type == PF_ADDR_TABLE)
6664 			pfr_update_stats(tr->src.addr.p.tbl,
6665 			    (s == NULL) ? pd.src :
6666 			    &s->key[(s->direction == PF_IN)]->addr[0],
6667 			    pd.af, pd.tot_len, dir == PF_OUT,
6668 			    r->action == PF_PASS, tr->src.neg);
6669 		if (tr->dst.addr.type == PF_ADDR_TABLE)
6670 			pfr_update_stats(tr->dst.addr.p.tbl,
6671 			    (s == NULL) ? pd.dst :
6672 			    &s->key[(s->direction == PF_IN)]->addr[1],
6673 			    pd.af, pd.tot_len, dir == PF_OUT,
6674 			    r->action == PF_PASS, tr->dst.neg);
6675 	}
6676 
6677 	switch (action) {
6678 	case PF_SYNPROXY_DROP:
6679 		m_freem(*m0);
6680 	case PF_DEFER:
6681 		*m0 = NULL;
6682 		action = PF_PASS;
6683 		break;
6684 	case PF_DROP:
6685 		m_freem(*m0);
6686 		*m0 = NULL;
6687 		break;
6688 	default:
6689 		/* pf_route6() returns unlocked. */
6690 		if (r->rt) {
6691 			pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd, inp);
6692 			return (action);
6693 		}
6694 		break;
6695 	}
6696 
6697 	if (s)
6698 		PF_STATE_UNLOCK(s);
6699 
6700 	/* If reassembled packet passed, create new fragments. */
6701 	if (action == PF_PASS && *m0 && (pflags & PFIL_FWD) &&
6702 	    (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL)
6703 		action = pf_refragment6(ifp, m0, mtag);
6704 
6705 	SDT_PROBE4(pf, ip, test6, done, action, reason, r, s);
6706 
6707 	return (action);
6708 }
6709 #endif /* INET6 */
6710