xref: /freebsd/sys/netpfil/pf/pf.c (revision d59a76183470685bdf0b88013d2baad1f04f030f)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2001 Daniel Hartmeier
5  * Copyright (c) 2002 - 2008 Henning Brauer
6  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  *
13  *    - Redistributions of source code must retain the above copyright
14  *      notice, this list of conditions and the following disclaimer.
15  *    - Redistributions in binary form must reproduce the above
16  *      copyright notice, this list of conditions and the following
17  *      disclaimer in the documentation and/or other materials provided
18  *      with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
28  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
30  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31  * POSSIBILITY OF SUCH DAMAGE.
32  *
33  * Effort sponsored in part by the Defense Advanced Research Projects
34  * Agency (DARPA) and Air Force Research Laboratory, Air Force
35  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
36  *
37  *	$OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
38  */
39 
40 #include <sys/cdefs.h>
41 #include "opt_bpf.h"
42 #include "opt_inet.h"
43 #include "opt_inet6.h"
44 #include "opt_pf.h"
45 #include "opt_sctp.h"
46 
47 #include <sys/param.h>
48 #include <sys/bus.h>
49 #include <sys/endian.h>
50 #include <sys/gsb_crc32.h>
51 #include <sys/hash.h>
52 #include <sys/interrupt.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/limits.h>
56 #include <sys/mbuf.h>
57 #include <sys/md5.h>
58 #include <sys/random.h>
59 #include <sys/refcount.h>
60 #include <sys/sdt.h>
61 #include <sys/socket.h>
62 #include <sys/sysctl.h>
63 #include <sys/taskqueue.h>
64 #include <sys/ucred.h>
65 
66 #include <net/if.h>
67 #include <net/if_var.h>
68 #include <net/if_private.h>
69 #include <net/if_types.h>
70 #include <net/if_vlan_var.h>
71 #include <net/route.h>
72 #include <net/route/nhop.h>
73 #include <net/vnet.h>
74 
75 #include <net/pfil.h>
76 #include <net/pfvar.h>
77 #include <net/if_pflog.h>
78 #include <net/if_pfsync.h>
79 
80 #include <netinet/in_pcb.h>
81 #include <netinet/in_var.h>
82 #include <netinet/in_fib.h>
83 #include <netinet/ip.h>
84 #include <netinet/ip_fw.h>
85 #include <netinet/ip_icmp.h>
86 #include <netinet/icmp_var.h>
87 #include <netinet/ip_var.h>
88 #include <netinet/tcp.h>
89 #include <netinet/tcp_fsm.h>
90 #include <netinet/tcp_seq.h>
91 #include <netinet/tcp_timer.h>
92 #include <netinet/tcp_var.h>
93 #include <netinet/udp.h>
94 #include <netinet/udp_var.h>
95 
96 /* dummynet */
97 #include <netinet/ip_dummynet.h>
98 #include <netinet/ip_fw.h>
99 #include <netpfil/ipfw/dn_heap.h>
100 #include <netpfil/ipfw/ip_fw_private.h>
101 #include <netpfil/ipfw/ip_dn_private.h>
102 
103 #ifdef INET6
104 #include <netinet/ip6.h>
105 #include <netinet/icmp6.h>
106 #include <netinet6/nd6.h>
107 #include <netinet6/ip6_var.h>
108 #include <netinet6/in6_pcb.h>
109 #include <netinet6/in6_fib.h>
110 #include <netinet6/scope6_var.h>
111 #endif /* INET6 */
112 
113 #include <netinet/sctp_header.h>
114 #include <netinet/sctp_crc32.h>
115 
116 #include <machine/in_cksum.h>
117 #include <security/mac/mac_framework.h>
118 
119 #define	DPFPRINTF(n, x)	if (V_pf_status.debug >= (n)) printf x
120 
121 SDT_PROVIDER_DEFINE(pf);
122 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *",
123     "struct pf_kstate *");
124 SDT_PROBE_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *",
125     "struct pf_kstate *");
126 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *",
127     "struct pf_state_key_cmp *", "int", "struct pf_pdesc *",
128     "struct pf_kstate *");
129 SDT_PROBE_DEFINE2(pf, ip, , bound_iface, "struct pf_kstate *",
130     "struct pfi_kkif *");
131 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *",
132     "struct pf_krule *", "struct mbuf *", "int");
133 SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t",
134     "struct pf_sctp_source *");
135 SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t",
136     "struct pf_kstate *", "struct pf_sctp_source *");
137 
138 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *",
139     "struct mbuf *");
140 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *");
141 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch,
142     "int", "struct pf_keth_rule *", "char *");
143 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *");
144 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match,
145     "int", "struct pf_keth_rule *");
146 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t");
147 
148 /*
149  * Global variables
150  */
151 
152 /* state tables */
153 VNET_DEFINE(struct pf_altqqueue,	 pf_altqs[4]);
154 VNET_DEFINE(struct pf_kpalist,		 pf_pabuf);
155 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_active);
156 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_active);
157 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_inactive);
158 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_inactive);
159 VNET_DEFINE(struct pf_kstatus,		 pf_status);
160 
161 VNET_DEFINE(u_int32_t,			 ticket_altqs_active);
162 VNET_DEFINE(u_int32_t,			 ticket_altqs_inactive);
163 VNET_DEFINE(int,			 altqs_inactive_open);
164 VNET_DEFINE(u_int32_t,			 ticket_pabuf);
165 
166 VNET_DEFINE(MD5_CTX,			 pf_tcp_secret_ctx);
167 #define	V_pf_tcp_secret_ctx		 VNET(pf_tcp_secret_ctx)
168 VNET_DEFINE(u_char,			 pf_tcp_secret[16]);
169 #define	V_pf_tcp_secret			 VNET(pf_tcp_secret)
170 VNET_DEFINE(int,			 pf_tcp_secret_init);
171 #define	V_pf_tcp_secret_init		 VNET(pf_tcp_secret_init)
172 VNET_DEFINE(int,			 pf_tcp_iss_off);
173 #define	V_pf_tcp_iss_off		 VNET(pf_tcp_iss_off)
174 VNET_DECLARE(int,			 pf_vnet_active);
175 #define	V_pf_vnet_active		 VNET(pf_vnet_active)
176 
177 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx);
178 #define V_pf_purge_idx	VNET(pf_purge_idx)
179 
180 #ifdef PF_WANT_32_TO_64_COUNTER
181 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter);
182 #define	V_pf_counter_periodic_iter	VNET(pf_counter_periodic_iter)
183 
184 VNET_DEFINE(struct allrulelist_head, pf_allrulelist);
185 VNET_DEFINE(size_t, pf_allrulecount);
186 VNET_DEFINE(struct pf_krule *, pf_rulemarker);
187 #endif
188 
189 struct pf_sctp_endpoint;
190 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint);
191 struct pf_sctp_source {
192 	sa_family_t			af;
193 	struct pf_addr			addr;
194 	TAILQ_ENTRY(pf_sctp_source)	entry;
195 };
196 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source);
197 struct pf_sctp_endpoint
198 {
199 	uint32_t		 v_tag;
200 	struct pf_sctp_sources	 sources;
201 	RB_ENTRY(pf_sctp_endpoint)	entry;
202 };
203 static int
204 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b)
205 {
206 	return (a->v_tag - b->v_tag);
207 }
208 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare);
209 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare);
210 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints);
211 #define V_pf_sctp_endpoints	VNET(pf_sctp_endpoints)
212 static struct mtx_padalign pf_sctp_endpoints_mtx;
213 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF);
214 #define	PF_SCTP_ENDPOINTS_LOCK()	mtx_lock(&pf_sctp_endpoints_mtx)
215 #define	PF_SCTP_ENDPOINTS_UNLOCK()	mtx_unlock(&pf_sctp_endpoints_mtx)
216 
217 /*
218  * Queue for pf_intr() sends.
219  */
220 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
221 struct pf_send_entry {
222 	STAILQ_ENTRY(pf_send_entry)	pfse_next;
223 	struct mbuf			*pfse_m;
224 	enum {
225 		PFSE_IP,
226 		PFSE_IP6,
227 		PFSE_ICMP,
228 		PFSE_ICMP6,
229 	}				pfse_type;
230 	struct {
231 		int		type;
232 		int		code;
233 		int		mtu;
234 	} icmpopts;
235 };
236 
237 STAILQ_HEAD(pf_send_head, pf_send_entry);
238 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue);
239 #define	V_pf_sendqueue	VNET(pf_sendqueue)
240 
241 static struct mtx_padalign pf_sendqueue_mtx;
242 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
243 #define	PF_SENDQ_LOCK()		mtx_lock(&pf_sendqueue_mtx)
244 #define	PF_SENDQ_UNLOCK()	mtx_unlock(&pf_sendqueue_mtx)
245 
246 /*
247  * Queue for pf_overload_task() tasks.
248  */
249 struct pf_overload_entry {
250 	SLIST_ENTRY(pf_overload_entry)	next;
251 	struct pf_addr  		addr;
252 	sa_family_t			af;
253 	uint8_t				dir;
254 	struct pf_krule  		*rule;
255 };
256 
257 SLIST_HEAD(pf_overload_head, pf_overload_entry);
258 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue);
259 #define V_pf_overloadqueue	VNET(pf_overloadqueue)
260 VNET_DEFINE_STATIC(struct task, pf_overloadtask);
261 #define	V_pf_overloadtask	VNET(pf_overloadtask)
262 
263 static struct mtx_padalign pf_overloadqueue_mtx;
264 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
265     "pf overload/flush queue", MTX_DEF);
266 #define	PF_OVERLOADQ_LOCK()	mtx_lock(&pf_overloadqueue_mtx)
267 #define	PF_OVERLOADQ_UNLOCK()	mtx_unlock(&pf_overloadqueue_mtx)
268 
269 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules);
270 struct mtx_padalign pf_unlnkdrules_mtx;
271 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
272     MTX_DEF);
273 
274 struct sx pf_config_lock;
275 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config");
276 
277 struct mtx_padalign pf_table_stats_lock;
278 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats",
279     MTX_DEF);
280 
281 VNET_DEFINE_STATIC(uma_zone_t,	pf_sources_z);
282 #define	V_pf_sources_z	VNET(pf_sources_z)
283 uma_zone_t		pf_mtag_z;
284 VNET_DEFINE(uma_zone_t,	 pf_state_z);
285 VNET_DEFINE(uma_zone_t,	 pf_state_key_z);
286 
287 VNET_DEFINE(struct unrhdr64, pf_stateid);
288 
289 static void		 pf_src_tree_remove_state(struct pf_kstate *);
290 static void		 pf_init_threshold(struct pf_threshold *, u_int32_t,
291 			    u_int32_t);
292 static void		 pf_add_threshold(struct pf_threshold *);
293 static int		 pf_check_threshold(struct pf_threshold *);
294 
295 static void		 pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
296 			    u_int16_t *, u_int16_t *, struct pf_addr *,
297 			    u_int16_t, u_int8_t, sa_family_t);
298 static int		 pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
299 			    struct tcphdr *, struct pf_state_peer *);
300 int			 pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *,
301 			    int *, u_int16_t *, u_int16_t *);
302 static void		 pf_change_icmp(struct pf_addr *, u_int16_t *,
303 			    struct pf_addr *, struct pf_addr *, u_int16_t,
304 			    u_int16_t *, u_int16_t *, u_int16_t *,
305 			    u_int16_t *, u_int8_t, sa_family_t);
306 static void		 pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
307 			    sa_family_t, struct pf_krule *, int);
308 static void		 pf_detach_state(struct pf_kstate *);
309 static int		 pf_state_key_attach(struct pf_state_key *,
310 			    struct pf_state_key *, struct pf_kstate *);
311 static void		 pf_state_key_detach(struct pf_kstate *, int);
312 static int		 pf_state_key_ctor(void *, int, void *, int);
313 static u_int32_t	 pf_tcp_iss(struct pf_pdesc *);
314 static __inline void	 pf_dummynet_flag_remove(struct mbuf *m,
315 			    struct pf_mtag *pf_mtag);
316 static int		 pf_dummynet(struct pf_pdesc *, struct pf_kstate *,
317 			    struct pf_krule *, struct mbuf **);
318 static int		 pf_dummynet_route(struct pf_pdesc *,
319 			    struct pf_kstate *, struct pf_krule *,
320 			    struct ifnet *, struct sockaddr *, struct mbuf **);
321 static int		 pf_test_eth_rule(int, struct pfi_kkif *,
322 			    struct mbuf **);
323 static int		 pf_test_rule(struct pf_krule **, struct pf_kstate **,
324 			    struct pfi_kkif *, struct mbuf *, int,
325 			    struct pf_pdesc *, struct pf_krule **,
326 			    struct pf_kruleset **, struct inpcb *);
327 static int		 pf_create_state(struct pf_krule *, struct pf_krule *,
328 			    struct pf_krule *, struct pf_pdesc *,
329 			    struct pf_ksrc_node *, struct pf_state_key *,
330 			    struct pf_state_key *, struct mbuf *, int,
331 			    u_int16_t, u_int16_t, int *, struct pfi_kkif *,
332 			    struct pf_kstate **, int, u_int16_t, u_int16_t,
333 			    int, struct pf_krule_slist *);
334 static int		 pf_state_key_addr_setup(struct pf_pdesc *, struct mbuf *,
335 			    int, struct pf_state_key_cmp *, int, struct pf_addr *,
336 			    int, struct pf_addr *, int);
337 static int		 pf_test_fragment(struct pf_krule **, struct pfi_kkif *,
338 			    struct mbuf *, void *, struct pf_pdesc *,
339 			    struct pf_krule **, struct pf_kruleset **);
340 static int		 pf_tcp_track_full(struct pf_kstate **,
341 			    struct pfi_kkif *, struct mbuf *, int,
342 			    struct pf_pdesc *, u_short *, int *);
343 static int		 pf_tcp_track_sloppy(struct pf_kstate **,
344 			    struct pf_pdesc *, u_short *);
345 static int		 pf_test_state_tcp(struct pf_kstate **,
346 			    struct pfi_kkif *, struct mbuf *, int,
347 			    void *, struct pf_pdesc *, u_short *);
348 static int		 pf_test_state_udp(struct pf_kstate **,
349 			    struct pfi_kkif *, struct mbuf *, int,
350 			    void *, struct pf_pdesc *);
351 int			 pf_icmp_state_lookup(struct pf_state_key_cmp *,
352 			    struct pf_pdesc *, struct pf_kstate **, struct mbuf *,
353 			    int, int, struct pfi_kkif *, u_int16_t, u_int16_t,
354 			    int, int *, int, int);
355 static int		 pf_test_state_icmp(struct pf_kstate **,
356 			    struct pfi_kkif *, struct mbuf *, int,
357 			    void *, struct pf_pdesc *, u_short *);
358 static void		 pf_sctp_multihome_detach_addr(const struct pf_kstate *);
359 static void		 pf_sctp_multihome_delayed(struct pf_pdesc *, int,
360 			    struct pfi_kkif *, struct pf_kstate *, int);
361 static int		 pf_test_state_sctp(struct pf_kstate **,
362 			    struct pfi_kkif *, struct mbuf *, int,
363 			    void *, struct pf_pdesc *, u_short *);
364 static int		 pf_test_state_other(struct pf_kstate **,
365 			    struct pfi_kkif *, struct mbuf *, struct pf_pdesc *);
366 static u_int16_t	 pf_calc_mss(struct pf_addr *, sa_family_t,
367 				int, u_int16_t);
368 static int		 pf_check_proto_cksum(struct mbuf *, int, int,
369 			    u_int8_t, sa_family_t);
370 static void		 pf_print_state_parts(struct pf_kstate *,
371 			    struct pf_state_key *, struct pf_state_key *);
372 static void		 pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t,
373 			    bool, u_int8_t);
374 static struct pf_kstate	*pf_find_state(struct pfi_kkif *,
375 			    const struct pf_state_key_cmp *, u_int);
376 static int		 pf_src_connlimit(struct pf_kstate **);
377 static void		 pf_overload_task(void *v, int pending);
378 static u_short		 pf_insert_src_node(struct pf_ksrc_node **,
379 			    struct pf_krule *, struct pf_addr *, sa_family_t);
380 static u_int		 pf_purge_expired_states(u_int, int);
381 static void		 pf_purge_unlinked_rules(void);
382 static int		 pf_mtag_uminit(void *, int, int);
383 static void		 pf_mtag_free(struct m_tag *);
384 static void		 pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *,
385 			    int, struct pf_state_key *);
386 #ifdef INET
387 static void		 pf_route(struct mbuf **, struct pf_krule *,
388 			    struct ifnet *, struct pf_kstate *,
389 			    struct pf_pdesc *, struct inpcb *);
390 #endif /* INET */
391 #ifdef INET6
392 static void		 pf_change_a6(struct pf_addr *, u_int16_t *,
393 			    struct pf_addr *, u_int8_t);
394 static void		 pf_route6(struct mbuf **, struct pf_krule *,
395 			    struct ifnet *, struct pf_kstate *,
396 			    struct pf_pdesc *, struct inpcb *);
397 #endif /* INET6 */
398 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t);
399 
400 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
401 
402 extern int pf_end_threads;
403 extern struct proc *pf_purge_proc;
404 
405 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
406 
407 enum { PF_ICMP_MULTI_NONE, PF_ICMP_MULTI_LINK };
408 
409 #define	PACKET_UNDO_NAT(_m, _pd, _off, _s)		\
410 	do {								\
411 		struct pf_state_key *nk;				\
412 		if ((pd->dir) == PF_OUT)					\
413 			nk = (_s)->key[PF_SK_STACK];			\
414 		else							\
415 			nk = (_s)->key[PF_SK_WIRE];			\
416 		pf_packet_rework_nat(_m, _pd, _off, nk);		\
417 	} while (0)
418 
419 #define	PACKET_LOOPED(pd)	((pd)->pf_mtag &&			\
420 				 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED)
421 
422 #define	STATE_LOOKUP(i, k, s, pd)					\
423 	do {								\
424 		(s) = pf_find_state((i), (k), (pd->dir));			\
425 		SDT_PROBE5(pf, ip, state, lookup, i, k, (pd->dir), pd, (s));	\
426 		if ((s) == NULL)					\
427 			return (PF_DROP);				\
428 		if (PACKET_LOOPED(pd))					\
429 			return (PF_PASS);				\
430 	} while (0)
431 
432 static struct pfi_kkif *
433 BOUND_IFACE(struct pf_kstate *st, struct pfi_kkif *k)
434 {
435 	SDT_PROBE2(pf, ip, , bound_iface, st, k);
436 
437 	/* Floating unless otherwise specified. */
438 	if (! (st->rule.ptr->rule_flag & PFRULE_IFBOUND))
439 		return (V_pfi_all);
440 
441 	/*
442 	 * Initially set to all, because we don't know what interface we'll be
443 	 * sending this out when we create the state.
444 	 */
445 	if (st->rule.ptr->rt == PF_REPLYTO)
446 		return (V_pfi_all);
447 
448 	/* Don't overrule the interface for states created on incoming packets. */
449 	if (st->direction == PF_IN)
450 		return (k);
451 
452 	/* No route-to, so don't overrule. */
453 	if (st->rt != PF_ROUTETO)
454 		return (k);
455 
456 	/* Bind to the route-to interface. */
457 	return (st->rt_kif);
458 }
459 
460 #define	STATE_INC_COUNTERS(s)						\
461 	do {								\
462 		struct pf_krule_item *mrm;				\
463 		counter_u64_add(s->rule.ptr->states_cur, 1);		\
464 		counter_u64_add(s->rule.ptr->states_tot, 1);		\
465 		if (s->anchor.ptr != NULL) {				\
466 			counter_u64_add(s->anchor.ptr->states_cur, 1);	\
467 			counter_u64_add(s->anchor.ptr->states_tot, 1);	\
468 		}							\
469 		if (s->nat_rule.ptr != NULL) {				\
470 			counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
471 			counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
472 		}							\
473 		SLIST_FOREACH(mrm, &s->match_rules, entry) {		\
474 			counter_u64_add(mrm->r->states_cur, 1);		\
475 			counter_u64_add(mrm->r->states_tot, 1);		\
476 		}							\
477 	} while (0)
478 
479 #define	STATE_DEC_COUNTERS(s)						\
480 	do {								\
481 		struct pf_krule_item *mrm;				\
482 		if (s->nat_rule.ptr != NULL)				\
483 			counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
484 		if (s->anchor.ptr != NULL)				\
485 			counter_u64_add(s->anchor.ptr->states_cur, -1);	\
486 		counter_u64_add(s->rule.ptr->states_cur, -1);		\
487 		SLIST_FOREACH(mrm, &s->match_rules, entry)		\
488 			counter_u64_add(mrm->r->states_cur, -1);	\
489 	} while (0)
490 
491 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
492 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items");
493 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
494 VNET_DEFINE(struct pf_idhash *, pf_idhash);
495 VNET_DEFINE(struct pf_srchash *, pf_srchash);
496 
497 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
498     "pf(4)");
499 
500 VNET_DEFINE(u_long, pf_hashmask);
501 VNET_DEFINE(u_long, pf_srchashmask);
502 VNET_DEFINE_STATIC(u_long, pf_hashsize);
503 #define V_pf_hashsize	VNET(pf_hashsize)
504 VNET_DEFINE_STATIC(u_long, pf_srchashsize);
505 #define V_pf_srchashsize	VNET(pf_srchashsize)
506 u_long	pf_ioctl_maxcount = 65535;
507 
508 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN,
509     &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable");
510 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN,
511     &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable");
512 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN,
513     &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call");
514 
515 VNET_DEFINE(void *, pf_swi_cookie);
516 VNET_DEFINE(struct intr_event *, pf_swi_ie);
517 
518 VNET_DEFINE(uint32_t, pf_hashseed);
519 #define	V_pf_hashseed	VNET(pf_hashseed)
520 
521 static void
522 pf_sctp_checksum(struct mbuf *m, int off)
523 {
524 	uint32_t sum = 0;
525 
526 	/* Zero out the checksum, to enable recalculation. */
527 	m_copyback(m, off + offsetof(struct sctphdr, checksum),
528 	    sizeof(sum), (caddr_t)&sum);
529 
530 	sum = sctp_calculate_cksum(m, off);
531 
532 	m_copyback(m, off + offsetof(struct sctphdr, checksum),
533 	    sizeof(sum), (caddr_t)&sum);
534 }
535 
536 int
537 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
538 {
539 
540 	switch (af) {
541 #ifdef INET
542 	case AF_INET:
543 		if (a->addr32[0] > b->addr32[0])
544 			return (1);
545 		if (a->addr32[0] < b->addr32[0])
546 			return (-1);
547 		break;
548 #endif /* INET */
549 #ifdef INET6
550 	case AF_INET6:
551 		if (a->addr32[3] > b->addr32[3])
552 			return (1);
553 		if (a->addr32[3] < b->addr32[3])
554 			return (-1);
555 		if (a->addr32[2] > b->addr32[2])
556 			return (1);
557 		if (a->addr32[2] < b->addr32[2])
558 			return (-1);
559 		if (a->addr32[1] > b->addr32[1])
560 			return (1);
561 		if (a->addr32[1] < b->addr32[1])
562 			return (-1);
563 		if (a->addr32[0] > b->addr32[0])
564 			return (1);
565 		if (a->addr32[0] < b->addr32[0])
566 			return (-1);
567 		break;
568 #endif /* INET6 */
569 	default:
570 		panic("%s: unknown address family %u", __func__, af);
571 	}
572 	return (0);
573 }
574 
575 static void
576 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off,
577 	struct pf_state_key *nk)
578 {
579 
580 	switch (pd->proto) {
581 	case IPPROTO_TCP: {
582 		struct tcphdr *th = &pd->hdr.tcp;
583 
584 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af))
585 			pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum,
586 			    &th->th_sum, &nk->addr[pd->sidx],
587 			    nk->port[pd->sidx], 0, pd->af);
588 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af))
589 			pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum,
590 			    &th->th_sum, &nk->addr[pd->didx],
591 			    nk->port[pd->didx], 0, pd->af);
592 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
593 		break;
594 	}
595 	case IPPROTO_UDP: {
596 		struct udphdr *uh = &pd->hdr.udp;
597 
598 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af))
599 			pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
600 			    &uh->uh_sum, &nk->addr[pd->sidx],
601 			    nk->port[pd->sidx], 1, pd->af);
602 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af))
603 			pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
604 			    &uh->uh_sum, &nk->addr[pd->didx],
605 			    nk->port[pd->didx], 1, pd->af);
606 		m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
607 		break;
608 	}
609 	case IPPROTO_SCTP: {
610 		struct sctphdr *sh = &pd->hdr.sctp;
611 		uint16_t checksum = 0;
612 
613 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) {
614 			pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum,
615 			    &checksum, &nk->addr[pd->sidx],
616 			    nk->port[pd->sidx], 1, pd->af);
617 		}
618 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) {
619 			pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum,
620 			    &checksum, &nk->addr[pd->didx],
621 			    nk->port[pd->didx], 1, pd->af);
622 		}
623 
624 		break;
625 	}
626 	case IPPROTO_ICMP: {
627 		struct icmp *ih = &pd->hdr.icmp;
628 
629 		if (nk->port[pd->sidx] != ih->icmp_id) {
630 			pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
631 			    ih->icmp_cksum, ih->icmp_id,
632 			    nk->port[pd->sidx], 0);
633 			ih->icmp_id = nk->port[pd->sidx];
634 			pd->sport = &ih->icmp_id;
635 
636 			m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih);
637 		}
638 		/* FALLTHROUGH */
639 	}
640 	default:
641 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) {
642 			switch (pd->af) {
643 			case AF_INET:
644 				pf_change_a(&pd->src->v4.s_addr,
645 				    pd->ip_sum, nk->addr[pd->sidx].v4.s_addr,
646 				    0);
647 				break;
648 			case AF_INET6:
649 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
650 				break;
651 			}
652 		}
653 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) {
654 			switch (pd->af) {
655 			case AF_INET:
656 				pf_change_a(&pd->dst->v4.s_addr,
657 				    pd->ip_sum, nk->addr[pd->didx].v4.s_addr,
658 				    0);
659 				break;
660 			case AF_INET6:
661 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
662 				break;
663 			}
664 		}
665 		break;
666 	}
667 }
668 
669 static __inline uint32_t
670 pf_hashkey(const struct pf_state_key *sk)
671 {
672 	uint32_t h;
673 
674 	h = murmur3_32_hash32((const uint32_t *)sk,
675 	    sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
676 	    V_pf_hashseed);
677 
678 	return (h & V_pf_hashmask);
679 }
680 
681 static __inline uint32_t
682 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
683 {
684 	uint32_t h;
685 
686 	switch (af) {
687 	case AF_INET:
688 		h = murmur3_32_hash32((uint32_t *)&addr->v4,
689 		    sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
690 		break;
691 	case AF_INET6:
692 		h = murmur3_32_hash32((uint32_t *)&addr->v6,
693 		    sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
694 		break;
695 	default:
696 		panic("%s: unknown address family %u", __func__, af);
697 	}
698 
699 	return (h & V_pf_srchashmask);
700 }
701 
702 #ifdef ALTQ
703 static int
704 pf_state_hash(struct pf_kstate *s)
705 {
706 	u_int32_t hv = (intptr_t)s / sizeof(*s);
707 
708 	hv ^= crc32(&s->src, sizeof(s->src));
709 	hv ^= crc32(&s->dst, sizeof(s->dst));
710 	if (hv == 0)
711 		hv = 1;
712 	return (hv);
713 }
714 #endif
715 
716 static __inline void
717 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate)
718 {
719 	if (which == PF_PEER_DST || which == PF_PEER_BOTH)
720 		s->dst.state = newstate;
721 	if (which == PF_PEER_DST)
722 		return;
723 	if (s->src.state == newstate)
724 		return;
725 	if (s->creatorid == V_pf_status.hostid &&
726 	    s->key[PF_SK_STACK] != NULL &&
727 	    s->key[PF_SK_STACK]->proto == IPPROTO_TCP &&
728 	    !(TCPS_HAVEESTABLISHED(s->src.state) ||
729 	    s->src.state == TCPS_CLOSED) &&
730 	    (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED))
731 		atomic_add_32(&V_pf_status.states_halfopen, -1);
732 
733 	s->src.state = newstate;
734 }
735 
736 #ifdef INET6
737 void
738 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
739 {
740 	switch (af) {
741 #ifdef INET
742 	case AF_INET:
743 		memcpy(&dst->v4, &src->v4, sizeof(dst->v4));
744 		break;
745 #endif /* INET */
746 	case AF_INET6:
747 		memcpy(&dst->v6, &src->v6, sizeof(dst->v6));
748 		break;
749 	}
750 }
751 #endif /* INET6 */
752 
753 static void
754 pf_init_threshold(struct pf_threshold *threshold,
755     u_int32_t limit, u_int32_t seconds)
756 {
757 	threshold->limit = limit * PF_THRESHOLD_MULT;
758 	threshold->seconds = seconds;
759 	threshold->count = 0;
760 	threshold->last = time_uptime;
761 }
762 
763 static void
764 pf_add_threshold(struct pf_threshold *threshold)
765 {
766 	u_int32_t t = time_uptime, diff = t - threshold->last;
767 
768 	if (diff >= threshold->seconds)
769 		threshold->count = 0;
770 	else
771 		threshold->count -= threshold->count * diff /
772 		    threshold->seconds;
773 	threshold->count += PF_THRESHOLD_MULT;
774 	threshold->last = t;
775 }
776 
777 static int
778 pf_check_threshold(struct pf_threshold *threshold)
779 {
780 	return (threshold->count > threshold->limit);
781 }
782 
783 static int
784 pf_src_connlimit(struct pf_kstate **state)
785 {
786 	struct pf_overload_entry *pfoe;
787 	int bad = 0;
788 
789 	PF_STATE_LOCK_ASSERT(*state);
790 	/*
791 	 * XXXKS: The src node is accessed unlocked!
792 	 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node);
793 	 */
794 
795 	(*state)->src_node->conn++;
796 	(*state)->src.tcp_est = 1;
797 	pf_add_threshold(&(*state)->src_node->conn_rate);
798 
799 	if ((*state)->rule.ptr->max_src_conn &&
800 	    (*state)->rule.ptr->max_src_conn <
801 	    (*state)->src_node->conn) {
802 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
803 		bad++;
804 	}
805 
806 	if ((*state)->rule.ptr->max_src_conn_rate.limit &&
807 	    pf_check_threshold(&(*state)->src_node->conn_rate)) {
808 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
809 		bad++;
810 	}
811 
812 	if (!bad)
813 		return (0);
814 
815 	/* Kill this state. */
816 	(*state)->timeout = PFTM_PURGE;
817 	pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED);
818 
819 	if ((*state)->rule.ptr->overload_tbl == NULL)
820 		return (1);
821 
822 	/* Schedule overloading and flushing task. */
823 	pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
824 	if (pfoe == NULL)
825 		return (1);	/* too bad :( */
826 
827 	bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
828 	pfoe->af = (*state)->key[PF_SK_WIRE]->af;
829 	pfoe->rule = (*state)->rule.ptr;
830 	pfoe->dir = (*state)->direction;
831 	PF_OVERLOADQ_LOCK();
832 	SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
833 	PF_OVERLOADQ_UNLOCK();
834 	taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
835 
836 	return (1);
837 }
838 
839 static void
840 pf_overload_task(void *v, int pending)
841 {
842 	struct pf_overload_head queue;
843 	struct pfr_addr p;
844 	struct pf_overload_entry *pfoe, *pfoe1;
845 	uint32_t killed = 0;
846 
847 	CURVNET_SET((struct vnet *)v);
848 
849 	PF_OVERLOADQ_LOCK();
850 	queue = V_pf_overloadqueue;
851 	SLIST_INIT(&V_pf_overloadqueue);
852 	PF_OVERLOADQ_UNLOCK();
853 
854 	bzero(&p, sizeof(p));
855 	SLIST_FOREACH(pfoe, &queue, next) {
856 		counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
857 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
858 			printf("%s: blocking address ", __func__);
859 			pf_print_host(&pfoe->addr, 0, pfoe->af);
860 			printf("\n");
861 		}
862 
863 		p.pfra_af = pfoe->af;
864 		switch (pfoe->af) {
865 #ifdef INET
866 		case AF_INET:
867 			p.pfra_net = 32;
868 			p.pfra_ip4addr = pfoe->addr.v4;
869 			break;
870 #endif
871 #ifdef INET6
872 		case AF_INET6:
873 			p.pfra_net = 128;
874 			p.pfra_ip6addr = pfoe->addr.v6;
875 			break;
876 #endif
877 		}
878 
879 		PF_RULES_WLOCK();
880 		pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
881 		PF_RULES_WUNLOCK();
882 	}
883 
884 	/*
885 	 * Remove those entries, that don't need flushing.
886 	 */
887 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
888 		if (pfoe->rule->flush == 0) {
889 			SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
890 			free(pfoe, M_PFTEMP);
891 		} else
892 			counter_u64_add(
893 			    V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
894 
895 	/* If nothing to flush, return. */
896 	if (SLIST_EMPTY(&queue)) {
897 		CURVNET_RESTORE();
898 		return;
899 	}
900 
901 	for (int i = 0; i <= V_pf_hashmask; i++) {
902 		struct pf_idhash *ih = &V_pf_idhash[i];
903 		struct pf_state_key *sk;
904 		struct pf_kstate *s;
905 
906 		PF_HASHROW_LOCK(ih);
907 		LIST_FOREACH(s, &ih->states, entry) {
908 		    sk = s->key[PF_SK_WIRE];
909 		    SLIST_FOREACH(pfoe, &queue, next)
910 			if (sk->af == pfoe->af &&
911 			    ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
912 			    pfoe->rule == s->rule.ptr) &&
913 			    ((pfoe->dir == PF_OUT &&
914 			    PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
915 			    (pfoe->dir == PF_IN &&
916 			    PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
917 				s->timeout = PFTM_PURGE;
918 				pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED);
919 				killed++;
920 			}
921 		}
922 		PF_HASHROW_UNLOCK(ih);
923 	}
924 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
925 		free(pfoe, M_PFTEMP);
926 	if (V_pf_status.debug >= PF_DEBUG_MISC)
927 		printf("%s: %u states killed", __func__, killed);
928 
929 	CURVNET_RESTORE();
930 }
931 
932 /*
933  * Can return locked on failure, so that we can consistently
934  * allocate and insert a new one.
935  */
936 struct pf_ksrc_node *
937 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af,
938 	struct pf_srchash **sh, bool returnlocked)
939 {
940 	struct pf_ksrc_node *n;
941 
942 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
943 
944 	*sh = &V_pf_srchash[pf_hashsrc(src, af)];
945 	PF_HASHROW_LOCK(*sh);
946 	LIST_FOREACH(n, &(*sh)->nodes, entry)
947 		if (n->rule.ptr == rule && n->af == af &&
948 		    ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
949 		    (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
950 			break;
951 
952 	if (n != NULL) {
953 		n->states++;
954 		PF_HASHROW_UNLOCK(*sh);
955 	} else if (returnlocked == false)
956 		PF_HASHROW_UNLOCK(*sh);
957 
958 	return (n);
959 }
960 
961 static void
962 pf_free_src_node(struct pf_ksrc_node *sn)
963 {
964 
965 	for (int i = 0; i < 2; i++) {
966 		counter_u64_free(sn->bytes[i]);
967 		counter_u64_free(sn->packets[i]);
968 	}
969 	uma_zfree(V_pf_sources_z, sn);
970 }
971 
972 static u_short
973 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule,
974     struct pf_addr *src, sa_family_t af)
975 {
976 	u_short			 reason = 0;
977 	struct pf_srchash	*sh = NULL;
978 
979 	KASSERT((rule->rule_flag & PFRULE_SRCTRACK ||
980 	    rule->rpool.opts & PF_POOL_STICKYADDR),
981 	    ("%s for non-tracking rule %p", __func__, rule));
982 
983 	if (*sn == NULL)
984 		*sn = pf_find_src_node(src, rule, af, &sh, true);
985 
986 	if (*sn == NULL) {
987 		PF_HASHROW_ASSERT(sh);
988 
989 		if (rule->max_src_nodes &&
990 		    counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) {
991 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1);
992 			PF_HASHROW_UNLOCK(sh);
993 			reason = PFRES_SRCLIMIT;
994 			goto done;
995 		}
996 
997 		(*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
998 		if ((*sn) == NULL) {
999 			PF_HASHROW_UNLOCK(sh);
1000 			reason = PFRES_MEMORY;
1001 			goto done;
1002 		}
1003 
1004 		for (int i = 0; i < 2; i++) {
1005 			(*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT);
1006 			(*sn)->packets[i] = counter_u64_alloc(M_NOWAIT);
1007 
1008 			if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) {
1009 				pf_free_src_node(*sn);
1010 				PF_HASHROW_UNLOCK(sh);
1011 				reason = PFRES_MEMORY;
1012 				goto done;
1013 			}
1014 		}
1015 
1016 		pf_init_threshold(&(*sn)->conn_rate,
1017 		    rule->max_src_conn_rate.limit,
1018 		    rule->max_src_conn_rate.seconds);
1019 
1020 		MPASS((*sn)->lock == NULL);
1021 		(*sn)->lock = &sh->lock;
1022 
1023 		(*sn)->af = af;
1024 		(*sn)->rule.ptr = rule;
1025 		PF_ACPY(&(*sn)->addr, src, af);
1026 		LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
1027 		(*sn)->creation = time_uptime;
1028 		(*sn)->ruletype = rule->action;
1029 		(*sn)->states = 1;
1030 		if ((*sn)->rule.ptr != NULL)
1031 			counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
1032 		PF_HASHROW_UNLOCK(sh);
1033 		counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
1034 	} else {
1035 		if (rule->max_src_states &&
1036 		    (*sn)->states >= rule->max_src_states) {
1037 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
1038 			    1);
1039 			reason = PFRES_SRCLIMIT;
1040 			goto done;
1041 		}
1042 	}
1043 done:
1044 	return (reason);
1045 }
1046 
1047 void
1048 pf_unlink_src_node(struct pf_ksrc_node *src)
1049 {
1050 	PF_SRC_NODE_LOCK_ASSERT(src);
1051 
1052 	LIST_REMOVE(src, entry);
1053 	if (src->rule.ptr)
1054 		counter_u64_add(src->rule.ptr->src_nodes, -1);
1055 }
1056 
1057 u_int
1058 pf_free_src_nodes(struct pf_ksrc_node_list *head)
1059 {
1060 	struct pf_ksrc_node *sn, *tmp;
1061 	u_int count = 0;
1062 
1063 	LIST_FOREACH_SAFE(sn, head, entry, tmp) {
1064 		pf_free_src_node(sn);
1065 		count++;
1066 	}
1067 
1068 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
1069 
1070 	return (count);
1071 }
1072 
1073 void
1074 pf_mtag_initialize(void)
1075 {
1076 
1077 	pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
1078 	    sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
1079 	    UMA_ALIGN_PTR, 0);
1080 }
1081 
1082 /* Per-vnet data storage structures initialization. */
1083 void
1084 pf_initialize(void)
1085 {
1086 	struct pf_keyhash	*kh;
1087 	struct pf_idhash	*ih;
1088 	struct pf_srchash	*sh;
1089 	u_int i;
1090 
1091 	if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize))
1092 		V_pf_hashsize = PF_HASHSIZ;
1093 	if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize))
1094 		V_pf_srchashsize = PF_SRCHASHSIZ;
1095 
1096 	V_pf_hashseed = arc4random();
1097 
1098 	/* States and state keys storage. */
1099 	V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate),
1100 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1101 	V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
1102 	uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
1103 	uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
1104 
1105 	V_pf_state_key_z = uma_zcreate("pf state keys",
1106 	    sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
1107 	    UMA_ALIGN_PTR, 0);
1108 
1109 	V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash),
1110 	    M_PFHASH, M_NOWAIT | M_ZERO);
1111 	V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash),
1112 	    M_PFHASH, M_NOWAIT | M_ZERO);
1113 	if (V_pf_keyhash == NULL || V_pf_idhash == NULL) {
1114 		printf("pf: Unable to allocate memory for "
1115 		    "state_hashsize %lu.\n", V_pf_hashsize);
1116 
1117 		free(V_pf_keyhash, M_PFHASH);
1118 		free(V_pf_idhash, M_PFHASH);
1119 
1120 		V_pf_hashsize = PF_HASHSIZ;
1121 		V_pf_keyhash = mallocarray(V_pf_hashsize,
1122 		    sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO);
1123 		V_pf_idhash = mallocarray(V_pf_hashsize,
1124 		    sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO);
1125 	}
1126 
1127 	V_pf_hashmask = V_pf_hashsize - 1;
1128 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
1129 	    i++, kh++, ih++) {
1130 		mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
1131 		mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
1132 	}
1133 
1134 	/* Source nodes. */
1135 	V_pf_sources_z = uma_zcreate("pf source nodes",
1136 	    sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
1137 	    0);
1138 	V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
1139 	uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
1140 	uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
1141 
1142 	V_pf_srchash = mallocarray(V_pf_srchashsize,
1143 	    sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO);
1144 	if (V_pf_srchash == NULL) {
1145 		printf("pf: Unable to allocate memory for "
1146 		    "source_hashsize %lu.\n", V_pf_srchashsize);
1147 
1148 		V_pf_srchashsize = PF_SRCHASHSIZ;
1149 		V_pf_srchash = mallocarray(V_pf_srchashsize,
1150 		    sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO);
1151 	}
1152 
1153 	V_pf_srchashmask = V_pf_srchashsize - 1;
1154 	for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++)
1155 		mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
1156 
1157 	/* ALTQ */
1158 	TAILQ_INIT(&V_pf_altqs[0]);
1159 	TAILQ_INIT(&V_pf_altqs[1]);
1160 	TAILQ_INIT(&V_pf_altqs[2]);
1161 	TAILQ_INIT(&V_pf_altqs[3]);
1162 	TAILQ_INIT(&V_pf_pabuf);
1163 	V_pf_altqs_active = &V_pf_altqs[0];
1164 	V_pf_altq_ifs_active = &V_pf_altqs[1];
1165 	V_pf_altqs_inactive = &V_pf_altqs[2];
1166 	V_pf_altq_ifs_inactive = &V_pf_altqs[3];
1167 
1168 	/* Send & overload+flush queues. */
1169 	STAILQ_INIT(&V_pf_sendqueue);
1170 	SLIST_INIT(&V_pf_overloadqueue);
1171 	TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
1172 
1173 	/* Unlinked, but may be referenced rules. */
1174 	TAILQ_INIT(&V_pf_unlinked_rules);
1175 }
1176 
1177 void
1178 pf_mtag_cleanup(void)
1179 {
1180 
1181 	uma_zdestroy(pf_mtag_z);
1182 }
1183 
1184 void
1185 pf_cleanup(void)
1186 {
1187 	struct pf_keyhash	*kh;
1188 	struct pf_idhash	*ih;
1189 	struct pf_srchash	*sh;
1190 	struct pf_send_entry	*pfse, *next;
1191 	u_int i;
1192 
1193 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
1194 	    i++, kh++, ih++) {
1195 		KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
1196 		    __func__));
1197 		KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
1198 		    __func__));
1199 		mtx_destroy(&kh->lock);
1200 		mtx_destroy(&ih->lock);
1201 	}
1202 	free(V_pf_keyhash, M_PFHASH);
1203 	free(V_pf_idhash, M_PFHASH);
1204 
1205 	for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
1206 		KASSERT(LIST_EMPTY(&sh->nodes),
1207 		    ("%s: source node hash not empty", __func__));
1208 		mtx_destroy(&sh->lock);
1209 	}
1210 	free(V_pf_srchash, M_PFHASH);
1211 
1212 	STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
1213 		m_freem(pfse->pfse_m);
1214 		free(pfse, M_PFTEMP);
1215 	}
1216 	MPASS(RB_EMPTY(&V_pf_sctp_endpoints));
1217 
1218 	uma_zdestroy(V_pf_sources_z);
1219 	uma_zdestroy(V_pf_state_z);
1220 	uma_zdestroy(V_pf_state_key_z);
1221 }
1222 
1223 static int
1224 pf_mtag_uminit(void *mem, int size, int how)
1225 {
1226 	struct m_tag *t;
1227 
1228 	t = (struct m_tag *)mem;
1229 	t->m_tag_cookie = MTAG_ABI_COMPAT;
1230 	t->m_tag_id = PACKET_TAG_PF;
1231 	t->m_tag_len = sizeof(struct pf_mtag);
1232 	t->m_tag_free = pf_mtag_free;
1233 
1234 	return (0);
1235 }
1236 
1237 static void
1238 pf_mtag_free(struct m_tag *t)
1239 {
1240 
1241 	uma_zfree(pf_mtag_z, t);
1242 }
1243 
1244 struct pf_mtag *
1245 pf_get_mtag(struct mbuf *m)
1246 {
1247 	struct m_tag *mtag;
1248 
1249 	if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
1250 		return ((struct pf_mtag *)(mtag + 1));
1251 
1252 	mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
1253 	if (mtag == NULL)
1254 		return (NULL);
1255 	bzero(mtag + 1, sizeof(struct pf_mtag));
1256 	m_tag_prepend(m, mtag);
1257 
1258 	return ((struct pf_mtag *)(mtag + 1));
1259 }
1260 
1261 static int
1262 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
1263     struct pf_kstate *s)
1264 {
1265 	struct pf_keyhash	*khs, *khw, *kh;
1266 	struct pf_state_key	*sk, *cur;
1267 	struct pf_kstate	*si, *olds = NULL;
1268 	int idx;
1269 
1270 	NET_EPOCH_ASSERT();
1271 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1272 	KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
1273 	KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
1274 
1275 	/*
1276 	 * We need to lock hash slots of both keys. To avoid deadlock
1277 	 * we always lock the slot with lower address first. Unlock order
1278 	 * isn't important.
1279 	 *
1280 	 * We also need to lock ID hash slot before dropping key
1281 	 * locks. On success we return with ID hash slot locked.
1282 	 */
1283 
1284 	if (skw == sks) {
1285 		khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
1286 		PF_HASHROW_LOCK(khs);
1287 	} else {
1288 		khs = &V_pf_keyhash[pf_hashkey(sks)];
1289 		khw = &V_pf_keyhash[pf_hashkey(skw)];
1290 		if (khs == khw) {
1291 			PF_HASHROW_LOCK(khs);
1292 		} else if (khs < khw) {
1293 			PF_HASHROW_LOCK(khs);
1294 			PF_HASHROW_LOCK(khw);
1295 		} else {
1296 			PF_HASHROW_LOCK(khw);
1297 			PF_HASHROW_LOCK(khs);
1298 		}
1299 	}
1300 
1301 #define	KEYS_UNLOCK()	do {			\
1302 	if (khs != khw) {			\
1303 		PF_HASHROW_UNLOCK(khs);		\
1304 		PF_HASHROW_UNLOCK(khw);		\
1305 	} else					\
1306 		PF_HASHROW_UNLOCK(khs);		\
1307 } while (0)
1308 
1309 	/*
1310 	 * First run: start with wire key.
1311 	 */
1312 	sk = skw;
1313 	kh = khw;
1314 	idx = PF_SK_WIRE;
1315 
1316 	MPASS(s->lock == NULL);
1317 	s->lock = &V_pf_idhash[PF_IDHASH(s)].lock;
1318 
1319 keyattach:
1320 	LIST_FOREACH(cur, &kh->keys, entry)
1321 		if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
1322 			break;
1323 
1324 	if (cur != NULL) {
1325 		/* Key exists. Check for same kif, if none, add to key. */
1326 		TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
1327 			struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
1328 
1329 			PF_HASHROW_LOCK(ih);
1330 			if (si->kif == s->kif &&
1331 			    si->direction == s->direction) {
1332 				if (sk->proto == IPPROTO_TCP &&
1333 				    si->src.state >= TCPS_FIN_WAIT_2 &&
1334 				    si->dst.state >= TCPS_FIN_WAIT_2) {
1335 					/*
1336 					 * New state matches an old >FIN_WAIT_2
1337 					 * state. We can't drop key hash locks,
1338 					 * thus we can't unlink it properly.
1339 					 *
1340 					 * As a workaround we drop it into
1341 					 * TCPS_CLOSED state, schedule purge
1342 					 * ASAP and push it into the very end
1343 					 * of the slot TAILQ, so that it won't
1344 					 * conflict with our new state.
1345 					 */
1346 					pf_set_protostate(si, PF_PEER_BOTH,
1347 					    TCPS_CLOSED);
1348 					si->timeout = PFTM_PURGE;
1349 					olds = si;
1350 				} else {
1351 					if (V_pf_status.debug >= PF_DEBUG_MISC) {
1352 						printf("pf: %s key attach "
1353 						    "failed on %s: ",
1354 						    (idx == PF_SK_WIRE) ?
1355 						    "wire" : "stack",
1356 						    s->kif->pfik_name);
1357 						pf_print_state_parts(s,
1358 						    (idx == PF_SK_WIRE) ?
1359 						    sk : NULL,
1360 						    (idx == PF_SK_STACK) ?
1361 						    sk : NULL);
1362 						printf(", existing: ");
1363 						pf_print_state_parts(si,
1364 						    (idx == PF_SK_WIRE) ?
1365 						    sk : NULL,
1366 						    (idx == PF_SK_STACK) ?
1367 						    sk : NULL);
1368 						printf("\n");
1369 					}
1370 					s->timeout = PFTM_UNLINKED;
1371 					PF_HASHROW_UNLOCK(ih);
1372 					KEYS_UNLOCK();
1373 					uma_zfree(V_pf_state_key_z, sk);
1374 					if (idx == PF_SK_STACK)
1375 						pf_detach_state(s);
1376 					return (EEXIST); /* collision! */
1377 				}
1378 			}
1379 			PF_HASHROW_UNLOCK(ih);
1380 		}
1381 		uma_zfree(V_pf_state_key_z, sk);
1382 		s->key[idx] = cur;
1383 	} else {
1384 		LIST_INSERT_HEAD(&kh->keys, sk, entry);
1385 		s->key[idx] = sk;
1386 	}
1387 
1388 stateattach:
1389 	/* List is sorted, if-bound states before floating. */
1390 	if (s->kif == V_pfi_all)
1391 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1392 	else
1393 		TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1394 
1395 	if (olds) {
1396 		TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1397 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1398 		    key_list[idx]);
1399 		olds = NULL;
1400 	}
1401 
1402 	/*
1403 	 * Attach done. See how should we (or should not?)
1404 	 * attach a second key.
1405 	 */
1406 	if (sks == skw) {
1407 		s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1408 		idx = PF_SK_STACK;
1409 		sks = NULL;
1410 		goto stateattach;
1411 	} else if (sks != NULL) {
1412 		/*
1413 		 * Continue attaching with stack key.
1414 		 */
1415 		sk = sks;
1416 		kh = khs;
1417 		idx = PF_SK_STACK;
1418 		sks = NULL;
1419 		goto keyattach;
1420 	}
1421 
1422 	PF_STATE_LOCK(s);
1423 	KEYS_UNLOCK();
1424 
1425 	KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1426 	    ("%s failure", __func__));
1427 
1428 	return (0);
1429 #undef	KEYS_UNLOCK
1430 }
1431 
1432 static void
1433 pf_detach_state(struct pf_kstate *s)
1434 {
1435 	struct pf_state_key *sks = s->key[PF_SK_STACK];
1436 	struct pf_keyhash *kh;
1437 
1438 	NET_EPOCH_ASSERT();
1439 	MPASS(s->timeout >= PFTM_MAX);
1440 
1441 	pf_sctp_multihome_detach_addr(s);
1442 
1443 	if ((s->state_flags & PFSTATE_PFLOW) && V_pflow_export_state_ptr)
1444 		V_pflow_export_state_ptr(s);
1445 
1446 	if (sks != NULL) {
1447 		kh = &V_pf_keyhash[pf_hashkey(sks)];
1448 		PF_HASHROW_LOCK(kh);
1449 		if (s->key[PF_SK_STACK] != NULL)
1450 			pf_state_key_detach(s, PF_SK_STACK);
1451 		/*
1452 		 * If both point to same key, then we are done.
1453 		 */
1454 		if (sks == s->key[PF_SK_WIRE]) {
1455 			pf_state_key_detach(s, PF_SK_WIRE);
1456 			PF_HASHROW_UNLOCK(kh);
1457 			return;
1458 		}
1459 		PF_HASHROW_UNLOCK(kh);
1460 	}
1461 
1462 	if (s->key[PF_SK_WIRE] != NULL) {
1463 		kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1464 		PF_HASHROW_LOCK(kh);
1465 		if (s->key[PF_SK_WIRE] != NULL)
1466 			pf_state_key_detach(s, PF_SK_WIRE);
1467 		PF_HASHROW_UNLOCK(kh);
1468 	}
1469 }
1470 
1471 static void
1472 pf_state_key_detach(struct pf_kstate *s, int idx)
1473 {
1474 	struct pf_state_key *sk = s->key[idx];
1475 #ifdef INVARIANTS
1476 	struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1477 
1478 	PF_HASHROW_ASSERT(kh);
1479 #endif
1480 	TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1481 	s->key[idx] = NULL;
1482 
1483 	if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1484 		LIST_REMOVE(sk, entry);
1485 		uma_zfree(V_pf_state_key_z, sk);
1486 	}
1487 }
1488 
1489 static int
1490 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1491 {
1492 	struct pf_state_key *sk = mem;
1493 
1494 	bzero(sk, sizeof(struct pf_state_key_cmp));
1495 	TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1496 	TAILQ_INIT(&sk->states[PF_SK_STACK]);
1497 
1498 	return (0);
1499 }
1500 
1501 static int
1502 pf_state_key_addr_setup(struct pf_pdesc *pd, struct mbuf *m, int off,
1503     struct pf_state_key_cmp *key, int sidx, struct pf_addr *saddr,
1504     int didx, struct pf_addr *daddr, int multi)
1505 {
1506 #ifdef INET6
1507 	struct nd_neighbor_solicit nd;
1508 	struct pf_addr *target;
1509 	u_short action, reason;
1510 
1511 	if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6)
1512 		goto copy;
1513 
1514 	switch (pd->hdr.icmp6.icmp6_type) {
1515 	case ND_NEIGHBOR_SOLICIT:
1516 		if (multi)
1517 			return (-1);
1518 		if (!pf_pull_hdr(m, off, &nd, sizeof(nd), &action, &reason, pd->af))
1519 			return (-1);
1520 		target = (struct pf_addr *)&nd.nd_ns_target;
1521 		daddr = target;
1522 		break;
1523 	case ND_NEIGHBOR_ADVERT:
1524 		if (multi)
1525 			return (-1);
1526 		if (!pf_pull_hdr(m, off, &nd, sizeof(nd), &action, &reason, pd->af))
1527 			return (-1);
1528 		target = (struct pf_addr *)&nd.nd_ns_target;
1529 		saddr = target;
1530 		if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) {
1531 			key->addr[didx].addr32[0] = 0;
1532 			key->addr[didx].addr32[1] = 0;
1533 			key->addr[didx].addr32[2] = 0;
1534 			key->addr[didx].addr32[3] = 0;
1535 			daddr = NULL; /* overwritten */
1536 		}
1537 		break;
1538 	default:
1539 		if (multi == PF_ICMP_MULTI_LINK) {
1540 			key->addr[sidx].addr32[0] = IPV6_ADDR_INT32_MLL;
1541 			key->addr[sidx].addr32[1] = 0;
1542 			key->addr[sidx].addr32[2] = 0;
1543 			key->addr[sidx].addr32[3] = IPV6_ADDR_INT32_ONE;
1544 			saddr = NULL; /* overwritten */
1545 		}
1546 	}
1547 copy:
1548 #endif
1549 	if (saddr)
1550 		PF_ACPY(&key->addr[sidx], saddr, pd->af);
1551 	if (daddr)
1552 		PF_ACPY(&key->addr[didx], daddr, pd->af);
1553 
1554 	return (0);
1555 }
1556 
1557 struct pf_state_key *
1558 pf_state_key_setup(struct pf_pdesc *pd, struct mbuf *m, int off,
1559     struct pf_addr *saddr, struct pf_addr *daddr, u_int16_t sport,
1560     u_int16_t dport)
1561 {
1562 	struct pf_state_key *sk;
1563 
1564 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1565 	if (sk == NULL)
1566 		return (NULL);
1567 
1568 	if (pf_state_key_addr_setup(pd, m, off, (struct pf_state_key_cmp *)sk,
1569 	    pd->sidx, pd->src, pd->didx, pd->dst, 0)) {
1570 		uma_zfree(V_pf_state_key_z, sk);
1571 		return (NULL);
1572 	}
1573 
1574 	sk->port[pd->sidx] = sport;
1575 	sk->port[pd->didx] = dport;
1576 	sk->proto = pd->proto;
1577 	sk->af = pd->af;
1578 
1579 	return (sk);
1580 }
1581 
1582 struct pf_state_key *
1583 pf_state_key_clone(const struct pf_state_key *orig)
1584 {
1585 	struct pf_state_key *sk;
1586 
1587 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1588 	if (sk == NULL)
1589 		return (NULL);
1590 
1591 	bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1592 
1593 	return (sk);
1594 }
1595 
1596 int
1597 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif,
1598     struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s)
1599 {
1600 	struct pf_idhash *ih;
1601 	struct pf_kstate *cur;
1602 	int error;
1603 
1604 	NET_EPOCH_ASSERT();
1605 
1606 	KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1607 	    ("%s: sks not pristine", __func__));
1608 	KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1609 	    ("%s: skw not pristine", __func__));
1610 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1611 
1612 	s->kif = kif;
1613 	s->orig_kif = orig_kif;
1614 
1615 	if (s->id == 0 && s->creatorid == 0) {
1616 		s->id = alloc_unr64(&V_pf_stateid);
1617 		s->id = htobe64(s->id);
1618 		s->creatorid = V_pf_status.hostid;
1619 	}
1620 
1621 	/* Returns with ID locked on success. */
1622 	if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1623 		return (error);
1624 
1625 	ih = &V_pf_idhash[PF_IDHASH(s)];
1626 	PF_HASHROW_ASSERT(ih);
1627 	LIST_FOREACH(cur, &ih->states, entry)
1628 		if (cur->id == s->id && cur->creatorid == s->creatorid)
1629 			break;
1630 
1631 	if (cur != NULL) {
1632 		s->timeout = PFTM_UNLINKED;
1633 		PF_HASHROW_UNLOCK(ih);
1634 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1635 			printf("pf: state ID collision: "
1636 			    "id: %016llx creatorid: %08x\n",
1637 			    (unsigned long long)be64toh(s->id),
1638 			    ntohl(s->creatorid));
1639 		}
1640 		pf_detach_state(s);
1641 		return (EEXIST);
1642 	}
1643 	LIST_INSERT_HEAD(&ih->states, s, entry);
1644 	/* One for keys, one for ID hash. */
1645 	refcount_init(&s->refs, 2);
1646 
1647 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1648 	if (V_pfsync_insert_state_ptr != NULL)
1649 		V_pfsync_insert_state_ptr(s);
1650 
1651 	/* Returns locked. */
1652 	return (0);
1653 }
1654 
1655 /*
1656  * Find state by ID: returns with locked row on success.
1657  */
1658 struct pf_kstate *
1659 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1660 {
1661 	struct pf_idhash *ih;
1662 	struct pf_kstate *s;
1663 
1664 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1665 
1666 	ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))];
1667 
1668 	PF_HASHROW_LOCK(ih);
1669 	LIST_FOREACH(s, &ih->states, entry)
1670 		if (s->id == id && s->creatorid == creatorid)
1671 			break;
1672 
1673 	if (s == NULL)
1674 		PF_HASHROW_UNLOCK(ih);
1675 
1676 	return (s);
1677 }
1678 
1679 /*
1680  * Find state by key.
1681  * Returns with ID hash slot locked on success.
1682  */
1683 static struct pf_kstate *
1684 pf_find_state(struct pfi_kkif *kif, const struct pf_state_key_cmp *key,
1685     u_int dir)
1686 {
1687 	struct pf_keyhash	*kh;
1688 	struct pf_state_key	*sk;
1689 	struct pf_kstate	*s;
1690 	int idx;
1691 
1692 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1693 
1694 	kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)];
1695 
1696 	PF_HASHROW_LOCK(kh);
1697 	LIST_FOREACH(sk, &kh->keys, entry)
1698 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1699 			break;
1700 	if (sk == NULL) {
1701 		PF_HASHROW_UNLOCK(kh);
1702 		return (NULL);
1703 	}
1704 
1705 	idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1706 
1707 	/* List is sorted, if-bound states before floating ones. */
1708 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1709 		if (s->kif == V_pfi_all || s->kif == kif || s->orig_kif == kif) {
1710 			PF_STATE_LOCK(s);
1711 			PF_HASHROW_UNLOCK(kh);
1712 			if (__predict_false(s->timeout >= PFTM_MAX)) {
1713 				/*
1714 				 * State is either being processed by
1715 				 * pf_unlink_state() in an other thread, or
1716 				 * is scheduled for immediate expiry.
1717 				 */
1718 				PF_STATE_UNLOCK(s);
1719 				return (NULL);
1720 			}
1721 			return (s);
1722 		}
1723 	PF_HASHROW_UNLOCK(kh);
1724 
1725 	return (NULL);
1726 }
1727 
1728 /*
1729  * Returns with ID hash slot locked on success.
1730  */
1731 struct pf_kstate *
1732 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more)
1733 {
1734 	struct pf_keyhash	*kh;
1735 	struct pf_state_key	*sk;
1736 	struct pf_kstate	*s, *ret = NULL;
1737 	int			 idx, inout = 0;
1738 
1739 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1740 
1741 	kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)];
1742 
1743 	PF_HASHROW_LOCK(kh);
1744 	LIST_FOREACH(sk, &kh->keys, entry)
1745 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1746 			break;
1747 	if (sk == NULL) {
1748 		PF_HASHROW_UNLOCK(kh);
1749 		return (NULL);
1750 	}
1751 	switch (dir) {
1752 	case PF_IN:
1753 		idx = PF_SK_WIRE;
1754 		break;
1755 	case PF_OUT:
1756 		idx = PF_SK_STACK;
1757 		break;
1758 	case PF_INOUT:
1759 		idx = PF_SK_WIRE;
1760 		inout = 1;
1761 		break;
1762 	default:
1763 		panic("%s: dir %u", __func__, dir);
1764 	}
1765 second_run:
1766 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1767 		if (more == NULL) {
1768 			PF_STATE_LOCK(s);
1769 			PF_HASHROW_UNLOCK(kh);
1770 			return (s);
1771 		}
1772 
1773 		if (ret)
1774 			(*more)++;
1775 		else {
1776 			ret = s;
1777 			PF_STATE_LOCK(s);
1778 		}
1779 	}
1780 	if (inout == 1) {
1781 		inout = 0;
1782 		idx = PF_SK_STACK;
1783 		goto second_run;
1784 	}
1785 	PF_HASHROW_UNLOCK(kh);
1786 
1787 	return (ret);
1788 }
1789 
1790 /*
1791  * FIXME
1792  * This routine is inefficient -- locks the state only to unlock immediately on
1793  * return.
1794  * It is racy -- after the state is unlocked nothing stops other threads from
1795  * removing it.
1796  */
1797 bool
1798 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir)
1799 {
1800 	struct pf_kstate *s;
1801 
1802 	s = pf_find_state_all(key, dir, NULL);
1803 	if (s != NULL) {
1804 		PF_STATE_UNLOCK(s);
1805 		return (true);
1806 	}
1807 	return (false);
1808 }
1809 
1810 /* END state table stuff */
1811 
1812 static void
1813 pf_send(struct pf_send_entry *pfse)
1814 {
1815 
1816 	PF_SENDQ_LOCK();
1817 	STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1818 	PF_SENDQ_UNLOCK();
1819 	swi_sched(V_pf_swi_cookie, 0);
1820 }
1821 
1822 static bool
1823 pf_isforlocal(struct mbuf *m, int af)
1824 {
1825 	switch (af) {
1826 #ifdef INET
1827 	case AF_INET: {
1828 		struct ip *ip = mtod(m, struct ip *);
1829 
1830 		return (in_localip(ip->ip_dst));
1831 	}
1832 #endif
1833 #ifdef INET6
1834 	case AF_INET6: {
1835 		struct ip6_hdr *ip6;
1836 		struct in6_ifaddr *ia;
1837 		ip6 = mtod(m, struct ip6_hdr *);
1838 		ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1839 		if (ia == NULL)
1840 			return (false);
1841 		return (! (ia->ia6_flags & IN6_IFF_NOTREADY));
1842 	}
1843 #endif
1844 	default:
1845 		panic("Unsupported af %d", af);
1846 	}
1847 
1848 	return (false);
1849 }
1850 
1851 int
1852 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type,
1853     int *icmp_dir, int *multi, u_int16_t *virtual_id, u_int16_t *virtual_type)
1854 {
1855 	/*
1856 	 * ICMP types marked with PF_OUT are typically responses to
1857 	 * PF_IN, and will match states in the opposite direction.
1858 	 * PF_IN ICMP types need to match a state with that type.
1859 	 */
1860 	*icmp_dir = PF_OUT;
1861 	*multi = PF_ICMP_MULTI_LINK;
1862 	/* Queries (and responses) */
1863 	switch (pd->af) {
1864 #ifdef INET
1865 	case AF_INET:
1866 		switch (type) {
1867 		case ICMP_ECHO:
1868 			*icmp_dir = PF_IN;
1869 		case ICMP_ECHOREPLY:
1870 			*virtual_type = ICMP_ECHO;
1871 			*virtual_id = pd->hdr.icmp.icmp_id;
1872 			break;
1873 
1874 		case ICMP_TSTAMP:
1875 			*icmp_dir = PF_IN;
1876 		case ICMP_TSTAMPREPLY:
1877 			*virtual_type = ICMP_TSTAMP;
1878 			*virtual_id = pd->hdr.icmp.icmp_id;
1879 			break;
1880 
1881 		case ICMP_IREQ:
1882 			*icmp_dir = PF_IN;
1883 		case ICMP_IREQREPLY:
1884 			*virtual_type = ICMP_IREQ;
1885 			*virtual_id = pd->hdr.icmp.icmp_id;
1886 			break;
1887 
1888 		case ICMP_MASKREQ:
1889 			*icmp_dir = PF_IN;
1890 		case ICMP_MASKREPLY:
1891 			*virtual_type = ICMP_MASKREQ;
1892 			*virtual_id = pd->hdr.icmp.icmp_id;
1893 			break;
1894 
1895 		case ICMP_IPV6_WHEREAREYOU:
1896 			*icmp_dir = PF_IN;
1897 		case ICMP_IPV6_IAMHERE:
1898 			*virtual_type = ICMP_IPV6_WHEREAREYOU;
1899 			*virtual_id = 0; /* Nothing sane to match on! */
1900 			break;
1901 
1902 		case ICMP_MOBILE_REGREQUEST:
1903 			*icmp_dir = PF_IN;
1904 		case ICMP_MOBILE_REGREPLY:
1905 			*virtual_type = ICMP_MOBILE_REGREQUEST;
1906 			*virtual_id = 0; /* Nothing sane to match on! */
1907 			break;
1908 
1909 		case ICMP_ROUTERSOLICIT:
1910 			*icmp_dir = PF_IN;
1911 		case ICMP_ROUTERADVERT:
1912 			*virtual_type = ICMP_ROUTERSOLICIT;
1913 			*virtual_id = 0; /* Nothing sane to match on! */
1914 			break;
1915 
1916 		/* These ICMP types map to other connections */
1917 		case ICMP_UNREACH:
1918 		case ICMP_SOURCEQUENCH:
1919 		case ICMP_REDIRECT:
1920 		case ICMP_TIMXCEED:
1921 		case ICMP_PARAMPROB:
1922 			/* These will not be used, but set them anyway */
1923 			*icmp_dir = PF_IN;
1924 			*virtual_type = type;
1925 			*virtual_id = 0;
1926 			HTONS(*virtual_type);
1927 			return (1);  /* These types match to another state */
1928 
1929 		/*
1930 		 * All remaining ICMP types get their own states,
1931 		 * and will only match in one direction.
1932 		 */
1933 		default:
1934 			*icmp_dir = PF_IN;
1935 			*virtual_type = type;
1936 			*virtual_id = 0;
1937 			break;
1938 		}
1939 		break;
1940 #endif /* INET */
1941 #ifdef INET6
1942 	case AF_INET6:
1943 		switch (type) {
1944 		case ICMP6_ECHO_REQUEST:
1945 			*icmp_dir = PF_IN;
1946 		case ICMP6_ECHO_REPLY:
1947 			*virtual_type = ICMP6_ECHO_REQUEST;
1948 			*virtual_id = pd->hdr.icmp6.icmp6_id;
1949 			break;
1950 
1951 		case MLD_LISTENER_QUERY:
1952 		case MLD_LISTENER_REPORT: {
1953 			/*
1954 			 * Listener Report can be sent by clients
1955 			 * without an associated Listener Query.
1956 			 * In addition to that, when Report is sent as a
1957 			 * reply to a Query its source and destination
1958 			 * address are different.
1959 			 */
1960 			*icmp_dir = PF_IN;
1961 			*virtual_type = MLD_LISTENER_QUERY;
1962 			*virtual_id = 0;
1963 			break;
1964 		}
1965 		case MLD_MTRACE:
1966 			*icmp_dir = PF_IN;
1967 		case MLD_MTRACE_RESP:
1968 			*virtual_type = MLD_MTRACE;
1969 			*virtual_id = 0; /* Nothing sane to match on! */
1970 			break;
1971 
1972 		case ND_NEIGHBOR_SOLICIT:
1973 			*icmp_dir = PF_IN;
1974 		case ND_NEIGHBOR_ADVERT: {
1975 			*virtual_type = ND_NEIGHBOR_SOLICIT;
1976 			*virtual_id = 0;
1977 			break;
1978 		}
1979 
1980 		/*
1981 		 * These ICMP types map to other connections.
1982 		 * ND_REDIRECT can't be in this list because the triggering
1983 		 * packet header is optional.
1984 		 */
1985 		case ICMP6_DST_UNREACH:
1986 		case ICMP6_PACKET_TOO_BIG:
1987 		case ICMP6_TIME_EXCEEDED:
1988 		case ICMP6_PARAM_PROB:
1989 			/* These will not be used, but set them anyway */
1990 			*icmp_dir = PF_IN;
1991 			*virtual_type = type;
1992 			*virtual_id = 0;
1993 			HTONS(*virtual_type);
1994 			return (1);  /* These types match to another state */
1995 		/*
1996 		 * All remaining ICMP6 types get their own states,
1997 		 * and will only match in one direction.
1998 		 */
1999 		default:
2000 			*icmp_dir = PF_IN;
2001 			*virtual_type = type;
2002 			*virtual_id = 0;
2003 			break;
2004 		}
2005 		break;
2006 #endif /* INET6 */
2007 	default:
2008 		*icmp_dir = PF_IN;
2009 		*virtual_type = type;
2010 		*virtual_id = 0;
2011 		break;
2012 	}
2013 	HTONS(*virtual_type);
2014 	return (0);  /* These types match to their own state */
2015 }
2016 
2017 void
2018 pf_intr(void *v)
2019 {
2020 	struct epoch_tracker et;
2021 	struct pf_send_head queue;
2022 	struct pf_send_entry *pfse, *next;
2023 
2024 	CURVNET_SET((struct vnet *)v);
2025 
2026 	PF_SENDQ_LOCK();
2027 	queue = V_pf_sendqueue;
2028 	STAILQ_INIT(&V_pf_sendqueue);
2029 	PF_SENDQ_UNLOCK();
2030 
2031 	NET_EPOCH_ENTER(et);
2032 
2033 	STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
2034 		switch (pfse->pfse_type) {
2035 #ifdef INET
2036 		case PFSE_IP: {
2037 			if (pf_isforlocal(pfse->pfse_m, AF_INET)) {
2038 				pfse->pfse_m->m_flags |= M_SKIP_FIREWALL;
2039 				pfse->pfse_m->m_pkthdr.csum_flags |=
2040 				    CSUM_IP_VALID | CSUM_IP_CHECKED;
2041 				ip_input(pfse->pfse_m);
2042 			} else {
2043 				ip_output(pfse->pfse_m, NULL, NULL, 0, NULL,
2044 				    NULL);
2045 			}
2046 			break;
2047 		}
2048 		case PFSE_ICMP:
2049 			icmp_error(pfse->pfse_m, pfse->icmpopts.type,
2050 			    pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
2051 			break;
2052 #endif /* INET */
2053 #ifdef INET6
2054 		case PFSE_IP6:
2055 			if (pf_isforlocal(pfse->pfse_m, AF_INET6)) {
2056 				pfse->pfse_m->m_flags |= M_SKIP_FIREWALL;
2057 				ip6_input(pfse->pfse_m);
2058 			} else {
2059 				ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL,
2060 				    NULL, NULL);
2061 			}
2062 			break;
2063 		case PFSE_ICMP6:
2064 			icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
2065 			    pfse->icmpopts.code, pfse->icmpopts.mtu);
2066 			break;
2067 #endif /* INET6 */
2068 		default:
2069 			panic("%s: unknown type", __func__);
2070 		}
2071 		free(pfse, M_PFTEMP);
2072 	}
2073 	NET_EPOCH_EXIT(et);
2074 	CURVNET_RESTORE();
2075 }
2076 
2077 #define	pf_purge_thread_period	(hz / 10)
2078 
2079 #ifdef PF_WANT_32_TO_64_COUNTER
2080 static void
2081 pf_status_counter_u64_periodic(void)
2082 {
2083 
2084 	PF_RULES_RASSERT();
2085 
2086 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) {
2087 		return;
2088 	}
2089 
2090 	for (int i = 0; i < FCNT_MAX; i++) {
2091 		pf_counter_u64_periodic(&V_pf_status.fcounters[i]);
2092 	}
2093 }
2094 
2095 static void
2096 pf_kif_counter_u64_periodic(void)
2097 {
2098 	struct pfi_kkif *kif;
2099 	size_t r, run;
2100 
2101 	PF_RULES_RASSERT();
2102 
2103 	if (__predict_false(V_pf_allkifcount == 0)) {
2104 		return;
2105 	}
2106 
2107 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) {
2108 		return;
2109 	}
2110 
2111 	run = V_pf_allkifcount / 10;
2112 	if (run < 5)
2113 		run = 5;
2114 
2115 	for (r = 0; r < run; r++) {
2116 		kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist);
2117 		if (kif == NULL) {
2118 			LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist);
2119 			LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist);
2120 			break;
2121 		}
2122 
2123 		LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist);
2124 		LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist);
2125 
2126 		for (int i = 0; i < 2; i++) {
2127 			for (int j = 0; j < 2; j++) {
2128 				for (int k = 0; k < 2; k++) {
2129 					pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]);
2130 					pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]);
2131 				}
2132 			}
2133 		}
2134 	}
2135 }
2136 
2137 static void
2138 pf_rule_counter_u64_periodic(void)
2139 {
2140 	struct pf_krule *rule;
2141 	size_t r, run;
2142 
2143 	PF_RULES_RASSERT();
2144 
2145 	if (__predict_false(V_pf_allrulecount == 0)) {
2146 		return;
2147 	}
2148 
2149 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) {
2150 		return;
2151 	}
2152 
2153 	run = V_pf_allrulecount / 10;
2154 	if (run < 5)
2155 		run = 5;
2156 
2157 	for (r = 0; r < run; r++) {
2158 		rule = LIST_NEXT(V_pf_rulemarker, allrulelist);
2159 		if (rule == NULL) {
2160 			LIST_REMOVE(V_pf_rulemarker, allrulelist);
2161 			LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist);
2162 			break;
2163 		}
2164 
2165 		LIST_REMOVE(V_pf_rulemarker, allrulelist);
2166 		LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist);
2167 
2168 		pf_counter_u64_periodic(&rule->evaluations);
2169 		for (int i = 0; i < 2; i++) {
2170 			pf_counter_u64_periodic(&rule->packets[i]);
2171 			pf_counter_u64_periodic(&rule->bytes[i]);
2172 		}
2173 	}
2174 }
2175 
2176 static void
2177 pf_counter_u64_periodic_main(void)
2178 {
2179 	PF_RULES_RLOCK_TRACKER;
2180 
2181 	V_pf_counter_periodic_iter++;
2182 
2183 	PF_RULES_RLOCK();
2184 	pf_counter_u64_critical_enter();
2185 	pf_status_counter_u64_periodic();
2186 	pf_kif_counter_u64_periodic();
2187 	pf_rule_counter_u64_periodic();
2188 	pf_counter_u64_critical_exit();
2189 	PF_RULES_RUNLOCK();
2190 }
2191 #else
2192 #define	pf_counter_u64_periodic_main()	do { } while (0)
2193 #endif
2194 
2195 void
2196 pf_purge_thread(void *unused __unused)
2197 {
2198 	struct epoch_tracker	 et;
2199 
2200 	VNET_ITERATOR_DECL(vnet_iter);
2201 
2202 	sx_xlock(&pf_end_lock);
2203 	while (pf_end_threads == 0) {
2204 		sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period);
2205 
2206 		VNET_LIST_RLOCK();
2207 		NET_EPOCH_ENTER(et);
2208 		VNET_FOREACH(vnet_iter) {
2209 			CURVNET_SET(vnet_iter);
2210 
2211 			/* Wait until V_pf_default_rule is initialized. */
2212 			if (V_pf_vnet_active == 0) {
2213 				CURVNET_RESTORE();
2214 				continue;
2215 			}
2216 
2217 			pf_counter_u64_periodic_main();
2218 
2219 			/*
2220 			 *  Process 1/interval fraction of the state
2221 			 * table every run.
2222 			 */
2223 			V_pf_purge_idx =
2224 			    pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask /
2225 			    (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
2226 
2227 			/*
2228 			 * Purge other expired types every
2229 			 * PFTM_INTERVAL seconds.
2230 			 */
2231 			if (V_pf_purge_idx == 0) {
2232 				/*
2233 				 * Order is important:
2234 				 * - states and src nodes reference rules
2235 				 * - states and rules reference kifs
2236 				 */
2237 				pf_purge_expired_fragments();
2238 				pf_purge_expired_src_nodes();
2239 				pf_purge_unlinked_rules();
2240 				pfi_kkif_purge();
2241 			}
2242 			CURVNET_RESTORE();
2243 		}
2244 		NET_EPOCH_EXIT(et);
2245 		VNET_LIST_RUNLOCK();
2246 	}
2247 
2248 	pf_end_threads++;
2249 	sx_xunlock(&pf_end_lock);
2250 	kproc_exit(0);
2251 }
2252 
2253 void
2254 pf_unload_vnet_purge(void)
2255 {
2256 
2257 	/*
2258 	 * To cleanse up all kifs and rules we need
2259 	 * two runs: first one clears reference flags,
2260 	 * then pf_purge_expired_states() doesn't
2261 	 * raise them, and then second run frees.
2262 	 */
2263 	pf_purge_unlinked_rules();
2264 	pfi_kkif_purge();
2265 
2266 	/*
2267 	 * Now purge everything.
2268 	 */
2269 	pf_purge_expired_states(0, V_pf_hashmask);
2270 	pf_purge_fragments(UINT_MAX);
2271 	pf_purge_expired_src_nodes();
2272 
2273 	/*
2274 	 * Now all kifs & rules should be unreferenced,
2275 	 * thus should be successfully freed.
2276 	 */
2277 	pf_purge_unlinked_rules();
2278 	pfi_kkif_purge();
2279 }
2280 
2281 u_int32_t
2282 pf_state_expires(const struct pf_kstate *state)
2283 {
2284 	u_int32_t	timeout;
2285 	u_int32_t	start;
2286 	u_int32_t	end;
2287 	u_int32_t	states;
2288 
2289 	/* handle all PFTM_* > PFTM_MAX here */
2290 	if (state->timeout == PFTM_PURGE)
2291 		return (time_uptime);
2292 	KASSERT(state->timeout != PFTM_UNLINKED,
2293 	    ("pf_state_expires: timeout == PFTM_UNLINKED"));
2294 	KASSERT((state->timeout < PFTM_MAX),
2295 	    ("pf_state_expires: timeout > PFTM_MAX"));
2296 	timeout = state->rule.ptr->timeout[state->timeout];
2297 	if (!timeout)
2298 		timeout = V_pf_default_rule.timeout[state->timeout];
2299 	start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
2300 	if (start && state->rule.ptr != &V_pf_default_rule) {
2301 		end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
2302 		states = counter_u64_fetch(state->rule.ptr->states_cur);
2303 	} else {
2304 		start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
2305 		end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
2306 		states = V_pf_status.states;
2307 	}
2308 	if (end && states > start && start < end) {
2309 		if (states < end) {
2310 			timeout = (u_int64_t)timeout * (end - states) /
2311 			    (end - start);
2312 			return ((state->expire / 1000) + timeout);
2313 		}
2314 		else
2315 			return (time_uptime);
2316 	}
2317 	return ((state->expire / 1000) + timeout);
2318 }
2319 
2320 void
2321 pf_purge_expired_src_nodes(void)
2322 {
2323 	struct pf_ksrc_node_list	 freelist;
2324 	struct pf_srchash	*sh;
2325 	struct pf_ksrc_node	*cur, *next;
2326 	int i;
2327 
2328 	LIST_INIT(&freelist);
2329 	for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
2330 	    PF_HASHROW_LOCK(sh);
2331 	    LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
2332 		if (cur->states == 0 && cur->expire <= time_uptime) {
2333 			pf_unlink_src_node(cur);
2334 			LIST_INSERT_HEAD(&freelist, cur, entry);
2335 		} else if (cur->rule.ptr != NULL)
2336 			cur->rule.ptr->rule_ref |= PFRULE_REFS;
2337 	    PF_HASHROW_UNLOCK(sh);
2338 	}
2339 
2340 	pf_free_src_nodes(&freelist);
2341 
2342 	V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
2343 }
2344 
2345 static void
2346 pf_src_tree_remove_state(struct pf_kstate *s)
2347 {
2348 	struct pf_ksrc_node *sn;
2349 	uint32_t timeout;
2350 
2351 	timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
2352 	    s->rule.ptr->timeout[PFTM_SRC_NODE] :
2353 	    V_pf_default_rule.timeout[PFTM_SRC_NODE];
2354 
2355 	if (s->src_node != NULL) {
2356 		sn = s->src_node;
2357 		PF_SRC_NODE_LOCK(sn);
2358 		if (s->src.tcp_est)
2359 			--sn->conn;
2360 		if (--sn->states == 0)
2361 			sn->expire = time_uptime + timeout;
2362 		PF_SRC_NODE_UNLOCK(sn);
2363 	}
2364 	if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
2365 		sn = s->nat_src_node;
2366 		PF_SRC_NODE_LOCK(sn);
2367 		if (--sn->states == 0)
2368 			sn->expire = time_uptime + timeout;
2369 		PF_SRC_NODE_UNLOCK(sn);
2370 	}
2371 	s->src_node = s->nat_src_node = NULL;
2372 }
2373 
2374 /*
2375  * Unlink and potentilly free a state. Function may be
2376  * called with ID hash row locked, but always returns
2377  * unlocked, since it needs to go through key hash locking.
2378  */
2379 int
2380 pf_unlink_state(struct pf_kstate *s)
2381 {
2382 	struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
2383 
2384 	NET_EPOCH_ASSERT();
2385 	PF_HASHROW_ASSERT(ih);
2386 
2387 	if (s->timeout == PFTM_UNLINKED) {
2388 		/*
2389 		 * State is being processed
2390 		 * by pf_unlink_state() in
2391 		 * an other thread.
2392 		 */
2393 		PF_HASHROW_UNLOCK(ih);
2394 		return (0);	/* XXXGL: undefined actually */
2395 	}
2396 
2397 	if (s->src.state == PF_TCPS_PROXY_DST) {
2398 		/* XXX wire key the right one? */
2399 		pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af,
2400 		    &s->key[PF_SK_WIRE]->addr[1],
2401 		    &s->key[PF_SK_WIRE]->addr[0],
2402 		    s->key[PF_SK_WIRE]->port[1],
2403 		    s->key[PF_SK_WIRE]->port[0],
2404 		    s->src.seqhi, s->src.seqlo + 1,
2405 		    TH_RST|TH_ACK, 0, 0, 0, true, s->tag, 0, s->act.rtableid);
2406 	}
2407 
2408 	LIST_REMOVE(s, entry);
2409 	pf_src_tree_remove_state(s);
2410 
2411 	if (V_pfsync_delete_state_ptr != NULL)
2412 		V_pfsync_delete_state_ptr(s);
2413 
2414 	STATE_DEC_COUNTERS(s);
2415 
2416 	s->timeout = PFTM_UNLINKED;
2417 
2418 	/* Ensure we remove it from the list of halfopen states, if needed. */
2419 	if (s->key[PF_SK_STACK] != NULL &&
2420 	    s->key[PF_SK_STACK]->proto == IPPROTO_TCP)
2421 		pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED);
2422 
2423 	PF_HASHROW_UNLOCK(ih);
2424 
2425 	pf_detach_state(s);
2426 	/* pf_state_insert() initialises refs to 2 */
2427 	return (pf_release_staten(s, 2));
2428 }
2429 
2430 struct pf_kstate *
2431 pf_alloc_state(int flags)
2432 {
2433 
2434 	return (uma_zalloc(V_pf_state_z, flags | M_ZERO));
2435 }
2436 
2437 void
2438 pf_free_state(struct pf_kstate *cur)
2439 {
2440 	struct pf_krule_item *ri;
2441 
2442 	KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
2443 	KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
2444 	    cur->timeout));
2445 
2446 	while ((ri = SLIST_FIRST(&cur->match_rules))) {
2447 		SLIST_REMOVE_HEAD(&cur->match_rules, entry);
2448 		free(ri, M_PF_RULE_ITEM);
2449 	}
2450 
2451 	pf_normalize_tcp_cleanup(cur);
2452 	uma_zfree(V_pf_state_z, cur);
2453 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
2454 }
2455 
2456 /*
2457  * Called only from pf_purge_thread(), thus serialized.
2458  */
2459 static u_int
2460 pf_purge_expired_states(u_int i, int maxcheck)
2461 {
2462 	struct pf_idhash *ih;
2463 	struct pf_kstate *s;
2464 	struct pf_krule_item *mrm;
2465 	size_t count __unused;
2466 
2467 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2468 
2469 	/*
2470 	 * Go through hash and unlink states that expire now.
2471 	 */
2472 	while (maxcheck > 0) {
2473 		count = 0;
2474 		ih = &V_pf_idhash[i];
2475 
2476 		/* only take the lock if we expect to do work */
2477 		if (!LIST_EMPTY(&ih->states)) {
2478 relock:
2479 			PF_HASHROW_LOCK(ih);
2480 			LIST_FOREACH(s, &ih->states, entry) {
2481 				if (pf_state_expires(s) <= time_uptime) {
2482 					V_pf_status.states -=
2483 					    pf_unlink_state(s);
2484 					goto relock;
2485 				}
2486 				s->rule.ptr->rule_ref |= PFRULE_REFS;
2487 				if (s->nat_rule.ptr != NULL)
2488 					s->nat_rule.ptr->rule_ref |= PFRULE_REFS;
2489 				if (s->anchor.ptr != NULL)
2490 					s->anchor.ptr->rule_ref |= PFRULE_REFS;
2491 				s->kif->pfik_flags |= PFI_IFLAG_REFS;
2492 				SLIST_FOREACH(mrm, &s->match_rules, entry)
2493 					mrm->r->rule_ref |= PFRULE_REFS;
2494 				if (s->rt_kif)
2495 					s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
2496 				count++;
2497 			}
2498 			PF_HASHROW_UNLOCK(ih);
2499 		}
2500 
2501 		SDT_PROBE2(pf, purge, state, rowcount, i, count);
2502 
2503 		/* Return when we hit end of hash. */
2504 		if (++i > V_pf_hashmask) {
2505 			V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2506 			return (0);
2507 		}
2508 
2509 		maxcheck--;
2510 	}
2511 
2512 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2513 
2514 	return (i);
2515 }
2516 
2517 static void
2518 pf_purge_unlinked_rules(void)
2519 {
2520 	struct pf_krulequeue tmpq;
2521 	struct pf_krule *r, *r1;
2522 
2523 	/*
2524 	 * If we have overloading task pending, then we'd
2525 	 * better skip purging this time. There is a tiny
2526 	 * probability that overloading task references
2527 	 * an already unlinked rule.
2528 	 */
2529 	PF_OVERLOADQ_LOCK();
2530 	if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
2531 		PF_OVERLOADQ_UNLOCK();
2532 		return;
2533 	}
2534 	PF_OVERLOADQ_UNLOCK();
2535 
2536 	/*
2537 	 * Do naive mark-and-sweep garbage collecting of old rules.
2538 	 * Reference flag is raised by pf_purge_expired_states()
2539 	 * and pf_purge_expired_src_nodes().
2540 	 *
2541 	 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
2542 	 * use a temporary queue.
2543 	 */
2544 	TAILQ_INIT(&tmpq);
2545 	PF_UNLNKDRULES_LOCK();
2546 	TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
2547 		if (!(r->rule_ref & PFRULE_REFS)) {
2548 			TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
2549 			TAILQ_INSERT_TAIL(&tmpq, r, entries);
2550 		} else
2551 			r->rule_ref &= ~PFRULE_REFS;
2552 	}
2553 	PF_UNLNKDRULES_UNLOCK();
2554 
2555 	if (!TAILQ_EMPTY(&tmpq)) {
2556 		PF_CONFIG_LOCK();
2557 		PF_RULES_WLOCK();
2558 		TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
2559 			TAILQ_REMOVE(&tmpq, r, entries);
2560 			pf_free_rule(r);
2561 		}
2562 		PF_RULES_WUNLOCK();
2563 		PF_CONFIG_UNLOCK();
2564 	}
2565 }
2566 
2567 void
2568 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
2569 {
2570 	switch (af) {
2571 #ifdef INET
2572 	case AF_INET: {
2573 		u_int32_t a = ntohl(addr->addr32[0]);
2574 		printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
2575 		    (a>>8)&255, a&255);
2576 		if (p) {
2577 			p = ntohs(p);
2578 			printf(":%u", p);
2579 		}
2580 		break;
2581 	}
2582 #endif /* INET */
2583 #ifdef INET6
2584 	case AF_INET6: {
2585 		u_int16_t b;
2586 		u_int8_t i, curstart, curend, maxstart, maxend;
2587 		curstart = curend = maxstart = maxend = 255;
2588 		for (i = 0; i < 8; i++) {
2589 			if (!addr->addr16[i]) {
2590 				if (curstart == 255)
2591 					curstart = i;
2592 				curend = i;
2593 			} else {
2594 				if ((curend - curstart) >
2595 				    (maxend - maxstart)) {
2596 					maxstart = curstart;
2597 					maxend = curend;
2598 				}
2599 				curstart = curend = 255;
2600 			}
2601 		}
2602 		if ((curend - curstart) >
2603 		    (maxend - maxstart)) {
2604 			maxstart = curstart;
2605 			maxend = curend;
2606 		}
2607 		for (i = 0; i < 8; i++) {
2608 			if (i >= maxstart && i <= maxend) {
2609 				if (i == 0)
2610 					printf(":");
2611 				if (i == maxend)
2612 					printf(":");
2613 			} else {
2614 				b = ntohs(addr->addr16[i]);
2615 				printf("%x", b);
2616 				if (i < 7)
2617 					printf(":");
2618 			}
2619 		}
2620 		if (p) {
2621 			p = ntohs(p);
2622 			printf("[%u]", p);
2623 		}
2624 		break;
2625 	}
2626 #endif /* INET6 */
2627 	}
2628 }
2629 
2630 void
2631 pf_print_state(struct pf_kstate *s)
2632 {
2633 	pf_print_state_parts(s, NULL, NULL);
2634 }
2635 
2636 static void
2637 pf_print_state_parts(struct pf_kstate *s,
2638     struct pf_state_key *skwp, struct pf_state_key *sksp)
2639 {
2640 	struct pf_state_key *skw, *sks;
2641 	u_int8_t proto, dir;
2642 
2643 	/* Do our best to fill these, but they're skipped if NULL */
2644 	skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
2645 	sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
2646 	proto = skw ? skw->proto : (sks ? sks->proto : 0);
2647 	dir = s ? s->direction : 0;
2648 
2649 	switch (proto) {
2650 	case IPPROTO_IPV4:
2651 		printf("IPv4");
2652 		break;
2653 	case IPPROTO_IPV6:
2654 		printf("IPv6");
2655 		break;
2656 	case IPPROTO_TCP:
2657 		printf("TCP");
2658 		break;
2659 	case IPPROTO_UDP:
2660 		printf("UDP");
2661 		break;
2662 	case IPPROTO_ICMP:
2663 		printf("ICMP");
2664 		break;
2665 	case IPPROTO_ICMPV6:
2666 		printf("ICMPv6");
2667 		break;
2668 	default:
2669 		printf("%u", proto);
2670 		break;
2671 	}
2672 	switch (dir) {
2673 	case PF_IN:
2674 		printf(" in");
2675 		break;
2676 	case PF_OUT:
2677 		printf(" out");
2678 		break;
2679 	}
2680 	if (skw) {
2681 		printf(" wire: ");
2682 		pf_print_host(&skw->addr[0], skw->port[0], skw->af);
2683 		printf(" ");
2684 		pf_print_host(&skw->addr[1], skw->port[1], skw->af);
2685 	}
2686 	if (sks) {
2687 		printf(" stack: ");
2688 		if (sks != skw) {
2689 			pf_print_host(&sks->addr[0], sks->port[0], sks->af);
2690 			printf(" ");
2691 			pf_print_host(&sks->addr[1], sks->port[1], sks->af);
2692 		} else
2693 			printf("-");
2694 	}
2695 	if (s) {
2696 		if (proto == IPPROTO_TCP) {
2697 			printf(" [lo=%u high=%u win=%u modulator=%u",
2698 			    s->src.seqlo, s->src.seqhi,
2699 			    s->src.max_win, s->src.seqdiff);
2700 			if (s->src.wscale && s->dst.wscale)
2701 				printf(" wscale=%u",
2702 				    s->src.wscale & PF_WSCALE_MASK);
2703 			printf("]");
2704 			printf(" [lo=%u high=%u win=%u modulator=%u",
2705 			    s->dst.seqlo, s->dst.seqhi,
2706 			    s->dst.max_win, s->dst.seqdiff);
2707 			if (s->src.wscale && s->dst.wscale)
2708 				printf(" wscale=%u",
2709 				s->dst.wscale & PF_WSCALE_MASK);
2710 			printf("]");
2711 		}
2712 		printf(" %u:%u", s->src.state, s->dst.state);
2713 	}
2714 }
2715 
2716 void
2717 pf_print_flags(u_int8_t f)
2718 {
2719 	if (f)
2720 		printf(" ");
2721 	if (f & TH_FIN)
2722 		printf("F");
2723 	if (f & TH_SYN)
2724 		printf("S");
2725 	if (f & TH_RST)
2726 		printf("R");
2727 	if (f & TH_PUSH)
2728 		printf("P");
2729 	if (f & TH_ACK)
2730 		printf("A");
2731 	if (f & TH_URG)
2732 		printf("U");
2733 	if (f & TH_ECE)
2734 		printf("E");
2735 	if (f & TH_CWR)
2736 		printf("W");
2737 }
2738 
2739 #define	PF_SET_SKIP_STEPS(i)					\
2740 	do {							\
2741 		while (head[i] != cur) {			\
2742 			head[i]->skip[i].ptr = cur;		\
2743 			head[i] = TAILQ_NEXT(head[i], entries);	\
2744 		}						\
2745 	} while (0)
2746 
2747 void
2748 pf_calc_skip_steps(struct pf_krulequeue *rules)
2749 {
2750 	struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT];
2751 	int i;
2752 
2753 	cur = TAILQ_FIRST(rules);
2754 	prev = cur;
2755 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2756 		head[i] = cur;
2757 	while (cur != NULL) {
2758 		if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
2759 			PF_SET_SKIP_STEPS(PF_SKIP_IFP);
2760 		if (cur->direction != prev->direction)
2761 			PF_SET_SKIP_STEPS(PF_SKIP_DIR);
2762 		if (cur->af != prev->af)
2763 			PF_SET_SKIP_STEPS(PF_SKIP_AF);
2764 		if (cur->proto != prev->proto)
2765 			PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
2766 		if (cur->src.neg != prev->src.neg ||
2767 		    pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
2768 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
2769 		if (cur->src.port[0] != prev->src.port[0] ||
2770 		    cur->src.port[1] != prev->src.port[1] ||
2771 		    cur->src.port_op != prev->src.port_op)
2772 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
2773 		if (cur->dst.neg != prev->dst.neg ||
2774 		    pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
2775 			PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
2776 		if (cur->dst.port[0] != prev->dst.port[0] ||
2777 		    cur->dst.port[1] != prev->dst.port[1] ||
2778 		    cur->dst.port_op != prev->dst.port_op)
2779 			PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
2780 
2781 		prev = cur;
2782 		cur = TAILQ_NEXT(cur, entries);
2783 	}
2784 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2785 		PF_SET_SKIP_STEPS(i);
2786 }
2787 
2788 int
2789 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
2790 {
2791 	if (aw1->type != aw2->type)
2792 		return (1);
2793 	switch (aw1->type) {
2794 	case PF_ADDR_ADDRMASK:
2795 	case PF_ADDR_RANGE:
2796 		if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
2797 			return (1);
2798 		if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
2799 			return (1);
2800 		return (0);
2801 	case PF_ADDR_DYNIFTL:
2802 		return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
2803 	case PF_ADDR_NOROUTE:
2804 	case PF_ADDR_URPFFAILED:
2805 		return (0);
2806 	case PF_ADDR_TABLE:
2807 		return (aw1->p.tbl != aw2->p.tbl);
2808 	default:
2809 		printf("invalid address type: %d\n", aw1->type);
2810 		return (1);
2811 	}
2812 }
2813 
2814 /**
2815  * Checksum updates are a little complicated because the checksum in the TCP/UDP
2816  * header isn't always a full checksum. In some cases (i.e. output) it's a
2817  * pseudo-header checksum, which is a partial checksum over src/dst IP
2818  * addresses, protocol number and length.
2819  *
2820  * That means we have the following cases:
2821  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2822  *  	checksums, we need to update the checksum whenever we change anything.
2823  *  * Output (i.e. the checksum is a pseudo-header checksum):
2824  *  	x The field being updated is src/dst address or affects the length of
2825  *  	the packet. We need to update the pseudo-header checksum (note that this
2826  *  	checksum is not ones' complement).
2827  *  	x Some other field is being modified (e.g. src/dst port numbers): We
2828  *  	don't have to update anything.
2829  **/
2830 u_int16_t
2831 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2832 {
2833 	u_int32_t x;
2834 
2835 	x = cksum + old - new;
2836 	x = (x + (x >> 16)) & 0xffff;
2837 
2838 	/* optimise: eliminate a branch when not udp */
2839 	if (udp && cksum == 0x0000)
2840 		return cksum;
2841 	if (udp && x == 0x0000)
2842 		x = 0xffff;
2843 
2844 	return (u_int16_t)(x);
2845 }
2846 
2847 static void
2848 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi,
2849     u_int8_t udp)
2850 {
2851 	u_int16_t old = htons(hi ? (*f << 8) : *f);
2852 	u_int16_t new = htons(hi ? ( v << 8) :  v);
2853 
2854 	if (*f == v)
2855 		return;
2856 
2857 	*f = v;
2858 
2859 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2860 		return;
2861 
2862 	*cksum = pf_cksum_fixup(*cksum, old, new, udp);
2863 }
2864 
2865 void
2866 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v,
2867     bool hi, u_int8_t udp)
2868 {
2869 	u_int8_t *fb = (u_int8_t *)f;
2870 	u_int8_t *vb = (u_int8_t *)&v;
2871 
2872 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2873 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2874 }
2875 
2876 void
2877 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v,
2878     bool hi, u_int8_t udp)
2879 {
2880 	u_int8_t *fb = (u_int8_t *)f;
2881 	u_int8_t *vb = (u_int8_t *)&v;
2882 
2883 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2884 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2885 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2886 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2887 }
2888 
2889 u_int16_t
2890 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2891         u_int16_t new, u_int8_t udp)
2892 {
2893 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2894 		return (cksum);
2895 
2896 	return (pf_cksum_fixup(cksum, old, new, udp));
2897 }
2898 
2899 static void
2900 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2901         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2902         sa_family_t af)
2903 {
2904 	struct pf_addr	ao;
2905 	u_int16_t	po = *p;
2906 
2907 	PF_ACPY(&ao, a, af);
2908 	PF_ACPY(a, an, af);
2909 
2910 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2911 		*pc = ~*pc;
2912 
2913 	*p = pn;
2914 
2915 	switch (af) {
2916 #ifdef INET
2917 	case AF_INET:
2918 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2919 		    ao.addr16[0], an->addr16[0], 0),
2920 		    ao.addr16[1], an->addr16[1], 0);
2921 		*p = pn;
2922 
2923 		*pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2924 		    ao.addr16[0], an->addr16[0], u),
2925 		    ao.addr16[1], an->addr16[1], u);
2926 
2927 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2928 		break;
2929 #endif /* INET */
2930 #ifdef INET6
2931 	case AF_INET6:
2932 		*pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2933 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2934 		    pf_cksum_fixup(pf_cksum_fixup(*pc,
2935 		    ao.addr16[0], an->addr16[0], u),
2936 		    ao.addr16[1], an->addr16[1], u),
2937 		    ao.addr16[2], an->addr16[2], u),
2938 		    ao.addr16[3], an->addr16[3], u),
2939 		    ao.addr16[4], an->addr16[4], u),
2940 		    ao.addr16[5], an->addr16[5], u),
2941 		    ao.addr16[6], an->addr16[6], u),
2942 		    ao.addr16[7], an->addr16[7], u);
2943 
2944 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2945 		break;
2946 #endif /* INET6 */
2947 	}
2948 
2949 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2950 	    CSUM_DELAY_DATA_IPV6)) {
2951 		*pc = ~*pc;
2952 		if (! *pc)
2953 			*pc = 0xffff;
2954 	}
2955 }
2956 
2957 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2958 void
2959 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2960 {
2961 	u_int32_t	ao;
2962 
2963 	memcpy(&ao, a, sizeof(ao));
2964 	memcpy(a, &an, sizeof(u_int32_t));
2965 	*c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2966 	    ao % 65536, an % 65536, u);
2967 }
2968 
2969 void
2970 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2971 {
2972 	u_int32_t	ao;
2973 
2974 	memcpy(&ao, a, sizeof(ao));
2975 	memcpy(a, &an, sizeof(u_int32_t));
2976 
2977 	*c = pf_proto_cksum_fixup(m,
2978 	    pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2979 	    ao % 65536, an % 65536, udp);
2980 }
2981 
2982 #ifdef INET6
2983 static void
2984 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2985 {
2986 	struct pf_addr	ao;
2987 
2988 	PF_ACPY(&ao, a, AF_INET6);
2989 	PF_ACPY(a, an, AF_INET6);
2990 
2991 	*c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2992 	    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2993 	    pf_cksum_fixup(pf_cksum_fixup(*c,
2994 	    ao.addr16[0], an->addr16[0], u),
2995 	    ao.addr16[1], an->addr16[1], u),
2996 	    ao.addr16[2], an->addr16[2], u),
2997 	    ao.addr16[3], an->addr16[3], u),
2998 	    ao.addr16[4], an->addr16[4], u),
2999 	    ao.addr16[5], an->addr16[5], u),
3000 	    ao.addr16[6], an->addr16[6], u),
3001 	    ao.addr16[7], an->addr16[7], u);
3002 }
3003 #endif /* INET6 */
3004 
3005 static void
3006 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
3007     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
3008     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
3009 {
3010 	struct pf_addr	oia, ooa;
3011 
3012 	PF_ACPY(&oia, ia, af);
3013 	if (oa)
3014 		PF_ACPY(&ooa, oa, af);
3015 
3016 	/* Change inner protocol port, fix inner protocol checksum. */
3017 	if (ip != NULL) {
3018 		u_int16_t	oip = *ip;
3019 		u_int32_t	opc;
3020 
3021 		if (pc != NULL)
3022 			opc = *pc;
3023 		*ip = np;
3024 		if (pc != NULL)
3025 			*pc = pf_cksum_fixup(*pc, oip, *ip, u);
3026 		*ic = pf_cksum_fixup(*ic, oip, *ip, 0);
3027 		if (pc != NULL)
3028 			*ic = pf_cksum_fixup(*ic, opc, *pc, 0);
3029 	}
3030 	/* Change inner ip address, fix inner ip and icmp checksums. */
3031 	PF_ACPY(ia, na, af);
3032 	switch (af) {
3033 #ifdef INET
3034 	case AF_INET: {
3035 		u_int32_t	 oh2c = *h2c;
3036 
3037 		*h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
3038 		    oia.addr16[0], ia->addr16[0], 0),
3039 		    oia.addr16[1], ia->addr16[1], 0);
3040 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
3041 		    oia.addr16[0], ia->addr16[0], 0),
3042 		    oia.addr16[1], ia->addr16[1], 0);
3043 		*ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
3044 		break;
3045 	}
3046 #endif /* INET */
3047 #ifdef INET6
3048 	case AF_INET6:
3049 		*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
3050 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
3051 		    pf_cksum_fixup(pf_cksum_fixup(*ic,
3052 		    oia.addr16[0], ia->addr16[0], u),
3053 		    oia.addr16[1], ia->addr16[1], u),
3054 		    oia.addr16[2], ia->addr16[2], u),
3055 		    oia.addr16[3], ia->addr16[3], u),
3056 		    oia.addr16[4], ia->addr16[4], u),
3057 		    oia.addr16[5], ia->addr16[5], u),
3058 		    oia.addr16[6], ia->addr16[6], u),
3059 		    oia.addr16[7], ia->addr16[7], u);
3060 		break;
3061 #endif /* INET6 */
3062 	}
3063 	/* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
3064 	if (oa) {
3065 		PF_ACPY(oa, na, af);
3066 		switch (af) {
3067 #ifdef INET
3068 		case AF_INET:
3069 			*hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
3070 			    ooa.addr16[0], oa->addr16[0], 0),
3071 			    ooa.addr16[1], oa->addr16[1], 0);
3072 			break;
3073 #endif /* INET */
3074 #ifdef INET6
3075 		case AF_INET6:
3076 			*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
3077 			    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
3078 			    pf_cksum_fixup(pf_cksum_fixup(*ic,
3079 			    ooa.addr16[0], oa->addr16[0], u),
3080 			    ooa.addr16[1], oa->addr16[1], u),
3081 			    ooa.addr16[2], oa->addr16[2], u),
3082 			    ooa.addr16[3], oa->addr16[3], u),
3083 			    ooa.addr16[4], oa->addr16[4], u),
3084 			    ooa.addr16[5], oa->addr16[5], u),
3085 			    ooa.addr16[6], oa->addr16[6], u),
3086 			    ooa.addr16[7], oa->addr16[7], u);
3087 			break;
3088 #endif /* INET6 */
3089 		}
3090 	}
3091 }
3092 
3093 /*
3094  * Need to modulate the sequence numbers in the TCP SACK option
3095  * (credits to Krzysztof Pfaff for report and patch)
3096  */
3097 static int
3098 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
3099     struct tcphdr *th, struct pf_state_peer *dst)
3100 {
3101 	int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
3102 	u_int8_t opts[TCP_MAXOLEN], *opt = opts;
3103 	int copyback = 0, i, olen;
3104 	struct sackblk sack;
3105 
3106 #define	TCPOLEN_SACKLEN	(TCPOLEN_SACK + 2)
3107 	if (hlen < TCPOLEN_SACKLEN ||
3108 	    !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
3109 		return 0;
3110 
3111 	while (hlen >= TCPOLEN_SACKLEN) {
3112 		size_t startoff = opt - opts;
3113 		olen = opt[1];
3114 		switch (*opt) {
3115 		case TCPOPT_EOL:	/* FALLTHROUGH */
3116 		case TCPOPT_NOP:
3117 			opt++;
3118 			hlen--;
3119 			break;
3120 		case TCPOPT_SACK:
3121 			if (olen > hlen)
3122 				olen = hlen;
3123 			if (olen >= TCPOLEN_SACKLEN) {
3124 				for (i = 2; i + TCPOLEN_SACK <= olen;
3125 				    i += TCPOLEN_SACK) {
3126 					memcpy(&sack, &opt[i], sizeof(sack));
3127 					pf_patch_32_unaligned(m,
3128 					    &th->th_sum, &sack.start,
3129 					    htonl(ntohl(sack.start) - dst->seqdiff),
3130 					    PF_ALGNMNT(startoff),
3131 					    0);
3132 					pf_patch_32_unaligned(m, &th->th_sum,
3133 					    &sack.end,
3134 					    htonl(ntohl(sack.end) - dst->seqdiff),
3135 					    PF_ALGNMNT(startoff),
3136 					    0);
3137 					memcpy(&opt[i], &sack, sizeof(sack));
3138 				}
3139 				copyback = 1;
3140 			}
3141 			/* FALLTHROUGH */
3142 		default:
3143 			if (olen < 2)
3144 				olen = 2;
3145 			hlen -= olen;
3146 			opt += olen;
3147 		}
3148 	}
3149 
3150 	if (copyback)
3151 		m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
3152 	return (copyback);
3153 }
3154 
3155 struct mbuf *
3156 pf_build_tcp(const struct pf_krule *r, sa_family_t af,
3157     const struct pf_addr *saddr, const struct pf_addr *daddr,
3158     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
3159     u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl,
3160     bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid)
3161 {
3162 	struct mbuf	*m;
3163 	int		 len, tlen;
3164 #ifdef INET
3165 	struct ip	*h = NULL;
3166 #endif /* INET */
3167 #ifdef INET6
3168 	struct ip6_hdr	*h6 = NULL;
3169 #endif /* INET6 */
3170 	struct tcphdr	*th;
3171 	char		*opt;
3172 	struct pf_mtag  *pf_mtag;
3173 
3174 	len = 0;
3175 	th = NULL;
3176 
3177 	/* maximum segment size tcp option */
3178 	tlen = sizeof(struct tcphdr);
3179 	if (mss)
3180 		tlen += 4;
3181 
3182 	switch (af) {
3183 #ifdef INET
3184 	case AF_INET:
3185 		len = sizeof(struct ip) + tlen;
3186 		break;
3187 #endif /* INET */
3188 #ifdef INET6
3189 	case AF_INET6:
3190 		len = sizeof(struct ip6_hdr) + tlen;
3191 		break;
3192 #endif /* INET6 */
3193 	default:
3194 		panic("%s: unsupported af %d", __func__, af);
3195 	}
3196 
3197 	m = m_gethdr(M_NOWAIT, MT_DATA);
3198 	if (m == NULL)
3199 		return (NULL);
3200 
3201 #ifdef MAC
3202 	mac_netinet_firewall_send(m);
3203 #endif
3204 	if ((pf_mtag = pf_get_mtag(m)) == NULL) {
3205 		m_freem(m);
3206 		return (NULL);
3207 	}
3208 	if (skip_firewall)
3209 		m->m_flags |= M_SKIP_FIREWALL;
3210 	pf_mtag->tag = mtag_tag;
3211 	pf_mtag->flags = mtag_flags;
3212 
3213 	if (rtableid >= 0)
3214 		M_SETFIB(m, rtableid);
3215 
3216 #ifdef ALTQ
3217 	if (r != NULL && r->qid) {
3218 		pf_mtag->qid = r->qid;
3219 
3220 		/* add hints for ecn */
3221 		pf_mtag->hdr = mtod(m, struct ip *);
3222 	}
3223 #endif /* ALTQ */
3224 	m->m_data += max_linkhdr;
3225 	m->m_pkthdr.len = m->m_len = len;
3226 	/* The rest of the stack assumes a rcvif, so provide one.
3227 	 * This is a locally generated packet, so .. close enough. */
3228 	m->m_pkthdr.rcvif = V_loif;
3229 	bzero(m->m_data, len);
3230 	switch (af) {
3231 #ifdef INET
3232 	case AF_INET:
3233 		h = mtod(m, struct ip *);
3234 
3235 		/* IP header fields included in the TCP checksum */
3236 		h->ip_p = IPPROTO_TCP;
3237 		h->ip_len = htons(tlen);
3238 		h->ip_src.s_addr = saddr->v4.s_addr;
3239 		h->ip_dst.s_addr = daddr->v4.s_addr;
3240 
3241 		th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
3242 		break;
3243 #endif /* INET */
3244 #ifdef INET6
3245 	case AF_INET6:
3246 		h6 = mtod(m, struct ip6_hdr *);
3247 
3248 		/* IP header fields included in the TCP checksum */
3249 		h6->ip6_nxt = IPPROTO_TCP;
3250 		h6->ip6_plen = htons(tlen);
3251 		memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
3252 		memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
3253 
3254 		th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
3255 		break;
3256 #endif /* INET6 */
3257 	}
3258 
3259 	/* TCP header */
3260 	th->th_sport = sport;
3261 	th->th_dport = dport;
3262 	th->th_seq = htonl(seq);
3263 	th->th_ack = htonl(ack);
3264 	th->th_off = tlen >> 2;
3265 	th->th_flags = tcp_flags;
3266 	th->th_win = htons(win);
3267 
3268 	if (mss) {
3269 		opt = (char *)(th + 1);
3270 		opt[0] = TCPOPT_MAXSEG;
3271 		opt[1] = 4;
3272 		HTONS(mss);
3273 		bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
3274 	}
3275 
3276 	switch (af) {
3277 #ifdef INET
3278 	case AF_INET:
3279 		/* TCP checksum */
3280 		th->th_sum = in_cksum(m, len);
3281 
3282 		/* Finish the IP header */
3283 		h->ip_v = 4;
3284 		h->ip_hl = sizeof(*h) >> 2;
3285 		h->ip_tos = IPTOS_LOWDELAY;
3286 		h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
3287 		h->ip_len = htons(len);
3288 		h->ip_ttl = ttl ? ttl : V_ip_defttl;
3289 		h->ip_sum = 0;
3290 		break;
3291 #endif /* INET */
3292 #ifdef INET6
3293 	case AF_INET6:
3294 		/* TCP checksum */
3295 		th->th_sum = in6_cksum(m, IPPROTO_TCP,
3296 		    sizeof(struct ip6_hdr), tlen);
3297 
3298 		h6->ip6_vfc |= IPV6_VERSION;
3299 		h6->ip6_hlim = IPV6_DEFHLIM;
3300 		break;
3301 #endif /* INET6 */
3302 	}
3303 
3304 	return (m);
3305 }
3306 
3307 static void
3308 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd,
3309     uint8_t ttl, int rtableid)
3310 {
3311 	struct mbuf		*m;
3312 #ifdef INET
3313 	struct ip		*h = NULL;
3314 #endif /* INET */
3315 #ifdef INET6
3316 	struct ip6_hdr		*h6 = NULL;
3317 #endif /* INET6 */
3318 	struct sctphdr		*hdr;
3319 	struct sctp_chunkhdr	*chunk;
3320 	struct pf_send_entry	*pfse;
3321 	int			 off = 0;
3322 
3323 	MPASS(af == pd->af);
3324 
3325 	m = m_gethdr(M_NOWAIT, MT_DATA);
3326 	if (m == NULL)
3327 		return;
3328 
3329 	m->m_data += max_linkhdr;
3330 	m->m_flags |= M_SKIP_FIREWALL;
3331 	/* The rest of the stack assumes a rcvif, so provide one.
3332 	 * This is a locally generated packet, so .. close enough. */
3333 	m->m_pkthdr.rcvif = V_loif;
3334 
3335 	/* IPv4|6 header */
3336 	switch (af) {
3337 #ifdef INET
3338 	case AF_INET:
3339 		bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk));
3340 
3341 		h = mtod(m, struct ip *);
3342 
3343 		/* IP header fields included in the TCP checksum */
3344 
3345 		h->ip_p = IPPROTO_SCTP;
3346 		h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk));
3347 		h->ip_ttl = ttl ? ttl : V_ip_defttl;
3348 		h->ip_src = pd->dst->v4;
3349 		h->ip_dst = pd->src->v4;
3350 
3351 		off += sizeof(struct ip);
3352 		break;
3353 #endif /* INET */
3354 #ifdef INET6
3355 	case AF_INET6:
3356 		bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk));
3357 
3358 		h6 = mtod(m, struct ip6_hdr *);
3359 
3360 		/* IP header fields included in the TCP checksum */
3361 		h6->ip6_vfc |= IPV6_VERSION;
3362 		h6->ip6_nxt = IPPROTO_SCTP;
3363 		h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk));
3364 		h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim;
3365 		memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr));
3366 		memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr));
3367 
3368 		off += sizeof(struct ip6_hdr);
3369 		break;
3370 #endif /* INET6 */
3371 	}
3372 
3373 	/* SCTP header */
3374 	hdr = mtodo(m, off);
3375 
3376 	hdr->src_port = pd->hdr.sctp.dest_port;
3377 	hdr->dest_port = pd->hdr.sctp.src_port;
3378 	hdr->v_tag = pd->sctp_initiate_tag;
3379 	hdr->checksum = 0;
3380 
3381 	/* Abort chunk. */
3382 	off += sizeof(struct sctphdr);
3383 	chunk = mtodo(m, off);
3384 
3385 	chunk->chunk_type = SCTP_ABORT_ASSOCIATION;
3386 	chunk->chunk_length = htons(sizeof(*chunk));
3387 
3388 	/* SCTP checksum */
3389 	off += sizeof(*chunk);
3390 	m->m_pkthdr.len = m->m_len = off;
3391 
3392 	pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk));
3393 
3394 	if (rtableid >= 0)
3395 		M_SETFIB(m, rtableid);
3396 
3397 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3398 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3399 	if (pfse == NULL) {
3400 		m_freem(m);
3401 		return;
3402 	}
3403 
3404 	switch (af) {
3405 #ifdef INET
3406 	case AF_INET:
3407 		pfse->pfse_type = PFSE_IP;
3408 		break;
3409 #endif /* INET */
3410 #ifdef INET6
3411 	case AF_INET6:
3412 		pfse->pfse_type = PFSE_IP6;
3413 		break;
3414 #endif /* INET6 */
3415 	}
3416 
3417 	pfse->pfse_m = m;
3418 	pf_send(pfse);
3419 }
3420 
3421 void
3422 pf_send_tcp(const struct pf_krule *r, sa_family_t af,
3423     const struct pf_addr *saddr, const struct pf_addr *daddr,
3424     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
3425     u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl,
3426     bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid)
3427 {
3428 	struct pf_send_entry *pfse;
3429 	struct mbuf	*m;
3430 
3431 	m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags,
3432 	    win, mss, ttl, skip_firewall, mtag_tag, mtag_flags, rtableid);
3433 	if (m == NULL)
3434 		return;
3435 
3436 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3437 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3438 	if (pfse == NULL) {
3439 		m_freem(m);
3440 		return;
3441 	}
3442 
3443 	switch (af) {
3444 #ifdef INET
3445 	case AF_INET:
3446 		pfse->pfse_type = PFSE_IP;
3447 		break;
3448 #endif /* INET */
3449 #ifdef INET6
3450 	case AF_INET6:
3451 		pfse->pfse_type = PFSE_IP6;
3452 		break;
3453 #endif /* INET6 */
3454 	}
3455 
3456 	pfse->pfse_m = m;
3457 	pf_send(pfse);
3458 }
3459 
3460 static void
3461 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd,
3462     struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th,
3463     struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
3464     u_short *reason, int rtableid)
3465 {
3466 	struct pf_addr	* const saddr = pd->src;
3467 	struct pf_addr	* const daddr = pd->dst;
3468 	sa_family_t	 af = pd->af;
3469 
3470 	/* undo NAT changes, if they have taken place */
3471 	if (nr != NULL) {
3472 		PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3473 		PF_ACPY(daddr, &sk->addr[pd->didx], af);
3474 		if (pd->sport)
3475 			*pd->sport = sk->port[pd->sidx];
3476 		if (pd->dport)
3477 			*pd->dport = sk->port[pd->didx];
3478 		if (pd->proto_sum)
3479 			*pd->proto_sum = bproto_sum;
3480 		if (pd->ip_sum)
3481 			*pd->ip_sum = bip_sum;
3482 		m_copyback(m, off, hdrlen, pd->hdr.any);
3483 	}
3484 	if (pd->proto == IPPROTO_TCP &&
3485 	    ((r->rule_flag & PFRULE_RETURNRST) ||
3486 	    (r->rule_flag & PFRULE_RETURN)) &&
3487 	    !(th->th_flags & TH_RST)) {
3488 		u_int32_t	 ack = ntohl(th->th_seq) + pd->p_len;
3489 		int		 len = 0;
3490 #ifdef INET
3491 		struct ip	*h4;
3492 #endif
3493 #ifdef INET6
3494 		struct ip6_hdr	*h6;
3495 #endif
3496 
3497 		switch (af) {
3498 #ifdef INET
3499 		case AF_INET:
3500 			h4 = mtod(m, struct ip *);
3501 			len = ntohs(h4->ip_len) - off;
3502 			break;
3503 #endif
3504 #ifdef INET6
3505 		case AF_INET6:
3506 			h6 = mtod(m, struct ip6_hdr *);
3507 			len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3508 			break;
3509 #endif
3510 		}
3511 
3512 		if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3513 			REASON_SET(reason, PFRES_PROTCKSUM);
3514 		else {
3515 			if (th->th_flags & TH_SYN)
3516 				ack++;
3517 			if (th->th_flags & TH_FIN)
3518 				ack++;
3519 			pf_send_tcp(r, af, pd->dst,
3520 				pd->src, th->th_dport, th->th_sport,
3521 				ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3522 				r->return_ttl, true, 0, 0, rtableid);
3523 		}
3524 	} else if (pd->proto == IPPROTO_SCTP &&
3525 	    (r->rule_flag & PFRULE_RETURN)) {
3526 		pf_send_sctp_abort(af, pd, r->return_ttl, rtableid);
3527 	} else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3528 		r->return_icmp)
3529 		pf_send_icmp(m, r->return_icmp >> 8,
3530 			r->return_icmp & 255, af, r, rtableid);
3531 	else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3532 		r->return_icmp6)
3533 		pf_send_icmp(m, r->return_icmp6 >> 8,
3534 			r->return_icmp6 & 255, af, r, rtableid);
3535 }
3536 
3537 static int
3538 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
3539 {
3540 	struct m_tag *mtag;
3541 	u_int8_t mpcp;
3542 
3543 	mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
3544 	if (mtag == NULL)
3545 		return (0);
3546 
3547 	if (prio == PF_PRIO_ZERO)
3548 		prio = 0;
3549 
3550 	mpcp = *(uint8_t *)(mtag + 1);
3551 
3552 	return (mpcp == prio);
3553 }
3554 
3555 static int
3556 pf_icmp_to_bandlim(uint8_t type)
3557 {
3558 	switch (type) {
3559 		case ICMP_ECHO:
3560 		case ICMP_ECHOREPLY:
3561 			return (BANDLIM_ICMP_ECHO);
3562 		case ICMP_TSTAMP:
3563 		case ICMP_TSTAMPREPLY:
3564 			return (BANDLIM_ICMP_TSTAMP);
3565 		case ICMP_UNREACH:
3566 		default:
3567 			return (BANDLIM_ICMP_UNREACH);
3568 	}
3569 }
3570 
3571 static void
3572 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
3573     struct pf_krule *r, int rtableid)
3574 {
3575 	struct pf_send_entry *pfse;
3576 	struct mbuf *m0;
3577 	struct pf_mtag *pf_mtag;
3578 
3579 	/* ICMP packet rate limitation. */
3580 #ifdef INET6
3581 	if (af == AF_INET6) {
3582 		if (icmp6_ratelimit(NULL, type, code))
3583 			return;
3584 	}
3585 #endif
3586 #ifdef INET
3587 	if (af == AF_INET) {
3588 		if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0)
3589 			return;
3590 	}
3591 #endif
3592 
3593 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3594 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3595 	if (pfse == NULL)
3596 		return;
3597 
3598 	if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
3599 		free(pfse, M_PFTEMP);
3600 		return;
3601 	}
3602 
3603 	if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
3604 		free(pfse, M_PFTEMP);
3605 		return;
3606 	}
3607 	/* XXX: revisit */
3608 	m0->m_flags |= M_SKIP_FIREWALL;
3609 
3610 	if (rtableid >= 0)
3611 		M_SETFIB(m0, rtableid);
3612 
3613 #ifdef ALTQ
3614 	if (r->qid) {
3615 		pf_mtag->qid = r->qid;
3616 		/* add hints for ecn */
3617 		pf_mtag->hdr = mtod(m0, struct ip *);
3618 	}
3619 #endif /* ALTQ */
3620 
3621 	switch (af) {
3622 #ifdef INET
3623 	case AF_INET:
3624 		pfse->pfse_type = PFSE_ICMP;
3625 		break;
3626 #endif /* INET */
3627 #ifdef INET6
3628 	case AF_INET6:
3629 		pfse->pfse_type = PFSE_ICMP6;
3630 		break;
3631 #endif /* INET6 */
3632 	}
3633 	pfse->pfse_m = m0;
3634 	pfse->icmpopts.type = type;
3635 	pfse->icmpopts.code = code;
3636 	pf_send(pfse);
3637 }
3638 
3639 /*
3640  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
3641  * If n is 0, they match if they are equal. If n is != 0, they match if they
3642  * are different.
3643  */
3644 int
3645 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
3646     struct pf_addr *b, sa_family_t af)
3647 {
3648 	int	match = 0;
3649 
3650 	switch (af) {
3651 #ifdef INET
3652 	case AF_INET:
3653 		if (IN_ARE_MASKED_ADDR_EQUAL(a->v4, b->v4, m->v4))
3654 			match++;
3655 		break;
3656 #endif /* INET */
3657 #ifdef INET6
3658 	case AF_INET6:
3659 		if (IN6_ARE_MASKED_ADDR_EQUAL(&a->v6, &b->v6, &m->v6))
3660 			match++;
3661 		break;
3662 #endif /* INET6 */
3663 	}
3664 	if (match) {
3665 		if (n)
3666 			return (0);
3667 		else
3668 			return (1);
3669 	} else {
3670 		if (n)
3671 			return (1);
3672 		else
3673 			return (0);
3674 	}
3675 }
3676 
3677 /*
3678  * Return 1 if b <= a <= e, otherwise return 0.
3679  */
3680 int
3681 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
3682     struct pf_addr *a, sa_family_t af)
3683 {
3684 	switch (af) {
3685 #ifdef INET
3686 	case AF_INET:
3687 		if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
3688 		    (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
3689 			return (0);
3690 		break;
3691 #endif /* INET */
3692 #ifdef INET6
3693 	case AF_INET6: {
3694 		int	i;
3695 
3696 		/* check a >= b */
3697 		for (i = 0; i < 4; ++i)
3698 			if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
3699 				break;
3700 			else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
3701 				return (0);
3702 		/* check a <= e */
3703 		for (i = 0; i < 4; ++i)
3704 			if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
3705 				break;
3706 			else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
3707 				return (0);
3708 		break;
3709 	}
3710 #endif /* INET6 */
3711 	}
3712 	return (1);
3713 }
3714 
3715 static int
3716 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
3717 {
3718 	switch (op) {
3719 	case PF_OP_IRG:
3720 		return ((p > a1) && (p < a2));
3721 	case PF_OP_XRG:
3722 		return ((p < a1) || (p > a2));
3723 	case PF_OP_RRG:
3724 		return ((p >= a1) && (p <= a2));
3725 	case PF_OP_EQ:
3726 		return (p == a1);
3727 	case PF_OP_NE:
3728 		return (p != a1);
3729 	case PF_OP_LT:
3730 		return (p < a1);
3731 	case PF_OP_LE:
3732 		return (p <= a1);
3733 	case PF_OP_GT:
3734 		return (p > a1);
3735 	case PF_OP_GE:
3736 		return (p >= a1);
3737 	}
3738 	return (0); /* never reached */
3739 }
3740 
3741 int
3742 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
3743 {
3744 	NTOHS(a1);
3745 	NTOHS(a2);
3746 	NTOHS(p);
3747 	return (pf_match(op, a1, a2, p));
3748 }
3749 
3750 static int
3751 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
3752 {
3753 	if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
3754 		return (0);
3755 	return (pf_match(op, a1, a2, u));
3756 }
3757 
3758 static int
3759 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
3760 {
3761 	if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
3762 		return (0);
3763 	return (pf_match(op, a1, a2, g));
3764 }
3765 
3766 int
3767 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag)
3768 {
3769 	if (*tag == -1)
3770 		*tag = mtag;
3771 
3772 	return ((!r->match_tag_not && r->match_tag == *tag) ||
3773 	    (r->match_tag_not && r->match_tag != *tag));
3774 }
3775 
3776 int
3777 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
3778 {
3779 
3780 	KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
3781 
3782 	if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
3783 		return (ENOMEM);
3784 
3785 	pd->pf_mtag->tag = tag;
3786 
3787 	return (0);
3788 }
3789 
3790 #define	PF_ANCHOR_STACKSIZE	32
3791 struct pf_kanchor_stackframe {
3792 	struct pf_kruleset	*rs;
3793 	struct pf_krule		*r;	/* XXX: + match bit */
3794 	struct pf_kanchor	*child;
3795 };
3796 
3797 /*
3798  * XXX: We rely on malloc(9) returning pointer aligned addresses.
3799  */
3800 #define	PF_ANCHORSTACK_MATCH	0x00000001
3801 #define	PF_ANCHORSTACK_MASK	(PF_ANCHORSTACK_MATCH)
3802 
3803 #define	PF_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
3804 #define	PF_ANCHOR_RULE(f)	(struct pf_krule *)			\
3805 				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
3806 #define	PF_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 			\
3807 				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
3808 } while (0)
3809 
3810 void
3811 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth,
3812     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
3813     int *match)
3814 {
3815 	struct pf_kanchor_stackframe	*f;
3816 
3817 	PF_RULES_RASSERT();
3818 
3819 	if (match)
3820 		*match = 0;
3821 	if (*depth >= PF_ANCHOR_STACKSIZE) {
3822 		printf("%s: anchor stack overflow on %s\n",
3823 		    __func__, (*r)->anchor->name);
3824 		*r = TAILQ_NEXT(*r, entries);
3825 		return;
3826 	} else if (*depth == 0 && a != NULL)
3827 		*a = *r;
3828 	f = stack + (*depth)++;
3829 	f->rs = *rs;
3830 	f->r = *r;
3831 	if ((*r)->anchor_wildcard) {
3832 		struct pf_kanchor_node *parent = &(*r)->anchor->children;
3833 
3834 		if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) {
3835 			*r = NULL;
3836 			return;
3837 		}
3838 		*rs = &f->child->ruleset;
3839 	} else {
3840 		f->child = NULL;
3841 		*rs = &(*r)->anchor->ruleset;
3842 	}
3843 	*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
3844 }
3845 
3846 int
3847 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth,
3848     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
3849     int *match)
3850 {
3851 	struct pf_kanchor_stackframe	*f;
3852 	struct pf_krule *fr;
3853 	int quick = 0;
3854 
3855 	PF_RULES_RASSERT();
3856 
3857 	do {
3858 		if (*depth <= 0)
3859 			break;
3860 		f = stack + *depth - 1;
3861 		fr = PF_ANCHOR_RULE(f);
3862 		if (f->child != NULL) {
3863 			/*
3864 			 * This block traverses through
3865 			 * a wildcard anchor.
3866 			 */
3867 			if (match != NULL && *match) {
3868 				/*
3869 				 * If any of "*" matched, then
3870 				 * "foo/ *" matched, mark frame
3871 				 * appropriately.
3872 				 */
3873 				PF_ANCHOR_SET_MATCH(f);
3874 				*match = 0;
3875 			}
3876 			f->child = RB_NEXT(pf_kanchor_node,
3877 			    &fr->anchor->children, f->child);
3878 			if (f->child != NULL) {
3879 				*rs = &f->child->ruleset;
3880 				*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
3881 				if (*r == NULL)
3882 					continue;
3883 				else
3884 					break;
3885 			}
3886 		}
3887 		(*depth)--;
3888 		if (*depth == 0 && a != NULL)
3889 			*a = NULL;
3890 		*rs = f->rs;
3891 		if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
3892 			quick = fr->quick;
3893 		*r = TAILQ_NEXT(fr, entries);
3894 	} while (*r == NULL);
3895 
3896 	return (quick);
3897 }
3898 
3899 struct pf_keth_anchor_stackframe {
3900 	struct pf_keth_ruleset	*rs;
3901 	struct pf_keth_rule	*r;	/* XXX: + match bit */
3902 	struct pf_keth_anchor	*child;
3903 };
3904 
3905 #define	PF_ETH_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
3906 #define	PF_ETH_ANCHOR_RULE(f)	(struct pf_keth_rule *)			\
3907 				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
3908 #define	PF_ETH_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 		\
3909 				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
3910 } while (0)
3911 
3912 void
3913 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth,
3914     struct pf_keth_ruleset **rs, struct pf_keth_rule **r,
3915     struct pf_keth_rule **a, int *match)
3916 {
3917 	struct pf_keth_anchor_stackframe	*f;
3918 
3919 	NET_EPOCH_ASSERT();
3920 
3921 	if (match)
3922 		*match = 0;
3923 	if (*depth >= PF_ANCHOR_STACKSIZE) {
3924 		printf("%s: anchor stack overflow on %s\n",
3925 		    __func__, (*r)->anchor->name);
3926 		*r = TAILQ_NEXT(*r, entries);
3927 		return;
3928 	} else if (*depth == 0 && a != NULL)
3929 		*a = *r;
3930 	f = stack + (*depth)++;
3931 	f->rs = *rs;
3932 	f->r = *r;
3933 	if ((*r)->anchor_wildcard) {
3934 		struct pf_keth_anchor_node *parent = &(*r)->anchor->children;
3935 
3936 		if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) {
3937 			*r = NULL;
3938 			return;
3939 		}
3940 		*rs = &f->child->ruleset;
3941 	} else {
3942 		f->child = NULL;
3943 		*rs = &(*r)->anchor->ruleset;
3944 	}
3945 	*r = TAILQ_FIRST((*rs)->active.rules);
3946 }
3947 
3948 int
3949 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth,
3950     struct pf_keth_ruleset **rs, struct pf_keth_rule **r,
3951     struct pf_keth_rule **a, int *match)
3952 {
3953 	struct pf_keth_anchor_stackframe	*f;
3954 	struct pf_keth_rule *fr;
3955 	int quick = 0;
3956 
3957 	NET_EPOCH_ASSERT();
3958 
3959 	do {
3960 		if (*depth <= 0)
3961 			break;
3962 		f = stack + *depth - 1;
3963 		fr = PF_ETH_ANCHOR_RULE(f);
3964 		if (f->child != NULL) {
3965 			/*
3966 			 * This block traverses through
3967 			 * a wildcard anchor.
3968 			 */
3969 			if (match != NULL && *match) {
3970 				/*
3971 				 * If any of "*" matched, then
3972 				 * "foo/ *" matched, mark frame
3973 				 * appropriately.
3974 				 */
3975 				PF_ETH_ANCHOR_SET_MATCH(f);
3976 				*match = 0;
3977 			}
3978 			f->child = RB_NEXT(pf_keth_anchor_node,
3979 			    &fr->anchor->children, f->child);
3980 			if (f->child != NULL) {
3981 				*rs = &f->child->ruleset;
3982 				*r = TAILQ_FIRST((*rs)->active.rules);
3983 				if (*r == NULL)
3984 					continue;
3985 				else
3986 					break;
3987 			}
3988 		}
3989 		(*depth)--;
3990 		if (*depth == 0 && a != NULL)
3991 			*a = NULL;
3992 		*rs = f->rs;
3993 		if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match))
3994 			quick = fr->quick;
3995 		*r = TAILQ_NEXT(fr, entries);
3996 	} while (*r == NULL);
3997 
3998 	return (quick);
3999 }
4000 
4001 #ifdef INET6
4002 void
4003 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
4004     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
4005 {
4006 	switch (af) {
4007 #ifdef INET
4008 	case AF_INET:
4009 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
4010 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
4011 		break;
4012 #endif /* INET */
4013 	case AF_INET6:
4014 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
4015 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
4016 		naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
4017 		((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
4018 		naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
4019 		((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
4020 		naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
4021 		((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
4022 		break;
4023 	}
4024 }
4025 
4026 void
4027 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
4028 {
4029 	switch (af) {
4030 #ifdef INET
4031 	case AF_INET:
4032 		addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
4033 		break;
4034 #endif /* INET */
4035 	case AF_INET6:
4036 		if (addr->addr32[3] == 0xffffffff) {
4037 			addr->addr32[3] = 0;
4038 			if (addr->addr32[2] == 0xffffffff) {
4039 				addr->addr32[2] = 0;
4040 				if (addr->addr32[1] == 0xffffffff) {
4041 					addr->addr32[1] = 0;
4042 					addr->addr32[0] =
4043 					    htonl(ntohl(addr->addr32[0]) + 1);
4044 				} else
4045 					addr->addr32[1] =
4046 					    htonl(ntohl(addr->addr32[1]) + 1);
4047 			} else
4048 				addr->addr32[2] =
4049 				    htonl(ntohl(addr->addr32[2]) + 1);
4050 		} else
4051 			addr->addr32[3] =
4052 			    htonl(ntohl(addr->addr32[3]) + 1);
4053 		break;
4054 	}
4055 }
4056 #endif /* INET6 */
4057 
4058 void
4059 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a)
4060 {
4061 	/*
4062 	 * Modern rules use the same flags in rules as they do in states.
4063 	 */
4064 	a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID|
4065 	    PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO));
4066 
4067 	/*
4068 	 * Old-style scrub rules have different flags which need to be translated.
4069 	 */
4070 	if (r->rule_flag & PFRULE_RANDOMID)
4071 		a->flags |= PFSTATE_RANDOMID;
4072 	if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) {
4073 		a->flags |= PFSTATE_SETTOS;
4074 		a->set_tos = r->set_tos;
4075 	}
4076 
4077 	if (r->qid)
4078 		a->qid = r->qid;
4079 	if (r->pqid)
4080 		a->pqid = r->pqid;
4081 	if (r->rtableid >= 0)
4082 		a->rtableid = r->rtableid;
4083 	a->log |= r->log;
4084 	if (r->min_ttl)
4085 		a->min_ttl = r->min_ttl;
4086 	if (r->max_mss)
4087 		a->max_mss = r->max_mss;
4088 	if (r->dnpipe)
4089 		a->dnpipe = r->dnpipe;
4090 	if (r->dnrpipe)
4091 		a->dnrpipe = r->dnrpipe;
4092 	if (r->dnpipe || r->dnrpipe) {
4093 		if (r->free_flags & PFRULE_DN_IS_PIPE)
4094 			a->flags |= PFSTATE_DN_IS_PIPE;
4095 		else
4096 			a->flags &= ~PFSTATE_DN_IS_PIPE;
4097 	}
4098 	if (r->scrub_flags & PFSTATE_SETPRIO) {
4099 		a->set_prio[0] = r->set_prio[0];
4100 		a->set_prio[1] = r->set_prio[1];
4101 	}
4102 }
4103 
4104 int
4105 pf_socket_lookup(struct pf_pdesc *pd, struct mbuf *m)
4106 {
4107 	struct pf_addr		*saddr, *daddr;
4108 	u_int16_t		 sport, dport;
4109 	struct inpcbinfo	*pi;
4110 	struct inpcb		*inp;
4111 
4112 	pd->lookup.uid = UID_MAX;
4113 	pd->lookup.gid = GID_MAX;
4114 
4115 	switch (pd->proto) {
4116 	case IPPROTO_TCP:
4117 		sport = pd->hdr.tcp.th_sport;
4118 		dport = pd->hdr.tcp.th_dport;
4119 		pi = &V_tcbinfo;
4120 		break;
4121 	case IPPROTO_UDP:
4122 		sport = pd->hdr.udp.uh_sport;
4123 		dport = pd->hdr.udp.uh_dport;
4124 		pi = &V_udbinfo;
4125 		break;
4126 	default:
4127 		return (-1);
4128 	}
4129 	if (pd->dir == PF_IN) {
4130 		saddr = pd->src;
4131 		daddr = pd->dst;
4132 	} else {
4133 		u_int16_t	p;
4134 
4135 		p = sport;
4136 		sport = dport;
4137 		dport = p;
4138 		saddr = pd->dst;
4139 		daddr = pd->src;
4140 	}
4141 	switch (pd->af) {
4142 #ifdef INET
4143 	case AF_INET:
4144 		inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
4145 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
4146 		if (inp == NULL) {
4147 			inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
4148 			   daddr->v4, dport, INPLOOKUP_WILDCARD |
4149 			   INPLOOKUP_RLOCKPCB, NULL, m);
4150 			if (inp == NULL)
4151 				return (-1);
4152 		}
4153 		break;
4154 #endif /* INET */
4155 #ifdef INET6
4156 	case AF_INET6:
4157 		inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
4158 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
4159 		if (inp == NULL) {
4160 			inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
4161 			    &daddr->v6, dport, INPLOOKUP_WILDCARD |
4162 			    INPLOOKUP_RLOCKPCB, NULL, m);
4163 			if (inp == NULL)
4164 				return (-1);
4165 		}
4166 		break;
4167 #endif /* INET6 */
4168 
4169 	default:
4170 		return (-1);
4171 	}
4172 	INP_RLOCK_ASSERT(inp);
4173 	pd->lookup.uid = inp->inp_cred->cr_uid;
4174 	pd->lookup.gid = inp->inp_cred->cr_groups[0];
4175 	INP_RUNLOCK(inp);
4176 
4177 	return (1);
4178 }
4179 
4180 u_int8_t
4181 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
4182 {
4183 	int		 hlen;
4184 	u_int8_t	 hdr[60];
4185 	u_int8_t	*opt, optlen;
4186 	u_int8_t	 wscale = 0;
4187 
4188 	hlen = th_off << 2;		/* hlen <= sizeof(hdr) */
4189 	if (hlen <= sizeof(struct tcphdr))
4190 		return (0);
4191 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
4192 		return (0);
4193 	opt = hdr + sizeof(struct tcphdr);
4194 	hlen -= sizeof(struct tcphdr);
4195 	while (hlen >= 3) {
4196 		switch (*opt) {
4197 		case TCPOPT_EOL:
4198 		case TCPOPT_NOP:
4199 			++opt;
4200 			--hlen;
4201 			break;
4202 		case TCPOPT_WINDOW:
4203 			wscale = opt[2];
4204 			if (wscale > TCP_MAX_WINSHIFT)
4205 				wscale = TCP_MAX_WINSHIFT;
4206 			wscale |= PF_WSCALE_FLAG;
4207 			/* FALLTHROUGH */
4208 		default:
4209 			optlen = opt[1];
4210 			if (optlen < 2)
4211 				optlen = 2;
4212 			hlen -= optlen;
4213 			opt += optlen;
4214 			break;
4215 		}
4216 	}
4217 	return (wscale);
4218 }
4219 
4220 u_int16_t
4221 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
4222 {
4223 	int		 hlen;
4224 	u_int8_t	 hdr[60];
4225 	u_int8_t	*opt, optlen;
4226 	u_int16_t	 mss = V_tcp_mssdflt;
4227 
4228 	hlen = th_off << 2;	/* hlen <= sizeof(hdr) */
4229 	if (hlen <= sizeof(struct tcphdr))
4230 		return (0);
4231 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
4232 		return (0);
4233 	opt = hdr + sizeof(struct tcphdr);
4234 	hlen -= sizeof(struct tcphdr);
4235 	while (hlen >= TCPOLEN_MAXSEG) {
4236 		switch (*opt) {
4237 		case TCPOPT_EOL:
4238 		case TCPOPT_NOP:
4239 			++opt;
4240 			--hlen;
4241 			break;
4242 		case TCPOPT_MAXSEG:
4243 			bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
4244 			NTOHS(mss);
4245 			/* FALLTHROUGH */
4246 		default:
4247 			optlen = opt[1];
4248 			if (optlen < 2)
4249 				optlen = 2;
4250 			hlen -= optlen;
4251 			opt += optlen;
4252 			break;
4253 		}
4254 	}
4255 	return (mss);
4256 }
4257 
4258 static u_int16_t
4259 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
4260 {
4261 	struct nhop_object *nh;
4262 #ifdef INET6
4263 	struct in6_addr		dst6;
4264 	uint32_t		scopeid;
4265 #endif /* INET6 */
4266 	int			 hlen = 0;
4267 	uint16_t		 mss = 0;
4268 
4269 	NET_EPOCH_ASSERT();
4270 
4271 	switch (af) {
4272 #ifdef INET
4273 	case AF_INET:
4274 		hlen = sizeof(struct ip);
4275 		nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0);
4276 		if (nh != NULL)
4277 			mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
4278 		break;
4279 #endif /* INET */
4280 #ifdef INET6
4281 	case AF_INET6:
4282 		hlen = sizeof(struct ip6_hdr);
4283 		in6_splitscope(&addr->v6, &dst6, &scopeid);
4284 		nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0);
4285 		if (nh != NULL)
4286 			mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
4287 		break;
4288 #endif /* INET6 */
4289 	}
4290 
4291 	mss = max(V_tcp_mssdflt, mss);
4292 	mss = min(mss, offer);
4293 	mss = max(mss, 64);		/* sanity - at least max opt space */
4294 	return (mss);
4295 }
4296 
4297 static u_int32_t
4298 pf_tcp_iss(struct pf_pdesc *pd)
4299 {
4300 	MD5_CTX ctx;
4301 	u_int32_t digest[4];
4302 
4303 	if (V_pf_tcp_secret_init == 0) {
4304 		arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
4305 		MD5Init(&V_pf_tcp_secret_ctx);
4306 		MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
4307 		    sizeof(V_pf_tcp_secret));
4308 		V_pf_tcp_secret_init = 1;
4309 	}
4310 
4311 	ctx = V_pf_tcp_secret_ctx;
4312 
4313 	MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short));
4314 	MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short));
4315 	if (pd->af == AF_INET6) {
4316 		MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
4317 		MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
4318 	} else {
4319 		MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
4320 		MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
4321 	}
4322 	MD5Final((u_char *)digest, &ctx);
4323 	V_pf_tcp_iss_off += 4096;
4324 #define	ISN_RANDOM_INCREMENT (4096 - 1)
4325 	return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
4326 	    V_pf_tcp_iss_off);
4327 #undef	ISN_RANDOM_INCREMENT
4328 }
4329 
4330 static bool
4331 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r)
4332 {
4333 	bool match = true;
4334 
4335 	/* Always matches if not set */
4336 	if (! r->isset)
4337 		return (!r->neg);
4338 
4339 	for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4340 		if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) {
4341 			match = false;
4342 			break;
4343 		}
4344 	}
4345 
4346 	return (match ^ r->neg);
4347 }
4348 
4349 static int
4350 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag)
4351 {
4352 	if (*tag == -1)
4353 		*tag = mtag;
4354 
4355 	return ((!r->match_tag_not && r->match_tag == *tag) ||
4356 	    (r->match_tag_not && r->match_tag != *tag));
4357 }
4358 
4359 static void
4360 pf_bridge_to(struct ifnet *ifp, struct mbuf *m)
4361 {
4362 	/* If we don't have the interface drop the packet. */
4363 	if (ifp == NULL) {
4364 		m_freem(m);
4365 		return;
4366 	}
4367 
4368 	switch (ifp->if_type) {
4369 	case IFT_ETHER:
4370 	case IFT_XETHER:
4371 	case IFT_L2VLAN:
4372 	case IFT_BRIDGE:
4373 	case IFT_IEEE8023ADLAG:
4374 		break;
4375 	default:
4376 		m_freem(m);
4377 		return;
4378 	}
4379 
4380 	ifp->if_transmit(ifp, m);
4381 }
4382 
4383 static int
4384 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0)
4385 {
4386 #ifdef INET
4387 	struct ip ip;
4388 #endif
4389 #ifdef INET6
4390 	struct ip6_hdr ip6;
4391 #endif
4392 	struct mbuf *m = *m0;
4393 	struct ether_header *e;
4394 	struct pf_keth_rule *r, *rm, *a = NULL;
4395 	struct pf_keth_ruleset *ruleset = NULL;
4396 	struct pf_mtag *mtag;
4397 	struct pf_keth_ruleq *rules;
4398 	struct pf_addr *src = NULL, *dst = NULL;
4399 	struct pfi_kkif *bridge_to;
4400 	sa_family_t af = 0;
4401 	uint16_t proto;
4402 	int asd = 0, match = 0;
4403 	int tag = -1;
4404 	uint8_t action;
4405 	struct pf_keth_anchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
4406 
4407 	MPASS(kif->pfik_ifp->if_vnet == curvnet);
4408 	NET_EPOCH_ASSERT();
4409 
4410 	PF_RULES_RLOCK_TRACKER;
4411 
4412 	SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m);
4413 
4414 	mtag = pf_find_mtag(m);
4415 	if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
4416 		/* Dummynet re-injects packets after they've
4417 		 * completed their delay. We've already
4418 		 * processed them, so pass unconditionally. */
4419 
4420 		/* But only once. We may see the packet multiple times (e.g.
4421 		 * PFIL_IN/PFIL_OUT). */
4422 		pf_dummynet_flag_remove(m, mtag);
4423 
4424 		return (PF_PASS);
4425 	}
4426 
4427 	ruleset = V_pf_keth;
4428 	rules = ck_pr_load_ptr(&ruleset->active.rules);
4429 	r = TAILQ_FIRST(rules);
4430 	rm = NULL;
4431 
4432 	if (__predict_false(m->m_len < sizeof(struct ether_header)) &&
4433 	    (m = *m0 = m_pullup(*m0, sizeof(struct ether_header))) == NULL) {
4434 		DPFPRINTF(PF_DEBUG_URGENT,
4435 		    ("pf_test_eth_rule: m_len < sizeof(struct ether_header)"
4436 		     ", pullup failed\n"));
4437 		return (PF_DROP);
4438 	}
4439 	e = mtod(m, struct ether_header *);
4440 	proto = ntohs(e->ether_type);
4441 
4442 	switch (proto) {
4443 #ifdef INET
4444 	case ETHERTYPE_IP: {
4445 		if (m_length(m, NULL) < (sizeof(struct ether_header) +
4446 		    sizeof(ip)))
4447 			return (PF_DROP);
4448 
4449 		af = AF_INET;
4450 		m_copydata(m, sizeof(struct ether_header), sizeof(ip),
4451 		    (caddr_t)&ip);
4452 		src = (struct pf_addr *)&ip.ip_src;
4453 		dst = (struct pf_addr *)&ip.ip_dst;
4454 		break;
4455 	}
4456 #endif /* INET */
4457 #ifdef INET6
4458 	case ETHERTYPE_IPV6: {
4459 		if (m_length(m, NULL) < (sizeof(struct ether_header) +
4460 		    sizeof(ip6)))
4461 			return (PF_DROP);
4462 
4463 		af = AF_INET6;
4464 		m_copydata(m, sizeof(struct ether_header), sizeof(ip6),
4465 		    (caddr_t)&ip6);
4466 		src = (struct pf_addr *)&ip6.ip6_src;
4467 		dst = (struct pf_addr *)&ip6.ip6_dst;
4468 		break;
4469 	}
4470 #endif /* INET6 */
4471 	}
4472 
4473 	PF_RULES_RLOCK();
4474 
4475 	while (r != NULL) {
4476 		counter_u64_add(r->evaluations, 1);
4477 		SDT_PROBE2(pf, eth, test_rule, test, r->nr, r);
4478 
4479 		if (pfi_kkif_match(r->kif, kif) == r->ifnot) {
4480 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4481 			    "kif");
4482 			r = r->skip[PFE_SKIP_IFP].ptr;
4483 		}
4484 		else if (r->direction && r->direction != dir) {
4485 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4486 			    "dir");
4487 			r = r->skip[PFE_SKIP_DIR].ptr;
4488 		}
4489 		else if (r->proto && r->proto != proto) {
4490 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4491 			    "proto");
4492 			r = r->skip[PFE_SKIP_PROTO].ptr;
4493 		}
4494 		else if (! pf_match_eth_addr(e->ether_shost, &r->src)) {
4495 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4496 			    "src");
4497 			r = r->skip[PFE_SKIP_SRC_ADDR].ptr;
4498 		}
4499 		else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) {
4500 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4501 			    "dst");
4502 			r = r->skip[PFE_SKIP_DST_ADDR].ptr;
4503 		}
4504 		else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af,
4505 		    r->ipsrc.neg, kif, M_GETFIB(m))) {
4506 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4507 			    "ip_src");
4508 			r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr;
4509 		}
4510 		else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af,
4511 		    r->ipdst.neg, kif, M_GETFIB(m))) {
4512 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4513 			    "ip_dst");
4514 			r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr;
4515 		}
4516 		else if (r->match_tag && !pf_match_eth_tag(m, r, &tag,
4517 		    mtag ? mtag->tag : 0)) {
4518 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4519 			    "match_tag");
4520 			r = TAILQ_NEXT(r, entries);
4521 		}
4522 		else {
4523 			if (r->tag)
4524 				tag = r->tag;
4525 			if (r->anchor == NULL) {
4526 				/* Rule matches */
4527 				rm = r;
4528 
4529 				SDT_PROBE2(pf, eth, test_rule, match, r->nr, r);
4530 
4531 				if (r->quick)
4532 					break;
4533 
4534 				r = TAILQ_NEXT(r, entries);
4535 			} else {
4536 				pf_step_into_keth_anchor(anchor_stack, &asd,
4537 				    &ruleset, &r, &a, &match);
4538 			}
4539 		}
4540 		if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd,
4541 		    &ruleset, &r, &a, &match))
4542 			break;
4543 	}
4544 
4545 	r = rm;
4546 
4547 	SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r);
4548 
4549 	/* Default to pass. */
4550 	if (r == NULL) {
4551 		PF_RULES_RUNLOCK();
4552 		return (PF_PASS);
4553 	}
4554 
4555 	/* Execute action. */
4556 	counter_u64_add(r->packets[dir == PF_OUT], 1);
4557 	counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL));
4558 	pf_update_timestamp(r);
4559 
4560 	/* Shortcut. Don't tag if we're just going to drop anyway. */
4561 	if (r->action == PF_DROP) {
4562 		PF_RULES_RUNLOCK();
4563 		return (PF_DROP);
4564 	}
4565 
4566 	if (tag > 0) {
4567 		if (mtag == NULL)
4568 			mtag = pf_get_mtag(m);
4569 		if (mtag == NULL) {
4570 			PF_RULES_RUNLOCK();
4571 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4572 			return (PF_DROP);
4573 		}
4574 		mtag->tag = tag;
4575 	}
4576 
4577 	if (r->qid != 0) {
4578 		if (mtag == NULL)
4579 			mtag = pf_get_mtag(m);
4580 		if (mtag == NULL) {
4581 			PF_RULES_RUNLOCK();
4582 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4583 			return (PF_DROP);
4584 		}
4585 		mtag->qid = r->qid;
4586 	}
4587 
4588 	action = r->action;
4589 	bridge_to = r->bridge_to;
4590 
4591 	/* Dummynet */
4592 	if (r->dnpipe) {
4593 		struct ip_fw_args dnflow;
4594 
4595 		/* Drop packet if dummynet is not loaded. */
4596 		if (ip_dn_io_ptr == NULL) {
4597 			PF_RULES_RUNLOCK();
4598 			m_freem(m);
4599 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4600 			return (PF_DROP);
4601 		}
4602 		if (mtag == NULL)
4603 			mtag = pf_get_mtag(m);
4604 		if (mtag == NULL) {
4605 			PF_RULES_RUNLOCK();
4606 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4607 			return (PF_DROP);
4608 		}
4609 
4610 		bzero(&dnflow, sizeof(dnflow));
4611 
4612 		/* We don't have port numbers here, so we set 0.  That means
4613 		 * that we'll be somewhat limited in distinguishing flows (i.e.
4614 		 * only based on IP addresses, not based on port numbers), but
4615 		 * it's better than nothing. */
4616 		dnflow.f_id.dst_port = 0;
4617 		dnflow.f_id.src_port = 0;
4618 		dnflow.f_id.proto = 0;
4619 
4620 		dnflow.rule.info = r->dnpipe;
4621 		dnflow.rule.info |= IPFW_IS_DUMMYNET;
4622 		if (r->dnflags & PFRULE_DN_IS_PIPE)
4623 			dnflow.rule.info |= IPFW_IS_PIPE;
4624 
4625 		dnflow.f_id.extra = dnflow.rule.info;
4626 
4627 		dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT;
4628 		dnflow.flags |= IPFW_ARGS_ETHER;
4629 		dnflow.ifp = kif->pfik_ifp;
4630 
4631 		switch (af) {
4632 		case AF_INET:
4633 			dnflow.f_id.addr_type = 4;
4634 			dnflow.f_id.src_ip = src->v4.s_addr;
4635 			dnflow.f_id.dst_ip = dst->v4.s_addr;
4636 			break;
4637 		case AF_INET6:
4638 			dnflow.flags |= IPFW_ARGS_IP6;
4639 			dnflow.f_id.addr_type = 6;
4640 			dnflow.f_id.src_ip6 = src->v6;
4641 			dnflow.f_id.dst_ip6 = dst->v6;
4642 			break;
4643 		}
4644 
4645 		PF_RULES_RUNLOCK();
4646 
4647 		mtag->flags |= PF_MTAG_FLAG_DUMMYNET;
4648 		ip_dn_io_ptr(m0, &dnflow);
4649 		if (*m0 != NULL)
4650 			pf_dummynet_flag_remove(m, mtag);
4651 	} else {
4652 		PF_RULES_RUNLOCK();
4653 	}
4654 
4655 	if (action == PF_PASS && bridge_to) {
4656 		pf_bridge_to(bridge_to->pfik_ifp, *m0);
4657 		*m0 = NULL; /* We've eaten the packet. */
4658 	}
4659 
4660 	return (action);
4661 }
4662 
4663 static int
4664 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, struct pfi_kkif *kif,
4665     struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_krule **am,
4666     struct pf_kruleset **rsm, struct inpcb *inp)
4667 {
4668 	struct pf_krule		*nr = NULL;
4669 	struct pf_addr		* const saddr = pd->src;
4670 	struct pf_addr		* const daddr = pd->dst;
4671 	sa_family_t		 af = pd->af;
4672 	struct pf_krule		*r, *a = NULL;
4673 	struct pf_kruleset	*ruleset = NULL;
4674 	struct pf_krule_slist	 match_rules;
4675 	struct pf_krule_item	*ri;
4676 	struct pf_ksrc_node	*nsn = NULL;
4677 	struct tcphdr		*th = &pd->hdr.tcp;
4678 	struct pf_state_key	*sk = NULL, *nk = NULL;
4679 	u_short			 reason, transerror;
4680 	int			 rewrite = 0, hdrlen = 0;
4681 	int			 tag = -1;
4682 	int			 asd = 0;
4683 	int			 match = 0;
4684 	int			 state_icmp = 0, icmp_dir, multi;
4685 	u_int16_t		 sport = 0, dport = 0, virtual_type, virtual_id;
4686 	u_int16_t		 bproto_sum = 0, bip_sum = 0;
4687 	u_int8_t		 icmptype = 0, icmpcode = 0;
4688 	struct pf_kanchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
4689 
4690 	PF_RULES_RASSERT();
4691 
4692 	SLIST_INIT(&match_rules);
4693 
4694 	if (inp != NULL) {
4695 		INP_LOCK_ASSERT(inp);
4696 		pd->lookup.uid = inp->inp_cred->cr_uid;
4697 		pd->lookup.gid = inp->inp_cred->cr_groups[0];
4698 		pd->lookup.done = 1;
4699 	}
4700 
4701 	switch (pd->proto) {
4702 	case IPPROTO_TCP:
4703 		sport = th->th_sport;
4704 		dport = th->th_dport;
4705 		hdrlen = sizeof(*th);
4706 		break;
4707 	case IPPROTO_UDP:
4708 		sport = pd->hdr.udp.uh_sport;
4709 		dport = pd->hdr.udp.uh_dport;
4710 		hdrlen = sizeof(pd->hdr.udp);
4711 		break;
4712 	case IPPROTO_SCTP:
4713 		sport = pd->hdr.sctp.src_port;
4714 		dport = pd->hdr.sctp.dest_port;
4715 		hdrlen = sizeof(pd->hdr.sctp);
4716 		break;
4717 #ifdef INET
4718 	case IPPROTO_ICMP:
4719 		if (pd->af != AF_INET)
4720 			break;
4721 		hdrlen = sizeof(pd->hdr.icmp);
4722 		icmptype = pd->hdr.icmp.icmp_type;
4723 		icmpcode = pd->hdr.icmp.icmp_code;
4724 		state_icmp = pf_icmp_mapping(pd, icmptype,
4725 		    &icmp_dir, &multi, &virtual_id, &virtual_type);
4726 		if (icmp_dir == PF_IN) {
4727 			sport = virtual_id;
4728 			dport = virtual_type;
4729 		} else {
4730 			sport = virtual_type;
4731 			dport = virtual_id;
4732 		}
4733 		break;
4734 #endif /* INET */
4735 #ifdef INET6
4736 	case IPPROTO_ICMPV6:
4737 		if (af != AF_INET6)
4738 			break;
4739 		hdrlen = sizeof(pd->hdr.icmp6);
4740 		icmptype = pd->hdr.icmp6.icmp6_type;
4741 		icmpcode = pd->hdr.icmp6.icmp6_code;
4742 		state_icmp = pf_icmp_mapping(pd, icmptype,
4743 		    &icmp_dir, &multi, &virtual_id, &virtual_type);
4744 		if (icmp_dir == PF_IN) {
4745 			sport = virtual_id;
4746 			dport = virtual_type;
4747 		} else {
4748 			sport = virtual_type;
4749 			dport = virtual_id;
4750 		}
4751 
4752 		break;
4753 #endif /* INET6 */
4754 	default:
4755 		sport = dport = hdrlen = 0;
4756 		break;
4757 	}
4758 
4759 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
4760 
4761 	/* check packet for BINAT/NAT/RDR */
4762 	transerror = pf_get_translation(pd, m, off, kif, &nsn, &sk,
4763 	    &nk, saddr, daddr, sport, dport, anchor_stack, &nr);
4764 	switch (transerror) {
4765 	default:
4766 		/* A translation error occurred. */
4767 		REASON_SET(&reason, transerror);
4768 		goto cleanup;
4769 	case PFRES_MAX:
4770 		/* No match. */
4771 		break;
4772 	case PFRES_MATCH:
4773 		KASSERT(sk != NULL, ("%s: null sk", __func__));
4774 		KASSERT(nk != NULL, ("%s: null nk", __func__));
4775 
4776 		if (nr->log) {
4777 			PFLOG_PACKET(kif, m, af, PF_PASS, PFRES_MATCH, nr, a,
4778 			    ruleset, pd, 1);
4779 		}
4780 
4781 		if (pd->ip_sum)
4782 			bip_sum = *pd->ip_sum;
4783 
4784 		switch (pd->proto) {
4785 		case IPPROTO_TCP:
4786 			bproto_sum = th->th_sum;
4787 			pd->proto_sum = &th->th_sum;
4788 
4789 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
4790 			    nk->port[pd->sidx] != sport) {
4791 				pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
4792 				    &th->th_sum, &nk->addr[pd->sidx],
4793 				    nk->port[pd->sidx], 0, af);
4794 				pd->sport = &th->th_sport;
4795 				sport = th->th_sport;
4796 			}
4797 
4798 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
4799 			    nk->port[pd->didx] != dport) {
4800 				pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
4801 				    &th->th_sum, &nk->addr[pd->didx],
4802 				    nk->port[pd->didx], 0, af);
4803 				dport = th->th_dport;
4804 				pd->dport = &th->th_dport;
4805 			}
4806 			rewrite++;
4807 			break;
4808 		case IPPROTO_UDP:
4809 			bproto_sum = pd->hdr.udp.uh_sum;
4810 			pd->proto_sum = &pd->hdr.udp.uh_sum;
4811 
4812 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
4813 			    nk->port[pd->sidx] != sport) {
4814 				pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport,
4815 				    pd->ip_sum, &pd->hdr.udp.uh_sum,
4816 				    &nk->addr[pd->sidx],
4817 				    nk->port[pd->sidx], 1, af);
4818 				sport = pd->hdr.udp.uh_sport;
4819 				pd->sport = &pd->hdr.udp.uh_sport;
4820 			}
4821 
4822 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
4823 			    nk->port[pd->didx] != dport) {
4824 				pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport,
4825 				    pd->ip_sum, &pd->hdr.udp.uh_sum,
4826 				    &nk->addr[pd->didx],
4827 				    nk->port[pd->didx], 1, af);
4828 				dport = pd->hdr.udp.uh_dport;
4829 				pd->dport = &pd->hdr.udp.uh_dport;
4830 			}
4831 			rewrite++;
4832 			break;
4833 		case IPPROTO_SCTP: {
4834 			uint16_t checksum = 0;
4835 
4836 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
4837 			    nk->port[pd->sidx] != sport) {
4838 				pf_change_ap(m, saddr, &pd->hdr.sctp.src_port,
4839 				    pd->ip_sum, &checksum,
4840 				    &nk->addr[pd->sidx],
4841 				    nk->port[pd->sidx], 1, af);
4842 			}
4843 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
4844 			    nk->port[pd->didx] != dport) {
4845 				pf_change_ap(m, daddr, &pd->hdr.sctp.dest_port,
4846 				    pd->ip_sum, &checksum,
4847 				    &nk->addr[pd->didx],
4848 				    nk->port[pd->didx], 1, af);
4849 			}
4850 			break;
4851 		}
4852 #ifdef INET
4853 		case IPPROTO_ICMP:
4854 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
4855 				pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
4856 				    nk->addr[pd->sidx].v4.s_addr, 0);
4857 
4858 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
4859 				pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
4860 				    nk->addr[pd->didx].v4.s_addr, 0);
4861 
4862 			if (virtual_type == htons(ICMP_ECHO) &&
4863 			     nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) {
4864 				pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
4865 				    pd->hdr.icmp.icmp_cksum, sport,
4866 				    nk->port[pd->sidx], 0);
4867 				pd->hdr.icmp.icmp_id = nk->port[pd->sidx];
4868 				pd->sport = &pd->hdr.icmp.icmp_id;
4869 			}
4870 			m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
4871 			break;
4872 #endif /* INET */
4873 #ifdef INET6
4874 		case IPPROTO_ICMPV6:
4875 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
4876 				pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum,
4877 				    &nk->addr[pd->sidx], 0);
4878 
4879 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
4880 				pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum,
4881 				    &nk->addr[pd->didx], 0);
4882 			rewrite++;
4883 			break;
4884 #endif /* INET */
4885 		default:
4886 			switch (af) {
4887 #ifdef INET
4888 			case AF_INET:
4889 				if (PF_ANEQ(saddr,
4890 				    &nk->addr[pd->sidx], AF_INET))
4891 					pf_change_a(&saddr->v4.s_addr,
4892 					    pd->ip_sum,
4893 					    nk->addr[pd->sidx].v4.s_addr, 0);
4894 
4895 				if (PF_ANEQ(daddr,
4896 				    &nk->addr[pd->didx], AF_INET))
4897 					pf_change_a(&daddr->v4.s_addr,
4898 					    pd->ip_sum,
4899 					    nk->addr[pd->didx].v4.s_addr, 0);
4900 				break;
4901 #endif /* INET */
4902 #ifdef INET6
4903 			case AF_INET6:
4904 				if (PF_ANEQ(saddr,
4905 				    &nk->addr[pd->sidx], AF_INET6))
4906 					PF_ACPY(saddr, &nk->addr[pd->sidx], af);
4907 
4908 				if (PF_ANEQ(daddr,
4909 				    &nk->addr[pd->didx], AF_INET6))
4910 					PF_ACPY(daddr, &nk->addr[pd->didx], af);
4911 				break;
4912 #endif /* INET */
4913 			}
4914 			break;
4915 		}
4916 		if (nr->natpass)
4917 			r = NULL;
4918 		pd->nat_rule = nr;
4919 	}
4920 
4921 	while (r != NULL) {
4922 		pf_counter_u64_add(&r->evaluations, 1);
4923 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
4924 			r = r->skip[PF_SKIP_IFP].ptr;
4925 		else if (r->direction && r->direction != pd->dir)
4926 			r = r->skip[PF_SKIP_DIR].ptr;
4927 		else if (r->af && r->af != af)
4928 			r = r->skip[PF_SKIP_AF].ptr;
4929 		else if (r->proto && r->proto != pd->proto)
4930 			r = r->skip[PF_SKIP_PROTO].ptr;
4931 		else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
4932 		    r->src.neg, kif, M_GETFIB(m)))
4933 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
4934 		/* tcp/udp only. port_op always 0 in other cases */
4935 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
4936 		    r->src.port[0], r->src.port[1], sport))
4937 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
4938 		else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
4939 		    r->dst.neg, NULL, M_GETFIB(m)))
4940 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
4941 		/* tcp/udp only. port_op always 0 in other cases */
4942 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
4943 		    r->dst.port[0], r->dst.port[1], dport))
4944 			r = r->skip[PF_SKIP_DST_PORT].ptr;
4945 		/* icmp only. type always 0 in other cases */
4946 		else if (r->type && r->type != icmptype + 1)
4947 			r = TAILQ_NEXT(r, entries);
4948 		/* icmp only. type always 0 in other cases */
4949 		else if (r->code && r->code != icmpcode + 1)
4950 			r = TAILQ_NEXT(r, entries);
4951 		else if (r->tos && !(r->tos == pd->tos))
4952 			r = TAILQ_NEXT(r, entries);
4953 		else if (r->rule_flag & PFRULE_FRAGMENT)
4954 			r = TAILQ_NEXT(r, entries);
4955 		else if (pd->proto == IPPROTO_TCP &&
4956 		    (r->flagset & th->th_flags) != r->flags)
4957 			r = TAILQ_NEXT(r, entries);
4958 		/* tcp/udp only. uid.op always 0 in other cases */
4959 		else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
4960 		    pf_socket_lookup(pd, m), 1)) &&
4961 		    !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
4962 		    pd->lookup.uid))
4963 			r = TAILQ_NEXT(r, entries);
4964 		/* tcp/udp only. gid.op always 0 in other cases */
4965 		else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
4966 		    pf_socket_lookup(pd, m), 1)) &&
4967 		    !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
4968 		    pd->lookup.gid))
4969 			r = TAILQ_NEXT(r, entries);
4970 		else if (r->prio &&
4971 		    !pf_match_ieee8021q_pcp(r->prio, m))
4972 			r = TAILQ_NEXT(r, entries);
4973 		else if (r->prob &&
4974 		    r->prob <= arc4random())
4975 			r = TAILQ_NEXT(r, entries);
4976 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
4977 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
4978 			r = TAILQ_NEXT(r, entries);
4979 		else if (r->os_fingerprint != PF_OSFP_ANY &&
4980 		    (pd->proto != IPPROTO_TCP || !pf_osfp_match(
4981 		    pf_osfp_fingerprint(pd, m, off, th),
4982 		    r->os_fingerprint)))
4983 			r = TAILQ_NEXT(r, entries);
4984 		else {
4985 			if (r->tag)
4986 				tag = r->tag;
4987 			if (r->anchor == NULL) {
4988 				if (r->action == PF_MATCH) {
4989 					ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO);
4990 					if (ri == NULL) {
4991 						REASON_SET(&reason, PFRES_MEMORY);
4992 						goto cleanup;
4993 					}
4994 					ri->r = r;
4995 					SLIST_INSERT_HEAD(&match_rules, ri, entry);
4996 					pf_counter_u64_critical_enter();
4997 					pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
4998 					pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
4999 					pf_counter_u64_critical_exit();
5000 					pf_rule_to_actions(r, &pd->act);
5001 					if (r->log)
5002 						PFLOG_PACKET(kif, m, af,
5003 						    r->action, PFRES_MATCH, r,
5004 						    a, ruleset, pd, 1);
5005 				} else {
5006 					match = 1;
5007 					*rm = r;
5008 					*am = a;
5009 					*rsm = ruleset;
5010 				}
5011 				if ((*rm)->quick)
5012 					break;
5013 				r = TAILQ_NEXT(r, entries);
5014 			} else
5015 				pf_step_into_anchor(anchor_stack, &asd,
5016 				    &ruleset, PF_RULESET_FILTER, &r, &a,
5017 				    &match);
5018 		}
5019 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
5020 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
5021 			break;
5022 	}
5023 	r = *rm;
5024 	a = *am;
5025 	ruleset = *rsm;
5026 
5027 	REASON_SET(&reason, PFRES_MATCH);
5028 
5029 	/* apply actions for last matching pass/block rule */
5030 	pf_rule_to_actions(r, &pd->act);
5031 
5032 	if (r->log) {
5033 		if (rewrite)
5034 			m_copyback(m, off, hdrlen, pd->hdr.any);
5035 		PFLOG_PACKET(kif, m, af, r->action, reason, r, a, ruleset, pd, 1);
5036 	}
5037 
5038 	if ((r->action == PF_DROP) &&
5039 	    ((r->rule_flag & PFRULE_RETURNRST) ||
5040 	    (r->rule_flag & PFRULE_RETURNICMP) ||
5041 	    (r->rule_flag & PFRULE_RETURN))) {
5042 		pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum,
5043 		    bip_sum, hdrlen, &reason, r->rtableid);
5044 	}
5045 
5046 	if (r->action == PF_DROP)
5047 		goto cleanup;
5048 
5049 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
5050 		REASON_SET(&reason, PFRES_MEMORY);
5051 		goto cleanup;
5052 	}
5053 	if (pd->act.rtableid >= 0)
5054 		M_SETFIB(m, pd->act.rtableid);
5055 
5056 	if (!state_icmp && (r->keep_state || nr != NULL ||
5057 	    (pd->flags & PFDESC_TCP_NORM))) {
5058 		int action;
5059 		action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
5060 		    sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
5061 		    hdrlen, &match_rules);
5062 		if (action != PF_PASS) {
5063 			if (action == PF_DROP &&
5064 			    (r->rule_flag & PFRULE_RETURN))
5065 				pf_return(r, nr, pd, sk, off, m, th, kif,
5066 				    bproto_sum, bip_sum, hdrlen, &reason,
5067 				    pd->act.rtableid);
5068 			return (action);
5069 		}
5070 	} else {
5071 		while ((ri = SLIST_FIRST(&match_rules))) {
5072 			SLIST_REMOVE_HEAD(&match_rules, entry);
5073 			free(ri, M_PF_RULE_ITEM);
5074 		}
5075 
5076 		uma_zfree(V_pf_state_key_z, sk);
5077 		uma_zfree(V_pf_state_key_z, nk);
5078 	}
5079 
5080 	/* copy back packet headers if we performed NAT operations */
5081 	if (rewrite)
5082 		m_copyback(m, off, hdrlen, pd->hdr.any);
5083 
5084 	if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
5085 	    pd->dir == PF_OUT &&
5086 	    V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m))
5087 		/*
5088 		 * We want the state created, but we dont
5089 		 * want to send this in case a partner
5090 		 * firewall has to know about it to allow
5091 		 * replies through it.
5092 		 */
5093 		return (PF_DEFER);
5094 
5095 	return (PF_PASS);
5096 
5097 cleanup:
5098 	while ((ri = SLIST_FIRST(&match_rules))) {
5099 		SLIST_REMOVE_HEAD(&match_rules, entry);
5100 		free(ri, M_PF_RULE_ITEM);
5101 	}
5102 
5103 	uma_zfree(V_pf_state_key_z, sk);
5104 	uma_zfree(V_pf_state_key_z, nk);
5105 	return (PF_DROP);
5106 }
5107 
5108 static int
5109 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a,
5110     struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk,
5111     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
5112     u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm,
5113     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
5114     struct pf_krule_slist *match_rules)
5115 {
5116 	struct pf_kstate	*s = NULL;
5117 	struct pf_ksrc_node	*sn = NULL;
5118 	struct tcphdr		*th = &pd->hdr.tcp;
5119 	u_int16_t		 mss = V_tcp_mssdflt;
5120 	u_short			 reason, sn_reason;
5121 	struct pf_krule_item	*ri;
5122 
5123 	/* check maximums */
5124 	if (r->max_states &&
5125 	    (counter_u64_fetch(r->states_cur) >= r->max_states)) {
5126 		counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
5127 		REASON_SET(&reason, PFRES_MAXSTATES);
5128 		goto csfailed;
5129 	}
5130 	/* src node for filter rule */
5131 	if ((r->rule_flag & PFRULE_SRCTRACK ||
5132 	    r->rpool.opts & PF_POOL_STICKYADDR) &&
5133 	    (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) {
5134 		REASON_SET(&reason, sn_reason);
5135 		goto csfailed;
5136 	}
5137 	/* src node for translation rule */
5138 	if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
5139 	    (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx],
5140 	    pd->af)) != 0 ) {
5141 		REASON_SET(&reason, sn_reason);
5142 		goto csfailed;
5143 	}
5144 	s = pf_alloc_state(M_NOWAIT);
5145 	if (s == NULL) {
5146 		REASON_SET(&reason, PFRES_MEMORY);
5147 		goto csfailed;
5148 	}
5149 	s->rule.ptr = r;
5150 	s->nat_rule.ptr = nr;
5151 	s->anchor.ptr = a;
5152 	bcopy(match_rules, &s->match_rules, sizeof(s->match_rules));
5153 	memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions));
5154 
5155 	STATE_INC_COUNTERS(s);
5156 	if (r->allow_opts)
5157 		s->state_flags |= PFSTATE_ALLOWOPTS;
5158 	if (r->rule_flag & PFRULE_STATESLOPPY)
5159 		s->state_flags |= PFSTATE_SLOPPY;
5160 	if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */
5161 		s->state_flags |= PFSTATE_SCRUB_TCP;
5162 	if ((r->rule_flag & PFRULE_PFLOW) ||
5163 	    (nr != NULL && nr->rule_flag & PFRULE_PFLOW))
5164 		s->state_flags |= PFSTATE_PFLOW;
5165 
5166 	s->act.log = pd->act.log & PF_LOG_ALL;
5167 	s->sync_state = PFSYNC_S_NONE;
5168 	s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */
5169 
5170 	if (nr != NULL)
5171 		s->act.log |= nr->log & PF_LOG_ALL;
5172 	switch (pd->proto) {
5173 	case IPPROTO_TCP:
5174 		s->src.seqlo = ntohl(th->th_seq);
5175 		s->src.seqhi = s->src.seqlo + pd->p_len + 1;
5176 		if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
5177 		    r->keep_state == PF_STATE_MODULATE) {
5178 			/* Generate sequence number modulator */
5179 			if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
5180 			    0)
5181 				s->src.seqdiff = 1;
5182 			pf_change_proto_a(m, &th->th_seq, &th->th_sum,
5183 			    htonl(s->src.seqlo + s->src.seqdiff), 0);
5184 			*rewrite = 1;
5185 		} else
5186 			s->src.seqdiff = 0;
5187 		if (th->th_flags & TH_SYN) {
5188 			s->src.seqhi++;
5189 			s->src.wscale = pf_get_wscale(m, off,
5190 			    th->th_off, pd->af);
5191 		}
5192 		s->src.max_win = MAX(ntohs(th->th_win), 1);
5193 		if (s->src.wscale & PF_WSCALE_MASK) {
5194 			/* Remove scale factor from initial window */
5195 			int win = s->src.max_win;
5196 			win += 1 << (s->src.wscale & PF_WSCALE_MASK);
5197 			s->src.max_win = (win - 1) >>
5198 			    (s->src.wscale & PF_WSCALE_MASK);
5199 		}
5200 		if (th->th_flags & TH_FIN)
5201 			s->src.seqhi++;
5202 		s->dst.seqhi = 1;
5203 		s->dst.max_win = 1;
5204 		pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT);
5205 		pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED);
5206 		s->timeout = PFTM_TCP_FIRST_PACKET;
5207 		atomic_add_32(&V_pf_status.states_halfopen, 1);
5208 		break;
5209 	case IPPROTO_UDP:
5210 		pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE);
5211 		pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC);
5212 		s->timeout = PFTM_UDP_FIRST_PACKET;
5213 		break;
5214 	case IPPROTO_SCTP:
5215 		pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT);
5216 		pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED);
5217 		s->timeout = PFTM_SCTP_FIRST_PACKET;
5218 		break;
5219 	case IPPROTO_ICMP:
5220 #ifdef INET6
5221 	case IPPROTO_ICMPV6:
5222 #endif
5223 		s->timeout = PFTM_ICMP_FIRST_PACKET;
5224 		break;
5225 	default:
5226 		pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE);
5227 		pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC);
5228 		s->timeout = PFTM_OTHER_FIRST_PACKET;
5229 	}
5230 
5231 	if (r->rt) {
5232 		/* pf_map_addr increases the reason counters */
5233 		if ((reason = pf_map_addr(pd->af, r, pd->src, &s->rt_addr,
5234 		    &s->rt_kif, NULL, &sn)) != 0)
5235 			goto csfailed;
5236 		s->rt = r->rt;
5237 	}
5238 
5239 	s->creation = s->expire = pf_get_uptime();
5240 
5241 	if (sn != NULL)
5242 		s->src_node = sn;
5243 	if (nsn != NULL) {
5244 		/* XXX We only modify one side for now. */
5245 		PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
5246 		s->nat_src_node = nsn;
5247 	}
5248 	if (pd->proto == IPPROTO_TCP) {
5249 		if (s->state_flags & PFSTATE_SCRUB_TCP &&
5250 		    pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) {
5251 			REASON_SET(&reason, PFRES_MEMORY);
5252 			goto drop;
5253 		}
5254 		if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub &&
5255 		    pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
5256 		    &s->src, &s->dst, rewrite)) {
5257 			/* This really shouldn't happen!!! */
5258 			DPFPRINTF(PF_DEBUG_URGENT,
5259 			    ("pf_normalize_tcp_stateful failed on first "
5260 			     "pkt\n"));
5261 			goto drop;
5262 		}
5263 	} else if (pd->proto == IPPROTO_SCTP) {
5264 		if (pf_normalize_sctp_init(m, off, pd, &s->src, &s->dst))
5265 			goto drop;
5266 		if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP)))
5267 			goto drop;
5268 	}
5269 	s->direction = pd->dir;
5270 
5271 	/*
5272 	 * sk/nk could already been setup by pf_get_translation().
5273 	 */
5274 	if (nr == NULL) {
5275 		KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
5276 		    __func__, nr, sk, nk));
5277 		sk = pf_state_key_setup(pd, m, off, pd->src, pd->dst, sport, dport);
5278 		if (sk == NULL)
5279 			goto csfailed;
5280 		nk = sk;
5281 	} else
5282 		KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
5283 		    __func__, nr, sk, nk));
5284 
5285 	/* Swap sk/nk for PF_OUT. */
5286 	if (pf_state_insert(BOUND_IFACE(s, kif), kif,
5287 	    (pd->dir == PF_IN) ? sk : nk,
5288 	    (pd->dir == PF_IN) ? nk : sk, s)) {
5289 		REASON_SET(&reason, PFRES_STATEINS);
5290 		goto drop;
5291 	} else
5292 		*sm = s;
5293 
5294 	if (tag > 0)
5295 		s->tag = tag;
5296 	if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
5297 	    TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
5298 		pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC);
5299 		/* undo NAT changes, if they have taken place */
5300 		if (nr != NULL) {
5301 			struct pf_state_key *skt = s->key[PF_SK_WIRE];
5302 			if (pd->dir == PF_OUT)
5303 				skt = s->key[PF_SK_STACK];
5304 			PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
5305 			PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
5306 			if (pd->sport)
5307 				*pd->sport = skt->port[pd->sidx];
5308 			if (pd->dport)
5309 				*pd->dport = skt->port[pd->didx];
5310 			if (pd->proto_sum)
5311 				*pd->proto_sum = bproto_sum;
5312 			if (pd->ip_sum)
5313 				*pd->ip_sum = bip_sum;
5314 			m_copyback(m, off, hdrlen, pd->hdr.any);
5315 		}
5316 		s->src.seqhi = htonl(arc4random());
5317 		/* Find mss option */
5318 		int rtid = M_GETFIB(m);
5319 		mss = pf_get_mss(m, off, th->th_off, pd->af);
5320 		mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
5321 		mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
5322 		s->src.mss = mss;
5323 		pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport,
5324 		    th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
5325 		    TH_SYN|TH_ACK, 0, s->src.mss, 0, true, 0, 0,
5326 		    pd->act.rtableid);
5327 		REASON_SET(&reason, PFRES_SYNPROXY);
5328 		return (PF_SYNPROXY_DROP);
5329 	}
5330 
5331 	return (PF_PASS);
5332 
5333 csfailed:
5334 	while ((ri = SLIST_FIRST(match_rules))) {
5335 		SLIST_REMOVE_HEAD(match_rules, entry);
5336 		free(ri, M_PF_RULE_ITEM);
5337 	}
5338 
5339 	uma_zfree(V_pf_state_key_z, sk);
5340 	uma_zfree(V_pf_state_key_z, nk);
5341 
5342 	if (sn != NULL) {
5343 		PF_SRC_NODE_LOCK(sn);
5344 		if (--sn->states == 0 && sn->expire == 0) {
5345 			pf_unlink_src_node(sn);
5346 			uma_zfree(V_pf_sources_z, sn);
5347 			counter_u64_add(
5348 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
5349 		}
5350 		PF_SRC_NODE_UNLOCK(sn);
5351 	}
5352 
5353 	if (nsn != sn && nsn != NULL) {
5354 		PF_SRC_NODE_LOCK(nsn);
5355 		if (--nsn->states == 0 && nsn->expire == 0) {
5356 			pf_unlink_src_node(nsn);
5357 			uma_zfree(V_pf_sources_z, nsn);
5358 			counter_u64_add(
5359 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
5360 		}
5361 		PF_SRC_NODE_UNLOCK(nsn);
5362 	}
5363 
5364 drop:
5365 	if (s != NULL) {
5366 		pf_src_tree_remove_state(s);
5367 		s->timeout = PFTM_UNLINKED;
5368 		STATE_DEC_COUNTERS(s);
5369 		pf_free_state(s);
5370 	}
5371 
5372 	return (PF_DROP);
5373 }
5374 
5375 static int
5376 pf_test_fragment(struct pf_krule **rm, struct pfi_kkif *kif,
5377     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am,
5378     struct pf_kruleset **rsm)
5379 {
5380 	struct pf_krule		*r, *a = NULL;
5381 	struct pf_kruleset	*ruleset = NULL;
5382 	struct pf_krule_slist	 match_rules;
5383 	struct pf_krule_item	*ri;
5384 	sa_family_t		 af = pd->af;
5385 	u_short			 reason;
5386 	int			 tag = -1;
5387 	int			 asd = 0;
5388 	int			 match = 0;
5389 	struct pf_kanchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
5390 
5391 	PF_RULES_RASSERT();
5392 
5393 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
5394 	SLIST_INIT(&match_rules);
5395 	while (r != NULL) {
5396 		pf_counter_u64_add(&r->evaluations, 1);
5397 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
5398 			r = r->skip[PF_SKIP_IFP].ptr;
5399 		else if (r->direction && r->direction != pd->dir)
5400 			r = r->skip[PF_SKIP_DIR].ptr;
5401 		else if (r->af && r->af != af)
5402 			r = r->skip[PF_SKIP_AF].ptr;
5403 		else if (r->proto && r->proto != pd->proto)
5404 			r = r->skip[PF_SKIP_PROTO].ptr;
5405 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
5406 		    r->src.neg, kif, M_GETFIB(m)))
5407 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
5408 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
5409 		    r->dst.neg, NULL, M_GETFIB(m)))
5410 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
5411 		else if (r->tos && !(r->tos == pd->tos))
5412 			r = TAILQ_NEXT(r, entries);
5413 		else if (r->os_fingerprint != PF_OSFP_ANY)
5414 			r = TAILQ_NEXT(r, entries);
5415 		else if (pd->proto == IPPROTO_UDP &&
5416 		    (r->src.port_op || r->dst.port_op))
5417 			r = TAILQ_NEXT(r, entries);
5418 		else if (pd->proto == IPPROTO_TCP &&
5419 		    (r->src.port_op || r->dst.port_op || r->flagset))
5420 			r = TAILQ_NEXT(r, entries);
5421 		else if ((pd->proto == IPPROTO_ICMP ||
5422 		    pd->proto == IPPROTO_ICMPV6) &&
5423 		    (r->type || r->code))
5424 			r = TAILQ_NEXT(r, entries);
5425 		else if (r->prio &&
5426 		    !pf_match_ieee8021q_pcp(r->prio, m))
5427 			r = TAILQ_NEXT(r, entries);
5428 		else if (r->prob && r->prob <=
5429 		    (arc4random() % (UINT_MAX - 1) + 1))
5430 			r = TAILQ_NEXT(r, entries);
5431 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
5432 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
5433 			r = TAILQ_NEXT(r, entries);
5434 		else {
5435 			if (r->anchor == NULL) {
5436 				if (r->action == PF_MATCH) {
5437 					ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO);
5438 					if (ri == NULL) {
5439 						REASON_SET(&reason, PFRES_MEMORY);
5440 						goto cleanup;
5441 					}
5442 					ri->r = r;
5443 					SLIST_INSERT_HEAD(&match_rules, ri, entry);
5444 					pf_counter_u64_critical_enter();
5445 					pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
5446 					pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
5447 					pf_counter_u64_critical_exit();
5448 					pf_rule_to_actions(r, &pd->act);
5449 					if (r->log)
5450 						PFLOG_PACKET(kif, m, af,
5451 						    r->action, PFRES_MATCH, r,
5452 						    a, ruleset, pd, 1);
5453 				} else {
5454 					match = 1;
5455 					*rm = r;
5456 					*am = a;
5457 					*rsm = ruleset;
5458 				}
5459 				if ((*rm)->quick)
5460 					break;
5461 				r = TAILQ_NEXT(r, entries);
5462 			} else
5463 				pf_step_into_anchor(anchor_stack, &asd,
5464 				    &ruleset, PF_RULESET_FILTER, &r, &a,
5465 				    &match);
5466 		}
5467 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
5468 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
5469 			break;
5470 	}
5471 	r = *rm;
5472 	a = *am;
5473 	ruleset = *rsm;
5474 
5475 	REASON_SET(&reason, PFRES_MATCH);
5476 
5477 	/* apply actions for last matching pass/block rule */
5478 	pf_rule_to_actions(r, &pd->act);
5479 
5480 	if (r->log)
5481 		PFLOG_PACKET(kif, m, af, r->action, reason, r, a, ruleset, pd, 1);
5482 
5483 	if (r->action != PF_PASS)
5484 		return (PF_DROP);
5485 
5486 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
5487 		REASON_SET(&reason, PFRES_MEMORY);
5488 		goto cleanup;
5489 	}
5490 
5491 	return (PF_PASS);
5492 
5493 cleanup:
5494 	while ((ri = SLIST_FIRST(&match_rules))) {
5495 		SLIST_REMOVE_HEAD(&match_rules, entry);
5496 		free(ri, M_PF_RULE_ITEM);
5497 	}
5498 
5499 	return (PF_DROP);
5500 }
5501 
5502 static int
5503 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif,
5504     struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason,
5505     int *copyback)
5506 {
5507 	struct tcphdr		*th = &pd->hdr.tcp;
5508 	struct pf_state_peer	*src, *dst;
5509 	u_int16_t		 win = ntohs(th->th_win);
5510 	u_int32_t		 ack, end, data_end, seq, orig_seq;
5511 	u_int8_t		 sws, dws, psrc, pdst;
5512 	int			 ackskew;
5513 
5514 	if (pd->dir == (*state)->direction) {
5515 		src = &(*state)->src;
5516 		dst = &(*state)->dst;
5517 		psrc = PF_PEER_SRC;
5518 		pdst = PF_PEER_DST;
5519 	} else {
5520 		src = &(*state)->dst;
5521 		dst = &(*state)->src;
5522 		psrc = PF_PEER_DST;
5523 		pdst = PF_PEER_SRC;
5524 	}
5525 
5526 	if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
5527 		sws = src->wscale & PF_WSCALE_MASK;
5528 		dws = dst->wscale & PF_WSCALE_MASK;
5529 	} else
5530 		sws = dws = 0;
5531 
5532 	/*
5533 	 * Sequence tracking algorithm from Guido van Rooij's paper:
5534 	 *   http://www.madison-gurkha.com/publications/tcp_filtering/
5535 	 *	tcp_filtering.ps
5536 	 */
5537 
5538 	orig_seq = seq = ntohl(th->th_seq);
5539 	if (src->seqlo == 0) {
5540 		/* First packet from this end. Set its state */
5541 
5542 		if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) &&
5543 		    src->scrub == NULL) {
5544 			if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
5545 				REASON_SET(reason, PFRES_MEMORY);
5546 				return (PF_DROP);
5547 			}
5548 		}
5549 
5550 		/* Deferred generation of sequence number modulator */
5551 		if (dst->seqdiff && !src->seqdiff) {
5552 			/* use random iss for the TCP server */
5553 			while ((src->seqdiff = arc4random() - seq) == 0)
5554 				;
5555 			ack = ntohl(th->th_ack) - dst->seqdiff;
5556 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
5557 			    src->seqdiff), 0);
5558 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
5559 			*copyback = 1;
5560 		} else {
5561 			ack = ntohl(th->th_ack);
5562 		}
5563 
5564 		end = seq + pd->p_len;
5565 		if (th->th_flags & TH_SYN) {
5566 			end++;
5567 			if (dst->wscale & PF_WSCALE_FLAG) {
5568 				src->wscale = pf_get_wscale(m, off, th->th_off,
5569 				    pd->af);
5570 				if (src->wscale & PF_WSCALE_FLAG) {
5571 					/* Remove scale factor from initial
5572 					 * window */
5573 					sws = src->wscale & PF_WSCALE_MASK;
5574 					win = ((u_int32_t)win + (1 << sws) - 1)
5575 					    >> sws;
5576 					dws = dst->wscale & PF_WSCALE_MASK;
5577 				} else {
5578 					/* fixup other window */
5579 					dst->max_win = MIN(TCP_MAXWIN,
5580 					    (u_int32_t)dst->max_win <<
5581 					    (dst->wscale & PF_WSCALE_MASK));
5582 					/* in case of a retrans SYN|ACK */
5583 					dst->wscale = 0;
5584 				}
5585 			}
5586 		}
5587 		data_end = end;
5588 		if (th->th_flags & TH_FIN)
5589 			end++;
5590 
5591 		src->seqlo = seq;
5592 		if (src->state < TCPS_SYN_SENT)
5593 			pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
5594 
5595 		/*
5596 		 * May need to slide the window (seqhi may have been set by
5597 		 * the crappy stack check or if we picked up the connection
5598 		 * after establishment)
5599 		 */
5600 		if (src->seqhi == 1 ||
5601 		    SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
5602 			src->seqhi = end + MAX(1, dst->max_win << dws);
5603 		if (win > src->max_win)
5604 			src->max_win = win;
5605 
5606 	} else {
5607 		ack = ntohl(th->th_ack) - dst->seqdiff;
5608 		if (src->seqdiff) {
5609 			/* Modulate sequence numbers */
5610 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
5611 			    src->seqdiff), 0);
5612 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
5613 			*copyback = 1;
5614 		}
5615 		end = seq + pd->p_len;
5616 		if (th->th_flags & TH_SYN)
5617 			end++;
5618 		data_end = end;
5619 		if (th->th_flags & TH_FIN)
5620 			end++;
5621 	}
5622 
5623 	if ((th->th_flags & TH_ACK) == 0) {
5624 		/* Let it pass through the ack skew check */
5625 		ack = dst->seqlo;
5626 	} else if ((ack == 0 &&
5627 	    (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
5628 	    /* broken tcp stacks do not set ack */
5629 	    (dst->state < TCPS_SYN_SENT)) {
5630 		/*
5631 		 * Many stacks (ours included) will set the ACK number in an
5632 		 * FIN|ACK if the SYN times out -- no sequence to ACK.
5633 		 */
5634 		ack = dst->seqlo;
5635 	}
5636 
5637 	if (seq == end) {
5638 		/* Ease sequencing restrictions on no data packets */
5639 		seq = src->seqlo;
5640 		data_end = end = seq;
5641 	}
5642 
5643 	ackskew = dst->seqlo - ack;
5644 
5645 	/*
5646 	 * Need to demodulate the sequence numbers in any TCP SACK options
5647 	 * (Selective ACK). We could optionally validate the SACK values
5648 	 * against the current ACK window, either forwards or backwards, but
5649 	 * I'm not confident that SACK has been implemented properly
5650 	 * everywhere. It wouldn't surprise me if several stacks accidentally
5651 	 * SACK too far backwards of previously ACKed data. There really aren't
5652 	 * any security implications of bad SACKing unless the target stack
5653 	 * doesn't validate the option length correctly. Someone trying to
5654 	 * spoof into a TCP connection won't bother blindly sending SACK
5655 	 * options anyway.
5656 	 */
5657 	if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
5658 		if (pf_modulate_sack(m, off, pd, th, dst))
5659 			*copyback = 1;
5660 	}
5661 
5662 #define	MAXACKWINDOW (0xffff + 1500)	/* 1500 is an arbitrary fudge factor */
5663 	if (SEQ_GEQ(src->seqhi, data_end) &&
5664 	    /* Last octet inside other's window space */
5665 	    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
5666 	    /* Retrans: not more than one window back */
5667 	    (ackskew >= -MAXACKWINDOW) &&
5668 	    /* Acking not more than one reassembled fragment backwards */
5669 	    (ackskew <= (MAXACKWINDOW << sws)) &&
5670 	    /* Acking not more than one window forward */
5671 	    ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
5672 	    (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) {
5673 	    /* Require an exact/+1 sequence match on resets when possible */
5674 
5675 		if (dst->scrub || src->scrub) {
5676 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
5677 			    *state, src, dst, copyback))
5678 				return (PF_DROP);
5679 		}
5680 
5681 		/* update max window */
5682 		if (src->max_win < win)
5683 			src->max_win = win;
5684 		/* synchronize sequencing */
5685 		if (SEQ_GT(end, src->seqlo))
5686 			src->seqlo = end;
5687 		/* slide the window of what the other end can send */
5688 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
5689 			dst->seqhi = ack + MAX((win << sws), 1);
5690 
5691 		/* update states */
5692 		if (th->th_flags & TH_SYN)
5693 			if (src->state < TCPS_SYN_SENT)
5694 				pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
5695 		if (th->th_flags & TH_FIN)
5696 			if (src->state < TCPS_CLOSING)
5697 				pf_set_protostate(*state, psrc, TCPS_CLOSING);
5698 		if (th->th_flags & TH_ACK) {
5699 			if (dst->state == TCPS_SYN_SENT) {
5700 				pf_set_protostate(*state, pdst,
5701 				    TCPS_ESTABLISHED);
5702 				if (src->state == TCPS_ESTABLISHED &&
5703 				    (*state)->src_node != NULL &&
5704 				    pf_src_connlimit(state)) {
5705 					REASON_SET(reason, PFRES_SRCLIMIT);
5706 					return (PF_DROP);
5707 				}
5708 			} else if (dst->state == TCPS_CLOSING)
5709 				pf_set_protostate(*state, pdst,
5710 				    TCPS_FIN_WAIT_2);
5711 		}
5712 		if (th->th_flags & TH_RST)
5713 			pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
5714 
5715 		/* update expire time */
5716 		(*state)->expire = pf_get_uptime();
5717 		if (src->state >= TCPS_FIN_WAIT_2 &&
5718 		    dst->state >= TCPS_FIN_WAIT_2)
5719 			(*state)->timeout = PFTM_TCP_CLOSED;
5720 		else if (src->state >= TCPS_CLOSING &&
5721 		    dst->state >= TCPS_CLOSING)
5722 			(*state)->timeout = PFTM_TCP_FIN_WAIT;
5723 		else if (src->state < TCPS_ESTABLISHED ||
5724 		    dst->state < TCPS_ESTABLISHED)
5725 			(*state)->timeout = PFTM_TCP_OPENING;
5726 		else if (src->state >= TCPS_CLOSING ||
5727 		    dst->state >= TCPS_CLOSING)
5728 			(*state)->timeout = PFTM_TCP_CLOSING;
5729 		else
5730 			(*state)->timeout = PFTM_TCP_ESTABLISHED;
5731 
5732 		/* Fall through to PASS packet */
5733 
5734 	} else if ((dst->state < TCPS_SYN_SENT ||
5735 		dst->state >= TCPS_FIN_WAIT_2 ||
5736 		src->state >= TCPS_FIN_WAIT_2) &&
5737 	    SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) &&
5738 	    /* Within a window forward of the originating packet */
5739 	    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
5740 	    /* Within a window backward of the originating packet */
5741 
5742 		/*
5743 		 * This currently handles three situations:
5744 		 *  1) Stupid stacks will shotgun SYNs before their peer
5745 		 *     replies.
5746 		 *  2) When PF catches an already established stream (the
5747 		 *     firewall rebooted, the state table was flushed, routes
5748 		 *     changed...)
5749 		 *  3) Packets get funky immediately after the connection
5750 		 *     closes (this should catch Solaris spurious ACK|FINs
5751 		 *     that web servers like to spew after a close)
5752 		 *
5753 		 * This must be a little more careful than the above code
5754 		 * since packet floods will also be caught here. We don't
5755 		 * update the TTL here to mitigate the damage of a packet
5756 		 * flood and so the same code can handle awkward establishment
5757 		 * and a loosened connection close.
5758 		 * In the establishment case, a correct peer response will
5759 		 * validate the connection, go through the normal state code
5760 		 * and keep updating the state TTL.
5761 		 */
5762 
5763 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
5764 			printf("pf: loose state match: ");
5765 			pf_print_state(*state);
5766 			pf_print_flags(th->th_flags);
5767 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
5768 			    "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
5769 			    pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
5770 			    (unsigned long long)(*state)->packets[1],
5771 			    pd->dir == PF_IN ? "in" : "out",
5772 			    pd->dir == (*state)->direction ? "fwd" : "rev");
5773 		}
5774 
5775 		if (dst->scrub || src->scrub) {
5776 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
5777 			    *state, src, dst, copyback))
5778 				return (PF_DROP);
5779 		}
5780 
5781 		/* update max window */
5782 		if (src->max_win < win)
5783 			src->max_win = win;
5784 		/* synchronize sequencing */
5785 		if (SEQ_GT(end, src->seqlo))
5786 			src->seqlo = end;
5787 		/* slide the window of what the other end can send */
5788 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
5789 			dst->seqhi = ack + MAX((win << sws), 1);
5790 
5791 		/*
5792 		 * Cannot set dst->seqhi here since this could be a shotgunned
5793 		 * SYN and not an already established connection.
5794 		 */
5795 
5796 		if (th->th_flags & TH_FIN)
5797 			if (src->state < TCPS_CLOSING)
5798 				pf_set_protostate(*state, psrc, TCPS_CLOSING);
5799 		if (th->th_flags & TH_RST)
5800 			pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
5801 
5802 		/* Fall through to PASS packet */
5803 
5804 	} else {
5805 		if ((*state)->dst.state == TCPS_SYN_SENT &&
5806 		    (*state)->src.state == TCPS_SYN_SENT) {
5807 			/* Send RST for state mismatches during handshake */
5808 			if (!(th->th_flags & TH_RST))
5809 				pf_send_tcp((*state)->rule.ptr, pd->af,
5810 				    pd->dst, pd->src, th->th_dport,
5811 				    th->th_sport, ntohl(th->th_ack), 0,
5812 				    TH_RST, 0, 0,
5813 				    (*state)->rule.ptr->return_ttl, true, 0, 0,
5814 				    (*state)->act.rtableid);
5815 			src->seqlo = 0;
5816 			src->seqhi = 1;
5817 			src->max_win = 1;
5818 		} else if (V_pf_status.debug >= PF_DEBUG_MISC) {
5819 			printf("pf: BAD state: ");
5820 			pf_print_state(*state);
5821 			pf_print_flags(th->th_flags);
5822 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
5823 			    "pkts=%llu:%llu dir=%s,%s\n",
5824 			    seq, orig_seq, ack, pd->p_len, ackskew,
5825 			    (unsigned long long)(*state)->packets[0],
5826 			    (unsigned long long)(*state)->packets[1],
5827 			    pd->dir == PF_IN ? "in" : "out",
5828 			    pd->dir == (*state)->direction ? "fwd" : "rev");
5829 			printf("pf: State failure on: %c %c %c %c | %c %c\n",
5830 			    SEQ_GEQ(src->seqhi, data_end) ? ' ' : '1',
5831 			    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
5832 			    ' ': '2',
5833 			    (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
5834 			    (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
5835 			    SEQ_GEQ(src->seqhi + MAXACKWINDOW, data_end) ?' ' :'5',
5836 			    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
5837 		}
5838 		REASON_SET(reason, PFRES_BADSTATE);
5839 		return (PF_DROP);
5840 	}
5841 
5842 	return (PF_PASS);
5843 }
5844 
5845 static int
5846 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason)
5847 {
5848 	struct tcphdr		*th = &pd->hdr.tcp;
5849 	struct pf_state_peer	*src, *dst;
5850 	u_int8_t		 psrc, pdst;
5851 
5852 	if (pd->dir == (*state)->direction) {
5853 		src = &(*state)->src;
5854 		dst = &(*state)->dst;
5855 		psrc = PF_PEER_SRC;
5856 		pdst = PF_PEER_DST;
5857 	} else {
5858 		src = &(*state)->dst;
5859 		dst = &(*state)->src;
5860 		psrc = PF_PEER_DST;
5861 		pdst = PF_PEER_SRC;
5862 	}
5863 
5864 	if (th->th_flags & TH_SYN)
5865 		if (src->state < TCPS_SYN_SENT)
5866 			pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
5867 	if (th->th_flags & TH_FIN)
5868 		if (src->state < TCPS_CLOSING)
5869 			pf_set_protostate(*state, psrc, TCPS_CLOSING);
5870 	if (th->th_flags & TH_ACK) {
5871 		if (dst->state == TCPS_SYN_SENT) {
5872 			pf_set_protostate(*state, pdst, TCPS_ESTABLISHED);
5873 			if (src->state == TCPS_ESTABLISHED &&
5874 			    (*state)->src_node != NULL &&
5875 			    pf_src_connlimit(state)) {
5876 				REASON_SET(reason, PFRES_SRCLIMIT);
5877 				return (PF_DROP);
5878 			}
5879 		} else if (dst->state == TCPS_CLOSING) {
5880 			pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2);
5881 		} else if (src->state == TCPS_SYN_SENT &&
5882 		    dst->state < TCPS_SYN_SENT) {
5883 			/*
5884 			 * Handle a special sloppy case where we only see one
5885 			 * half of the connection. If there is a ACK after
5886 			 * the initial SYN without ever seeing a packet from
5887 			 * the destination, set the connection to established.
5888 			 */
5889 			pf_set_protostate(*state, PF_PEER_BOTH,
5890 			    TCPS_ESTABLISHED);
5891 			dst->state = src->state = TCPS_ESTABLISHED;
5892 			if ((*state)->src_node != NULL &&
5893 			    pf_src_connlimit(state)) {
5894 				REASON_SET(reason, PFRES_SRCLIMIT);
5895 				return (PF_DROP);
5896 			}
5897 		} else if (src->state == TCPS_CLOSING &&
5898 		    dst->state == TCPS_ESTABLISHED &&
5899 		    dst->seqlo == 0) {
5900 			/*
5901 			 * Handle the closing of half connections where we
5902 			 * don't see the full bidirectional FIN/ACK+ACK
5903 			 * handshake.
5904 			 */
5905 			pf_set_protostate(*state, pdst, TCPS_CLOSING);
5906 		}
5907 	}
5908 	if (th->th_flags & TH_RST)
5909 		pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
5910 
5911 	/* update expire time */
5912 	(*state)->expire = pf_get_uptime();
5913 	if (src->state >= TCPS_FIN_WAIT_2 &&
5914 	    dst->state >= TCPS_FIN_WAIT_2)
5915 		(*state)->timeout = PFTM_TCP_CLOSED;
5916 	else if (src->state >= TCPS_CLOSING &&
5917 	    dst->state >= TCPS_CLOSING)
5918 		(*state)->timeout = PFTM_TCP_FIN_WAIT;
5919 	else if (src->state < TCPS_ESTABLISHED ||
5920 	    dst->state < TCPS_ESTABLISHED)
5921 		(*state)->timeout = PFTM_TCP_OPENING;
5922 	else if (src->state >= TCPS_CLOSING ||
5923 	    dst->state >= TCPS_CLOSING)
5924 		(*state)->timeout = PFTM_TCP_CLOSING;
5925 	else
5926 		(*state)->timeout = PFTM_TCP_ESTABLISHED;
5927 
5928 	return (PF_PASS);
5929 }
5930 
5931 static int
5932 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason)
5933 {
5934 	struct pf_state_key	*sk = (*state)->key[pd->didx];
5935 	struct tcphdr		*th = &pd->hdr.tcp;
5936 
5937 	if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
5938 		if (pd->dir != (*state)->direction) {
5939 			REASON_SET(reason, PFRES_SYNPROXY);
5940 			return (PF_SYNPROXY_DROP);
5941 		}
5942 		if (th->th_flags & TH_SYN) {
5943 			if (ntohl(th->th_seq) != (*state)->src.seqlo) {
5944 				REASON_SET(reason, PFRES_SYNPROXY);
5945 				return (PF_DROP);
5946 			}
5947 			pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
5948 			    pd->src, th->th_dport, th->th_sport,
5949 			    (*state)->src.seqhi, ntohl(th->th_seq) + 1,
5950 			    TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, true, 0, 0,
5951 			    (*state)->act.rtableid);
5952 			REASON_SET(reason, PFRES_SYNPROXY);
5953 			return (PF_SYNPROXY_DROP);
5954 		} else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK ||
5955 		    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
5956 		    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
5957 			REASON_SET(reason, PFRES_SYNPROXY);
5958 			return (PF_DROP);
5959 		} else if ((*state)->src_node != NULL &&
5960 		    pf_src_connlimit(state)) {
5961 			REASON_SET(reason, PFRES_SRCLIMIT);
5962 			return (PF_DROP);
5963 		} else
5964 			pf_set_protostate(*state, PF_PEER_SRC,
5965 			    PF_TCPS_PROXY_DST);
5966 	}
5967 	if ((*state)->src.state == PF_TCPS_PROXY_DST) {
5968 		if (pd->dir == (*state)->direction) {
5969 			if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
5970 			    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
5971 			    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
5972 				REASON_SET(reason, PFRES_SYNPROXY);
5973 				return (PF_DROP);
5974 			}
5975 			(*state)->src.max_win = MAX(ntohs(th->th_win), 1);
5976 			if ((*state)->dst.seqhi == 1)
5977 				(*state)->dst.seqhi = htonl(arc4random());
5978 			pf_send_tcp((*state)->rule.ptr, pd->af,
5979 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
5980 			    sk->port[pd->sidx], sk->port[pd->didx],
5981 			    (*state)->dst.seqhi, 0, TH_SYN, 0,
5982 			    (*state)->src.mss, 0, false, (*state)->tag, 0,
5983 			    (*state)->act.rtableid);
5984 			REASON_SET(reason, PFRES_SYNPROXY);
5985 			return (PF_SYNPROXY_DROP);
5986 		} else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
5987 		    (TH_SYN|TH_ACK)) ||
5988 		    (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
5989 			REASON_SET(reason, PFRES_SYNPROXY);
5990 			return (PF_DROP);
5991 		} else {
5992 			(*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
5993 			(*state)->dst.seqlo = ntohl(th->th_seq);
5994 			pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
5995 			    pd->src, th->th_dport, th->th_sport,
5996 			    ntohl(th->th_ack), ntohl(th->th_seq) + 1,
5997 			    TH_ACK, (*state)->src.max_win, 0, 0, false,
5998 			    (*state)->tag, 0, (*state)->act.rtableid);
5999 			pf_send_tcp((*state)->rule.ptr, pd->af,
6000 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
6001 			    sk->port[pd->sidx], sk->port[pd->didx],
6002 			    (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
6003 			    TH_ACK, (*state)->dst.max_win, 0, 0, true, 0, 0,
6004 			    (*state)->act.rtableid);
6005 			(*state)->src.seqdiff = (*state)->dst.seqhi -
6006 			    (*state)->src.seqlo;
6007 			(*state)->dst.seqdiff = (*state)->src.seqhi -
6008 			    (*state)->dst.seqlo;
6009 			(*state)->src.seqhi = (*state)->src.seqlo +
6010 			    (*state)->dst.max_win;
6011 			(*state)->dst.seqhi = (*state)->dst.seqlo +
6012 			    (*state)->src.max_win;
6013 			(*state)->src.wscale = (*state)->dst.wscale = 0;
6014 			pf_set_protostate(*state, PF_PEER_BOTH,
6015 			    TCPS_ESTABLISHED);
6016 			REASON_SET(reason, PFRES_SYNPROXY);
6017 			return (PF_SYNPROXY_DROP);
6018 		}
6019 	}
6020 
6021 	return (PF_PASS);
6022 }
6023 
6024 static int
6025 pf_test_state_tcp(struct pf_kstate **state, struct pfi_kkif *kif,
6026     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
6027     u_short *reason)
6028 {
6029 	struct pf_state_key_cmp	 key;
6030 	struct tcphdr		*th = &pd->hdr.tcp;
6031 	int			 copyback = 0;
6032 	int			 action;
6033 	struct pf_state_peer	*src, *dst;
6034 
6035 	bzero(&key, sizeof(key));
6036 	key.af = pd->af;
6037 	key.proto = IPPROTO_TCP;
6038 	if (pd->dir == PF_IN)	{	/* wire side, straight */
6039 		PF_ACPY(&key.addr[0], pd->src, key.af);
6040 		PF_ACPY(&key.addr[1], pd->dst, key.af);
6041 		key.port[0] = th->th_sport;
6042 		key.port[1] = th->th_dport;
6043 	} else {			/* stack side, reverse */
6044 		PF_ACPY(&key.addr[1], pd->src, key.af);
6045 		PF_ACPY(&key.addr[0], pd->dst, key.af);
6046 		key.port[1] = th->th_sport;
6047 		key.port[0] = th->th_dport;
6048 	}
6049 
6050 	STATE_LOOKUP(kif, &key, *state, pd);
6051 
6052 	if (pd->dir == (*state)->direction) {
6053 		src = &(*state)->src;
6054 		dst = &(*state)->dst;
6055 	} else {
6056 		src = &(*state)->dst;
6057 		dst = &(*state)->src;
6058 	}
6059 
6060 	if ((action = pf_synproxy(pd, state, reason)) != PF_PASS)
6061 		return (action);
6062 
6063 	if (dst->state >= TCPS_FIN_WAIT_2 &&
6064 	    src->state >= TCPS_FIN_WAIT_2 &&
6065 	    (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) ||
6066 	    ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK &&
6067 	    pf_syncookie_check(pd) && pd->dir == PF_IN))) {
6068 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
6069 			printf("pf: state reuse ");
6070 			pf_print_state(*state);
6071 			pf_print_flags(th->th_flags);
6072 			printf("\n");
6073 		}
6074 		/* XXX make sure it's the same direction ?? */
6075 		pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED);
6076 		pf_unlink_state(*state);
6077 		*state = NULL;
6078 		return (PF_DROP);
6079 	}
6080 
6081 	if ((*state)->state_flags & PFSTATE_SLOPPY) {
6082 		if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP)
6083 			return (PF_DROP);
6084 	} else {
6085 		if (pf_tcp_track_full(state, kif, m, off, pd, reason,
6086 		    &copyback) == PF_DROP)
6087 			return (PF_DROP);
6088 	}
6089 
6090 	/* translate source/destination address, if necessary */
6091 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
6092 		struct pf_state_key *nk = (*state)->key[pd->didx];
6093 
6094 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
6095 		    nk->port[pd->sidx] != th->th_sport)
6096 			pf_change_ap(m, pd->src, &th->th_sport,
6097 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
6098 			    nk->port[pd->sidx], 0, pd->af);
6099 
6100 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
6101 		    nk->port[pd->didx] != th->th_dport)
6102 			pf_change_ap(m, pd->dst, &th->th_dport,
6103 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
6104 			    nk->port[pd->didx], 0, pd->af);
6105 		copyback = 1;
6106 	}
6107 
6108 	/* Copyback sequence modulation or stateful scrub changes if needed */
6109 	if (copyback)
6110 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
6111 
6112 	return (PF_PASS);
6113 }
6114 
6115 static int
6116 pf_test_state_udp(struct pf_kstate **state, struct pfi_kkif *kif,
6117     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
6118 {
6119 	struct pf_state_peer	*src, *dst;
6120 	struct pf_state_key_cmp	 key;
6121 	struct udphdr		*uh = &pd->hdr.udp;
6122 	uint8_t			 psrc, pdst;
6123 
6124 	bzero(&key, sizeof(key));
6125 	key.af = pd->af;
6126 	key.proto = IPPROTO_UDP;
6127 	if (pd->dir == PF_IN)	{	/* wire side, straight */
6128 		PF_ACPY(&key.addr[0], pd->src, key.af);
6129 		PF_ACPY(&key.addr[1], pd->dst, key.af);
6130 		key.port[0] = uh->uh_sport;
6131 		key.port[1] = uh->uh_dport;
6132 	} else {			/* stack side, reverse */
6133 		PF_ACPY(&key.addr[1], pd->src, key.af);
6134 		PF_ACPY(&key.addr[0], pd->dst, key.af);
6135 		key.port[1] = uh->uh_sport;
6136 		key.port[0] = uh->uh_dport;
6137 	}
6138 
6139 	STATE_LOOKUP(kif, &key, *state, pd);
6140 
6141 	if (pd->dir == (*state)->direction) {
6142 		src = &(*state)->src;
6143 		dst = &(*state)->dst;
6144 		psrc = PF_PEER_SRC;
6145 		pdst = PF_PEER_DST;
6146 	} else {
6147 		src = &(*state)->dst;
6148 		dst = &(*state)->src;
6149 		psrc = PF_PEER_DST;
6150 		pdst = PF_PEER_SRC;
6151 	}
6152 
6153 	/* update states */
6154 	if (src->state < PFUDPS_SINGLE)
6155 		pf_set_protostate(*state, psrc, PFUDPS_SINGLE);
6156 	if (dst->state == PFUDPS_SINGLE)
6157 		pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE);
6158 
6159 	/* update expire time */
6160 	(*state)->expire = pf_get_uptime();
6161 	if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
6162 		(*state)->timeout = PFTM_UDP_MULTIPLE;
6163 	else
6164 		(*state)->timeout = PFTM_UDP_SINGLE;
6165 
6166 	/* translate source/destination address, if necessary */
6167 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
6168 		struct pf_state_key *nk = (*state)->key[pd->didx];
6169 
6170 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
6171 		    nk->port[pd->sidx] != uh->uh_sport)
6172 			pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
6173 			    &uh->uh_sum, &nk->addr[pd->sidx],
6174 			    nk->port[pd->sidx], 1, pd->af);
6175 
6176 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
6177 		    nk->port[pd->didx] != uh->uh_dport)
6178 			pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
6179 			    &uh->uh_sum, &nk->addr[pd->didx],
6180 			    nk->port[pd->didx], 1, pd->af);
6181 		m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
6182 	}
6183 
6184 	return (PF_PASS);
6185 }
6186 
6187 static int
6188 pf_test_state_sctp(struct pf_kstate **state, struct pfi_kkif *kif,
6189     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
6190 {
6191 	struct pf_state_key_cmp	 key;
6192 	struct pf_state_peer	*src, *dst;
6193 	struct sctphdr		*sh = &pd->hdr.sctp;
6194 	u_int8_t		 psrc; //, pdst;
6195 
6196 	bzero(&key, sizeof(key));
6197 	key.af = pd->af;
6198 	key.proto = IPPROTO_SCTP;
6199 	if (pd->dir == PF_IN)	{	/* wire side, straight */
6200 		PF_ACPY(&key.addr[0], pd->src, key.af);
6201 		PF_ACPY(&key.addr[1], pd->dst, key.af);
6202 		key.port[0] = sh->src_port;
6203 		key.port[1] = sh->dest_port;
6204 	} else {			/* stack side, reverse */
6205 		PF_ACPY(&key.addr[1], pd->src, key.af);
6206 		PF_ACPY(&key.addr[0], pd->dst, key.af);
6207 		key.port[1] = sh->src_port;
6208 		key.port[0] = sh->dest_port;
6209 	}
6210 
6211 	STATE_LOOKUP(kif, &key, *state, pd);
6212 
6213 	if (pd->dir == (*state)->direction) {
6214 		src = &(*state)->src;
6215 		dst = &(*state)->dst;
6216 		psrc = PF_PEER_SRC;
6217 	} else {
6218 		src = &(*state)->dst;
6219 		dst = &(*state)->src;
6220 		psrc = PF_PEER_DST;
6221 	}
6222 
6223 	if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) &&
6224 	    (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) &&
6225 	    pd->sctp_flags & PFDESC_SCTP_INIT) {
6226 		pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED);
6227 		pf_unlink_state(*state);
6228 		*state = NULL;
6229 		return (PF_DROP);
6230 	}
6231 
6232 	/* Track state. */
6233 	if (pd->sctp_flags & PFDESC_SCTP_INIT) {
6234 		if (src->state < SCTP_COOKIE_WAIT) {
6235 			pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT);
6236 			(*state)->timeout = PFTM_SCTP_OPENING;
6237 		}
6238 	}
6239 	if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) {
6240 		MPASS(dst->scrub != NULL);
6241 		if (dst->scrub->pfss_v_tag == 0)
6242 			dst->scrub->pfss_v_tag = pd->sctp_initiate_tag;
6243 	}
6244 
6245 	if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) {
6246 		if (src->state < SCTP_ESTABLISHED) {
6247 			pf_set_protostate(*state, psrc, SCTP_ESTABLISHED);
6248 			(*state)->timeout = PFTM_SCTP_ESTABLISHED;
6249 		}
6250 	}
6251 	if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT |
6252 	    PFDESC_SCTP_SHUTDOWN_COMPLETE)) {
6253 		if (src->state < SCTP_SHUTDOWN_PENDING) {
6254 			pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING);
6255 			(*state)->timeout = PFTM_SCTP_CLOSING;
6256 		}
6257 	}
6258 	if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) {
6259 		pf_set_protostate(*state, psrc, SCTP_CLOSED);
6260 		(*state)->timeout = PFTM_SCTP_CLOSED;
6261 	}
6262 
6263 	if (src->scrub != NULL) {
6264 		if (src->scrub->pfss_v_tag == 0) {
6265 			src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag;
6266 		} else  if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag)
6267 			return (PF_DROP);
6268 	}
6269 
6270 	(*state)->expire = pf_get_uptime();
6271 
6272 	/* translate source/destination address, if necessary */
6273 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
6274 		uint16_t checksum = 0;
6275 		struct pf_state_key *nk = (*state)->key[pd->didx];
6276 
6277 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
6278 		    nk->port[pd->sidx] != pd->hdr.sctp.src_port) {
6279 			pf_change_ap(m, pd->src, &pd->hdr.sctp.src_port,
6280 			    pd->ip_sum, &checksum, &nk->addr[pd->sidx],
6281 			    nk->port[pd->sidx], 1, pd->af);
6282 		}
6283 
6284 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
6285 		    nk->port[pd->didx] != pd->hdr.sctp.dest_port) {
6286 			pf_change_ap(m, pd->dst, &pd->hdr.sctp.dest_port,
6287 			    pd->ip_sum, &checksum, &nk->addr[pd->didx],
6288 			    nk->port[pd->didx], 1, pd->af);
6289 		}
6290 	}
6291 
6292 	return (PF_PASS);
6293 }
6294 
6295 static void
6296 pf_sctp_multihome_detach_addr(const struct pf_kstate *s)
6297 {
6298 	struct pf_sctp_endpoint key;
6299 	struct pf_sctp_endpoint *ep;
6300 	struct pf_state_key *sks = s->key[PF_SK_STACK];
6301 	struct pf_sctp_source *i, *tmp;
6302 
6303 	if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL)
6304 		return;
6305 
6306 	PF_SCTP_ENDPOINTS_LOCK();
6307 
6308 	key.v_tag = s->dst.scrub->pfss_v_tag;
6309 	ep  = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6310 	if (ep != NULL) {
6311 		TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) {
6312 			if (pf_addr_cmp(&i->addr,
6313 			    &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT],
6314 			    s->key[PF_SK_WIRE]->af) == 0) {
6315 				SDT_PROBE3(pf, sctp, multihome, remove,
6316 				    key.v_tag, s, i);
6317 				TAILQ_REMOVE(&ep->sources, i, entry);
6318 				free(i, M_PFTEMP);
6319 				break;
6320 			}
6321 		}
6322 
6323 		if (TAILQ_EMPTY(&ep->sources)) {
6324 			RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
6325 			free(ep, M_PFTEMP);
6326 		}
6327 	}
6328 
6329 	/* Other direction. */
6330 	key.v_tag = s->src.scrub->pfss_v_tag;
6331 	ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6332 	if (ep != NULL) {
6333 		TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) {
6334 			if (pf_addr_cmp(&i->addr,
6335 			    &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN],
6336 			    s->key[PF_SK_WIRE]->af) == 0) {
6337 				SDT_PROBE3(pf, sctp, multihome, remove,
6338 				    key.v_tag, s, i);
6339 				TAILQ_REMOVE(&ep->sources, i, entry);
6340 				free(i, M_PFTEMP);
6341 				break;
6342 			}
6343 		}
6344 
6345 		if (TAILQ_EMPTY(&ep->sources)) {
6346 			RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
6347 			free(ep, M_PFTEMP);
6348 		}
6349 	}
6350 
6351 	PF_SCTP_ENDPOINTS_UNLOCK();
6352 }
6353 
6354 static void
6355 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag)
6356 {
6357 	struct pf_sctp_endpoint key = {
6358 		.v_tag = v_tag,
6359 	};
6360 	struct pf_sctp_source *i;
6361 	struct pf_sctp_endpoint *ep;
6362 
6363 	PF_SCTP_ENDPOINTS_LOCK();
6364 
6365 	ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6366 	if (ep == NULL) {
6367 		ep = malloc(sizeof(struct pf_sctp_endpoint),
6368 		    M_PFTEMP, M_NOWAIT);
6369 		if (ep == NULL) {
6370 			PF_SCTP_ENDPOINTS_UNLOCK();
6371 			return;
6372 		}
6373 
6374 		ep->v_tag = v_tag;
6375 		TAILQ_INIT(&ep->sources);
6376 		RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
6377 	}
6378 
6379 	/* Avoid inserting duplicates. */
6380 	TAILQ_FOREACH(i, &ep->sources, entry) {
6381 		if (pf_addr_cmp(&i->addr, a, pd->af) == 0) {
6382 			PF_SCTP_ENDPOINTS_UNLOCK();
6383 			return;
6384 		}
6385 	}
6386 
6387 	i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT);
6388 	if (i == NULL) {
6389 		PF_SCTP_ENDPOINTS_UNLOCK();
6390 		return;
6391 	}
6392 
6393 	i->af = pd->af;
6394 	memcpy(&i->addr, a, sizeof(*a));
6395 	TAILQ_INSERT_TAIL(&ep->sources, i, entry);
6396 	SDT_PROBE2(pf, sctp, multihome, add, v_tag, i);
6397 
6398 	PF_SCTP_ENDPOINTS_UNLOCK();
6399 }
6400 
6401 static void
6402 pf_sctp_multihome_delayed(struct pf_pdesc *pd, int off, struct pfi_kkif *kif,
6403     struct pf_kstate *s, int action)
6404 {
6405 	struct pf_sctp_multihome_job	*j, *tmp;
6406 	struct pf_sctp_source		*i;
6407 	int			 ret __unused;
6408 	struct pf_kstate	*sm = NULL;
6409 	struct pf_krule		*ra = NULL;
6410 	struct pf_krule		*r = &V_pf_default_rule;
6411 	struct pf_kruleset	*rs = NULL;
6412 	bool do_extra = true;
6413 
6414 	PF_RULES_RLOCK_TRACKER;
6415 
6416 again:
6417 	TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) {
6418 		if (s == NULL || action != PF_PASS)
6419 			goto free;
6420 
6421 		/* Confirm we don't recurse here. */
6422 		MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP));
6423 
6424 		switch (j->op) {
6425 		case  SCTP_ADD_IP_ADDRESS: {
6426 			uint32_t v_tag = pd->sctp_initiate_tag;
6427 
6428 			if (v_tag == 0) {
6429 				if (s->direction == pd->dir)
6430 					v_tag = s->src.scrub->pfss_v_tag;
6431 				else
6432 					v_tag = s->dst.scrub->pfss_v_tag;
6433 			}
6434 
6435 			/*
6436 			 * Avoid duplicating states. We'll already have
6437 			 * created a state based on the source address of
6438 			 * the packet, but SCTP endpoints may also list this
6439 			 * address again in the INIT(_ACK) parameters.
6440 			 */
6441 			if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) {
6442 				break;
6443 			}
6444 
6445 			j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP;
6446 			PF_RULES_RLOCK();
6447 			sm = NULL;
6448 			/*
6449 			 * New connections need to be floating, because
6450 			 * we cannot know what interfaces it will use.
6451 			 * That's why we pass V_pfi_all rather than kif.
6452 			 */
6453 			ret = pf_test_rule(&r, &sm, V_pfi_all,
6454 			    j->m, off, &j->pd, &ra, &rs, NULL);
6455 			PF_RULES_RUNLOCK();
6456 			SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->m, ret);
6457 			if (ret != PF_DROP && sm != NULL) {
6458 				/* Inherit v_tag values. */
6459 				if (sm->direction == s->direction) {
6460 					sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag;
6461 					sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag;
6462 				} else {
6463 					sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag;
6464 					sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag;
6465 				}
6466 				PF_STATE_UNLOCK(sm);
6467 			} else {
6468 				/* If we try duplicate inserts? */
6469 				break;
6470 			}
6471 
6472 			/* Only add the address if we've actually allowed the state. */
6473 			pf_sctp_multihome_add_addr(pd, &j->src, v_tag);
6474 
6475 			if (! do_extra) {
6476 				break;
6477 			}
6478 			/*
6479 			 * We need to do this for each of our source addresses.
6480 			 * Find those based on the verification tag.
6481 			 */
6482 			struct pf_sctp_endpoint key = {
6483 				.v_tag = pd->hdr.sctp.v_tag,
6484 			};
6485 			struct pf_sctp_endpoint *ep;
6486 
6487 			PF_SCTP_ENDPOINTS_LOCK();
6488 			ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6489 			if (ep == NULL) {
6490 				PF_SCTP_ENDPOINTS_UNLOCK();
6491 				break;
6492 			}
6493 			MPASS(ep != NULL);
6494 
6495 			TAILQ_FOREACH(i, &ep->sources, entry) {
6496 				struct pf_sctp_multihome_job *nj;
6497 
6498 				/* SCTP can intermingle IPv4 and IPv6. */
6499 				if (i->af != pd->af)
6500 					continue;
6501 
6502 				nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO);
6503 				if (! nj) {
6504 					continue;
6505 				}
6506 				memcpy(&nj->pd, &j->pd, sizeof(j->pd));
6507 				memcpy(&nj->src, &j->src, sizeof(nj->src));
6508 				nj->pd.src = &nj->src;
6509 				// New destination address!
6510 				memcpy(&nj->dst, &i->addr, sizeof(nj->dst));
6511 				nj->pd.dst = &nj->dst;
6512 				nj->m = j->m;
6513 				nj->op = j->op;
6514 
6515 				TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next);
6516 			}
6517 			PF_SCTP_ENDPOINTS_UNLOCK();
6518 
6519 			break;
6520 		}
6521 		case SCTP_DEL_IP_ADDRESS: {
6522 			struct pf_state_key_cmp key;
6523 			uint8_t psrc;
6524 
6525 			bzero(&key, sizeof(key));
6526 			key.af = j->pd.af;
6527 			key.proto = IPPROTO_SCTP;
6528 			if (j->pd.dir == PF_IN)	{	/* wire side, straight */
6529 				PF_ACPY(&key.addr[0], j->pd.src, key.af);
6530 				PF_ACPY(&key.addr[1], j->pd.dst, key.af);
6531 				key.port[0] = j->pd.hdr.sctp.src_port;
6532 				key.port[1] = j->pd.hdr.sctp.dest_port;
6533 			} else {			/* stack side, reverse */
6534 				PF_ACPY(&key.addr[1], j->pd.src, key.af);
6535 				PF_ACPY(&key.addr[0], j->pd.dst, key.af);
6536 				key.port[1] = j->pd.hdr.sctp.src_port;
6537 				key.port[0] = j->pd.hdr.sctp.dest_port;
6538 			}
6539 
6540 			sm = pf_find_state(kif, &key, j->pd.dir);
6541 			if (sm != NULL) {
6542 				PF_STATE_LOCK_ASSERT(sm);
6543 				if (j->pd.dir == sm->direction) {
6544 					psrc = PF_PEER_SRC;
6545 				} else {
6546 					psrc = PF_PEER_DST;
6547 				}
6548 				pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING);
6549 				sm->timeout = PFTM_SCTP_CLOSING;
6550 				PF_STATE_UNLOCK(sm);
6551 			}
6552 			break;
6553 		default:
6554 			panic("Unknown op %#x", j->op);
6555 		}
6556 	}
6557 
6558 	free:
6559 		TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next);
6560 		free(j, M_PFTEMP);
6561 	}
6562 
6563 	/* We may have inserted extra work while processing the list. */
6564 	if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) {
6565 		do_extra = false;
6566 		goto again;
6567 	}
6568 }
6569 
6570 static int
6571 pf_multihome_scan(struct mbuf *m, int start, int len, struct pf_pdesc *pd,
6572     struct pfi_kkif *kif, int op)
6573 {
6574 	int			 off = 0;
6575 	struct pf_sctp_multihome_job	*job;
6576 
6577 	while (off < len) {
6578 		struct sctp_paramhdr h;
6579 
6580 		if (!pf_pull_hdr(m, start + off, &h, sizeof(h), NULL, NULL,
6581 		    pd->af))
6582 			return (PF_DROP);
6583 
6584 		/* Parameters are at least 4 bytes. */
6585 		if (ntohs(h.param_length) < 4)
6586 			return (PF_DROP);
6587 
6588 		switch (ntohs(h.param_type)) {
6589 		case  SCTP_IPV4_ADDRESS: {
6590 			struct in_addr t;
6591 
6592 			if (ntohs(h.param_length) !=
6593 			    (sizeof(struct sctp_paramhdr) + sizeof(t)))
6594 				return (PF_DROP);
6595 
6596 			if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t),
6597 			    NULL, NULL, pd->af))
6598 				return (PF_DROP);
6599 
6600 			if (in_nullhost(t))
6601 				t.s_addr = pd->src->v4.s_addr;
6602 
6603 			/*
6604 			 * We hold the state lock (idhash) here, which means
6605 			 * that we can't acquire the keyhash, or we'll get a
6606 			 * LOR (and potentially double-lock things too). We also
6607 			 * can't release the state lock here, so instead we'll
6608 			 * enqueue this for async handling.
6609 			 * There's a relatively small race here, in that a
6610 			 * packet using the new addresses could arrive already,
6611 			 * but that's just though luck for it.
6612 			 */
6613 			job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO);
6614 			if (! job)
6615 				return (PF_DROP);
6616 
6617 			memcpy(&job->pd, pd, sizeof(*pd));
6618 
6619 			// New source address!
6620 			memcpy(&job->src, &t, sizeof(t));
6621 			job->pd.src = &job->src;
6622 			memcpy(&job->dst, pd->dst, sizeof(job->dst));
6623 			job->pd.dst = &job->dst;
6624 			job->m = m;
6625 			job->op = op;
6626 
6627 			TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next);
6628 			break;
6629 		}
6630 #ifdef INET6
6631 		case SCTP_IPV6_ADDRESS: {
6632 			struct in6_addr t;
6633 
6634 			if (ntohs(h.param_length) !=
6635 			    (sizeof(struct sctp_paramhdr) + sizeof(t)))
6636 				return (PF_DROP);
6637 
6638 			if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t),
6639 			    NULL, NULL, pd->af))
6640 				return (PF_DROP);
6641 			if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0)
6642 				break;
6643 			if (memcmp(&t, &in6addr_any, sizeof(t)) == 0)
6644 				memcpy(&t, &pd->src->v6, sizeof(t));
6645 
6646 			job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO);
6647 			if (! job)
6648 				return (PF_DROP);
6649 
6650 			memcpy(&job->pd, pd, sizeof(*pd));
6651 			memcpy(&job->src, &t, sizeof(t));
6652 			job->pd.src = &job->src;
6653 			memcpy(&job->dst, pd->dst, sizeof(job->dst));
6654 			job->pd.dst = &job->dst;
6655 			job->m = m;
6656 			job->op = op;
6657 
6658 			TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next);
6659 			break;
6660 		}
6661 #endif
6662 		case SCTP_ADD_IP_ADDRESS: {
6663 			int ret;
6664 			struct sctp_asconf_paramhdr ah;
6665 
6666 			if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah),
6667 			    NULL, NULL, pd->af))
6668 				return (PF_DROP);
6669 
6670 			ret = pf_multihome_scan(m, start + off + sizeof(ah),
6671 			    ntohs(ah.ph.param_length) - sizeof(ah), pd, kif,
6672 			    SCTP_ADD_IP_ADDRESS);
6673 			if (ret != PF_PASS)
6674 				return (ret);
6675 			break;
6676 		}
6677 		case SCTP_DEL_IP_ADDRESS: {
6678 			int ret;
6679 			struct sctp_asconf_paramhdr ah;
6680 
6681 			if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah),
6682 			    NULL, NULL, pd->af))
6683 				return (PF_DROP);
6684 			ret = pf_multihome_scan(m, start + off + sizeof(ah),
6685 			    ntohs(ah.ph.param_length) - sizeof(ah), pd, kif,
6686 			    SCTP_DEL_IP_ADDRESS);
6687 			if (ret != PF_PASS)
6688 				return (ret);
6689 			break;
6690 		}
6691 		default:
6692 			break;
6693 		}
6694 
6695 		off += roundup(ntohs(h.param_length), 4);
6696 	}
6697 
6698 	return (PF_PASS);
6699 }
6700 int
6701 pf_multihome_scan_init(struct mbuf *m, int start, int len, struct pf_pdesc *pd,
6702     struct pfi_kkif *kif)
6703 {
6704 	start += sizeof(struct sctp_init_chunk);
6705 	len -= sizeof(struct sctp_init_chunk);
6706 
6707 	return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS));
6708 }
6709 
6710 int
6711 pf_multihome_scan_asconf(struct mbuf *m, int start, int len,
6712     struct pf_pdesc *pd, struct pfi_kkif *kif)
6713 {
6714 	start += sizeof(struct sctp_asconf_chunk);
6715 	len -= sizeof(struct sctp_asconf_chunk);
6716 
6717 	return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS));
6718 }
6719 
6720 int
6721 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd,
6722     struct pf_kstate **state, struct mbuf *m, int off, int direction,
6723     struct pfi_kkif *kif, u_int16_t icmpid, u_int16_t type, int icmp_dir,
6724     int *iidx, int multi, int inner)
6725 {
6726 	key->af = pd->af;
6727 	key->proto = pd->proto;
6728 	if (icmp_dir == PF_IN) {
6729 		*iidx = pd->sidx;
6730 		key->port[pd->sidx] = icmpid;
6731 		key->port[pd->didx] = type;
6732 	} else {
6733 		*iidx = pd->didx;
6734 		key->port[pd->sidx] = type;
6735 		key->port[pd->didx] = icmpid;
6736 	}
6737 	if (pf_state_key_addr_setup(pd, m, off, key, pd->sidx, pd->src,
6738 	    pd->didx, pd->dst, multi))
6739 		return (PF_DROP);
6740 
6741 	STATE_LOOKUP(kif, key, *state, pd);
6742 
6743 	if ((*state)->state_flags & PFSTATE_SLOPPY)
6744 		return (-1);
6745 
6746 	/* Is this ICMP message flowing in right direction? */
6747 	if ((*state)->rule.ptr->type &&
6748 	    (((!inner && (*state)->direction == direction) ||
6749 	    (inner && (*state)->direction != direction)) ?
6750 	    PF_IN : PF_OUT) != icmp_dir) {
6751 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
6752 			printf("pf: icmp type %d in wrong direction (%d): ",
6753 			    icmp_dir, pd->dir);
6754 			pf_print_state(*state);
6755 			printf("\n");
6756 		}
6757 		PF_STATE_UNLOCK(*state);
6758 		*state = NULL;
6759 		return (PF_DROP);
6760 	}
6761 	return (-1);
6762 }
6763 
6764 static int
6765 pf_test_state_icmp(struct pf_kstate **state, struct pfi_kkif *kif,
6766     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
6767 {
6768 	struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
6769 	u_int16_t	*icmpsum, virtual_id, virtual_type;
6770 	u_int8_t	 icmptype, icmpcode;
6771 	int		 icmp_dir, iidx, ret, multi;
6772 	struct pf_state_key_cmp key;
6773 #ifdef INET
6774 	u_int16_t	 icmpid;
6775 #endif
6776 
6777 	MPASS(*state == NULL);
6778 
6779 	bzero(&key, sizeof(key));
6780 	switch (pd->proto) {
6781 #ifdef INET
6782 	case IPPROTO_ICMP:
6783 		icmptype = pd->hdr.icmp.icmp_type;
6784 		icmpcode = pd->hdr.icmp.icmp_code;
6785 		icmpid = pd->hdr.icmp.icmp_id;
6786 		icmpsum = &pd->hdr.icmp.icmp_cksum;
6787 		break;
6788 #endif /* INET */
6789 #ifdef INET6
6790 	case IPPROTO_ICMPV6:
6791 		icmptype = pd->hdr.icmp6.icmp6_type;
6792 		icmpcode = pd->hdr.icmp6.icmp6_code;
6793 #ifdef INET
6794 		icmpid = pd->hdr.icmp6.icmp6_id;
6795 #endif
6796 		icmpsum = &pd->hdr.icmp6.icmp6_cksum;
6797 		break;
6798 #endif /* INET6 */
6799 	}
6800 
6801 	if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &multi,
6802 	    &virtual_id, &virtual_type) == 0) {
6803 		/*
6804 		 * ICMP query/reply message not related to a TCP/UDP packet.
6805 		 * Search for an ICMP state.
6806 		 */
6807 		ret = pf_icmp_state_lookup(&key, pd, state, m, off, pd->dir,
6808 		    kif, virtual_id, virtual_type, icmp_dir, &iidx,
6809 		    PF_ICMP_MULTI_NONE, 0);
6810 		if (ret >= 0) {
6811 			MPASS(*state == NULL);
6812 			if (ret == PF_DROP && pd->af == AF_INET6 &&
6813 			    icmp_dir == PF_OUT) {
6814 				ret = pf_icmp_state_lookup(&key, pd, state, m, off,
6815 				    pd->dir, kif, virtual_id, virtual_type,
6816 				    icmp_dir, &iidx, multi, 0);
6817 				if (ret >= 0) {
6818 					MPASS(*state == NULL);
6819 					return (ret);
6820 				}
6821 			} else
6822 				return (ret);
6823 		}
6824 
6825 		(*state)->expire = pf_get_uptime();
6826 		(*state)->timeout = PFTM_ICMP_ERROR_REPLY;
6827 
6828 		/* translate source/destination address, if necessary */
6829 		if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
6830 			struct pf_state_key *nk = (*state)->key[pd->didx];
6831 
6832 			switch (pd->af) {
6833 #ifdef INET
6834 			case AF_INET:
6835 				if (PF_ANEQ(pd->src,
6836 				    &nk->addr[pd->sidx], AF_INET))
6837 					pf_change_a(&saddr->v4.s_addr,
6838 					    pd->ip_sum,
6839 					    nk->addr[pd->sidx].v4.s_addr, 0);
6840 
6841 				if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
6842 				    AF_INET))
6843 					pf_change_a(&daddr->v4.s_addr,
6844 					    pd->ip_sum,
6845 					    nk->addr[pd->didx].v4.s_addr, 0);
6846 
6847 				if (nk->port[iidx] !=
6848 				    pd->hdr.icmp.icmp_id) {
6849 					pd->hdr.icmp.icmp_cksum =
6850 					    pf_cksum_fixup(
6851 					    pd->hdr.icmp.icmp_cksum, icmpid,
6852 					    nk->port[iidx], 0);
6853 					pd->hdr.icmp.icmp_id =
6854 					    nk->port[iidx];
6855 				}
6856 
6857 				m_copyback(m, off, ICMP_MINLEN,
6858 				    (caddr_t )&pd->hdr.icmp);
6859 				break;
6860 #endif /* INET */
6861 #ifdef INET6
6862 			case AF_INET6:
6863 				if (PF_ANEQ(pd->src,
6864 				    &nk->addr[pd->sidx], AF_INET6))
6865 					pf_change_a6(saddr,
6866 					    &pd->hdr.icmp6.icmp6_cksum,
6867 					    &nk->addr[pd->sidx], 0);
6868 
6869 				if (PF_ANEQ(pd->dst,
6870 				    &nk->addr[pd->didx], AF_INET6))
6871 					pf_change_a6(daddr,
6872 					    &pd->hdr.icmp6.icmp6_cksum,
6873 					    &nk->addr[pd->didx], 0);
6874 
6875 				m_copyback(m, off, sizeof(struct icmp6_hdr),
6876 				    (caddr_t )&pd->hdr.icmp6);
6877 				break;
6878 #endif /* INET6 */
6879 			}
6880 		}
6881 		return (PF_PASS);
6882 
6883 	} else {
6884 		/*
6885 		 * ICMP error message in response to a TCP/UDP packet.
6886 		 * Extract the inner TCP/UDP header and search for that state.
6887 		 */
6888 
6889 		struct pf_pdesc	pd2;
6890 		bzero(&pd2, sizeof pd2);
6891 #ifdef INET
6892 		struct ip	h2;
6893 #endif /* INET */
6894 #ifdef INET6
6895 		struct ip6_hdr	h2_6;
6896 		int		terminal = 0;
6897 #endif /* INET6 */
6898 		int		ipoff2 = 0;
6899 		int		off2 = 0;
6900 
6901 		pd2.af = pd->af;
6902 		pd2.dir = pd->dir;
6903 		/* Payload packet is from the opposite direction. */
6904 		pd2.sidx = (pd->dir == PF_IN) ? 1 : 0;
6905 		pd2.didx = (pd->dir == PF_IN) ? 0 : 1;
6906 		switch (pd->af) {
6907 #ifdef INET
6908 		case AF_INET:
6909 			/* offset of h2 in mbuf chain */
6910 			ipoff2 = off + ICMP_MINLEN;
6911 
6912 			if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
6913 			    NULL, reason, pd2.af)) {
6914 				DPFPRINTF(PF_DEBUG_MISC,
6915 				    ("pf: ICMP error message too short "
6916 				    "(ip)\n"));
6917 				return (PF_DROP);
6918 			}
6919 			/*
6920 			 * ICMP error messages don't refer to non-first
6921 			 * fragments
6922 			 */
6923 			if (h2.ip_off & htons(IP_OFFMASK)) {
6924 				REASON_SET(reason, PFRES_FRAG);
6925 				return (PF_DROP);
6926 			}
6927 
6928 			/* offset of protocol header that follows h2 */
6929 			off2 = ipoff2 + (h2.ip_hl << 2);
6930 
6931 			pd2.proto = h2.ip_p;
6932 			pd2.src = (struct pf_addr *)&h2.ip_src;
6933 			pd2.dst = (struct pf_addr *)&h2.ip_dst;
6934 			pd2.ip_sum = &h2.ip_sum;
6935 			break;
6936 #endif /* INET */
6937 #ifdef INET6
6938 		case AF_INET6:
6939 			ipoff2 = off + sizeof(struct icmp6_hdr);
6940 
6941 			if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
6942 			    NULL, reason, pd2.af)) {
6943 				DPFPRINTF(PF_DEBUG_MISC,
6944 				    ("pf: ICMP error message too short "
6945 				    "(ip6)\n"));
6946 				return (PF_DROP);
6947 			}
6948 			pd2.proto = h2_6.ip6_nxt;
6949 			pd2.src = (struct pf_addr *)&h2_6.ip6_src;
6950 			pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
6951 			pd2.ip_sum = NULL;
6952 			off2 = ipoff2 + sizeof(h2_6);
6953 			do {
6954 				switch (pd2.proto) {
6955 				case IPPROTO_FRAGMENT:
6956 					/*
6957 					 * ICMPv6 error messages for
6958 					 * non-first fragments
6959 					 */
6960 					REASON_SET(reason, PFRES_FRAG);
6961 					return (PF_DROP);
6962 				case IPPROTO_AH:
6963 				case IPPROTO_HOPOPTS:
6964 				case IPPROTO_ROUTING:
6965 				case IPPROTO_DSTOPTS: {
6966 					/* get next header and header length */
6967 					struct ip6_ext opt6;
6968 
6969 					if (!pf_pull_hdr(m, off2, &opt6,
6970 					    sizeof(opt6), NULL, reason,
6971 					    pd2.af)) {
6972 						DPFPRINTF(PF_DEBUG_MISC,
6973 						    ("pf: ICMPv6 short opt\n"));
6974 						return (PF_DROP);
6975 					}
6976 					if (pd2.proto == IPPROTO_AH)
6977 						off2 += (opt6.ip6e_len + 2) * 4;
6978 					else
6979 						off2 += (opt6.ip6e_len + 1) * 8;
6980 					pd2.proto = opt6.ip6e_nxt;
6981 					/* goto the next header */
6982 					break;
6983 				}
6984 				default:
6985 					terminal++;
6986 					break;
6987 				}
6988 			} while (!terminal);
6989 			break;
6990 #endif /* INET6 */
6991 		}
6992 
6993 		if (PF_ANEQ(pd->dst, pd2.src, pd->af)) {
6994 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
6995 				printf("pf: BAD ICMP %d:%d outer dst: ",
6996 				    icmptype, icmpcode);
6997 				pf_print_host(pd->src, 0, pd->af);
6998 				printf(" -> ");
6999 				pf_print_host(pd->dst, 0, pd->af);
7000 				printf(" inner src: ");
7001 				pf_print_host(pd2.src, 0, pd2.af);
7002 				printf(" -> ");
7003 				pf_print_host(pd2.dst, 0, pd2.af);
7004 				printf("\n");
7005 			}
7006 			REASON_SET(reason, PFRES_BADSTATE);
7007 			return (PF_DROP);
7008 		}
7009 
7010 		switch (pd2.proto) {
7011 		case IPPROTO_TCP: {
7012 			struct tcphdr		 th;
7013 			u_int32_t		 seq;
7014 			struct pf_state_peer	*src, *dst;
7015 			u_int8_t		 dws;
7016 			int			 copyback = 0;
7017 
7018 			/*
7019 			 * Only the first 8 bytes of the TCP header can be
7020 			 * expected. Don't access any TCP header fields after
7021 			 * th_seq, an ackskew test is not possible.
7022 			 */
7023 			if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
7024 			    pd2.af)) {
7025 				DPFPRINTF(PF_DEBUG_MISC,
7026 				    ("pf: ICMP error message too short "
7027 				    "(tcp)\n"));
7028 				return (PF_DROP);
7029 			}
7030 
7031 			key.af = pd2.af;
7032 			key.proto = IPPROTO_TCP;
7033 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
7034 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
7035 			key.port[pd2.sidx] = th.th_sport;
7036 			key.port[pd2.didx] = th.th_dport;
7037 
7038 			STATE_LOOKUP(kif, &key, *state, pd);
7039 
7040 			if (pd->dir == (*state)->direction) {
7041 				src = &(*state)->dst;
7042 				dst = &(*state)->src;
7043 			} else {
7044 				src = &(*state)->src;
7045 				dst = &(*state)->dst;
7046 			}
7047 
7048 			if (src->wscale && dst->wscale)
7049 				dws = dst->wscale & PF_WSCALE_MASK;
7050 			else
7051 				dws = 0;
7052 
7053 			/* Demodulate sequence number */
7054 			seq = ntohl(th.th_seq) - src->seqdiff;
7055 			if (src->seqdiff) {
7056 				pf_change_a(&th.th_seq, icmpsum,
7057 				    htonl(seq), 0);
7058 				copyback = 1;
7059 			}
7060 
7061 			if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
7062 			    (!SEQ_GEQ(src->seqhi, seq) ||
7063 			    !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
7064 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
7065 					printf("pf: BAD ICMP %d:%d ",
7066 					    icmptype, icmpcode);
7067 					pf_print_host(pd->src, 0, pd->af);
7068 					printf(" -> ");
7069 					pf_print_host(pd->dst, 0, pd->af);
7070 					printf(" state: ");
7071 					pf_print_state(*state);
7072 					printf(" seq=%u\n", seq);
7073 				}
7074 				REASON_SET(reason, PFRES_BADSTATE);
7075 				return (PF_DROP);
7076 			} else {
7077 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
7078 					printf("pf: OK ICMP %d:%d ",
7079 					    icmptype, icmpcode);
7080 					pf_print_host(pd->src, 0, pd->af);
7081 					printf(" -> ");
7082 					pf_print_host(pd->dst, 0, pd->af);
7083 					printf(" state: ");
7084 					pf_print_state(*state);
7085 					printf(" seq=%u\n", seq);
7086 				}
7087 			}
7088 
7089 			/* translate source/destination address, if necessary */
7090 			if ((*state)->key[PF_SK_WIRE] !=
7091 			    (*state)->key[PF_SK_STACK]) {
7092 				struct pf_state_key *nk =
7093 				    (*state)->key[pd->didx];
7094 
7095 				if (PF_ANEQ(pd2.src,
7096 				    &nk->addr[pd2.sidx], pd2.af) ||
7097 				    nk->port[pd2.sidx] != th.th_sport)
7098 					pf_change_icmp(pd2.src, &th.th_sport,
7099 					    daddr, &nk->addr[pd2.sidx],
7100 					    nk->port[pd2.sidx], NULL,
7101 					    pd2.ip_sum, icmpsum,
7102 					    pd->ip_sum, 0, pd2.af);
7103 
7104 				if (PF_ANEQ(pd2.dst,
7105 				    &nk->addr[pd2.didx], pd2.af) ||
7106 				    nk->port[pd2.didx] != th.th_dport)
7107 					pf_change_icmp(pd2.dst, &th.th_dport,
7108 					    saddr, &nk->addr[pd2.didx],
7109 					    nk->port[pd2.didx], NULL,
7110 					    pd2.ip_sum, icmpsum,
7111 					    pd->ip_sum, 0, pd2.af);
7112 				copyback = 1;
7113 			}
7114 
7115 			if (copyback) {
7116 				switch (pd2.af) {
7117 #ifdef INET
7118 				case AF_INET:
7119 					m_copyback(m, off, ICMP_MINLEN,
7120 					    (caddr_t )&pd->hdr.icmp);
7121 					m_copyback(m, ipoff2, sizeof(h2),
7122 					    (caddr_t )&h2);
7123 					break;
7124 #endif /* INET */
7125 #ifdef INET6
7126 				case AF_INET6:
7127 					m_copyback(m, off,
7128 					    sizeof(struct icmp6_hdr),
7129 					    (caddr_t )&pd->hdr.icmp6);
7130 					m_copyback(m, ipoff2, sizeof(h2_6),
7131 					    (caddr_t )&h2_6);
7132 					break;
7133 #endif /* INET6 */
7134 				}
7135 				m_copyback(m, off2, 8, (caddr_t)&th);
7136 			}
7137 
7138 			return (PF_PASS);
7139 			break;
7140 		}
7141 		case IPPROTO_UDP: {
7142 			struct udphdr		uh;
7143 
7144 			if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
7145 			    NULL, reason, pd2.af)) {
7146 				DPFPRINTF(PF_DEBUG_MISC,
7147 				    ("pf: ICMP error message too short "
7148 				    "(udp)\n"));
7149 				return (PF_DROP);
7150 			}
7151 
7152 			key.af = pd2.af;
7153 			key.proto = IPPROTO_UDP;
7154 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
7155 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
7156 			key.port[pd2.sidx] = uh.uh_sport;
7157 			key.port[pd2.didx] = uh.uh_dport;
7158 
7159 			STATE_LOOKUP(kif, &key, *state, pd);
7160 
7161 			/* translate source/destination address, if necessary */
7162 			if ((*state)->key[PF_SK_WIRE] !=
7163 			    (*state)->key[PF_SK_STACK]) {
7164 				struct pf_state_key *nk =
7165 				    (*state)->key[pd->didx];
7166 
7167 				if (PF_ANEQ(pd2.src,
7168 				    &nk->addr[pd2.sidx], pd2.af) ||
7169 				    nk->port[pd2.sidx] != uh.uh_sport)
7170 					pf_change_icmp(pd2.src, &uh.uh_sport,
7171 					    daddr, &nk->addr[pd2.sidx],
7172 					    nk->port[pd2.sidx], &uh.uh_sum,
7173 					    pd2.ip_sum, icmpsum,
7174 					    pd->ip_sum, 1, pd2.af);
7175 
7176 				if (PF_ANEQ(pd2.dst,
7177 				    &nk->addr[pd2.didx], pd2.af) ||
7178 				    nk->port[pd2.didx] != uh.uh_dport)
7179 					pf_change_icmp(pd2.dst, &uh.uh_dport,
7180 					    saddr, &nk->addr[pd2.didx],
7181 					    nk->port[pd2.didx], &uh.uh_sum,
7182 					    pd2.ip_sum, icmpsum,
7183 					    pd->ip_sum, 1, pd2.af);
7184 
7185 				switch (pd2.af) {
7186 #ifdef INET
7187 				case AF_INET:
7188 					m_copyback(m, off, ICMP_MINLEN,
7189 					    (caddr_t )&pd->hdr.icmp);
7190 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
7191 					break;
7192 #endif /* INET */
7193 #ifdef INET6
7194 				case AF_INET6:
7195 					m_copyback(m, off,
7196 					    sizeof(struct icmp6_hdr),
7197 					    (caddr_t )&pd->hdr.icmp6);
7198 					m_copyback(m, ipoff2, sizeof(h2_6),
7199 					    (caddr_t )&h2_6);
7200 					break;
7201 #endif /* INET6 */
7202 				}
7203 				m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
7204 			}
7205 			return (PF_PASS);
7206 			break;
7207 		}
7208 #ifdef INET
7209 		case IPPROTO_ICMP: {
7210 			struct icmp	*iih = &pd2.hdr.icmp;
7211 
7212 			if (!pf_pull_hdr(m, off2, iih, ICMP_MINLEN,
7213 			    NULL, reason, pd2.af)) {
7214 				DPFPRINTF(PF_DEBUG_MISC,
7215 				    ("pf: ICMP error message too short i"
7216 				    "(icmp)\n"));
7217 				return (PF_DROP);
7218 			}
7219 
7220 			icmpid = iih->icmp_id;
7221 			pf_icmp_mapping(&pd2, iih->icmp_type,
7222 			    &icmp_dir, &multi, &virtual_id, &virtual_type);
7223 
7224 			ret = pf_icmp_state_lookup(&key, &pd2, state, m, off,
7225 			    pd2.dir, kif, virtual_id, virtual_type,
7226 			    icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1);
7227 			if (ret >= 0) {
7228 				MPASS(*state == NULL);
7229 				return (ret);
7230 			}
7231 
7232 			/* translate source/destination address, if necessary */
7233 			if ((*state)->key[PF_SK_WIRE] !=
7234 			    (*state)->key[PF_SK_STACK]) {
7235 				struct pf_state_key *nk =
7236 				    (*state)->key[pd->didx];
7237 
7238 				if (PF_ANEQ(pd2.src,
7239 				    &nk->addr[pd2.sidx], pd2.af) ||
7240 				    (virtual_type == htons(ICMP_ECHO) &&
7241 				    nk->port[iidx] != iih->icmp_id))
7242 					pf_change_icmp(pd2.src,
7243 					    (virtual_type == htons(ICMP_ECHO)) ?
7244 					    &iih->icmp_id : NULL,
7245 					    daddr, &nk->addr[pd2.sidx],
7246 					    (virtual_type == htons(ICMP_ECHO)) ?
7247 					    nk->port[iidx] : 0, NULL,
7248 					    pd2.ip_sum, icmpsum,
7249 					    pd->ip_sum, 0, AF_INET);
7250 
7251 				if (PF_ANEQ(pd2.dst,
7252 				    &nk->addr[pd2.didx], pd2.af))
7253 					pf_change_icmp(pd2.dst, NULL, NULL,
7254 					    &nk->addr[pd2.didx], 0, NULL,
7255 					    pd2.ip_sum, icmpsum, pd->ip_sum, 0,
7256 					    AF_INET);
7257 
7258 				m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
7259 				m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
7260 				m_copyback(m, off2, ICMP_MINLEN, (caddr_t)iih);
7261 			}
7262 			return (PF_PASS);
7263 			break;
7264 		}
7265 #endif /* INET */
7266 #ifdef INET6
7267 		case IPPROTO_ICMPV6: {
7268 			struct icmp6_hdr	*iih = &pd2.hdr.icmp6;
7269 
7270 			if (!pf_pull_hdr(m, off2, iih,
7271 			    sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
7272 				DPFPRINTF(PF_DEBUG_MISC,
7273 				    ("pf: ICMP error message too short "
7274 				    "(icmp6)\n"));
7275 				return (PF_DROP);
7276 			}
7277 
7278 			pf_icmp_mapping(&pd2, iih->icmp6_type,
7279 			    &icmp_dir, &multi, &virtual_id, &virtual_type);
7280 
7281 			ret = pf_icmp_state_lookup(&key, &pd2, state, m, off,
7282 			    pd->dir, kif, virtual_id, virtual_type,
7283 			    icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1);
7284 			if (ret >= 0) {
7285 				MPASS(*state == NULL);
7286 				if (ret == PF_DROP && pd2.af == AF_INET6 &&
7287 				    icmp_dir == PF_OUT) {
7288 					ret = pf_icmp_state_lookup(&key, &pd2,
7289 					    state, m, off, pd->dir, kif,
7290 					    virtual_id, virtual_type,
7291 					    icmp_dir, &iidx, multi, 1);
7292 					if (ret >= 0) {
7293 						MPASS(*state == NULL);
7294 						return (ret);
7295 					}
7296 				} else
7297 					return (ret);
7298 			}
7299 
7300 			/* translate source/destination address, if necessary */
7301 			if ((*state)->key[PF_SK_WIRE] !=
7302 			    (*state)->key[PF_SK_STACK]) {
7303 				struct pf_state_key *nk =
7304 				    (*state)->key[pd->didx];
7305 
7306 				if (PF_ANEQ(pd2.src,
7307 				    &nk->addr[pd2.sidx], pd2.af) ||
7308 				    ((virtual_type == htons(ICMP6_ECHO_REQUEST)) &&
7309 				    nk->port[pd2.sidx] != iih->icmp6_id))
7310 					pf_change_icmp(pd2.src,
7311 					    (virtual_type == htons(ICMP6_ECHO_REQUEST))
7312 					    ? &iih->icmp6_id : NULL,
7313 					    daddr, &nk->addr[pd2.sidx],
7314 					    (virtual_type == htons(ICMP6_ECHO_REQUEST))
7315 					    ? nk->port[iidx] : 0, NULL,
7316 					    pd2.ip_sum, icmpsum,
7317 					    pd->ip_sum, 0, AF_INET6);
7318 
7319 				if (PF_ANEQ(pd2.dst,
7320 				    &nk->addr[pd2.didx], pd2.af))
7321 					pf_change_icmp(pd2.dst, NULL, NULL,
7322 					    &nk->addr[pd2.didx], 0, NULL,
7323 					    pd2.ip_sum, icmpsum,
7324 					    pd->ip_sum, 0, AF_INET6);
7325 
7326 				m_copyback(m, off, sizeof(struct icmp6_hdr),
7327 				    (caddr_t)&pd->hdr.icmp6);
7328 				m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
7329 				m_copyback(m, off2, sizeof(struct icmp6_hdr),
7330 				    (caddr_t)iih);
7331 			}
7332 			return (PF_PASS);
7333 			break;
7334 		}
7335 #endif /* INET6 */
7336 		default: {
7337 			key.af = pd2.af;
7338 			key.proto = pd2.proto;
7339 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
7340 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
7341 			key.port[0] = key.port[1] = 0;
7342 
7343 			STATE_LOOKUP(kif, &key, *state, pd);
7344 
7345 			/* translate source/destination address, if necessary */
7346 			if ((*state)->key[PF_SK_WIRE] !=
7347 			    (*state)->key[PF_SK_STACK]) {
7348 				struct pf_state_key *nk =
7349 				    (*state)->key[pd->didx];
7350 
7351 				if (PF_ANEQ(pd2.src,
7352 				    &nk->addr[pd2.sidx], pd2.af))
7353 					pf_change_icmp(pd2.src, NULL, daddr,
7354 					    &nk->addr[pd2.sidx], 0, NULL,
7355 					    pd2.ip_sum, icmpsum,
7356 					    pd->ip_sum, 0, pd2.af);
7357 
7358 				if (PF_ANEQ(pd2.dst,
7359 				    &nk->addr[pd2.didx], pd2.af))
7360 					pf_change_icmp(pd2.dst, NULL, saddr,
7361 					    &nk->addr[pd2.didx], 0, NULL,
7362 					    pd2.ip_sum, icmpsum,
7363 					    pd->ip_sum, 0, pd2.af);
7364 
7365 				switch (pd2.af) {
7366 #ifdef INET
7367 				case AF_INET:
7368 					m_copyback(m, off, ICMP_MINLEN,
7369 					    (caddr_t)&pd->hdr.icmp);
7370 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
7371 					break;
7372 #endif /* INET */
7373 #ifdef INET6
7374 				case AF_INET6:
7375 					m_copyback(m, off,
7376 					    sizeof(struct icmp6_hdr),
7377 					    (caddr_t )&pd->hdr.icmp6);
7378 					m_copyback(m, ipoff2, sizeof(h2_6),
7379 					    (caddr_t )&h2_6);
7380 					break;
7381 #endif /* INET6 */
7382 				}
7383 			}
7384 			return (PF_PASS);
7385 			break;
7386 		}
7387 		}
7388 	}
7389 }
7390 
7391 static int
7392 pf_test_state_other(struct pf_kstate **state, struct pfi_kkif *kif,
7393     struct mbuf *m, struct pf_pdesc *pd)
7394 {
7395 	struct pf_state_peer	*src, *dst;
7396 	struct pf_state_key_cmp	 key;
7397 	uint8_t			 psrc, pdst;
7398 
7399 	bzero(&key, sizeof(key));
7400 	key.af = pd->af;
7401 	key.proto = pd->proto;
7402 	if (pd->dir == PF_IN)	{
7403 		PF_ACPY(&key.addr[0], pd->src, key.af);
7404 		PF_ACPY(&key.addr[1], pd->dst, key.af);
7405 		key.port[0] = key.port[1] = 0;
7406 	} else {
7407 		PF_ACPY(&key.addr[1], pd->src, key.af);
7408 		PF_ACPY(&key.addr[0], pd->dst, key.af);
7409 		key.port[1] = key.port[0] = 0;
7410 	}
7411 
7412 	STATE_LOOKUP(kif, &key, *state, pd);
7413 
7414 	if (pd->dir == (*state)->direction) {
7415 		src = &(*state)->src;
7416 		dst = &(*state)->dst;
7417 		psrc = PF_PEER_SRC;
7418 		pdst = PF_PEER_DST;
7419 	} else {
7420 		src = &(*state)->dst;
7421 		dst = &(*state)->src;
7422 		psrc = PF_PEER_DST;
7423 		pdst = PF_PEER_SRC;
7424 	}
7425 
7426 	/* update states */
7427 	if (src->state < PFOTHERS_SINGLE)
7428 		pf_set_protostate(*state, psrc, PFOTHERS_SINGLE);
7429 	if (dst->state == PFOTHERS_SINGLE)
7430 		pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE);
7431 
7432 	/* update expire time */
7433 	(*state)->expire = pf_get_uptime();
7434 	if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
7435 		(*state)->timeout = PFTM_OTHER_MULTIPLE;
7436 	else
7437 		(*state)->timeout = PFTM_OTHER_SINGLE;
7438 
7439 	/* translate source/destination address, if necessary */
7440 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
7441 		struct pf_state_key *nk = (*state)->key[pd->didx];
7442 
7443 		KASSERT(nk, ("%s: nk is null", __func__));
7444 		KASSERT(pd, ("%s: pd is null", __func__));
7445 		KASSERT(pd->src, ("%s: pd->src is null", __func__));
7446 		KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
7447 		switch (pd->af) {
7448 #ifdef INET
7449 		case AF_INET:
7450 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
7451 				pf_change_a(&pd->src->v4.s_addr,
7452 				    pd->ip_sum,
7453 				    nk->addr[pd->sidx].v4.s_addr,
7454 				    0);
7455 
7456 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
7457 				pf_change_a(&pd->dst->v4.s_addr,
7458 				    pd->ip_sum,
7459 				    nk->addr[pd->didx].v4.s_addr,
7460 				    0);
7461 
7462 			break;
7463 #endif /* INET */
7464 #ifdef INET6
7465 		case AF_INET6:
7466 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
7467 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
7468 
7469 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
7470 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
7471 #endif /* INET6 */
7472 		}
7473 	}
7474 	return (PF_PASS);
7475 }
7476 
7477 /*
7478  * ipoff and off are measured from the start of the mbuf chain.
7479  * h must be at "ipoff" on the mbuf chain.
7480  */
7481 void *
7482 pf_pull_hdr(const struct mbuf *m, int off, void *p, int len,
7483     u_short *actionp, u_short *reasonp, sa_family_t af)
7484 {
7485 	switch (af) {
7486 #ifdef INET
7487 	case AF_INET: {
7488 		const struct ip	*h = mtod(m, struct ip *);
7489 		u_int16_t	 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
7490 
7491 		if (fragoff) {
7492 			if (fragoff >= len)
7493 				ACTION_SET(actionp, PF_PASS);
7494 			else {
7495 				ACTION_SET(actionp, PF_DROP);
7496 				REASON_SET(reasonp, PFRES_FRAG);
7497 			}
7498 			return (NULL);
7499 		}
7500 		if (m->m_pkthdr.len < off + len ||
7501 		    ntohs(h->ip_len) < off + len) {
7502 			ACTION_SET(actionp, PF_DROP);
7503 			REASON_SET(reasonp, PFRES_SHORT);
7504 			return (NULL);
7505 		}
7506 		break;
7507 	}
7508 #endif /* INET */
7509 #ifdef INET6
7510 	case AF_INET6: {
7511 		const struct ip6_hdr	*h = mtod(m, struct ip6_hdr *);
7512 
7513 		if (m->m_pkthdr.len < off + len ||
7514 		    (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
7515 		    (unsigned)(off + len)) {
7516 			ACTION_SET(actionp, PF_DROP);
7517 			REASON_SET(reasonp, PFRES_SHORT);
7518 			return (NULL);
7519 		}
7520 		break;
7521 	}
7522 #endif /* INET6 */
7523 	}
7524 	m_copydata(m, off, len, p);
7525 	return (p);
7526 }
7527 
7528 int
7529 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif,
7530     int rtableid)
7531 {
7532 	struct ifnet		*ifp;
7533 
7534 	/*
7535 	 * Skip check for addresses with embedded interface scope,
7536 	 * as they would always match anyway.
7537 	 */
7538 	if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
7539 		return (1);
7540 
7541 	if (af != AF_INET && af != AF_INET6)
7542 		return (0);
7543 
7544 	if (kif == V_pfi_all)
7545 		return (1);
7546 
7547 	/* Skip checks for ipsec interfaces */
7548 	if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
7549 		return (1);
7550 
7551 	ifp = (kif != NULL) ? kif->pfik_ifp : NULL;
7552 
7553 	switch (af) {
7554 #ifdef INET6
7555 	case AF_INET6:
7556 		return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE,
7557 		    ifp));
7558 #endif
7559 #ifdef INET
7560 	case AF_INET:
7561 		return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE,
7562 		    ifp));
7563 #endif
7564 	}
7565 
7566 	return (0);
7567 }
7568 
7569 #ifdef INET
7570 static void
7571 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp,
7572     struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp)
7573 {
7574 	struct mbuf		*m0, *m1, *md;
7575 	struct sockaddr_in	dst;
7576 	struct ip		*ip;
7577 	struct pfi_kkif		*nkif = NULL;
7578 	struct ifnet		*ifp = NULL;
7579 	struct pf_addr		 naddr;
7580 	struct pf_ksrc_node	*sn = NULL;
7581 	int			 error = 0;
7582 	uint16_t		 ip_len, ip_off;
7583 	uint16_t		 tmp;
7584 	int			 r_rt, r_dir;
7585 
7586 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
7587 
7588 	if (s) {
7589 		r_rt = s->rt;
7590 		r_dir = s->direction;
7591 	} else {
7592 		r_rt = r->rt;
7593 		r_dir = r->direction;
7594 	}
7595 
7596 	KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT ||
7597 	    r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction",
7598 	    __func__));
7599 
7600 	if ((pd->pf_mtag == NULL &&
7601 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
7602 	    pd->pf_mtag->routed++ > 3) {
7603 		m0 = *m;
7604 		*m = NULL;
7605 		goto bad_locked;
7606 	}
7607 
7608 	if (r_rt == PF_DUPTO) {
7609 		if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) {
7610 			if (s == NULL) {
7611 				ifp = r->rpool.cur->kif ?
7612 				    r->rpool.cur->kif->pfik_ifp : NULL;
7613 			} else {
7614 				ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7615 				/* If pfsync'd */
7616 				if (ifp == NULL && r->rpool.cur != NULL)
7617 					ifp = r->rpool.cur->kif ?
7618 					    r->rpool.cur->kif->pfik_ifp : NULL;
7619 				PF_STATE_UNLOCK(s);
7620 			}
7621 			if (ifp == oifp) {
7622 				/* When the 2nd interface is not skipped */
7623 				return;
7624 			} else {
7625 				m0 = *m;
7626 				*m = NULL;
7627 				goto bad;
7628 			}
7629 		} else {
7630 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED;
7631 			if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) {
7632 				if (s)
7633 					PF_STATE_UNLOCK(s);
7634 				return;
7635 			}
7636 		}
7637 	} else {
7638 		if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) {
7639 			pf_dummynet(pd, s, r, m);
7640 			if (s)
7641 				PF_STATE_UNLOCK(s);
7642 			return;
7643 		}
7644 		m0 = *m;
7645 	}
7646 
7647 	ip = mtod(m0, struct ip *);
7648 
7649 	bzero(&dst, sizeof(dst));
7650 	dst.sin_family = AF_INET;
7651 	dst.sin_len = sizeof(dst);
7652 	dst.sin_addr = ip->ip_dst;
7653 
7654 	bzero(&naddr, sizeof(naddr));
7655 
7656 	if (s == NULL) {
7657 		if (TAILQ_EMPTY(&r->rpool.list)) {
7658 			DPFPRINTF(PF_DEBUG_URGENT,
7659 			    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
7660 			goto bad_locked;
7661 		}
7662 		pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
7663 		    &naddr, &nkif, NULL, &sn);
7664 		if (!PF_AZERO(&naddr, AF_INET))
7665 			dst.sin_addr.s_addr = naddr.v4.s_addr;
7666 		ifp = nkif ? nkif->pfik_ifp : NULL;
7667 	} else {
7668 		struct pfi_kkif *kif;
7669 
7670 		if (!PF_AZERO(&s->rt_addr, AF_INET))
7671 			dst.sin_addr.s_addr =
7672 			    s->rt_addr.v4.s_addr;
7673 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7674 		kif = s->rt_kif;
7675 		/* If pfsync'd */
7676 		if (ifp == NULL && r->rpool.cur != NULL) {
7677 			ifp = r->rpool.cur->kif ?
7678 			    r->rpool.cur->kif->pfik_ifp : NULL;
7679 			kif = r->rpool.cur->kif;
7680 		}
7681 		if (ifp != NULL && kif != NULL &&
7682 		    r->rule_flag & PFRULE_IFBOUND &&
7683 		    r->rt == PF_REPLYTO &&
7684 		    s->kif == V_pfi_all) {
7685 			s->kif = kif;
7686 			s->orig_kif = oifp->if_pf_kif;
7687 		}
7688 
7689 		PF_STATE_UNLOCK(s);
7690 	}
7691 
7692 	if (ifp == NULL)
7693 		goto bad;
7694 
7695 	if (pd->dir == PF_IN) {
7696 		if (pf_test(PF_OUT, PFIL_FWD, ifp, &m0, inp, &pd->act) != PF_PASS)
7697 			goto bad;
7698 		else if (m0 == NULL)
7699 			goto done;
7700 		if (m0->m_len < sizeof(struct ip)) {
7701 			DPFPRINTF(PF_DEBUG_URGENT,
7702 			    ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
7703 			goto bad;
7704 		}
7705 		ip = mtod(m0, struct ip *);
7706 	}
7707 
7708 	if (ifp->if_flags & IFF_LOOPBACK)
7709 		m0->m_flags |= M_SKIP_FIREWALL;
7710 
7711 	ip_len = ntohs(ip->ip_len);
7712 	ip_off = ntohs(ip->ip_off);
7713 
7714 	/* Copied from FreeBSD 10.0-CURRENT ip_output. */
7715 	m0->m_pkthdr.csum_flags |= CSUM_IP;
7716 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
7717 		in_delayed_cksum(m0);
7718 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
7719 	}
7720 	if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
7721 		pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2));
7722 		m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
7723 	}
7724 
7725 	if (pd->dir == PF_IN) {
7726 		/*
7727 		 * Make sure dummynet gets the correct direction, in case it needs to
7728 		 * re-inject later.
7729 		 */
7730 		pd->dir = PF_OUT;
7731 
7732 		/*
7733 		 * The following processing is actually the rest of the inbound processing, even
7734 		 * though we've marked it as outbound (so we don't look through dummynet) and it
7735 		 * happens after the outbound processing (pf_test(PF_OUT) above).
7736 		 * Swap the dummynet pipe numbers, because it's going to come to the wrong
7737 		 * conclusion about what direction it's processing, and we can't fix it or it
7738 		 * will re-inject incorrectly. Swapping the pipe numbers means that its incorrect
7739 		 * decision will pick the right pipe, and everything will mostly work as expected.
7740 		 */
7741 		tmp = pd->act.dnrpipe;
7742 		pd->act.dnrpipe = pd->act.dnpipe;
7743 		pd->act.dnpipe = tmp;
7744 	}
7745 
7746 	/*
7747 	 * If small enough for interface, or the interface will take
7748 	 * care of the fragmentation for us, we can just send directly.
7749 	 */
7750 	if (ip_len <= ifp->if_mtu ||
7751 	    (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
7752 		ip->ip_sum = 0;
7753 		if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
7754 			ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
7755 			m0->m_pkthdr.csum_flags &= ~CSUM_IP;
7756 		}
7757 		m_clrprotoflags(m0);	/* Avoid confusing lower layers. */
7758 
7759 		md = m0;
7760 		error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md);
7761 		if (md != NULL)
7762 			error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL);
7763 		goto done;
7764 	}
7765 
7766 	/* Balk when DF bit is set or the interface didn't support TSO. */
7767 	if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
7768 		error = EMSGSIZE;
7769 		KMOD_IPSTAT_INC(ips_cantfrag);
7770 		if (r_rt != PF_DUPTO) {
7771 			if (s && pd->nat_rule != NULL)
7772 				PACKET_UNDO_NAT(m0, pd,
7773 				    (ip->ip_hl << 2) + (ip_off & IP_OFFMASK),
7774 				    s);
7775 
7776 			icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
7777 			    ifp->if_mtu);
7778 			goto done;
7779 		} else
7780 			goto bad;
7781 	}
7782 
7783 	error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
7784 	if (error)
7785 		goto bad;
7786 
7787 	for (; m0; m0 = m1) {
7788 		m1 = m0->m_nextpkt;
7789 		m0->m_nextpkt = NULL;
7790 		if (error == 0) {
7791 			m_clrprotoflags(m0);
7792 			md = m0;
7793 			pd->pf_mtag = pf_find_mtag(md);
7794 			error = pf_dummynet_route(pd, s, r, ifp,
7795 			    sintosa(&dst), &md);
7796 			if (md != NULL)
7797 				error = (*ifp->if_output)(ifp, md,
7798 				    sintosa(&dst), NULL);
7799 		} else
7800 			m_freem(m0);
7801 	}
7802 
7803 	if (error == 0)
7804 		KMOD_IPSTAT_INC(ips_fragmented);
7805 
7806 done:
7807 	if (r_rt != PF_DUPTO)
7808 		*m = NULL;
7809 	return;
7810 
7811 bad_locked:
7812 	if (s)
7813 		PF_STATE_UNLOCK(s);
7814 bad:
7815 	m_freem(m0);
7816 	goto done;
7817 }
7818 #endif /* INET */
7819 
7820 #ifdef INET6
7821 static void
7822 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp,
7823     struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp)
7824 {
7825 	struct mbuf		*m0, *md;
7826 	struct sockaddr_in6	dst;
7827 	struct ip6_hdr		*ip6;
7828 	struct pfi_kkif		*nkif = NULL;
7829 	struct ifnet		*ifp = NULL;
7830 	struct pf_addr		 naddr;
7831 	struct pf_ksrc_node	*sn = NULL;
7832 	int			 r_rt, r_dir;
7833 
7834 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
7835 
7836 	if (s) {
7837 		r_rt = s->rt;
7838 		r_dir = s->direction;
7839 	} else {
7840 		r_rt = r->rt;
7841 		r_dir = r->direction;
7842 	}
7843 
7844 	KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT ||
7845 	    r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction",
7846 	    __func__));
7847 
7848 	if ((pd->pf_mtag == NULL &&
7849 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
7850 	    pd->pf_mtag->routed++ > 3) {
7851 		m0 = *m;
7852 		*m = NULL;
7853 		goto bad_locked;
7854 	}
7855 
7856 	if (r_rt == PF_DUPTO) {
7857 		if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) {
7858 			if (s == NULL) {
7859 				ifp = r->rpool.cur->kif ?
7860 				    r->rpool.cur->kif->pfik_ifp : NULL;
7861 			} else {
7862 				ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7863 				/* If pfsync'd */
7864 				if (ifp == NULL && r->rpool.cur != NULL)
7865 					ifp = r->rpool.cur->kif ?
7866 					    r->rpool.cur->kif->pfik_ifp : NULL;
7867 				PF_STATE_UNLOCK(s);
7868 			}
7869 			if (ifp == oifp) {
7870 				/* When the 2nd interface is not skipped */
7871 				return;
7872 			} else {
7873 				m0 = *m;
7874 				*m = NULL;
7875 				goto bad;
7876 			}
7877 		} else {
7878 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED;
7879 			if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) {
7880 				if (s)
7881 					PF_STATE_UNLOCK(s);
7882 				return;
7883 			}
7884 		}
7885 	} else {
7886 		if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) {
7887 			pf_dummynet(pd, s, r, m);
7888 			if (s)
7889 				PF_STATE_UNLOCK(s);
7890 			return;
7891 		}
7892 		m0 = *m;
7893 	}
7894 
7895 	ip6 = mtod(m0, struct ip6_hdr *);
7896 
7897 	bzero(&dst, sizeof(dst));
7898 	dst.sin6_family = AF_INET6;
7899 	dst.sin6_len = sizeof(dst);
7900 	dst.sin6_addr = ip6->ip6_dst;
7901 
7902 	bzero(&naddr, sizeof(naddr));
7903 
7904 	if (s == NULL) {
7905 		if (TAILQ_EMPTY(&r->rpool.list)) {
7906 			DPFPRINTF(PF_DEBUG_URGENT,
7907 			    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
7908 			goto bad_locked;
7909 		}
7910 		pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
7911 		    &naddr, &nkif, NULL, &sn);
7912 		if (!PF_AZERO(&naddr, AF_INET6))
7913 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
7914 			    &naddr, AF_INET6);
7915 		ifp = nkif ? nkif->pfik_ifp : NULL;
7916 	} else {
7917 		struct pfi_kkif *kif;
7918 
7919 		if (!PF_AZERO(&s->rt_addr, AF_INET6))
7920 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
7921 			    &s->rt_addr, AF_INET6);
7922 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7923 		kif = s->rt_kif;
7924 		/* If pfsync'd */
7925 		if (ifp == NULL && r->rpool.cur != NULL) {
7926 			ifp = r->rpool.cur->kif ?
7927 			    r->rpool.cur->kif->pfik_ifp : NULL;
7928 			kif = r->rpool.cur->kif;
7929 		}
7930 		if (ifp != NULL && kif != NULL &&
7931 		    r->rule_flag & PFRULE_IFBOUND &&
7932 		    r->rt == PF_REPLYTO &&
7933 		    s->kif == V_pfi_all) {
7934 			s->kif = kif;
7935 			s->orig_kif = oifp->if_pf_kif;
7936 		}
7937 	}
7938 
7939 	if (s)
7940 		PF_STATE_UNLOCK(s);
7941 
7942 	if (ifp == NULL)
7943 		goto bad;
7944 
7945 	if (pd->dir == PF_IN) {
7946 		if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp, &pd->act) != PF_PASS)
7947 			goto bad;
7948 		else if (m0 == NULL)
7949 			goto done;
7950 		if (m0->m_len < sizeof(struct ip6_hdr)) {
7951 			DPFPRINTF(PF_DEBUG_URGENT,
7952 			    ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
7953 			    __func__));
7954 			goto bad;
7955 		}
7956 		ip6 = mtod(m0, struct ip6_hdr *);
7957 	}
7958 
7959 	if (ifp->if_flags & IFF_LOOPBACK)
7960 		m0->m_flags |= M_SKIP_FIREWALL;
7961 
7962 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
7963 	    ~ifp->if_hwassist) {
7964 		uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
7965 		in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
7966 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
7967 	}
7968 
7969 	/*
7970 	 * If the packet is too large for the outgoing interface,
7971 	 * send back an icmp6 error.
7972 	 */
7973 	if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
7974 		dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
7975 	if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) {
7976 		md = m0;
7977 		pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md);
7978 		if (md != NULL)
7979 			nd6_output_ifp(ifp, ifp, md, &dst, NULL);
7980 	}
7981 	else {
7982 		in6_ifstat_inc(ifp, ifs6_in_toobig);
7983 		if (r_rt != PF_DUPTO) {
7984 			if (s && pd->nat_rule != NULL)
7985 				PACKET_UNDO_NAT(m0, pd,
7986 				    ((caddr_t)ip6 - m0->m_data) +
7987 				    sizeof(struct ip6_hdr), s);
7988 
7989 			icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
7990 		} else
7991 			goto bad;
7992 	}
7993 
7994 done:
7995 	if (r_rt != PF_DUPTO)
7996 		*m = NULL;
7997 	return;
7998 
7999 bad_locked:
8000 	if (s)
8001 		PF_STATE_UNLOCK(s);
8002 bad:
8003 	m_freem(m0);
8004 	goto done;
8005 }
8006 #endif /* INET6 */
8007 
8008 /*
8009  * FreeBSD supports cksum offloads for the following drivers.
8010  *  em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4)
8011  *
8012  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
8013  *  network driver performed cksum including pseudo header, need to verify
8014  *   csum_data
8015  * CSUM_DATA_VALID :
8016  *  network driver performed cksum, needs to additional pseudo header
8017  *  cksum computation with partial csum_data(i.e. lack of H/W support for
8018  *  pseudo header, for instance sk(4) and possibly gem(4))
8019  *
8020  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
8021  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
8022  * TCP/UDP layer.
8023  * Also, set csum_data to 0xffff to force cksum validation.
8024  */
8025 static int
8026 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
8027 {
8028 	u_int16_t sum = 0;
8029 	int hw_assist = 0;
8030 	struct ip *ip;
8031 
8032 	if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
8033 		return (1);
8034 	if (m->m_pkthdr.len < off + len)
8035 		return (1);
8036 
8037 	switch (p) {
8038 	case IPPROTO_TCP:
8039 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
8040 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
8041 				sum = m->m_pkthdr.csum_data;
8042 			} else {
8043 				ip = mtod(m, struct ip *);
8044 				sum = in_pseudo(ip->ip_src.s_addr,
8045 				ip->ip_dst.s_addr, htonl((u_short)len +
8046 				m->m_pkthdr.csum_data + IPPROTO_TCP));
8047 			}
8048 			sum ^= 0xffff;
8049 			++hw_assist;
8050 		}
8051 		break;
8052 	case IPPROTO_UDP:
8053 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
8054 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
8055 				sum = m->m_pkthdr.csum_data;
8056 			} else {
8057 				ip = mtod(m, struct ip *);
8058 				sum = in_pseudo(ip->ip_src.s_addr,
8059 				ip->ip_dst.s_addr, htonl((u_short)len +
8060 				m->m_pkthdr.csum_data + IPPROTO_UDP));
8061 			}
8062 			sum ^= 0xffff;
8063 			++hw_assist;
8064 		}
8065 		break;
8066 	case IPPROTO_ICMP:
8067 #ifdef INET6
8068 	case IPPROTO_ICMPV6:
8069 #endif /* INET6 */
8070 		break;
8071 	default:
8072 		return (1);
8073 	}
8074 
8075 	if (!hw_assist) {
8076 		switch (af) {
8077 		case AF_INET:
8078 			if (p == IPPROTO_ICMP) {
8079 				if (m->m_len < off)
8080 					return (1);
8081 				m->m_data += off;
8082 				m->m_len -= off;
8083 				sum = in_cksum(m, len);
8084 				m->m_data -= off;
8085 				m->m_len += off;
8086 			} else {
8087 				if (m->m_len < sizeof(struct ip))
8088 					return (1);
8089 				sum = in4_cksum(m, p, off, len);
8090 			}
8091 			break;
8092 #ifdef INET6
8093 		case AF_INET6:
8094 			if (m->m_len < sizeof(struct ip6_hdr))
8095 				return (1);
8096 			sum = in6_cksum(m, p, off, len);
8097 			break;
8098 #endif /* INET6 */
8099 		default:
8100 			return (1);
8101 		}
8102 	}
8103 	if (sum) {
8104 		switch (p) {
8105 		case IPPROTO_TCP:
8106 		    {
8107 			KMOD_TCPSTAT_INC(tcps_rcvbadsum);
8108 			break;
8109 		    }
8110 		case IPPROTO_UDP:
8111 		    {
8112 			KMOD_UDPSTAT_INC(udps_badsum);
8113 			break;
8114 		    }
8115 #ifdef INET
8116 		case IPPROTO_ICMP:
8117 		    {
8118 			KMOD_ICMPSTAT_INC(icps_checksum);
8119 			break;
8120 		    }
8121 #endif
8122 #ifdef INET6
8123 		case IPPROTO_ICMPV6:
8124 		    {
8125 			KMOD_ICMP6STAT_INC(icp6s_checksum);
8126 			break;
8127 		    }
8128 #endif /* INET6 */
8129 		}
8130 		return (1);
8131 	} else {
8132 		if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
8133 			m->m_pkthdr.csum_flags |=
8134 			    (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
8135 			m->m_pkthdr.csum_data = 0xffff;
8136 		}
8137 	}
8138 	return (0);
8139 }
8140 
8141 static bool
8142 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r,
8143     const struct pf_kstate *s, struct ip_fw_args *dnflow)
8144 {
8145 	int dndir = r->direction;
8146 
8147 	if (s && dndir == PF_INOUT) {
8148 		dndir = s->direction;
8149 	} else if (dndir == PF_INOUT) {
8150 		/* Assume primary direction. Happens when we've set dnpipe in
8151 		 * the ethernet level code. */
8152 		dndir = pd->dir;
8153 	}
8154 
8155 	if (pd->pf_mtag->flags & PF_MTAG_FLAG_DUMMYNETED)
8156 		return (false);
8157 
8158 	memset(dnflow, 0, sizeof(*dnflow));
8159 
8160 	if (pd->dport != NULL)
8161 		dnflow->f_id.dst_port = ntohs(*pd->dport);
8162 	if (pd->sport != NULL)
8163 		dnflow->f_id.src_port = ntohs(*pd->sport);
8164 
8165 	if (pd->dir == PF_IN)
8166 		dnflow->flags |= IPFW_ARGS_IN;
8167 	else
8168 		dnflow->flags |= IPFW_ARGS_OUT;
8169 
8170 	if (pd->dir != dndir && pd->act.dnrpipe) {
8171 		dnflow->rule.info = pd->act.dnrpipe;
8172 	}
8173 	else if (pd->dir == dndir && pd->act.dnpipe) {
8174 		dnflow->rule.info = pd->act.dnpipe;
8175 	}
8176 	else {
8177 		return (false);
8178 	}
8179 
8180 	dnflow->rule.info |= IPFW_IS_DUMMYNET;
8181 	if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE)
8182 		dnflow->rule.info |= IPFW_IS_PIPE;
8183 
8184 	dnflow->f_id.proto = pd->proto;
8185 	dnflow->f_id.extra = dnflow->rule.info;
8186 	switch (pd->af) {
8187 	case AF_INET:
8188 		dnflow->f_id.addr_type = 4;
8189 		dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr);
8190 		dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr);
8191 		break;
8192 	case AF_INET6:
8193 		dnflow->flags |= IPFW_ARGS_IP6;
8194 		dnflow->f_id.addr_type = 6;
8195 		dnflow->f_id.src_ip6 = pd->src->v6;
8196 		dnflow->f_id.dst_ip6 = pd->dst->v6;
8197 		break;
8198 	default:
8199 		panic("Invalid AF");
8200 		break;
8201 	}
8202 
8203 	return (true);
8204 }
8205 
8206 int
8207 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0,
8208     struct inpcb *inp)
8209 {
8210 	struct pfi_kkif		*kif;
8211 	struct mbuf		*m = *m0;
8212 
8213 	M_ASSERTPKTHDR(m);
8214 	MPASS(ifp->if_vnet == curvnet);
8215 	NET_EPOCH_ASSERT();
8216 
8217 	if (!V_pf_status.running)
8218 		return (PF_PASS);
8219 
8220 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
8221 
8222 	if (kif == NULL) {
8223 		DPFPRINTF(PF_DEBUG_URGENT,
8224 		    ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname));
8225 		return (PF_DROP);
8226 	}
8227 	if (kif->pfik_flags & PFI_IFLAG_SKIP)
8228 		return (PF_PASS);
8229 
8230 	if (m->m_flags & M_SKIP_FIREWALL)
8231 		return (PF_PASS);
8232 
8233 	/* Stateless! */
8234 	return (pf_test_eth_rule(dir, kif, m0));
8235 }
8236 
8237 static __inline void
8238 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag)
8239 {
8240 	struct m_tag *mtag;
8241 
8242 	pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET;
8243 
8244 	/* dummynet adds this tag, but pf does not need it,
8245 	 * and keeping it creates unexpected behavior,
8246 	 * e.g. in case of divert(4) usage right after dummynet. */
8247 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
8248 	if (mtag != NULL)
8249 		m_tag_delete(m, mtag);
8250 }
8251 
8252 static int
8253 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s,
8254     struct pf_krule *r, struct mbuf **m0)
8255 {
8256 	return (pf_dummynet_route(pd, s, r, NULL, NULL, m0));
8257 }
8258 
8259 static int
8260 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s,
8261     struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa,
8262     struct mbuf **m0)
8263 {
8264 	NET_EPOCH_ASSERT();
8265 
8266 	if (pd->act.dnpipe || pd->act.dnrpipe) {
8267 		struct ip_fw_args dnflow;
8268 		if (ip_dn_io_ptr == NULL) {
8269 			m_freem(*m0);
8270 			*m0 = NULL;
8271 			return (ENOMEM);
8272 		}
8273 
8274 		if (pd->pf_mtag == NULL &&
8275 		    ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) {
8276 			m_freem(*m0);
8277 			*m0 = NULL;
8278 			return (ENOMEM);
8279 		}
8280 
8281 		if (ifp != NULL) {
8282 			pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO;
8283 
8284 			pd->pf_mtag->if_index = ifp->if_index;
8285 			pd->pf_mtag->if_idxgen = ifp->if_idxgen;
8286 
8287 			MPASS(sa != NULL);
8288 
8289 			if (pd->af == AF_INET)
8290 				memcpy(&pd->pf_mtag->dst, sa,
8291 				    sizeof(struct sockaddr_in));
8292 			else
8293 				memcpy(&pd->pf_mtag->dst, sa,
8294 				    sizeof(struct sockaddr_in6));
8295 		}
8296 
8297 		if (s != NULL && s->nat_rule.ptr != NULL &&
8298 		    s->nat_rule.ptr->action == PF_RDR &&
8299 		    (
8300 #ifdef INET
8301 		    (pd->af == AF_INET && IN_LOOPBACK(ntohl(pd->dst->v4.s_addr))) ||
8302 #endif
8303 		    (pd->af == AF_INET6 && IN6_IS_ADDR_LOOPBACK(&pd->dst->v6)))) {
8304 			/*
8305 			 * If we're redirecting to loopback mark this packet
8306 			 * as being local. Otherwise it might get dropped
8307 			 * if dummynet re-injects.
8308 			 */
8309 			(*m0)->m_pkthdr.rcvif = V_loif;
8310 		}
8311 
8312 		if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) {
8313 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET;
8314 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNETED;
8315 			ip_dn_io_ptr(m0, &dnflow);
8316 			if (*m0 != NULL) {
8317 				pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
8318 				pf_dummynet_flag_remove(*m0, pd->pf_mtag);
8319 			}
8320 		}
8321 	}
8322 
8323 	return (0);
8324 }
8325 
8326 #ifdef INET
8327 int
8328 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0,
8329     struct inpcb *inp, struct pf_rule_actions *default_actions)
8330 {
8331 	struct pfi_kkif		*kif;
8332 	u_short			 action, reason = 0;
8333 	struct mbuf		*m = *m0;
8334 	struct ip		*h = NULL;
8335 	struct m_tag		*mtag;
8336 	struct pf_krule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
8337 	struct pf_kstate	*s = NULL;
8338 	struct pf_kruleset	*ruleset = NULL;
8339 	struct pf_pdesc		 pd;
8340 	int			 off, dirndx, use_2nd_queue = 0;
8341 	uint16_t		 tag;
8342 	uint8_t			 rt;
8343 
8344 	PF_RULES_RLOCK_TRACKER;
8345 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir));
8346 	M_ASSERTPKTHDR(m);
8347 
8348 	if (!V_pf_status.running)
8349 		return (PF_PASS);
8350 
8351 	PF_RULES_RLOCK();
8352 
8353 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
8354 
8355 	if (__predict_false(kif == NULL)) {
8356 		DPFPRINTF(PF_DEBUG_URGENT,
8357 		    ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
8358 		PF_RULES_RUNLOCK();
8359 		return (PF_DROP);
8360 	}
8361 	if (kif->pfik_flags & PFI_IFLAG_SKIP) {
8362 		PF_RULES_RUNLOCK();
8363 		return (PF_PASS);
8364 	}
8365 
8366 	if (m->m_flags & M_SKIP_FIREWALL) {
8367 		PF_RULES_RUNLOCK();
8368 		return (PF_PASS);
8369 	}
8370 
8371 	memset(&pd, 0, sizeof(pd));
8372 	TAILQ_INIT(&pd.sctp_multihome_jobs);
8373 	if (default_actions != NULL)
8374 		memcpy(&pd.act, default_actions, sizeof(pd.act));
8375 	pd.pf_mtag = pf_find_mtag(m);
8376 
8377 	if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) {
8378 		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
8379 
8380 		ifp = ifnet_byindexgen(pd.pf_mtag->if_index,
8381 		    pd.pf_mtag->if_idxgen);
8382 		if (ifp == NULL || ifp->if_flags & IFF_DYING) {
8383 			PF_RULES_RUNLOCK();
8384 			m_freem(*m0);
8385 			*m0 = NULL;
8386 			return (PF_PASS);
8387 		}
8388 		PF_RULES_RUNLOCK();
8389 		(ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL);
8390 		*m0 = NULL;
8391 		return (PF_PASS);
8392 	}
8393 
8394 	if (pd.pf_mtag && pd.pf_mtag->dnpipe) {
8395 		pd.act.dnpipe = pd.pf_mtag->dnpipe;
8396 		pd.act.flags = pd.pf_mtag->dnflags;
8397 	}
8398 
8399 	if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL &&
8400 	    pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
8401 		/* Dummynet re-injects packets after they've
8402 		 * completed their delay. We've already
8403 		 * processed them, so pass unconditionally. */
8404 
8405 		/* But only once. We may see the packet multiple times (e.g.
8406 		 * PFIL_IN/PFIL_OUT). */
8407 		pf_dummynet_flag_remove(m, pd.pf_mtag);
8408 		PF_RULES_RUNLOCK();
8409 
8410 		return (PF_PASS);
8411 	}
8412 
8413 	pd.sport = pd.dport = NULL;
8414 	pd.proto_sum = NULL;
8415 	pd.dir = dir;
8416 	pd.sidx = (dir == PF_IN) ? 0 : 1;
8417 	pd.didx = (dir == PF_IN) ? 1 : 0;
8418 	pd.af = AF_INET;
8419 	pd.act.rtableid = -1;
8420 
8421 	if (__predict_false(m->m_len < sizeof(struct ip)) &&
8422 	    (m = *m0 = m_pullup(*m0, sizeof(struct ip))) == NULL) {
8423 		DPFPRINTF(PF_DEBUG_URGENT,
8424 		    ("pf_test: m_len < sizeof(struct ip), pullup failed\n"));
8425 		PF_RULES_RUNLOCK();
8426 		return (PF_DROP);
8427 	}
8428 	h = mtod(m, struct ip *);
8429 	off = h->ip_hl << 2;
8430 
8431 	if (__predict_false(ip_divert_ptr != NULL) &&
8432 	    ((mtag = m_tag_locate(m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) {
8433 		struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1);
8434 		if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) ||
8435 		    (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) {
8436 			if (pd.pf_mtag == NULL &&
8437 			    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8438 				action = PF_DROP;
8439 				goto done;
8440 			}
8441 			pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED;
8442 		}
8443 		if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) {
8444 			m->m_flags |= M_FASTFWD_OURS;
8445 			pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT;
8446 		}
8447 		m_tag_delete(m, mtag);
8448 
8449 		mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
8450 		if (mtag != NULL)
8451 			m_tag_delete(m, mtag);
8452 	} else if (pf_normalize_ip(m0, kif, &reason, &pd) != PF_PASS) {
8453 		/* We do IP header normalization and packet reassembly here */
8454 		action = PF_DROP;
8455 		goto done;
8456 	}
8457 	m = *m0;	/* pf_normalize messes with m0 */
8458 	h = mtod(m, struct ip *);
8459 
8460 	off = h->ip_hl << 2;
8461 	if (off < (int)sizeof(struct ip)) {
8462 		action = PF_DROP;
8463 		REASON_SET(&reason, PFRES_SHORT);
8464 		pd.act.log = PF_LOG_FORCE;
8465 		goto done;
8466 	}
8467 
8468 	pd.src = (struct pf_addr *)&h->ip_src;
8469 	pd.dst = (struct pf_addr *)&h->ip_dst;
8470 	pd.ip_sum = &h->ip_sum;
8471 	pd.proto = h->ip_p;
8472 	pd.tos = h->ip_tos & ~IPTOS_ECN_MASK;
8473 	pd.tot_len = ntohs(h->ip_len);
8474 
8475 	/* handle fragments that didn't get reassembled by normalization */
8476 	if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
8477 		action = pf_test_fragment(&r, kif, m, h, &pd, &a, &ruleset);
8478 		goto done;
8479 	}
8480 
8481 	switch (h->ip_p) {
8482 	case IPPROTO_TCP: {
8483 		if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp),
8484 		    &action, &reason, AF_INET)) {
8485 			if (action != PF_PASS)
8486 				pd.act.log = PF_LOG_FORCE;
8487 			goto done;
8488 		}
8489 		pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2);
8490 
8491 		pd.sport = &pd.hdr.tcp.th_sport;
8492 		pd.dport = &pd.hdr.tcp.th_dport;
8493 
8494 		/* Respond to SYN with a syncookie. */
8495 		if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN &&
8496 		    pd.dir == PF_IN && pf_synflood_check(&pd)) {
8497 			pf_syncookie_send(m, off, &pd);
8498 			action = PF_DROP;
8499 			break;
8500 		}
8501 
8502 		if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0)
8503 			use_2nd_queue = 1;
8504 		action = pf_normalize_tcp(kif, m, 0, off, h, &pd);
8505 		if (action == PF_DROP)
8506 			goto done;
8507 		action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason);
8508 		if (action == PF_PASS) {
8509 			if (V_pfsync_update_state_ptr != NULL)
8510 				V_pfsync_update_state_ptr(s);
8511 			r = s->rule.ptr;
8512 			a = s->anchor.ptr;
8513 		} else if (s == NULL) {
8514 			/* Validate remote SYN|ACK, re-create original SYN if
8515 			 * valid. */
8516 			if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) ==
8517 			    TH_ACK && pf_syncookie_validate(&pd) &&
8518 			    pd.dir == PF_IN) {
8519 				struct mbuf *msyn;
8520 
8521 				msyn = pf_syncookie_recreate_syn(h->ip_ttl, off,
8522 				    &pd);
8523 				if (msyn == NULL) {
8524 					action = PF_DROP;
8525 					break;
8526 				}
8527 
8528 				action = pf_test(dir, pflags, ifp, &msyn, inp,
8529 				    &pd.act);
8530 				m_freem(msyn);
8531 				if (action != PF_PASS)
8532 					break;
8533 
8534 				action = pf_test_state_tcp(&s, kif, m, off, h,
8535 				    &pd, &reason);
8536 				if (action != PF_PASS || s == NULL) {
8537 					action = PF_DROP;
8538 					break;
8539 				}
8540 
8541 				s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1;
8542 				s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1;
8543 				pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST);
8544 				action = pf_synproxy(&pd, &s, &reason);
8545 				break;
8546 			} else {
8547 				action = pf_test_rule(&r, &s, kif, m, off, &pd,
8548 				    &a, &ruleset, inp);
8549 			}
8550 		}
8551 		break;
8552 	}
8553 
8554 	case IPPROTO_UDP: {
8555 		if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp),
8556 		    &action, &reason, AF_INET)) {
8557 			if (action != PF_PASS)
8558 				pd.act.log = PF_LOG_FORCE;
8559 			goto done;
8560 		}
8561 		pd.sport = &pd.hdr.udp.uh_sport;
8562 		pd.dport = &pd.hdr.udp.uh_dport;
8563 		if (pd.hdr.udp.uh_dport == 0 ||
8564 		    ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off ||
8565 		    ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) {
8566 			action = PF_DROP;
8567 			REASON_SET(&reason, PFRES_SHORT);
8568 			goto done;
8569 		}
8570 		action = pf_test_state_udp(&s, kif, m, off, h, &pd);
8571 		if (action == PF_PASS) {
8572 			if (V_pfsync_update_state_ptr != NULL)
8573 				V_pfsync_update_state_ptr(s);
8574 			r = s->rule.ptr;
8575 			a = s->anchor.ptr;
8576 		} else if (s == NULL)
8577 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8578 			    &a, &ruleset, inp);
8579 		break;
8580 	}
8581 
8582 	case IPPROTO_SCTP: {
8583 		if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp),
8584 		    &action, &reason, AF_INET)) {
8585 			if (action != PF_PASS)
8586 				pd.act.log |= PF_LOG_FORCE;
8587 			goto done;
8588 		}
8589 		pd.p_len = pd.tot_len - off;
8590 
8591 		pd.sport = &pd.hdr.sctp.src_port;
8592 		pd.dport = &pd.hdr.sctp.dest_port;
8593 		if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) {
8594 			action = PF_DROP;
8595 			REASON_SET(&reason, PFRES_SHORT);
8596 			goto done;
8597 		}
8598 		action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd);
8599 		if (action == PF_DROP)
8600 			goto done;
8601 		action = pf_test_state_sctp(&s, kif, m, off, h, &pd,
8602 		    &reason);
8603 		if (action == PF_PASS) {
8604 			if (V_pfsync_update_state_ptr != NULL)
8605 				V_pfsync_update_state_ptr(s);
8606 			r = s->rule.ptr;
8607 			a = s->anchor.ptr;
8608 		} else if (s == NULL) {
8609 			action = pf_test_rule(&r, &s, kif, m, off,
8610 			    &pd, &a, &ruleset, inp);
8611 		}
8612 		break;
8613 	}
8614 
8615 	case IPPROTO_ICMP: {
8616 		if (!pf_pull_hdr(m, off, &pd.hdr.icmp, ICMP_MINLEN,
8617 		    &action, &reason, AF_INET)) {
8618 			if (action != PF_PASS)
8619 				pd.act.log = PF_LOG_FORCE;
8620 			goto done;
8621 		}
8622 		action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason);
8623 		if (action == PF_PASS) {
8624 			if (V_pfsync_update_state_ptr != NULL)
8625 				V_pfsync_update_state_ptr(s);
8626 			r = s->rule.ptr;
8627 			a = s->anchor.ptr;
8628 		} else if (s == NULL)
8629 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8630 			    &a, &ruleset, inp);
8631 		break;
8632 	}
8633 
8634 #ifdef INET6
8635 	case IPPROTO_ICMPV6: {
8636 		action = PF_DROP;
8637 		DPFPRINTF(PF_DEBUG_MISC,
8638 		    ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
8639 		goto done;
8640 	}
8641 #endif
8642 
8643 	default:
8644 		action = pf_test_state_other(&s, kif, m, &pd);
8645 		if (action == PF_PASS) {
8646 			if (V_pfsync_update_state_ptr != NULL)
8647 				V_pfsync_update_state_ptr(s);
8648 			r = s->rule.ptr;
8649 			a = s->anchor.ptr;
8650 		} else if (s == NULL)
8651 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8652 			    &a, &ruleset, inp);
8653 		break;
8654 	}
8655 
8656 done:
8657 	PF_RULES_RUNLOCK();
8658 	if (action == PF_PASS && h->ip_hl > 5 &&
8659 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
8660 		action = PF_DROP;
8661 		REASON_SET(&reason, PFRES_IPOPTIONS);
8662 		pd.act.log = PF_LOG_FORCE;
8663 		DPFPRINTF(PF_DEBUG_MISC,
8664 		    ("pf: dropping packet with ip options\n"));
8665 	}
8666 
8667 	if (s) {
8668 		uint8_t log = pd.act.log;
8669 		memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions));
8670 		pd.act.log |= log;
8671 		tag = s->tag;
8672 		rt = s->rt;
8673 	} else {
8674 		tag = r->tag;
8675 		rt = r->rt;
8676 	}
8677 
8678 	if (tag > 0 && pf_tag_packet(m, &pd, tag)) {
8679 		action = PF_DROP;
8680 		REASON_SET(&reason, PFRES_MEMORY);
8681 	}
8682 
8683 	pf_scrub_ip(&m, &pd);
8684 	if (pd.proto == IPPROTO_TCP && pd.act.max_mss)
8685 		pf_normalize_mss(m, off, &pd);
8686 
8687 	if (pd.act.rtableid >= 0)
8688 		M_SETFIB(m, pd.act.rtableid);
8689 
8690 	if (pd.act.flags & PFSTATE_SETPRIO) {
8691 		if (pd.tos & IPTOS_LOWDELAY)
8692 			use_2nd_queue = 1;
8693 		if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) {
8694 			action = PF_DROP;
8695 			REASON_SET(&reason, PFRES_MEMORY);
8696 			pd.act.log = PF_LOG_FORCE;
8697 			DPFPRINTF(PF_DEBUG_MISC,
8698 			    ("pf: failed to allocate 802.1q mtag\n"));
8699 		}
8700 	}
8701 
8702 #ifdef ALTQ
8703 	if (action == PF_PASS && pd.act.qid) {
8704 		if (pd.pf_mtag == NULL &&
8705 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8706 			action = PF_DROP;
8707 			REASON_SET(&reason, PFRES_MEMORY);
8708 		} else {
8709 			if (s != NULL)
8710 				pd.pf_mtag->qid_hash = pf_state_hash(s);
8711 			if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY))
8712 				pd.pf_mtag->qid = pd.act.pqid;
8713 			else
8714 				pd.pf_mtag->qid = pd.act.qid;
8715 			/* Add hints for ecn. */
8716 			pd.pf_mtag->hdr = h;
8717 		}
8718 	}
8719 #endif /* ALTQ */
8720 
8721 	/*
8722 	 * connections redirected to loopback should not match sockets
8723 	 * bound specifically to loopback due to security implications,
8724 	 * see tcp_input() and in_pcblookup_listen().
8725 	 */
8726 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
8727 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
8728 	    (s->nat_rule.ptr->action == PF_RDR ||
8729 	    s->nat_rule.ptr->action == PF_BINAT) &&
8730 	    IN_LOOPBACK(ntohl(pd.dst->v4.s_addr)))
8731 		m->m_flags |= M_SKIP_FIREWALL;
8732 
8733 	if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS &&
8734 	    r->divert.port && !PACKET_LOOPED(&pd)) {
8735 		mtag = m_tag_alloc(MTAG_PF_DIVERT, 0,
8736 		    sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO);
8737 		if (mtag != NULL) {
8738 			((struct pf_divert_mtag *)(mtag+1))->port =
8739 			    ntohs(r->divert.port);
8740 			((struct pf_divert_mtag *)(mtag+1))->idir =
8741 			    (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN :
8742 			    PF_DIVERT_MTAG_DIR_OUT;
8743 
8744 			if (s)
8745 				PF_STATE_UNLOCK(s);
8746 
8747 			m_tag_prepend(m, mtag);
8748 			if (m->m_flags & M_FASTFWD_OURS) {
8749 				if (pd.pf_mtag == NULL &&
8750 				    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8751 					action = PF_DROP;
8752 					REASON_SET(&reason, PFRES_MEMORY);
8753 					pd.act.log = PF_LOG_FORCE;
8754 					DPFPRINTF(PF_DEBUG_MISC,
8755 					    ("pf: failed to allocate tag\n"));
8756 				} else {
8757 					pd.pf_mtag->flags |=
8758 					    PF_MTAG_FLAG_FASTFWD_OURS_PRESENT;
8759 					m->m_flags &= ~M_FASTFWD_OURS;
8760 				}
8761 			}
8762 			ip_divert_ptr(*m0, dir == PF_IN);
8763 			*m0 = NULL;
8764 
8765 			return (action);
8766 		} else {
8767 			/* XXX: ipfw has the same behaviour! */
8768 			action = PF_DROP;
8769 			REASON_SET(&reason, PFRES_MEMORY);
8770 			pd.act.log = PF_LOG_FORCE;
8771 			DPFPRINTF(PF_DEBUG_MISC,
8772 			    ("pf: failed to allocate divert tag\n"));
8773 		}
8774 	}
8775 	/* this flag will need revising if the pkt is forwarded */
8776 	if (pd.pf_mtag)
8777 		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED;
8778 
8779 	if (pd.act.log) {
8780 		struct pf_krule		*lr;
8781 		struct pf_krule_item	*ri;
8782 
8783 		if (s != NULL && s->nat_rule.ptr != NULL &&
8784 		    s->nat_rule.ptr->log & PF_LOG_ALL)
8785 			lr = s->nat_rule.ptr;
8786 		else
8787 			lr = r;
8788 
8789 		if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL)
8790 			PFLOG_PACKET(kif, m, AF_INET, action, reason, lr, a,
8791 			    ruleset, &pd, (s == NULL));
8792 		if (s) {
8793 			SLIST_FOREACH(ri, &s->match_rules, entry)
8794 				if (ri->r->log & PF_LOG_ALL)
8795 					PFLOG_PACKET(kif, m, AF_INET, action,
8796 					    reason, ri->r, a, ruleset, &pd, 0);
8797 		}
8798 	}
8799 
8800 	pf_counter_u64_critical_enter();
8801 	pf_counter_u64_add_protected(&kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS],
8802 	    pd.tot_len);
8803 	pf_counter_u64_add_protected(&kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS],
8804 	    1);
8805 
8806 	if (action == PF_PASS || r->action == PF_DROP) {
8807 		dirndx = (dir == PF_OUT);
8808 		pf_counter_u64_add_protected(&r->packets[dirndx], 1);
8809 		pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len);
8810 		pf_update_timestamp(r);
8811 
8812 		if (a != NULL) {
8813 			pf_counter_u64_add_protected(&a->packets[dirndx], 1);
8814 			pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len);
8815 		}
8816 		if (s != NULL) {
8817 			struct pf_krule_item	*ri;
8818 
8819 			if (s->nat_rule.ptr != NULL) {
8820 				pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx],
8821 				    1);
8822 				pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx],
8823 				    pd.tot_len);
8824 			}
8825 			if (s->src_node != NULL) {
8826 				counter_u64_add(s->src_node->packets[dirndx],
8827 				    1);
8828 				counter_u64_add(s->src_node->bytes[dirndx],
8829 				    pd.tot_len);
8830 			}
8831 			if (s->nat_src_node != NULL) {
8832 				counter_u64_add(s->nat_src_node->packets[dirndx],
8833 				    1);
8834 				counter_u64_add(s->nat_src_node->bytes[dirndx],
8835 				    pd.tot_len);
8836 			}
8837 			dirndx = (dir == s->direction) ? 0 : 1;
8838 			s->packets[dirndx]++;
8839 			s->bytes[dirndx] += pd.tot_len;
8840 			SLIST_FOREACH(ri, &s->match_rules, entry) {
8841 				pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1);
8842 				pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd.tot_len);
8843 			}
8844 		}
8845 		tr = r;
8846 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
8847 		if (nr != NULL && r == &V_pf_default_rule)
8848 			tr = nr;
8849 		if (tr->src.addr.type == PF_ADDR_TABLE)
8850 			pfr_update_stats(tr->src.addr.p.tbl,
8851 			    (s == NULL) ? pd.src :
8852 			    &s->key[(s->direction == PF_IN)]->
8853 				addr[(s->direction == PF_OUT)],
8854 			    pd.af, pd.tot_len, dir == PF_OUT,
8855 			    r->action == PF_PASS, tr->src.neg);
8856 		if (tr->dst.addr.type == PF_ADDR_TABLE)
8857 			pfr_update_stats(tr->dst.addr.p.tbl,
8858 			    (s == NULL) ? pd.dst :
8859 			    &s->key[(s->direction == PF_IN)]->
8860 				addr[(s->direction == PF_IN)],
8861 			    pd.af, pd.tot_len, dir == PF_OUT,
8862 			    r->action == PF_PASS, tr->dst.neg);
8863 	}
8864 	pf_counter_u64_critical_exit();
8865 
8866 	switch (action) {
8867 	case PF_SYNPROXY_DROP:
8868 		m_freem(*m0);
8869 	case PF_DEFER:
8870 		*m0 = NULL;
8871 		action = PF_PASS;
8872 		break;
8873 	case PF_DROP:
8874 		m_freem(*m0);
8875 		*m0 = NULL;
8876 		break;
8877 	default:
8878 		/* pf_route() returns unlocked. */
8879 		if (rt) {
8880 			pf_route(m0, r, kif->pfik_ifp, s, &pd, inp);
8881 			goto out;
8882 		}
8883 		if (pf_dummynet(&pd, s, r, m0) != 0) {
8884 			action = PF_DROP;
8885 			REASON_SET(&reason, PFRES_MEMORY);
8886 		}
8887 		break;
8888 	}
8889 
8890 	SDT_PROBE4(pf, ip, test, done, action, reason, r, s);
8891 
8892 	if (s && action != PF_DROP) {
8893 		if (!s->if_index_in && dir == PF_IN)
8894 			s->if_index_in = ifp->if_index;
8895 		else if (!s->if_index_out && dir == PF_OUT)
8896 			s->if_index_out = ifp->if_index;
8897 	}
8898 
8899 	if (s)
8900 		PF_STATE_UNLOCK(s);
8901 
8902 out:
8903 	pf_sctp_multihome_delayed(&pd, off, kif, s, action);
8904 
8905 	return (action);
8906 }
8907 #endif /* INET */
8908 
8909 #ifdef INET6
8910 int
8911 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp,
8912     struct pf_rule_actions *default_actions)
8913 {
8914 	struct pfi_kkif		*kif;
8915 	u_short			 action, reason = 0;
8916 	struct mbuf		*m = *m0, *n = NULL;
8917 	struct m_tag		*mtag;
8918 	struct ip6_hdr		*h = NULL;
8919 	struct pf_krule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
8920 	struct pf_kstate	*s = NULL;
8921 	struct pf_kruleset	*ruleset = NULL;
8922 	struct pf_pdesc		 pd;
8923 	int			 off, terminal = 0, dirndx, rh_cnt = 0, use_2nd_queue = 0;
8924 	uint16_t		 tag;
8925 	uint8_t			 rt;
8926 
8927 	PF_RULES_RLOCK_TRACKER;
8928 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir));
8929 	M_ASSERTPKTHDR(m);
8930 
8931 	if (!V_pf_status.running)
8932 		return (PF_PASS);
8933 
8934 	PF_RULES_RLOCK();
8935 
8936 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
8937 	if (__predict_false(kif == NULL)) {
8938 		DPFPRINTF(PF_DEBUG_URGENT,
8939 		    ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
8940 		PF_RULES_RUNLOCK();
8941 		return (PF_DROP);
8942 	}
8943 	if (kif->pfik_flags & PFI_IFLAG_SKIP) {
8944 		PF_RULES_RUNLOCK();
8945 		return (PF_PASS);
8946 	}
8947 
8948 	if (m->m_flags & M_SKIP_FIREWALL) {
8949 		PF_RULES_RUNLOCK();
8950 		return (PF_PASS);
8951 	}
8952 
8953 	/*
8954 	 * If we end up changing IP addresses (e.g. binat) the stack may get
8955 	 * confused and fail to send the icmp6 packet too big error. Just send
8956 	 * it here, before we do any NAT.
8957 	 */
8958 	if (dir == PF_OUT && pflags & PFIL_FWD && IN6_LINKMTU(ifp) < pf_max_frag_size(m)) {
8959 		PF_RULES_RUNLOCK();
8960 		*m0 = NULL;
8961 		icmp6_error(m, ICMP6_PACKET_TOO_BIG, 0, IN6_LINKMTU(ifp));
8962 		return (PF_DROP);
8963 	}
8964 
8965 	memset(&pd, 0, sizeof(pd));
8966 	TAILQ_INIT(&pd.sctp_multihome_jobs);
8967 	if (default_actions != NULL)
8968 		memcpy(&pd.act, default_actions, sizeof(pd.act));
8969 	pd.pf_mtag = pf_find_mtag(m);
8970 
8971 	if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) {
8972 		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
8973 
8974 		ifp = ifnet_byindexgen(pd.pf_mtag->if_index,
8975 		    pd.pf_mtag->if_idxgen);
8976 		if (ifp == NULL || ifp->if_flags & IFF_DYING) {
8977 			PF_RULES_RUNLOCK();
8978 			m_freem(*m0);
8979 			*m0 = NULL;
8980 			return (PF_PASS);
8981 		}
8982 		PF_RULES_RUNLOCK();
8983 		nd6_output_ifp(ifp, ifp, m,
8984                     (struct sockaddr_in6 *)&pd.pf_mtag->dst, NULL);
8985 		*m0 = NULL;
8986 		return (PF_PASS);
8987 	}
8988 
8989 	if (pd.pf_mtag && pd.pf_mtag->dnpipe) {
8990 		pd.act.dnpipe = pd.pf_mtag->dnpipe;
8991 		pd.act.flags = pd.pf_mtag->dnflags;
8992 	}
8993 
8994 	if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL &&
8995 	    pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
8996 		pf_dummynet_flag_remove(m, pd.pf_mtag);
8997 		/* Dummynet re-injects packets after they've
8998 		 * completed their delay. We've already
8999 		 * processed them, so pass unconditionally. */
9000 		PF_RULES_RUNLOCK();
9001 		return (PF_PASS);
9002 	}
9003 
9004 	pd.sport = pd.dport = NULL;
9005 	pd.ip_sum = NULL;
9006 	pd.proto_sum = NULL;
9007 	pd.dir = dir;
9008 	pd.sidx = (dir == PF_IN) ? 0 : 1;
9009 	pd.didx = (dir == PF_IN) ? 1 : 0;
9010 	pd.af = AF_INET6;
9011 	pd.act.rtableid = -1;
9012 
9013 	if (__predict_false(m->m_len < sizeof(struct ip6_hdr)) &&
9014 	    (m = *m0 = m_pullup(*m0, sizeof(struct ip6_hdr))) == NULL) {
9015 		DPFPRINTF(PF_DEBUG_URGENT,
9016 		    ("pf_test6: m_len < sizeof(struct ip6_hdr)"
9017 		     ", pullup failed\n"));
9018 		PF_RULES_RUNLOCK();
9019 		return (PF_DROP);
9020 	}
9021 	h = mtod(m, struct ip6_hdr *);
9022 	off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
9023 
9024 	/* We do IP header normalization and packet reassembly here */
9025 	if (pf_normalize_ip6(m0, kif, &reason, &pd) != PF_PASS) {
9026 		action = PF_DROP;
9027 		goto done;
9028 	}
9029 	m = *m0;	/* pf_normalize messes with m0 */
9030 	h = mtod(m, struct ip6_hdr *);
9031 	off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
9032 
9033 	/*
9034 	 * we do not support jumbogram.  if we keep going, zero ip6_plen
9035 	 * will do something bad, so drop the packet for now.
9036 	 */
9037 	if (htons(h->ip6_plen) == 0) {
9038 		action = PF_DROP;
9039 		REASON_SET(&reason, PFRES_NORM);	/*XXX*/
9040 		goto done;
9041 	}
9042 
9043 	pd.src = (struct pf_addr *)&h->ip6_src;
9044 	pd.dst = (struct pf_addr *)&h->ip6_dst;
9045 	pd.tos = IPV6_DSCP(h);
9046 	pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
9047 
9048 	pd.proto = h->ip6_nxt;
9049 	do {
9050 		switch (pd.proto) {
9051 		case IPPROTO_FRAGMENT:
9052 			action = pf_test_fragment(&r, kif, m, h, &pd, &a,
9053 			    &ruleset);
9054 			if (action == PF_DROP)
9055 				REASON_SET(&reason, PFRES_FRAG);
9056 			goto done;
9057 		case IPPROTO_ROUTING: {
9058 			struct ip6_rthdr rthdr;
9059 
9060 			if (rh_cnt++) {
9061 				DPFPRINTF(PF_DEBUG_MISC,
9062 				    ("pf: IPv6 more than one rthdr\n"));
9063 				action = PF_DROP;
9064 				REASON_SET(&reason, PFRES_IPOPTIONS);
9065 				pd.act.log = PF_LOG_FORCE;
9066 				goto done;
9067 			}
9068 			if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
9069 			    &reason, pd.af)) {
9070 				DPFPRINTF(PF_DEBUG_MISC,
9071 				    ("pf: IPv6 short rthdr\n"));
9072 				action = PF_DROP;
9073 				REASON_SET(&reason, PFRES_SHORT);
9074 				pd.act.log = PF_LOG_FORCE;
9075 				goto done;
9076 			}
9077 			if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
9078 				DPFPRINTF(PF_DEBUG_MISC,
9079 				    ("pf: IPv6 rthdr0\n"));
9080 				action = PF_DROP;
9081 				REASON_SET(&reason, PFRES_IPOPTIONS);
9082 				pd.act.log = PF_LOG_FORCE;
9083 				goto done;
9084 			}
9085 			/* FALLTHROUGH */
9086 		}
9087 		case IPPROTO_AH:
9088 		case IPPROTO_HOPOPTS:
9089 		case IPPROTO_DSTOPTS: {
9090 			/* get next header and header length */
9091 			struct ip6_ext	opt6;
9092 
9093 			if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
9094 			    NULL, &reason, pd.af)) {
9095 				DPFPRINTF(PF_DEBUG_MISC,
9096 				    ("pf: IPv6 short opt\n"));
9097 				action = PF_DROP;
9098 				pd.act.log = PF_LOG_FORCE;
9099 				goto done;
9100 			}
9101 			if (pd.proto == IPPROTO_AH)
9102 				off += (opt6.ip6e_len + 2) * 4;
9103 			else
9104 				off += (opt6.ip6e_len + 1) * 8;
9105 			pd.proto = opt6.ip6e_nxt;
9106 			/* goto the next header */
9107 			break;
9108 		}
9109 		default:
9110 			terminal++;
9111 			break;
9112 		}
9113 	} while (!terminal);
9114 
9115 	/* if there's no routing header, use unmodified mbuf for checksumming */
9116 	if (!n)
9117 		n = m;
9118 
9119 	switch (pd.proto) {
9120 	case IPPROTO_TCP: {
9121 		if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp),
9122 		    &action, &reason, AF_INET6)) {
9123 			if (action != PF_PASS)
9124 				pd.act.log |= PF_LOG_FORCE;
9125 			goto done;
9126 		}
9127 		pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2);
9128 		pd.sport = &pd.hdr.tcp.th_sport;
9129 		pd.dport = &pd.hdr.tcp.th_dport;
9130 
9131 		/* Respond to SYN with a syncookie. */
9132 		if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN &&
9133 		    pd.dir == PF_IN && pf_synflood_check(&pd)) {
9134 			pf_syncookie_send(m, off, &pd);
9135 			action = PF_DROP;
9136 			break;
9137 		}
9138 
9139 		action = pf_normalize_tcp(kif, m, 0, off, h, &pd);
9140 		if (action == PF_DROP)
9141 			goto done;
9142 		action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason);
9143 		if (action == PF_PASS) {
9144 			if (V_pfsync_update_state_ptr != NULL)
9145 				V_pfsync_update_state_ptr(s);
9146 			r = s->rule.ptr;
9147 			a = s->anchor.ptr;
9148 		} else if (s == NULL) {
9149 			/* Validate remote SYN|ACK, re-create original SYN if
9150 			 * valid. */
9151 			if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) ==
9152 			    TH_ACK && pf_syncookie_validate(&pd) &&
9153 			    pd.dir == PF_IN) {
9154 				struct mbuf *msyn;
9155 
9156 				msyn = pf_syncookie_recreate_syn(h->ip6_hlim,
9157 				    off, &pd);
9158 				if (msyn == NULL) {
9159 					action = PF_DROP;
9160 					break;
9161 				}
9162 
9163 				action = pf_test6(dir, pflags, ifp, &msyn, inp,
9164 				    &pd.act);
9165 				m_freem(msyn);
9166 				if (action != PF_PASS)
9167 					break;
9168 
9169 				action = pf_test_state_tcp(&s, kif, m, off, h,
9170 				    &pd, &reason);
9171 				if (action != PF_PASS || s == NULL) {
9172 					action = PF_DROP;
9173 					break;
9174 				}
9175 
9176 				s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1;
9177 				s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1;
9178 				pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST);
9179 
9180 				action = pf_synproxy(&pd, &s, &reason);
9181 				break;
9182 			} else {
9183 				action = pf_test_rule(&r, &s, kif, m, off, &pd,
9184 				    &a, &ruleset, inp);
9185 			}
9186 		}
9187 		break;
9188 	}
9189 
9190 	case IPPROTO_UDP: {
9191 		if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp),
9192 		    &action, &reason, AF_INET6)) {
9193 			if (action != PF_PASS)
9194 				pd.act.log |= PF_LOG_FORCE;
9195 			goto done;
9196 		}
9197 		pd.sport = &pd.hdr.udp.uh_sport;
9198 		pd.dport = &pd.hdr.udp.uh_dport;
9199 		if (pd.hdr.udp.uh_dport == 0 ||
9200 		    ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off ||
9201 		    ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) {
9202 			action = PF_DROP;
9203 			REASON_SET(&reason, PFRES_SHORT);
9204 			goto done;
9205 		}
9206 		action = pf_test_state_udp(&s, kif, m, off, h, &pd);
9207 		if (action == PF_PASS) {
9208 			if (V_pfsync_update_state_ptr != NULL)
9209 				V_pfsync_update_state_ptr(s);
9210 			r = s->rule.ptr;
9211 			a = s->anchor.ptr;
9212 		} else if (s == NULL)
9213 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
9214 			    &a, &ruleset, inp);
9215 		break;
9216 	}
9217 
9218 	case IPPROTO_SCTP: {
9219 		if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp),
9220 		    &action, &reason, AF_INET6)) {
9221 			if (action != PF_PASS)
9222 				pd.act.log |= PF_LOG_FORCE;
9223 			goto done;
9224 		}
9225 		pd.sport = &pd.hdr.sctp.src_port;
9226 		pd.dport = &pd.hdr.sctp.dest_port;
9227 		if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) {
9228 			action = PF_DROP;
9229 			REASON_SET(&reason, PFRES_SHORT);
9230 			goto done;
9231 		}
9232 		action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd);
9233 		if (action == PF_DROP)
9234 			goto done;
9235 		action = pf_test_state_sctp(&s, kif, m, off, h, &pd,
9236 		    &reason);
9237 		if (action == PF_PASS) {
9238 			if (V_pfsync_update_state_ptr != NULL)
9239 				V_pfsync_update_state_ptr(s);
9240 			r = s->rule.ptr;
9241 			a = s->anchor.ptr;
9242 		} else if (s == NULL) {
9243 			action = pf_test_rule(&r, &s, kif, m, off,
9244 			    &pd, &a, &ruleset, inp);
9245 		}
9246 		break;
9247 	}
9248 
9249 	case IPPROTO_ICMP: {
9250 		action = PF_DROP;
9251 		DPFPRINTF(PF_DEBUG_MISC,
9252 		    ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
9253 		goto done;
9254 	}
9255 
9256 	case IPPROTO_ICMPV6: {
9257 		if (!pf_pull_hdr(m, off, &pd.hdr.icmp6, sizeof(pd.hdr.icmp6),
9258 		    &action, &reason, AF_INET6)) {
9259 			if (action != PF_PASS)
9260 				pd.act.log |= PF_LOG_FORCE;
9261 			goto done;
9262 		}
9263 		action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason);
9264 		if (action == PF_PASS) {
9265 			if (V_pfsync_update_state_ptr != NULL)
9266 				V_pfsync_update_state_ptr(s);
9267 			r = s->rule.ptr;
9268 			a = s->anchor.ptr;
9269 		} else if (s == NULL)
9270 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
9271 			    &a, &ruleset, inp);
9272 		break;
9273 	}
9274 
9275 	default:
9276 		action = pf_test_state_other(&s, kif, m, &pd);
9277 		if (action == PF_PASS) {
9278 			if (V_pfsync_update_state_ptr != NULL)
9279 				V_pfsync_update_state_ptr(s);
9280 			r = s->rule.ptr;
9281 			a = s->anchor.ptr;
9282 		} else if (s == NULL)
9283 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
9284 			    &a, &ruleset, inp);
9285 		break;
9286 	}
9287 
9288 done:
9289 	PF_RULES_RUNLOCK();
9290 	if (n != m) {
9291 		m_freem(n);
9292 		n = NULL;
9293 	}
9294 
9295 	/* handle dangerous IPv6 extension headers. */
9296 	if (action == PF_PASS && rh_cnt &&
9297 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
9298 		action = PF_DROP;
9299 		REASON_SET(&reason, PFRES_IPOPTIONS);
9300 		pd.act.log = r->log;
9301 		DPFPRINTF(PF_DEBUG_MISC,
9302 		    ("pf: dropping packet with dangerous v6 headers\n"));
9303 	}
9304 
9305 	if (s) {
9306 		uint8_t log = pd.act.log;
9307 		memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions));
9308 		pd.act.log |= log;
9309 		tag = s->tag;
9310 		rt = s->rt;
9311 	} else {
9312 		tag = r->tag;
9313 		rt = r->rt;
9314 	}
9315 
9316 	if (tag > 0 && pf_tag_packet(m, &pd, tag)) {
9317 		action = PF_DROP;
9318 		REASON_SET(&reason, PFRES_MEMORY);
9319 	}
9320 
9321 	pf_scrub_ip6(&m, &pd);
9322 	if (pd.proto == IPPROTO_TCP && pd.act.max_mss)
9323 		pf_normalize_mss(m, off, &pd);
9324 
9325 	if (pd.act.rtableid >= 0)
9326 		M_SETFIB(m, pd.act.rtableid);
9327 
9328 	if (pd.act.flags & PFSTATE_SETPRIO) {
9329 		if (pd.tos & IPTOS_LOWDELAY)
9330 			use_2nd_queue = 1;
9331 		if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) {
9332 			action = PF_DROP;
9333 			REASON_SET(&reason, PFRES_MEMORY);
9334 			pd.act.log = PF_LOG_FORCE;
9335 			DPFPRINTF(PF_DEBUG_MISC,
9336 			    ("pf: failed to allocate 802.1q mtag\n"));
9337 		}
9338 	}
9339 
9340 #ifdef ALTQ
9341 	if (action == PF_PASS && pd.act.qid) {
9342 		if (pd.pf_mtag == NULL &&
9343 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
9344 			action = PF_DROP;
9345 			REASON_SET(&reason, PFRES_MEMORY);
9346 		} else {
9347 			if (s != NULL)
9348 				pd.pf_mtag->qid_hash = pf_state_hash(s);
9349 			if (pd.tos & IPTOS_LOWDELAY)
9350 				pd.pf_mtag->qid = pd.act.pqid;
9351 			else
9352 				pd.pf_mtag->qid = pd.act.qid;
9353 			/* Add hints for ecn. */
9354 			pd.pf_mtag->hdr = h;
9355 		}
9356 	}
9357 #endif /* ALTQ */
9358 
9359 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
9360 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
9361 	    (s->nat_rule.ptr->action == PF_RDR ||
9362 	    s->nat_rule.ptr->action == PF_BINAT) &&
9363 	    IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
9364 		m->m_flags |= M_SKIP_FIREWALL;
9365 
9366 	/* XXX: Anybody working on it?! */
9367 	if (r->divert.port)
9368 		printf("pf: divert(9) is not supported for IPv6\n");
9369 
9370 	if (pd.act.log) {
9371 		struct pf_krule		*lr;
9372 		struct pf_krule_item	*ri;
9373 
9374 		if (s != NULL && s->nat_rule.ptr != NULL &&
9375 		    s->nat_rule.ptr->log & PF_LOG_ALL)
9376 			lr = s->nat_rule.ptr;
9377 		else
9378 			lr = r;
9379 
9380 		if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL)
9381 			PFLOG_PACKET(kif, m, AF_INET6, action, reason, lr, a, ruleset,
9382 			    &pd, (s == NULL));
9383 		if (s) {
9384 			SLIST_FOREACH(ri, &s->match_rules, entry)
9385 				if (ri->r->log & PF_LOG_ALL)
9386 					PFLOG_PACKET(kif, m, AF_INET6, action, reason,
9387 					    ri->r, a, ruleset, &pd, 0);
9388 		}
9389 	}
9390 
9391 	pf_counter_u64_critical_enter();
9392 	pf_counter_u64_add_protected(&kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS],
9393 	    pd.tot_len);
9394 	pf_counter_u64_add_protected(&kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS],
9395 	    1);
9396 
9397 	if (action == PF_PASS || r->action == PF_DROP) {
9398 		dirndx = (dir == PF_OUT);
9399 		pf_counter_u64_add_protected(&r->packets[dirndx], 1);
9400 		pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len);
9401 		if (a != NULL) {
9402 			pf_counter_u64_add_protected(&a->packets[dirndx], 1);
9403 			pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len);
9404 		}
9405 		if (s != NULL) {
9406 			if (s->nat_rule.ptr != NULL) {
9407 				pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx],
9408 				    1);
9409 				pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx],
9410 				    pd.tot_len);
9411 			}
9412 			if (s->src_node != NULL) {
9413 				counter_u64_add(s->src_node->packets[dirndx],
9414 				    1);
9415 				counter_u64_add(s->src_node->bytes[dirndx],
9416 				    pd.tot_len);
9417 			}
9418 			if (s->nat_src_node != NULL) {
9419 				counter_u64_add(s->nat_src_node->packets[dirndx],
9420 				    1);
9421 				counter_u64_add(s->nat_src_node->bytes[dirndx],
9422 				    pd.tot_len);
9423 			}
9424 			dirndx = (dir == s->direction) ? 0 : 1;
9425 			s->packets[dirndx]++;
9426 			s->bytes[dirndx] += pd.tot_len;
9427 		}
9428 		tr = r;
9429 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
9430 		if (nr != NULL && r == &V_pf_default_rule)
9431 			tr = nr;
9432 		if (tr->src.addr.type == PF_ADDR_TABLE)
9433 			pfr_update_stats(tr->src.addr.p.tbl,
9434 			    (s == NULL) ? pd.src :
9435 			    &s->key[(s->direction == PF_IN)]->addr[0],
9436 			    pd.af, pd.tot_len, dir == PF_OUT,
9437 			    r->action == PF_PASS, tr->src.neg);
9438 		if (tr->dst.addr.type == PF_ADDR_TABLE)
9439 			pfr_update_stats(tr->dst.addr.p.tbl,
9440 			    (s == NULL) ? pd.dst :
9441 			    &s->key[(s->direction == PF_IN)]->addr[1],
9442 			    pd.af, pd.tot_len, dir == PF_OUT,
9443 			    r->action == PF_PASS, tr->dst.neg);
9444 	}
9445 	pf_counter_u64_critical_exit();
9446 
9447 	switch (action) {
9448 	case PF_SYNPROXY_DROP:
9449 		m_freem(*m0);
9450 	case PF_DEFER:
9451 		*m0 = NULL;
9452 		action = PF_PASS;
9453 		break;
9454 	case PF_DROP:
9455 		m_freem(*m0);
9456 		*m0 = NULL;
9457 		break;
9458 	default:
9459 		/* pf_route6() returns unlocked. */
9460 		if (rt) {
9461 			pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp);
9462 			goto out;
9463 		}
9464 		if (pf_dummynet(&pd, s, r, m0) != 0) {
9465 			action = PF_DROP;
9466 			REASON_SET(&reason, PFRES_MEMORY);
9467 		}
9468 		break;
9469 	}
9470 
9471 	if (s && action != PF_DROP) {
9472 		if (!s->if_index_in && dir == PF_IN)
9473 			s->if_index_in = ifp->if_index;
9474 		else if (!s->if_index_out && dir == PF_OUT)
9475 			s->if_index_out = ifp->if_index;
9476 	}
9477 
9478 	if (s)
9479 		PF_STATE_UNLOCK(s);
9480 
9481 	/* If reassembled packet passed, create new fragments. */
9482 	if (action == PF_PASS && *m0 && dir == PF_OUT &&
9483 	    (mtag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL)
9484 		action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD);
9485 
9486 out:
9487 	SDT_PROBE4(pf, ip, test6, done, action, reason, r, s);
9488 
9489 	pf_sctp_multihome_delayed(&pd, off, kif, s, action);
9490 
9491 	return (action);
9492 }
9493 #endif /* INET6 */
9494