xref: /dragonfly/sys/net/pf/pf.c (revision b272101a)
1 /*
2  * Copyright (c) 2004 The DragonFly Project.  All rights reserved.
3  *
4  * Copyright (c) 2001 Daniel Hartmeier
5  * Copyright (c) 2002 - 2008 Henning Brauer
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  *
12  *    - Redistributions of source code must retain the above copyright
13  *      notice, this list of conditions and the following disclaimer.
14  *    - Redistributions in binary form must reproduce the above
15  *      copyright notice, this list of conditions and the following
16  *      disclaimer in the documentation and/or other materials provided
17  *      with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
22  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
23  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
24  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
25  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
26  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
27  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
29  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  *
32  * Effort sponsored in part by the Defense Advanced Research Projects
33  * Agency (DARPA) and Air Force Research Laboratory, Air Force
34  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
35  *
36  */
37 
38 #include "opt_inet.h"
39 #include "opt_inet6.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/filio.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/kernel.h>
49 #include <sys/time.h>
50 #include <sys/sysctl.h>
51 #include <sys/endian.h>
52 #include <sys/proc.h>
53 #include <sys/kthread.h>
54 #include <sys/spinlock.h>
55 
56 #include <sys/md5.h>
57 
58 #include <net/if.h>
59 #include <net/if_types.h>
60 #include <net/bpf.h>
61 #include <net/netisr2.h>
62 #include <net/route.h>
63 
64 #include <netinet/in.h>
65 #include <netinet/in_var.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/ip.h>
68 #include <netinet/ip_var.h>
69 #include <netinet/tcp.h>
70 #include <netinet/tcp_seq.h>
71 #include <netinet/udp.h>
72 #include <netinet/ip_icmp.h>
73 #include <netinet/in_pcb.h>
74 #include <netinet/tcp_timer.h>
75 #include <netinet/tcp_var.h>
76 #include <netinet/udp_var.h>
77 #include <netinet/icmp_var.h>
78 #include <netinet/if_ether.h>
79 
80 #include <net/pf/pfvar.h>
81 #include <net/pf/if_pflog.h>
82 
83 #include <net/pf/if_pfsync.h>
84 
85 #ifdef INET6
86 #include <netinet/ip6.h>
87 #include <netinet/icmp6.h>
88 #include <netinet6/nd6.h>
89 #include <netinet6/ip6_var.h>
90 #include <netinet6/in6_pcb.h>
91 #endif /* INET6 */
92 
93 #include <sys/in_cksum.h>
94 #include <sys/ucred.h>
95 #include <machine/limits.h>
96 #include <sys/msgport2.h>
97 #include <sys/spinlock2.h>
98 #include <net/netmsg2.h>
99 #include <net/toeplitz2.h>
100 
101 extern int ip_optcopy(struct ip *, struct ip *);
102 extern int debug_pfugidhack;
103 
104 /*
105  * pf_token - shared lock for cpu-localized operations,
106  *	      exclusive lock otherwise.
107  *
108  * pf_gtoken- exclusive lock used for initialization.
109  */
110 struct lwkt_token pf_token = LWKT_TOKEN_INITIALIZER(pf_token);
111 struct lwkt_token pf_gtoken = LWKT_TOKEN_INITIALIZER(pf_gtoken);
112 
113 #define DPFPRINTF(n, x)	if (pf_status.debug >= (n)) kprintf x
114 
115 #define FAIL(code)	{ error = (code); goto done; }
116 
117 /*
118  * Global variables
119  */
120 
121 /* mask radix tree */
122 struct radix_node_head	*pf_maskhead;
123 
124 /* state tables */
125 struct pf_state_tree	 *pf_statetbl;		/* incls one global table */
126 struct pf_state		**purge_cur;
127 struct pf_altqqueue	 pf_altqs[2];
128 struct pf_palist	 pf_pabuf;
129 struct pf_altqqueue	*pf_altqs_active;
130 struct pf_altqqueue	*pf_altqs_inactive;
131 struct pf_status	 pf_status;
132 
133 u_int32_t		 ticket_altqs_active;
134 u_int32_t		 ticket_altqs_inactive;
135 int			 altqs_inactive_open;
136 u_int32_t		 ticket_pabuf;
137 
138 MD5_CTX			 pf_tcp_secret_ctx;
139 u_char			 pf_tcp_secret[16];
140 int			 pf_tcp_secret_init;
141 int			 pf_tcp_iss_off;
142 
143 struct pf_anchor_stackframe {
144 	struct pf_ruleset			*rs;
145 	struct pf_rule				*r;
146 	struct pf_anchor_node			*parent;
147 	struct pf_anchor			*child;
148 } pf_anchor_stack[64];
149 
150 struct malloc_type	 *pf_src_tree_pl, *pf_rule_pl, *pf_pooladdr_pl;
151 struct malloc_type	 *pf_state_pl, *pf_state_key_pl, *pf_state_item_pl;
152 struct malloc_type	 *pf_altq_pl;
153 
154 void			 pf_print_host(struct pf_addr *, u_int16_t, u_int8_t);
155 
156 void			 pf_init_threshold(struct pf_threshold *, u_int32_t,
157 			    u_int32_t);
158 void			 pf_add_threshold(struct pf_threshold *);
159 int			 pf_check_threshold(struct pf_threshold *);
160 
161 void			 pf_change_ap(struct pf_addr *, u_int16_t *,
162 			    u_int16_t *, u_int16_t *, struct pf_addr *,
163 			    u_int16_t, u_int8_t, sa_family_t);
164 int			 pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
165 			    struct tcphdr *, struct pf_state_peer *);
166 #ifdef INET6
167 void			 pf_change_a6(struct pf_addr *, u_int16_t *,
168 			    struct pf_addr *, u_int8_t);
169 #endif /* INET6 */
170 void			 pf_change_icmp(struct pf_addr *, u_int16_t *,
171 			    struct pf_addr *, struct pf_addr *, u_int16_t,
172 			    u_int16_t *, u_int16_t *, u_int16_t *,
173 			    u_int16_t *, u_int8_t, sa_family_t);
174 void			 pf_send_tcp(const struct pf_rule *, sa_family_t,
175 			    const struct pf_addr *, const struct pf_addr *,
176 			    u_int16_t, u_int16_t, u_int32_t, u_int32_t,
177 			    u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
178 			    u_int16_t, struct ether_header *, struct ifnet *);
179 void			 pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
180 			    sa_family_t, struct pf_rule *);
181 struct pf_rule		*pf_match_translation(struct pf_pdesc *, struct mbuf *,
182 			    int, int, struct pfi_kif *,
183 			    struct pf_addr *, u_int16_t, struct pf_addr *,
184 			    u_int16_t, int);
185 struct pf_rule		*pf_get_translation(struct pf_pdesc *, struct mbuf *,
186 			    int, int, struct pfi_kif *, struct pf_src_node **,
187 			    struct pf_state_key **, struct pf_state_key **,
188 			    struct pf_state_key **, struct pf_state_key **,
189 			    struct pf_addr *, struct pf_addr *,
190 			    u_int16_t, u_int16_t);
191 void			 pf_detach_state(struct pf_state *);
192 int			 pf_state_key_setup(struct pf_pdesc *, struct pf_rule *,
193 			    struct pf_state_key **, struct pf_state_key **,
194 			    struct pf_state_key **, struct pf_state_key **,
195 			    struct pf_addr *, struct pf_addr *,
196 			    u_int16_t, u_int16_t);
197 void			 pf_state_key_detach(struct pf_state *, int);
198 u_int32_t		 pf_tcp_iss(struct pf_pdesc *);
199 int			 pf_test_rule(struct pf_rule **, struct pf_state **,
200 			    int, struct pfi_kif *, struct mbuf *, int,
201 			    void *, struct pf_pdesc *, struct pf_rule **,
202 			    struct pf_ruleset **, struct ifqueue *, struct inpcb *);
203 static __inline int	 pf_create_state(struct pf_rule *, struct pf_rule *,
204 			    struct pf_rule *, struct pf_pdesc *,
205 			    struct pf_src_node *, struct pf_state_key *,
206 			    struct pf_state_key *, struct pf_state_key *,
207 			    struct pf_state_key *, struct mbuf *, int,
208 			    u_int16_t, u_int16_t, int *, struct pfi_kif *,
209 			    struct pf_state **, int, u_int16_t, u_int16_t,
210 			    int);
211 int			 pf_test_fragment(struct pf_rule **, int,
212 			    struct pfi_kif *, struct mbuf *, void *,
213 			    struct pf_pdesc *, struct pf_rule **,
214 			    struct pf_ruleset **);
215 int			 pf_tcp_track_full(struct pf_state_peer *,
216 			    struct pf_state_peer *, struct pf_state **,
217 			    struct pfi_kif *, struct mbuf *, int,
218 			    struct pf_pdesc *, u_short *, int *);
219 int			pf_tcp_track_sloppy(struct pf_state_peer *,
220 			    struct pf_state_peer *, struct pf_state **,
221 			    struct pf_pdesc *, u_short *);
222 int			 pf_test_state_tcp(struct pf_state **, int,
223 			    struct pfi_kif *, struct mbuf *, int,
224 			    void *, struct pf_pdesc *, u_short *);
225 int			 pf_test_state_udp(struct pf_state **, int,
226 			    struct pfi_kif *, struct mbuf *, int,
227 			    void *, struct pf_pdesc *);
228 int			 pf_test_state_icmp(struct pf_state **, int,
229 			    struct pfi_kif *, struct mbuf *, int,
230 			    void *, struct pf_pdesc *, u_short *);
231 int			 pf_test_state_other(struct pf_state **, int,
232 			    struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
233 void			 pf_step_into_anchor(int *, struct pf_ruleset **, int,
234 			    struct pf_rule **, struct pf_rule **, int *);
235 int			 pf_step_out_of_anchor(int *, struct pf_ruleset **,
236 			     int, struct pf_rule **, struct pf_rule **,
237 			     int *);
238 void			 pf_hash(struct pf_addr *, struct pf_addr *,
239 			    struct pf_poolhashkey *, sa_family_t);
240 int			 pf_map_addr(u_int8_t, struct pf_rule *,
241 			    struct pf_addr *, struct pf_addr *,
242 			    struct pf_addr *, struct pf_src_node **);
243 int			 pf_get_sport(struct pf_pdesc *,
244 			    sa_family_t, u_int8_t, struct pf_rule *,
245 			    struct pf_addr *, struct pf_addr *,
246 			    u_int16_t, u_int16_t,
247 			    struct pf_addr *, u_int16_t *,
248 			    u_int16_t, u_int16_t,
249 			    struct pf_src_node **);
250 void			 pf_route(struct mbuf **, struct pf_rule *, int,
251 			    struct ifnet *, struct pf_state *,
252 			    struct pf_pdesc *);
253 void			 pf_route6(struct mbuf **, struct pf_rule *, int,
254 			    struct ifnet *, struct pf_state *,
255 			    struct pf_pdesc *);
256 u_int8_t		 pf_get_wscale(struct mbuf *, int, u_int16_t,
257 			    sa_family_t);
258 u_int16_t		 pf_get_mss(struct mbuf *, int, u_int16_t,
259 			    sa_family_t);
260 u_int16_t		 pf_calc_mss(struct pf_addr *, sa_family_t,
261 				u_int16_t);
262 void			 pf_set_rt_ifp(struct pf_state *,
263 			    struct pf_addr *);
264 int			 pf_check_proto_cksum(struct mbuf *, int, int,
265 			    u_int8_t, sa_family_t);
266 struct pf_divert	*pf_get_divert(struct mbuf *);
267 void			 pf_print_state_parts(struct pf_state *,
268 			    struct pf_state_key *, struct pf_state_key *);
269 int			 pf_addr_wrap_neq(struct pf_addr_wrap *,
270 			    struct pf_addr_wrap *);
271 struct pf_state		*pf_find_state(struct pfi_kif *,
272 			    struct pf_state_key_cmp *, u_int, struct mbuf *);
273 int			 pf_src_connlimit(struct pf_state *);
274 int			 pf_check_congestion(struct ifqueue *);
275 
276 extern int pf_end_threads;
277 
278 struct pf_pool_limit pf_pool_limits[PF_LIMIT_MAX] = {
279 	{ &pf_state_pl, PFSTATE_HIWAT },
280 	{ &pf_src_tree_pl, PFSNODE_HIWAT },
281 	{ &pf_frent_pl, PFFRAG_FRENT_HIWAT },
282 	{ &pfr_ktable_pl, PFR_KTABLE_HIWAT },
283 	{ &pfr_kentry_pl, PFR_KENTRY_HIWAT }
284 };
285 
286 /*
287  * If route-to and direction is out we match with no further processing
288  *	(rt_kif must be assigned and not equal to the out interface)
289  * If reply-to and direction is in we match with no further processing
290  *	(rt_kif must be assigned and not equal to the in interface)
291  */
292 #define STATE_LOOKUP(i, k, d, s, m)					\
293 	do {								\
294 		s = pf_find_state(i, k, d, m);				\
295 		if (s == NULL || (s)->timeout == PFTM_PURGE)		\
296 			return (PF_DROP);				\
297 		if (d == PF_OUT &&					\
298 		    (((s)->rule.ptr->rt == PF_ROUTETO &&		\
299 		    (s)->rule.ptr->direction == PF_OUT) ||		\
300 		    ((s)->rule.ptr->rt == PF_REPLYTO &&			\
301 		    (s)->rule.ptr->direction == PF_IN)) &&		\
302 		    (s)->rt_kif != NULL &&				\
303 		    (s)->rt_kif != i)					\
304 			return (PF_PASS);				\
305 	} while (0)
306 
307 #define BOUND_IFACE(r, k) \
308 	((r)->rule_flag & PFRULE_IFBOUND) ? (k) : pfi_all
309 
310 #define STATE_INC_COUNTERS(s)				\
311 	do {						\
312 		atomic_add_int(&s->rule.ptr->states_cur, 1);	\
313 		s->rule.ptr->states_tot++;		\
314 		if (s->anchor.ptr != NULL) {		\
315 			atomic_add_int(&s->anchor.ptr->states_cur, 1);	\
316 			s->anchor.ptr->states_tot++;	\
317 		}					\
318 		if (s->nat_rule.ptr != NULL) {		\
319 			atomic_add_int(&s->nat_rule.ptr->states_cur, 1); \
320 			s->nat_rule.ptr->states_tot++;	\
321 		}					\
322 	} while (0)
323 
324 #define STATE_DEC_COUNTERS(s)				\
325 	do {						\
326 		if (s->nat_rule.ptr != NULL)		\
327 			atomic_add_int(&s->nat_rule.ptr->states_cur, -1); \
328 		if (s->anchor.ptr != NULL)		\
329 			atomic_add_int(&s->anchor.ptr->states_cur, -1);	\
330 		atomic_add_int(&s->rule.ptr->states_cur, -1);		\
331 	} while (0)
332 
333 static MALLOC_DEFINE(M_PFSTATEPL, "pfstatepl", "pf state pool list");
334 static MALLOC_DEFINE(M_PFSRCTREEPL, "pfsrctpl", "pf source tree pool list");
335 static MALLOC_DEFINE(M_PFSTATEKEYPL, "pfstatekeypl", "pf state key pool list");
336 static MALLOC_DEFINE(M_PFSTATEITEMPL, "pfstateitempl", "pf state item pool list");
337 
338 static __inline int pf_src_compare(struct pf_src_node *, struct pf_src_node *);
339 static __inline int pf_state_compare_key(struct pf_state_key *,
340 				struct pf_state_key *);
341 static __inline int pf_state_compare_rkey(struct pf_state_key *,
342 				struct pf_state_key *);
343 static __inline int pf_state_compare_id(struct pf_state *,
344 				struct pf_state *);
345 
346 struct pf_src_tree *tree_src_tracking;
347 struct pf_state_tree_id *tree_id;
348 struct pf_state_queue *state_list;
349 struct pf_counters *pf_counters;
350 
351 RB_GENERATE(pf_src_tree, pf_src_node, entry, pf_src_compare);
352 RB_GENERATE(pf_state_tree, pf_state_key, entry, pf_state_compare_key);
353 RB_GENERATE(pf_state_rtree, pf_state_key, entry, pf_state_compare_rkey);
354 RB_GENERATE(pf_state_tree_id, pf_state, entry_id, pf_state_compare_id);
355 
356 static __inline int
pf_src_compare(struct pf_src_node * a,struct pf_src_node * b)357 pf_src_compare(struct pf_src_node *a, struct pf_src_node *b)
358 {
359 	int	diff;
360 
361 	if (a->rule.ptr > b->rule.ptr)
362 		return (1);
363 	if (a->rule.ptr < b->rule.ptr)
364 		return (-1);
365 	if ((diff = a->af - b->af) != 0)
366 		return (diff);
367 	switch (a->af) {
368 #ifdef INET
369 	case AF_INET:
370 		if (a->addr.addr32[0] > b->addr.addr32[0])
371 			return (1);
372 		if (a->addr.addr32[0] < b->addr.addr32[0])
373 			return (-1);
374 		break;
375 #endif /* INET */
376 #ifdef INET6
377 	case AF_INET6:
378 		if (a->addr.addr32[3] > b->addr.addr32[3])
379 			return (1);
380 		if (a->addr.addr32[3] < b->addr.addr32[3])
381 			return (-1);
382 		if (a->addr.addr32[2] > b->addr.addr32[2])
383 			return (1);
384 		if (a->addr.addr32[2] < b->addr.addr32[2])
385 			return (-1);
386 		if (a->addr.addr32[1] > b->addr.addr32[1])
387 			return (1);
388 		if (a->addr.addr32[1] < b->addr.addr32[1])
389 			return (-1);
390 		if (a->addr.addr32[0] > b->addr.addr32[0])
391 			return (1);
392 		if (a->addr.addr32[0] < b->addr.addr32[0])
393 			return (-1);
394 		break;
395 #endif /* INET6 */
396 	}
397 	return (0);
398 }
399 
400 u_int32_t
pf_state_hash(struct pf_state_key * sk)401 pf_state_hash(struct pf_state_key *sk)
402 {
403 	u_int32_t hv = (u_int32_t)(((intptr_t)sk >> 6) ^ ((intptr_t)sk >> 15));
404 	if (hv == 0)	/* disallow 0 */
405 		hv = 1;
406 	return(hv);
407 }
408 
409 #ifdef INET6
410 void
pf_addrcpy(struct pf_addr * dst,struct pf_addr * src,sa_family_t af)411 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
412 {
413 	switch (af) {
414 #ifdef INET
415 	case AF_INET:
416 		dst->addr32[0] = src->addr32[0];
417 		break;
418 #endif /* INET */
419 	case AF_INET6:
420 		dst->addr32[0] = src->addr32[0];
421 		dst->addr32[1] = src->addr32[1];
422 		dst->addr32[2] = src->addr32[2];
423 		dst->addr32[3] = src->addr32[3];
424 		break;
425 	}
426 }
427 #endif /* INET6 */
428 
429 void
pf_init_threshold(struct pf_threshold * threshold,u_int32_t limit,u_int32_t seconds)430 pf_init_threshold(struct pf_threshold *threshold,
431     u_int32_t limit, u_int32_t seconds)
432 {
433 	threshold->limit = limit * PF_THRESHOLD_MULT;
434 	threshold->seconds = seconds;
435 	threshold->count = 0;
436 	threshold->last = time_second;
437 }
438 
439 void
pf_add_threshold(struct pf_threshold * threshold)440 pf_add_threshold(struct pf_threshold *threshold)
441 {
442 	u_int32_t t = time_second, diff = t - threshold->last;
443 
444 	if (diff >= threshold->seconds)
445 		threshold->count = 0;
446 	else
447 		threshold->count -= threshold->count * diff /
448 		    threshold->seconds;
449 	threshold->count += PF_THRESHOLD_MULT;
450 	threshold->last = t;
451 }
452 
453 int
pf_check_threshold(struct pf_threshold * threshold)454 pf_check_threshold(struct pf_threshold *threshold)
455 {
456 	return (threshold->count > threshold->limit);
457 }
458 
459 int
pf_src_connlimit(struct pf_state * state)460 pf_src_connlimit(struct pf_state *state)
461 {
462 	int bad = 0;
463 	int cpu = mycpu->gd_cpuid;
464 
465 	atomic_add_int(&state->src_node->conn, 1);
466 	state->src.tcp_est = 1;
467 	pf_add_threshold(&state->src_node->conn_rate);
468 
469 	if (state->rule.ptr->max_src_conn &&
470 	    state->rule.ptr->max_src_conn <
471 	    state->src_node->conn) {
472 		PF_INC_LCOUNTER(LCNT_SRCCONN);
473 		bad++;
474 	}
475 
476 	if (state->rule.ptr->max_src_conn_rate.limit &&
477 	    pf_check_threshold(&state->src_node->conn_rate)) {
478 		PF_INC_LCOUNTER(LCNT_SRCCONNRATE);
479 		bad++;
480 	}
481 
482 	if (!bad)
483 		return 0;
484 
485 	if (state->rule.ptr->overload_tbl) {
486 		struct pfr_addr p;
487 		u_int32_t	killed = 0;
488 
489 		PF_INC_LCOUNTER(LCNT_OVERLOAD_TABLE);
490 		if (pf_status.debug >= PF_DEBUG_MISC) {
491 			kprintf("pf_src_connlimit: blocking address ");
492 			pf_print_host(&state->src_node->addr, 0,
493 			    state->key[PF_SK_WIRE]->af);
494 		}
495 
496 		bzero(&p, sizeof(p));
497 		p.pfra_af = state->key[PF_SK_WIRE]->af;
498 		switch (state->key[PF_SK_WIRE]->af) {
499 #ifdef INET
500 		case AF_INET:
501 			p.pfra_net = 32;
502 			p.pfra_ip4addr = state->src_node->addr.v4;
503 			break;
504 #endif /* INET */
505 #ifdef INET6
506 		case AF_INET6:
507 			p.pfra_net = 128;
508 			p.pfra_ip6addr = state->src_node->addr.v6;
509 			break;
510 #endif /* INET6 */
511 		}
512 
513 		pfr_insert_kentry(state->rule.ptr->overload_tbl,
514 		    &p, time_second);
515 
516 		/* kill existing states if that's required. */
517 		if (state->rule.ptr->flush) {
518 			struct pf_state_key *sk;
519 			struct pf_state *st;
520 
521 			PF_INC_LCOUNTER(LCNT_OVERLOAD_FLUSH);
522 			RB_FOREACH(st, pf_state_tree_id, &tree_id[cpu]) {
523 				sk = st->key[PF_SK_WIRE];
524 				/*
525 				 * Kill states from this source.  (Only those
526 				 * from the same rule if PF_FLUSH_GLOBAL is not
527 				 * set).  (Only on current cpu).
528 				 */
529 				if (sk->af ==
530 				    state->key[PF_SK_WIRE]->af &&
531 				    ((state->direction == PF_OUT &&
532 				    PF_AEQ(&state->src_node->addr,
533 					&sk->addr[0], sk->af)) ||
534 				    (state->direction == PF_IN &&
535 				    PF_AEQ(&state->src_node->addr,
536 					&sk->addr[1], sk->af))) &&
537 				    (state->rule.ptr->flush &
538 				    PF_FLUSH_GLOBAL ||
539 				    state->rule.ptr == st->rule.ptr)) {
540 					st->timeout = PFTM_PURGE;
541 					st->src.state = st->dst.state =
542 					    TCPS_CLOSED;
543 					killed++;
544 				}
545 			}
546 			if (pf_status.debug >= PF_DEBUG_MISC)
547 				kprintf(", %u states killed", killed);
548 		}
549 		if (pf_status.debug >= PF_DEBUG_MISC)
550 			kprintf("\n");
551 	}
552 
553 	/* kill this state */
554 	state->timeout = PFTM_PURGE;
555 	state->src.state = state->dst.state = TCPS_CLOSED;
556 
557 	return 1;
558 }
559 
560 int
pf_insert_src_node(struct pf_src_node ** sn,struct pf_rule * rule,struct pf_addr * src,sa_family_t af)561 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
562     struct pf_addr *src, sa_family_t af)
563 {
564 	struct pf_src_node	k;
565 	int cpu = mycpu->gd_cpuid;
566 
567 	bzero(&k, sizeof(k));	/* avoid gcc warnings */
568 	if (*sn == NULL) {
569 		k.af = af;
570 		PF_ACPY(&k.addr, src, af);
571 		if (rule->rule_flag & PFRULE_RULESRCTRACK ||
572 		    rule->rpool.opts & PF_POOL_STICKYADDR)
573 			k.rule.ptr = rule;
574 		else
575 			k.rule.ptr = NULL;
576 		PF_INC_SCOUNTER(SCNT_SRC_NODE_SEARCH);
577 		*sn = RB_FIND(pf_src_tree, &tree_src_tracking[cpu], &k);
578 	}
579 	if (*sn == NULL) {
580 		if (!rule->max_src_nodes ||
581 		    rule->src_nodes < rule->max_src_nodes)
582 			(*sn) = kmalloc(sizeof(struct pf_src_node),
583 					M_PFSRCTREEPL, M_NOWAIT|M_ZERO);
584 		else
585 			PF_INC_LCOUNTER(LCNT_SRCNODES);
586 		if ((*sn) == NULL)
587 			return (-1);
588 
589 		pf_init_threshold(&(*sn)->conn_rate,
590 		    rule->max_src_conn_rate.limit,
591 		    rule->max_src_conn_rate.seconds);
592 
593 		(*sn)->af = af;
594 		if (rule->rule_flag & PFRULE_RULESRCTRACK ||
595 		    rule->rpool.opts & PF_POOL_STICKYADDR)
596 			(*sn)->rule.ptr = rule;
597 		else
598 			(*sn)->rule.ptr = NULL;
599 		PF_ACPY(&(*sn)->addr, src, af);
600 		if (RB_INSERT(pf_src_tree,
601 		    &tree_src_tracking[cpu], *sn) != NULL) {
602 			if (pf_status.debug >= PF_DEBUG_MISC) {
603 				kprintf("pf: src_tree insert failed: ");
604 				pf_print_host(&(*sn)->addr, 0, af);
605 				kprintf("\n");
606 			}
607 			kfree(*sn, M_PFSRCTREEPL);
608 			return (-1);
609 		}
610 
611 		/*
612 		 * Atomic op required to increment src_nodes in the rule
613 		 * because we hold a shared token here (decrements will use
614 		 * an exclusive token).
615 		 */
616 		(*sn)->creation = time_second;
617 		(*sn)->ruletype = rule->action;
618 		if ((*sn)->rule.ptr != NULL)
619 			atomic_add_int(&(*sn)->rule.ptr->src_nodes, 1);
620 		PF_INC_SCOUNTER(SCNT_SRC_NODE_INSERT);
621 		atomic_add_int(&pf_status.src_nodes, 1);
622 	} else {
623 		if (rule->max_src_states &&
624 		    (*sn)->states >= rule->max_src_states) {
625 			PF_INC_LCOUNTER(LCNT_SRCSTATES);
626 			return (-1);
627 		}
628 	}
629 	return (0);
630 }
631 
632 /*
633  * state table (indexed by the pf_state_key structure), normal RBTREE
634  * comparison.
635  */
636 static __inline int
pf_state_compare_key(struct pf_state_key * a,struct pf_state_key * b)637 pf_state_compare_key(struct pf_state_key *a, struct pf_state_key *b)
638 {
639 	int	diff;
640 
641 	if ((diff = a->proto - b->proto) != 0)
642 		return (diff);
643 	if ((diff = a->af - b->af) != 0)
644 		return (diff);
645 	switch (a->af) {
646 #ifdef INET
647 	case AF_INET:
648 		if (a->addr[0].addr32[0] > b->addr[0].addr32[0])
649 			return (1);
650 		if (a->addr[0].addr32[0] < b->addr[0].addr32[0])
651 			return (-1);
652 		if (a->addr[1].addr32[0] > b->addr[1].addr32[0])
653 			return (1);
654 		if (a->addr[1].addr32[0] < b->addr[1].addr32[0])
655 			return (-1);
656 		break;
657 #endif /* INET */
658 #ifdef INET6
659 	case AF_INET6:
660 		if (a->addr[0].addr32[3] > b->addr[0].addr32[3])
661 			return (1);
662 		if (a->addr[0].addr32[3] < b->addr[0].addr32[3])
663 			return (-1);
664 		if (a->addr[1].addr32[3] > b->addr[1].addr32[3])
665 			return (1);
666 		if (a->addr[1].addr32[3] < b->addr[1].addr32[3])
667 			return (-1);
668 		if (a->addr[0].addr32[2] > b->addr[0].addr32[2])
669 			return (1);
670 		if (a->addr[0].addr32[2] < b->addr[0].addr32[2])
671 			return (-1);
672 		if (a->addr[1].addr32[2] > b->addr[1].addr32[2])
673 			return (1);
674 		if (a->addr[1].addr32[2] < b->addr[1].addr32[2])
675 			return (-1);
676 		if (a->addr[0].addr32[1] > b->addr[0].addr32[1])
677 			return (1);
678 		if (a->addr[0].addr32[1] < b->addr[0].addr32[1])
679 			return (-1);
680 		if (a->addr[1].addr32[1] > b->addr[1].addr32[1])
681 			return (1);
682 		if (a->addr[1].addr32[1] < b->addr[1].addr32[1])
683 			return (-1);
684 		if (a->addr[0].addr32[0] > b->addr[0].addr32[0])
685 			return (1);
686 		if (a->addr[0].addr32[0] < b->addr[0].addr32[0])
687 			return (-1);
688 		if (a->addr[1].addr32[0] > b->addr[1].addr32[0])
689 			return (1);
690 		if (a->addr[1].addr32[0] < b->addr[1].addr32[0])
691 			return (-1);
692 		break;
693 #endif /* INET6 */
694 	}
695 
696 	if ((diff = a->port[0] - b->port[0]) != 0)
697 		return (diff);
698 	if ((diff = a->port[1] - b->port[1]) != 0)
699 		return (diff);
700 
701 	return (0);
702 }
703 
704 /*
705  * Used for RB_FIND only, compare in the reverse direction.  The
706  * element to be reversed is always (a), since we obviously can't
707  * reverse the state tree depicted by (b).
708  */
709 static __inline int
pf_state_compare_rkey(struct pf_state_key * a,struct pf_state_key * b)710 pf_state_compare_rkey(struct pf_state_key *a, struct pf_state_key *b)
711 {
712 	int	diff;
713 
714 	if ((diff = a->proto - b->proto) != 0)
715 		return (diff);
716 	if ((diff = a->af - b->af) != 0)
717 		return (diff);
718 	switch (a->af) {
719 #ifdef INET
720 	case AF_INET:
721 		if (a->addr[1].addr32[0] > b->addr[0].addr32[0])
722 			return (1);
723 		if (a->addr[1].addr32[0] < b->addr[0].addr32[0])
724 			return (-1);
725 		if (a->addr[0].addr32[0] > b->addr[1].addr32[0])
726 			return (1);
727 		if (a->addr[0].addr32[0] < b->addr[1].addr32[0])
728 			return (-1);
729 		break;
730 #endif /* INET */
731 #ifdef INET6
732 	case AF_INET6:
733 		if (a->addr[1].addr32[3] > b->addr[0].addr32[3])
734 			return (1);
735 		if (a->addr[1].addr32[3] < b->addr[0].addr32[3])
736 			return (-1);
737 		if (a->addr[0].addr32[3] > b->addr[1].addr32[3])
738 			return (1);
739 		if (a->addr[0].addr32[3] < b->addr[1].addr32[3])
740 			return (-1);
741 		if (a->addr[1].addr32[2] > b->addr[0].addr32[2])
742 			return (1);
743 		if (a->addr[1].addr32[2] < b->addr[0].addr32[2])
744 			return (-1);
745 		if (a->addr[0].addr32[2] > b->addr[1].addr32[2])
746 			return (1);
747 		if (a->addr[0].addr32[2] < b->addr[1].addr32[2])
748 			return (-1);
749 		if (a->addr[1].addr32[1] > b->addr[0].addr32[1])
750 			return (1);
751 		if (a->addr[1].addr32[1] < b->addr[0].addr32[1])
752 			return (-1);
753 		if (a->addr[0].addr32[1] > b->addr[1].addr32[1])
754 			return (1);
755 		if (a->addr[0].addr32[1] < b->addr[1].addr32[1])
756 			return (-1);
757 		if (a->addr[1].addr32[0] > b->addr[0].addr32[0])
758 			return (1);
759 		if (a->addr[1].addr32[0] < b->addr[0].addr32[0])
760 			return (-1);
761 		if (a->addr[0].addr32[0] > b->addr[1].addr32[0])
762 			return (1);
763 		if (a->addr[0].addr32[0] < b->addr[1].addr32[0])
764 			return (-1);
765 		break;
766 #endif /* INET6 */
767 	}
768 
769 	if ((diff = a->port[1] - b->port[0]) != 0)
770 		return (diff);
771 	if ((diff = a->port[0] - b->port[1]) != 0)
772 		return (diff);
773 
774 	return (0);
775 }
776 
777 static __inline int
pf_state_compare_id(struct pf_state * a,struct pf_state * b)778 pf_state_compare_id(struct pf_state *a, struct pf_state *b)
779 {
780 	if (a->id > b->id)
781 		return (1);
782 	if (a->id < b->id)
783 		return (-1);
784 	if (a->creatorid > b->creatorid)
785 		return (1);
786 	if (a->creatorid < b->creatorid)
787 		return (-1);
788 
789 	return (0);
790 }
791 
792 int
pf_state_key_attach(struct pf_state_key * sk,struct pf_state * s,int idx)793 pf_state_key_attach(struct pf_state_key *sk, struct pf_state *s, int idx)
794 {
795 	struct pf_state_item	*si;
796 	struct pf_state_key     *cur;
797 	int cpu;
798 	int error;
799 
800 	/*
801 	 * PFSTATE_STACK_GLOBAL is set when the state might not hash to the
802 	 * current cpu.  The keys are managed on the global statetbl tree
803 	 * for this case.  Only translations (RDR, NAT) can cause this.
804 	 *
805 	 * When this flag is not set we must still check the global statetbl
806 	 * for a collision, and if we find one we set the HALF_DUPLEX flag
807 	 * in the state.
808 	 */
809 	if (s->state_flags & PFSTATE_STACK_GLOBAL) {
810 		cpu = ncpus;
811 		lockmgr(&pf_global_statetbl_lock, LK_EXCLUSIVE);
812 	} else {
813 		cpu = mycpu->gd_cpuid;
814 		lockmgr(&pf_global_statetbl_lock, LK_SHARED);
815 	}
816 	KKASSERT(s->key[idx] == NULL);	/* XXX handle this? */
817 
818 	if (pf_status.debug >= PF_DEBUG_MISC) {
819 		kprintf("state_key attach cpu %d (%08x:%d) %s (%08x:%d)\n",
820 			cpu,
821 			ntohl(sk->addr[0].addr32[0]), ntohs(sk->port[0]),
822 			(idx == PF_SK_WIRE ? "->" : "<-"),
823 			ntohl(sk->addr[1].addr32[0]), ntohs(sk->port[1]));
824 	}
825 
826 	/*
827 	 * Check whether (e.g.) a PASS rule being put on a per-cpu tree
828 	 * collides with a translation rule on the global tree.  This is
829 	 * NOT an error.  We *WANT* to establish state for this case so the
830 	 * packet path is short-cutted and doesn't need to scan the ruleset
831 	 * on every packet.  But the established state will only see one
832 	 * side of a two-way packet conversation.  To prevent this from
833 	 * causing problems (e.g. generating a RST), we force PFSTATE_SLOPPY
834 	 * to be set on the established state.
835 	 *
836 	 * A collision against RDR state can only occur with a PASS IN in the
837 	 * opposite direction or a PASS OUT in the forwards direction.  This
838 	 * is because RDRs are processed on the input side.
839 	 *
840 	 * A collision against NAT state can only occur with a PASS IN in the
841 	 * forwards direction or a PASS OUT in the opposite direction.  This
842 	 * is because NATs are processed on the output side.
843 	 *
844 	 * In both situations we need to do a reverse addr/port test because
845 	 * the PASS IN or PASS OUT only establishes if it doesn't match the
846 	 * established RDR state in the forwards direction.  The direction
847 	 * flag has to be ignored (it will be one way for a PASS IN and the
848 	 * other way for a PASS OUT).
849 	 *
850 	 * pf_global_statetbl_lock will be locked shared when testing and
851 	 * not entering into the global state table.
852 	 */
853 	if (cpu != ncpus &&
854 	    (cur = RB_FIND(pf_state_rtree,
855 			   (struct pf_state_rtree *)&pf_statetbl[ncpus],
856 			   sk)) != NULL) {
857 		TAILQ_FOREACH(si, &cur->states, entry) {
858 			/*
859 			 * NOTE: We must ignore direction mismatches.
860 			 */
861 			if (si->s->kif == s->kif) {
862 				s->state_flags |= PFSTATE_HALF_DUPLEX |
863 						  PFSTATE_SLOPPY;
864 				if (pf_status.debug >= PF_DEBUG_MISC) {
865 					kprintf(
866 					    "pf: %s key attach collision "
867 					    "on %s: ",
868 					    (idx == PF_SK_WIRE) ?
869 					    "wire" : "stack",
870 					    s->kif->pfik_name);
871 					pf_print_state_parts(s,
872 					    (idx == PF_SK_WIRE) ? sk : NULL,
873 					    (idx == PF_SK_STACK) ? sk : NULL);
874 					kprintf("\n");
875 				}
876 				break;
877 			}
878 		}
879 	}
880 
881 	/*
882 	 * Enter into either the per-cpu or the global state table.
883 	 *
884 	 * pf_global_statetbl_lock will be locked exclusively when entering
885 	 * into the global state table.
886 	 */
887 	if ((cur = RB_INSERT(pf_state_tree, &pf_statetbl[cpu], sk)) != NULL) {
888 		/* key exists. check for same kif, if none, add to key */
889 		TAILQ_FOREACH(si, &cur->states, entry) {
890 			if (si->s->kif == s->kif &&
891 			    si->s->direction == s->direction) {
892 				if (pf_status.debug >= PF_DEBUG_MISC) {
893 					kprintf(
894 					    "pf: %s key attach failed on %s: ",
895 					    (idx == PF_SK_WIRE) ?
896 					    "wire" : "stack",
897 					    s->kif->pfik_name);
898 					pf_print_state_parts(s,
899 					    (idx == PF_SK_WIRE) ? sk : NULL,
900 					    (idx == PF_SK_STACK) ? sk : NULL);
901 					kprintf("\n");
902 				}
903 				kfree(sk, M_PFSTATEKEYPL);
904 				error = -1;
905 				goto failed;	/* collision! */
906 			}
907 		}
908 		kfree(sk, M_PFSTATEKEYPL);
909 
910 		s->key[idx] = cur;
911 	} else {
912 		s->key[idx] = sk;
913 	}
914 
915 	if ((si = kmalloc(sizeof(struct pf_state_item),
916 			  M_PFSTATEITEMPL, M_NOWAIT)) == NULL) {
917 		pf_state_key_detach(s, idx);
918 		error = -1;
919 		goto failed;	/* collision! */
920 	}
921 	si->s = s;
922 
923 	/* list is sorted, if-bound states before floating */
924 	if (s->kif == pfi_all)
925 		TAILQ_INSERT_TAIL(&s->key[idx]->states, si, entry);
926 	else
927 		TAILQ_INSERT_HEAD(&s->key[idx]->states, si, entry);
928 
929 	error = 0;
930 failed:
931 	lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
932 	return error;
933 }
934 
935 /*
936  * NOTE: Can only be called indirectly via the purge thread with pf_token
937  *	 exclusively locked.
938  */
939 void
pf_detach_state(struct pf_state * s)940 pf_detach_state(struct pf_state *s)
941 {
942 	if (s->key[PF_SK_WIRE] == s->key[PF_SK_STACK])
943 		s->key[PF_SK_WIRE] = NULL;
944 
945 	if (s->key[PF_SK_STACK] != NULL)
946 		pf_state_key_detach(s, PF_SK_STACK);
947 
948 	if (s->key[PF_SK_WIRE] != NULL)
949 		pf_state_key_detach(s, PF_SK_WIRE);
950 }
951 
952 /*
953  * NOTE: Can only be called indirectly via the purge thread with pf_token
954  *	 exclusively locked.
955  */
956 void
pf_state_key_detach(struct pf_state * s,int idx)957 pf_state_key_detach(struct pf_state *s, int idx)
958 {
959 	struct pf_state_item	*si;
960 	int cpu;
961 
962 	/*
963 	 * PFSTATE_STACK_GLOBAL is set for translations when the translated
964 	 * address/port is not localized to the same cpu that the untranslated
965 	 * address/port is on.  The wire pf_state_key is managed on the global
966 	 * statetbl tree for this case.
967 	 */
968 	if (s->state_flags & PFSTATE_STACK_GLOBAL) {
969 		cpu = ncpus;
970 		lockmgr(&pf_global_statetbl_lock, LK_EXCLUSIVE);
971 	} else {
972 		cpu = mycpu->gd_cpuid;
973 	}
974 
975 	si = TAILQ_FIRST(&s->key[idx]->states);
976 	while (si && si->s != s)
977 		si = TAILQ_NEXT(si, entry);
978 
979 	if (si) {
980 		TAILQ_REMOVE(&s->key[idx]->states, si, entry);
981 		kfree(si, M_PFSTATEITEMPL);
982 	}
983 
984 	if (TAILQ_EMPTY(&s->key[idx]->states)) {
985 		RB_REMOVE(pf_state_tree, &pf_statetbl[cpu], s->key[idx]);
986 		if (s->key[idx]->reverse)
987 			s->key[idx]->reverse->reverse = NULL;
988 		if (s->key[idx]->inp)
989 			s->key[idx]->inp->inp_pf_sk = NULL;
990 		kfree(s->key[idx], M_PFSTATEKEYPL);
991 	}
992 	s->key[idx] = NULL;
993 
994 	if (s->state_flags & PFSTATE_STACK_GLOBAL)
995 		lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
996 }
997 
998 struct pf_state_key *
pf_alloc_state_key(int pool_flags)999 pf_alloc_state_key(int pool_flags)
1000 {
1001 	struct pf_state_key	*sk;
1002 
1003 	sk = kmalloc(sizeof(struct pf_state_key), M_PFSTATEKEYPL, pool_flags);
1004 	if (sk) {
1005 		TAILQ_INIT(&sk->states);
1006 	}
1007 	return (sk);
1008 }
1009 
1010 int
pf_state_key_setup(struct pf_pdesc * pd,struct pf_rule * nr,struct pf_state_key ** skw,struct pf_state_key ** sks,struct pf_state_key ** skp,struct pf_state_key ** nkp,struct pf_addr * saddr,struct pf_addr * daddr,u_int16_t sport,u_int16_t dport)1011 pf_state_key_setup(struct pf_pdesc *pd, struct pf_rule *nr,
1012 	struct pf_state_key **skw, struct pf_state_key **sks,
1013 	struct pf_state_key **skp, struct pf_state_key **nkp,
1014 	struct pf_addr *saddr, struct pf_addr *daddr,
1015 	u_int16_t sport, u_int16_t dport)
1016 {
1017 	KKASSERT((*skp == NULL && *nkp == NULL));
1018 
1019 	if ((*skp = pf_alloc_state_key(M_NOWAIT | M_ZERO)) == NULL)
1020 		return (ENOMEM);
1021 
1022 	PF_ACPY(&(*skp)->addr[pd->sidx], saddr, pd->af);
1023 	PF_ACPY(&(*skp)->addr[pd->didx], daddr, pd->af);
1024 	(*skp)->port[pd->sidx] = sport;
1025 	(*skp)->port[pd->didx] = dport;
1026 	(*skp)->proto = pd->proto;
1027 	(*skp)->af = pd->af;
1028 
1029 	if (nr != NULL) {
1030 		if ((*nkp = pf_alloc_state_key(M_NOWAIT | M_ZERO)) == NULL)
1031 			return (ENOMEM); /* caller must handle cleanup */
1032 
1033 		/* XXX maybe just bcopy and TAILQ_INIT(&(*nkp)->states) */
1034 		PF_ACPY(&(*nkp)->addr[0], &(*skp)->addr[0], pd->af);
1035 		PF_ACPY(&(*nkp)->addr[1], &(*skp)->addr[1], pd->af);
1036 		(*nkp)->port[0] = (*skp)->port[0];
1037 		(*nkp)->port[1] = (*skp)->port[1];
1038 		(*nkp)->proto = pd->proto;
1039 		(*nkp)->af = pd->af;
1040 	} else {
1041 		*nkp = *skp;
1042 	}
1043 
1044 	if (pd->dir == PF_IN) {
1045 		*skw = *skp;
1046 		*sks = *nkp;
1047 	} else {
1048 		*sks = *skp;
1049 		*skw = *nkp;
1050 	}
1051 	return (0);
1052 }
1053 
1054 /*
1055  * Insert pf_state with one or two state keys (allowing a reverse path lookup
1056  * which is used by NAT).  In the NAT case skw is the initiator (?) and
1057  * sks is the target.
1058  */
1059 int
pf_state_insert(struct pfi_kif * kif,struct pf_state_key * skw,struct pf_state_key * sks,struct pf_state * s)1060 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1061 		struct pf_state_key *sks, struct pf_state *s)
1062 {
1063 	int cpu = mycpu->gd_cpuid;
1064 
1065 	s->kif = kif;
1066 	s->cpuid = cpu;
1067 
1068 	if (skw == sks) {
1069 		if (pf_state_key_attach(skw, s, PF_SK_WIRE))
1070 			return (-1);
1071 		s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1072 	} else {
1073 		/*
1074 		skw->reverse = sks;
1075 		sks->reverse = skw;
1076 		*/
1077 		if (pf_state_key_attach(skw, s, PF_SK_WIRE)) {
1078 			kfree(sks, M_PFSTATEKEYPL);
1079 			return (-1);
1080 		}
1081 		if (pf_state_key_attach(sks, s, PF_SK_STACK)) {
1082 			pf_state_key_detach(s, PF_SK_WIRE);
1083 			return (-1);
1084 		}
1085 	}
1086 
1087 	if (s->id == 0 && s->creatorid == 0) {
1088 		u_int64_t sid;
1089 
1090 		sid = atomic_fetchadd_long(&pf_status.stateid, 1);
1091 		s->id = htobe64(sid);
1092 		s->creatorid = pf_status.hostid;
1093 	}
1094 
1095 	/*
1096 	 * Calculate hash code for altq
1097 	 */
1098 	s->hash = crc32(s->key[PF_SK_WIRE], PF_STATE_KEY_HASH_LENGTH);
1099 
1100 	if (RB_INSERT(pf_state_tree_id, &tree_id[cpu], s) != NULL) {
1101 		if (pf_status.debug >= PF_DEBUG_MISC) {
1102 			kprintf("pf: state insert failed: "
1103 			    "id: %016jx creatorid: %08x",
1104 			      (uintmax_t)be64toh(s->id), ntohl(s->creatorid));
1105 			if (s->sync_flags & PFSTATE_FROMSYNC)
1106 				kprintf(" (from sync)");
1107 			kprintf("\n");
1108 		}
1109 		pf_detach_state(s);
1110 		return (-1);
1111 	}
1112 	TAILQ_INSERT_TAIL(&state_list[cpu], s, entry_list);
1113 	PF_INC_FCOUNTER(FCNT_STATE_INSERT);
1114 	atomic_add_int(&pf_status.states, 1);
1115 	pfi_kif_ref(kif, PFI_KIF_REF_STATE);
1116 	pfsync_insert_state(s);
1117 	return (0);
1118 }
1119 
1120 struct pf_state *
pf_find_state_byid(struct pf_state_cmp * key)1121 pf_find_state_byid(struct pf_state_cmp *key)
1122 {
1123 	int cpu = mycpu->gd_cpuid;
1124 
1125 	PF_INC_FCOUNTER(FCNT_STATE_SEARCH);
1126 
1127 	return (RB_FIND(pf_state_tree_id, &tree_id[cpu],
1128 			(struct pf_state *)key));
1129 }
1130 
1131 /*
1132  * WARNING! May return a state structure that was localized to another cpu,
1133  *	    destruction is typically protected by the callers pf_token.
1134  *	    The element can only be destroyed
1135  */
1136 struct pf_state *
pf_find_state(struct pfi_kif * kif,struct pf_state_key_cmp * key,u_int dir,struct mbuf * m)1137 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir,
1138 	      struct mbuf *m)
1139 {
1140 	struct pf_state_key	*skey = (void *)key;
1141 	struct pf_state_key	*sk;
1142 	struct pf_state_item	*si;
1143 	struct pf_state *s;
1144 	int cpu = mycpu->gd_cpuid;
1145 	int globalstl = 0;
1146 
1147 	PF_INC_FCOUNTER(FCNT_STATE_SEARCH);
1148 
1149 	if (dir == PF_OUT && m->m_pkthdr.pf.statekey &&
1150 	    ((struct pf_state_key *)m->m_pkthdr.pf.statekey)->reverse) {
1151 		sk = ((struct pf_state_key *)m->m_pkthdr.pf.statekey)->reverse;
1152 	} else {
1153 		sk = RB_FIND(pf_state_tree, &pf_statetbl[cpu], skey);
1154 		if (sk == NULL) {
1155 			lockmgr(&pf_global_statetbl_lock, LK_SHARED);
1156 			sk = RB_FIND(pf_state_tree, &pf_statetbl[ncpus], skey);
1157 			if (sk == NULL) {
1158 				lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
1159 				return (NULL);
1160 			}
1161 			globalstl = 1;
1162 		}
1163 		if (dir == PF_OUT && m->m_pkthdr.pf.statekey) {
1164 			((struct pf_state_key *)
1165 			    m->m_pkthdr.pf.statekey)->reverse = sk;
1166 			sk->reverse = m->m_pkthdr.pf.statekey;
1167 		}
1168 	}
1169 	if (dir == PF_OUT)
1170 		m->m_pkthdr.pf.statekey = NULL;
1171 
1172 	/* list is sorted, if-bound states before floating ones */
1173 	TAILQ_FOREACH(si, &sk->states, entry) {
1174 		if ((si->s->kif == pfi_all || si->s->kif == kif) &&
1175 		    sk == (dir == PF_IN ? si->s->key[PF_SK_WIRE] :
1176 					  si->s->key[PF_SK_STACK])) {
1177 			break;
1178 		}
1179 	}
1180 
1181 	/*
1182 	 * Extract state before potentially releasing the global statetbl
1183 	 * lock.  Ignore the state if the create is still in-progress as
1184 	 * it can be deleted out from under us by the owning localized cpu.
1185 	 * However, if CREATEINPROG is not set, state can only be deleted
1186 	 * by the purge thread which we are protected from via our shared
1187 	 * pf_token.
1188 	 */
1189 	if (si) {
1190 		s = si->s;
1191 		if (s && (s->state_flags & PFSTATE_CREATEINPROG))
1192 			s = NULL;
1193 	} else {
1194 		s = NULL;
1195 	}
1196 	if (globalstl)
1197 		lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
1198 	return s;
1199 }
1200 
1201 /*
1202  * WARNING! May return a state structure that was localized to another cpu,
1203  *	    destruction is typically protected by the callers pf_token.
1204  */
1205 struct pf_state *
pf_find_state_all(struct pf_state_key_cmp * key,u_int dir,int * more)1206 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1207 {
1208 	struct pf_state_key	*skey = (void *)key;
1209 	struct pf_state_key	*sk;
1210 	struct pf_state_item	*si, *ret = NULL;
1211 	struct pf_state		*s;
1212 	int cpu = mycpu->gd_cpuid;
1213 	int globalstl = 0;
1214 
1215 	PF_INC_FCOUNTER(FCNT_STATE_SEARCH);
1216 
1217 	sk = RB_FIND(pf_state_tree, &pf_statetbl[cpu], skey);
1218 	if (sk == NULL) {
1219 		lockmgr(&pf_global_statetbl_lock, LK_SHARED);
1220 		sk = RB_FIND(pf_state_tree, &pf_statetbl[ncpus], skey);
1221 		globalstl = 1;
1222 	}
1223 	if (sk != NULL) {
1224 		TAILQ_FOREACH(si, &sk->states, entry)
1225 			if (dir == PF_INOUT ||
1226 			    (sk == (dir == PF_IN ? si->s->key[PF_SK_WIRE] :
1227 			    si->s->key[PF_SK_STACK]))) {
1228 				if (more == NULL) {
1229 					ret = si;
1230 					break;
1231 				}
1232 				if (ret)
1233 					(*more)++;
1234 				else
1235 					ret = si;
1236 			}
1237 	}
1238 
1239 	/*
1240 	 * Extract state before potentially releasing the global statetbl
1241 	 * lock.  Ignore the state if the create is still in-progress as
1242 	 * it can be deleted out from under us by the owning localized cpu.
1243 	 * However, if CREATEINPROG is not set, state can only be deleted
1244 	 * by the purge thread which we are protected from via our shared
1245 	 * pf_token.
1246 	 */
1247 	if (ret) {
1248 		s = ret->s;
1249 		if (s && (s->state_flags & PFSTATE_CREATEINPROG))
1250 			s = NULL;
1251 	} else {
1252 		s = NULL;
1253 	}
1254 	if (globalstl)
1255 		lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
1256 	return s;
1257 }
1258 
1259 /* END state table stuff */
1260 
1261 void
pf_purge_thread(void * v)1262 pf_purge_thread(void *v)
1263 {
1264 	globaldata_t save_gd = mycpu;
1265 	int nloops = 0;
1266 	int locked = 0;
1267 	int nn;
1268 	int endingit;
1269 
1270 	for (;;) {
1271 		tsleep(pf_purge_thread, PWAIT, "pftm", 1 * hz);
1272 
1273 		endingit = pf_end_threads;
1274 
1275 		for (nn = 0; nn < ncpus; ++nn) {
1276 			lwkt_setcpu_self(globaldata_find(nn));
1277 
1278 			lwkt_gettoken(&pf_token);
1279 			lockmgr(&pf_consistency_lock, LK_EXCLUSIVE);
1280 			crit_enter();
1281 
1282 			/*
1283 			 * process a fraction of the state table every second
1284 			 */
1285 			if(!pf_purge_expired_states(
1286 				1 + (pf_status.states /
1287 				     pf_default_rule.timeout[
1288 					PFTM_INTERVAL]), 0)) {
1289 				pf_purge_expired_states(
1290 					1 + (pf_status.states /
1291 					     pf_default_rule.timeout[
1292 						PFTM_INTERVAL]), 1);
1293 			}
1294 
1295 			/*
1296 			 * purge other expired types every PFTM_INTERVAL
1297 			 * seconds
1298 			 */
1299 			if (++nloops >=
1300 			    pf_default_rule.timeout[PFTM_INTERVAL]) {
1301 				pf_purge_expired_fragments();
1302 				if (!pf_purge_expired_src_nodes(locked)) {
1303 					pf_purge_expired_src_nodes(1);
1304 				}
1305 				nloops = 0;
1306 			}
1307 
1308 			/*
1309 			 * If terminating the thread, clean everything out
1310 			 * (on all cpus).
1311 			 */
1312 			if (endingit) {
1313 				pf_purge_expired_states(pf_status.states, 0);
1314 				pf_purge_expired_fragments();
1315 				pf_purge_expired_src_nodes(1);
1316 			}
1317 
1318 			crit_exit();
1319 			lockmgr(&pf_consistency_lock, LK_RELEASE);
1320 			lwkt_reltoken(&pf_token);
1321 		}
1322 		lwkt_setcpu_self(save_gd);
1323 		if (endingit)
1324 			break;
1325 	}
1326 
1327 	/*
1328 	 * Thread termination
1329 	 */
1330 	pf_end_threads++;
1331 	wakeup(pf_purge_thread);
1332 	kthread_exit();
1333 }
1334 
1335 u_int32_t
pf_state_expires(const struct pf_state * state)1336 pf_state_expires(const struct pf_state *state)
1337 {
1338 	u_int32_t	timeout;
1339 	u_int32_t	start;
1340 	u_int32_t	end;
1341 	u_int32_t	states;
1342 
1343 	/* handle all PFTM_* > PFTM_MAX here */
1344 	if (state->timeout == PFTM_PURGE)
1345 		return (time_second);
1346 	if (state->timeout == PFTM_UNTIL_PACKET)
1347 		return (0);
1348 	KKASSERT(state->timeout != PFTM_UNLINKED);
1349 	KKASSERT(state->timeout < PFTM_MAX);
1350 	timeout = state->rule.ptr->timeout[state->timeout];
1351 	if (!timeout)
1352 		timeout = pf_default_rule.timeout[state->timeout];
1353 	start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1354 	if (start) {
1355 		end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1356 		states = state->rule.ptr->states_cur;
1357 	} else {
1358 		start = pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1359 		end = pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1360 		states = pf_status.states;
1361 	}
1362 
1363 	/*
1364 	 * If the number of states exceeds allowed values, adaptively
1365 	 * timeout the state more quickly.  This can be very dangerous
1366 	 * to legitimate connections, however, so defray the timeout
1367 	 * based on the packet count.
1368 	 *
1369 	 * Retain from 0-100% based on number of states.
1370 	 *
1371 	 * Recover up to 50% of the lost portion if there was
1372 	 * packet traffic (100 pkts = 50%).
1373 	 */
1374 	if (end && states > start && start < end) {
1375 		u_int32_t n;			/* timeout retention 0-100% */
1376 		u_int64_t pkts;
1377 #if 0
1378 		static struct krate boorate = { .freq = 1 };
1379 #endif
1380 
1381 		/*
1382 		 * Reduce timeout by n% (0-100)
1383 		 */
1384 		n = (states - start) * 100 / (end - start);
1385 		if (n > 100)
1386 			n = 0;
1387 		else
1388 			n = 100 - n;
1389 
1390 		/*
1391 		 * But claw back some of the reduction based on packet
1392 		 * count associated with the state.
1393 		 */
1394 		pkts = state->packets[0] + state->packets[1];
1395 		if (pkts > 100)
1396 			pkts = 100;
1397 #if 0
1398 		krateprintf(&boorate, "timeout %-4u n=%u pkts=%-3lu -> %lu\n",
1399 			timeout, n, pkts, n + (100 - n) * pkts / 200);
1400 #endif
1401 
1402 		n += (100 - n) * pkts / 200;	/* recover by up-to 50% */
1403 		timeout = timeout * n / 100;
1404 
1405 	}
1406 	return (state->expire + timeout);
1407 }
1408 
1409 /*
1410  * (called with exclusive pf_token)
1411  */
1412 int
pf_purge_expired_src_nodes(int waslocked)1413 pf_purge_expired_src_nodes(int waslocked)
1414 {
1415 	struct pf_src_node *cur, *next;
1416 	int locked = waslocked;
1417 	int cpu = mycpu->gd_cpuid;
1418 
1419 	for (cur = RB_MIN(pf_src_tree, &tree_src_tracking[cpu]);
1420 	     cur;
1421 	     cur = next) {
1422 		next = RB_NEXT(pf_src_tree, &tree_src_tracking[cpu], cur);
1423 
1424 		if (cur->states <= 0 && cur->expire <= time_second) {
1425 			 if (!locked) {
1426 				 lockmgr(&pf_consistency_lock, LK_EXCLUSIVE);
1427 			 	 next = RB_NEXT(pf_src_tree,
1428 				     &tree_src_tracking[cpu], cur);
1429 				 locked = 1;
1430 			 }
1431 			 if (cur->rule.ptr != NULL) {
1432 				/*
1433 				 * decrements in rule should be ok, token is
1434 				 * held exclusively in this code path.
1435 				 */
1436 				 atomic_add_int(&cur->rule.ptr->src_nodes, -1);
1437 				 if (cur->rule.ptr->states_cur <= 0 &&
1438 				     cur->rule.ptr->max_src_nodes <= 0)
1439 					 pf_rm_rule(NULL, cur->rule.ptr);
1440 			 }
1441 			 RB_REMOVE(pf_src_tree, &tree_src_tracking[cpu], cur);
1442 			 PF_INC_SCOUNTER(SCNT_SRC_NODE_REMOVALS);
1443 			 atomic_add_int(&pf_status.src_nodes, -1);
1444 			 kfree(cur, M_PFSRCTREEPL);
1445 		}
1446 	}
1447 	if (locked && !waslocked)
1448 		lockmgr(&pf_consistency_lock, LK_RELEASE);
1449 	return(1);
1450 }
1451 
1452 void
pf_src_tree_remove_state(struct pf_state * s)1453 pf_src_tree_remove_state(struct pf_state *s)
1454 {
1455 	u_int32_t timeout;
1456 
1457 	if (s->src_node != NULL) {
1458 		if (s->src.tcp_est)
1459 			atomic_add_int(&s->src_node->conn, -1);
1460 		if (--s->src_node->states <= 0) {
1461 			timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1462 			if (!timeout) {
1463 				timeout =
1464 				    pf_default_rule.timeout[PFTM_SRC_NODE];
1465 			}
1466 			s->src_node->expire = time_second + timeout;
1467 		}
1468 	}
1469 	if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1470 		if (--s->nat_src_node->states <= 0) {
1471 			timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1472 			if (!timeout)
1473 				timeout =
1474 				    pf_default_rule.timeout[PFTM_SRC_NODE];
1475 			s->nat_src_node->expire = time_second + timeout;
1476 		}
1477 	}
1478 	s->src_node = s->nat_src_node = NULL;
1479 }
1480 
1481 /* callers should be at crit_enter() */
1482 void
pf_unlink_state(struct pf_state * cur)1483 pf_unlink_state(struct pf_state *cur)
1484 {
1485 	int cpu = mycpu->gd_cpuid;
1486 
1487 	if (cur->src.state == PF_TCPS_PROXY_DST) {
1488 		/* XXX wire key the right one? */
1489 		pf_send_tcp(cur->rule.ptr, cur->key[PF_SK_WIRE]->af,
1490 		    &cur->key[PF_SK_WIRE]->addr[1],
1491 		    &cur->key[PF_SK_WIRE]->addr[0],
1492 		    cur->key[PF_SK_WIRE]->port[1],
1493 		    cur->key[PF_SK_WIRE]->port[0],
1494 		    cur->src.seqhi, cur->src.seqlo + 1,
1495 		    TH_RST|TH_ACK, 0, 0, 0, 1, cur->tag, NULL, NULL);
1496 	}
1497 	RB_REMOVE(pf_state_tree_id, &tree_id[cpu], cur);
1498 	if (cur->creatorid == pf_status.hostid)
1499 		pfsync_delete_state(cur);
1500 	cur->timeout = PFTM_UNLINKED;
1501 	pf_src_tree_remove_state(cur);
1502 	pf_detach_state(cur);
1503 }
1504 
1505 /*
1506  * callers should be at crit_enter() and hold pf_consistency_lock exclusively.
1507  * pf_token must also be held exclusively.
1508  */
1509 void
pf_free_state(struct pf_state * cur)1510 pf_free_state(struct pf_state *cur)
1511 {
1512 	int cpu = mycpu->gd_cpuid;
1513 
1514 	KKASSERT(cur->cpuid == cpu);
1515 
1516 	if (pfsyncif != NULL &&
1517 	    (pfsyncif->sc_bulk_send_next == cur ||
1518 	    pfsyncif->sc_bulk_terminator == cur))
1519 		return;
1520 	KKASSERT(cur->timeout == PFTM_UNLINKED);
1521 	/*
1522 	 * decrements in rule should be ok, token is
1523 	 * held exclusively in this code path.
1524 	 */
1525 	if (--cur->rule.ptr->states_cur <= 0 &&
1526 	    cur->rule.ptr->src_nodes <= 0)
1527 		pf_rm_rule(NULL, cur->rule.ptr);
1528 	if (cur->nat_rule.ptr != NULL) {
1529 		if (--cur->nat_rule.ptr->states_cur <= 0 &&
1530 			cur->nat_rule.ptr->src_nodes <= 0) {
1531 			pf_rm_rule(NULL, cur->nat_rule.ptr);
1532 		}
1533 	}
1534 	if (cur->anchor.ptr != NULL) {
1535 		if (--cur->anchor.ptr->states_cur <= 0)
1536 			pf_rm_rule(NULL, cur->anchor.ptr);
1537 	}
1538 	pf_normalize_tcp_cleanup(cur);
1539 	pfi_kif_unref(cur->kif, PFI_KIF_REF_STATE);
1540 
1541 	/*
1542 	 * We may be freeing pf_purge_expired_states()'s saved scan entry,
1543 	 * adjust it if necessary.
1544 	 */
1545 	if (purge_cur[cpu] == cur) {
1546 		kprintf("PURGE CONFLICT\n");
1547 		purge_cur[cpu] = TAILQ_NEXT(purge_cur[cpu], entry_list);
1548 	}
1549 	TAILQ_REMOVE(&state_list[cpu], cur, entry_list);
1550 	if (cur->tag)
1551 		pf_tag_unref(cur->tag);
1552 	kfree(cur, M_PFSTATEPL);
1553 	PF_INC_FCOUNTER(FCNT_STATE_REMOVALS);
1554 	atomic_add_int(&pf_status.states, -1);
1555 }
1556 
1557 int
pf_purge_expired_states(u_int32_t maxcheck,int waslocked)1558 pf_purge_expired_states(u_int32_t maxcheck, int waslocked)
1559 {
1560 	struct pf_state		*cur;
1561 	int locked = waslocked;
1562 	int cpu = mycpu->gd_cpuid;
1563 
1564 	while (maxcheck--) {
1565 		/*
1566 		 * Wrap to start of list when we hit the end
1567 		 */
1568 		cur = purge_cur[cpu];
1569 		if (cur == NULL) {
1570 			cur = TAILQ_FIRST(&state_list[cpu]);
1571 			if (cur == NULL)
1572 				break;	/* list empty */
1573 		}
1574 
1575 		/*
1576 		 * Setup next (purge_cur) while we process this one.  If
1577 		 * we block and something else deletes purge_cur,
1578 		 * pf_free_state() will adjust it further ahead.
1579 		 */
1580 		purge_cur[cpu] = TAILQ_NEXT(cur, entry_list);
1581 
1582 		if (cur->timeout == PFTM_UNLINKED) {
1583 			/* free unlinked state */
1584 			if (! locked) {
1585 				lockmgr(&pf_consistency_lock, LK_EXCLUSIVE);
1586 				locked = 1;
1587 			}
1588 			pf_free_state(cur);
1589 		} else if (pf_state_expires(cur) <= time_second) {
1590 			/* unlink and free expired state */
1591 			pf_unlink_state(cur);
1592 			if (! locked) {
1593 				if (!lockmgr(&pf_consistency_lock, LK_EXCLUSIVE))
1594 					return (0);
1595 				locked = 1;
1596 			}
1597 			pf_free_state(cur);
1598 		}
1599 	}
1600 
1601 	if (locked)
1602 		lockmgr(&pf_consistency_lock, LK_RELEASE);
1603 	return (1);
1604 }
1605 
1606 int
pf_tbladdr_setup(struct pf_ruleset * rs,struct pf_addr_wrap * aw)1607 pf_tbladdr_setup(struct pf_ruleset *rs, struct pf_addr_wrap *aw)
1608 {
1609 	if (aw->type != PF_ADDR_TABLE)
1610 		return (0);
1611 	if ((aw->p.tbl = pfr_attach_table(rs, aw->v.tblname)) == NULL)
1612 		return (1);
1613 	return (0);
1614 }
1615 
1616 void
pf_tbladdr_remove(struct pf_addr_wrap * aw)1617 pf_tbladdr_remove(struct pf_addr_wrap *aw)
1618 {
1619 	if (aw->type != PF_ADDR_TABLE || aw->p.tbl == NULL)
1620 		return;
1621 	pfr_detach_table(aw->p.tbl);
1622 	aw->p.tbl = NULL;
1623 }
1624 
1625 void
pf_tbladdr_copyout(struct pf_addr_wrap * aw)1626 pf_tbladdr_copyout(struct pf_addr_wrap *aw)
1627 {
1628 	struct pfr_ktable *kt = aw->p.tbl;
1629 
1630 	if (aw->type != PF_ADDR_TABLE || kt == NULL)
1631 		return;
1632 	if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL)
1633 		kt = kt->pfrkt_root;
1634 	aw->p.tbl = NULL;
1635 	aw->p.tblcnt = (kt->pfrkt_flags & PFR_TFLAG_ACTIVE) ?
1636 		kt->pfrkt_cnt : -1;
1637 }
1638 
1639 void
pf_print_host(struct pf_addr * addr,u_int16_t p,sa_family_t af)1640 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1641 {
1642 	switch (af) {
1643 #ifdef INET
1644 	case AF_INET: {
1645 		u_int32_t a = ntohl(addr->addr32[0]);
1646 		kprintf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1647 		    (a>>8)&255, a&255);
1648 		if (p) {
1649 			p = ntohs(p);
1650 			kprintf(":%u", p);
1651 		}
1652 		break;
1653 	}
1654 #endif /* INET */
1655 #ifdef INET6
1656 	case AF_INET6: {
1657 		u_int16_t b;
1658 		u_int8_t i, curstart, curend, maxstart, maxend;
1659 		curstart = curend = maxstart = maxend = 255;
1660 		for (i = 0; i < 8; i++) {
1661 			if (!addr->addr16[i]) {
1662 				if (curstart == 255)
1663 					curstart = i;
1664 				curend = i;
1665 			} else {
1666 				if ((curend - curstart) >
1667 				    (maxend - maxstart)) {
1668 					maxstart = curstart;
1669 					maxend = curend;
1670 				}
1671 				curstart = curend = 255;
1672 			}
1673 		}
1674 		if ((curend - curstart) >
1675 		    (maxend - maxstart)) {
1676 			maxstart = curstart;
1677 			maxend = curend;
1678 		}
1679 		for (i = 0; i < 8; i++) {
1680 			if (i >= maxstart && i <= maxend) {
1681 				if (i == 0)
1682 					kprintf(":");
1683 				if (i == maxend)
1684 					kprintf(":");
1685 			} else {
1686 				b = ntohs(addr->addr16[i]);
1687 				kprintf("%x", b);
1688 				if (i < 7)
1689 					kprintf(":");
1690 			}
1691 		}
1692 		if (p) {
1693 			p = ntohs(p);
1694 			kprintf("[%u]", p);
1695 		}
1696 		break;
1697 	}
1698 #endif /* INET6 */
1699 	}
1700 }
1701 
1702 void
pf_print_state(struct pf_state * s)1703 pf_print_state(struct pf_state *s)
1704 {
1705 	pf_print_state_parts(s, NULL, NULL);
1706 }
1707 
1708 void
pf_print_state_parts(struct pf_state * s,struct pf_state_key * skwp,struct pf_state_key * sksp)1709 pf_print_state_parts(struct pf_state *s,
1710     struct pf_state_key *skwp, struct pf_state_key *sksp)
1711 {
1712 	struct pf_state_key *skw, *sks;
1713 	u_int8_t proto, dir;
1714 
1715 	/* Do our best to fill these, but they're skipped if NULL */
1716 	skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1717 	sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1718 	proto = skw ? skw->proto : (sks ? sks->proto : 0);
1719 	dir = s ? s->direction : 0;
1720 
1721 	switch (proto) {
1722 	case IPPROTO_TCP:
1723 		kprintf("TCP ");
1724 		break;
1725 	case IPPROTO_UDP:
1726 		kprintf("UDP ");
1727 		break;
1728 	case IPPROTO_ICMP:
1729 		kprintf("ICMP ");
1730 		break;
1731 	case IPPROTO_ICMPV6:
1732 		kprintf("ICMPV6 ");
1733 		break;
1734 	default:
1735 		kprintf("%u ", skw->proto);
1736 		break;
1737 	}
1738 	switch (dir) {
1739 	case PF_IN:
1740 		kprintf(" in");
1741 		break;
1742 	case PF_OUT:
1743 		kprintf(" out");
1744 		break;
1745 	}
1746 	if (skw) {
1747 		kprintf(" wire: ");
1748 		pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1749 		kprintf(" ");
1750 		pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1751 	}
1752 	if (sks) {
1753 		kprintf(" stack: ");
1754 		if (sks != skw) {
1755 			pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1756 			kprintf(" ");
1757 			pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1758 		} else
1759 			kprintf("-");
1760 	}
1761 	if (s) {
1762 		if (proto == IPPROTO_TCP) {
1763 			kprintf(" [lo=%u high=%u win=%u modulator=%u",
1764 			    s->src.seqlo, s->src.seqhi,
1765 			    s->src.max_win, s->src.seqdiff);
1766 			if (s->src.wscale && s->dst.wscale)
1767 				kprintf(" wscale=%u",
1768 				    s->src.wscale & PF_WSCALE_MASK);
1769 			kprintf("]");
1770 			kprintf(" [lo=%u high=%u win=%u modulator=%u",
1771 			    s->dst.seqlo, s->dst.seqhi,
1772 			    s->dst.max_win, s->dst.seqdiff);
1773 			if (s->src.wscale && s->dst.wscale)
1774 				kprintf(" wscale=%u",
1775 				s->dst.wscale & PF_WSCALE_MASK);
1776 			kprintf("]");
1777 		}
1778 		kprintf(" %u:%u", s->src.state, s->dst.state);
1779 	}
1780 }
1781 
1782 void
pf_print_flags(u_int8_t f)1783 pf_print_flags(u_int8_t f)
1784 {
1785 	if (f)
1786 		kprintf(" ");
1787 	if (f & TH_FIN)
1788 		kprintf("F");
1789 	if (f & TH_SYN)
1790 		kprintf("S");
1791 	if (f & TH_RST)
1792 		kprintf("R");
1793 	if (f & TH_PUSH)
1794 		kprintf("P");
1795 	if (f & TH_ACK)
1796 		kprintf("A");
1797 	if (f & TH_URG)
1798 		kprintf("U");
1799 	if (f & TH_ECE)
1800 		kprintf("E");
1801 	if (f & TH_CWR)
1802 		kprintf("W");
1803 }
1804 
1805 #define	PF_SET_SKIP_STEPS(i)					\
1806 	do {							\
1807 		while (head[i] != cur) {			\
1808 			head[i]->skip[i].ptr = cur;		\
1809 			head[i] = TAILQ_NEXT(head[i], entries);	\
1810 		}						\
1811 	} while (0)
1812 
1813 void
pf_calc_skip_steps(struct pf_rulequeue * rules)1814 pf_calc_skip_steps(struct pf_rulequeue *rules)
1815 {
1816 	struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1817 	int i;
1818 
1819 	cur = TAILQ_FIRST(rules);
1820 	prev = cur;
1821 	for (i = 0; i < PF_SKIP_COUNT; ++i)
1822 		head[i] = cur;
1823 	while (cur != NULL) {
1824 
1825 		if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1826 			PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1827 		if (cur->direction != prev->direction)
1828 			PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1829 		if (cur->af != prev->af)
1830 			PF_SET_SKIP_STEPS(PF_SKIP_AF);
1831 		if (cur->proto != prev->proto)
1832 			PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
1833 		if (cur->src.neg != prev->src.neg ||
1834 		    pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
1835 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
1836 		if (cur->src.port[0] != prev->src.port[0] ||
1837 		    cur->src.port[1] != prev->src.port[1] ||
1838 		    cur->src.port_op != prev->src.port_op)
1839 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
1840 		if (cur->dst.neg != prev->dst.neg ||
1841 		    pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
1842 			PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
1843 		if (cur->dst.port[0] != prev->dst.port[0] ||
1844 		    cur->dst.port[1] != prev->dst.port[1] ||
1845 		    cur->dst.port_op != prev->dst.port_op)
1846 			PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
1847 
1848 		prev = cur;
1849 		cur = TAILQ_NEXT(cur, entries);
1850 	}
1851 	for (i = 0; i < PF_SKIP_COUNT; ++i)
1852 		PF_SET_SKIP_STEPS(i);
1853 }
1854 
1855 int
pf_addr_wrap_neq(struct pf_addr_wrap * aw1,struct pf_addr_wrap * aw2)1856 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
1857 {
1858 	if (aw1->type != aw2->type)
1859 		return (1);
1860 	switch (aw1->type) {
1861 	case PF_ADDR_ADDRMASK:
1862 	case PF_ADDR_RANGE:
1863 		if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
1864 			return (1);
1865 		if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
1866 			return (1);
1867 		return (0);
1868 	case PF_ADDR_DYNIFTL:
1869 		return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
1870 	case PF_ADDR_NOROUTE:
1871 	case PF_ADDR_URPFFAILED:
1872 		return (0);
1873 	case PF_ADDR_TABLE:
1874 		return (aw1->p.tbl != aw2->p.tbl);
1875 	case PF_ADDR_RTLABEL:
1876 		return (aw1->v.rtlabel != aw2->v.rtlabel);
1877 	default:
1878 		kprintf("invalid address type: %d\n", aw1->type);
1879 		return (1);
1880 	}
1881 }
1882 
1883 u_int16_t
pf_cksum_fixup(u_int16_t cksum,u_int16_t old,u_int16_t new,u_int8_t udp)1884 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
1885 {
1886 	u_int32_t	l;
1887 
1888 	if (udp && !cksum)
1889 		return (0x0000);
1890 	l = cksum + old - new;
1891 	l = (l >> 16) + (l & 65535);
1892 	l = l & 65535;
1893 	if (udp && !l)
1894 		return (0xFFFF);
1895 	return (l);
1896 }
1897 
1898 void
pf_change_ap(struct pf_addr * a,u_int16_t * p,u_int16_t * ic,u_int16_t * pc,struct pf_addr * an,u_int16_t pn,u_int8_t u,sa_family_t af)1899 pf_change_ap(struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc,
1900     struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af)
1901 {
1902 	struct pf_addr	ao;
1903 	u_int16_t	po = *p;
1904 
1905 	PF_ACPY(&ao, a, af);
1906 	PF_ACPY(a, an, af);
1907 
1908 	*p = pn;
1909 
1910 	switch (af) {
1911 #ifdef INET
1912 	case AF_INET:
1913 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
1914 		    ao.addr16[0], an->addr16[0], 0),
1915 		    ao.addr16[1], an->addr16[1], 0);
1916 		*p = pn;
1917 		*pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
1918 		    ao.addr16[0], an->addr16[0], u),
1919 		    ao.addr16[1], an->addr16[1], u),
1920 		    po, pn, u);
1921 		break;
1922 #endif /* INET */
1923 #ifdef INET6
1924 	case AF_INET6:
1925 		*pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1926 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1927 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
1928 		    ao.addr16[0], an->addr16[0], u),
1929 		    ao.addr16[1], an->addr16[1], u),
1930 		    ao.addr16[2], an->addr16[2], u),
1931 		    ao.addr16[3], an->addr16[3], u),
1932 		    ao.addr16[4], an->addr16[4], u),
1933 		    ao.addr16[5], an->addr16[5], u),
1934 		    ao.addr16[6], an->addr16[6], u),
1935 		    ao.addr16[7], an->addr16[7], u),
1936 		    po, pn, u);
1937 		break;
1938 #endif /* INET6 */
1939 	}
1940 }
1941 
1942 
1943 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
1944 void
pf_change_a(void * a,u_int16_t * c,u_int32_t an,u_int8_t u)1945 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
1946 {
1947 	u_int32_t	ao;
1948 
1949 	memcpy(&ao, a, sizeof(ao));
1950 	memcpy(a, &an, sizeof(u_int32_t));
1951 	*c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
1952 	    ao % 65536, an % 65536, u);
1953 }
1954 
1955 #ifdef INET6
1956 void
pf_change_a6(struct pf_addr * a,u_int16_t * c,struct pf_addr * an,u_int8_t u)1957 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
1958 {
1959 	struct pf_addr	ao;
1960 
1961 	PF_ACPY(&ao, a, AF_INET6);
1962 	PF_ACPY(a, an, AF_INET6);
1963 
1964 	*c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1965 	    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1966 	    pf_cksum_fixup(pf_cksum_fixup(*c,
1967 	    ao.addr16[0], an->addr16[0], u),
1968 	    ao.addr16[1], an->addr16[1], u),
1969 	    ao.addr16[2], an->addr16[2], u),
1970 	    ao.addr16[3], an->addr16[3], u),
1971 	    ao.addr16[4], an->addr16[4], u),
1972 	    ao.addr16[5], an->addr16[5], u),
1973 	    ao.addr16[6], an->addr16[6], u),
1974 	    ao.addr16[7], an->addr16[7], u);
1975 }
1976 #endif /* INET6 */
1977 
1978 void
pf_change_icmp(struct pf_addr * ia,u_int16_t * ip,struct pf_addr * oa,struct pf_addr * na,u_int16_t np,u_int16_t * pc,u_int16_t * h2c,u_int16_t * ic,u_int16_t * hc,u_int8_t u,sa_family_t af)1979 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
1980     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
1981     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
1982 {
1983 	struct pf_addr	oia, ooa;
1984 
1985 	PF_ACPY(&oia, ia, af);
1986 	if (oa)
1987 		PF_ACPY(&ooa, oa, af);
1988 
1989 	/* Change inner protocol port, fix inner protocol checksum. */
1990 	if (ip != NULL) {
1991 		u_int16_t	oip = *ip;
1992 		u_int32_t	opc = 0;
1993 
1994 		if (pc != NULL)
1995 			opc = *pc;
1996 		*ip = np;
1997 		if (pc != NULL)
1998 			*pc = pf_cksum_fixup(*pc, oip, *ip, u);
1999 		*ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2000 		if (pc != NULL)
2001 			*ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2002 	}
2003 	/* Change inner ip address, fix inner ip and icmp checksums. */
2004 	PF_ACPY(ia, na, af);
2005 	switch (af) {
2006 #ifdef INET
2007 	case AF_INET: {
2008 		u_int32_t	 oh2c = *h2c;
2009 
2010 		*h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2011 		    oia.addr16[0], ia->addr16[0], 0),
2012 		    oia.addr16[1], ia->addr16[1], 0);
2013 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2014 		    oia.addr16[0], ia->addr16[0], 0),
2015 		    oia.addr16[1], ia->addr16[1], 0);
2016 		*ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2017 		break;
2018 	}
2019 #endif /* INET */
2020 #ifdef INET6
2021 	case AF_INET6:
2022 		*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2023 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2024 		    pf_cksum_fixup(pf_cksum_fixup(*ic,
2025 		    oia.addr16[0], ia->addr16[0], u),
2026 		    oia.addr16[1], ia->addr16[1], u),
2027 		    oia.addr16[2], ia->addr16[2], u),
2028 		    oia.addr16[3], ia->addr16[3], u),
2029 		    oia.addr16[4], ia->addr16[4], u),
2030 		    oia.addr16[5], ia->addr16[5], u),
2031 		    oia.addr16[6], ia->addr16[6], u),
2032 		    oia.addr16[7], ia->addr16[7], u);
2033 		break;
2034 #endif /* INET6 */
2035 	}
2036 	/* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2037 	if (oa) {
2038 		PF_ACPY(oa, na, af);
2039 		switch (af) {
2040 #ifdef INET
2041 		case AF_INET:
2042 			*hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2043 			    ooa.addr16[0], oa->addr16[0], 0),
2044 			    ooa.addr16[1], oa->addr16[1], 0);
2045 			break;
2046 #endif /* INET */
2047 #ifdef INET6
2048 		case AF_INET6:
2049 			*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2050 			    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2051 			    pf_cksum_fixup(pf_cksum_fixup(*ic,
2052 			    ooa.addr16[0], oa->addr16[0], u),
2053 			    ooa.addr16[1], oa->addr16[1], u),
2054 			    ooa.addr16[2], oa->addr16[2], u),
2055 			    ooa.addr16[3], oa->addr16[3], u),
2056 			    ooa.addr16[4], oa->addr16[4], u),
2057 			    ooa.addr16[5], oa->addr16[5], u),
2058 			    ooa.addr16[6], oa->addr16[6], u),
2059 			    ooa.addr16[7], oa->addr16[7], u);
2060 			break;
2061 #endif /* INET6 */
2062 		}
2063 	}
2064 }
2065 
2066 
2067 /*
2068  * Need to modulate the sequence numbers in the TCP SACK option
2069  * (credits to Krzysztof Pfaff for report and patch)
2070  */
2071 int
pf_modulate_sack(struct mbuf * m,int off,struct pf_pdesc * pd,struct tcphdr * th,struct pf_state_peer * dst)2072 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2073     struct tcphdr *th, struct pf_state_peer *dst)
2074 {
2075 	int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2076 	u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2077 	int copyback = 0, i, olen;
2078 	struct raw_sackblock sack;
2079 
2080 #define TCPOLEN_SACKLEN	(TCPOLEN_SACK + 2)
2081 	if (hlen < TCPOLEN_SACKLEN ||
2082 	    !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2083 		return 0;
2084 
2085 	while (hlen >= TCPOLEN_SACKLEN) {
2086 		olen = opt[1];
2087 		switch (*opt) {
2088 		case TCPOPT_EOL:	/* FALLTHROUGH */
2089 		case TCPOPT_NOP:
2090 			opt++;
2091 			hlen--;
2092 			break;
2093 		case TCPOPT_SACK:
2094 			if (olen > hlen)
2095 				olen = hlen;
2096 			if (olen >= TCPOLEN_SACKLEN) {
2097 				for (i = 2; i + TCPOLEN_SACK <= olen;
2098 				    i += TCPOLEN_SACK) {
2099 					memcpy(&sack, &opt[i], sizeof(sack));
2100 					pf_change_a(&sack.rblk_start, &th->th_sum,
2101 					    htonl(ntohl(sack.rblk_start) -
2102 					    dst->seqdiff), 0);
2103 					pf_change_a(&sack.rblk_end, &th->th_sum,
2104 					    htonl(ntohl(sack.rblk_end) -
2105 					    dst->seqdiff), 0);
2106 					memcpy(&opt[i], &sack, sizeof(sack));
2107 				}
2108 				copyback = 1;
2109 			}
2110 			/* FALLTHROUGH */
2111 		default:
2112 			if (olen < 2)
2113 				olen = 2;
2114 			hlen -= olen;
2115 			opt += olen;
2116 		}
2117 	}
2118 
2119 	if (copyback)
2120 		m_copyback(m, off + sizeof(*th), thoptlen, opts);
2121 	return (copyback);
2122 }
2123 
2124 void
pf_send_tcp(const struct pf_rule * r,sa_family_t af,const struct pf_addr * saddr,const struct pf_addr * daddr,u_int16_t sport,u_int16_t dport,u_int32_t seq,u_int32_t ack,u_int8_t flags,u_int16_t win,u_int16_t mss,u_int8_t ttl,int tag,u_int16_t rtag,struct ether_header * eh,struct ifnet * ifp)2125 pf_send_tcp(const struct pf_rule *r, sa_family_t af,
2126     const struct pf_addr *saddr, const struct pf_addr *daddr,
2127     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2128     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2129     u_int16_t rtag, struct ether_header *eh, struct ifnet *ifp)
2130 {
2131 	struct mbuf	*m;
2132 	int		 len = 0, tlen;
2133 #ifdef INET
2134 	struct ip	*h = NULL;
2135 #endif /* INET */
2136 #ifdef INET6
2137 	struct ip6_hdr	*h6 = NULL;
2138 #endif /* INET6 */
2139 	struct tcphdr	*th = NULL;
2140 	char		*opt;
2141 
2142 	ASSERT_LWKT_TOKEN_HELD(&pf_token);
2143 
2144 	/* maximum segment size tcp option */
2145 	tlen = sizeof(struct tcphdr);
2146 	if (mss)
2147 		tlen += 4;
2148 
2149 	switch (af) {
2150 #ifdef INET
2151 	case AF_INET:
2152 		len = sizeof(struct ip) + tlen;
2153 		break;
2154 #endif /* INET */
2155 #ifdef INET6
2156 	case AF_INET6:
2157 		len = sizeof(struct ip6_hdr) + tlen;
2158 		break;
2159 #endif /* INET6 */
2160 	}
2161 
2162 	/*
2163 	 * Create outgoing mbuf.
2164 	 *
2165 	 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
2166 	 * so make sure pf.flags is clear.
2167 	 */
2168 	m = m_gethdr(M_NOWAIT, MT_HEADER);
2169 	if (m == NULL) {
2170 		return;
2171 	}
2172 	if (tag)
2173 		m->m_pkthdr.fw_flags |= PF_MBUF_TAGGED;
2174 	m->m_pkthdr.pf.flags = 0;
2175 	m->m_pkthdr.pf.tag = rtag;
2176 	/* XXX Recheck when upgrading to > 4.4 */
2177 	m->m_pkthdr.pf.statekey = NULL;
2178 	if (r != NULL && r->rtableid >= 0)
2179 		m->m_pkthdr.pf.rtableid = r->rtableid;
2180 
2181 #ifdef ALTQ
2182 	if (r != NULL && r->qid) {
2183 		m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE;
2184 		m->m_pkthdr.pf.qid = r->qid;
2185 		m->m_pkthdr.pf.ecn_af = af;
2186 		m->m_pkthdr.pf.hdr = mtod(m, struct ip *);
2187 	}
2188 #endif /* ALTQ */
2189 	m->m_data += max_linkhdr;
2190 	m->m_pkthdr.len = m->m_len = len;
2191 	m->m_pkthdr.rcvif = NULL;
2192 	bzero(m->m_data, len);
2193 	switch (af) {
2194 #ifdef INET
2195 	case AF_INET:
2196 		h = mtod(m, struct ip *);
2197 
2198 		/* IP header fields included in the TCP checksum */
2199 		h->ip_p = IPPROTO_TCP;
2200 		h->ip_len = htons(tlen);
2201 		h->ip_src.s_addr = saddr->v4.s_addr;
2202 		h->ip_dst.s_addr = daddr->v4.s_addr;
2203 
2204 		th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2205 		break;
2206 #endif /* INET */
2207 #ifdef INET6
2208 	case AF_INET6:
2209 		h6 = mtod(m, struct ip6_hdr *);
2210 
2211 		/* IP header fields included in the TCP checksum */
2212 		h6->ip6_nxt = IPPROTO_TCP;
2213 		h6->ip6_plen = htons(tlen);
2214 		memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2215 		memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2216 
2217 		th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2218 		break;
2219 #endif /* INET6 */
2220 	}
2221 
2222 	/* TCP header */
2223 	th->th_sport = sport;
2224 	th->th_dport = dport;
2225 	th->th_seq = htonl(seq);
2226 	th->th_ack = htonl(ack);
2227 	th->th_off = tlen >> 2;
2228 	th->th_flags = flags;
2229 	th->th_win = htons(win);
2230 
2231 	if (mss) {
2232 		opt = (char *)(th + 1);
2233 		opt[0] = TCPOPT_MAXSEG;
2234 		opt[1] = 4;
2235 		mss = htons(mss);
2236 		bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2237 	}
2238 
2239 	switch (af) {
2240 #ifdef INET
2241 	case AF_INET:
2242 		/* TCP checksum */
2243 		th->th_sum = in_cksum(m, len);
2244 
2245 		/* Finish the IP header */
2246 		h->ip_v = 4;
2247 		h->ip_hl = sizeof(*h) >> 2;
2248 		h->ip_tos = IPTOS_LOWDELAY;
2249 		h->ip_len = htons(len);
2250 		h->ip_off = path_mtu_discovery ? htons(IP_DF) : 0;
2251 		h->ip_ttl = ttl ? ttl : ip_defttl;
2252 		h->ip_sum = 0;
2253 		if (eh == NULL) {
2254 			lwkt_reltoken(&pf_token);
2255 			ip_output(m, NULL, NULL, 0, NULL, NULL);
2256 			lwkt_gettoken(&pf_token);
2257 		} else {
2258 			struct route		 ro;
2259 			struct rtentry		 rt;
2260 			struct ether_header	*e = (void *)ro.ro_dst.sa_data;
2261 
2262 			if (ifp == NULL) {
2263 				m_freem(m);
2264 				return;
2265 			}
2266 			rt.rt_ifp = ifp;
2267 			ro.ro_rt = &rt;
2268 			ro.ro_dst.sa_len = sizeof(ro.ro_dst);
2269 			ro.ro_dst.sa_family = pseudo_AF_HDRCMPLT;
2270 			bcopy(eh->ether_dhost, e->ether_shost, ETHER_ADDR_LEN);
2271 			bcopy(eh->ether_shost, e->ether_dhost, ETHER_ADDR_LEN);
2272 			e->ether_type = eh->ether_type;
2273 			/* XXX_IMPORT: later */
2274 			lwkt_reltoken(&pf_token);
2275 			ip_output(m, NULL, &ro, 0, NULL, NULL);
2276 			lwkt_gettoken(&pf_token);
2277 		}
2278 		break;
2279 #endif /* INET */
2280 #ifdef INET6
2281 	case AF_INET6:
2282 		/* TCP checksum */
2283 		th->th_sum = in6_cksum(m, IPPROTO_TCP,
2284 		    sizeof(struct ip6_hdr), tlen);
2285 
2286 		h6->ip6_vfc |= IPV6_VERSION;
2287 		h6->ip6_hlim = IPV6_DEFHLIM;
2288 
2289 		lwkt_reltoken(&pf_token);
2290 		ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
2291 		lwkt_gettoken(&pf_token);
2292 		break;
2293 #endif /* INET6 */
2294 	}
2295 }
2296 
2297 void
pf_send_icmp(struct mbuf * m,u_int8_t type,u_int8_t code,sa_family_t af,struct pf_rule * r)2298 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2299     struct pf_rule *r)
2300 {
2301 	struct mbuf	*m0;
2302 
2303 	/*
2304 	 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
2305 	 * so make sure pf.flags is clear.
2306 	 */
2307 	if ((m0 = m_copym(m, 0, M_COPYALL, M_NOWAIT)) == NULL)
2308 		return;
2309 
2310 	m0->m_pkthdr.fw_flags |= PF_MBUF_TAGGED;
2311 	m0->m_pkthdr.pf.flags = 0;
2312 	/* XXX Re-Check when Upgrading to > 4.4 */
2313 	m0->m_pkthdr.pf.statekey = NULL;
2314 
2315 	if (r->rtableid >= 0)
2316 		m0->m_pkthdr.pf.rtableid = r->rtableid;
2317 
2318 #ifdef ALTQ
2319 	if (r->qid) {
2320 		m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE;
2321 		m0->m_pkthdr.pf.qid = r->qid;
2322 		m0->m_pkthdr.pf.ecn_af = af;
2323 		m0->m_pkthdr.pf.hdr = mtod(m0, struct ip *);
2324 	}
2325 #endif /* ALTQ */
2326 
2327 	switch (af) {
2328 #ifdef INET
2329 	case AF_INET:
2330 		icmp_error(m0, type, code, 0, 0);
2331 		break;
2332 #endif /* INET */
2333 #ifdef INET6
2334 	case AF_INET6:
2335 		icmp6_error(m0, type, code, 0);
2336 		break;
2337 #endif /* INET6 */
2338 	}
2339 }
2340 
2341 /*
2342  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2343  * If n is 0, they match if they are equal. If n is != 0, they match if they
2344  * are different.
2345  */
2346 int
pf_match_addr(u_int8_t n,struct pf_addr * a,struct pf_addr * m,struct pf_addr * b,sa_family_t af)2347 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2348     struct pf_addr *b, sa_family_t af)
2349 {
2350 	int	match = 0;
2351 
2352 	switch (af) {
2353 #ifdef INET
2354 	case AF_INET:
2355 		if ((a->addr32[0] & m->addr32[0]) ==
2356 		    (b->addr32[0] & m->addr32[0]))
2357 			match++;
2358 		break;
2359 #endif /* INET */
2360 #ifdef INET6
2361 	case AF_INET6:
2362 		if (((a->addr32[0] & m->addr32[0]) ==
2363 		     (b->addr32[0] & m->addr32[0])) &&
2364 		    ((a->addr32[1] & m->addr32[1]) ==
2365 		     (b->addr32[1] & m->addr32[1])) &&
2366 		    ((a->addr32[2] & m->addr32[2]) ==
2367 		     (b->addr32[2] & m->addr32[2])) &&
2368 		    ((a->addr32[3] & m->addr32[3]) ==
2369 		     (b->addr32[3] & m->addr32[3])))
2370 			match++;
2371 		break;
2372 #endif /* INET6 */
2373 	}
2374 	if (match) {
2375 		if (n)
2376 			return (0);
2377 		else
2378 			return (1);
2379 	} else {
2380 		if (n)
2381 			return (1);
2382 		else
2383 			return (0);
2384 	}
2385 }
2386 
2387 /*
2388  * Return 1 if b <= a <= e, otherwise return 0.
2389  */
2390 int
pf_match_addr_range(struct pf_addr * b,struct pf_addr * e,struct pf_addr * a,sa_family_t af)2391 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2392     struct pf_addr *a, sa_family_t af)
2393 {
2394 	switch (af) {
2395 #ifdef INET
2396 	case AF_INET:
2397 		if ((a->addr32[0] < b->addr32[0]) ||
2398 		    (a->addr32[0] > e->addr32[0]))
2399 			return (0);
2400 		break;
2401 #endif /* INET */
2402 #ifdef INET6
2403 	case AF_INET6: {
2404 		int	i;
2405 
2406 		/* check a >= b */
2407 		for (i = 0; i < 4; ++i)
2408 			if (a->addr32[i] > b->addr32[i])
2409 				break;
2410 			else if (a->addr32[i] < b->addr32[i])
2411 				return (0);
2412 		/* check a <= e */
2413 		for (i = 0; i < 4; ++i)
2414 			if (a->addr32[i] < e->addr32[i])
2415 				break;
2416 			else if (a->addr32[i] > e->addr32[i])
2417 				return (0);
2418 		break;
2419 	}
2420 #endif /* INET6 */
2421 	}
2422 	return (1);
2423 }
2424 
2425 int
pf_match(u_int8_t op,u_int32_t a1,u_int32_t a2,u_int32_t p)2426 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2427 {
2428 	switch (op) {
2429 	case PF_OP_IRG:
2430 		return ((p > a1) && (p < a2));
2431 	case PF_OP_XRG:
2432 		return ((p < a1) || (p > a2));
2433 	case PF_OP_RRG:
2434 		return ((p >= a1) && (p <= a2));
2435 	case PF_OP_EQ:
2436 		return (p == a1);
2437 	case PF_OP_NE:
2438 		return (p != a1);
2439 	case PF_OP_LT:
2440 		return (p < a1);
2441 	case PF_OP_LE:
2442 		return (p <= a1);
2443 	case PF_OP_GT:
2444 		return (p > a1);
2445 	case PF_OP_GE:
2446 		return (p >= a1);
2447 	}
2448 	return (0); /* never reached */
2449 }
2450 
2451 int
pf_match_port(u_int8_t op,u_int16_t a1,u_int16_t a2,u_int16_t p)2452 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2453 {
2454 	a1 = ntohs(a1);
2455 	a2 = ntohs(a2);
2456 	p = ntohs(p);
2457 	return (pf_match(op, a1, a2, p));
2458 }
2459 
2460 int
pf_match_uid(u_int8_t op,uid_t a1,uid_t a2,uid_t u)2461 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2462 {
2463 	if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2464 		return (0);
2465 	return (pf_match(op, a1, a2, u));
2466 }
2467 
2468 int
pf_match_gid(u_int8_t op,gid_t a1,gid_t a2,gid_t g)2469 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2470 {
2471 	if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2472 		return (0);
2473 	return (pf_match(op, a1, a2, g));
2474 }
2475 
2476 int
pf_match_tag(struct mbuf * m,struct pf_rule * r,int * tag)2477 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag)
2478 {
2479 	if (*tag == -1)
2480 		*tag = m->m_pkthdr.pf.tag;
2481 
2482 	return ((!r->match_tag_not && r->match_tag == *tag) ||
2483 	    (r->match_tag_not && r->match_tag != *tag));
2484 }
2485 
2486 int
pf_tag_packet(struct mbuf * m,int tag,int rtableid)2487 pf_tag_packet(struct mbuf *m, int tag, int rtableid)
2488 {
2489 	if (tag <= 0 && rtableid < 0)
2490 		return (0);
2491 
2492 	if (tag > 0)
2493 		m->m_pkthdr.pf.tag = tag;
2494 	if (rtableid >= 0)
2495 		m->m_pkthdr.pf.rtableid = rtableid;
2496 
2497 	return (0);
2498 }
2499 
2500 void
pf_step_into_anchor(int * depth,struct pf_ruleset ** rs,int n,struct pf_rule ** r,struct pf_rule ** a,int * match)2501 pf_step_into_anchor(int *depth, struct pf_ruleset **rs, int n,
2502     struct pf_rule **r, struct pf_rule **a, int *match)
2503 {
2504 	struct pf_anchor_stackframe	*f;
2505 
2506 	(*r)->anchor->match = 0;
2507 	if (match)
2508 		*match = 0;
2509 	if (*depth >= NELEM(pf_anchor_stack)) {
2510 		kprintf("pf_step_into_anchor: stack overflow\n");
2511 		*r = TAILQ_NEXT(*r, entries);
2512 		return;
2513 	} else if (*depth == 0 && a != NULL)
2514 		*a = *r;
2515 	f = pf_anchor_stack + (*depth)++;
2516 	f->rs = *rs;
2517 	f->r = *r;
2518 	if ((*r)->anchor_wildcard) {
2519 		f->parent = &(*r)->anchor->children;
2520 		if ((f->child = RB_MIN(pf_anchor_node, f->parent)) ==
2521 		    NULL) {
2522 			*r = NULL;
2523 			return;
2524 		}
2525 		*rs = &f->child->ruleset;
2526 	} else {
2527 		f->parent = NULL;
2528 		f->child = NULL;
2529 		*rs = &(*r)->anchor->ruleset;
2530 	}
2531 	*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2532 }
2533 
2534 int
pf_step_out_of_anchor(int * depth,struct pf_ruleset ** rs,int n,struct pf_rule ** r,struct pf_rule ** a,int * match)2535 pf_step_out_of_anchor(int *depth, struct pf_ruleset **rs, int n,
2536     struct pf_rule **r, struct pf_rule **a, int *match)
2537 {
2538 	struct pf_anchor_stackframe	*f;
2539 	int quick = 0;
2540 
2541 	do {
2542 		if (*depth <= 0)
2543 			break;
2544 		f = pf_anchor_stack + *depth - 1;
2545 		if (f->parent != NULL && f->child != NULL) {
2546 			if (f->child->match ||
2547 			    (match != NULL && *match)) {
2548 				f->r->anchor->match = 1;
2549 				*match = 0;
2550 			}
2551 			f->child = RB_NEXT(pf_anchor_node, f->parent, f->child);
2552 			if (f->child != NULL) {
2553 				*rs = &f->child->ruleset;
2554 				*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2555 				if (*r == NULL)
2556 					continue;
2557 				else
2558 					break;
2559 			}
2560 		}
2561 		(*depth)--;
2562 		if (*depth == 0 && a != NULL)
2563 			*a = NULL;
2564 		*rs = f->rs;
2565 		if (f->r->anchor->match || (match != NULL && *match))
2566 			quick = f->r->quick;
2567 		*r = TAILQ_NEXT(f->r, entries);
2568 	} while (*r == NULL);
2569 
2570 	return (quick);
2571 }
2572 
2573 #ifdef INET6
2574 void
pf_poolmask(struct pf_addr * naddr,struct pf_addr * raddr,struct pf_addr * rmask,struct pf_addr * saddr,sa_family_t af)2575 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2576     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2577 {
2578 	switch (af) {
2579 #ifdef INET
2580 	case AF_INET:
2581 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2582 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2583 		break;
2584 #endif /* INET */
2585 	case AF_INET6:
2586 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2587 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2588 		naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2589 		((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2590 		naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2591 		((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2592 		naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2593 		((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2594 		break;
2595 	}
2596 }
2597 
2598 void
pf_addr_inc(struct pf_addr * addr,sa_family_t af)2599 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2600 {
2601 	switch (af) {
2602 #ifdef INET
2603 	case AF_INET:
2604 		addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2605 		break;
2606 #endif /* INET */
2607 	case AF_INET6:
2608 		if (addr->addr32[3] == 0xffffffff) {
2609 			addr->addr32[3] = 0;
2610 			if (addr->addr32[2] == 0xffffffff) {
2611 				addr->addr32[2] = 0;
2612 				if (addr->addr32[1] == 0xffffffff) {
2613 					addr->addr32[1] = 0;
2614 					addr->addr32[0] =
2615 					    htonl(ntohl(addr->addr32[0]) + 1);
2616 				} else
2617 					addr->addr32[1] =
2618 					    htonl(ntohl(addr->addr32[1]) + 1);
2619 			} else
2620 				addr->addr32[2] =
2621 				    htonl(ntohl(addr->addr32[2]) + 1);
2622 		} else
2623 			addr->addr32[3] =
2624 			    htonl(ntohl(addr->addr32[3]) + 1);
2625 		break;
2626 	}
2627 }
2628 #endif /* INET6 */
2629 
2630 #define mix(a,b,c) \
2631 	do {					\
2632 		a -= b; a -= c; a ^= (c >> 13);	\
2633 		b -= c; b -= a; b ^= (a << 8);	\
2634 		c -= a; c -= b; c ^= (b >> 13);	\
2635 		a -= b; a -= c; a ^= (c >> 12);	\
2636 		b -= c; b -= a; b ^= (a << 16);	\
2637 		c -= a; c -= b; c ^= (b >> 5);	\
2638 		a -= b; a -= c; a ^= (c >> 3);	\
2639 		b -= c; b -= a; b ^= (a << 10);	\
2640 		c -= a; c -= b; c ^= (b >> 15);	\
2641 	} while (0)
2642 
2643 /*
2644  * hash function based on bridge_hash in if_bridge.c
2645  */
2646 void
pf_hash(struct pf_addr * inaddr,struct pf_addr * hash,struct pf_poolhashkey * key,sa_family_t af)2647 pf_hash(struct pf_addr *inaddr, struct pf_addr *hash,
2648     struct pf_poolhashkey *key, sa_family_t af)
2649 {
2650 	u_int32_t	a = 0x9e3779b9, b = 0x9e3779b9, c = key->key32[0];
2651 
2652 	switch (af) {
2653 #ifdef INET
2654 	case AF_INET:
2655 		a += inaddr->addr32[0];
2656 		b += key->key32[1];
2657 		mix(a, b, c);
2658 		hash->addr32[0] = c + key->key32[2];
2659 		break;
2660 #endif /* INET */
2661 #ifdef INET6
2662 	case AF_INET6:
2663 		a += inaddr->addr32[0];
2664 		b += inaddr->addr32[2];
2665 		mix(a, b, c);
2666 		hash->addr32[0] = c;
2667 		a += inaddr->addr32[1];
2668 		b += inaddr->addr32[3];
2669 		c += key->key32[1];
2670 		mix(a, b, c);
2671 		hash->addr32[1] = c;
2672 		a += inaddr->addr32[2];
2673 		b += inaddr->addr32[1];
2674 		c += key->key32[2];
2675 		mix(a, b, c);
2676 		hash->addr32[2] = c;
2677 		a += inaddr->addr32[3];
2678 		b += inaddr->addr32[0];
2679 		c += key->key32[3];
2680 		mix(a, b, c);
2681 		hash->addr32[3] = c;
2682 		break;
2683 #endif /* INET6 */
2684 	}
2685 }
2686 
2687 int
pf_map_addr(sa_family_t af,struct pf_rule * r,struct pf_addr * saddr,struct pf_addr * naddr,struct pf_addr * init_addr,struct pf_src_node ** sn)2688 pf_map_addr(sa_family_t af, struct pf_rule *r, struct pf_addr *saddr,
2689     struct pf_addr *naddr, struct pf_addr *init_addr, struct pf_src_node **sn)
2690 {
2691 	unsigned char		 hash[16];
2692 	struct pf_pool		*rpool = &r->rpool;
2693 	struct pf_pooladdr	*acur = rpool->cur;
2694 	struct pf_pooladdr	*cur;
2695 	struct pf_addr		*raddr;
2696 	struct pf_addr		*rmask;
2697 	struct pf_addr		counter;
2698 	struct pf_src_node	 k;
2699 	int cpu = mycpu->gd_cpuid;
2700 	int tblidx;
2701 
2702 	bzero(hash, sizeof(hash));	/* avoid gcc warnings */
2703 
2704 	/*
2705 	 * NOTE! rpool->cur and rpool->tblidx can be iterators and thus
2706 	 *	 may represent a SMP race due to the shared nature of the
2707 	 *	 rpool structure.  We allow the race and ensure that updates
2708 	 *	 do not create a fatal condition.
2709 	 */
2710 	cpu_ccfence();
2711 	cur = acur;
2712 	raddr = &cur->addr.v.a.addr;
2713 	rmask = &cur->addr.v.a.mask;
2714 
2715 	if (*sn == NULL && r->rpool.opts & PF_POOL_STICKYADDR &&
2716 	    (r->rpool.opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) {
2717 		k.af = af;
2718 		PF_ACPY(&k.addr, saddr, af);
2719 		if (r->rule_flag & PFRULE_RULESRCTRACK ||
2720 		    r->rpool.opts & PF_POOL_STICKYADDR)
2721 			k.rule.ptr = r;
2722 		else
2723 			k.rule.ptr = NULL;
2724 		PF_INC_SCOUNTER(SCNT_SRC_NODE_SEARCH);
2725 		*sn = RB_FIND(pf_src_tree, &tree_src_tracking[cpu], &k);
2726 		if (*sn != NULL && !PF_AZERO(&(*sn)->raddr, af)) {
2727 			PF_ACPY(naddr, &(*sn)->raddr, af);
2728 			if (pf_status.debug >= PF_DEBUG_MISC) {
2729 				kprintf("pf_map_addr: src tracking maps ");
2730 				pf_print_host(&k.addr, 0, af);
2731 				kprintf(" to ");
2732 				pf_print_host(naddr, 0, af);
2733 				kprintf("\n");
2734 			}
2735 			return (0);
2736 		}
2737 	}
2738 
2739 	if (cur->addr.type == PF_ADDR_NOROUTE)
2740 		return (1);
2741 	if (cur->addr.type == PF_ADDR_DYNIFTL) {
2742 		switch (af) {
2743 #ifdef INET
2744 		case AF_INET:
2745 			if (cur->addr.p.dyn->pfid_acnt4 < 1 &&
2746 			    (rpool->opts & PF_POOL_TYPEMASK) !=
2747 			    PF_POOL_ROUNDROBIN)
2748 				return (1);
2749 			raddr = &cur->addr.p.dyn->pfid_addr4;
2750 			rmask = &cur->addr.p.dyn->pfid_mask4;
2751 			break;
2752 #endif /* INET */
2753 #ifdef INET6
2754 		case AF_INET6:
2755 			if (cur->addr.p.dyn->pfid_acnt6 < 1 &&
2756 			    (rpool->opts & PF_POOL_TYPEMASK) !=
2757 			    PF_POOL_ROUNDROBIN)
2758 				return (1);
2759 			raddr = &cur->addr.p.dyn->pfid_addr6;
2760 			rmask = &cur->addr.p.dyn->pfid_mask6;
2761 			break;
2762 #endif /* INET6 */
2763 		}
2764 	} else if (cur->addr.type == PF_ADDR_TABLE) {
2765 		if ((rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_ROUNDROBIN)
2766 			return (1); /* unsupported */
2767 	} else {
2768 		raddr = &cur->addr.v.a.addr;
2769 		rmask = &cur->addr.v.a.mask;
2770 	}
2771 
2772 	switch (rpool->opts & PF_POOL_TYPEMASK) {
2773 	case PF_POOL_NONE:
2774 		PF_ACPY(naddr, raddr, af);
2775 		break;
2776 	case PF_POOL_BITMASK:
2777 		PF_POOLMASK(naddr, raddr, rmask, saddr, af);
2778 		break;
2779 	case PF_POOL_RANDOM:
2780 		if (init_addr != NULL && PF_AZERO(init_addr, af)) {
2781 			switch (af) {
2782 #ifdef INET
2783 			case AF_INET:
2784 				counter.addr32[0] = htonl(karc4random());
2785 				break;
2786 #endif /* INET */
2787 #ifdef INET6
2788 			case AF_INET6:
2789 				if (rmask->addr32[3] != 0xffffffff)
2790 					counter.addr32[3] =
2791 						htonl(karc4random());
2792 				else
2793 					break;
2794 				if (rmask->addr32[2] != 0xffffffff)
2795 					counter.addr32[2] =
2796 						htonl(karc4random());
2797 				else
2798 					break;
2799 				if (rmask->addr32[1] != 0xffffffff)
2800 					counter.addr32[1] =
2801 						htonl(karc4random());
2802 				else
2803 					break;
2804 				if (rmask->addr32[0] != 0xffffffff)
2805 					counter.addr32[0] =
2806 						htonl(karc4random());
2807 				break;
2808 #endif /* INET6 */
2809 			}
2810 			PF_POOLMASK(naddr, raddr, rmask, &counter, af);
2811 			PF_ACPY(init_addr, naddr, af);
2812 
2813 		} else {
2814 			counter = rpool->counter;
2815 			cpu_ccfence();
2816 			PF_AINC(&counter, af);
2817 			PF_POOLMASK(naddr, raddr, rmask, &counter, af);
2818 			rpool->counter = counter;
2819 		}
2820 		break;
2821 	case PF_POOL_SRCHASH:
2822 		pf_hash(saddr, (struct pf_addr *)&hash, &rpool->key, af);
2823 		PF_POOLMASK(naddr, raddr, rmask, (struct pf_addr *)&hash, af);
2824 		break;
2825 	case PF_POOL_ROUNDROBIN:
2826 		tblidx = rpool->tblidx;
2827 		counter = rpool->counter;
2828 		if (cur->addr.type == PF_ADDR_TABLE) {
2829 			if (!pfr_pool_get(cur->addr.p.tbl,
2830 			    &tblidx, &counter,
2831 			    &raddr, &rmask, af)) {
2832 				goto get_addr;
2833 			}
2834 		} else if (cur->addr.type == PF_ADDR_DYNIFTL) {
2835 			if (!pfr_pool_get(cur->addr.p.dyn->pfid_kt,
2836 			    &tblidx, &counter,
2837 			    &raddr, &rmask, af)) {
2838 				goto get_addr;
2839 			}
2840 		} else if (pf_match_addr(0, raddr, rmask,
2841 					 &counter, af)) {
2842 			goto get_addr;
2843 		}
2844 
2845 	try_next:
2846 		if ((cur = TAILQ_NEXT(cur, entries)) == NULL)
2847 			cur = TAILQ_FIRST(&rpool->list);
2848 		if (cur->addr.type == PF_ADDR_TABLE) {
2849 			tblidx = -1;
2850 			if (pfr_pool_get(cur->addr.p.tbl,
2851 			    &tblidx, &counter,
2852 			    &raddr, &rmask, af)) {
2853 				/* table contains no address of type 'af' */
2854 				if (cur != acur)
2855 					goto try_next;
2856 				return (1);
2857 			}
2858 		} else if (cur->addr.type == PF_ADDR_DYNIFTL) {
2859 			tblidx = -1;
2860 			if (pfr_pool_get(cur->addr.p.dyn->pfid_kt,
2861 			    &tblidx, &counter,
2862 			    &raddr, &rmask, af)) {
2863 				/* table contains no address of type 'af' */
2864 				if (cur != acur)
2865 					goto try_next;
2866 				return (1);
2867 			}
2868 		} else {
2869 			raddr = &cur->addr.v.a.addr;
2870 			rmask = &cur->addr.v.a.mask;
2871 			PF_ACPY(&counter, raddr, af);
2872 		}
2873 
2874 	get_addr:
2875 		rpool->cur = cur;
2876 		rpool->tblidx = tblidx;
2877 		PF_ACPY(naddr, &counter, af);
2878 		if (init_addr != NULL && PF_AZERO(init_addr, af))
2879 			PF_ACPY(init_addr, naddr, af);
2880 		PF_AINC(&counter, af);
2881 		rpool->counter = counter;
2882 		break;
2883 	}
2884 	if (*sn != NULL)
2885 		PF_ACPY(&(*sn)->raddr, naddr, af);
2886 
2887 	if (pf_status.debug >= PF_DEBUG_MISC &&
2888 	    (rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) {
2889 		kprintf("pf_map_addr: selected address ");
2890 		pf_print_host(naddr, 0, af);
2891 		kprintf("\n");
2892 	}
2893 
2894 	return (0);
2895 }
2896 
2897 int
pf_get_sport(struct pf_pdesc * pd,sa_family_t af,u_int8_t proto,struct pf_rule * r,struct pf_addr * saddr,struct pf_addr * daddr,u_int16_t sport,u_int16_t dport,struct pf_addr * naddr,u_int16_t * nport,u_int16_t low,u_int16_t high,struct pf_src_node ** sn)2898 pf_get_sport(struct pf_pdesc *pd, sa_family_t af,
2899 	     u_int8_t proto, struct pf_rule *r,
2900 	     struct pf_addr *saddr, struct pf_addr *daddr,
2901 	     u_int16_t sport, u_int16_t dport,
2902 	     struct pf_addr *naddr, u_int16_t *nport,
2903 	     u_int16_t low, u_int16_t high, struct pf_src_node **sn)
2904 {
2905 	struct pf_state_key_cmp	key;
2906 	struct pf_addr		init_addr;
2907 	u_int16_t		cut;
2908 	u_int32_t		hash_base = 0;
2909 	int			do_hash = 0;
2910 
2911 	bzero(&init_addr, sizeof(init_addr));
2912 	if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn))
2913 		return (1);
2914 
2915 	if (proto == IPPROTO_ICMP) {
2916 		low = 1;
2917 		high = 65535;
2918 	}
2919 
2920 	bzero(&key, sizeof(key));
2921 	key.af = af;
2922 	key.proto = proto;
2923 	key.port[0] = dport;
2924 	PF_ACPY(&key.addr[0], daddr, key.af);
2925 
2926 	do {
2927 		PF_ACPY(&key.addr[1], naddr, key.af);
2928 
2929 		/*
2930 		 * We want to select a port that calculates to a toeplitz hash
2931 		 * that masks to the same cpu, otherwise the response may
2932 		 * not see the new state.
2933 		 *
2934 		 * We can still do this even if the kernel is disregarding
2935 		 * the hash and vectoring the packets to a specific cpu,
2936 		 * but it will reduce the number of ports we can use.
2937 		 */
2938 		switch(af) {
2939 		case AF_INET:
2940 			if (proto == IPPROTO_TCP) {
2941 				do_hash = 1;
2942 				hash_base = toeplitz_piecemeal_port(dport) ^
2943 				    toeplitz_piecemeal_addr(daddr->v4.s_addr) ^
2944 				    toeplitz_piecemeal_addr(naddr->v4.s_addr);
2945 			}
2946 			break;
2947 		case AF_INET6:
2948 			/* XXX TODO XXX */
2949 		default:
2950 			/* XXX TODO XXX */
2951 			break;
2952 		}
2953 
2954 		/*
2955 		 * port search; start random, step;
2956 		 * similar 2 portloop in in_pcbbind
2957 		 *
2958 		 * WARNING! We try to match such that the kernel will
2959 		 *	    dispatch the translated host/port to the same
2960 		 *	    cpu, but this might not be possible.
2961 		 *
2962 		 *	    In the case where the port is fixed, or for the
2963 		 *	    UDP case (whos toeplitz does not incorporate the
2964 		 *	    port), we set not_cpu_localized which ultimately
2965 		 *	    causes the pf_state_tree element
2966 		 *
2967 		 * XXX fixed ports present a problem for cpu localization.
2968 		 */
2969 		if (!(proto == IPPROTO_TCP ||
2970 		      proto == IPPROTO_UDP ||
2971 		      proto == IPPROTO_ICMP)) {
2972 			/*
2973 			 * non-specific protocol, leave port intact.
2974 			 */
2975 			key.port[1] = sport;
2976 			if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
2977 				*nport = sport;
2978 				pd->not_cpu_localized = 1;
2979 				return (0);
2980 			}
2981 		} else if (low == 0 && high == 0) {
2982 			/*
2983 			 * static-port same as originator.
2984 			 */
2985 			key.port[1] = sport;
2986 			if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
2987 				*nport = sport;
2988 				pd->not_cpu_localized = 1;
2989 				return (0);
2990 			}
2991 		} else if (low == high) {
2992 			/*
2993 			 * specific port as specified.
2994 			 */
2995 			key.port[1] = htons(low);
2996 			if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
2997 				*nport = htons(low);
2998 				pd->not_cpu_localized = 1;
2999 				return (0);
3000 			}
3001 		} else {
3002 			/*
3003 			 * normal dynamic port
3004 			 */
3005 			u_int16_t tmp;
3006 
3007 			if (low > high) {
3008 				tmp = low;
3009 				low = high;
3010 				high = tmp;
3011 			}
3012 			/* low < high */
3013 			cut = htonl(karc4random()) % (1 + high - low) + low;
3014 			/* low <= cut <= high */
3015 			for (tmp = cut; tmp <= high; ++(tmp)) {
3016 				key.port[1] = htons(tmp);
3017 				if (do_hash) {
3018 					uint32_t hash;
3019 
3020 					hash = hash_base ^
3021 					toeplitz_piecemeal_port(key.port[1]);
3022 					if (netisr_hashcpu(hash) != mycpuid)
3023 						continue;
3024 				}
3025 				if (pf_find_state_all(&key, PF_IN, NULL) ==
3026 				    NULL && !in_baddynamic(tmp, proto)) {
3027 					if (proto == IPPROTO_UDP)
3028 						pd->not_cpu_localized = 1;
3029 					*nport = htons(tmp);
3030 					return (0);
3031 				}
3032 			}
3033 			for (tmp = cut - 1; tmp >= low; --(tmp)) {
3034 				key.port[1] = htons(tmp);
3035 				if (do_hash) {
3036 					uint32_t hash;
3037 
3038 					hash = hash_base ^
3039 					toeplitz_piecemeal_port(key.port[1]);
3040 					if (netisr_hashcpu(hash) != mycpuid)
3041 						continue;
3042 				}
3043 				if (pf_find_state_all(&key, PF_IN, NULL) ==
3044 				    NULL && !in_baddynamic(tmp, proto)) {
3045 					if (proto == IPPROTO_UDP)
3046 						pd->not_cpu_localized = 1;
3047 					*nport = htons(tmp);
3048 					return (0);
3049 				}
3050 			}
3051 		}
3052 
3053 		/*
3054 		 * Next address
3055 		 */
3056 		switch (r->rpool.opts & PF_POOL_TYPEMASK) {
3057 		case PF_POOL_RANDOM:
3058 		case PF_POOL_ROUNDROBIN:
3059 			if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn))
3060 				return (1);
3061 			break;
3062 		case PF_POOL_NONE:
3063 		case PF_POOL_SRCHASH:
3064 		case PF_POOL_BITMASK:
3065 		default:
3066 			return (1);
3067 		}
3068 	} while (! PF_AEQ(&init_addr, naddr, af) );
3069 	return (1);					/* none available */
3070 }
3071 
3072 struct pf_rule *
pf_match_translation(struct pf_pdesc * pd,struct mbuf * m,int off,int direction,struct pfi_kif * kif,struct pf_addr * saddr,u_int16_t sport,struct pf_addr * daddr,u_int16_t dport,int rs_num)3073 pf_match_translation(struct pf_pdesc *pd, struct mbuf *m, int off,
3074     int direction, struct pfi_kif *kif, struct pf_addr *saddr, u_int16_t sport,
3075     struct pf_addr *daddr, u_int16_t dport, int rs_num)
3076 {
3077 	struct pf_rule		*r, *rm = NULL;
3078 	struct pf_ruleset	*ruleset = NULL;
3079 	int			 tag = -1;
3080 	int			 rtableid = -1;
3081 	int			 asd = 0;
3082 
3083 	r = TAILQ_FIRST(pf_main_ruleset.rules[rs_num].active.ptr);
3084 	while (r && rm == NULL) {
3085 		struct pf_rule_addr	*src = NULL, *dst = NULL;
3086 		struct pf_addr_wrap	*xdst = NULL;
3087 		struct pf_pooladdr	*cur;
3088 
3089 		if (r->action == PF_BINAT && direction == PF_IN) {
3090 			src = &r->dst;
3091 			cur = r->rpool.cur;	/* SMP race possible */
3092 			cpu_ccfence();
3093 			if (cur)
3094 				xdst = &cur->addr;
3095 		} else {
3096 			src = &r->src;
3097 			dst = &r->dst;
3098 		}
3099 
3100 		r->evaluations++;
3101 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
3102 			r = r->skip[PF_SKIP_IFP].ptr;
3103 		else if (r->direction && r->direction != direction)
3104 			r = r->skip[PF_SKIP_DIR].ptr;
3105 		else if (r->af && r->af != pd->af)
3106 			r = r->skip[PF_SKIP_AF].ptr;
3107 		else if (r->proto && r->proto != pd->proto)
3108 			r = r->skip[PF_SKIP_PROTO].ptr;
3109 		else if (PF_MISMATCHAW(&src->addr, saddr, pd->af,
3110 		    src->neg, kif))
3111 			r = r->skip[src == &r->src ? PF_SKIP_SRC_ADDR :
3112 			    PF_SKIP_DST_ADDR].ptr;
3113 		else if (src->port_op && !pf_match_port(src->port_op,
3114 		    src->port[0], src->port[1], sport))
3115 			r = r->skip[src == &r->src ? PF_SKIP_SRC_PORT :
3116 			    PF_SKIP_DST_PORT].ptr;
3117 		else if (dst != NULL &&
3118 		    PF_MISMATCHAW(&dst->addr, daddr, pd->af, dst->neg, NULL))
3119 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
3120 		else if (xdst != NULL && PF_MISMATCHAW(xdst, daddr, pd->af,
3121 		    0, NULL))
3122 			r = TAILQ_NEXT(r, entries);
3123 		else if (dst != NULL && dst->port_op &&
3124 		    !pf_match_port(dst->port_op, dst->port[0],
3125 		    dst->port[1], dport))
3126 			r = r->skip[PF_SKIP_DST_PORT].ptr;
3127 		else if (r->match_tag && !pf_match_tag(m, r, &tag))
3128 			r = TAILQ_NEXT(r, entries);
3129 		else if (r->os_fingerprint != PF_OSFP_ANY && (pd->proto !=
3130 		    IPPROTO_TCP || !pf_osfp_match(pf_osfp_fingerprint(pd, m,
3131 		    off, pd->hdr.tcp), r->os_fingerprint)))
3132 			r = TAILQ_NEXT(r, entries);
3133 		else {
3134 			if (r->tag)
3135 				tag = r->tag;
3136 			if (r->rtableid >= 0)
3137 				rtableid = r->rtableid;
3138 			if (r->anchor == NULL) {
3139 				rm = r;
3140 			} else
3141 				pf_step_into_anchor(&asd, &ruleset, rs_num,
3142 				    &r, NULL, NULL);
3143 		}
3144 		if (r == NULL)
3145 			pf_step_out_of_anchor(&asd, &ruleset, rs_num, &r,
3146 			    NULL, NULL);
3147 	}
3148 	if (pf_tag_packet(m, tag, rtableid))
3149 		return (NULL);
3150 	if (rm != NULL && (rm->action == PF_NONAT ||
3151 	    rm->action == PF_NORDR || rm->action == PF_NOBINAT))
3152 		return (NULL);
3153 	return (rm);
3154 }
3155 
3156 struct pf_rule *
pf_get_translation(struct pf_pdesc * pd,struct mbuf * m,int off,int direction,struct pfi_kif * kif,struct pf_src_node ** sn,struct pf_state_key ** skw,struct pf_state_key ** sks,struct pf_state_key ** skp,struct pf_state_key ** nkp,struct pf_addr * saddr,struct pf_addr * daddr,u_int16_t sport,u_int16_t dport)3157 pf_get_translation(struct pf_pdesc *pd, struct mbuf *m, int off, int direction,
3158     struct pfi_kif *kif, struct pf_src_node **sn,
3159     struct pf_state_key **skw, struct pf_state_key **sks,
3160     struct pf_state_key **skp, struct pf_state_key **nkp,
3161     struct pf_addr *saddr, struct pf_addr *daddr,
3162     u_int16_t sport, u_int16_t dport)
3163 {
3164 	struct pf_rule	*r = NULL;
3165 
3166 	if (direction == PF_OUT) {
3167 		r = pf_match_translation(pd, m, off, direction, kif, saddr,
3168 		    sport, daddr, dport, PF_RULESET_BINAT);
3169 		if (r == NULL)
3170 			r = pf_match_translation(pd, m, off, direction, kif,
3171 			    saddr, sport, daddr, dport, PF_RULESET_NAT);
3172 	} else {
3173 		r = pf_match_translation(pd, m, off, direction, kif, saddr,
3174 		    sport, daddr, dport, PF_RULESET_RDR);
3175 		if (r == NULL)
3176 			r = pf_match_translation(pd, m, off, direction, kif,
3177 			    saddr, sport, daddr, dport, PF_RULESET_BINAT);
3178 	}
3179 
3180 	if (r != NULL) {
3181 		struct pf_addr	*naddr;
3182 		u_int16_t	*nport;
3183 
3184 		if (pf_state_key_setup(pd, r, skw, sks, skp, nkp,
3185 		    saddr, daddr, sport, dport))
3186 			return r;
3187 
3188 		/* XXX We only modify one side for now. */
3189 		naddr = &(*nkp)->addr[1];
3190 		nport = &(*nkp)->port[1];
3191 
3192 		/*
3193 		 * NOTE: Currently all translations will clear
3194 		 *	 BRIDGE_MBUF_TAGGED, telling the bridge to
3195 		 *	 ignore the original input encapsulation.
3196 		 */
3197 		switch (r->action) {
3198 		case PF_NONAT:
3199 		case PF_NOBINAT:
3200 		case PF_NORDR:
3201 			return (NULL);
3202 		case PF_NAT:
3203 			m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
3204 			if (pf_get_sport(pd, pd->af, pd->proto, r,
3205 			    saddr, daddr, sport, dport,
3206 			    naddr, nport, r->rpool.proxy_port[0],
3207 			    r->rpool.proxy_port[1], sn)) {
3208 				DPFPRINTF(PF_DEBUG_MISC,
3209 				    ("pf: NAT proxy port allocation "
3210 				    "(%u-%u) failed\n",
3211 				    r->rpool.proxy_port[0],
3212 				    r->rpool.proxy_port[1]));
3213 				return (NULL);
3214 			}
3215 			break;
3216 		case PF_BINAT:
3217 			m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
3218 			switch (direction) {
3219 			case PF_OUT:
3220 				if (r->rpool.cur->addr.type == PF_ADDR_DYNIFTL){
3221 					switch (pd->af) {
3222 #ifdef INET
3223 					case AF_INET:
3224 						if (r->rpool.cur->addr.p.dyn->
3225 						    pfid_acnt4 < 1)
3226 							return (NULL);
3227 						PF_POOLMASK(naddr,
3228 						    &r->rpool.cur->addr.p.dyn->
3229 						    pfid_addr4,
3230 						    &r->rpool.cur->addr.p.dyn->
3231 						    pfid_mask4,
3232 						    saddr, AF_INET);
3233 						break;
3234 #endif /* INET */
3235 #ifdef INET6
3236 					case AF_INET6:
3237 						if (r->rpool.cur->addr.p.dyn->
3238 						    pfid_acnt6 < 1)
3239 							return (NULL);
3240 						PF_POOLMASK(naddr,
3241 						    &r->rpool.cur->addr.p.dyn->
3242 						    pfid_addr6,
3243 						    &r->rpool.cur->addr.p.dyn->
3244 						    pfid_mask6,
3245 						    saddr, AF_INET6);
3246 						break;
3247 #endif /* INET6 */
3248 					}
3249 				} else
3250 					PF_POOLMASK(naddr,
3251 					    &r->rpool.cur->addr.v.a.addr,
3252 					    &r->rpool.cur->addr.v.a.mask,
3253 					    saddr, pd->af);
3254 				break;
3255 			case PF_IN:
3256 				if (r->src.addr.type == PF_ADDR_DYNIFTL) {
3257 					switch (pd->af) {
3258 #ifdef INET
3259 					case AF_INET:
3260 						if (r->src.addr.p.dyn->
3261 						    pfid_acnt4 < 1)
3262 							return (NULL);
3263 						PF_POOLMASK(naddr,
3264 						    &r->src.addr.p.dyn->
3265 						    pfid_addr4,
3266 						    &r->src.addr.p.dyn->
3267 						    pfid_mask4,
3268 						    daddr, AF_INET);
3269 						break;
3270 #endif /* INET */
3271 #ifdef INET6
3272 					case AF_INET6:
3273 						if (r->src.addr.p.dyn->
3274 						    pfid_acnt6 < 1)
3275 							return (NULL);
3276 						PF_POOLMASK(naddr,
3277 						    &r->src.addr.p.dyn->
3278 						    pfid_addr6,
3279 						    &r->src.addr.p.dyn->
3280 						    pfid_mask6,
3281 						    daddr, AF_INET6);
3282 						break;
3283 #endif /* INET6 */
3284 					}
3285 				} else
3286 					PF_POOLMASK(naddr,
3287 					    &r->src.addr.v.a.addr,
3288 					    &r->src.addr.v.a.mask, daddr,
3289 					    pd->af);
3290 				break;
3291 			}
3292 			break;
3293 		case PF_RDR: {
3294 			m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
3295 			if (pf_map_addr(pd->af, r, saddr, naddr, NULL, sn))
3296 				return (NULL);
3297 			if ((r->rpool.opts & PF_POOL_TYPEMASK) ==
3298 			    PF_POOL_BITMASK)
3299 				PF_POOLMASK(naddr, naddr,
3300 				    &r->rpool.cur->addr.v.a.mask, daddr,
3301 				    pd->af);
3302 
3303 			if (r->rpool.proxy_port[1]) {
3304 				u_int32_t	tmp_nport;
3305 
3306 				tmp_nport = ((ntohs(dport) -
3307 				    ntohs(r->dst.port[0])) %
3308 				    (r->rpool.proxy_port[1] -
3309 				    r->rpool.proxy_port[0] + 1)) +
3310 				    r->rpool.proxy_port[0];
3311 
3312 				/* wrap around if necessary */
3313 				if (tmp_nport > 65535)
3314 					tmp_nport -= 65535;
3315 				*nport = htons((u_int16_t)tmp_nport);
3316 			} else if (r->rpool.proxy_port[0]) {
3317 				*nport = htons(r->rpool.proxy_port[0]);
3318 			}
3319 			pd->not_cpu_localized = 1;
3320 			break;
3321 		}
3322 		default:
3323 			return (NULL);
3324 		}
3325 	}
3326 
3327 	return (r);
3328 }
3329 
3330 struct netmsg_hashlookup {
3331 	struct netmsg_base	base;
3332 	struct inpcb		**nm_pinp;
3333 	struct inpcbinfo    	*nm_pcbinfo;
3334 	struct pf_addr		*nm_saddr;
3335 	struct pf_addr		*nm_daddr;
3336 	uint16_t		nm_sport;
3337 	uint16_t		nm_dport;
3338 	sa_family_t		nm_af;
3339 };
3340 
3341 #ifdef PF_SOCKET_LOOKUP_DOMSG
3342 static void
in_pcblookup_hash_handler(netmsg_t msg)3343 in_pcblookup_hash_handler(netmsg_t msg)
3344 {
3345 	struct netmsg_hashlookup *rmsg = (struct netmsg_hashlookup *)msg;
3346 
3347 	if (rmsg->nm_af == AF_INET)
3348 		*rmsg->nm_pinp = in_pcblookup_hash(rmsg->nm_pcbinfo,
3349 		    rmsg->nm_saddr->v4, rmsg->nm_sport, rmsg->nm_daddr->v4,
3350 		    rmsg->nm_dport, INPLOOKUP_WILDCARD, NULL);
3351 #ifdef INET6
3352 	else
3353 		*rmsg->nm_pinp = in6_pcblookup_hash(rmsg->nm_pcbinfo,
3354 		    &rmsg->nm_saddr->v6, rmsg->nm_sport, &rmsg->nm_daddr->v6,
3355 		    rmsg->nm_dport, INPLOOKUP_WILDCARD, NULL);
3356 #endif /* INET6 */
3357 	lwkt_replymsg(&rmsg->base.lmsg, 0);
3358 }
3359 #endif	/* PF_SOCKET_LOOKUP_DOMSG */
3360 
3361 int
pf_socket_lookup(int direction,struct pf_pdesc * pd)3362 pf_socket_lookup(int direction, struct pf_pdesc *pd)
3363 {
3364 	struct pf_addr		*saddr, *daddr;
3365 	u_int16_t		 sport, dport;
3366 	struct inpcbinfo	*pi;
3367 	struct inpcb		*inp;
3368 	struct netmsg_hashlookup *msg = NULL;
3369 #ifdef PF_SOCKET_LOOKUP_DOMSG
3370 	struct netmsg_hashlookup msg0;
3371 #endif
3372 	int			 pi_cpu = 0;
3373 
3374 	if (pd == NULL)
3375 		return (-1);
3376 	pd->lookup.uid = UID_MAX;
3377 	pd->lookup.gid = GID_MAX;
3378 	pd->lookup.pid = NO_PID;
3379 	if (direction == PF_IN) {
3380 		saddr = pd->src;
3381 		daddr = pd->dst;
3382 	} else {
3383 		saddr = pd->dst;
3384 		daddr = pd->src;
3385 	}
3386 	switch (pd->proto) {
3387 	case IPPROTO_TCP:
3388 		if (pd->hdr.tcp == NULL)
3389 			return (-1);
3390 		sport = pd->hdr.tcp->th_sport;
3391 		dport = pd->hdr.tcp->th_dport;
3392 
3393 		pi_cpu = tcp_addrcpu(saddr->v4.s_addr, sport, daddr->v4.s_addr, dport);
3394 		pi = &tcbinfo[pi_cpu];
3395 		/*
3396 		 * Our netstack runs lockless on MP systems
3397 		 * (only for TCP connections at the moment).
3398 		 *
3399 		 * As we are not allowed to read another CPU's tcbinfo,
3400 		 * we have to ask that CPU via remote call to search the
3401 		 * table for us.
3402 		 *
3403 		 * Prepare a msg iff data belongs to another CPU.
3404 		 */
3405 		if (pi_cpu != mycpu->gd_cpuid) {
3406 #ifdef PF_SOCKET_LOOKUP_DOMSG
3407 			/*
3408 			 * NOTE:
3409 			 *
3410 			 * Following lwkt_domsg() is dangerous and could
3411 			 * lockup the network system, e.g.
3412 			 *
3413 			 * On 2 CPU system:
3414 			 * netisr0 domsg to netisr1 (due to lookup)
3415 			 * netisr1 domsg to netisr0 (due to lookup)
3416 			 *
3417 			 * We simply return -1 here, since we are probably
3418 			 * called before NAT, so the TCP packet should
3419 			 * already be on the correct CPU.
3420 			 */
3421 			msg = &msg0;
3422 			netmsg_init(&msg->base, NULL, &curthread->td_msgport,
3423 				    0, in_pcblookup_hash_handler);
3424 			msg->nm_pinp = &inp;
3425 			msg->nm_pcbinfo = pi;
3426 			msg->nm_saddr = saddr;
3427 			msg->nm_sport = sport;
3428 			msg->nm_daddr = daddr;
3429 			msg->nm_dport = dport;
3430 			msg->nm_af = pd->af;
3431 #else	/* !PF_SOCKET_LOOKUP_DOMSG */
3432 			kprintf("pf_socket_lookup: tcp packet not on the "
3433 				"correct cpu %d, cur cpu %d\n",
3434 				pi_cpu, mycpuid);
3435 			print_backtrace(-1);
3436 			return -1;
3437 #endif	/* PF_SOCKET_LOOKUP_DOMSG */
3438 		}
3439 		break;
3440 	case IPPROTO_UDP:
3441 		if (pd->hdr.udp == NULL)
3442 			return (-1);
3443 		sport = pd->hdr.udp->uh_sport;
3444 		dport = pd->hdr.udp->uh_dport;
3445 		pi = &udbinfo[mycpuid];
3446 		break;
3447 	default:
3448 		return (-1);
3449 	}
3450 	if (direction != PF_IN) {
3451 		u_int16_t	p;
3452 
3453 		p = sport;
3454 		sport = dport;
3455 		dport = p;
3456 	}
3457 	switch (pd->af) {
3458 #ifdef INET6
3459 	case AF_INET6:
3460 		/*
3461 		 * Query other CPU, second part
3462 		 *
3463 		 * msg only gets initialized when:
3464 		 * 1) packet is TCP
3465 		 * 2) the info belongs to another CPU
3466 		 *
3467 		 * Use some switch/case magic to avoid code duplication.
3468 		 */
3469 		if (msg == NULL) {
3470 			inp = in6_pcblookup_hash(pi, &saddr->v6, sport,
3471 			    &daddr->v6, dport, INPLOOKUP_WILDCARD, NULL);
3472 
3473 			if (inp == NULL)
3474 				return (-1);
3475 			break;
3476 		}
3477 		/* FALLTHROUGH if SMP and on other CPU */
3478 #endif /* INET6 */
3479 	case AF_INET:
3480 		if (msg != NULL) {
3481 			lwkt_domsg(netisr_cpuport(pi_cpu),
3482 				     &msg->base.lmsg, 0);
3483 		} else
3484 		{
3485 			inp = in_pcblookup_hash(pi, saddr->v4, sport, daddr->v4,
3486 			    dport, INPLOOKUP_WILDCARD, NULL);
3487 		}
3488 		if (inp == NULL)
3489 			return (-1);
3490 		break;
3491 
3492 	default:
3493 		return (-1);
3494 	}
3495 	pd->lookup.uid = inp->inp_socket->so_cred->cr_uid;
3496 	pd->lookup.gid = inp->inp_socket->so_cred->cr_groups[0];
3497 	return (1);
3498 }
3499 
3500 u_int8_t
pf_get_wscale(struct mbuf * m,int off,u_int16_t th_off,sa_family_t af)3501 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3502 {
3503 	int		 hlen;
3504 	u_int8_t	 hdr[60];
3505 	u_int8_t	*opt, optlen;
3506 	u_int8_t	 wscale = 0;
3507 
3508 	hlen = th_off << 2;		/* hlen <= sizeof(hdr) */
3509 	if (hlen <= sizeof(struct tcphdr))
3510 		return (0);
3511 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3512 		return (0);
3513 	opt = hdr + sizeof(struct tcphdr);
3514 	hlen -= sizeof(struct tcphdr);
3515 	while (hlen >= 3) {
3516 		switch (*opt) {
3517 		case TCPOPT_EOL:
3518 		case TCPOPT_NOP:
3519 			++opt;
3520 			--hlen;
3521 			break;
3522 		case TCPOPT_WINDOW:
3523 			wscale = opt[2];
3524 			if (wscale > TCP_MAX_WINSHIFT)
3525 				wscale = TCP_MAX_WINSHIFT;
3526 			wscale |= PF_WSCALE_FLAG;
3527 			/* FALLTHROUGH */
3528 		default:
3529 			optlen = opt[1];
3530 			if (optlen < 2)
3531 				optlen = 2;
3532 			hlen -= optlen;
3533 			opt += optlen;
3534 			break;
3535 		}
3536 	}
3537 	return (wscale);
3538 }
3539 
3540 u_int16_t
pf_get_mss(struct mbuf * m,int off,u_int16_t th_off,sa_family_t af)3541 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3542 {
3543 	int		 hlen;
3544 	u_int8_t	 hdr[60];
3545 	u_int8_t	*opt, optlen;
3546 	u_int16_t	 mss = tcp_mssdflt;
3547 
3548 	hlen = th_off << 2;	/* hlen <= sizeof(hdr) */
3549 	if (hlen <= sizeof(struct tcphdr))
3550 		return (0);
3551 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3552 		return (0);
3553 	opt = hdr + sizeof(struct tcphdr);
3554 	hlen -= sizeof(struct tcphdr);
3555 	while (hlen >= TCPOLEN_MAXSEG) {
3556 		switch (*opt) {
3557 		case TCPOPT_EOL:
3558 		case TCPOPT_NOP:
3559 			++opt;
3560 			--hlen;
3561 			break;
3562 		case TCPOPT_MAXSEG:
3563 			bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
3564 			/* FALLTHROUGH */
3565 		default:
3566 			optlen = opt[1];
3567 			if (optlen < 2)
3568 				optlen = 2;
3569 			hlen -= optlen;
3570 			opt += optlen;
3571 			break;
3572 		}
3573 	}
3574 	return (mss);
3575 }
3576 
3577 u_int16_t
pf_calc_mss(struct pf_addr * addr,sa_family_t af,u_int16_t offer)3578 pf_calc_mss(struct pf_addr *addr, sa_family_t af, u_int16_t offer)
3579 {
3580 #ifdef INET
3581 	struct sockaddr_in	*dst;
3582 	struct route		 ro;
3583 #endif /* INET */
3584 #ifdef INET6
3585 	struct sockaddr_in6	*dst6;
3586 	struct route_in6	 ro6;
3587 #endif /* INET6 */
3588 	struct rtentry		*rt = NULL;
3589 	int			 hlen = 0;
3590 	u_int16_t		 mss = tcp_mssdflt;
3591 
3592 	switch (af) {
3593 #ifdef INET
3594 	case AF_INET:
3595 		hlen = sizeof(struct ip);
3596 		bzero(&ro, sizeof(ro));
3597 		dst = (struct sockaddr_in *)&ro.ro_dst;
3598 		dst->sin_family = AF_INET;
3599 		dst->sin_len = sizeof(*dst);
3600 		dst->sin_addr = addr->v4;
3601 		rtalloc_ign(&ro, (RTF_CLONING | RTF_PRCLONING));
3602 		rt = ro.ro_rt;
3603 		break;
3604 #endif /* INET */
3605 #ifdef INET6
3606 	case AF_INET6:
3607 		hlen = sizeof(struct ip6_hdr);
3608 		bzero(&ro6, sizeof(ro6));
3609 		dst6 = (struct sockaddr_in6 *)&ro6.ro_dst;
3610 		dst6->sin6_family = AF_INET6;
3611 		dst6->sin6_len = sizeof(*dst6);
3612 		dst6->sin6_addr = addr->v6;
3613 		rtalloc_ign((struct route *)&ro6, (RTF_CLONING | RTF_PRCLONING));
3614 		rt = ro6.ro_rt;
3615 		break;
3616 #endif /* INET6 */
3617 	}
3618 
3619 	if (rt && rt->rt_ifp) {
3620 		mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr);
3621 		mss = max(tcp_mssdflt, mss);
3622 		RTFREE(rt);
3623 	}
3624 	mss = min(mss, offer);
3625 	mss = max(mss, 64);		/* sanity - at least max opt space */
3626 	return (mss);
3627 }
3628 
3629 void
pf_set_rt_ifp(struct pf_state * s,struct pf_addr * saddr)3630 pf_set_rt_ifp(struct pf_state *s, struct pf_addr *saddr)
3631 {
3632 	struct pf_rule *r = s->rule.ptr;
3633 
3634 	s->rt_kif = NULL;
3635 	if (!r->rt || r->rt == PF_FASTROUTE)
3636 		return;
3637 	switch (s->key[PF_SK_WIRE]->af) {
3638 #ifdef INET
3639 	case AF_INET:
3640 		pf_map_addr(AF_INET, r, saddr, &s->rt_addr, NULL,
3641 		    &s->nat_src_node);
3642 		s->rt_kif = r->rpool.cur->kif;
3643 		break;
3644 #endif /* INET */
3645 #ifdef INET6
3646 	case AF_INET6:
3647 		pf_map_addr(AF_INET6, r, saddr, &s->rt_addr, NULL,
3648 		    &s->nat_src_node);
3649 		s->rt_kif = r->rpool.cur->kif;
3650 		break;
3651 #endif /* INET6 */
3652 	}
3653 }
3654 
3655 u_int32_t
pf_tcp_iss(struct pf_pdesc * pd)3656 pf_tcp_iss(struct pf_pdesc *pd)
3657 {
3658 	MD5_CTX ctx;
3659 	u_int32_t digest[4];
3660 
3661 	if (pf_tcp_secret_init == 0) {
3662 		lwkt_gettoken(&pf_gtoken);
3663 		if (pf_tcp_secret_init == 0) {
3664 			karc4random_buf(pf_tcp_secret, sizeof(pf_tcp_secret));
3665 			MD5Init(&pf_tcp_secret_ctx);
3666 			MD5Update(&pf_tcp_secret_ctx, pf_tcp_secret,
3667 			    sizeof(pf_tcp_secret));
3668 			pf_tcp_secret_init = 1;
3669 		}
3670 		lwkt_reltoken(&pf_gtoken);
3671 	}
3672 	ctx = pf_tcp_secret_ctx;
3673 
3674 	MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
3675 	MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
3676 	if (pd->af == AF_INET6) {
3677 		MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3678 		MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3679 	} else {
3680 		MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3681 		MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3682 	}
3683 	MD5Final((u_char *)digest, &ctx);
3684 	pf_tcp_iss_off += 4096;
3685 
3686 	return (digest[0] + pd->hdr.tcp->th_seq + pf_tcp_iss_off);
3687 }
3688 
3689 int
pf_test_rule(struct pf_rule ** rm,struct pf_state ** sm,int direction,struct pfi_kif * kif,struct mbuf * m,int off,void * h,struct pf_pdesc * pd,struct pf_rule ** am,struct pf_ruleset ** rsm,struct ifqueue * ifq,struct inpcb * inp)3690 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
3691     struct pfi_kif *kif, struct mbuf *m, int off, void *h,
3692     struct pf_pdesc *pd, struct pf_rule **am, struct pf_ruleset **rsm,
3693     struct ifqueue *ifq, struct inpcb *inp)
3694 {
3695 	struct pf_rule		*nr = NULL;
3696 	struct pf_addr		*saddr = pd->src, *daddr = pd->dst;
3697 	sa_family_t		 af = pd->af;
3698 	struct pf_rule		*r, *a = NULL;
3699 	struct pf_ruleset	*ruleset = NULL;
3700 	struct pf_src_node	*nsn = NULL;
3701 	struct tcphdr		*th = pd->hdr.tcp;
3702 	struct pf_state_key	*skw = NULL, *sks = NULL;
3703 	struct pf_state_key	*sk = NULL, *nk = NULL;
3704 	u_short			 reason;
3705 	int			 rewrite = 0, hdrlen = 0;
3706 	int			 tag = -1, rtableid = -1;
3707 	int			 asd = 0;
3708 	int			 match = 0;
3709 	int			 state_icmp = 0;
3710 	u_int16_t		 sport = 0, dport = 0;
3711 	u_int16_t		 bproto_sum = 0, bip_sum = 0;
3712 	u_int8_t		 icmptype = 0, icmpcode = 0;
3713 
3714 
3715 	if (direction == PF_IN && pf_check_congestion(ifq)) {
3716 		REASON_SET(&reason, PFRES_CONGEST);
3717 		return (PF_DROP);
3718 	}
3719 
3720 	if (inp != NULL)
3721 		pd->lookup.done = pf_socket_lookup(direction, pd);
3722 	else if (debug_pfugidhack) {
3723 		DPFPRINTF(PF_DEBUG_MISC, ("pf: unlocked lookup\n"));
3724 		pd->lookup.done = pf_socket_lookup(direction, pd);
3725 	}
3726 
3727 	switch (pd->proto) {
3728 	case IPPROTO_TCP:
3729 		sport = th->th_sport;
3730 		dport = th->th_dport;
3731 		hdrlen = sizeof(*th);
3732 		break;
3733 	case IPPROTO_UDP:
3734 		sport = pd->hdr.udp->uh_sport;
3735 		dport = pd->hdr.udp->uh_dport;
3736 		hdrlen = sizeof(*pd->hdr.udp);
3737 		break;
3738 #ifdef INET
3739 	case IPPROTO_ICMP:
3740 		if (pd->af != AF_INET)
3741 			break;
3742 		sport = dport = pd->hdr.icmp->icmp_id;
3743 		hdrlen = sizeof(*pd->hdr.icmp);
3744 		icmptype = pd->hdr.icmp->icmp_type;
3745 		icmpcode = pd->hdr.icmp->icmp_code;
3746 
3747 		if (icmptype == ICMP_UNREACH ||
3748 		    icmptype == ICMP_SOURCEQUENCH ||
3749 		    icmptype == ICMP_REDIRECT ||
3750 		    icmptype == ICMP_TIMXCEED ||
3751 		    icmptype == ICMP_PARAMPROB)
3752 			state_icmp++;
3753 		break;
3754 #endif /* INET */
3755 #ifdef INET6
3756 	case IPPROTO_ICMPV6:
3757 		if (af != AF_INET6)
3758 			break;
3759 		sport = dport = pd->hdr.icmp6->icmp6_id;
3760 		hdrlen = sizeof(*pd->hdr.icmp6);
3761 		icmptype = pd->hdr.icmp6->icmp6_type;
3762 		icmpcode = pd->hdr.icmp6->icmp6_code;
3763 
3764 		if (icmptype == ICMP6_DST_UNREACH ||
3765 		    icmptype == ICMP6_PACKET_TOO_BIG ||
3766 		    icmptype == ICMP6_TIME_EXCEEDED ||
3767 		    icmptype == ICMP6_PARAM_PROB)
3768 			state_icmp++;
3769 		break;
3770 #endif /* INET6 */
3771 	default:
3772 		sport = dport = hdrlen = 0;
3773 		break;
3774 	}
3775 
3776 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3777 
3778 	/* check packet for BINAT/NAT/RDR */
3779 	if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn,
3780 	    &skw, &sks, &sk, &nk, saddr, daddr, sport, dport)) != NULL) {
3781 		if (nk == NULL || sk == NULL) {
3782 			REASON_SET(&reason, PFRES_MEMORY);
3783 			goto cleanup;
3784 		}
3785 
3786 		if (pd->ip_sum)
3787 			bip_sum = *pd->ip_sum;
3788 
3789 		m->m_flags &= ~M_HASH;
3790 		switch (pd->proto) {
3791 		case IPPROTO_TCP:
3792 			bproto_sum = th->th_sum;
3793 			pd->proto_sum = &th->th_sum;
3794 
3795 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3796 			    nk->port[pd->sidx] != sport) {
3797 				pf_change_ap(saddr, &th->th_sport, pd->ip_sum,
3798 				    &th->th_sum, &nk->addr[pd->sidx],
3799 				    nk->port[pd->sidx], 0, af);
3800 				pd->sport = &th->th_sport;
3801 				sport = th->th_sport;
3802 			}
3803 
3804 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3805 			    nk->port[pd->didx] != dport) {
3806 				pf_change_ap(daddr, &th->th_dport, pd->ip_sum,
3807 				    &th->th_sum, &nk->addr[pd->didx],
3808 				    nk->port[pd->didx], 0, af);
3809 				dport = th->th_dport;
3810 				pd->dport = &th->th_dport;
3811 			}
3812 			rewrite++;
3813 			break;
3814 		case IPPROTO_UDP:
3815 			bproto_sum = pd->hdr.udp->uh_sum;
3816 			pd->proto_sum = &pd->hdr.udp->uh_sum;
3817 
3818 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3819 			    nk->port[pd->sidx] != sport) {
3820 				pf_change_ap(saddr, &pd->hdr.udp->uh_sport,
3821 				    pd->ip_sum, &pd->hdr.udp->uh_sum,
3822 				    &nk->addr[pd->sidx],
3823 				    nk->port[pd->sidx], 1, af);
3824 				sport = pd->hdr.udp->uh_sport;
3825 				pd->sport = &pd->hdr.udp->uh_sport;
3826 			}
3827 
3828 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3829 			    nk->port[pd->didx] != dport) {
3830 				pf_change_ap(daddr, &pd->hdr.udp->uh_dport,
3831 				    pd->ip_sum, &pd->hdr.udp->uh_sum,
3832 				    &nk->addr[pd->didx],
3833 				    nk->port[pd->didx], 1, af);
3834 				dport = pd->hdr.udp->uh_dport;
3835 				pd->dport = &pd->hdr.udp->uh_dport;
3836 			}
3837 			rewrite++;
3838 			break;
3839 #ifdef INET
3840 		case IPPROTO_ICMP:
3841 			nk->port[0] = nk->port[1];
3842 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3843 				pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3844 				    nk->addr[pd->sidx].v4.s_addr, 0);
3845 
3846 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3847 				pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3848 				    nk->addr[pd->didx].v4.s_addr, 0);
3849 
3850 			if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3851 				pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3852 				    pd->hdr.icmp->icmp_cksum, sport,
3853 				    nk->port[1], 0);
3854 				pd->hdr.icmp->icmp_id = nk->port[1];
3855 				pd->sport = &pd->hdr.icmp->icmp_id;
3856 			}
3857 			m_copyback(m, off, ICMP_MINLEN, pd->hdr.icmp);
3858 			break;
3859 #endif /* INET */
3860 #ifdef INET6
3861 		case IPPROTO_ICMPV6:
3862 			nk->port[0] = nk->port[1];
3863 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3864 				pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3865 				    &nk->addr[pd->sidx], 0);
3866 
3867 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3868 				pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3869 				    &nk->addr[pd->didx], 0);
3870 			rewrite++;
3871 			break;
3872 #endif /* INET */
3873 		default:
3874 			switch (af) {
3875 #ifdef INET
3876 			case AF_INET:
3877 				if (PF_ANEQ(saddr,
3878 				    &nk->addr[pd->sidx], AF_INET))
3879 					pf_change_a(&saddr->v4.s_addr,
3880 					    pd->ip_sum,
3881 					    nk->addr[pd->sidx].v4.s_addr, 0);
3882 
3883 				if (PF_ANEQ(daddr,
3884 				    &nk->addr[pd->didx], AF_INET))
3885 					pf_change_a(&daddr->v4.s_addr,
3886 					    pd->ip_sum,
3887 					    nk->addr[pd->didx].v4.s_addr, 0);
3888 				break;
3889 #endif /* INET */
3890 #ifdef INET6
3891 			case AF_INET6:
3892 				if (PF_ANEQ(saddr,
3893 				    &nk->addr[pd->sidx], AF_INET6))
3894 					PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3895 
3896 				if (PF_ANEQ(daddr,
3897 				    &nk->addr[pd->didx], AF_INET6))
3898 					PF_ACPY(saddr, &nk->addr[pd->didx], af);
3899 				break;
3900 #endif /* INET */
3901 			}
3902 			break;
3903 		}
3904 		if (nr->natpass)
3905 			r = NULL;
3906 		pd->nat_rule = nr;
3907 	}
3908 
3909 	while (r != NULL) {
3910 		r->evaluations++;
3911 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
3912 			r = r->skip[PF_SKIP_IFP].ptr;
3913 		else if (r->direction && r->direction != direction)
3914 			r = r->skip[PF_SKIP_DIR].ptr;
3915 		else if (r->af && r->af != af)
3916 			r = r->skip[PF_SKIP_AF].ptr;
3917 		else if (r->proto && r->proto != pd->proto)
3918 			r = r->skip[PF_SKIP_PROTO].ptr;
3919 		else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3920 		    r->src.neg, kif))
3921 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3922 		/* tcp/udp only. port_op always 0 in other cases */
3923 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
3924 		    r->src.port[0], r->src.port[1], sport))
3925 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
3926 		else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3927 		    r->dst.neg, NULL))
3928 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
3929 		/* tcp/udp only. port_op always 0 in other cases */
3930 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3931 		    r->dst.port[0], r->dst.port[1], dport))
3932 			r = r->skip[PF_SKIP_DST_PORT].ptr;
3933 		/* icmp only. type always 0 in other cases */
3934 		else if (r->type && r->type != icmptype + 1)
3935 			r = TAILQ_NEXT(r, entries);
3936 		/* icmp only. type always 0 in other cases */
3937 		else if (r->code && r->code != icmpcode + 1)
3938 			r = TAILQ_NEXT(r, entries);
3939 		else if (r->tos && !(r->tos == pd->tos))
3940 			r = TAILQ_NEXT(r, entries);
3941 		else if (r->rule_flag & PFRULE_FRAGMENT)
3942 			r = TAILQ_NEXT(r, entries);
3943 		else if (pd->proto == IPPROTO_TCP &&
3944 		    (r->flagset & th->th_flags) != r->flags)
3945 			r = TAILQ_NEXT(r, entries);
3946 		/* tcp/udp only. uid.op always 0 in other cases */
3947 		else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3948 		    pf_socket_lookup(direction, pd), 1)) &&
3949 		    !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3950 		    pd->lookup.uid))
3951 			r = TAILQ_NEXT(r, entries);
3952 		/* tcp/udp only. gid.op always 0 in other cases */
3953 		else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3954 		    pf_socket_lookup(direction, pd), 1)) &&
3955 		    !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3956 		    pd->lookup.gid))
3957 			r = TAILQ_NEXT(r, entries);
3958 		else if (r->prob &&
3959 		  r->prob <= karc4random())
3960 			r = TAILQ_NEXT(r, entries);
3961 		else if (r->match_tag && !pf_match_tag(m, r, &tag))
3962 			r = TAILQ_NEXT(r, entries);
3963 		else if (r->os_fingerprint != PF_OSFP_ANY &&
3964 		    (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3965 		    pf_osfp_fingerprint(pd, m, off, th),
3966 		    r->os_fingerprint)))
3967 			r = TAILQ_NEXT(r, entries);
3968 		else {
3969 			if (r->tag)
3970 				tag = r->tag;
3971 			if (r->rtableid >= 0)
3972 				rtableid = r->rtableid;
3973 			if (r->anchor == NULL) {
3974 				match = 1;
3975 				*rm = r;
3976 				*am = a;
3977 				*rsm = ruleset;
3978 				if ((*rm)->quick)
3979 					break;
3980 				r = TAILQ_NEXT(r, entries);
3981 			} else
3982 				pf_step_into_anchor(&asd, &ruleset,
3983 				    PF_RULESET_FILTER, &r, &a, &match);
3984 		}
3985 		if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset,
3986 		    PF_RULESET_FILTER, &r, &a, &match))
3987 			break;
3988 	}
3989 	r = *rm;
3990 	a = *am;
3991 	ruleset = *rsm;
3992 
3993 	REASON_SET(&reason, PFRES_MATCH);
3994 
3995 	if (r->log || (nr != NULL && nr->log)) {
3996 		if (rewrite)
3997 			m_copyback(m, off, hdrlen, pd->hdr.any);
3998 		PFLOG_PACKET(kif, h, m, af, direction, reason, r->log ? r : nr,
3999 		    a, ruleset, pd);
4000 	}
4001 
4002 	if ((r->action == PF_DROP) &&
4003 	    ((r->rule_flag & PFRULE_RETURNRST) ||
4004 	    (r->rule_flag & PFRULE_RETURNICMP) ||
4005 	    (r->rule_flag & PFRULE_RETURN))) {
4006 		/* undo NAT changes, if they have taken place */
4007 		if (nr != NULL) {
4008 			PF_ACPY(saddr, &sk->addr[pd->sidx], af);
4009 			PF_ACPY(daddr, &sk->addr[pd->didx], af);
4010 			if (pd->sport)
4011 				*pd->sport = sk->port[pd->sidx];
4012 			if (pd->dport)
4013 				*pd->dport = sk->port[pd->didx];
4014 			if (pd->proto_sum)
4015 				*pd->proto_sum = bproto_sum;
4016 			if (pd->ip_sum)
4017 				*pd->ip_sum = bip_sum;
4018 			m_copyback(m, off, hdrlen, pd->hdr.any);
4019 		}
4020 		if (pd->proto == IPPROTO_TCP &&
4021 		    ((r->rule_flag & PFRULE_RETURNRST) ||
4022 		    (r->rule_flag & PFRULE_RETURN)) &&
4023 		    !(th->th_flags & TH_RST)) {
4024 			u_int32_t	 ack = ntohl(th->th_seq) + pd->p_len;
4025 			int		 len = 0;
4026 			struct ip	*h4;
4027 #ifdef INET6
4028 			struct ip6_hdr	*h6;
4029 #endif
4030 			switch (af) {
4031 			case AF_INET:
4032 				h4 = mtod(m, struct ip *);
4033 				len = ntohs(h4->ip_len) - off;
4034 				break;
4035 #ifdef INET6
4036 			case AF_INET6:
4037 				h6 = mtod(m, struct ip6_hdr *);
4038 				len = h6->ip6_plen - (off - sizeof(*h6));
4039 				break;
4040 #endif
4041 			}
4042 
4043 			if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
4044 				REASON_SET(&reason, PFRES_PROTCKSUM);
4045 			else {
4046 				if (th->th_flags & TH_SYN)
4047 					ack++;
4048 				if (th->th_flags & TH_FIN)
4049 					ack++;
4050 				pf_send_tcp(r, af, pd->dst,
4051 				    pd->src, th->th_dport, th->th_sport,
4052 				    ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
4053 				    r->return_ttl, 1, 0, pd->eh, kif->pfik_ifp);
4054 			}
4055 		} else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
4056 		    r->return_icmp)
4057 			pf_send_icmp(m, r->return_icmp >> 8,
4058 			    r->return_icmp & 255, af, r);
4059 		else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
4060 		    r->return_icmp6)
4061 			pf_send_icmp(m, r->return_icmp6 >> 8,
4062 			    r->return_icmp6 & 255, af, r);
4063 	}
4064 
4065 	if (r->action == PF_DROP)
4066 		goto cleanup;
4067 
4068 	if (pf_tag_packet(m, tag, rtableid)) {
4069 		REASON_SET(&reason, PFRES_MEMORY);
4070 		goto cleanup;
4071 	}
4072 
4073 	if (!state_icmp && (r->keep_state || nr != NULL ||
4074 	    (pd->flags & PFDESC_TCP_NORM))) {
4075 		int action;
4076 		action = pf_create_state(r, nr, a, pd, nsn, skw, sks, nk, sk, m,
4077 		    off, sport, dport, &rewrite, kif, sm, tag, bproto_sum,
4078 		    bip_sum, hdrlen);
4079 		if (action != PF_PASS)
4080 			return (action);
4081 	}
4082 
4083 	/* copy back packet headers if we performed NAT operations */
4084 	if (rewrite)
4085 		m_copyback(m, off, hdrlen, pd->hdr.any);
4086 
4087 	return (PF_PASS);
4088 
4089 cleanup:
4090 	if (sk != NULL)
4091 		kfree(sk, M_PFSTATEKEYPL);
4092 	if (nk != NULL)
4093 		kfree(nk, M_PFSTATEKEYPL);
4094 	return (PF_DROP);
4095 }
4096 
4097 static __inline int
pf_create_state(struct pf_rule * r,struct pf_rule * nr,struct pf_rule * a,struct pf_pdesc * pd,struct pf_src_node * nsn,struct pf_state_key * skw,struct pf_state_key * sks,struct pf_state_key * nk,struct pf_state_key * sk,struct mbuf * m,int off,u_int16_t sport,u_int16_t dport,int * rewrite,struct pfi_kif * kif,struct pf_state ** sm,int tag,u_int16_t bproto_sum,u_int16_t bip_sum,int hdrlen)4098 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
4099     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *skw,
4100     struct pf_state_key *sks, struct pf_state_key *nk, struct pf_state_key *sk,
4101     struct mbuf *m, int off, u_int16_t sport, u_int16_t dport, int *rewrite,
4102     struct pfi_kif *kif, struct pf_state **sm, int tag, u_int16_t bproto_sum,
4103     u_int16_t bip_sum, int hdrlen)
4104 {
4105 	struct pf_state		*s = NULL;
4106 	struct pf_src_node	*sn = NULL;
4107 	struct tcphdr		*th = pd->hdr.tcp;
4108 	u_int16_t		 mss = tcp_mssdflt;
4109 	u_short			 reason;
4110 	int cpu = mycpu->gd_cpuid;
4111 
4112 	/* check maximums */
4113 	if (r->max_states && (r->states_cur >= r->max_states)) {
4114 		PF_INC_LCOUNTER(LCNT_STATES);
4115 		REASON_SET(&reason, PFRES_MAXSTATES);
4116 		return (PF_DROP);
4117 	}
4118 	/* src node for filter rule */
4119 	if ((r->rule_flag & PFRULE_SRCTRACK ||
4120 	    r->rpool.opts & PF_POOL_STICKYADDR) &&
4121 	    pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
4122 		REASON_SET(&reason, PFRES_SRCLIMIT);
4123 		goto csfailed;
4124 	}
4125 	/* src node for translation rule */
4126 	if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
4127 	    pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
4128 		REASON_SET(&reason, PFRES_SRCLIMIT);
4129 		goto csfailed;
4130 	}
4131 	s = kmalloc(sizeof(struct pf_state), M_PFSTATEPL, M_NOWAIT|M_ZERO);
4132 	if (s == NULL) {
4133 		REASON_SET(&reason, PFRES_MEMORY);
4134 		goto csfailed;
4135 	}
4136 	lockinit(&s->lk, "pfstlk", 0, 0);
4137 	s->id = 0; /* XXX Do we really need that? not in OpenBSD */
4138 	s->creatorid = 0;
4139 	s->rule.ptr = r;
4140 	s->nat_rule.ptr = nr;
4141 	s->anchor.ptr = a;
4142 	s->state_flags = PFSTATE_CREATEINPROG;
4143 	STATE_INC_COUNTERS(s);
4144 	if (r->allow_opts)
4145 		s->state_flags |= PFSTATE_ALLOWOPTS;
4146 	if (r->rule_flag & PFRULE_STATESLOPPY)
4147 		s->state_flags |= PFSTATE_SLOPPY;
4148 	if (pd->not_cpu_localized)
4149 		s->state_flags |= PFSTATE_STACK_GLOBAL;
4150 
4151 	s->log = r->log & PF_LOG_ALL;
4152 	if (nr != NULL)
4153 		s->log |= nr->log & PF_LOG_ALL;
4154 	switch (pd->proto) {
4155 	case IPPROTO_TCP:
4156 		s->src.seqlo = ntohl(th->th_seq);
4157 		s->src.seqhi = s->src.seqlo + pd->p_len + 1;
4158 		if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
4159 		    r->keep_state == PF_STATE_MODULATE) {
4160 			/* Generate sequence number modulator */
4161 			if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
4162 			    0)
4163 				s->src.seqdiff = 1;
4164 			pf_change_a(&th->th_seq, &th->th_sum,
4165 			    htonl(s->src.seqlo + s->src.seqdiff), 0);
4166 			*rewrite = 1;
4167 		} else
4168 			s->src.seqdiff = 0;
4169 		if (th->th_flags & TH_SYN) {
4170 			s->src.seqhi++;
4171 			s->src.wscale = pf_get_wscale(m, off,
4172 			    th->th_off, pd->af);
4173 		}
4174 		s->src.max_win = MAX(ntohs(th->th_win), 1);
4175 		if (s->src.wscale & PF_WSCALE_MASK) {
4176 			/* Remove scale factor from initial window */
4177 			int win = s->src.max_win;
4178 			win += 1 << (s->src.wscale & PF_WSCALE_MASK);
4179 			s->src.max_win = (win - 1) >>
4180 			    (s->src.wscale & PF_WSCALE_MASK);
4181 		}
4182 		if (th->th_flags & TH_FIN)
4183 			s->src.seqhi++;
4184 		s->dst.seqhi = 1;
4185 		s->dst.max_win = 1;
4186 		s->src.state = TCPS_SYN_SENT;
4187 		s->dst.state = TCPS_CLOSED;
4188 		s->timeout = PFTM_TCP_FIRST_PACKET;
4189 		break;
4190 	case IPPROTO_UDP:
4191 		s->src.state = PFUDPS_SINGLE;
4192 		s->dst.state = PFUDPS_NO_TRAFFIC;
4193 		s->timeout = PFTM_UDP_FIRST_PACKET;
4194 		break;
4195 	case IPPROTO_ICMP:
4196 #ifdef INET6
4197 	case IPPROTO_ICMPV6:
4198 #endif
4199 		s->timeout = PFTM_ICMP_FIRST_PACKET;
4200 		break;
4201 	default:
4202 		s->src.state = PFOTHERS_SINGLE;
4203 		s->dst.state = PFOTHERS_NO_TRAFFIC;
4204 		s->timeout = PFTM_OTHER_FIRST_PACKET;
4205 	}
4206 
4207 	s->creation = time_second;
4208 	s->expire = time_second;
4209 
4210 	if (sn != NULL) {
4211 		s->src_node = sn;
4212 		s->src_node->states++;
4213 	}
4214 	if (nsn != NULL) {
4215 		/* XXX We only modify one side for now. */
4216 		PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
4217 		s->nat_src_node = nsn;
4218 		s->nat_src_node->states++;
4219 	}
4220 	if (pd->proto == IPPROTO_TCP) {
4221 		if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
4222 		    off, pd, th, &s->src, &s->dst)) {
4223 			REASON_SET(&reason, PFRES_MEMORY);
4224 			pf_src_tree_remove_state(s);
4225 			STATE_DEC_COUNTERS(s);
4226 			kfree(s, M_PFSTATEPL);
4227 			return (PF_DROP);
4228 		}
4229 		if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
4230 		    pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
4231 		    &s->src, &s->dst, rewrite)) {
4232 			/* This really shouldn't happen!!! */
4233 			DPFPRINTF(PF_DEBUG_URGENT,
4234 			    ("pf_normalize_tcp_stateful failed on first pkt"));
4235 			pf_normalize_tcp_cleanup(s);
4236 			pf_src_tree_remove_state(s);
4237 			STATE_DEC_COUNTERS(s);
4238 			kfree(s, M_PFSTATEPL);
4239 			return (PF_DROP);
4240 		}
4241 	}
4242 	s->direction = pd->dir;
4243 
4244 	if (sk == NULL && pf_state_key_setup(pd, nr, &skw, &sks, &sk, &nk,
4245 					     pd->src, pd->dst, sport, dport)) {
4246 		REASON_SET(&reason, PFRES_MEMORY);
4247 		goto csfailed;
4248 	}
4249 
4250 	if (pf_state_insert(BOUND_IFACE(r, kif), skw, sks, s)) {
4251 		if (pd->proto == IPPROTO_TCP)
4252 			pf_normalize_tcp_cleanup(s);
4253 		REASON_SET(&reason, PFRES_STATEINS);
4254 		pf_src_tree_remove_state(s);
4255 		STATE_DEC_COUNTERS(s);
4256 		kfree(s, M_PFSTATEPL);
4257 		return (PF_DROP);
4258 	} else
4259 		*sm = s;
4260 
4261 	pf_set_rt_ifp(s, pd->src);	/* needs s->state_key set */
4262 	if (tag > 0) {
4263 		pf_tag_ref(tag);
4264 		s->tag = tag;
4265 	}
4266 	if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
4267 	    TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
4268 		s->src.state = PF_TCPS_PROXY_SRC;
4269 		/* undo NAT changes, if they have taken place */
4270 		if (nr != NULL) {
4271 			struct pf_state_key *skt = s->key[PF_SK_WIRE];
4272 			if (pd->dir == PF_OUT)
4273 				skt = s->key[PF_SK_STACK];
4274 			PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
4275 			PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
4276 			if (pd->sport)
4277 				*pd->sport = skt->port[pd->sidx];
4278 			if (pd->dport)
4279 				*pd->dport = skt->port[pd->didx];
4280 			if (pd->proto_sum)
4281 				*pd->proto_sum = bproto_sum;
4282 			if (pd->ip_sum)
4283 				*pd->ip_sum = bip_sum;
4284 			m->m_flags &= ~M_HASH;
4285 			m_copyback(m, off, hdrlen, pd->hdr.any);
4286 		}
4287 		s->src.seqhi = htonl(karc4random());
4288 		/* Find mss option */
4289 		mss = pf_get_mss(m, off, th->th_off, pd->af);
4290 		mss = pf_calc_mss(pd->src, pd->af, mss);
4291 		mss = pf_calc_mss(pd->dst, pd->af, mss);
4292 		s->src.mss = mss;
4293 		s->state_flags &= ~PFSTATE_CREATEINPROG;
4294 		pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport,
4295 			    th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
4296 			    TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL, NULL);
4297 		REASON_SET(&reason, PFRES_SYNPROXY);
4298 		return (PF_SYNPROXY_DROP);
4299 	}
4300 
4301 	s->state_flags &= ~PFSTATE_CREATEINPROG;
4302 	return (PF_PASS);
4303 
4304 csfailed:
4305 	if (sk != NULL)
4306 		kfree(sk, M_PFSTATEKEYPL);
4307 	if (nk != NULL)
4308 		kfree(nk, M_PFSTATEKEYPL);
4309 
4310 	if (sn != NULL && sn->states == 0 && sn->expire == 0) {
4311 		RB_REMOVE(pf_src_tree, &tree_src_tracking[cpu], sn);
4312 		PF_INC_SCOUNTER(SCNT_SRC_NODE_REMOVALS);
4313 		atomic_add_int(&pf_status.src_nodes, -1);
4314 		kfree(sn, M_PFSRCTREEPL);
4315 	}
4316 	if (nsn != sn && nsn != NULL && nsn->states == 0 && nsn->expire == 0) {
4317 		RB_REMOVE(pf_src_tree, &tree_src_tracking[cpu], nsn);
4318 		PF_INC_SCOUNTER(SCNT_SRC_NODE_REMOVALS);
4319 		atomic_add_int(&pf_status.src_nodes, -1);
4320 		kfree(nsn, M_PFSRCTREEPL);
4321 	}
4322 	if (s) {
4323 		pf_src_tree_remove_state(s);
4324 		STATE_DEC_COUNTERS(s);
4325 		kfree(s, M_PFSTATEPL);
4326 	}
4327 
4328 	return (PF_DROP);
4329 }
4330 
4331 int
pf_test_fragment(struct pf_rule ** rm,int direction,struct pfi_kif * kif,struct mbuf * m,void * h,struct pf_pdesc * pd,struct pf_rule ** am,struct pf_ruleset ** rsm)4332 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
4333     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
4334     struct pf_ruleset **rsm)
4335 {
4336 	struct pf_rule		*r, *a = NULL;
4337 	struct pf_ruleset	*ruleset = NULL;
4338 	sa_family_t		 af = pd->af;
4339 	u_short			 reason;
4340 	int			 tag = -1;
4341 	int			 asd = 0;
4342 	int			 match = 0;
4343 
4344 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
4345 	while (r != NULL) {
4346 		r->evaluations++;
4347 		if (pfi_kif_match(r->kif, kif) == r->ifnot)
4348 			r = r->skip[PF_SKIP_IFP].ptr;
4349 		else if (r->direction && r->direction != direction)
4350 			r = r->skip[PF_SKIP_DIR].ptr;
4351 		else if (r->af && r->af != af)
4352 			r = r->skip[PF_SKIP_AF].ptr;
4353 		else if (r->proto && r->proto != pd->proto)
4354 			r = r->skip[PF_SKIP_PROTO].ptr;
4355 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
4356 		    r->src.neg, kif))
4357 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
4358 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
4359 		    r->dst.neg, NULL))
4360 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
4361 		else if (r->tos && !(r->tos == pd->tos))
4362 			r = TAILQ_NEXT(r, entries);
4363 		else if (r->os_fingerprint != PF_OSFP_ANY)
4364 			r = TAILQ_NEXT(r, entries);
4365 		else if (pd->proto == IPPROTO_UDP &&
4366 		    (r->src.port_op || r->dst.port_op))
4367 			r = TAILQ_NEXT(r, entries);
4368 		else if (pd->proto == IPPROTO_TCP &&
4369 		    (r->src.port_op || r->dst.port_op || r->flagset))
4370 			r = TAILQ_NEXT(r, entries);
4371 		else if ((pd->proto == IPPROTO_ICMP ||
4372 		    pd->proto == IPPROTO_ICMPV6) &&
4373 		    (r->type || r->code))
4374 			r = TAILQ_NEXT(r, entries);
4375 		else if (r->prob && r->prob <= karc4random())
4376 			r = TAILQ_NEXT(r, entries);
4377 		else if (r->match_tag && !pf_match_tag(m, r, &tag))
4378 			r = TAILQ_NEXT(r, entries);
4379 		else {
4380 			if (r->anchor == NULL) {
4381 				match = 1;
4382 				*rm = r;
4383 				*am = a;
4384 				*rsm = ruleset;
4385 				if ((*rm)->quick)
4386 					break;
4387 				r = TAILQ_NEXT(r, entries);
4388 			} else
4389 				pf_step_into_anchor(&asd, &ruleset,
4390 				    PF_RULESET_FILTER, &r, &a, &match);
4391 		}
4392 		if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset,
4393 		    PF_RULESET_FILTER, &r, &a, &match))
4394 			break;
4395 	}
4396 	r = *rm;
4397 	a = *am;
4398 	ruleset = *rsm;
4399 
4400 	REASON_SET(&reason, PFRES_MATCH);
4401 
4402 	if (r->log)
4403 		PFLOG_PACKET(kif, h, m, af, direction, reason, r, a, ruleset,
4404 		    pd);
4405 
4406 	if (r->action != PF_PASS)
4407 		return (PF_DROP);
4408 
4409 	if (pf_tag_packet(m, tag, -1)) {
4410 		REASON_SET(&reason, PFRES_MEMORY);
4411 		return (PF_DROP);
4412 	}
4413 
4414 	return (PF_PASS);
4415 }
4416 
4417 /*
4418  * Called with state locked
4419  */
4420 int
pf_tcp_track_full(struct pf_state_peer * src,struct pf_state_peer * dst,struct pf_state ** state,struct pfi_kif * kif,struct mbuf * m,int off,struct pf_pdesc * pd,u_short * reason,int * copyback)4421 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
4422 	struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
4423 	struct pf_pdesc *pd, u_short *reason, int *copyback)
4424 {
4425 	struct tcphdr		*th = pd->hdr.tcp;
4426 	u_int16_t		 win = ntohs(th->th_win);
4427 	u_int32_t		 ack, end, seq, orig_seq;
4428 	u_int8_t		 sws, dws;
4429 	int			 ackskew;
4430 
4431 	if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
4432 		sws = src->wscale & PF_WSCALE_MASK;
4433 		dws = dst->wscale & PF_WSCALE_MASK;
4434 	} else {
4435 		sws = dws = 0;
4436 	}
4437 
4438 	/*
4439 	 * Sequence tracking algorithm from Guido van Rooij's paper:
4440 	 *   http://www.madison-gurkha.com/publications/tcp_filtering/
4441 	 *	tcp_filtering.ps
4442 	 */
4443 
4444 	orig_seq = seq = ntohl(th->th_seq);
4445 	if (src->seqlo == 0) {
4446 		/* First packet from this end. Set its state */
4447 
4448 		if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
4449 		    src->scrub == NULL) {
4450 			if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
4451 				REASON_SET(reason, PFRES_MEMORY);
4452 				return (PF_DROP);
4453 			}
4454 		}
4455 
4456 		/* Deferred generation of sequence number modulator */
4457 		if (dst->seqdiff && !src->seqdiff) {
4458 			/* use random iss for the TCP server */
4459 			while ((src->seqdiff = karc4random() - seq) == 0)
4460 				;
4461 			ack = ntohl(th->th_ack) - dst->seqdiff;
4462 			pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
4463 			    src->seqdiff), 0);
4464 			pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
4465 			*copyback = 1;
4466 		} else {
4467 			ack = ntohl(th->th_ack);
4468 		}
4469 
4470 		end = seq + pd->p_len;
4471 		if (th->th_flags & TH_SYN) {
4472 			end++;
4473 			(*state)->sync_flags |= PFSTATE_GOT_SYN2;
4474 			if (dst->wscale & PF_WSCALE_FLAG) {
4475 				src->wscale = pf_get_wscale(m, off, th->th_off,
4476 				    pd->af);
4477 				if (src->wscale & PF_WSCALE_FLAG) {
4478 					/* Remove scale factor from initial
4479 					 * window */
4480 					sws = src->wscale & PF_WSCALE_MASK;
4481 					win = ((u_int32_t)win + (1 << sws) - 1)
4482 					    >> sws;
4483 					dws = dst->wscale & PF_WSCALE_MASK;
4484 				} else {
4485 					/* fixup other window */
4486 					dst->max_win <<= dst->wscale &
4487 					    PF_WSCALE_MASK;
4488 					/* in case of a retrans SYN|ACK */
4489 					dst->wscale = 0;
4490 				}
4491 			}
4492 		}
4493 		if (th->th_flags & TH_FIN)
4494 			end++;
4495 
4496 		src->seqlo = seq;
4497 		if (src->state < TCPS_SYN_SENT)
4498 			src->state = TCPS_SYN_SENT;
4499 
4500 		/*
4501 		 * May need to slide the window (seqhi may have been set by
4502 		 * the crappy stack check or if we picked up the connection
4503 		 * after establishment)
4504 		 */
4505 		if (src->seqhi == 1 ||
4506 		    SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
4507 			src->seqhi = end + MAX(1, dst->max_win << dws);
4508 		if (win > src->max_win)
4509 			src->max_win = win;
4510 
4511 	} else {
4512 		ack = ntohl(th->th_ack) - dst->seqdiff;
4513 		if (src->seqdiff) {
4514 			/* Modulate sequence numbers */
4515 			pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
4516 			    src->seqdiff), 0);
4517 			pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
4518 			*copyback = 1;
4519 		}
4520 		end = seq + pd->p_len;
4521 		if (th->th_flags & TH_SYN)
4522 			end++;
4523 		if (th->th_flags & TH_FIN)
4524 			end++;
4525 	}
4526 
4527 	if ((th->th_flags & TH_ACK) == 0) {
4528 		/* Let it pass through the ack skew check */
4529 		ack = dst->seqlo;
4530 	} else if ((ack == 0 &&
4531 	    (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
4532 	    /* broken tcp stacks do not set ack */
4533 	    (dst->state < TCPS_SYN_SENT)) {
4534 		/*
4535 		 * Many stacks (ours included) will set the ACK number in an
4536 		 * FIN|ACK if the SYN times out -- no sequence to ACK.
4537 		 */
4538 		ack = dst->seqlo;
4539 	}
4540 
4541 	if (seq == end) {
4542 		/* Ease sequencing restrictions on no data packets */
4543 		seq = src->seqlo;
4544 		end = seq;
4545 	}
4546 
4547 	ackskew = dst->seqlo - ack;
4548 
4549 
4550 	/*
4551 	 * Need to demodulate the sequence numbers in any TCP SACK options
4552 	 * (Selective ACK). We could optionally validate the SACK values
4553 	 * against the current ACK window, either forwards or backwards, but
4554 	 * I'm not confident that SACK has been implemented properly
4555 	 * everywhere. It wouldn't surprise me if several stacks accidently
4556 	 * SACK too far backwards of previously ACKed data. There really aren't
4557 	 * any security implications of bad SACKing unless the target stack
4558 	 * doesn't validate the option length correctly. Someone trying to
4559 	 * spoof into a TCP connection won't bother blindly sending SACK
4560 	 * options anyway.
4561 	 */
4562 	if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
4563 		if (pf_modulate_sack(m, off, pd, th, dst))
4564 			*copyback = 1;
4565 	}
4566 
4567 
4568 #define MAXACKWINDOW (0xffff + 1500)	/* 1500 is an arbitrary fudge factor */
4569 	if (SEQ_GEQ(src->seqhi, end) &&
4570 	    /* Last octet inside other's window space */
4571 	    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
4572 	    /* Retrans: not more than one window back */
4573 	    (ackskew >= -MAXACKWINDOW) &&
4574 	    /* Acking not more than one reassembled fragment backwards */
4575 	    (ackskew <= (MAXACKWINDOW << sws)) &&
4576 	    /* Acking not more than one window forward */
4577 	    ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
4578 	    (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo)))
4579 	{
4580 		/*
4581 		 * Require an exact/+1 sequence match on resets
4582 		 * when possible
4583 		 */
4584 		if (dst->scrub || src->scrub) {
4585 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4586 			    *state, src, dst, copyback))
4587 				return (PF_DROP);
4588 		}
4589 
4590 		/* update max window */
4591 		if (src->max_win < win)
4592 			src->max_win = win;
4593 		/* synchronize sequencing */
4594 		if (SEQ_GT(end, src->seqlo))
4595 			src->seqlo = end;
4596 		/* slide the window of what the other end can send */
4597 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4598 			dst->seqhi = ack + MAX((win << sws), 1);
4599 
4600 
4601 		/* update states */
4602 		if (th->th_flags & TH_SYN)
4603 			if (src->state < TCPS_SYN_SENT)
4604 				src->state = TCPS_SYN_SENT;
4605 		if (th->th_flags & TH_FIN)
4606 			if (src->state < TCPS_CLOSING)
4607 				src->state = TCPS_CLOSING;
4608 		if (th->th_flags & TH_ACK) {
4609 			if (dst->state == TCPS_SYN_SENT) {
4610 				dst->state = TCPS_ESTABLISHED;
4611 				if (src->state == TCPS_ESTABLISHED &&
4612 				    (*state)->src_node != NULL &&
4613 				    pf_src_connlimit(*state)) {
4614 					REASON_SET(reason, PFRES_SRCLIMIT);
4615 					return (PF_DROP);
4616 				}
4617 			} else if (dst->state == TCPS_CLOSING)
4618 				dst->state = TCPS_FIN_WAIT_2;
4619 		}
4620 		if (th->th_flags & TH_RST)
4621 			src->state = dst->state = TCPS_TIME_WAIT;
4622 
4623 		/* update expire time */
4624 		(*state)->expire = time_second;
4625 		if (src->state >= TCPS_FIN_WAIT_2 &&
4626 		    dst->state >= TCPS_FIN_WAIT_2)
4627 			(*state)->timeout = PFTM_TCP_CLOSED;
4628 		else if (src->state >= TCPS_CLOSING &&
4629 		    dst->state >= TCPS_CLOSING)
4630 			(*state)->timeout = PFTM_TCP_FIN_WAIT;
4631 		else if (src->state < TCPS_ESTABLISHED ||
4632 		    dst->state < TCPS_ESTABLISHED)
4633 			(*state)->timeout = PFTM_TCP_OPENING;
4634 		else if (src->state >= TCPS_CLOSING ||
4635 		    dst->state >= TCPS_CLOSING)
4636 			(*state)->timeout = PFTM_TCP_CLOSING;
4637 		else if ((th->th_flags & TH_SYN) &&
4638 			 ((*state)->state_flags & PFSTATE_SLOPPY))
4639 			(*state)->timeout = PFTM_TCP_FIRST_PACKET;
4640 		else
4641 			(*state)->timeout = PFTM_TCP_ESTABLISHED;
4642 
4643 		/* Fall through to PASS packet */
4644 
4645 	} else if ((dst->state < TCPS_SYN_SENT ||
4646 		dst->state >= TCPS_FIN_WAIT_2 ||
4647 		src->state >= TCPS_FIN_WAIT_2) &&
4648 	    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4649 	    /* Within a window forward of the originating packet */
4650 	    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4651 	    /* Within a window backward of the originating packet */
4652 
4653 		/*
4654 		 * This currently handles three situations:
4655 		 *  1) Stupid stacks will shotgun SYNs before their peer
4656 		 *     replies.
4657 		 *  2) When PF catches an already established stream (the
4658 		 *     firewall rebooted, the state table was flushed, routes
4659 		 *     changed...)
4660 		 *  3) Packets get funky immediately after the connection
4661 		 *     closes (this should catch Solaris spurious ACK|FINs
4662 		 *     that web servers like to spew after a close)
4663 		 *
4664 		 * This must be a little more careful than the above code
4665 		 * since packet floods will also be caught here. We don't
4666 		 * update the TTL here to mitigate the damage of a packet
4667 		 * flood and so the same code can handle awkward establishment
4668 		 * and a loosened connection close.
4669 		 * In the establishment case, a correct peer response will
4670 		 * validate the connection, go through the normal state code
4671 		 * and keep updating the state TTL.
4672 		 */
4673 
4674 		if (pf_status.debug >= PF_DEBUG_MISC) {
4675 			kprintf("pf: loose state match: ");
4676 			pf_print_state(*state);
4677 			pf_print_flags(th->th_flags);
4678 			kprintf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4679 			    "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, pd->p_len,
4680 			    ackskew, (unsigned long long)(*state)->packets[0],
4681 			    (unsigned long long)(*state)->packets[1],
4682 			    pd->dir == PF_IN ? "in" : "out",
4683 			    pd->dir == (*state)->direction ? "fwd" : "rev");
4684 		}
4685 
4686 		if (dst->scrub || src->scrub) {
4687 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4688 			    *state, src, dst, copyback))
4689 				return (PF_DROP);
4690 		}
4691 
4692 		/* update max window */
4693 		if (src->max_win < win)
4694 			src->max_win = win;
4695 		/* synchronize sequencing */
4696 		if (SEQ_GT(end, src->seqlo))
4697 			src->seqlo = end;
4698 		/* slide the window of what the other end can send */
4699 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4700 			dst->seqhi = ack + MAX((win << sws), 1);
4701 
4702 		/*
4703 		 * Cannot set dst->seqhi here since this could be a shotgunned
4704 		 * SYN and not an already established connection.
4705 		 */
4706 
4707 		if (th->th_flags & TH_FIN)
4708 			if (src->state < TCPS_CLOSING)
4709 				src->state = TCPS_CLOSING;
4710 		if (th->th_flags & TH_RST)
4711 			src->state = dst->state = TCPS_TIME_WAIT;
4712 
4713 		/* Fall through to PASS packet */
4714 
4715 	} else if ((*state)->pickup_mode == PF_PICKUPS_HASHONLY ||
4716 		    ((*state)->pickup_mode == PF_PICKUPS_ENABLED &&
4717 		     ((*state)->sync_flags & PFSTATE_GOT_SYN_MASK) !=
4718 		      PFSTATE_GOT_SYN_MASK)) {
4719 		/*
4720 		 * If pickup mode is hash only, do not fail on sequence checks.
4721 		 *
4722 		 * If pickup mode is enabled and we did not see the SYN in
4723 		 * both direction, do not fail on sequence checks because
4724 		 * we do not have complete information on window scale.
4725 		 *
4726 		 * Adjust expiration and fall through to PASS packet.
4727 		 * XXX Add a FIN check to reduce timeout?
4728 		 */
4729 		(*state)->expire = time_second;
4730 	} else  {
4731 		/*
4732 		 * Failure processing
4733 		 */
4734 		if ((*state)->dst.state == TCPS_SYN_SENT &&
4735 		    (*state)->src.state == TCPS_SYN_SENT) {
4736 			/* Send RST for state mismatches during handshake */
4737 			if (!(th->th_flags & TH_RST))
4738 				pf_send_tcp((*state)->rule.ptr, pd->af,
4739 				    pd->dst, pd->src, th->th_dport,
4740 				    th->th_sport, ntohl(th->th_ack), 0,
4741 				    TH_RST, 0, 0,
4742 				    (*state)->rule.ptr->return_ttl, 1, 0,
4743 				    pd->eh, kif->pfik_ifp);
4744 			src->seqlo = 0;
4745 			src->seqhi = 1;
4746 			src->max_win = 1;
4747 		} else if (pf_status.debug >= PF_DEBUG_MISC) {
4748 			kprintf("pf: BAD state: ");
4749 			pf_print_state(*state);
4750 			pf_print_flags(th->th_flags);
4751 			kprintf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4752 			    "pkts=%llu:%llu dir=%s,%s\n",
4753 			    seq, orig_seq, ack, pd->p_len, ackskew,
4754 			    (unsigned long long)(*state)->packets[0],
4755 				(unsigned long long)(*state)->packets[1],
4756 			    pd->dir == PF_IN ? "in" : "out",
4757 			    pd->dir == (*state)->direction ? "fwd" : "rev");
4758 			kprintf("pf: State failure on: %c %c %c %c | %c %c\n",
4759 			    SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4760 			    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4761 			    ' ': '2',
4762 			    (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4763 			    (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4764 			    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4765 			    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4766 		}
4767 		REASON_SET(reason, PFRES_BADSTATE);
4768 		return (PF_DROP);
4769 	}
4770 
4771 	return (PF_PASS);
4772 }
4773 
4774 /*
4775  * Called with state locked
4776  */
4777 int
pf_tcp_track_sloppy(struct pf_state_peer * src,struct pf_state_peer * dst,struct pf_state ** state,struct pf_pdesc * pd,u_short * reason)4778 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4779 	struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4780 {
4781 	struct tcphdr		*th = pd->hdr.tcp;
4782 
4783 	if (th->th_flags & TH_SYN)
4784 		if (src->state < TCPS_SYN_SENT)
4785 			src->state = TCPS_SYN_SENT;
4786 	if (th->th_flags & TH_FIN)
4787 		if (src->state < TCPS_CLOSING)
4788 			src->state = TCPS_CLOSING;
4789 	if (th->th_flags & TH_ACK) {
4790 		if (dst->state == TCPS_SYN_SENT) {
4791 			dst->state = TCPS_ESTABLISHED;
4792 			if (src->state == TCPS_ESTABLISHED &&
4793 			    (*state)->src_node != NULL &&
4794 			    pf_src_connlimit(*state)) {
4795 				REASON_SET(reason, PFRES_SRCLIMIT);
4796 				return (PF_DROP);
4797 			}
4798 		} else if (dst->state == TCPS_CLOSING) {
4799 			dst->state = TCPS_FIN_WAIT_2;
4800 		} else if (src->state == TCPS_SYN_SENT &&
4801 		    dst->state < TCPS_SYN_SENT) {
4802 			/*
4803 			 * Handle a special sloppy case where we only see one
4804 			 * half of the connection. If there is a ACK after
4805 			 * the initial SYN without ever seeing a packet from
4806 			 * the destination, set the connection to established.
4807 			 */
4808 			dst->state = src->state = TCPS_ESTABLISHED;
4809 			if ((*state)->src_node != NULL &&
4810 			    pf_src_connlimit(*state)) {
4811 				REASON_SET(reason, PFRES_SRCLIMIT);
4812 				return (PF_DROP);
4813 			}
4814 		} else if (src->state == TCPS_CLOSING &&
4815 		    dst->state == TCPS_ESTABLISHED &&
4816 		    dst->seqlo == 0) {
4817 			/*
4818 			 * Handle the closing of half connections where we
4819 			 * don't see the full bidirectional FIN/ACK+ACK
4820 			 * handshake.
4821 			 */
4822 			dst->state = TCPS_CLOSING;
4823 		}
4824 	}
4825 	if (th->th_flags & TH_RST)
4826 		src->state = dst->state = TCPS_TIME_WAIT;
4827 
4828 	/* update expire time */
4829 	(*state)->expire = time_second;
4830 	if (src->state >= TCPS_FIN_WAIT_2 &&
4831 	    dst->state >= TCPS_FIN_WAIT_2)
4832 		(*state)->timeout = PFTM_TCP_CLOSED;
4833 	else if (src->state >= TCPS_CLOSING &&
4834 	    dst->state >= TCPS_CLOSING)
4835 		(*state)->timeout = PFTM_TCP_FIN_WAIT;
4836 	else if (src->state < TCPS_ESTABLISHED ||
4837 	    dst->state < TCPS_ESTABLISHED)
4838 		(*state)->timeout = PFTM_TCP_OPENING;
4839 	else if (src->state >= TCPS_CLOSING ||
4840 	    dst->state >= TCPS_CLOSING)
4841 		(*state)->timeout = PFTM_TCP_CLOSING;
4842 	else if ((th->th_flags & TH_SYN) &&
4843 		 ((*state)->state_flags & PFSTATE_SLOPPY))
4844 		(*state)->timeout = PFTM_TCP_FIRST_PACKET;
4845 	else
4846 		(*state)->timeout = PFTM_TCP_ESTABLISHED;
4847 
4848 	return (PF_PASS);
4849 }
4850 
4851 /*
4852  * Test TCP connection state.  Caller must hold the state locked.
4853  */
4854 int
pf_test_state_tcp(struct pf_state ** state,int direction,struct pfi_kif * kif,struct mbuf * m,int off,void * h,struct pf_pdesc * pd,u_short * reason)4855 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4856 		  struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4857 		  u_short *reason)
4858 {
4859 	struct pf_state_key_cmp	 key;
4860 	struct tcphdr		*th = pd->hdr.tcp;
4861 	int			 copyback = 0;
4862 	int			 error;
4863 	struct pf_state_peer	*src, *dst;
4864 	struct pf_state_key	*sk;
4865 
4866 	bzero(&key, sizeof(key));
4867 	key.af = pd->af;
4868 	key.proto = IPPROTO_TCP;
4869 	if (direction == PF_IN)	{	/* wire side, straight */
4870 		PF_ACPY(&key.addr[0], pd->src, key.af);
4871 		PF_ACPY(&key.addr[1], pd->dst, key.af);
4872 		key.port[0] = th->th_sport;
4873 		key.port[1] = th->th_dport;
4874 		if (pf_status.debug >= PF_DEBUG_MISC) {
4875 			kprintf("test-tcp IN (%08x:%d) -> (%08x:%d)\n",
4876 				ntohl(key.addr[0].addr32[0]),
4877 				ntohs(key.port[0]),
4878 				ntohl(key.addr[1].addr32[0]),
4879 				ntohs(key.port[1]));
4880 		}
4881 	} else {			/* stack side, reverse */
4882 		PF_ACPY(&key.addr[1], pd->src, key.af);
4883 		PF_ACPY(&key.addr[0], pd->dst, key.af);
4884 		key.port[1] = th->th_sport;
4885 		key.port[0] = th->th_dport;
4886 		if (pf_status.debug >= PF_DEBUG_MISC) {
4887 			kprintf("test-tcp OUT (%08x:%d) <- (%08x:%d)\n",
4888 				ntohl(key.addr[0].addr32[0]),
4889 				ntohs(key.port[0]),
4890 				ntohl(key.addr[1].addr32[0]),
4891 				ntohs(key.port[1]));
4892 		}
4893 	}
4894 
4895 	STATE_LOOKUP(kif, &key, direction, *state, m);
4896 	lockmgr(&(*state)->lk, LK_EXCLUSIVE);
4897 
4898 	if (direction == (*state)->direction) {
4899 		src = &(*state)->src;
4900 		dst = &(*state)->dst;
4901 	} else {
4902 		src = &(*state)->dst;
4903 		dst = &(*state)->src;
4904 	}
4905 
4906 	sk = (*state)->key[pd->didx];
4907 
4908 	if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4909 		if (direction != (*state)->direction) {
4910 			REASON_SET(reason, PFRES_SYNPROXY);
4911 			FAIL (PF_SYNPROXY_DROP);
4912 		}
4913 		if (th->th_flags & TH_SYN) {
4914 			if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4915 				REASON_SET(reason, PFRES_SYNPROXY);
4916 				FAIL (PF_DROP);
4917 			}
4918 			pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
4919 			    pd->src, th->th_dport, th->th_sport,
4920 			    (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4921 			    TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1,
4922 			    0, NULL, NULL);
4923 			REASON_SET(reason, PFRES_SYNPROXY);
4924 			FAIL (PF_SYNPROXY_DROP);
4925 		} else if (!(th->th_flags & TH_ACK) ||
4926 		    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4927 		    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4928 			REASON_SET(reason, PFRES_SYNPROXY);
4929 			FAIL (PF_DROP);
4930 		} else if ((*state)->src_node != NULL &&
4931 		    pf_src_connlimit(*state)) {
4932 			REASON_SET(reason, PFRES_SRCLIMIT);
4933 			FAIL (PF_DROP);
4934 		} else
4935 			(*state)->src.state = PF_TCPS_PROXY_DST;
4936 	}
4937 	if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4938 		if (direction == (*state)->direction) {
4939 			if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4940 			    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4941 			    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4942 				REASON_SET(reason, PFRES_SYNPROXY);
4943 				FAIL (PF_DROP);
4944 			}
4945 			(*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4946 			if ((*state)->dst.seqhi == 1)
4947 				(*state)->dst.seqhi = htonl(karc4random());
4948 			pf_send_tcp((*state)->rule.ptr, pd->af,
4949 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
4950 			    sk->port[pd->sidx], sk->port[pd->didx],
4951 			    (*state)->dst.seqhi, 0, TH_SYN, 0,
4952 			    (*state)->src.mss, 0, 0, (*state)->tag, NULL, NULL);
4953 			REASON_SET(reason, PFRES_SYNPROXY);
4954 			FAIL (PF_SYNPROXY_DROP);
4955 		} else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4956 		    (TH_SYN|TH_ACK)) ||
4957 		    (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4958 			REASON_SET(reason, PFRES_SYNPROXY);
4959 			FAIL (PF_DROP);
4960 		} else {
4961 			(*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4962 			(*state)->dst.seqlo = ntohl(th->th_seq);
4963 			pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
4964 			    pd->src, th->th_dport, th->th_sport,
4965 			    ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4966 			    TH_ACK, (*state)->src.max_win, 0, 0, 0,
4967 			    (*state)->tag, NULL, NULL);
4968 			pf_send_tcp((*state)->rule.ptr, pd->af,
4969 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
4970 			    sk->port[pd->sidx], sk->port[pd->didx],
4971 			    (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4972 			    TH_ACK, (*state)->dst.max_win, 0, 0, 1,
4973 			    0, NULL, NULL);
4974 			(*state)->src.seqdiff = (*state)->dst.seqhi -
4975 			    (*state)->src.seqlo;
4976 			(*state)->dst.seqdiff = (*state)->src.seqhi -
4977 			    (*state)->dst.seqlo;
4978 			(*state)->src.seqhi = (*state)->src.seqlo +
4979 			    (*state)->dst.max_win;
4980 			(*state)->dst.seqhi = (*state)->dst.seqlo +
4981 			    (*state)->src.max_win;
4982 			(*state)->src.wscale = (*state)->dst.wscale = 0;
4983 			(*state)->src.state = (*state)->dst.state =
4984 			    TCPS_ESTABLISHED;
4985 			REASON_SET(reason, PFRES_SYNPROXY);
4986 			FAIL (PF_SYNPROXY_DROP);
4987 		}
4988 	}
4989 
4990 	/*
4991 	 * Check for connection (addr+port pair) reuse.  We can't actually
4992 	 * unlink the state if we don't own it.
4993 	 */
4994 	if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4995 	    dst->state >= TCPS_FIN_WAIT_2 &&
4996 	    src->state >= TCPS_FIN_WAIT_2) {
4997 		if (pf_status.debug >= PF_DEBUG_MISC) {
4998 			kprintf("pf: state reuse ");
4999 			pf_print_state(*state);
5000 			pf_print_flags(th->th_flags);
5001 			kprintf("\n");
5002 		}
5003 		/* XXX make sure it's the same direction ?? */
5004 		(*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
5005 		if ((*state)->cpuid == mycpu->gd_cpuid) {
5006 			pf_unlink_state(*state);
5007 			*state = NULL;
5008 		} else {
5009 			(*state)->timeout = PFTM_PURGE;
5010 		}
5011 		FAIL (PF_DROP);
5012 	}
5013 
5014 	if ((*state)->state_flags & PFSTATE_SLOPPY) {
5015 		if (pf_tcp_track_sloppy(src, dst, state, pd,
5016 					reason) == PF_DROP) {
5017 			FAIL (PF_DROP);
5018 		}
5019 	} else {
5020 		if (pf_tcp_track_full(src, dst, state, kif, m, off, pd,
5021 				      reason, &copyback) == PF_DROP) {
5022 			FAIL (PF_DROP);
5023 		}
5024 	}
5025 
5026 	/* translate source/destination address, if necessary */
5027 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5028 		struct pf_state_key *nk = (*state)->key[pd->didx];
5029 
5030 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
5031 		    nk->port[pd->sidx] != th->th_sport)  {
5032 			/*
5033 			 * The translated source address may be completely
5034 			 * unrelated to the saved link header, make sure
5035 			 * a bridge doesn't try to use it.
5036 			 */
5037 			m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
5038 			pf_change_ap(pd->src, &th->th_sport, pd->ip_sum,
5039 			    &th->th_sum, &nk->addr[pd->sidx],
5040 			    nk->port[pd->sidx], 0, pd->af);
5041 		}
5042 
5043 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
5044 		    nk->port[pd->didx] != th->th_dport) {
5045 			/*
5046 			 * If we don't redispatch the packet will go into
5047 			 * the protocol stack on the wrong cpu for the
5048 			 * post-translated address.
5049 			 */
5050 			pf_change_ap(pd->dst, &th->th_dport, pd->ip_sum,
5051 			    &th->th_sum, &nk->addr[pd->didx],
5052 			    nk->port[pd->didx], 0, pd->af);
5053 		}
5054 		copyback = 1;
5055 	}
5056 
5057 	/* Copyback sequence modulation or stateful scrub changes if needed */
5058 	if (copyback) {
5059 		m->m_flags &= ~M_HASH;
5060 		m_copyback(m, off, sizeof(*th), th);
5061 	}
5062 
5063 	pfsync_update_state(*state);
5064 	error = PF_PASS;
5065 done:
5066 	if (*state)
5067 		lockmgr(&(*state)->lk, LK_RELEASE);
5068 	return (error);
5069 }
5070 
5071 /*
5072  * Test UDP connection state.  Caller must hold the state locked.
5073  */
5074 int
pf_test_state_udp(struct pf_state ** state,int direction,struct pfi_kif * kif,struct mbuf * m,int off,void * h,struct pf_pdesc * pd)5075 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
5076 		  struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
5077 {
5078 	struct pf_state_peer	*src, *dst;
5079 	struct pf_state_key_cmp	 key;
5080 	struct udphdr		*uh = pd->hdr.udp;
5081 
5082 	bzero(&key, sizeof(key));
5083 	key.af = pd->af;
5084 	key.proto = IPPROTO_UDP;
5085 	if (direction == PF_IN)	{	/* wire side, straight */
5086 		PF_ACPY(&key.addr[0], pd->src, key.af);
5087 		PF_ACPY(&key.addr[1], pd->dst, key.af);
5088 		key.port[0] = uh->uh_sport;
5089 		key.port[1] = uh->uh_dport;
5090 	} else {			/* stack side, reverse */
5091 		PF_ACPY(&key.addr[1], pd->src, key.af);
5092 		PF_ACPY(&key.addr[0], pd->dst, key.af);
5093 		key.port[1] = uh->uh_sport;
5094 		key.port[0] = uh->uh_dport;
5095 	}
5096 
5097 	STATE_LOOKUP(kif, &key, direction, *state, m);
5098 	lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5099 
5100 	if (direction == (*state)->direction) {
5101 		src = &(*state)->src;
5102 		dst = &(*state)->dst;
5103 	} else {
5104 		src = &(*state)->dst;
5105 		dst = &(*state)->src;
5106 	}
5107 
5108 	/* update states */
5109 	if (src->state < PFUDPS_SINGLE)
5110 		src->state = PFUDPS_SINGLE;
5111 	if (dst->state == PFUDPS_SINGLE)
5112 		dst->state = PFUDPS_MULTIPLE;
5113 
5114 	/* update expire time */
5115 	(*state)->expire = time_second;
5116 	if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
5117 		(*state)->timeout = PFTM_UDP_MULTIPLE;
5118 	else
5119 		(*state)->timeout = PFTM_UDP_SINGLE;
5120 
5121 	/* translate source/destination address, if necessary */
5122 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5123 		struct pf_state_key *nk = (*state)->key[pd->didx];
5124 
5125 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
5126 		    nk->port[pd->sidx] != uh->uh_sport) {
5127 			/*
5128 			 * The translated source address may be completely
5129 			 * unrelated to the saved link header, make sure
5130 			 * a bridge doesn't try to use it.
5131 			 */
5132 			m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
5133 			m->m_flags &= ~M_HASH;
5134 			pf_change_ap(pd->src, &uh->uh_sport, pd->ip_sum,
5135 			    &uh->uh_sum, &nk->addr[pd->sidx],
5136 			    nk->port[pd->sidx], 1, pd->af);
5137 		}
5138 
5139 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
5140 		    nk->port[pd->didx] != uh->uh_dport) {
5141 			/*
5142 			 * If we don't redispatch the packet will go into
5143 			 * the protocol stack on the wrong cpu for the
5144 			 * post-translated address.
5145 			 */
5146 			m->m_flags &= ~M_HASH;
5147 			pf_change_ap(pd->dst, &uh->uh_dport, pd->ip_sum,
5148 			    &uh->uh_sum, &nk->addr[pd->didx],
5149 			    nk->port[pd->didx], 1, pd->af);
5150 		}
5151 		m_copyback(m, off, sizeof(*uh), uh);
5152 	}
5153 
5154 	pfsync_update_state(*state);
5155 	lockmgr(&(*state)->lk, LK_RELEASE);
5156 	return (PF_PASS);
5157 }
5158 
5159 /*
5160  * Test ICMP connection state.  Caller must hold the state locked.
5161  */
5162 int
pf_test_state_icmp(struct pf_state ** state,int direction,struct pfi_kif * kif,struct mbuf * m,int off,void * h,struct pf_pdesc * pd,u_short * reason)5163 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
5164 		   struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
5165 		   u_short *reason)
5166 {
5167 	struct pf_addr	*saddr = pd->src, *daddr = pd->dst;
5168 	u_int16_t	 icmpid = 0, *icmpsum = NULL;
5169 	u_int8_t	 icmptype = 0;
5170 	int		 state_icmp = 0;
5171 	int		 error;
5172 	struct pf_state_key_cmp key;
5173 
5174 	bzero(&key, sizeof(key));
5175 
5176 	switch (pd->proto) {
5177 #ifdef INET
5178 	case IPPROTO_ICMP:
5179 		icmptype = pd->hdr.icmp->icmp_type;
5180 		icmpid = pd->hdr.icmp->icmp_id;
5181 		icmpsum = &pd->hdr.icmp->icmp_cksum;
5182 
5183 		if (icmptype == ICMP_UNREACH ||
5184 		    icmptype == ICMP_SOURCEQUENCH ||
5185 		    icmptype == ICMP_REDIRECT ||
5186 		    icmptype == ICMP_TIMXCEED ||
5187 		    icmptype == ICMP_PARAMPROB)
5188 			state_icmp++;
5189 		break;
5190 #endif /* INET */
5191 #ifdef INET6
5192 	case IPPROTO_ICMPV6:
5193 		icmptype = pd->hdr.icmp6->icmp6_type;
5194 		icmpid = pd->hdr.icmp6->icmp6_id;
5195 		icmpsum = &pd->hdr.icmp6->icmp6_cksum;
5196 
5197 		if (icmptype == ICMP6_DST_UNREACH ||
5198 		    icmptype == ICMP6_PACKET_TOO_BIG ||
5199 		    icmptype == ICMP6_TIME_EXCEEDED ||
5200 		    icmptype == ICMP6_PARAM_PROB)
5201 			state_icmp++;
5202 		break;
5203 #endif /* INET6 */
5204 	}
5205 
5206 	if (!state_icmp) {
5207 
5208 		/*
5209 		 * ICMP query/reply message not related to a TCP/UDP packet.
5210 		 * Search for an ICMP state.
5211 		 */
5212 		key.af = pd->af;
5213 		key.proto = pd->proto;
5214 		key.port[0] = key.port[1] = icmpid;
5215 		if (direction == PF_IN)	{	/* wire side, straight */
5216 			PF_ACPY(&key.addr[0], pd->src, key.af);
5217 			PF_ACPY(&key.addr[1], pd->dst, key.af);
5218 		} else {			/* stack side, reverse */
5219 			PF_ACPY(&key.addr[1], pd->src, key.af);
5220 			PF_ACPY(&key.addr[0], pd->dst, key.af);
5221 		}
5222 
5223 		STATE_LOOKUP(kif, &key, direction, *state, m);
5224 		lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5225 
5226 		(*state)->expire = time_second;
5227 		(*state)->timeout = PFTM_ICMP_ERROR_REPLY;
5228 
5229 		/* translate source/destination address, if necessary */
5230 		if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5231 			struct pf_state_key *nk = (*state)->key[pd->didx];
5232 
5233 			switch (pd->af) {
5234 #ifdef INET
5235 			case AF_INET:
5236 				if (PF_ANEQ(pd->src,
5237 				    &nk->addr[pd->sidx], AF_INET))
5238 					pf_change_a(&saddr->v4.s_addr,
5239 					    pd->ip_sum,
5240 					    nk->addr[pd->sidx].v4.s_addr, 0);
5241 
5242 				if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
5243 				    AF_INET))
5244 					pf_change_a(&daddr->v4.s_addr,
5245 					    pd->ip_sum,
5246 					    nk->addr[pd->didx].v4.s_addr, 0);
5247 
5248 				if (nk->port[0] !=
5249 				    pd->hdr.icmp->icmp_id) {
5250 					pd->hdr.icmp->icmp_cksum =
5251 					    pf_cksum_fixup(
5252 					    pd->hdr.icmp->icmp_cksum, icmpid,
5253 					    nk->port[pd->sidx], 0);
5254 					pd->hdr.icmp->icmp_id =
5255 					    nk->port[pd->sidx];
5256 				}
5257 
5258 				m->m_flags &= ~M_HASH;
5259 				m_copyback(m, off, ICMP_MINLEN, pd->hdr.icmp);
5260 				break;
5261 #endif /* INET */
5262 #ifdef INET6
5263 			case AF_INET6:
5264 				if (PF_ANEQ(pd->src,
5265 				    &nk->addr[pd->sidx], AF_INET6))
5266 					pf_change_a6(saddr,
5267 					    &pd->hdr.icmp6->icmp6_cksum,
5268 					    &nk->addr[pd->sidx], 0);
5269 
5270 				if (PF_ANEQ(pd->dst,
5271 				    &nk->addr[pd->didx], AF_INET6))
5272 					pf_change_a6(daddr,
5273 					    &pd->hdr.icmp6->icmp6_cksum,
5274 					    &nk->addr[pd->didx], 0);
5275 
5276 				m->m_flags &= ~M_HASH;
5277 				m_copyback(m, off, sizeof(struct icmp6_hdr),
5278 				    pd->hdr.icmp6);
5279 				break;
5280 #endif /* INET6 */
5281 			}
5282 		}
5283 	} else {
5284 		/*
5285 		 * ICMP error message in response to a TCP/UDP packet.
5286 		 * Extract the inner TCP/UDP header and search for that state.
5287 		 */
5288 
5289 		struct pf_pdesc	pd2;
5290 #ifdef INET
5291 		struct ip	h2;
5292 #endif /* INET */
5293 #ifdef INET6
5294 		struct ip6_hdr	h2_6;
5295 		int		terminal = 0;
5296 #endif /* INET6 */
5297 		int		ipoff2;
5298 		int		off2;
5299 
5300 		pd2.not_cpu_localized = 1;
5301 		pd2.af = pd->af;
5302 		/* Payload packet is from the opposite direction. */
5303 		pd2.sidx = (direction == PF_IN) ? 1 : 0;
5304 		pd2.didx = (direction == PF_IN) ? 0 : 1;
5305 		switch (pd->af) {
5306 #ifdef INET
5307 		case AF_INET:
5308 			/* offset of h2 in mbuf chain */
5309 			ipoff2 = off + ICMP_MINLEN;
5310 
5311 			if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
5312 			    NULL, reason, pd2.af)) {
5313 				DPFPRINTF(PF_DEBUG_MISC,
5314 				    ("pf: ICMP error message too short "
5315 				    "(ip)\n"));
5316 				FAIL (PF_DROP);
5317 			}
5318 			/*
5319 			 * ICMP error messages don't refer to non-first
5320 			 * fragments
5321 			 */
5322 			if (h2.ip_off & htons(IP_OFFMASK)) {
5323 				REASON_SET(reason, PFRES_FRAG);
5324 				FAIL (PF_DROP);
5325 			}
5326 
5327 			/* offset of protocol header that follows h2 */
5328 			off2 = ipoff2 + (h2.ip_hl << 2);
5329 
5330 			pd2.proto = h2.ip_p;
5331 			pd2.src = (struct pf_addr *)&h2.ip_src;
5332 			pd2.dst = (struct pf_addr *)&h2.ip_dst;
5333 			pd2.ip_sum = &h2.ip_sum;
5334 			break;
5335 #endif /* INET */
5336 #ifdef INET6
5337 		case AF_INET6:
5338 			ipoff2 = off + sizeof(struct icmp6_hdr);
5339 
5340 			if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
5341 			    NULL, reason, pd2.af)) {
5342 				DPFPRINTF(PF_DEBUG_MISC,
5343 				    ("pf: ICMP error message too short "
5344 				    "(ip6)\n"));
5345 				FAIL (PF_DROP);
5346 			}
5347 			pd2.proto = h2_6.ip6_nxt;
5348 			pd2.src = (struct pf_addr *)&h2_6.ip6_src;
5349 			pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
5350 			pd2.ip_sum = NULL;
5351 			off2 = ipoff2 + sizeof(h2_6);
5352 			do {
5353 				switch (pd2.proto) {
5354 				case IPPROTO_FRAGMENT:
5355 					/*
5356 					 * ICMPv6 error messages for
5357 					 * non-first fragments
5358 					 */
5359 					REASON_SET(reason, PFRES_FRAG);
5360 					FAIL (PF_DROP);
5361 				case IPPROTO_AH:
5362 				case IPPROTO_HOPOPTS:
5363 				case IPPROTO_ROUTING:
5364 				case IPPROTO_DSTOPTS: {
5365 					/* get next header and header length */
5366 					struct ip6_ext opt6;
5367 
5368 					if (!pf_pull_hdr(m, off2, &opt6,
5369 					    sizeof(opt6), NULL, reason,
5370 					    pd2.af)) {
5371 						DPFPRINTF(PF_DEBUG_MISC,
5372 						    ("pf: ICMPv6 short opt\n"));
5373 						FAIL (PF_DROP);
5374 					}
5375 					if (pd2.proto == IPPROTO_AH)
5376 						off2 += (opt6.ip6e_len + 2) * 4;
5377 					else
5378 						off2 += (opt6.ip6e_len + 1) * 8;
5379 					pd2.proto = opt6.ip6e_nxt;
5380 					/* goto the next header */
5381 					break;
5382 				}
5383 				default:
5384 					terminal++;
5385 					break;
5386 				}
5387 			} while (!terminal);
5388 			break;
5389 #endif /* INET6 */
5390 		default:
5391 			DPFPRINTF(PF_DEBUG_MISC,
5392 			    ("pf: ICMP AF %d unknown (ip6)\n", pd->af));
5393 			FAIL (PF_DROP);
5394 			break;
5395 		}
5396 
5397 		switch (pd2.proto) {
5398 		case IPPROTO_TCP: {
5399 			struct tcphdr		 th;
5400 			u_int32_t		 seq;
5401 			struct pf_state_peer	*src, *dst;
5402 			u_int8_t		 dws;
5403 			int			 copyback = 0;
5404 
5405 			/*
5406 			 * Only the first 8 bytes of the TCP header can be
5407 			 * expected. Don't access any TCP header fields after
5408 			 * th_seq, an ackskew test is not possible.
5409 			 */
5410 			if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
5411 			    pd2.af)) {
5412 				DPFPRINTF(PF_DEBUG_MISC,
5413 				    ("pf: ICMP error message too short "
5414 				    "(tcp)\n"));
5415 				FAIL (PF_DROP);
5416 			}
5417 
5418 			key.af = pd2.af;
5419 			key.proto = IPPROTO_TCP;
5420 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5421 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5422 			key.port[pd2.sidx] = th.th_sport;
5423 			key.port[pd2.didx] = th.th_dport;
5424 
5425 			STATE_LOOKUP(kif, &key, direction, *state, m);
5426 			lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5427 
5428 			if (direction == (*state)->direction) {
5429 				src = &(*state)->dst;
5430 				dst = &(*state)->src;
5431 			} else {
5432 				src = &(*state)->src;
5433 				dst = &(*state)->dst;
5434 			}
5435 
5436 			if (src->wscale && dst->wscale)
5437 				dws = dst->wscale & PF_WSCALE_MASK;
5438 			else
5439 				dws = 0;
5440 
5441 			/* Demodulate sequence number */
5442 			seq = ntohl(th.th_seq) - src->seqdiff;
5443 			if (src->seqdiff) {
5444 				pf_change_a(&th.th_seq, icmpsum,
5445 				    htonl(seq), 0);
5446 				copyback = 1;
5447 			}
5448 
5449 			if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
5450 			    (!SEQ_GEQ(src->seqhi, seq) ||
5451 			    !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
5452 				if (pf_status.debug >= PF_DEBUG_MISC) {
5453 					kprintf("pf: BAD ICMP %d:%d ",
5454 					    icmptype, pd->hdr.icmp->icmp_code);
5455 					pf_print_host(pd->src, 0, pd->af);
5456 					kprintf(" -> ");
5457 					pf_print_host(pd->dst, 0, pd->af);
5458 					kprintf(" state: ");
5459 					pf_print_state(*state);
5460 					kprintf(" seq=%u\n", seq);
5461 				}
5462 				REASON_SET(reason, PFRES_BADSTATE);
5463 				FAIL (PF_DROP);
5464 			} else {
5465 				if (pf_status.debug >= PF_DEBUG_MISC) {
5466 					kprintf("pf: OK ICMP %d:%d ",
5467 					    icmptype, pd->hdr.icmp->icmp_code);
5468 					pf_print_host(pd->src, 0, pd->af);
5469 					kprintf(" -> ");
5470 					pf_print_host(pd->dst, 0, pd->af);
5471 					kprintf(" state: ");
5472 					pf_print_state(*state);
5473 					kprintf(" seq=%u\n", seq);
5474 				}
5475 			}
5476 
5477 			/* translate source/destination address, if necessary */
5478 			if ((*state)->key[PF_SK_WIRE] !=
5479 			    (*state)->key[PF_SK_STACK]) {
5480 				struct pf_state_key *nk =
5481 				    (*state)->key[pd->didx];
5482 
5483 				if (PF_ANEQ(pd2.src,
5484 				    &nk->addr[pd2.sidx], pd2.af) ||
5485 				    nk->port[pd2.sidx] != th.th_sport)
5486 					pf_change_icmp(pd2.src, &th.th_sport,
5487 					    daddr, &nk->addr[pd2.sidx],
5488 					    nk->port[pd2.sidx], NULL,
5489 					    pd2.ip_sum, icmpsum,
5490 					    pd->ip_sum, 0, pd2.af);
5491 
5492 				if (PF_ANEQ(pd2.dst,
5493 				    &nk->addr[pd2.didx], pd2.af) ||
5494 				    nk->port[pd2.didx] != th.th_dport)
5495 					pf_change_icmp(pd2.dst, &th.th_dport,
5496 					    NULL, /* XXX Inbound NAT? */
5497 					    &nk->addr[pd2.didx],
5498 					    nk->port[pd2.didx], NULL,
5499 					    pd2.ip_sum, icmpsum,
5500 					    pd->ip_sum, 0, pd2.af);
5501 				copyback = 1;
5502 			}
5503 
5504 			if (copyback) {
5505 				switch (pd2.af) {
5506 #ifdef INET
5507 				case AF_INET:
5508 					m_copyback(m, off, ICMP_MINLEN,
5509 					    pd->hdr.icmp);
5510 					m_copyback(m, ipoff2, sizeof(h2),
5511 					    &h2);
5512 					break;
5513 #endif /* INET */
5514 #ifdef INET6
5515 				case AF_INET6:
5516 					m_copyback(m, off,
5517 					    sizeof(struct icmp6_hdr),
5518 					    pd->hdr.icmp6);
5519 					m_copyback(m, ipoff2, sizeof(h2_6),
5520 					    &h2_6);
5521 					break;
5522 #endif /* INET6 */
5523 				}
5524 				m->m_flags &= ~M_HASH;
5525 				m_copyback(m, off2, 8, &th);
5526 			}
5527 			break;
5528 		}
5529 		case IPPROTO_UDP: {
5530 			struct udphdr		uh;
5531 
5532 			if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
5533 			    NULL, reason, pd2.af)) {
5534 				DPFPRINTF(PF_DEBUG_MISC,
5535 				    ("pf: ICMP error message too short "
5536 				    "(udp)\n"));
5537 				return (PF_DROP);
5538 			}
5539 
5540 			key.af = pd2.af;
5541 			key.proto = IPPROTO_UDP;
5542 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5543 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5544 			key.port[pd2.sidx] = uh.uh_sport;
5545 			key.port[pd2.didx] = uh.uh_dport;
5546 
5547 			STATE_LOOKUP(kif, &key, direction, *state, m);
5548 			lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5549 
5550 			/* translate source/destination address, if necessary */
5551 			if ((*state)->key[PF_SK_WIRE] !=
5552 			    (*state)->key[PF_SK_STACK]) {
5553 				struct pf_state_key *nk =
5554 				    (*state)->key[pd->didx];
5555 
5556 				if (PF_ANEQ(pd2.src,
5557 				    &nk->addr[pd2.sidx], pd2.af) ||
5558 				    nk->port[pd2.sidx] != uh.uh_sport)
5559 					pf_change_icmp(pd2.src, &uh.uh_sport,
5560 					    daddr, &nk->addr[pd2.sidx],
5561 					    nk->port[pd2.sidx], &uh.uh_sum,
5562 					    pd2.ip_sum, icmpsum,
5563 					    pd->ip_sum, 1, pd2.af);
5564 
5565 				if (PF_ANEQ(pd2.dst,
5566 				    &nk->addr[pd2.didx], pd2.af) ||
5567 				    nk->port[pd2.didx] != uh.uh_dport)
5568 					pf_change_icmp(pd2.dst, &uh.uh_dport,
5569 					    NULL, /* XXX Inbound NAT? */
5570 					    &nk->addr[pd2.didx],
5571 					    nk->port[pd2.didx], &uh.uh_sum,
5572 					    pd2.ip_sum, icmpsum,
5573 					    pd->ip_sum, 1, pd2.af);
5574 
5575 				switch (pd2.af) {
5576 #ifdef INET
5577 				case AF_INET:
5578 					m_copyback(m, off, ICMP_MINLEN,
5579 					    pd->hdr.icmp);
5580 					m_copyback(m, ipoff2, sizeof(h2),
5581 					    &h2);
5582 					break;
5583 #endif /* INET */
5584 #ifdef INET6
5585 				case AF_INET6:
5586 					m_copyback(m, off,
5587 					    sizeof(struct icmp6_hdr),
5588 					    pd->hdr.icmp6);
5589 					m_copyback(m, ipoff2, sizeof(h2_6),
5590 					    &h2_6);
5591 					break;
5592 #endif /* INET6 */
5593 				}
5594 				m->m_flags &= ~M_HASH;
5595 				m_copyback(m, off2, sizeof(uh), &uh);
5596 			}
5597 			break;
5598 		}
5599 #ifdef INET
5600 		case IPPROTO_ICMP: {
5601 			struct icmp		iih;
5602 
5603 			if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
5604 			    NULL, reason, pd2.af)) {
5605 				DPFPRINTF(PF_DEBUG_MISC,
5606 				    ("pf: ICMP error message too short i"
5607 				    "(icmp)\n"));
5608 				return (PF_DROP);
5609 			}
5610 
5611 			key.af = pd2.af;
5612 			key.proto = IPPROTO_ICMP;
5613 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5614 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5615 			key.port[0] = key.port[1] = iih.icmp_id;
5616 
5617 			STATE_LOOKUP(kif, &key, direction, *state, m);
5618 			lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5619 
5620 			/* translate source/destination address, if necessary */
5621 			if ((*state)->key[PF_SK_WIRE] !=
5622 			    (*state)->key[PF_SK_STACK]) {
5623 				struct pf_state_key *nk =
5624 				    (*state)->key[pd->didx];
5625 
5626 				if (PF_ANEQ(pd2.src,
5627 				    &nk->addr[pd2.sidx], pd2.af) ||
5628 				    nk->port[pd2.sidx] != iih.icmp_id)
5629 					pf_change_icmp(pd2.src, &iih.icmp_id,
5630 					    daddr, &nk->addr[pd2.sidx],
5631 					    nk->port[pd2.sidx], NULL,
5632 					    pd2.ip_sum, icmpsum,
5633 					    pd->ip_sum, 0, AF_INET);
5634 
5635 				if (PF_ANEQ(pd2.dst,
5636 				    &nk->addr[pd2.didx], pd2.af) ||
5637 				    nk->port[pd2.didx] != iih.icmp_id)
5638 					pf_change_icmp(pd2.dst, &iih.icmp_id,
5639 					    NULL, /* XXX Inbound NAT? */
5640 					    &nk->addr[pd2.didx],
5641 					    nk->port[pd2.didx], NULL,
5642 					    pd2.ip_sum, icmpsum,
5643 					    pd->ip_sum, 0, AF_INET);
5644 
5645 				m_copyback(m, off, ICMP_MINLEN, pd->hdr.icmp);
5646 				m_copyback(m, ipoff2, sizeof(h2), &h2);
5647 				m_copyback(m, off2, ICMP_MINLEN, &iih);
5648 				m->m_flags &= ~M_HASH;
5649 			}
5650 			break;
5651 		}
5652 #endif /* INET */
5653 #ifdef INET6
5654 		case IPPROTO_ICMPV6: {
5655 			struct icmp6_hdr	iih;
5656 
5657 			if (!pf_pull_hdr(m, off2, &iih,
5658 			    sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
5659 				DPFPRINTF(PF_DEBUG_MISC,
5660 				    ("pf: ICMP error message too short "
5661 				    "(icmp6)\n"));
5662 				FAIL (PF_DROP);
5663 			}
5664 
5665 			key.af = pd2.af;
5666 			key.proto = IPPROTO_ICMPV6;
5667 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5668 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5669 			key.port[0] = key.port[1] = iih.icmp6_id;
5670 
5671 			STATE_LOOKUP(kif, &key, direction, *state, m);
5672 			lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5673 
5674 			/* translate source/destination address, if necessary */
5675 			if ((*state)->key[PF_SK_WIRE] !=
5676 			    (*state)->key[PF_SK_STACK]) {
5677 				struct pf_state_key *nk =
5678 				    (*state)->key[pd->didx];
5679 
5680 				if (PF_ANEQ(pd2.src,
5681 				    &nk->addr[pd2.sidx], pd2.af) ||
5682 				    nk->port[pd2.sidx] != iih.icmp6_id)
5683 					pf_change_icmp(pd2.src, &iih.icmp6_id,
5684 					    daddr, &nk->addr[pd2.sidx],
5685 					    nk->port[pd2.sidx], NULL,
5686 					    pd2.ip_sum, icmpsum,
5687 					    pd->ip_sum, 0, AF_INET6);
5688 
5689 				if (PF_ANEQ(pd2.dst,
5690 				    &nk->addr[pd2.didx], pd2.af) ||
5691 				    nk->port[pd2.didx] != iih.icmp6_id)
5692 					pf_change_icmp(pd2.dst, &iih.icmp6_id,
5693 					    NULL, /* XXX Inbound NAT? */
5694 					    &nk->addr[pd2.didx],
5695 					    nk->port[pd2.didx], NULL,
5696 					    pd2.ip_sum, icmpsum,
5697 					    pd->ip_sum, 0, AF_INET6);
5698 
5699 				m_copyback(m, off, sizeof(struct icmp6_hdr),
5700 				    pd->hdr.icmp6);
5701 				m_copyback(m, ipoff2, sizeof(h2_6), &h2_6);
5702 				m_copyback(m, off2, sizeof(struct icmp6_hdr),
5703 				    &iih);
5704 				m->m_flags &= ~M_HASH;
5705 			}
5706 			break;
5707 		}
5708 #endif /* INET6 */
5709 		default: {
5710 			key.af = pd2.af;
5711 			key.proto = pd2.proto;
5712 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5713 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5714 			key.port[0] = key.port[1] = 0;
5715 
5716 			STATE_LOOKUP(kif, &key, direction, *state, m);
5717 			lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5718 
5719 			/* translate source/destination address, if necessary */
5720 			if ((*state)->key[PF_SK_WIRE] !=
5721 			    (*state)->key[PF_SK_STACK]) {
5722 				struct pf_state_key *nk =
5723 				    (*state)->key[pd->didx];
5724 
5725 				if (PF_ANEQ(pd2.src,
5726 				    &nk->addr[pd2.sidx], pd2.af))
5727 					pf_change_icmp(pd2.src, NULL, daddr,
5728 					    &nk->addr[pd2.sidx], 0, NULL,
5729 					    pd2.ip_sum, icmpsum,
5730 					    pd->ip_sum, 0, pd2.af);
5731 
5732 				if (PF_ANEQ(pd2.dst,
5733 				    &nk->addr[pd2.didx], pd2.af))
5734 					pf_change_icmp(pd2.src, NULL,
5735 					    NULL, /* XXX Inbound NAT? */
5736 					    &nk->addr[pd2.didx], 0, NULL,
5737 					    pd2.ip_sum, icmpsum,
5738 					    pd->ip_sum, 0, pd2.af);
5739 
5740 				switch (pd2.af) {
5741 #ifdef INET
5742 				case AF_INET:
5743 					m_copyback(m, off, ICMP_MINLEN,
5744 					    pd->hdr.icmp);
5745 					m_copyback(m, ipoff2, sizeof(h2),
5746 					    &h2);
5747 					m->m_flags &= ~M_HASH;
5748 					break;
5749 #endif /* INET */
5750 #ifdef INET6
5751 				case AF_INET6:
5752 					m_copyback(m, off,
5753 					    sizeof(struct icmp6_hdr),
5754 					    pd->hdr.icmp6);
5755 					m_copyback(m, ipoff2, sizeof(h2_6),
5756 					    &h2_6);
5757 					m->m_flags &= ~M_HASH;
5758 					break;
5759 #endif /* INET6 */
5760 				}
5761 			}
5762 			break;
5763 		}
5764 		}
5765 	}
5766 
5767 	pfsync_update_state(*state);
5768 	error = PF_PASS;
5769 done:
5770 	if (*state)
5771 		lockmgr(&(*state)->lk, LK_RELEASE);
5772 	return (error);
5773 }
5774 
5775 /*
5776  * Test other connection state.  Caller must hold the state locked.
5777  */
5778 int
pf_test_state_other(struct pf_state ** state,int direction,struct pfi_kif * kif,struct mbuf * m,struct pf_pdesc * pd)5779 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
5780 		    struct mbuf *m, struct pf_pdesc *pd)
5781 {
5782 	struct pf_state_peer	*src, *dst;
5783 	struct pf_state_key_cmp	 key;
5784 
5785 	bzero(&key, sizeof(key));
5786 	key.af = pd->af;
5787 	key.proto = pd->proto;
5788 	if (direction == PF_IN)	{
5789 		PF_ACPY(&key.addr[0], pd->src, key.af);
5790 		PF_ACPY(&key.addr[1], pd->dst, key.af);
5791 		key.port[0] = key.port[1] = 0;
5792 	} else {
5793 		PF_ACPY(&key.addr[1], pd->src, key.af);
5794 		PF_ACPY(&key.addr[0], pd->dst, key.af);
5795 		key.port[1] = key.port[0] = 0;
5796 	}
5797 
5798 	STATE_LOOKUP(kif, &key, direction, *state, m);
5799 	lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5800 
5801 	if (direction == (*state)->direction) {
5802 		src = &(*state)->src;
5803 		dst = &(*state)->dst;
5804 	} else {
5805 		src = &(*state)->dst;
5806 		dst = &(*state)->src;
5807 	}
5808 
5809 	/* update states */
5810 	if (src->state < PFOTHERS_SINGLE)
5811 		src->state = PFOTHERS_SINGLE;
5812 	if (dst->state == PFOTHERS_SINGLE)
5813 		dst->state = PFOTHERS_MULTIPLE;
5814 
5815 	/* update expire time */
5816 	(*state)->expire = time_second;
5817 	if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5818 		(*state)->timeout = PFTM_OTHER_MULTIPLE;
5819 	else
5820 		(*state)->timeout = PFTM_OTHER_SINGLE;
5821 
5822 	/* translate source/destination address, if necessary */
5823 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5824 		struct pf_state_key *nk = (*state)->key[pd->didx];
5825 
5826 		KKASSERT(nk);
5827 		KKASSERT(pd);
5828 		KKASSERT(pd->src);
5829 		KKASSERT(pd->dst);
5830 		switch (pd->af) {
5831 #ifdef INET
5832 		case AF_INET:
5833 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5834 				pf_change_a(&pd->src->v4.s_addr,
5835 				    pd->ip_sum,
5836 				    nk->addr[pd->sidx].v4.s_addr,
5837 				    0);
5838 
5839 
5840 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5841 				pf_change_a(&pd->dst->v4.s_addr,
5842 				    pd->ip_sum,
5843 				    nk->addr[pd->didx].v4.s_addr,
5844 				    0);
5845 
5846 			break;
5847 #endif /* INET */
5848 #ifdef INET6
5849 		case AF_INET6:
5850 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET6))
5851 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5852 
5853 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET6))
5854 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5855 #endif /* INET6 */
5856 		}
5857 	}
5858 
5859 	pfsync_update_state(*state);
5860 	lockmgr(&(*state)->lk, LK_RELEASE);
5861 	return (PF_PASS);
5862 }
5863 
5864 /*
5865  * ipoff and off are measured from the start of the mbuf chain.
5866  * h must be at "ipoff" on the mbuf chain.
5867  */
5868 void *
pf_pull_hdr(struct mbuf * m,int off,void * p,int len,u_short * actionp,u_short * reasonp,sa_family_t af)5869 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5870     u_short *actionp, u_short *reasonp, sa_family_t af)
5871 {
5872 	switch (af) {
5873 #ifdef INET
5874 	case AF_INET: {
5875 		struct ip	*h = mtod(m, struct ip *);
5876 		u_int16_t	 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5877 
5878 		if (fragoff) {
5879 			if (fragoff >= len)
5880 				ACTION_SET(actionp, PF_PASS);
5881 			else {
5882 				ACTION_SET(actionp, PF_DROP);
5883 				REASON_SET(reasonp, PFRES_FRAG);
5884 			}
5885 			return (NULL);
5886 		}
5887 		if (m->m_pkthdr.len < off + len ||
5888 		    ntohs(h->ip_len) < off + len) {
5889 			ACTION_SET(actionp, PF_DROP);
5890 			REASON_SET(reasonp, PFRES_SHORT);
5891 			return (NULL);
5892 		}
5893 		break;
5894 	}
5895 #endif /* INET */
5896 #ifdef INET6
5897 	case AF_INET6: {
5898 		struct ip6_hdr	*h = mtod(m, struct ip6_hdr *);
5899 
5900 		if (m->m_pkthdr.len < off + len ||
5901 		    (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5902 		    (unsigned)(off + len)) {
5903 			ACTION_SET(actionp, PF_DROP);
5904 			REASON_SET(reasonp, PFRES_SHORT);
5905 			return (NULL);
5906 		}
5907 		break;
5908 	}
5909 #endif /* INET6 */
5910 	}
5911 	m_copydata(m, off, len, p);
5912 	return (p);
5913 }
5914 
5915 int
pf_routable(struct pf_addr * addr,sa_family_t af,struct pfi_kif * kif)5916 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif)
5917 {
5918 	struct sockaddr_in	*dst;
5919 	int			 ret = 1;
5920 	int			 check_mpath;
5921 #ifdef INET6
5922 	struct sockaddr_in6	*dst6;
5923 	struct route_in6	 ro;
5924 #else
5925 	struct route		 ro;
5926 #endif
5927 	struct radix_node	*rn;
5928 	struct rtentry		*rt;
5929 	struct ifnet		*ifp;
5930 
5931 	check_mpath = 0;
5932 	bzero(&ro, sizeof(ro));
5933 	switch (af) {
5934 	case AF_INET:
5935 		dst = satosin(&ro.ro_dst);
5936 		dst->sin_family = AF_INET;
5937 		dst->sin_len = sizeof(*dst);
5938 		dst->sin_addr = addr->v4;
5939 		break;
5940 #ifdef INET6
5941 	case AF_INET6:
5942 		/*
5943 		 * Skip check for addresses with embedded interface scope,
5944 		 * as they would always match anyway.
5945 		 */
5946 		if (IN6_IS_SCOPE_EMBED(&addr->v6))
5947 			goto out;
5948 		dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5949 		dst6->sin6_family = AF_INET6;
5950 		dst6->sin6_len = sizeof(*dst6);
5951 		dst6->sin6_addr = addr->v6;
5952 		break;
5953 #endif /* INET6 */
5954 	default:
5955 		return (0);
5956 	}
5957 
5958 	/* Skip checks for ipsec interfaces */
5959 	if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5960 		goto out;
5961 
5962 	rtalloc_ign((struct route *)&ro, 0);
5963 
5964 	if (ro.ro_rt != NULL) {
5965 		/* No interface given, this is a no-route check */
5966 		if (kif == NULL)
5967 			goto out;
5968 
5969 		if (kif->pfik_ifp == NULL) {
5970 			ret = 0;
5971 			goto out;
5972 		}
5973 
5974 		/* Perform uRPF check if passed input interface */
5975 		ret = 0;
5976 		rn = (struct radix_node *)ro.ro_rt;
5977 		do {
5978 			rt = (struct rtentry *)rn;
5979 			ifp = rt->rt_ifp;
5980 
5981 			if (kif->pfik_ifp == ifp)
5982 				ret = 1;
5983 			rn = NULL;
5984 		} while (check_mpath == 1 && rn != NULL && ret == 0);
5985 	} else
5986 		ret = 0;
5987 out:
5988 	if (ro.ro_rt != NULL)
5989 		RTFREE(ro.ro_rt);
5990 	return (ret);
5991 }
5992 
5993 int
pf_rtlabel_match(struct pf_addr * addr,sa_family_t af,struct pf_addr_wrap * aw)5994 pf_rtlabel_match(struct pf_addr *addr, sa_family_t af, struct pf_addr_wrap *aw)
5995 {
5996 	struct sockaddr_in	*dst;
5997 #ifdef INET6
5998 	struct sockaddr_in6	*dst6;
5999 	struct route_in6	 ro;
6000 #else
6001 	struct route		 ro;
6002 #endif
6003 	int			 ret = 0;
6004 
6005 	ASSERT_LWKT_TOKEN_HELD(&pf_token);
6006 
6007 	bzero(&ro, sizeof(ro));
6008 	switch (af) {
6009 	case AF_INET:
6010 		dst = satosin(&ro.ro_dst);
6011 		dst->sin_family = AF_INET;
6012 		dst->sin_len = sizeof(*dst);
6013 		dst->sin_addr = addr->v4;
6014 		break;
6015 #ifdef INET6
6016 	case AF_INET6:
6017 		dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
6018 		dst6->sin6_family = AF_INET6;
6019 		dst6->sin6_len = sizeof(*dst6);
6020 		dst6->sin6_addr = addr->v6;
6021 		break;
6022 #endif /* INET6 */
6023 	default:
6024 		return (0);
6025 	}
6026 
6027 rtalloc_ign((struct route *)&ro, (RTF_CLONING | RTF_PRCLONING));
6028 
6029 	if (ro.ro_rt != NULL) {
6030 		RTFREE(ro.ro_rt);
6031 	}
6032 
6033 	return (ret);
6034 }
6035 
6036 #ifdef INET
6037 void
pf_route(struct mbuf ** m,struct pf_rule * r,int dir,struct ifnet * oifp,struct pf_state * s,struct pf_pdesc * pd)6038 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
6039     struct pf_state *s, struct pf_pdesc *pd)
6040 {
6041 	struct mbuf		*m0, *m1;
6042 	struct route		 iproute;
6043 	struct route		*ro = NULL;
6044 	struct sockaddr_in	*dst;
6045 	struct ip		*ip;
6046 	struct ifnet		*ifp = NULL;
6047 	struct pf_addr		 naddr;
6048 	struct pf_src_node	*sn = NULL;
6049 	int			 error = 0;
6050 	int sw_csum;
6051 
6052 	ASSERT_LWKT_TOKEN_HELD(&pf_token);
6053 
6054 	if (m == NULL || *m == NULL || r == NULL ||
6055 	    (dir != PF_IN && dir != PF_OUT) || oifp == NULL)
6056 		panic("pf_route: invalid parameters");
6057 
6058 	if (((*m)->m_pkthdr.fw_flags & PF_MBUF_ROUTED) == 0) {
6059 		(*m)->m_pkthdr.fw_flags |= PF_MBUF_ROUTED;
6060 		(*m)->m_pkthdr.pf.routed = 1;
6061 	} else {
6062 		if ((*m)->m_pkthdr.pf.routed++ > 3) {
6063 			m0 = *m;
6064 			*m = NULL;
6065 			goto bad;
6066 		}
6067 	}
6068 
6069 	if (r->rt == PF_DUPTO) {
6070 		if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
6071 			return;
6072 		}
6073 	} else {
6074 		if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
6075 			return;
6076 		}
6077 		m0 = *m;
6078 	}
6079 
6080 	if (m0->m_len < sizeof(struct ip)) {
6081 		DPFPRINTF(PF_DEBUG_URGENT,
6082 		    ("pf_route: m0->m_len < sizeof(struct ip)\n"));
6083 		goto bad;
6084 	}
6085 
6086 	ip = mtod(m0, struct ip *);
6087 
6088 	ro = &iproute;
6089 	bzero((caddr_t)ro, sizeof(*ro));
6090 	dst = satosin(&ro->ro_dst);
6091 	dst->sin_family = AF_INET;
6092 	dst->sin_len = sizeof(*dst);
6093 	dst->sin_addr = ip->ip_dst;
6094 
6095 	if (r->rt == PF_FASTROUTE) {
6096 		rtalloc(ro);
6097 		if (ro->ro_rt == 0) {
6098 			ipstat.ips_noroute++;
6099 			goto bad;
6100 		}
6101 
6102 		ifp = ro->ro_rt->rt_ifp;
6103 		ro->ro_rt->rt_use++;
6104 
6105 		if (ro->ro_rt->rt_flags & RTF_GATEWAY)
6106 			dst = satosin(ro->ro_rt->rt_gateway);
6107 	} else {
6108 		if (TAILQ_EMPTY(&r->rpool.list)) {
6109 			DPFPRINTF(PF_DEBUG_URGENT,
6110 			    ("pf_route: TAILQ_EMPTY(&r->rpool.list)\n"));
6111 			goto bad;
6112 		}
6113 		if (s == NULL) {
6114 			pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
6115 			    &naddr, NULL, &sn);
6116 			if (!PF_AZERO(&naddr, AF_INET))
6117 				dst->sin_addr.s_addr = naddr.v4.s_addr;
6118 			ifp = r->rpool.cur->kif ?
6119 			    r->rpool.cur->kif->pfik_ifp : NULL;
6120 		} else {
6121 			if (!PF_AZERO(&s->rt_addr, AF_INET))
6122 				dst->sin_addr.s_addr =
6123 				    s->rt_addr.v4.s_addr;
6124 			ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
6125 		}
6126 	}
6127 	if (ifp == NULL)
6128 		goto bad;
6129 
6130 	if (oifp != ifp) {
6131 		if (pf_test(PF_OUT, ifp, &m0, NULL, NULL) != PF_PASS) {
6132 			goto bad;
6133 		} else if (m0 == NULL) {
6134 			goto done;
6135 		}
6136 		if (m0->m_len < sizeof(struct ip)) {
6137 			DPFPRINTF(PF_DEBUG_URGENT,
6138 			    ("pf_route: m0->m_len < sizeof(struct ip)\n"));
6139 			goto bad;
6140 		}
6141 		ip = mtod(m0, struct ip *);
6142 	}
6143 
6144 	/* Copied from FreeBSD 5.1-CURRENT ip_output. */
6145 	m0->m_pkthdr.csum_flags |= CSUM_IP;
6146 	sw_csum = m0->m_pkthdr.csum_flags & ~ifp->if_hwassist;
6147 	if (sw_csum & CSUM_DELAY_DATA) {
6148 		in_delayed_cksum(m0);
6149 		sw_csum &= ~CSUM_DELAY_DATA;
6150 	}
6151 	m0->m_pkthdr.csum_flags &= ifp->if_hwassist;
6152 	m0->m_pkthdr.csum_iphlen = (ip->ip_hl << 2);
6153 
6154 	/*
6155 	 * WARNING!  We cannot fragment if the packet was modified from an
6156 	 *	     original which expected to be using TSO.  In this
6157 	 *	     situation we pray that the target interface is
6158 	 *	     compatible with the originating interface.
6159 	 */
6160 	if (ntohs(ip->ip_len) <= ifp->if_mtu ||
6161 	    (m0->m_pkthdr.csum_flags & CSUM_TSO) ||
6162 	    ((ifp->if_hwassist & CSUM_FRAGMENT) &&
6163 		(ip->ip_off & htons(IP_DF)) == 0)) {
6164 		ip->ip_sum = 0;
6165 		if (sw_csum & CSUM_DELAY_IP) {
6166 			/* From KAME */
6167 			if (ip->ip_v == IPVERSION &&
6168 			    (ip->ip_hl << 2) == sizeof(*ip)) {
6169 				ip->ip_sum = in_cksum_hdr(ip);
6170 			} else {
6171 				ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
6172 			}
6173 		}
6174 		lwkt_reltoken(&pf_token);
6175 		error = ifp->if_output(ifp, m0, sintosa(dst), ro->ro_rt);
6176 		lwkt_gettoken(&pf_token);
6177 		goto done;
6178 	}
6179 
6180 	/*
6181 	 * Too large for interface; fragment if possible.
6182 	 * Must be able to put at least 8 bytes per fragment.
6183 	 */
6184 	if (ip->ip_off & htons(IP_DF)) {
6185 		ipstat.ips_cantfrag++;
6186 		if (r->rt != PF_DUPTO) {
6187 			icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
6188 				   ifp->if_mtu);
6189 			goto done;
6190 		} else
6191 			goto bad;
6192 	}
6193 
6194 	m1 = m0;
6195 	error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist, sw_csum);
6196 	if (error) {
6197 		goto bad;
6198 	}
6199 
6200 	for (m0 = m1; m0; m0 = m1) {
6201 		m1 = m0->m_nextpkt;
6202 		m0->m_nextpkt = 0;
6203 		if (error == 0) {
6204 			lwkt_reltoken(&pf_token);
6205 			error = (*ifp->if_output)(ifp, m0, sintosa(dst),
6206 						  NULL);
6207 			lwkt_gettoken(&pf_token);
6208 		} else
6209 			m_freem(m0);
6210 	}
6211 
6212 	if (error == 0)
6213 		ipstat.ips_fragmented++;
6214 
6215 done:
6216 	if (r->rt != PF_DUPTO)
6217 		*m = NULL;
6218 	if (ro == &iproute && ro->ro_rt)
6219 		RTFREE(ro->ro_rt);
6220 	return;
6221 
6222 bad:
6223 	m_freem(m0);
6224 	goto done;
6225 }
6226 #endif /* INET */
6227 
6228 #ifdef INET6
6229 void
pf_route6(struct mbuf ** m,struct pf_rule * r,int dir,struct ifnet * oifp,struct pf_state * s,struct pf_pdesc * pd)6230 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
6231     struct pf_state *s, struct pf_pdesc *pd)
6232 {
6233 	struct mbuf		*m0;
6234 	struct route_in6	 ip6route;
6235 	struct route_in6	*ro;
6236 	struct sockaddr_in6	*dst;
6237 	struct ip6_hdr		*ip6;
6238 	struct ifnet		*ifp = NULL;
6239 	struct pf_addr		 naddr;
6240 	struct pf_src_node	*sn = NULL;
6241 
6242 	if (m == NULL || *m == NULL || r == NULL ||
6243 	    (dir != PF_IN && dir != PF_OUT) || oifp == NULL)
6244 		panic("pf_route6: invalid parameters");
6245 
6246 	if (((*m)->m_pkthdr.fw_flags & PF_MBUF_ROUTED) == 0) {
6247 		(*m)->m_pkthdr.fw_flags |= PF_MBUF_ROUTED;
6248 		(*m)->m_pkthdr.pf.routed = 1;
6249 	} else {
6250 		if ((*m)->m_pkthdr.pf.routed++ > 3) {
6251 			m0 = *m;
6252 			*m = NULL;
6253 			goto bad;
6254 		}
6255 	}
6256 
6257 	if (r->rt == PF_DUPTO) {
6258 		if ((m0 = m_dup(*m, M_NOWAIT)) == NULL)
6259 			return;
6260 	} else {
6261 		if ((r->rt == PF_REPLYTO) == (r->direction == dir))
6262 			return;
6263 		m0 = *m;
6264 	}
6265 
6266 	if (m0->m_len < sizeof(struct ip6_hdr)) {
6267 		DPFPRINTF(PF_DEBUG_URGENT,
6268 		    ("pf_route6: m0->m_len < sizeof(struct ip6_hdr)\n"));
6269 		goto bad;
6270 	}
6271 	ip6 = mtod(m0, struct ip6_hdr *);
6272 
6273 	ro = &ip6route;
6274 	bzero((caddr_t)ro, sizeof(*ro));
6275 	dst = (struct sockaddr_in6 *)&ro->ro_dst;
6276 	dst->sin6_family = AF_INET6;
6277 	dst->sin6_len = sizeof(*dst);
6278 	dst->sin6_addr = ip6->ip6_dst;
6279 
6280 	/*
6281 	 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
6282 	 * so make sure pf.flags is clear.
6283 	 *
6284 	 * Cheat. XXX why only in the v6 case???
6285 	 */
6286 	if (r->rt == PF_FASTROUTE) {
6287 		m0->m_pkthdr.fw_flags |= PF_MBUF_TAGGED;
6288 		m0->m_pkthdr.pf.flags = 0;
6289 		/* XXX Re-Check when Upgrading to > 4.4 */
6290 		m0->m_pkthdr.pf.statekey = NULL;
6291 		ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL);
6292 		return;
6293 	}
6294 
6295 	if (TAILQ_EMPTY(&r->rpool.list)) {
6296 		DPFPRINTF(PF_DEBUG_URGENT,
6297 		    ("pf_route6: TAILQ_EMPTY(&r->rpool.list)\n"));
6298 		goto bad;
6299 	}
6300 	if (s == NULL) {
6301 		pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
6302 		    &naddr, NULL, &sn);
6303 		if (!PF_AZERO(&naddr, AF_INET6))
6304 			PF_ACPY((struct pf_addr *)&dst->sin6_addr,
6305 			    &naddr, AF_INET6);
6306 		ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
6307 	} else {
6308 		if (!PF_AZERO(&s->rt_addr, AF_INET6))
6309 			PF_ACPY((struct pf_addr *)&dst->sin6_addr,
6310 			    &s->rt_addr, AF_INET6);
6311 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
6312 	}
6313 	if (ifp == NULL)
6314 		goto bad;
6315 
6316 	if (oifp != ifp) {
6317 		if (pf_test6(PF_OUT, ifp, &m0, NULL, NULL) != PF_PASS) {
6318 			goto bad;
6319 		} else if (m0 == NULL) {
6320 			goto done;
6321 		}
6322 		if (m0->m_len < sizeof(struct ip6_hdr)) {
6323 			DPFPRINTF(PF_DEBUG_URGENT,
6324 			    ("pf_route6: m0->m_len < sizeof(struct ip6_hdr)\n"));
6325 			goto bad;
6326 		}
6327 		ip6 = mtod(m0, struct ip6_hdr *);
6328 	}
6329 
6330 	/*
6331 	 * If the packet is too large for the outgoing interface,
6332 	 * send back an icmp6 error.
6333 	 */
6334 	if (IN6_IS_SCOPE_EMBED(&dst->sin6_addr))
6335 		dst->sin6_addr.s6_addr16[1] = htons(ifp->if_index);
6336 	if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) {
6337 		nd6_output(ifp, ifp, m0, dst, NULL);
6338 	} else {
6339 		in6_ifstat_inc(ifp, ifs6_in_toobig);
6340 		if (r->rt != PF_DUPTO)
6341 			icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
6342 		else
6343 			goto bad;
6344 	}
6345 
6346 done:
6347 	if (r->rt != PF_DUPTO)
6348 		*m = NULL;
6349 	return;
6350 
6351 bad:
6352 	m_freem(m0);
6353 	goto done;
6354 }
6355 #endif /* INET6 */
6356 
6357 
6358 /*
6359  * check protocol (tcp/udp/icmp/icmp6) checksum and set mbuf flag
6360  *   off is the offset where the protocol header starts
6361  *   len is the total length of protocol header plus payload
6362  * returns 0 when the checksum is valid, otherwise returns 1.
6363  */
6364 /*
6365  * XXX
6366  * FreeBSD supports cksum offload for the following drivers.
6367  * em(4), gx(4), lge(4), nge(4), ti(4), xl(4)
6368  * If we can make full use of it we would outperform ipfw/ipfilter in
6369  * very heavy traffic.
6370  * I have not tested 'cause I don't have NICs that supports cksum offload.
6371  * (There might be problems. Typical phenomena would be
6372  *   1. No route message for UDP packet.
6373  *   2. No connection acceptance from external hosts regardless of rule set.)
6374  */
6375 int
pf_check_proto_cksum(struct mbuf * m,int off,int len,u_int8_t p,sa_family_t af)6376 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p,
6377     sa_family_t af)
6378 {
6379 	u_int16_t sum = 0;
6380 	int hw_assist = 0;
6381 	struct ip *ip;
6382 
6383 	if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
6384 		return (1);
6385 	if (m->m_pkthdr.len < off + len)
6386 		return (1);
6387 
6388 	switch (p) {
6389 	case IPPROTO_TCP:
6390 	case IPPROTO_UDP:
6391 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
6392 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
6393 				sum = m->m_pkthdr.csum_data;
6394 			} else {
6395 				ip = mtod(m, struct ip *);
6396 				sum = in_pseudo(ip->ip_src.s_addr,
6397 					ip->ip_dst.s_addr, htonl((u_short)len +
6398 					m->m_pkthdr.csum_data + p));
6399 			}
6400 			sum ^= 0xffff;
6401 			++hw_assist;
6402 		}
6403 		break;
6404 	case IPPROTO_ICMP:
6405 #ifdef INET6
6406 	case IPPROTO_ICMPV6:
6407 #endif /* INET6 */
6408 		break;
6409 	default:
6410 		return (1);
6411 	}
6412 
6413 	if (!hw_assist) {
6414 		switch (af) {
6415 		case AF_INET:
6416 			if (p == IPPROTO_ICMP) {
6417 				if (m->m_len < off)
6418 					return (1);
6419 				m->m_data += off;
6420 				m->m_len -= off;
6421 				sum = in_cksum(m, len);
6422 				m->m_data -= off;
6423 				m->m_len += off;
6424 			} else {
6425 				if (m->m_len < sizeof(struct ip))
6426 					return (1);
6427 				sum = in_cksum_range(m, p, off, len);
6428 				if (sum == 0) {
6429 					m->m_pkthdr.csum_flags |=
6430 					    (CSUM_DATA_VALID |
6431 					     CSUM_PSEUDO_HDR);
6432 					m->m_pkthdr.csum_data = 0xffff;
6433 				}
6434 			}
6435 			break;
6436 #ifdef INET6
6437 		case AF_INET6:
6438 			if (m->m_len < sizeof(struct ip6_hdr))
6439 				return (1);
6440 			sum = in6_cksum(m, p, off, len);
6441 			/*
6442 			 * XXX
6443 			 * IPv6 H/W cksum off-load not supported yet!
6444 			 *
6445 			 * if (sum == 0) {
6446 			 *	m->m_pkthdr.csum_flags |=
6447 			 *	    (CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
6448 			 *	m->m_pkthdr.csum_data = 0xffff;
6449 			 *}
6450 			 */
6451 			break;
6452 #endif /* INET6 */
6453 		default:
6454 			return (1);
6455 		}
6456 	}
6457 	if (sum) {
6458 		switch (p) {
6459 		case IPPROTO_TCP:
6460 			tcpstat.tcps_rcvbadsum++;
6461 			break;
6462 		case IPPROTO_UDP:
6463 			udp_stat.udps_badsum++;
6464 			break;
6465 		case IPPROTO_ICMP:
6466 			icmpstat.icps_checksum++;
6467 			break;
6468 #ifdef INET6
6469 		case IPPROTO_ICMPV6:
6470 			icmp6stat.icp6s_checksum++;
6471 			break;
6472 #endif /* INET6 */
6473 		}
6474 		return (1);
6475 	}
6476 	return (0);
6477 }
6478 
6479 struct pf_divert *
pf_find_divert(struct mbuf * m)6480 pf_find_divert(struct mbuf *m)
6481 {
6482 	struct m_tag    *mtag;
6483 
6484 	if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL)
6485 		return (NULL);
6486 
6487 	return ((struct pf_divert *)(mtag + 1));
6488 }
6489 
6490 struct pf_divert *
pf_get_divert(struct mbuf * m)6491 pf_get_divert(struct mbuf *m)
6492 {
6493 	struct m_tag    *mtag;
6494 
6495 	if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL) {
6496 		mtag = m_tag_get(PACKET_TAG_PF_DIVERT, sizeof(struct pf_divert),
6497 		    M_NOWAIT);
6498 		if (mtag == NULL)
6499 			return (NULL);
6500 		bzero(mtag + 1, sizeof(struct pf_divert));
6501 		m_tag_prepend(m, mtag);
6502 	}
6503 
6504 	return ((struct pf_divert *)(mtag + 1));
6505 }
6506 
6507 #ifdef INET
6508 
6509 /*
6510  * WARNING: pf_token held shared on entry, THIS IS CPU LOCALIZED CODE
6511  */
6512 int
pf_test(int dir,struct ifnet * ifp,struct mbuf ** m0,struct ether_header * eh,struct inpcb * inp)6513 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0,
6514     struct ether_header *eh, struct inpcb *inp)
6515 {
6516 	struct pfi_kif		*kif;
6517 	u_short			 action, reason = 0, log = 0;
6518 	struct mbuf		*m = *m0;
6519 	struct ip		*h = NULL;
6520 	struct pf_rule		*a = NULL, *r = &pf_default_rule, *tr, *nr;
6521 	struct pf_state		*s = NULL;
6522 	struct pf_ruleset	*ruleset = NULL;
6523 	struct pf_pdesc		 pd;
6524 	int			 off, dirndx;
6525 #ifdef ALTQ
6526 	int			 pqid = 0;
6527 #endif
6528 
6529 	if (m->m_pkthdr.fw_flags & IPFW_MBUF_CONTINUE) {
6530 		/* Skip us; continue in ipfw. */
6531 		return (PF_PASS);
6532 	}
6533 
6534 	if (!pf_status.running)
6535 		return (PF_PASS);
6536 
6537 	memset(&pd, 0, sizeof(pd));
6538 #ifdef foo
6539 	if (ifp->if_type == IFT_CARP && ifp->if_carpdev)
6540 		kif = (struct pfi_kif *)ifp->if_carpdev->if_pf_kif;
6541 	else
6542 #endif
6543 		kif = (struct pfi_kif *)ifp->if_pf_kif;
6544 
6545 	if (kif == NULL) {
6546 		DPFPRINTF(PF_DEBUG_URGENT,
6547 		    ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
6548 		return (PF_DROP);
6549 	}
6550 	if (kif->pfik_flags & PFI_IFLAG_SKIP)
6551 		return (PF_PASS);
6552 
6553 #ifdef DIAGNOSTIC
6554 	if ((m->m_flags & M_PKTHDR) == 0)
6555 		panic("non-M_PKTHDR is passed to pf_test");
6556 #endif /* DIAGNOSTIC */
6557 
6558 	if (m->m_pkthdr.len < (int)sizeof(*h)) {
6559 		action = PF_DROP;
6560 		REASON_SET(&reason, PFRES_SHORT);
6561 		log = 1;
6562 		goto done;
6563 	}
6564 
6565 	/*
6566 	 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
6567 	 * so make sure pf.flags is clear.
6568 	 */
6569 	if (m->m_pkthdr.fw_flags & PF_MBUF_TAGGED)
6570 		return (PF_PASS);
6571 	m->m_pkthdr.pf.flags = 0;
6572 	/* Re-Check when updating to > 4.4 */
6573 	m->m_pkthdr.pf.statekey = NULL;
6574 
6575 	/* We do IP header normalization and packet reassembly here */
6576 	if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
6577 		action = PF_DROP;
6578 		goto done;
6579 	}
6580 	m = *m0;	/* pf_normalize messes with m0 */
6581 	h = mtod(m, struct ip *);
6582 
6583 	off = h->ip_hl << 2;
6584 	if (off < (int)sizeof(*h)) {
6585 		action = PF_DROP;
6586 		REASON_SET(&reason, PFRES_SHORT);
6587 		log = 1;
6588 		goto done;
6589 	}
6590 
6591 	pd.src = (struct pf_addr *)&h->ip_src;
6592 	pd.dst = (struct pf_addr *)&h->ip_dst;
6593 	pd.sport = pd.dport = NULL;
6594 	pd.ip_sum = &h->ip_sum;
6595 	pd.proto_sum = NULL;
6596 	pd.proto = h->ip_p;
6597 	pd.dir = dir;
6598 	pd.sidx = (dir == PF_IN) ? 0 : 1;
6599 	pd.didx = (dir == PF_IN) ? 1 : 0;
6600 	pd.af = AF_INET;
6601 	pd.tos = h->ip_tos;
6602 	pd.tot_len = ntohs(h->ip_len);
6603 	pd.eh = eh;
6604 
6605 	/* handle fragments that didn't get reassembled by normalization */
6606 	if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
6607 		action = pf_test_fragment(&r, dir, kif, m, h,
6608 		    &pd, &a, &ruleset);
6609 		goto done;
6610 	}
6611 
6612 	switch (h->ip_p) {
6613 
6614 	case IPPROTO_TCP: {
6615 		struct tcphdr	th;
6616 
6617 		pd.hdr.tcp = &th;
6618 		if (!pf_pull_hdr(m, off, &th, sizeof(th),
6619 		    &action, &reason, AF_INET)) {
6620 			log = action != PF_PASS;
6621 			goto done;
6622 		}
6623 		pd.p_len = pd.tot_len - off - (th.th_off << 2);
6624 #ifdef ALTQ
6625 		if ((th.th_flags & TH_ACK) && pd.p_len == 0)
6626 			pqid = 1;
6627 #endif
6628 		action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6629 		if (action == PF_DROP)
6630 			goto done;
6631 		action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6632 					   &reason);
6633 		if (action == PF_PASS) {
6634 			r = s->rule.ptr;
6635 			a = s->anchor.ptr;
6636 			log = s->log;
6637 		} else if (s == NULL) {
6638 			action = pf_test_rule(&r, &s, dir, kif,
6639 					      m, off, h, &pd, &a,
6640 					      &ruleset, NULL, inp);
6641 		}
6642 		break;
6643 	}
6644 
6645 	case IPPROTO_UDP: {
6646 		struct udphdr	uh;
6647 
6648 		pd.hdr.udp = &uh;
6649 		if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6650 		    &action, &reason, AF_INET)) {
6651 			log = action != PF_PASS;
6652 			goto done;
6653 		}
6654 		if (uh.uh_dport == 0 ||
6655 		    ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6656 		    ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6657 			action = PF_DROP;
6658 			REASON_SET(&reason, PFRES_SHORT);
6659 			goto done;
6660 		}
6661 		action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6662 		if (action == PF_PASS) {
6663 			r = s->rule.ptr;
6664 			a = s->anchor.ptr;
6665 			log = s->log;
6666 		} else if (s == NULL) {
6667 			action = pf_test_rule(&r, &s, dir, kif,
6668 					      m, off, h, &pd, &a,
6669 					      &ruleset, NULL, inp);
6670 		}
6671 		break;
6672 	}
6673 
6674 	case IPPROTO_ICMP: {
6675 		struct icmp	ih;
6676 
6677 		pd.hdr.icmp = &ih;
6678 		if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
6679 		    &action, &reason, AF_INET)) {
6680 			log = action != PF_PASS;
6681 			goto done;
6682 		}
6683 		action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
6684 					    &reason);
6685 		if (action == PF_PASS) {
6686 			r = s->rule.ptr;
6687 			a = s->anchor.ptr;
6688 			log = s->log;
6689 		} else if (s == NULL) {
6690 			action = pf_test_rule(&r, &s, dir, kif,
6691 					      m, off, h, &pd, &a,
6692 					      &ruleset, NULL, inp);
6693 		}
6694 		break;
6695 	}
6696 
6697 	default:
6698 		action = pf_test_state_other(&s, dir, kif, m, &pd);
6699 		if (action == PF_PASS) {
6700 			r = s->rule.ptr;
6701 			a = s->anchor.ptr;
6702 			log = s->log;
6703 		} else if (s == NULL) {
6704 			action = pf_test_rule(&r, &s, dir, kif, m, off, h,
6705 					      &pd, &a, &ruleset, NULL, inp);
6706 		}
6707 		break;
6708 	}
6709 
6710 done:
6711 	if (action == PF_PASS && h->ip_hl > 5 &&
6712 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6713 		action = PF_DROP;
6714 		REASON_SET(&reason, PFRES_IPOPTIONS);
6715 		log = 1;
6716 		DPFPRINTF(PF_DEBUG_MISC,
6717 		    ("pf: dropping packet with ip options\n"));
6718 	}
6719 
6720 	if ((s && s->tag) || r->rtableid)
6721 		pf_tag_packet(m, s ? s->tag : 0, r->rtableid);
6722 
6723 #if 0
6724 	if (dir == PF_IN && s && s->key[PF_SK_STACK])
6725 		m->m_pkthdr.pf.statekey = s->key[PF_SK_STACK];
6726 #endif
6727 
6728 #ifdef ALTQ
6729 	/*
6730 	 * Generate a hash code and qid request for ALTQ.  A qid of 0
6731 	 * is allowed and will cause altq to select the default queue.
6732 	 */
6733 	if (action == PF_PASS) {
6734 		m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE;
6735 		if (pqid || (pd.tos & IPTOS_LOWDELAY))
6736 			m->m_pkthdr.pf.qid = r->pqid;
6737 		else
6738 			m->m_pkthdr.pf.qid = r->qid;
6739 		m->m_pkthdr.pf.ecn_af = AF_INET;
6740 		m->m_pkthdr.pf.hdr = h;
6741 		/* add connection hash for fairq */
6742 		if (s) {
6743 			/* for fairq */
6744 			m->m_pkthdr.pf.state_hash = s->hash;
6745 			m->m_pkthdr.pf.flags |= PF_TAG_STATE_HASHED;
6746 		}
6747 	}
6748 #endif /* ALTQ */
6749 
6750 	/*
6751 	 * connections redirected to loopback should not match sockets
6752 	 * bound specifically to loopback due to security implications,
6753 	 * see tcp_input() and in_pcblookup_listen().
6754 	 */
6755 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6756 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6757 	    (s->nat_rule.ptr->action == PF_RDR ||
6758 	    s->nat_rule.ptr->action == PF_BINAT) &&
6759 	    (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
6760 	{
6761 		m->m_pkthdr.pf.flags |= PF_TAG_TRANSLATE_LOCALHOST;
6762 	}
6763 
6764 	if (dir == PF_IN && action == PF_PASS && r->divert.port) {
6765 		struct pf_divert *divert;
6766 
6767 		if ((divert = pf_get_divert(m))) {
6768 			m->m_pkthdr.pf.flags |= PF_TAG_DIVERTED;
6769 			divert->port = r->divert.port;
6770 			divert->addr.ipv4 = r->divert.addr.v4;
6771 		}
6772 	}
6773 
6774 	if (log) {
6775 		struct pf_rule *lr;
6776 
6777 		if (s != NULL && s->nat_rule.ptr != NULL &&
6778 		    s->nat_rule.ptr->log & PF_LOG_ALL)
6779 			lr = s->nat_rule.ptr;
6780 		else
6781 			lr = r;
6782 		PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, lr, a, ruleset,
6783 		    &pd);
6784 	}
6785 
6786 	kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6787 	kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
6788 
6789 	if (action == PF_PASS || r->action == PF_DROP) {
6790 		dirndx = (dir == PF_OUT);
6791 		r->packets[dirndx]++;
6792 		r->bytes[dirndx] += pd.tot_len;
6793 		if (a != NULL) {
6794 			a->packets[dirndx]++;
6795 			a->bytes[dirndx] += pd.tot_len;
6796 		}
6797 		if (s != NULL) {
6798 			if (s->nat_rule.ptr != NULL) {
6799 				s->nat_rule.ptr->packets[dirndx]++;
6800 				s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6801 			}
6802 			if (s->src_node != NULL) {
6803 				s->src_node->packets[dirndx]++;
6804 				s->src_node->bytes[dirndx] += pd.tot_len;
6805 			}
6806 			if (s->nat_src_node != NULL) {
6807 				s->nat_src_node->packets[dirndx]++;
6808 				s->nat_src_node->bytes[dirndx] += pd.tot_len;
6809 			}
6810 			dirndx = (dir == s->direction) ? 0 : 1;
6811 			s->packets[dirndx]++;
6812 			s->bytes[dirndx] += pd.tot_len;
6813 		}
6814 		tr = r;
6815 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6816 		if (nr != NULL && r == &pf_default_rule)
6817 			tr = nr;
6818 		if (tr->src.addr.type == PF_ADDR_TABLE)
6819 			pfr_update_stats(tr->src.addr.p.tbl,
6820 			    (s == NULL) ? pd.src :
6821 			    &s->key[(s->direction == PF_IN)]->
6822 				addr[(s->direction == PF_OUT)],
6823 			    pd.af, pd.tot_len, dir == PF_OUT,
6824 			    r->action == PF_PASS, tr->src.neg);
6825 		if (tr->dst.addr.type == PF_ADDR_TABLE)
6826 			pfr_update_stats(tr->dst.addr.p.tbl,
6827 			    (s == NULL) ? pd.dst :
6828 			    &s->key[(s->direction == PF_IN)]->
6829 				addr[(s->direction == PF_IN)],
6830 			    pd.af, pd.tot_len, dir == PF_OUT,
6831 			    r->action == PF_PASS, tr->dst.neg);
6832 	}
6833 
6834 
6835 	if (action == PF_SYNPROXY_DROP) {
6836 		m_freem(*m0);
6837 		*m0 = NULL;
6838 		action = PF_PASS;
6839 	} else if (r->rt) {
6840 		/* pf_route can free the mbuf causing *m0 to become NULL */
6841 		pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
6842 	}
6843 
6844 	return (action);
6845 }
6846 #endif /* INET */
6847 
6848 #ifdef INET6
6849 
6850 /*
6851  * WARNING: pf_token held shared on entry, THIS IS CPU LOCALIZED CODE
6852  */
6853 int
pf_test6(int dir,struct ifnet * ifp,struct mbuf ** m0,struct ether_header * eh,struct inpcb * inp)6854 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0,
6855     struct ether_header *eh, struct inpcb *inp)
6856 {
6857 	struct pfi_kif		*kif;
6858 	u_short			 action, reason = 0, log = 0;
6859 	struct mbuf		*m = *m0, *n = NULL;
6860 	struct ip6_hdr		*h = NULL;
6861 	struct pf_rule		*a = NULL, *r = &pf_default_rule, *tr, *nr;
6862 	struct pf_state		*s = NULL;
6863 	struct pf_ruleset	*ruleset = NULL;
6864 	struct pf_pdesc		 pd;
6865 	int			 off, terminal = 0, dirndx, rh_cnt = 0;
6866 
6867 	if (!pf_status.running)
6868 		return (PF_PASS);
6869 
6870 	memset(&pd, 0, sizeof(pd));
6871 #ifdef foo
6872 	if (ifp->if_type == IFT_CARP && ifp->if_carpdev)
6873 		kif = (struct pfi_kif *)ifp->if_carpdev->if_pf_kif;
6874 	else
6875 #endif
6876 		kif = (struct pfi_kif *)ifp->if_pf_kif;
6877 
6878 	if (kif == NULL) {
6879 		DPFPRINTF(PF_DEBUG_URGENT,
6880 		    ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6881 		return (PF_DROP);
6882 	}
6883 	if (kif->pfik_flags & PFI_IFLAG_SKIP)
6884 		return (PF_PASS);
6885 
6886 #ifdef DIAGNOSTIC
6887 	if ((m->m_flags & M_PKTHDR) == 0)
6888 		panic("non-M_PKTHDR is passed to pf_test6");
6889 #endif /* DIAGNOSTIC */
6890 
6891 	if (m->m_pkthdr.len < (int)sizeof(*h)) {
6892 		action = PF_DROP;
6893 		REASON_SET(&reason, PFRES_SHORT);
6894 		log = 1;
6895 		goto done;
6896 	}
6897 
6898 	/*
6899 	 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
6900 	 * so make sure pf.flags is clear.
6901 	 */
6902 	if (m->m_pkthdr.fw_flags & PF_MBUF_TAGGED)
6903 		return (PF_PASS);
6904 	m->m_pkthdr.pf.flags = 0;
6905 	/* Re-Check when updating to > 4.4 */
6906 	m->m_pkthdr.pf.statekey = NULL;
6907 
6908 	/* We do IP header normalization and packet reassembly here */
6909 	if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6910 		action = PF_DROP;
6911 		goto done;
6912 	}
6913 	m = *m0;	/* pf_normalize messes with m0 */
6914 	h = mtod(m, struct ip6_hdr *);
6915 
6916 #if 1
6917 	/*
6918 	 * we do not support jumbogram yet.  if we keep going, zero ip6_plen
6919 	 * will do something bad, so drop the packet for now.
6920 	 */
6921 	if (htons(h->ip6_plen) == 0) {
6922 		action = PF_DROP;
6923 		REASON_SET(&reason, PFRES_NORM);	/*XXX*/
6924 		goto done;
6925 	}
6926 #endif
6927 
6928 	pd.src = (struct pf_addr *)&h->ip6_src;
6929 	pd.dst = (struct pf_addr *)&h->ip6_dst;
6930 	pd.sport = pd.dport = NULL;
6931 	pd.ip_sum = NULL;
6932 	pd.proto_sum = NULL;
6933 	pd.dir = dir;
6934 	pd.sidx = (dir == PF_IN) ? 0 : 1;
6935 	pd.didx = (dir == PF_IN) ? 1 : 0;
6936 	pd.af = AF_INET6;
6937 	pd.tos = 0;
6938 	pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6939 	pd.eh = eh;
6940 
6941 	off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6942 	pd.proto = h->ip6_nxt;
6943 	do {
6944 		switch (pd.proto) {
6945 		case IPPROTO_FRAGMENT:
6946 			action = pf_test_fragment(&r, dir, kif, m, h,
6947 			    &pd, &a, &ruleset);
6948 			if (action == PF_DROP)
6949 				REASON_SET(&reason, PFRES_FRAG);
6950 			goto done;
6951 		case IPPROTO_ROUTING: {
6952 			struct ip6_rthdr rthdr;
6953 
6954 			if (rh_cnt++) {
6955 				DPFPRINTF(PF_DEBUG_MISC,
6956 				    ("pf: IPv6 more than one rthdr\n"));
6957 				action = PF_DROP;
6958 				REASON_SET(&reason, PFRES_IPOPTIONS);
6959 				log = 1;
6960 				goto done;
6961 			}
6962 			if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6963 			    &reason, pd.af)) {
6964 				DPFPRINTF(PF_DEBUG_MISC,
6965 				    ("pf: IPv6 short rthdr\n"));
6966 				action = PF_DROP;
6967 				REASON_SET(&reason, PFRES_SHORT);
6968 				log = 1;
6969 				goto done;
6970 			}
6971 			if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6972 				DPFPRINTF(PF_DEBUG_MISC,
6973 				    ("pf: IPv6 rthdr0\n"));
6974 				action = PF_DROP;
6975 				REASON_SET(&reason, PFRES_IPOPTIONS);
6976 				log = 1;
6977 				goto done;
6978 			}
6979 			/* FALLTHROUGH */
6980 		}
6981 		case IPPROTO_AH:
6982 		case IPPROTO_HOPOPTS:
6983 		case IPPROTO_DSTOPTS: {
6984 			/* get next header and header length */
6985 			struct ip6_ext	opt6;
6986 
6987 			if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6988 			    NULL, &reason, pd.af)) {
6989 				DPFPRINTF(PF_DEBUG_MISC,
6990 				    ("pf: IPv6 short opt\n"));
6991 				action = PF_DROP;
6992 				log = 1;
6993 				goto done;
6994 			}
6995 			if (pd.proto == IPPROTO_AH)
6996 				off += (opt6.ip6e_len + 2) * 4;
6997 			else
6998 				off += (opt6.ip6e_len + 1) * 8;
6999 			pd.proto = opt6.ip6e_nxt;
7000 			/* goto the next header */
7001 			break;
7002 		}
7003 		default:
7004 			terminal++;
7005 			break;
7006 		}
7007 	} while (!terminal);
7008 
7009 	/* if there's no routing header, use unmodified mbuf for checksumming */
7010 	if (!n)
7011 		n = m;
7012 
7013 	switch (pd.proto) {
7014 
7015 	case IPPROTO_TCP: {
7016 		struct tcphdr	th;
7017 
7018 		pd.hdr.tcp = &th;
7019 		if (!pf_pull_hdr(m, off, &th, sizeof(th),
7020 		    &action, &reason, AF_INET6)) {
7021 			log = action != PF_PASS;
7022 			goto done;
7023 		}
7024 		pd.p_len = pd.tot_len - off - (th.th_off << 2);
7025 		action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
7026 		if (action == PF_DROP)
7027 			goto done;
7028 		action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
7029 					   &reason);
7030 		if (action == PF_PASS) {
7031 			r = s->rule.ptr;
7032 			a = s->anchor.ptr;
7033 			log = s->log;
7034 		} else if (s == NULL) {
7035 			action = pf_test_rule(&r, &s, dir, kif,
7036 					      m, off, h, &pd, &a,
7037 					      &ruleset, NULL, inp);
7038 		}
7039 		break;
7040 	}
7041 
7042 	case IPPROTO_UDP: {
7043 		struct udphdr	uh;
7044 
7045 		pd.hdr.udp = &uh;
7046 		if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
7047 		    &action, &reason, AF_INET6)) {
7048 			log = action != PF_PASS;
7049 			goto done;
7050 		}
7051 		if (uh.uh_dport == 0 ||
7052 		    ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
7053 		    ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
7054 			action = PF_DROP;
7055 			REASON_SET(&reason, PFRES_SHORT);
7056 			goto done;
7057 		}
7058 		action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
7059 		if (action == PF_PASS) {
7060 			r = s->rule.ptr;
7061 			a = s->anchor.ptr;
7062 			log = s->log;
7063 		} else if (s == NULL) {
7064 			action = pf_test_rule(&r, &s, dir, kif,
7065 					      m, off, h, &pd, &a,
7066 					      &ruleset, NULL, inp);
7067 		}
7068 		break;
7069 	}
7070 
7071 	case IPPROTO_ICMPV6: {
7072 		struct icmp6_hdr	ih;
7073 
7074 		pd.hdr.icmp6 = &ih;
7075 		if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
7076 		    &action, &reason, AF_INET6)) {
7077 			log = action != PF_PASS;
7078 			goto done;
7079 		}
7080 		action = pf_test_state_icmp(&s, dir, kif,
7081 					    m, off, h, &pd, &reason);
7082 		if (action == PF_PASS) {
7083 			r = s->rule.ptr;
7084 			a = s->anchor.ptr;
7085 			log = s->log;
7086 		} else if (s == NULL) {
7087 			action = pf_test_rule(&r, &s, dir, kif,
7088 					      m, off, h, &pd, &a,
7089 					      &ruleset, NULL, inp);
7090 		}
7091 		break;
7092 	}
7093 
7094 	default:
7095 		action = pf_test_state_other(&s, dir, kif, m, &pd);
7096 		if (action == PF_PASS) {
7097 			r = s->rule.ptr;
7098 			a = s->anchor.ptr;
7099 			log = s->log;
7100 		} else if (s == NULL) {
7101 			action = pf_test_rule(&r, &s, dir, kif, m, off, h,
7102 					      &pd, &a, &ruleset, NULL, inp);
7103 		}
7104 		break;
7105 	}
7106 
7107 done:
7108 	if (n != m) {
7109 		m_freem(n);
7110 		n = NULL;
7111 	}
7112 
7113 	/* handle dangerous IPv6 extension headers. */
7114 	if (action == PF_PASS && rh_cnt &&
7115 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
7116 		action = PF_DROP;
7117 		REASON_SET(&reason, PFRES_IPOPTIONS);
7118 		log = 1;
7119 		DPFPRINTF(PF_DEBUG_MISC,
7120 		    ("pf: dropping packet with dangerous v6 headers\n"));
7121 	}
7122 
7123 	if ((s && s->tag) || r->rtableid)
7124 		pf_tag_packet(m, s ? s->tag : 0, r->rtableid);
7125 
7126 #if 0
7127 	if (dir == PF_IN && s && s->key[PF_SK_STACK])
7128 		m->m_pkthdr.pf.statekey = s->key[PF_SK_STACK];
7129 #endif
7130 
7131 #ifdef ALTQ
7132 	/*
7133 	 * Generate a hash code and qid request for ALTQ.  A qid of 0
7134 	 * is allowed and will cause altq to select the default queue.
7135 	 */
7136 	if (action == PF_PASS) {
7137 		m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE;
7138 		if (pd.tos & IPTOS_LOWDELAY)
7139 			m->m_pkthdr.pf.qid = r->pqid;
7140 		else
7141 			m->m_pkthdr.pf.qid = r->qid;
7142 		m->m_pkthdr.pf.ecn_af = AF_INET6;
7143 		m->m_pkthdr.pf.hdr = h;
7144 		if (s) {
7145 			/* for fairq */
7146 			m->m_pkthdr.pf.state_hash = s->hash;
7147 			m->m_pkthdr.pf.flags |= PF_TAG_STATE_HASHED;
7148 		}
7149 	}
7150 #endif /* ALTQ */
7151 
7152 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
7153 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
7154 	    (s->nat_rule.ptr->action == PF_RDR ||
7155 	    s->nat_rule.ptr->action == PF_BINAT) &&
7156 	    IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
7157 	{
7158 		m->m_pkthdr.pf.flags |= PF_TAG_TRANSLATE_LOCALHOST;
7159 	}
7160 
7161 	if (dir == PF_IN && action == PF_PASS && r->divert.port) {
7162 		struct pf_divert *divert;
7163 
7164 		if ((divert = pf_get_divert(m))) {
7165 			m->m_pkthdr.pf.flags |= PF_TAG_DIVERTED;
7166 			divert->port = r->divert.port;
7167 			divert->addr.ipv6 = r->divert.addr.v6;
7168 		}
7169 	}
7170 
7171 	if (log) {
7172 		struct pf_rule *lr;
7173 
7174 		if (s != NULL && s->nat_rule.ptr != NULL &&
7175 		    s->nat_rule.ptr->log & PF_LOG_ALL)
7176 			lr = s->nat_rule.ptr;
7177 		else
7178 			lr = r;
7179 		PFLOG_PACKET(kif, h, m, AF_INET6, dir, reason, lr, a, ruleset,
7180 		    &pd);
7181 	}
7182 
7183 	kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
7184 	kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
7185 
7186 	if (action == PF_PASS || r->action == PF_DROP) {
7187 		dirndx = (dir == PF_OUT);
7188 		r->packets[dirndx]++;
7189 		r->bytes[dirndx] += pd.tot_len;
7190 		if (a != NULL) {
7191 			a->packets[dirndx]++;
7192 			a->bytes[dirndx] += pd.tot_len;
7193 		}
7194 		if (s != NULL) {
7195 			if (s->nat_rule.ptr != NULL) {
7196 				s->nat_rule.ptr->packets[dirndx]++;
7197 				s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
7198 			}
7199 			if (s->src_node != NULL) {
7200 				s->src_node->packets[dirndx]++;
7201 				s->src_node->bytes[dirndx] += pd.tot_len;
7202 			}
7203 			if (s->nat_src_node != NULL) {
7204 				s->nat_src_node->packets[dirndx]++;
7205 				s->nat_src_node->bytes[dirndx] += pd.tot_len;
7206 			}
7207 			dirndx = (dir == s->direction) ? 0 : 1;
7208 			s->packets[dirndx]++;
7209 			s->bytes[dirndx] += pd.tot_len;
7210 		}
7211 		tr = r;
7212 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
7213 		if (nr != NULL && r == &pf_default_rule)
7214 			tr = nr;
7215 		if (tr->src.addr.type == PF_ADDR_TABLE)
7216 			pfr_update_stats(tr->src.addr.p.tbl,
7217 			    (s == NULL) ? pd.src :
7218 			    &s->key[(s->direction == PF_IN)]->addr[0],
7219 			    pd.af, pd.tot_len, dir == PF_OUT,
7220 			    r->action == PF_PASS, tr->src.neg);
7221 		if (tr->dst.addr.type == PF_ADDR_TABLE)
7222 			pfr_update_stats(tr->dst.addr.p.tbl,
7223 			    (s == NULL) ? pd.dst :
7224 			    &s->key[(s->direction == PF_IN)]->addr[1],
7225 			    pd.af, pd.tot_len, dir == PF_OUT,
7226 			    r->action == PF_PASS, tr->dst.neg);
7227 	}
7228 
7229 
7230 	if (action == PF_SYNPROXY_DROP) {
7231 		m_freem(*m0);
7232 		*m0 = NULL;
7233 		action = PF_PASS;
7234 	} else if (r->rt)
7235 		/* pf_route6 can free the mbuf causing *m0 to become NULL */
7236 		pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
7237 
7238 	return (action);
7239 }
7240 #endif /* INET6 */
7241 
7242 int
pf_check_congestion(struct ifqueue * ifq)7243 pf_check_congestion(struct ifqueue *ifq)
7244 {
7245 		return (0);
7246 }
7247