1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2001 Daniel Hartmeier
5  * Copyright (c) 2002 - 2008 Henning Brauer
6  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  *
13  *    - Redistributions of source code must retain the above copyright
14  *      notice, this list of conditions and the following disclaimer.
15  *    - Redistributions in binary form must reproduce the above
16  *      copyright notice, this list of conditions and the following
17  *      disclaimer in the documentation and/or other materials provided
18  *      with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
28  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
30  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31  * POSSIBILITY OF SUCH DAMAGE.
32  *
33  * Effort sponsored in part by the Defense Advanced Research Projects
34  * Agency (DARPA) and Air Force Research Laboratory, Air Force
35  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
36  *
37  *	$OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
38  */
39 
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD: stable/12/sys/netpfil/pf/pf.c 373284 2023-12-05 18:31:13Z markj $");
42 
43 #include "opt_bpf.h"
44 #include "opt_inet.h"
45 #include "opt_inet6.h"
46 #include "opt_pf.h"
47 #include "opt_sctp.h"
48 
49 #include <sys/param.h>
50 #include <sys/bus.h>
51 #include <sys/endian.h>
52 #include <sys/gsb_crc32.h>
53 #include <sys/hash.h>
54 #include <sys/interrupt.h>
55 #include <sys/kernel.h>
56 #include <sys/kthread.h>
57 #include <sys/limits.h>
58 #include <sys/mbuf.h>
59 #include <sys/md5.h>
60 #include <sys/random.h>
61 #include <sys/refcount.h>
62 #include <sys/sdt.h>
63 #include <sys/socket.h>
64 #include <sys/sysctl.h>
65 #include <sys/taskqueue.h>
66 #include <sys/ucred.h>
67 
68 #include <net/if.h>
69 #include <net/if_var.h>
70 #include <net/if_types.h>
71 #include <net/if_vlan_var.h>
72 #include <net/route.h>
73 #include <net/radix_mpath.h>
74 #include <net/vnet.h>
75 
76 #include <net/pfil.h>
77 #include <net/pfvar.h>
78 #include <net/if_pflog.h>
79 #include <net/if_pfsync.h>
80 
81 #include <netinet/in_pcb.h>
82 #include <netinet/in_var.h>
83 #include <netinet/in_fib.h>
84 #include <netinet/ip.h>
85 #include <netinet/ip_fw.h>
86 #include <netinet/ip_icmp.h>
87 #include <netinet/icmp_var.h>
88 #include <netinet/ip_var.h>
89 #include <netinet/tcp.h>
90 #include <netinet/tcp_fsm.h>
91 #include <netinet/tcp_seq.h>
92 #include <netinet/tcp_timer.h>
93 #include <netinet/tcp_var.h>
94 #include <netinet/udp.h>
95 #include <netinet/udp_var.h>
96 
97 #include <netpfil/ipfw/ip_fw_private.h> /* XXX: only for DIR_IN/DIR_OUT */
98 
99 #ifdef INET6
100 #include <netinet/ip6.h>
101 #include <netinet/icmp6.h>
102 #include <netinet6/nd6.h>
103 #include <netinet6/ip6_var.h>
104 #include <netinet6/in6_pcb.h>
105 #include <netinet6/in6_fib.h>
106 #include <netinet6/scope6_var.h>
107 #endif /* INET6 */
108 
109 #if defined(SCTP) || defined(SCTP_SUPPORT)
110 #include <netinet/sctp_crc32.h>
111 #endif
112 
113 #include <machine/in_cksum.h>
114 #include <security/mac/mac_framework.h>
115 
116 #define	DPFPRINTF(n, x)	if (V_pf_status.debug >= (n)) printf x
117 
118 SDT_PROVIDER_DEFINE(pf);
119 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *",
120     "struct pf_kstate *");
121 SDT_PROBE_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *",
122     "struct pf_kstate *");
123 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *",
124     "struct pf_state_key_cmp *", "int", "struct pf_pdesc *",
125     "struct pf_kstate *");
126 
127 /*
128  * Global variables
129  */
130 
131 /* state tables */
132 VNET_DEFINE(struct pf_altqqueue,	 pf_altqs[4]);
133 VNET_DEFINE(struct pf_kpalist,		 pf_pabuf);
134 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_active);
135 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_active);
136 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_inactive);
137 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_inactive);
138 VNET_DEFINE(struct pf_kstatus,		 pf_status);
139 
140 VNET_DEFINE(u_int32_t,			 ticket_altqs_active);
141 VNET_DEFINE(u_int32_t,			 ticket_altqs_inactive);
142 VNET_DEFINE(int,			 altqs_inactive_open);
143 VNET_DEFINE(u_int32_t,			 ticket_pabuf);
144 
145 VNET_DEFINE(MD5_CTX,			 pf_tcp_secret_ctx);
146 #define	V_pf_tcp_secret_ctx		 VNET(pf_tcp_secret_ctx)
147 VNET_DEFINE(u_char,			 pf_tcp_secret[16]);
148 #define	V_pf_tcp_secret			 VNET(pf_tcp_secret)
149 VNET_DEFINE(int,			 pf_tcp_secret_init);
150 #define	V_pf_tcp_secret_init		 VNET(pf_tcp_secret_init)
151 VNET_DEFINE(int,			 pf_tcp_iss_off);
152 #define	V_pf_tcp_iss_off		 VNET(pf_tcp_iss_off)
153 VNET_DECLARE(int,			 pf_vnet_active);
154 #define	V_pf_vnet_active		 VNET(pf_vnet_active)
155 
156 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx);
157 #define V_pf_purge_idx	VNET(pf_purge_idx)
158 
159 #ifdef PF_WANT_32_TO_64_COUNTER
160 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter);
161 #define	V_pf_counter_periodic_iter	VNET(pf_counter_periodic_iter)
162 
163 VNET_DEFINE(struct allrulelist_head, pf_allrulelist);
164 VNET_DEFINE(size_t, pf_allrulecount);
165 VNET_DEFINE(struct pf_krule *, pf_rulemarker);
166 #endif
167 
168 /*
169  * Queue for pf_intr() sends.
170  */
171 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
172 struct pf_send_entry {
173 	STAILQ_ENTRY(pf_send_entry)	pfse_next;
174 	struct mbuf			*pfse_m;
175 	enum {
176 		PFSE_IP,
177 		PFSE_IP6,
178 		PFSE_ICMP,
179 		PFSE_ICMP6,
180 	}				pfse_type;
181 	struct {
182 		int		type;
183 		int		code;
184 		int		mtu;
185 	} icmpopts;
186 };
187 
188 STAILQ_HEAD(pf_send_head, pf_send_entry);
189 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue);
190 #define	V_pf_sendqueue	VNET(pf_sendqueue)
191 
192 static struct mtx_padalign pf_sendqueue_mtx;
193 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
194 #define	PF_SENDQ_LOCK()		mtx_lock(&pf_sendqueue_mtx)
195 #define	PF_SENDQ_UNLOCK()	mtx_unlock(&pf_sendqueue_mtx)
196 
197 /*
198  * Queue for pf_overload_task() tasks.
199  */
200 struct pf_overload_entry {
201 	SLIST_ENTRY(pf_overload_entry)	next;
202 	struct pf_addr  		addr;
203 	sa_family_t			af;
204 	uint8_t				dir;
205 	struct pf_krule  		*rule;
206 };
207 
208 SLIST_HEAD(pf_overload_head, pf_overload_entry);
209 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue);
210 #define V_pf_overloadqueue	VNET(pf_overloadqueue)
211 VNET_DEFINE_STATIC(struct task, pf_overloadtask);
212 #define	V_pf_overloadtask	VNET(pf_overloadtask)
213 
214 static struct mtx_padalign pf_overloadqueue_mtx;
215 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
216     "pf overload/flush queue", MTX_DEF);
217 #define	PF_OVERLOADQ_LOCK()	mtx_lock(&pf_overloadqueue_mtx)
218 #define	PF_OVERLOADQ_UNLOCK()	mtx_unlock(&pf_overloadqueue_mtx)
219 
220 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules);
221 struct mtx_padalign pf_unlnkdrules_mtx;
222 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
223     MTX_DEF);
224 
225 struct mtx_padalign pf_table_stats_lock;
226 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats",
227     MTX_DEF);
228 
229 VNET_DEFINE_STATIC(uma_zone_t,	pf_sources_z);
230 #define	V_pf_sources_z	VNET(pf_sources_z)
231 uma_zone_t		pf_mtag_z;
232 VNET_DEFINE(uma_zone_t,	 pf_state_z);
233 VNET_DEFINE(uma_zone_t,	 pf_state_key_z);
234 
235 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
236 #define	PFID_CPUBITS	8
237 #define	PFID_CPUSHIFT	(sizeof(uint64_t) * NBBY - PFID_CPUBITS)
238 #define	PFID_CPUMASK	((uint64_t)((1 << PFID_CPUBITS) - 1) <<	PFID_CPUSHIFT)
239 #define	PFID_MAXID	(~PFID_CPUMASK)
240 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU);
241 
242 static void		 pf_src_tree_remove_state(struct pf_kstate *);
243 static void		 pf_init_threshold(struct pf_threshold *, u_int32_t,
244 			    u_int32_t);
245 static void		 pf_add_threshold(struct pf_threshold *);
246 static int		 pf_check_threshold(struct pf_threshold *);
247 
248 static void		 pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
249 			    u_int16_t *, u_int16_t *, struct pf_addr *,
250 			    u_int16_t, u_int8_t, sa_family_t);
251 static int		 pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
252 			    struct tcphdr *, struct pf_state_peer *);
253 static void		 pf_change_icmp(struct pf_addr *, u_int16_t *,
254 			    struct pf_addr *, struct pf_addr *, u_int16_t,
255 			    u_int16_t *, u_int16_t *, u_int16_t *,
256 			    u_int16_t *, u_int8_t, sa_family_t);
257 static void		 pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
258 			    sa_family_t, struct pf_krule *);
259 static void		 pf_detach_state(struct pf_kstate *);
260 static int		 pf_state_key_attach(struct pf_state_key *,
261 			    struct pf_state_key *, struct pf_kstate *);
262 static void		 pf_state_key_detach(struct pf_kstate *, int);
263 static int		 pf_state_key_ctor(void *, int, void *, int);
264 static u_int32_t	 pf_tcp_iss(struct pf_pdesc *);
265 void			 pf_rule_to_actions(struct pf_krule *,
266 			    struct pf_rule_actions *);
267 static int		 pf_test_rule(struct pf_krule **, struct pf_kstate **,
268 			    int, struct pfi_kkif *, struct mbuf *, int,
269 			    struct pf_pdesc *, struct pf_krule **,
270 			    struct pf_kruleset **, struct inpcb *);
271 static int		 pf_create_state(struct pf_krule *, struct pf_krule *,
272 			    struct pf_krule *, struct pf_pdesc *,
273 			    struct pf_ksrc_node *, struct pf_state_key *,
274 			    struct pf_state_key *, struct mbuf *, int,
275 			    u_int16_t, u_int16_t, int *, struct pfi_kkif *,
276 			    struct pf_kstate **, int, u_int16_t, u_int16_t,
277 			    int);
278 static int		 pf_test_fragment(struct pf_krule **, int,
279 			    struct pfi_kkif *, struct mbuf *, void *,
280 			    struct pf_pdesc *, struct pf_krule **,
281 			    struct pf_kruleset **);
282 static int		 pf_tcp_track_full(struct pf_kstate **,
283 			    struct pfi_kkif *, struct mbuf *, int,
284 			    struct pf_pdesc *, u_short *, int *);
285 static int		 pf_tcp_track_sloppy(struct pf_kstate **,
286 			    struct pf_pdesc *, u_short *);
287 static int		 pf_test_state_tcp(struct pf_kstate **, int,
288 			    struct pfi_kkif *, struct mbuf *, int,
289 			    void *, struct pf_pdesc *, u_short *);
290 static int		 pf_test_state_udp(struct pf_kstate **, int,
291 			    struct pfi_kkif *, struct mbuf *, int,
292 			    void *, struct pf_pdesc *);
293 static int		 pf_test_state_icmp(struct pf_kstate **, int,
294 			    struct pfi_kkif *, struct mbuf *, int,
295 			    void *, struct pf_pdesc *, u_short *);
296 static int		 pf_test_state_other(struct pf_kstate **, int,
297 			    struct pfi_kkif *, struct mbuf *, struct pf_pdesc *);
298 static u_int16_t	 pf_calc_mss(struct pf_addr *, sa_family_t,
299 				int, u_int16_t);
300 static int		 pf_check_proto_cksum(struct mbuf *, int, int,
301 			    u_int8_t, sa_family_t);
302 static void		 pf_print_state_parts(struct pf_kstate *,
303 			    struct pf_state_key *, struct pf_state_key *);
304 static int		 pf_addr_wrap_neq(struct pf_addr_wrap *,
305 			    struct pf_addr_wrap *);
306 static void		 pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t,
307 			    bool, u_int8_t);
308 static struct pf_kstate	*pf_find_state(struct pfi_kkif *,
309 			    struct pf_state_key_cmp *, u_int);
310 static int		 pf_src_connlimit(struct pf_kstate **);
311 static void		 pf_overload_task(void *v, int pending);
312 static int		 pf_insert_src_node(struct pf_ksrc_node **,
313 			    struct pf_krule *, struct pf_addr *, sa_family_t);
314 static u_int		 pf_purge_expired_states(u_int, int);
315 static void		 pf_purge_unlinked_rules(void);
316 static int		 pf_mtag_uminit(void *, int, int);
317 static void		 pf_mtag_free(struct m_tag *);
318 static void		 pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *,
319 			    int, struct pf_state_key *);
320 #ifdef INET
321 static void		 pf_route(struct mbuf **, struct pf_krule *, int,
322 			    struct ifnet *, struct pf_kstate *,
323 			    struct pf_pdesc *, struct inpcb *);
324 #endif /* INET */
325 #ifdef INET6
326 static void		 pf_change_a6(struct pf_addr *, u_int16_t *,
327 			    struct pf_addr *, u_int8_t);
328 static void		 pf_route6(struct mbuf **, struct pf_krule *, int,
329 			    struct ifnet *, struct pf_kstate *,
330 			    struct pf_pdesc *, struct inpcb *);
331 #endif /* INET6 */
332 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t);
333 
334 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
335 
336 extern int pf_end_threads;
337 extern struct proc *pf_purge_proc;
338 
339 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
340 
341 #define	PACKET_UNDO_NAT(_m, _pd, _off, _s, _dir)		\
342 	do {								\
343 		struct pf_state_key *nk;				\
344 		if ((_dir) == PF_OUT)					\
345 			nk = (_s)->key[PF_SK_STACK];			\
346 		else							\
347 			nk = (_s)->key[PF_SK_WIRE];			\
348 		pf_packet_rework_nat(_m, _pd, _off, nk);		\
349 	} while (0)
350 
351 #define	PACKET_LOOPED(pd)	((pd)->pf_mtag &&			\
352 				 (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
353 
354 #define	STATE_LOOKUP(i, k, d, s, pd)					\
355 	do {								\
356 		(s) = pf_find_state((i), (k), (d));			\
357 		SDT_PROBE5(pf, ip, state, lookup, i, k, d, pd, (s));	\
358 		if ((s) == NULL)					\
359 			return (PF_DROP);				\
360 		if (PACKET_LOOPED(pd))					\
361 			return (PF_PASS);				\
362 	} while (0)
363 
364 #define	BOUND_IFACE(r, k) \
365 	((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
366 
367 #define	STATE_INC_COUNTERS(s)						\
368 	do {								\
369 		counter_u64_add(s->rule.ptr->states_cur, 1);		\
370 		counter_u64_add(s->rule.ptr->states_tot, 1);		\
371 		if (s->anchor.ptr != NULL) {				\
372 			counter_u64_add(s->anchor.ptr->states_cur, 1);	\
373 			counter_u64_add(s->anchor.ptr->states_tot, 1);	\
374 		}							\
375 		if (s->nat_rule.ptr != NULL) {				\
376 			counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
377 			counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
378 		}							\
379 	} while (0)
380 
381 #define	STATE_DEC_COUNTERS(s)						\
382 	do {								\
383 		if (s->nat_rule.ptr != NULL)				\
384 			counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
385 		if (s->anchor.ptr != NULL)				\
386 			counter_u64_add(s->anchor.ptr->states_cur, -1);	\
387 		counter_u64_add(s->rule.ptr->states_cur, -1);		\
388 	} while (0)
389 
390 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
391 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
392 VNET_DEFINE(struct pf_idhash *, pf_idhash);
393 VNET_DEFINE(struct pf_srchash *, pf_srchash);
394 
395 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)");
396 
397 u_long	pf_hashmask;
398 u_long	pf_srchashmask;
399 static u_long	pf_hashsize;
400 static u_long	pf_srchashsize;
401 u_long	pf_ioctl_maxcount = 65535;
402 
403 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
404     &pf_hashsize, 0, "Size of pf(4) states hashtable");
405 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
406     &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
407 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN,
408     &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call");
409 
410 VNET_DEFINE(void *, pf_swi_cookie);
411 VNET_DEFINE(struct intr_event *, pf_swi_ie);
412 
413 VNET_DEFINE(uint32_t, pf_hashseed);
414 #define	V_pf_hashseed	VNET(pf_hashseed)
415 
416 #ifdef __LP64__
417 static int
pf_bcmp_state_key(struct pf_state_key * k1_orig,struct pf_state_key_cmp * k2_orig)418 pf_bcmp_state_key(struct pf_state_key *k1_orig, struct pf_state_key_cmp *k2_orig)
419 {
420 	unsigned long *k1 = (unsigned long *)k1_orig;
421 	unsigned long *k2 = (unsigned long *)k2_orig;
422 
423 	if (k1[0] != k2[0])
424 		return (1);
425 
426 	if (k1[1] != k2[1])
427 		return (1);
428 
429 	if (k1[2] != k2[2])
430 		return (1);
431 
432 	if (k1[3] != k2[3])
433 		return (1);
434 
435 	if (k1[4] != k2[4])
436 		return (1);
437 
438 	return (0);
439 }
440 _Static_assert(sizeof(struct pf_state_key_cmp) == 40, "bad size of pf_state_key_cmp");
441 #else
442 static inline int
pf_bcmp_state_key(struct pf_state_key * k1_orig,struct pf_state_key_cmp * k2_orig)443 pf_bcmp_state_key(struct pf_state_key *k1_orig, struct pf_state_key_cmp *k2_orig)
444 {
445 
446 	return (bcmp(k1_orig, k2_orig, sizeof(struct pf_state_key_cmp)));
447 }
448 #endif
449 
450 int
pf_addr_cmp(struct pf_addr * a,struct pf_addr * b,sa_family_t af)451 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
452 {
453 
454 	switch (af) {
455 #ifdef INET
456 	case AF_INET:
457 		if (a->addr32[0] > b->addr32[0])
458 			return (1);
459 		if (a->addr32[0] < b->addr32[0])
460 			return (-1);
461 		break;
462 #endif /* INET */
463 #ifdef INET6
464 	case AF_INET6:
465 		if (a->addr32[3] > b->addr32[3])
466 			return (1);
467 		if (a->addr32[3] < b->addr32[3])
468 			return (-1);
469 		if (a->addr32[2] > b->addr32[2])
470 			return (1);
471 		if (a->addr32[2] < b->addr32[2])
472 			return (-1);
473 		if (a->addr32[1] > b->addr32[1])
474 			return (1);
475 		if (a->addr32[1] < b->addr32[1])
476 			return (-1);
477 		if (a->addr32[0] > b->addr32[0])
478 			return (1);
479 		if (a->addr32[0] < b->addr32[0])
480 			return (-1);
481 		break;
482 #endif /* INET6 */
483 	default:
484 		panic("%s: unknown address family %u", __func__, af);
485 	}
486 	return (0);
487 }
488 
489 static void
pf_packet_rework_nat(struct mbuf * m,struct pf_pdesc * pd,int off,struct pf_state_key * nk)490 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off,
491 	struct pf_state_key *nk)
492 {
493 
494 	switch (pd->proto) {
495 	case IPPROTO_TCP: {
496 		struct tcphdr *th = &pd->hdr.tcp;
497 
498 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af))
499 			pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum,
500 			    &th->th_sum, &nk->addr[pd->sidx],
501 			    nk->port[pd->sidx], 0, pd->af);
502 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af))
503 			pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum,
504 			    &th->th_sum, &nk->addr[pd->didx],
505 			    nk->port[pd->didx], 0, pd->af);
506 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
507 		break;
508 	}
509 	case IPPROTO_UDP: {
510 		struct udphdr *uh = &pd->hdr.udp;
511 
512 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af))
513 			pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
514 			    &uh->uh_sum, &nk->addr[pd->sidx],
515 			    nk->port[pd->sidx], 1, pd->af);
516 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af))
517 			pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
518 			    &uh->uh_sum, &nk->addr[pd->didx],
519 			    nk->port[pd->didx], 1, pd->af);
520 		m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
521 		break;
522 	}
523 	case IPPROTO_ICMP: {
524 		struct icmp *ih = &pd->hdr.icmp;
525 
526 		if (nk->port[pd->sidx] != ih->icmp_id) {
527 			pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
528 			    ih->icmp_cksum, ih->icmp_id,
529 			    nk->port[pd->sidx], 0);
530 			ih->icmp_id = nk->port[pd->sidx];
531 			pd->sport = &ih->icmp_id;
532 
533 			m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih);
534 		}
535 		/* FALLTHROUGH */
536 	}
537 	default:
538 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) {
539 			switch (pd->af) {
540 			case AF_INET:
541 				pf_change_a(&pd->src->v4.s_addr,
542 				    pd->ip_sum, nk->addr[pd->sidx].v4.s_addr,
543 				    0);
544 				break;
545 			case AF_INET6:
546 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
547 				break;
548 			}
549 		}
550 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) {
551 			switch (pd->af) {
552 			case AF_INET:
553 				pf_change_a(&pd->dst->v4.s_addr,
554 				    pd->ip_sum, nk->addr[pd->didx].v4.s_addr,
555 				    0);
556 				break;
557 			case AF_INET6:
558 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
559 				break;
560 			}
561 		}
562 		break;
563 	}
564 }
565 
566 static __inline uint32_t
pf_hashkey(struct pf_state_key * sk)567 pf_hashkey(struct pf_state_key *sk)
568 {
569 	uint32_t h;
570 
571 	h = murmur3_32_hash32((uint32_t *)sk,
572 	    sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
573 	    V_pf_hashseed);
574 
575 	return (h & pf_hashmask);
576 }
577 
578 static __inline uint32_t
pf_hashsrc(struct pf_addr * addr,sa_family_t af)579 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
580 {
581 	uint32_t h;
582 
583 	switch (af) {
584 	case AF_INET:
585 		h = murmur3_32_hash32((uint32_t *)&addr->v4,
586 		    sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
587 		break;
588 	case AF_INET6:
589 		h = murmur3_32_hash32((uint32_t *)&addr->v6,
590 		    sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
591 		break;
592 	default:
593 		panic("%s: unknown address family %u", __func__, af);
594 	}
595 
596 	return (h & pf_srchashmask);
597 }
598 
599 #ifdef ALTQ
600 static int
pf_state_hash(struct pf_kstate * s)601 pf_state_hash(struct pf_kstate *s)
602 {
603 	u_int32_t hv = (intptr_t)s / sizeof(*s);
604 
605 	hv ^= crc32(&s->src, sizeof(s->src));
606 	hv ^= crc32(&s->dst, sizeof(s->dst));
607 	if (hv == 0)
608 		hv = 1;
609 	return (hv);
610 }
611 #endif
612 
613 static __inline void
pf_set_protostate(struct pf_kstate * s,int which,u_int8_t newstate)614 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate)
615 {
616 	if (which == PF_PEER_DST || which == PF_PEER_BOTH)
617 		s->dst.state = newstate;
618 	if (which == PF_PEER_DST)
619 		return;
620 	if (s->src.state == newstate)
621 		return;
622 	if (s->creatorid == V_pf_status.hostid &&
623 	    s->key[PF_SK_STACK] != NULL &&
624 	    s->key[PF_SK_STACK]->proto == IPPROTO_TCP &&
625 	    !(TCPS_HAVEESTABLISHED(s->src.state) ||
626 	    s->src.state == TCPS_CLOSED) &&
627 	    (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED))
628 		atomic_add_32(&V_pf_status.states_halfopen, -1);
629 
630 	s->src.state = newstate;
631 }
632 
633 #ifdef INET6
634 void
pf_addrcpy(struct pf_addr * dst,struct pf_addr * src,sa_family_t af)635 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
636 {
637 	switch (af) {
638 #ifdef INET
639 	case AF_INET:
640 		dst->addr32[0] = src->addr32[0];
641 		break;
642 #endif /* INET */
643 	case AF_INET6:
644 		dst->addr32[0] = src->addr32[0];
645 		dst->addr32[1] = src->addr32[1];
646 		dst->addr32[2] = src->addr32[2];
647 		dst->addr32[3] = src->addr32[3];
648 		break;
649 	}
650 }
651 #endif /* INET6 */
652 
653 static void
pf_init_threshold(struct pf_threshold * threshold,u_int32_t limit,u_int32_t seconds)654 pf_init_threshold(struct pf_threshold *threshold,
655     u_int32_t limit, u_int32_t seconds)
656 {
657 	threshold->limit = limit * PF_THRESHOLD_MULT;
658 	threshold->seconds = seconds;
659 	threshold->count = 0;
660 	threshold->last = time_uptime;
661 }
662 
663 static void
pf_add_threshold(struct pf_threshold * threshold)664 pf_add_threshold(struct pf_threshold *threshold)
665 {
666 	u_int32_t t = time_uptime, diff = t - threshold->last;
667 
668 	if (diff >= threshold->seconds)
669 		threshold->count = 0;
670 	else
671 		threshold->count -= threshold->count * diff /
672 		    threshold->seconds;
673 	threshold->count += PF_THRESHOLD_MULT;
674 	threshold->last = t;
675 }
676 
677 static int
pf_check_threshold(struct pf_threshold * threshold)678 pf_check_threshold(struct pf_threshold *threshold)
679 {
680 	return (threshold->count > threshold->limit);
681 }
682 
683 static int
pf_src_connlimit(struct pf_kstate ** state)684 pf_src_connlimit(struct pf_kstate **state)
685 {
686 	struct pf_overload_entry *pfoe;
687 	int bad = 0;
688 
689 	PF_STATE_LOCK_ASSERT(*state);
690 
691 	(*state)->src_node->conn++;
692 	(*state)->src.tcp_est = 1;
693 	pf_add_threshold(&(*state)->src_node->conn_rate);
694 
695 	if ((*state)->rule.ptr->max_src_conn &&
696 	    (*state)->rule.ptr->max_src_conn <
697 	    (*state)->src_node->conn) {
698 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
699 		bad++;
700 	}
701 
702 	if ((*state)->rule.ptr->max_src_conn_rate.limit &&
703 	    pf_check_threshold(&(*state)->src_node->conn_rate)) {
704 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
705 		bad++;
706 	}
707 
708 	if (!bad)
709 		return (0);
710 
711 	/* Kill this state. */
712 	(*state)->timeout = PFTM_PURGE;
713 	pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED);
714 
715 	if ((*state)->rule.ptr->overload_tbl == NULL)
716 		return (1);
717 
718 	/* Schedule overloading and flushing task. */
719 	pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
720 	if (pfoe == NULL)
721 		return (1);	/* too bad :( */
722 
723 	bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
724 	pfoe->af = (*state)->key[PF_SK_WIRE]->af;
725 	pfoe->rule = (*state)->rule.ptr;
726 	pfoe->dir = (*state)->direction;
727 	PF_OVERLOADQ_LOCK();
728 	SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
729 	PF_OVERLOADQ_UNLOCK();
730 	taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
731 
732 	return (1);
733 }
734 
735 static void
pf_overload_task(void * v,int pending)736 pf_overload_task(void *v, int pending)
737 {
738 	struct pf_overload_head queue;
739 	struct pfr_addr p;
740 	struct pf_overload_entry *pfoe, *pfoe1;
741 	uint32_t killed = 0;
742 
743 	CURVNET_SET((struct vnet *)v);
744 
745 	PF_OVERLOADQ_LOCK();
746 	queue = V_pf_overloadqueue;
747 	SLIST_INIT(&V_pf_overloadqueue);
748 	PF_OVERLOADQ_UNLOCK();
749 
750 	bzero(&p, sizeof(p));
751 	SLIST_FOREACH(pfoe, &queue, next) {
752 		counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
753 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
754 			printf("%s: blocking address ", __func__);
755 			pf_print_host(&pfoe->addr, 0, pfoe->af);
756 			printf("\n");
757 		}
758 
759 		p.pfra_af = pfoe->af;
760 		switch (pfoe->af) {
761 #ifdef INET
762 		case AF_INET:
763 			p.pfra_net = 32;
764 			p.pfra_ip4addr = pfoe->addr.v4;
765 			break;
766 #endif
767 #ifdef INET6
768 		case AF_INET6:
769 			p.pfra_net = 128;
770 			p.pfra_ip6addr = pfoe->addr.v6;
771 			break;
772 #endif
773 		}
774 
775 		PF_RULES_WLOCK();
776 		pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
777 		PF_RULES_WUNLOCK();
778 	}
779 
780 	/*
781 	 * Remove those entries, that don't need flushing.
782 	 */
783 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
784 		if (pfoe->rule->flush == 0) {
785 			SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
786 			free(pfoe, M_PFTEMP);
787 		} else
788 			counter_u64_add(
789 			    V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
790 
791 	/* If nothing to flush, return. */
792 	if (SLIST_EMPTY(&queue)) {
793 		CURVNET_RESTORE();
794 		return;
795 	}
796 
797 	for (int i = 0; i <= pf_hashmask; i++) {
798 		struct pf_idhash *ih = &V_pf_idhash[i];
799 		struct pf_state_key *sk;
800 		struct pf_kstate *s;
801 
802 		PF_HASHROW_LOCK(ih);
803 		LIST_FOREACH(s, &ih->states, entry) {
804 		    sk = s->key[PF_SK_WIRE];
805 		    SLIST_FOREACH(pfoe, &queue, next)
806 			if (sk->af == pfoe->af &&
807 			    ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
808 			    pfoe->rule == s->rule.ptr) &&
809 			    ((pfoe->dir == PF_OUT &&
810 			    PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
811 			    (pfoe->dir == PF_IN &&
812 			    PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
813 				s->timeout = PFTM_PURGE;
814 				pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED);
815 				killed++;
816 			}
817 		}
818 		PF_HASHROW_UNLOCK(ih);
819 	}
820 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
821 		free(pfoe, M_PFTEMP);
822 	if (V_pf_status.debug >= PF_DEBUG_MISC)
823 		printf("%s: %u states killed", __func__, killed);
824 
825 	CURVNET_RESTORE();
826 }
827 
828 /*
829  * Can return locked on failure, so that we can consistently
830  * allocate and insert a new one.
831  */
832 struct pf_ksrc_node *
pf_find_src_node(struct pf_addr * src,struct pf_krule * rule,sa_family_t af,int returnlocked)833 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af,
834 	int returnlocked)
835 {
836 	struct pf_srchash *sh;
837 	struct pf_ksrc_node *n;
838 
839 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
840 
841 	sh = &V_pf_srchash[pf_hashsrc(src, af)];
842 	PF_HASHROW_LOCK(sh);
843 	LIST_FOREACH(n, &sh->nodes, entry)
844 		if (n->rule.ptr == rule && n->af == af &&
845 		    ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
846 		    (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
847 			break;
848 	if (n != NULL) {
849 		n->states++;
850 		PF_HASHROW_UNLOCK(sh);
851 	} else if (returnlocked == 0)
852 		PF_HASHROW_UNLOCK(sh);
853 
854 	return (n);
855 }
856 
857 static void
pf_free_src_node(struct pf_ksrc_node * sn)858 pf_free_src_node(struct pf_ksrc_node *sn)
859 {
860 
861 	for (int i = 0; i < 2; i++) {
862 		counter_u64_free(sn->bytes[i]);
863 		counter_u64_free(sn->packets[i]);
864 	}
865 	uma_zfree(V_pf_sources_z, sn);
866 }
867 
868 static int
pf_insert_src_node(struct pf_ksrc_node ** sn,struct pf_krule * rule,struct pf_addr * src,sa_family_t af)869 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule,
870     struct pf_addr *src, sa_family_t af)
871 {
872 
873 	KASSERT((rule->rule_flag & PFRULE_SRCTRACK ||
874 	    rule->rpool.opts & PF_POOL_STICKYADDR),
875 	    ("%s for non-tracking rule %p", __func__, rule));
876 
877 	if (*sn == NULL)
878 		*sn = pf_find_src_node(src, rule, af, 1);
879 
880 	if (*sn == NULL) {
881 		struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
882 
883 		PF_HASHROW_ASSERT(sh);
884 
885 		if (!rule->max_src_nodes ||
886 		    counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
887 			(*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
888 		else
889 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
890 			    1);
891 		if ((*sn) == NULL) {
892 			PF_HASHROW_UNLOCK(sh);
893 			return (-1);
894 		}
895 
896 		for (int i = 0; i < 2; i++) {
897 			(*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT);
898 			(*sn)->packets[i] = counter_u64_alloc(M_NOWAIT);
899 
900 			if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) {
901 				pf_free_src_node(*sn);
902 				PF_HASHROW_UNLOCK(sh);
903 				return (-1);
904 			}
905 		}
906 
907 		pf_init_threshold(&(*sn)->conn_rate,
908 		    rule->max_src_conn_rate.limit,
909 		    rule->max_src_conn_rate.seconds);
910 
911 		(*sn)->af = af;
912 		(*sn)->rule.ptr = rule;
913 		PF_ACPY(&(*sn)->addr, src, af);
914 		LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
915 		(*sn)->creation = time_uptime;
916 		(*sn)->ruletype = rule->action;
917 		(*sn)->states = 1;
918 		if ((*sn)->rule.ptr != NULL)
919 			counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
920 		PF_HASHROW_UNLOCK(sh);
921 		counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
922 	} else {
923 		if (rule->max_src_states &&
924 		    (*sn)->states >= rule->max_src_states) {
925 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
926 			    1);
927 			return (-1);
928 		}
929 	}
930 	return (0);
931 }
932 
933 void
pf_unlink_src_node(struct pf_ksrc_node * src)934 pf_unlink_src_node(struct pf_ksrc_node *src)
935 {
936 
937 	PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]);
938 	LIST_REMOVE(src, entry);
939 	if (src->rule.ptr)
940 		counter_u64_add(src->rule.ptr->src_nodes, -1);
941 }
942 
943 u_int
pf_free_src_nodes(struct pf_ksrc_node_list * head)944 pf_free_src_nodes(struct pf_ksrc_node_list *head)
945 {
946 	struct pf_ksrc_node *sn, *tmp;
947 	u_int count = 0;
948 
949 	LIST_FOREACH_SAFE(sn, head, entry, tmp) {
950 		pf_free_src_node(sn);
951 		count++;
952 	}
953 
954 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
955 
956 	return (count);
957 }
958 
959 void
pf_mtag_initialize(void)960 pf_mtag_initialize(void)
961 {
962 
963 	pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
964 	    sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
965 	    UMA_ALIGN_PTR, 0);
966 }
967 
968 /* Per-vnet data storage structures initialization. */
969 void
pf_initialize(void)970 pf_initialize(void)
971 {
972 	struct pf_keyhash	*kh;
973 	struct pf_idhash	*ih;
974 	struct pf_srchash	*sh;
975 	u_int i;
976 
977 	if (pf_hashsize == 0 || !powerof2(pf_hashsize))
978 		pf_hashsize = PF_HASHSIZ;
979 	if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
980 		pf_srchashsize = PF_SRCHASHSIZ;
981 
982 	V_pf_hashseed = arc4random();
983 
984 	/* States and state keys storage. */
985 	V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate),
986 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
987 	V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
988 	uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
989 	uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
990 
991 	V_pf_state_key_z = uma_zcreate("pf state keys",
992 	    sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
993 	    UMA_ALIGN_PTR, 0);
994 
995 	V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash),
996 	    M_PFHASH, M_NOWAIT | M_ZERO);
997 	V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash),
998 	    M_PFHASH, M_NOWAIT | M_ZERO);
999 	if (V_pf_keyhash == NULL || V_pf_idhash == NULL) {
1000 		printf("pf: Unable to allocate memory for "
1001 		    "state_hashsize %lu.\n", pf_hashsize);
1002 
1003 		free(V_pf_keyhash, M_PFHASH);
1004 		free(V_pf_idhash, M_PFHASH);
1005 
1006 		pf_hashsize = PF_HASHSIZ;
1007 		V_pf_keyhash = mallocarray(pf_hashsize,
1008 		    sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO);
1009 		V_pf_idhash = mallocarray(pf_hashsize,
1010 		    sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO);
1011 	}
1012 
1013 	pf_hashmask = pf_hashsize - 1;
1014 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
1015 	    i++, kh++, ih++) {
1016 		mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
1017 		mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
1018 	}
1019 
1020 	/* Source nodes. */
1021 	V_pf_sources_z = uma_zcreate("pf source nodes",
1022 	    sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
1023 	    0);
1024 	V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
1025 	uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
1026 	uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
1027 
1028 	V_pf_srchash = mallocarray(pf_srchashsize,
1029 	    sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO);
1030 	if (V_pf_srchash == NULL) {
1031 		printf("pf: Unable to allocate memory for "
1032 		    "source_hashsize %lu.\n", pf_srchashsize);
1033 
1034 		pf_srchashsize = PF_SRCHASHSIZ;
1035 		V_pf_srchash = mallocarray(pf_srchashsize,
1036 		    sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO);
1037 	}
1038 
1039 	pf_srchashmask = pf_srchashsize - 1;
1040 	for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
1041 		mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
1042 
1043 	/* ALTQ */
1044 	TAILQ_INIT(&V_pf_altqs[0]);
1045 	TAILQ_INIT(&V_pf_altqs[1]);
1046 	TAILQ_INIT(&V_pf_altqs[2]);
1047 	TAILQ_INIT(&V_pf_altqs[3]);
1048 	TAILQ_INIT(&V_pf_pabuf);
1049 	V_pf_altqs_active = &V_pf_altqs[0];
1050 	V_pf_altq_ifs_active = &V_pf_altqs[1];
1051 	V_pf_altqs_inactive = &V_pf_altqs[2];
1052 	V_pf_altq_ifs_inactive = &V_pf_altqs[3];
1053 
1054 	/* Send & overload+flush queues. */
1055 	STAILQ_INIT(&V_pf_sendqueue);
1056 	SLIST_INIT(&V_pf_overloadqueue);
1057 	TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
1058 
1059 	/* Unlinked, but may be referenced rules. */
1060 	TAILQ_INIT(&V_pf_unlinked_rules);
1061 }
1062 
1063 void
pf_mtag_cleanup(void)1064 pf_mtag_cleanup(void)
1065 {
1066 
1067 	uma_zdestroy(pf_mtag_z);
1068 }
1069 
1070 void
pf_cleanup(void)1071 pf_cleanup(void)
1072 {
1073 	struct pf_keyhash	*kh;
1074 	struct pf_idhash	*ih;
1075 	struct pf_srchash	*sh;
1076 	struct pf_send_entry	*pfse, *next;
1077 	u_int i;
1078 
1079 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
1080 	    i++, kh++, ih++) {
1081 		KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
1082 		    __func__));
1083 		KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
1084 		    __func__));
1085 		mtx_destroy(&kh->lock);
1086 		mtx_destroy(&ih->lock);
1087 	}
1088 	free(V_pf_keyhash, M_PFHASH);
1089 	free(V_pf_idhash, M_PFHASH);
1090 
1091 	for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1092 		KASSERT(LIST_EMPTY(&sh->nodes),
1093 		    ("%s: source node hash not empty", __func__));
1094 		mtx_destroy(&sh->lock);
1095 	}
1096 	free(V_pf_srchash, M_PFHASH);
1097 
1098 	STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
1099 		m_freem(pfse->pfse_m);
1100 		free(pfse, M_PFTEMP);
1101 	}
1102 
1103 	uma_zdestroy(V_pf_sources_z);
1104 	uma_zdestroy(V_pf_state_z);
1105 	uma_zdestroy(V_pf_state_key_z);
1106 }
1107 
1108 static int
pf_mtag_uminit(void * mem,int size,int how)1109 pf_mtag_uminit(void *mem, int size, int how)
1110 {
1111 	struct m_tag *t;
1112 
1113 	t = (struct m_tag *)mem;
1114 	t->m_tag_cookie = MTAG_ABI_COMPAT;
1115 	t->m_tag_id = PACKET_TAG_PF;
1116 	t->m_tag_len = sizeof(struct pf_mtag);
1117 	t->m_tag_free = pf_mtag_free;
1118 
1119 	return (0);
1120 }
1121 
1122 static void
pf_mtag_free(struct m_tag * t)1123 pf_mtag_free(struct m_tag *t)
1124 {
1125 
1126 	uma_zfree(pf_mtag_z, t);
1127 }
1128 
1129 struct pf_mtag *
pf_get_mtag(struct mbuf * m)1130 pf_get_mtag(struct mbuf *m)
1131 {
1132 	struct m_tag *mtag;
1133 
1134 	if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
1135 		return ((struct pf_mtag *)(mtag + 1));
1136 
1137 	mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
1138 	if (mtag == NULL)
1139 		return (NULL);
1140 	bzero(mtag + 1, sizeof(struct pf_mtag));
1141 	m_tag_prepend(m, mtag);
1142 
1143 	return ((struct pf_mtag *)(mtag + 1));
1144 }
1145 
1146 static int
pf_state_key_attach(struct pf_state_key * skw,struct pf_state_key * sks,struct pf_kstate * s)1147 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
1148     struct pf_kstate *s)
1149 {
1150 	struct pf_keyhash	*khs, *khw, *kh;
1151 	struct pf_state_key	*sk, *cur;
1152 	struct pf_kstate	*si, *olds = NULL;
1153 	int idx;
1154 
1155 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1156 	KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
1157 	KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
1158 
1159 	/*
1160 	 * We need to lock hash slots of both keys. To avoid deadlock
1161 	 * we always lock the slot with lower address first. Unlock order
1162 	 * isn't important.
1163 	 *
1164 	 * We also need to lock ID hash slot before dropping key
1165 	 * locks. On success we return with ID hash slot locked.
1166 	 */
1167 
1168 	if (skw == sks) {
1169 		khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
1170 		PF_HASHROW_LOCK(khs);
1171 	} else {
1172 		khs = &V_pf_keyhash[pf_hashkey(sks)];
1173 		khw = &V_pf_keyhash[pf_hashkey(skw)];
1174 		if (khs == khw) {
1175 			PF_HASHROW_LOCK(khs);
1176 		} else if (khs < khw) {
1177 			PF_HASHROW_LOCK(khs);
1178 			PF_HASHROW_LOCK(khw);
1179 		} else {
1180 			PF_HASHROW_LOCK(khw);
1181 			PF_HASHROW_LOCK(khs);
1182 		}
1183 	}
1184 
1185 #define	KEYS_UNLOCK()	do {			\
1186 	if (khs != khw) {			\
1187 		PF_HASHROW_UNLOCK(khs);		\
1188 		PF_HASHROW_UNLOCK(khw);		\
1189 	} else					\
1190 		PF_HASHROW_UNLOCK(khs);		\
1191 } while (0)
1192 
1193 	/*
1194 	 * First run: start with wire key.
1195 	 */
1196 	sk = skw;
1197 	kh = khw;
1198 	idx = PF_SK_WIRE;
1199 
1200 	MPASS(s->lock == NULL);
1201 	s->lock = &V_pf_idhash[PF_IDHASH(s)].lock;
1202 
1203 keyattach:
1204 	LIST_FOREACH(cur, &kh->keys, entry)
1205 		if (pf_bcmp_state_key(cur, (struct pf_state_key_cmp *)sk) == 0)
1206 			break;
1207 
1208 	if (cur != NULL) {
1209 		/* Key exists. Check for same kif, if none, add to key. */
1210 		TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
1211 			struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
1212 
1213 			PF_HASHROW_LOCK(ih);
1214 			if (si->kif == s->kif &&
1215 			    si->direction == s->direction) {
1216 				if (sk->proto == IPPROTO_TCP &&
1217 				    si->src.state >= TCPS_FIN_WAIT_2 &&
1218 				    si->dst.state >= TCPS_FIN_WAIT_2) {
1219 					/*
1220 					 * New state matches an old >FIN_WAIT_2
1221 					 * state. We can't drop key hash locks,
1222 					 * thus we can't unlink it properly.
1223 					 *
1224 					 * As a workaround we drop it into
1225 					 * TCPS_CLOSED state, schedule purge
1226 					 * ASAP and push it into the very end
1227 					 * of the slot TAILQ, so that it won't
1228 					 * conflict with our new state.
1229 					 */
1230 					pf_set_protostate(si, PF_PEER_BOTH,
1231 					    TCPS_CLOSED);
1232 					si->timeout = PFTM_PURGE;
1233 					olds = si;
1234 				} else {
1235 					if (V_pf_status.debug >= PF_DEBUG_MISC) {
1236 						printf("pf: %s key attach "
1237 						    "failed on %s: ",
1238 						    (idx == PF_SK_WIRE) ?
1239 						    "wire" : "stack",
1240 						    s->kif->pfik_name);
1241 						pf_print_state_parts(s,
1242 						    (idx == PF_SK_WIRE) ?
1243 						    sk : NULL,
1244 						    (idx == PF_SK_STACK) ?
1245 						    sk : NULL);
1246 						printf(", existing: ");
1247 						pf_print_state_parts(si,
1248 						    (idx == PF_SK_WIRE) ?
1249 						    sk : NULL,
1250 						    (idx == PF_SK_STACK) ?
1251 						    sk : NULL);
1252 						printf("\n");
1253 					}
1254 					PF_HASHROW_UNLOCK(ih);
1255 					KEYS_UNLOCK();
1256 					uma_zfree(V_pf_state_key_z, sk);
1257 					if (idx == PF_SK_STACK)
1258 						pf_detach_state(s);
1259 					return (EEXIST); /* collision! */
1260 				}
1261 			}
1262 			PF_HASHROW_UNLOCK(ih);
1263 		}
1264 		uma_zfree(V_pf_state_key_z, sk);
1265 		s->key[idx] = cur;
1266 	} else {
1267 		LIST_INSERT_HEAD(&kh->keys, sk, entry);
1268 		s->key[idx] = sk;
1269 	}
1270 
1271 stateattach:
1272 	/* List is sorted, if-bound states before floating. */
1273 	if (s->kif == V_pfi_all)
1274 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1275 	else
1276 		TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1277 
1278 	if (olds) {
1279 		TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1280 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1281 		    key_list[idx]);
1282 		olds = NULL;
1283 	}
1284 
1285 	/*
1286 	 * Attach done. See how should we (or should not?)
1287 	 * attach a second key.
1288 	 */
1289 	if (sks == skw) {
1290 		s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1291 		idx = PF_SK_STACK;
1292 		sks = NULL;
1293 		goto stateattach;
1294 	} else if (sks != NULL) {
1295 		/*
1296 		 * Continue attaching with stack key.
1297 		 */
1298 		sk = sks;
1299 		kh = khs;
1300 		idx = PF_SK_STACK;
1301 		sks = NULL;
1302 		goto keyattach;
1303 	}
1304 
1305 	PF_STATE_LOCK(s);
1306 	KEYS_UNLOCK();
1307 
1308 	KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1309 	    ("%s failure", __func__));
1310 
1311 	return (0);
1312 #undef	KEYS_UNLOCK
1313 }
1314 
1315 static void
pf_detach_state(struct pf_kstate * s)1316 pf_detach_state(struct pf_kstate *s)
1317 {
1318 	struct pf_state_key *sks = s->key[PF_SK_STACK];
1319 	struct pf_keyhash *kh;
1320 
1321 	if (sks != NULL) {
1322 		kh = &V_pf_keyhash[pf_hashkey(sks)];
1323 		PF_HASHROW_LOCK(kh);
1324 		if (s->key[PF_SK_STACK] != NULL)
1325 			pf_state_key_detach(s, PF_SK_STACK);
1326 		/*
1327 		 * If both point to same key, then we are done.
1328 		 */
1329 		if (sks == s->key[PF_SK_WIRE]) {
1330 			pf_state_key_detach(s, PF_SK_WIRE);
1331 			PF_HASHROW_UNLOCK(kh);
1332 			return;
1333 		}
1334 		PF_HASHROW_UNLOCK(kh);
1335 	}
1336 
1337 	if (s->key[PF_SK_WIRE] != NULL) {
1338 		kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1339 		PF_HASHROW_LOCK(kh);
1340 		if (s->key[PF_SK_WIRE] != NULL)
1341 			pf_state_key_detach(s, PF_SK_WIRE);
1342 		PF_HASHROW_UNLOCK(kh);
1343 	}
1344 }
1345 
1346 static void
pf_state_key_detach(struct pf_kstate * s,int idx)1347 pf_state_key_detach(struct pf_kstate *s, int idx)
1348 {
1349 	struct pf_state_key *sk = s->key[idx];
1350 #ifdef INVARIANTS
1351 	struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1352 
1353 	PF_HASHROW_ASSERT(kh);
1354 #endif
1355 	TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1356 	s->key[idx] = NULL;
1357 
1358 	if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1359 		LIST_REMOVE(sk, entry);
1360 		uma_zfree(V_pf_state_key_z, sk);
1361 	}
1362 }
1363 
1364 static int
pf_state_key_ctor(void * mem,int size,void * arg,int flags)1365 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1366 {
1367 	struct pf_state_key *sk = mem;
1368 
1369 	bzero(sk, sizeof(struct pf_state_key_cmp));
1370 	TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1371 	TAILQ_INIT(&sk->states[PF_SK_STACK]);
1372 
1373 	return (0);
1374 }
1375 
1376 struct pf_state_key *
pf_state_key_setup(struct pf_pdesc * pd,struct pf_addr * saddr,struct pf_addr * daddr,u_int16_t sport,u_int16_t dport)1377 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1378 	struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1379 {
1380 	struct pf_state_key *sk;
1381 
1382 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1383 	if (sk == NULL)
1384 		return (NULL);
1385 
1386 	PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1387 	PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1388 	sk->port[pd->sidx] = sport;
1389 	sk->port[pd->didx] = dport;
1390 	sk->proto = pd->proto;
1391 	sk->af = pd->af;
1392 
1393 	return (sk);
1394 }
1395 
1396 struct pf_state_key *
pf_state_key_clone(struct pf_state_key * orig)1397 pf_state_key_clone(struct pf_state_key *orig)
1398 {
1399 	struct pf_state_key *sk;
1400 
1401 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1402 	if (sk == NULL)
1403 		return (NULL);
1404 
1405 	bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1406 
1407 	return (sk);
1408 }
1409 
1410 int
pf_state_insert(struct pfi_kkif * kif,struct pfi_kkif * orig_kif,struct pf_state_key * skw,struct pf_state_key * sks,struct pf_kstate * s)1411 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif,
1412     struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s)
1413 {
1414 	struct pf_idhash *ih;
1415 	struct pf_kstate *cur;
1416 	int error;
1417 
1418 	KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1419 	    ("%s: sks not pristine", __func__));
1420 	KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1421 	    ("%s: skw not pristine", __func__));
1422 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1423 
1424 	s->kif = kif;
1425 	s->orig_kif = orig_kif;
1426 
1427 	if (s->id == 0 && s->creatorid == 0) {
1428 		/* XXX: should be atomic, but probability of collision low */
1429 		if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1430 			V_pf_stateid[curcpu] = 1;
1431 		s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1432 		s->id = htobe64(s->id);
1433 		s->creatorid = V_pf_status.hostid;
1434 	}
1435 
1436 	/* Returns with ID locked on success. */
1437 	if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1438 		return (error);
1439 
1440 	ih = &V_pf_idhash[PF_IDHASH(s)];
1441 	PF_HASHROW_ASSERT(ih);
1442 	LIST_FOREACH(cur, &ih->states, entry)
1443 		if (cur->id == s->id && cur->creatorid == s->creatorid)
1444 			break;
1445 
1446 	if (cur != NULL) {
1447 		PF_HASHROW_UNLOCK(ih);
1448 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1449 			printf("pf: state ID collision: "
1450 			    "id: %016llx creatorid: %08x\n",
1451 			    (unsigned long long)be64toh(s->id),
1452 			    ntohl(s->creatorid));
1453 		}
1454 		pf_detach_state(s);
1455 		return (EEXIST);
1456 	}
1457 	LIST_INSERT_HEAD(&ih->states, s, entry);
1458 	/* One for keys, one for ID hash. */
1459 	refcount_init(&s->refs, 2);
1460 
1461 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1462 	if (V_pfsync_insert_state_ptr != NULL)
1463 		V_pfsync_insert_state_ptr(s);
1464 
1465 	/* Returns locked. */
1466 	return (0);
1467 }
1468 
1469 /*
1470  * Find state by ID: returns with locked row on success.
1471  */
1472 struct pf_kstate *
pf_find_state_byid(uint64_t id,uint32_t creatorid)1473 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1474 {
1475 	struct pf_idhash *ih;
1476 	struct pf_kstate *s;
1477 
1478 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1479 
1480 	ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1481 
1482 	PF_HASHROW_LOCK(ih);
1483 	LIST_FOREACH(s, &ih->states, entry)
1484 		if (s->id == id && s->creatorid == creatorid)
1485 			break;
1486 
1487 	if (s == NULL)
1488 		PF_HASHROW_UNLOCK(ih);
1489 
1490 	return (s);
1491 }
1492 
1493 /*
1494  * Find state by key.
1495  * Returns with ID hash slot locked on success.
1496  */
1497 static struct pf_kstate *
pf_find_state(struct pfi_kkif * kif,struct pf_state_key_cmp * key,u_int dir)1498 pf_find_state(struct pfi_kkif *kif, struct pf_state_key_cmp *key, u_int dir)
1499 {
1500 	struct pf_keyhash	*kh;
1501 	struct pf_state_key	*sk;
1502 	struct pf_kstate	*s;
1503 	int idx;
1504 
1505 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1506 
1507 	kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1508 
1509 	PF_HASHROW_LOCK(kh);
1510 	LIST_FOREACH(sk, &kh->keys, entry)
1511 		if (pf_bcmp_state_key(sk, key) == 0)
1512 			break;
1513 	if (sk == NULL) {
1514 		PF_HASHROW_UNLOCK(kh);
1515 		return (NULL);
1516 	}
1517 
1518 	idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1519 
1520 	/* List is sorted, if-bound states before floating ones. */
1521 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1522 		if (s->kif == V_pfi_all || s->kif == kif) {
1523 			PF_STATE_LOCK(s);
1524 			PF_HASHROW_UNLOCK(kh);
1525 			if (__predict_false(s->timeout >= PFTM_MAX)) {
1526 				/*
1527 				 * State is either being processed by
1528 				 * pf_unlink_state() in an other thread, or
1529 				 * is scheduled for immediate expiry.
1530 				 */
1531 				PF_STATE_UNLOCK(s);
1532 				return (NULL);
1533 			}
1534 			return (s);
1535 		}
1536 	PF_HASHROW_UNLOCK(kh);
1537 
1538 	return (NULL);
1539 }
1540 
1541 struct pf_kstate *
pf_find_state_all(struct pf_state_key_cmp * key,u_int dir,int * more)1542 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1543 {
1544 	struct pf_keyhash	*kh;
1545 	struct pf_state_key	*sk;
1546 	struct pf_kstate	*s, *ret = NULL;
1547 	int			 idx, inout = 0;
1548 
1549 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1550 
1551 	kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1552 
1553 	PF_HASHROW_LOCK(kh);
1554 	LIST_FOREACH(sk, &kh->keys, entry)
1555 		if (pf_bcmp_state_key(sk, key) == 0)
1556 			break;
1557 	if (sk == NULL) {
1558 		PF_HASHROW_UNLOCK(kh);
1559 		return (NULL);
1560 	}
1561 	switch (dir) {
1562 	case PF_IN:
1563 		idx = PF_SK_WIRE;
1564 		break;
1565 	case PF_OUT:
1566 		idx = PF_SK_STACK;
1567 		break;
1568 	case PF_INOUT:
1569 		idx = PF_SK_WIRE;
1570 		inout = 1;
1571 		break;
1572 	default:
1573 		panic("%s: dir %u", __func__, dir);
1574 	}
1575 second_run:
1576 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1577 		if (more == NULL) {
1578 			PF_HASHROW_UNLOCK(kh);
1579 			return (s);
1580 		}
1581 
1582 		if (ret)
1583 			(*more)++;
1584 		else
1585 			ret = s;
1586 	}
1587 	if (inout == 1) {
1588 		inout = 0;
1589 		idx = PF_SK_STACK;
1590 		goto second_run;
1591 	}
1592 	PF_HASHROW_UNLOCK(kh);
1593 
1594 	return (ret);
1595 }
1596 
1597 bool
pf_find_state_all_exists(struct pf_state_key_cmp * key,u_int dir)1598 pf_find_state_all_exists(struct pf_state_key_cmp *key, u_int dir)
1599 {
1600 	struct pf_kstate *s;
1601 
1602 	s = pf_find_state_all(key, dir, NULL);
1603 	return (s != NULL);
1604 }
1605 
1606 /* END state table stuff */
1607 
1608 static void
pf_send(struct pf_send_entry * pfse)1609 pf_send(struct pf_send_entry *pfse)
1610 {
1611 
1612 	PF_SENDQ_LOCK();
1613 	STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1614 	PF_SENDQ_UNLOCK();
1615 	swi_sched(V_pf_swi_cookie, 0);
1616 }
1617 
1618 static bool
pf_isforlocal(struct mbuf * m,int af)1619 pf_isforlocal(struct mbuf *m, int af)
1620 {
1621 	switch (af) {
1622 #ifdef INET
1623 	case AF_INET: {
1624 		struct rm_priotracker in_ifa_tracker;
1625 		struct ip *ip;
1626 		struct in_ifaddr *ia = NULL;
1627 
1628 		ip = mtod(m, struct ip *);
1629 		IN_IFADDR_RLOCK(&in_ifa_tracker);
1630 		LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) {
1631 			if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr) {
1632 				IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1633 				return (true);
1634 			}
1635 		}
1636 		IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1637 		break;
1638 	}
1639 #endif
1640 #ifdef INET6
1641 	case AF_INET6: {
1642 		struct ip6_hdr *ip6;
1643 		struct in6_ifaddr *ia;
1644 		ip6 = mtod(m, struct ip6_hdr *);
1645 		ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1646 		if (ia == NULL)
1647 			return (false);
1648 		return (! (ia->ia6_flags & IN6_IFF_NOTREADY));
1649 	}
1650 #endif
1651 	default:
1652 		panic("Unsupported af %d", af);
1653 	}
1654 
1655 	return (false);
1656 }
1657 
1658 void
pf_intr(void * v)1659 pf_intr(void *v)
1660 {
1661 	struct pf_send_head queue;
1662 	struct pf_send_entry *pfse, *next;
1663 
1664 	CURVNET_SET((struct vnet *)v);
1665 
1666 	PF_SENDQ_LOCK();
1667 	queue = V_pf_sendqueue;
1668 	STAILQ_INIT(&V_pf_sendqueue);
1669 	PF_SENDQ_UNLOCK();
1670 
1671 	STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1672 		switch (pfse->pfse_type) {
1673 #ifdef INET
1674 		case PFSE_IP: {
1675 			if (pf_isforlocal(pfse->pfse_m, AF_INET)) {
1676 				pfse->pfse_m->m_flags |= M_SKIP_FIREWALL;
1677 				pfse->pfse_m->m_pkthdr.csum_flags |=
1678 				    CSUM_IP_VALID | CSUM_IP_CHECKED;
1679 				ip_input(pfse->pfse_m);
1680 			} else {
1681 				ip_output(pfse->pfse_m, NULL, NULL, 0, NULL,
1682 				    NULL);
1683 			}
1684 			break;
1685 		}
1686 		case PFSE_ICMP:
1687 			icmp_error(pfse->pfse_m, pfse->icmpopts.type,
1688 			    pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
1689 			break;
1690 #endif /* INET */
1691 #ifdef INET6
1692 		case PFSE_IP6:
1693 			if (pf_isforlocal(pfse->pfse_m, AF_INET6)) {
1694 				pfse->pfse_m->m_flags |= M_SKIP_FIREWALL;
1695 				ip6_input(pfse->pfse_m);
1696 			} else {
1697 				ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL,
1698 				    NULL, NULL);
1699 			}
1700 			break;
1701 		case PFSE_ICMP6:
1702 			icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
1703 			    pfse->icmpopts.code, pfse->icmpopts.mtu);
1704 			break;
1705 #endif /* INET6 */
1706 		default:
1707 			panic("%s: unknown type", __func__);
1708 		}
1709 		free(pfse, M_PFTEMP);
1710 	}
1711 	CURVNET_RESTORE();
1712 }
1713 
1714 #define	pf_purge_thread_period	(hz / 10)
1715 
1716 #ifdef PF_WANT_32_TO_64_COUNTER
1717 static void
pf_status_counter_u64_periodic(void)1718 pf_status_counter_u64_periodic(void)
1719 {
1720 
1721 	PF_RULES_RASSERT();
1722 
1723 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) {
1724 		return;
1725 	}
1726 
1727 	for (int i = 0; i < FCNT_MAX; i++) {
1728 		pf_counter_u64_periodic(&V_pf_status.fcounters[i]);
1729 	}
1730 }
1731 
1732 static void
pf_kif_counter_u64_periodic(void)1733 pf_kif_counter_u64_periodic(void)
1734 {
1735 	struct pfi_kkif *kif;
1736 	size_t r, run;
1737 
1738 	PF_RULES_RASSERT();
1739 
1740 	if (__predict_false(V_pf_allkifcount == 0)) {
1741 		return;
1742 	}
1743 
1744 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) {
1745 		return;
1746 	}
1747 
1748 	run = V_pf_allkifcount / 10;
1749 	if (run < 5)
1750 		run = 5;
1751 
1752 	for (r = 0; r < run; r++) {
1753 		kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist);
1754 		if (kif == NULL) {
1755 			LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist);
1756 			LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist);
1757 			break;
1758 		}
1759 
1760 		LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist);
1761 		LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist);
1762 
1763 		for (int i = 0; i < 2; i++) {
1764 			for (int j = 0; j < 2; j++) {
1765 				for (int k = 0; k < 2; k++) {
1766 					pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]);
1767 					pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]);
1768 				}
1769 			}
1770 		}
1771 	}
1772 }
1773 
1774 static void
pf_rule_counter_u64_periodic(void)1775 pf_rule_counter_u64_periodic(void)
1776 {
1777 	struct pf_krule *rule;
1778 	size_t r, run;
1779 
1780 	PF_RULES_RASSERT();
1781 
1782 	if (__predict_false(V_pf_allrulecount == 0)) {
1783 		return;
1784 	}
1785 
1786 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) {
1787 		return;
1788 	}
1789 
1790 	run = V_pf_allrulecount / 10;
1791 	if (run < 5)
1792 		run = 5;
1793 
1794 	for (r = 0; r < run; r++) {
1795 		rule = LIST_NEXT(V_pf_rulemarker, allrulelist);
1796 		if (rule == NULL) {
1797 			LIST_REMOVE(V_pf_rulemarker, allrulelist);
1798 			LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist);
1799 			break;
1800 		}
1801 
1802 		LIST_REMOVE(V_pf_rulemarker, allrulelist);
1803 		LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist);
1804 
1805 		pf_counter_u64_periodic(&rule->evaluations);
1806 		for (int i = 0; i < 2; i++) {
1807 			pf_counter_u64_periodic(&rule->packets[i]);
1808 			pf_counter_u64_periodic(&rule->bytes[i]);
1809 		}
1810 	}
1811 }
1812 
1813 static void
pf_counter_u64_periodic_main(void)1814 pf_counter_u64_periodic_main(void)
1815 {
1816 	PF_RULES_RLOCK_TRACKER;
1817 
1818 	V_pf_counter_periodic_iter++;
1819 
1820 	PF_RULES_RLOCK();
1821 	pf_counter_u64_critical_enter();
1822 	pf_status_counter_u64_periodic();
1823 	pf_kif_counter_u64_periodic();
1824 	pf_rule_counter_u64_periodic();
1825 	pf_counter_u64_critical_exit();
1826 	PF_RULES_RUNLOCK();
1827 }
1828 #else
1829 #define	pf_counter_u64_periodic_main()	do { } while (0)
1830 #endif
1831 
1832 void
pf_purge_thread(void * unused __unused)1833 pf_purge_thread(void *unused __unused)
1834 {
1835 	VNET_ITERATOR_DECL(vnet_iter);
1836 
1837 	sx_xlock(&pf_end_lock);
1838 	while (pf_end_threads == 0) {
1839 		sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period);
1840 
1841 		VNET_LIST_RLOCK();
1842 		VNET_FOREACH(vnet_iter) {
1843 			CURVNET_SET(vnet_iter);
1844 
1845 
1846 			/* Wait until V_pf_default_rule is initialized. */
1847 			if (V_pf_vnet_active == 0) {
1848 				CURVNET_RESTORE();
1849 				continue;
1850 			}
1851 
1852 			pf_counter_u64_periodic_main();
1853 
1854 			/*
1855 			 *  Process 1/interval fraction of the state
1856 			 * table every run.
1857 			 */
1858 			V_pf_purge_idx =
1859 			    pf_purge_expired_states(V_pf_purge_idx, pf_hashmask /
1860 			    (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1861 
1862 			/*
1863 			 * Purge other expired types every
1864 			 * PFTM_INTERVAL seconds.
1865 			 */
1866 			if (V_pf_purge_idx == 0) {
1867 				/*
1868 				 * Order is important:
1869 				 * - states and src nodes reference rules
1870 				 * - states and rules reference kifs
1871 				 */
1872 				pf_purge_expired_fragments();
1873 				pf_purge_expired_src_nodes();
1874 				pf_purge_unlinked_rules();
1875 				pfi_kkif_purge();
1876 			}
1877 			CURVNET_RESTORE();
1878 		}
1879 		VNET_LIST_RUNLOCK();
1880 	}
1881 
1882 	pf_end_threads++;
1883 	sx_xunlock(&pf_end_lock);
1884 	kproc_exit(0);
1885 }
1886 
1887 void
pf_unload_vnet_purge(void)1888 pf_unload_vnet_purge(void)
1889 {
1890 
1891 	/*
1892 	 * To cleanse up all kifs and rules we need
1893 	 * two runs: first one clears reference flags,
1894 	 * then pf_purge_expired_states() doesn't
1895 	 * raise them, and then second run frees.
1896 	 */
1897 	pf_purge_unlinked_rules();
1898 	pfi_kkif_purge();
1899 
1900 	/*
1901 	 * Now purge everything.
1902 	 */
1903 	pf_purge_expired_states(0, pf_hashmask);
1904 	pf_purge_fragments(UINT_MAX);
1905 	pf_purge_expired_src_nodes();
1906 
1907 	/*
1908 	 * Now all kifs & rules should be unreferenced,
1909 	 * thus should be successfully freed.
1910 	 */
1911 	pf_purge_unlinked_rules();
1912 	pfi_kkif_purge();
1913 }
1914 
1915 
1916 u_int32_t
pf_state_expires(const struct pf_kstate * state)1917 pf_state_expires(const struct pf_kstate *state)
1918 {
1919 	u_int32_t	timeout;
1920 	u_int32_t	start;
1921 	u_int32_t	end;
1922 	u_int32_t	states;
1923 
1924 	/* handle all PFTM_* > PFTM_MAX here */
1925 	if (state->timeout == PFTM_PURGE)
1926 		return (time_uptime);
1927 	KASSERT(state->timeout != PFTM_UNLINKED,
1928 	    ("pf_state_expires: timeout == PFTM_UNLINKED"));
1929 	KASSERT((state->timeout < PFTM_MAX),
1930 	    ("pf_state_expires: timeout > PFTM_MAX"));
1931 	timeout = state->rule.ptr->timeout[state->timeout];
1932 	if (!timeout)
1933 		timeout = V_pf_default_rule.timeout[state->timeout];
1934 	start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1935 	if (start && state->rule.ptr != &V_pf_default_rule) {
1936 		end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1937 		states = counter_u64_fetch(state->rule.ptr->states_cur);
1938 	} else {
1939 		start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1940 		end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1941 		states = V_pf_status.states;
1942 	}
1943 	if (end && states > start && start < end) {
1944 		if (states < end) {
1945 			timeout = (u_int64_t)timeout * (end - states) /
1946 			    (end - start);
1947 			return (state->expire + timeout);
1948 		}
1949 		else
1950 			return (time_uptime);
1951 	}
1952 	return (state->expire + timeout);
1953 }
1954 
1955 void
pf_purge_expired_src_nodes(void)1956 pf_purge_expired_src_nodes(void)
1957 {
1958 	struct pf_ksrc_node_list	 freelist;
1959 	struct pf_srchash	*sh;
1960 	struct pf_ksrc_node	*cur, *next;
1961 	int i;
1962 
1963 	LIST_INIT(&freelist);
1964 	for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1965 	    PF_HASHROW_LOCK(sh);
1966 	    LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1967 		if (cur->states == 0 && cur->expire <= time_uptime) {
1968 			pf_unlink_src_node(cur);
1969 			LIST_INSERT_HEAD(&freelist, cur, entry);
1970 		} else if (cur->rule.ptr != NULL)
1971 			cur->rule.ptr->rule_ref |= PFRULE_REFS;
1972 	    PF_HASHROW_UNLOCK(sh);
1973 	}
1974 
1975 	pf_free_src_nodes(&freelist);
1976 
1977 	V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1978 }
1979 
1980 static void
pf_src_tree_remove_state(struct pf_kstate * s)1981 pf_src_tree_remove_state(struct pf_kstate *s)
1982 {
1983 	struct pf_ksrc_node *sn;
1984 	struct pf_srchash *sh;
1985 	uint32_t timeout;
1986 
1987 	timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
1988 	    s->rule.ptr->timeout[PFTM_SRC_NODE] :
1989 	    V_pf_default_rule.timeout[PFTM_SRC_NODE];
1990 
1991 	if (s->src_node != NULL) {
1992 		sn = s->src_node;
1993 		sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1994 	    	PF_HASHROW_LOCK(sh);
1995 		if (s->src.tcp_est)
1996 			--sn->conn;
1997 		if (--sn->states == 0)
1998 			sn->expire = time_uptime + timeout;
1999 	    	PF_HASHROW_UNLOCK(sh);
2000 	}
2001 	if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
2002 		sn = s->nat_src_node;
2003 		sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
2004 	    	PF_HASHROW_LOCK(sh);
2005 		if (--sn->states == 0)
2006 			sn->expire = time_uptime + timeout;
2007 	    	PF_HASHROW_UNLOCK(sh);
2008 	}
2009 	s->src_node = s->nat_src_node = NULL;
2010 }
2011 
2012 /*
2013  * Unlink and potentilly free a state. Function may be
2014  * called with ID hash row locked, but always returns
2015  * unlocked, since it needs to go through key hash locking.
2016  */
2017 int
pf_unlink_state(struct pf_kstate * s,u_int flags)2018 pf_unlink_state(struct pf_kstate *s, u_int flags)
2019 {
2020 	struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
2021 
2022 	if ((flags & PF_ENTER_LOCKED) == 0)
2023 		PF_HASHROW_LOCK(ih);
2024 	else
2025 		PF_HASHROW_ASSERT(ih);
2026 
2027 	if (s->timeout == PFTM_UNLINKED) {
2028 		/*
2029 		 * State is being processed
2030 		 * by pf_unlink_state() in
2031 		 * an other thread.
2032 		 */
2033 		PF_HASHROW_UNLOCK(ih);
2034 		return (0);	/* XXXGL: undefined actually */
2035 	}
2036 
2037 	if (s->src.state == PF_TCPS_PROXY_DST) {
2038 		/* XXX wire key the right one? */
2039 		pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af,
2040 		    &s->key[PF_SK_WIRE]->addr[1],
2041 		    &s->key[PF_SK_WIRE]->addr[0],
2042 		    s->key[PF_SK_WIRE]->port[1],
2043 		    s->key[PF_SK_WIRE]->port[0],
2044 		    s->src.seqhi, s->src.seqlo + 1,
2045 		    TH_RST|TH_ACK, 0, 0, 0, 1, s->tag);
2046 	}
2047 
2048 	LIST_REMOVE(s, entry);
2049 	pf_src_tree_remove_state(s);
2050 
2051 	if (V_pfsync_delete_state_ptr != NULL)
2052 		V_pfsync_delete_state_ptr(s);
2053 
2054 	STATE_DEC_COUNTERS(s);
2055 
2056 	s->timeout = PFTM_UNLINKED;
2057 
2058 	/* Ensure we remove it from the list of halfopen states, if needed. */
2059 	if (s->key[PF_SK_STACK] != NULL &&
2060 	    s->key[PF_SK_STACK]->proto == IPPROTO_TCP)
2061 		pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED);
2062 
2063 	PF_HASHROW_UNLOCK(ih);
2064 
2065 	pf_detach_state(s);
2066 	/* pf_state_insert() initialises refs to 2 */
2067 	return (pf_release_staten(s, 2));
2068 }
2069 
2070 struct pf_kstate *
pf_alloc_state(int flags)2071 pf_alloc_state(int flags)
2072 {
2073 
2074 	return (uma_zalloc(V_pf_state_z, flags | M_ZERO));
2075 }
2076 
2077 void
pf_free_state(struct pf_kstate * cur)2078 pf_free_state(struct pf_kstate *cur)
2079 {
2080 
2081 	KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
2082 	KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
2083 	    cur->timeout));
2084 
2085 	pf_normalize_tcp_cleanup(cur);
2086 	uma_zfree(V_pf_state_z, cur);
2087 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
2088 }
2089 
2090 /*
2091  * Called only from pf_purge_thread(), thus serialized.
2092  */
2093 static u_int
pf_purge_expired_states(u_int i,int maxcheck)2094 pf_purge_expired_states(u_int i, int maxcheck)
2095 {
2096 	struct pf_idhash *ih;
2097 	struct pf_kstate *s;
2098 
2099 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2100 
2101 	/*
2102 	 * Go through hash and unlink states that expire now.
2103 	 */
2104 	while (maxcheck > 0) {
2105 
2106 		ih = &V_pf_idhash[i];
2107 
2108 		/* only take the lock if we expect to do work */
2109 		if (!LIST_EMPTY(&ih->states)) {
2110 relock:
2111 			PF_HASHROW_LOCK(ih);
2112 			LIST_FOREACH(s, &ih->states, entry) {
2113 				if (pf_state_expires(s) <= time_uptime) {
2114 					V_pf_status.states -=
2115 					    pf_unlink_state(s, PF_ENTER_LOCKED);
2116 					goto relock;
2117 				}
2118 				s->rule.ptr->rule_ref |= PFRULE_REFS;
2119 				if (s->nat_rule.ptr != NULL)
2120 					s->nat_rule.ptr->rule_ref |= PFRULE_REFS;
2121 				if (s->anchor.ptr != NULL)
2122 					s->anchor.ptr->rule_ref |= PFRULE_REFS;
2123 				s->kif->pfik_flags |= PFI_IFLAG_REFS;
2124 				if (s->rt_kif)
2125 					s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
2126 			}
2127 			PF_HASHROW_UNLOCK(ih);
2128 		}
2129 
2130 		/* Return when we hit end of hash. */
2131 		if (++i > pf_hashmask) {
2132 			V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2133 			return (0);
2134 		}
2135 
2136 		maxcheck--;
2137 	}
2138 
2139 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2140 
2141 	return (i);
2142 }
2143 
2144 static void
pf_purge_unlinked_rules(void)2145 pf_purge_unlinked_rules(void)
2146 {
2147 	struct pf_krulequeue tmpq;
2148 	struct pf_krule *r, *r1;
2149 
2150 	/*
2151 	 * If we have overloading task pending, then we'd
2152 	 * better skip purging this time. There is a tiny
2153 	 * probability that overloading task references
2154 	 * an already unlinked rule.
2155 	 */
2156 	PF_OVERLOADQ_LOCK();
2157 	if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
2158 		PF_OVERLOADQ_UNLOCK();
2159 		return;
2160 	}
2161 	PF_OVERLOADQ_UNLOCK();
2162 
2163 	/*
2164 	 * Do naive mark-and-sweep garbage collecting of old rules.
2165 	 * Reference flag is raised by pf_purge_expired_states()
2166 	 * and pf_purge_expired_src_nodes().
2167 	 *
2168 	 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
2169 	 * use a temporary queue.
2170 	 */
2171 	TAILQ_INIT(&tmpq);
2172 	PF_UNLNKDRULES_LOCK();
2173 	TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
2174 		if (!(r->rule_ref & PFRULE_REFS)) {
2175 			TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
2176 			TAILQ_INSERT_TAIL(&tmpq, r, entries);
2177 		} else
2178 			r->rule_ref &= ~PFRULE_REFS;
2179 	}
2180 	PF_UNLNKDRULES_UNLOCK();
2181 
2182 	if (!TAILQ_EMPTY(&tmpq)) {
2183 		PF_RULES_WLOCK();
2184 		TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
2185 			TAILQ_REMOVE(&tmpq, r, entries);
2186 			pf_free_rule(r);
2187 		}
2188 		PF_RULES_WUNLOCK();
2189 	}
2190 }
2191 
2192 void
pf_print_host(struct pf_addr * addr,u_int16_t p,sa_family_t af)2193 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
2194 {
2195 	switch (af) {
2196 #ifdef INET
2197 	case AF_INET: {
2198 		u_int32_t a = ntohl(addr->addr32[0]);
2199 		printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
2200 		    (a>>8)&255, a&255);
2201 		if (p) {
2202 			p = ntohs(p);
2203 			printf(":%u", p);
2204 		}
2205 		break;
2206 	}
2207 #endif /* INET */
2208 #ifdef INET6
2209 	case AF_INET6: {
2210 		u_int16_t b;
2211 		u_int8_t i, curstart, curend, maxstart, maxend;
2212 		curstart = curend = maxstart = maxend = 255;
2213 		for (i = 0; i < 8; i++) {
2214 			if (!addr->addr16[i]) {
2215 				if (curstart == 255)
2216 					curstart = i;
2217 				curend = i;
2218 			} else {
2219 				if ((curend - curstart) >
2220 				    (maxend - maxstart)) {
2221 					maxstart = curstart;
2222 					maxend = curend;
2223 				}
2224 				curstart = curend = 255;
2225 			}
2226 		}
2227 		if ((curend - curstart) >
2228 		    (maxend - maxstart)) {
2229 			maxstart = curstart;
2230 			maxend = curend;
2231 		}
2232 		for (i = 0; i < 8; i++) {
2233 			if (i >= maxstart && i <= maxend) {
2234 				if (i == 0)
2235 					printf(":");
2236 				if (i == maxend)
2237 					printf(":");
2238 			} else {
2239 				b = ntohs(addr->addr16[i]);
2240 				printf("%x", b);
2241 				if (i < 7)
2242 					printf(":");
2243 			}
2244 		}
2245 		if (p) {
2246 			p = ntohs(p);
2247 			printf("[%u]", p);
2248 		}
2249 		break;
2250 	}
2251 #endif /* INET6 */
2252 	}
2253 }
2254 
2255 void
pf_print_state(struct pf_kstate * s)2256 pf_print_state(struct pf_kstate *s)
2257 {
2258 	pf_print_state_parts(s, NULL, NULL);
2259 }
2260 
2261 static void
pf_print_state_parts(struct pf_kstate * s,struct pf_state_key * skwp,struct pf_state_key * sksp)2262 pf_print_state_parts(struct pf_kstate *s,
2263     struct pf_state_key *skwp, struct pf_state_key *sksp)
2264 {
2265 	struct pf_state_key *skw, *sks;
2266 	u_int8_t proto, dir;
2267 
2268 	/* Do our best to fill these, but they're skipped if NULL */
2269 	skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
2270 	sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
2271 	proto = skw ? skw->proto : (sks ? sks->proto : 0);
2272 	dir = s ? s->direction : 0;
2273 
2274 	switch (proto) {
2275 	case IPPROTO_IPV4:
2276 		printf("IPv4");
2277 		break;
2278 	case IPPROTO_IPV6:
2279 		printf("IPv6");
2280 		break;
2281 	case IPPROTO_TCP:
2282 		printf("TCP");
2283 		break;
2284 	case IPPROTO_UDP:
2285 		printf("UDP");
2286 		break;
2287 	case IPPROTO_ICMP:
2288 		printf("ICMP");
2289 		break;
2290 	case IPPROTO_ICMPV6:
2291 		printf("ICMPv6");
2292 		break;
2293 	default:
2294 		printf("%u", proto);
2295 		break;
2296 	}
2297 	switch (dir) {
2298 	case PF_IN:
2299 		printf(" in");
2300 		break;
2301 	case PF_OUT:
2302 		printf(" out");
2303 		break;
2304 	}
2305 	if (skw) {
2306 		printf(" wire: ");
2307 		pf_print_host(&skw->addr[0], skw->port[0], skw->af);
2308 		printf(" ");
2309 		pf_print_host(&skw->addr[1], skw->port[1], skw->af);
2310 	}
2311 	if (sks) {
2312 		printf(" stack: ");
2313 		if (sks != skw) {
2314 			pf_print_host(&sks->addr[0], sks->port[0], sks->af);
2315 			printf(" ");
2316 			pf_print_host(&sks->addr[1], sks->port[1], sks->af);
2317 		} else
2318 			printf("-");
2319 	}
2320 	if (s) {
2321 		if (proto == IPPROTO_TCP) {
2322 			printf(" [lo=%u high=%u win=%u modulator=%u",
2323 			    s->src.seqlo, s->src.seqhi,
2324 			    s->src.max_win, s->src.seqdiff);
2325 			if (s->src.wscale && s->dst.wscale)
2326 				printf(" wscale=%u",
2327 				    s->src.wscale & PF_WSCALE_MASK);
2328 			printf("]");
2329 			printf(" [lo=%u high=%u win=%u modulator=%u",
2330 			    s->dst.seqlo, s->dst.seqhi,
2331 			    s->dst.max_win, s->dst.seqdiff);
2332 			if (s->src.wscale && s->dst.wscale)
2333 				printf(" wscale=%u",
2334 				s->dst.wscale & PF_WSCALE_MASK);
2335 			printf("]");
2336 		}
2337 		printf(" %u:%u", s->src.state, s->dst.state);
2338 	}
2339 }
2340 
2341 void
pf_print_flags(u_int8_t f)2342 pf_print_flags(u_int8_t f)
2343 {
2344 	if (f)
2345 		printf(" ");
2346 	if (f & TH_FIN)
2347 		printf("F");
2348 	if (f & TH_SYN)
2349 		printf("S");
2350 	if (f & TH_RST)
2351 		printf("R");
2352 	if (f & TH_PUSH)
2353 		printf("P");
2354 	if (f & TH_ACK)
2355 		printf("A");
2356 	if (f & TH_URG)
2357 		printf("U");
2358 	if (f & TH_ECE)
2359 		printf("E");
2360 	if (f & TH_CWR)
2361 		printf("W");
2362 }
2363 
2364 #define	PF_SET_SKIP_STEPS(i)					\
2365 	do {							\
2366 		while (head[i] != cur) {			\
2367 			head[i]->skip[i].ptr = cur;		\
2368 			head[i] = TAILQ_NEXT(head[i], entries);	\
2369 		}						\
2370 	} while (0)
2371 
2372 void
pf_calc_skip_steps(struct pf_krulequeue * rules)2373 pf_calc_skip_steps(struct pf_krulequeue *rules)
2374 {
2375 	struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT];
2376 	int i;
2377 
2378 	cur = TAILQ_FIRST(rules);
2379 	prev = cur;
2380 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2381 		head[i] = cur;
2382 	while (cur != NULL) {
2383 
2384 		if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
2385 			PF_SET_SKIP_STEPS(PF_SKIP_IFP);
2386 		if (cur->direction != prev->direction)
2387 			PF_SET_SKIP_STEPS(PF_SKIP_DIR);
2388 		if (cur->af != prev->af)
2389 			PF_SET_SKIP_STEPS(PF_SKIP_AF);
2390 		if (cur->proto != prev->proto)
2391 			PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
2392 		if (cur->src.neg != prev->src.neg ||
2393 		    pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
2394 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
2395 		if (cur->src.port[0] != prev->src.port[0] ||
2396 		    cur->src.port[1] != prev->src.port[1] ||
2397 		    cur->src.port_op != prev->src.port_op)
2398 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
2399 		if (cur->dst.neg != prev->dst.neg ||
2400 		    pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
2401 			PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
2402 		if (cur->dst.port[0] != prev->dst.port[0] ||
2403 		    cur->dst.port[1] != prev->dst.port[1] ||
2404 		    cur->dst.port_op != prev->dst.port_op)
2405 			PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
2406 
2407 		prev = cur;
2408 		cur = TAILQ_NEXT(cur, entries);
2409 	}
2410 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2411 		PF_SET_SKIP_STEPS(i);
2412 }
2413 
2414 static int
pf_addr_wrap_neq(struct pf_addr_wrap * aw1,struct pf_addr_wrap * aw2)2415 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
2416 {
2417 	if (aw1->type != aw2->type)
2418 		return (1);
2419 	switch (aw1->type) {
2420 	case PF_ADDR_ADDRMASK:
2421 	case PF_ADDR_RANGE:
2422 		if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
2423 			return (1);
2424 		if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
2425 			return (1);
2426 		return (0);
2427 	case PF_ADDR_DYNIFTL:
2428 		return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
2429 	case PF_ADDR_NOROUTE:
2430 	case PF_ADDR_URPFFAILED:
2431 		return (0);
2432 	case PF_ADDR_TABLE:
2433 		return (aw1->p.tbl != aw2->p.tbl);
2434 	default:
2435 		printf("invalid address type: %d\n", aw1->type);
2436 		return (1);
2437 	}
2438 }
2439 
2440 /**
2441  * Checksum updates are a little complicated because the checksum in the TCP/UDP
2442  * header isn't always a full checksum. In some cases (i.e. output) it's a
2443  * pseudo-header checksum, which is a partial checksum over src/dst IP
2444  * addresses, protocol number and length.
2445  *
2446  * That means we have the following cases:
2447  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2448  *  	checksums, we need to update the checksum whenever we change anything.
2449  *  * Output (i.e. the checksum is a pseudo-header checksum):
2450  *  	x The field being updated is src/dst address or affects the length of
2451  *  	the packet. We need to update the pseudo-header checksum (note that this
2452  *  	checksum is not ones' complement).
2453  *  	x Some other field is being modified (e.g. src/dst port numbers): We
2454  *  	don't have to update anything.
2455  **/
2456 u_int16_t
pf_cksum_fixup(u_int16_t cksum,u_int16_t old,u_int16_t new,u_int8_t udp)2457 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2458 {
2459 	u_int32_t x;
2460 
2461 	x = cksum + old - new;
2462 	x = (x + (x >> 16)) & 0xffff;
2463 
2464 	/* optimise: eliminate a branch when not udp */
2465 	if (udp && cksum == 0x0000)
2466 		return cksum;
2467 	if (udp && x == 0x0000)
2468 		x = 0xffff;
2469 
2470 	return (u_int16_t)(x);
2471 }
2472 
2473 static void
pf_patch_8(struct mbuf * m,u_int16_t * cksum,u_int8_t * f,u_int8_t v,bool hi,u_int8_t udp)2474 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi,
2475     u_int8_t udp)
2476 {
2477 	u_int16_t old = htons(hi ? (*f << 8) : *f);
2478 	u_int16_t new = htons(hi ? ( v << 8) :  v);
2479 
2480 	if (*f == v)
2481 		return;
2482 
2483 	*f = v;
2484 
2485 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2486 		return;
2487 
2488 	*cksum = pf_cksum_fixup(*cksum, old, new, udp);
2489 }
2490 
2491 void
pf_patch_16_unaligned(struct mbuf * m,u_int16_t * cksum,void * f,u_int16_t v,bool hi,u_int8_t udp)2492 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v,
2493     bool hi, u_int8_t udp)
2494 {
2495 	u_int8_t *fb = (u_int8_t *)f;
2496 	u_int8_t *vb = (u_int8_t *)&v;
2497 
2498 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2499 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2500 }
2501 
2502 void
pf_patch_32_unaligned(struct mbuf * m,u_int16_t * cksum,void * f,u_int32_t v,bool hi,u_int8_t udp)2503 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v,
2504     bool hi, u_int8_t udp)
2505 {
2506 	u_int8_t *fb = (u_int8_t *)f;
2507 	u_int8_t *vb = (u_int8_t *)&v;
2508 
2509 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2510 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2511 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2512 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2513 }
2514 
2515 u_int16_t
pf_proto_cksum_fixup(struct mbuf * m,u_int16_t cksum,u_int16_t old,u_int16_t new,u_int8_t udp)2516 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2517         u_int16_t new, u_int8_t udp)
2518 {
2519 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2520 		return (cksum);
2521 
2522 	return (pf_cksum_fixup(cksum, old, new, udp));
2523 }
2524 
2525 static void
pf_change_ap(struct mbuf * m,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)2526 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2527         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2528         sa_family_t af)
2529 {
2530 	struct pf_addr	ao;
2531 	u_int16_t	po = *p;
2532 
2533 	PF_ACPY(&ao, a, af);
2534 	PF_ACPY(a, an, af);
2535 
2536 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2537 		*pc = ~*pc;
2538 
2539 	*p = pn;
2540 
2541 	switch (af) {
2542 #ifdef INET
2543 	case AF_INET:
2544 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2545 		    ao.addr16[0], an->addr16[0], 0),
2546 		    ao.addr16[1], an->addr16[1], 0);
2547 		*p = pn;
2548 
2549 		*pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2550 		    ao.addr16[0], an->addr16[0], u),
2551 		    ao.addr16[1], an->addr16[1], u);
2552 
2553 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2554 		break;
2555 #endif /* INET */
2556 #ifdef INET6
2557 	case AF_INET6:
2558 		*pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2559 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2560 		    pf_cksum_fixup(pf_cksum_fixup(*pc,
2561 		    ao.addr16[0], an->addr16[0], u),
2562 		    ao.addr16[1], an->addr16[1], u),
2563 		    ao.addr16[2], an->addr16[2], u),
2564 		    ao.addr16[3], an->addr16[3], u),
2565 		    ao.addr16[4], an->addr16[4], u),
2566 		    ao.addr16[5], an->addr16[5], u),
2567 		    ao.addr16[6], an->addr16[6], u),
2568 		    ao.addr16[7], an->addr16[7], u);
2569 
2570 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2571 		break;
2572 #endif /* INET6 */
2573 	}
2574 
2575 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2576 	    CSUM_DELAY_DATA_IPV6)) {
2577 		*pc = ~*pc;
2578 		if (! *pc)
2579 			*pc = 0xffff;
2580 	}
2581 }
2582 
2583 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2584 void
pf_change_a(void * a,u_int16_t * c,u_int32_t an,u_int8_t u)2585 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2586 {
2587 	u_int32_t	ao;
2588 
2589 	memcpy(&ao, a, sizeof(ao));
2590 	memcpy(a, &an, sizeof(u_int32_t));
2591 	*c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2592 	    ao % 65536, an % 65536, u);
2593 }
2594 
2595 void
pf_change_proto_a(struct mbuf * m,void * a,u_int16_t * c,u_int32_t an,u_int8_t udp)2596 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2597 {
2598 	u_int32_t	ao;
2599 
2600 	memcpy(&ao, a, sizeof(ao));
2601 	memcpy(a, &an, sizeof(u_int32_t));
2602 
2603 	*c = pf_proto_cksum_fixup(m,
2604 	    pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2605 	    ao % 65536, an % 65536, udp);
2606 }
2607 
2608 #ifdef INET6
2609 static void
pf_change_a6(struct pf_addr * a,u_int16_t * c,struct pf_addr * an,u_int8_t u)2610 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2611 {
2612 	struct pf_addr	ao;
2613 
2614 	PF_ACPY(&ao, a, AF_INET6);
2615 	PF_ACPY(a, an, AF_INET6);
2616 
2617 	*c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2618 	    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2619 	    pf_cksum_fixup(pf_cksum_fixup(*c,
2620 	    ao.addr16[0], an->addr16[0], u),
2621 	    ao.addr16[1], an->addr16[1], u),
2622 	    ao.addr16[2], an->addr16[2], u),
2623 	    ao.addr16[3], an->addr16[3], u),
2624 	    ao.addr16[4], an->addr16[4], u),
2625 	    ao.addr16[5], an->addr16[5], u),
2626 	    ao.addr16[6], an->addr16[6], u),
2627 	    ao.addr16[7], an->addr16[7], u);
2628 }
2629 #endif /* INET6 */
2630 
2631 static 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)2632 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2633     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2634     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2635 {
2636 	struct pf_addr	oia, ooa;
2637 
2638 	PF_ACPY(&oia, ia, af);
2639 	if (oa)
2640 		PF_ACPY(&ooa, oa, af);
2641 
2642 	/* Change inner protocol port, fix inner protocol checksum. */
2643 	if (ip != NULL) {
2644 		u_int16_t	oip = *ip;
2645 		u_int32_t	opc;
2646 
2647 		if (pc != NULL)
2648 			opc = *pc;
2649 		*ip = np;
2650 		if (pc != NULL)
2651 			*pc = pf_cksum_fixup(*pc, oip, *ip, u);
2652 		*ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2653 		if (pc != NULL)
2654 			*ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2655 	}
2656 	/* Change inner ip address, fix inner ip and icmp checksums. */
2657 	PF_ACPY(ia, na, af);
2658 	switch (af) {
2659 #ifdef INET
2660 	case AF_INET: {
2661 		u_int32_t	 oh2c = *h2c;
2662 
2663 		*h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2664 		    oia.addr16[0], ia->addr16[0], 0),
2665 		    oia.addr16[1], ia->addr16[1], 0);
2666 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2667 		    oia.addr16[0], ia->addr16[0], 0),
2668 		    oia.addr16[1], ia->addr16[1], 0);
2669 		*ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2670 		break;
2671 	}
2672 #endif /* INET */
2673 #ifdef INET6
2674 	case AF_INET6:
2675 		*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2676 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2677 		    pf_cksum_fixup(pf_cksum_fixup(*ic,
2678 		    oia.addr16[0], ia->addr16[0], u),
2679 		    oia.addr16[1], ia->addr16[1], u),
2680 		    oia.addr16[2], ia->addr16[2], u),
2681 		    oia.addr16[3], ia->addr16[3], u),
2682 		    oia.addr16[4], ia->addr16[4], u),
2683 		    oia.addr16[5], ia->addr16[5], u),
2684 		    oia.addr16[6], ia->addr16[6], u),
2685 		    oia.addr16[7], ia->addr16[7], u);
2686 		break;
2687 #endif /* INET6 */
2688 	}
2689 	/* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2690 	if (oa) {
2691 		PF_ACPY(oa, na, af);
2692 		switch (af) {
2693 #ifdef INET
2694 		case AF_INET:
2695 			*hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2696 			    ooa.addr16[0], oa->addr16[0], 0),
2697 			    ooa.addr16[1], oa->addr16[1], 0);
2698 			break;
2699 #endif /* INET */
2700 #ifdef INET6
2701 		case AF_INET6:
2702 			*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2703 			    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2704 			    pf_cksum_fixup(pf_cksum_fixup(*ic,
2705 			    ooa.addr16[0], oa->addr16[0], u),
2706 			    ooa.addr16[1], oa->addr16[1], u),
2707 			    ooa.addr16[2], oa->addr16[2], u),
2708 			    ooa.addr16[3], oa->addr16[3], u),
2709 			    ooa.addr16[4], oa->addr16[4], u),
2710 			    ooa.addr16[5], oa->addr16[5], u),
2711 			    ooa.addr16[6], oa->addr16[6], u),
2712 			    ooa.addr16[7], oa->addr16[7], u);
2713 			break;
2714 #endif /* INET6 */
2715 		}
2716 	}
2717 }
2718 
2719 
2720 /*
2721  * Need to modulate the sequence numbers in the TCP SACK option
2722  * (credits to Krzysztof Pfaff for report and patch)
2723  */
2724 static int
pf_modulate_sack(struct mbuf * m,int off,struct pf_pdesc * pd,struct tcphdr * th,struct pf_state_peer * dst)2725 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2726     struct tcphdr *th, struct pf_state_peer *dst)
2727 {
2728 	int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2729 	u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2730 	int copyback = 0, i, olen;
2731 	struct sackblk sack;
2732 
2733 #define	TCPOLEN_SACKLEN	(TCPOLEN_SACK + 2)
2734 	if (hlen < TCPOLEN_SACKLEN ||
2735 	    !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2736 		return 0;
2737 
2738 	while (hlen >= TCPOLEN_SACKLEN) {
2739 		size_t startoff = opt - opts;
2740 		olen = opt[1];
2741 		switch (*opt) {
2742 		case TCPOPT_EOL:	/* FALLTHROUGH */
2743 		case TCPOPT_NOP:
2744 			opt++;
2745 			hlen--;
2746 			break;
2747 		case TCPOPT_SACK:
2748 			if (olen > hlen)
2749 				olen = hlen;
2750 			if (olen >= TCPOLEN_SACKLEN) {
2751 				for (i = 2; i + TCPOLEN_SACK <= olen;
2752 				    i += TCPOLEN_SACK) {
2753 					memcpy(&sack, &opt[i], sizeof(sack));
2754 					pf_patch_32_unaligned(m,
2755 					    &th->th_sum, &sack.start,
2756 					    htonl(ntohl(sack.start) - dst->seqdiff),
2757 					    PF_ALGNMNT(startoff),
2758 					    0);
2759 					pf_patch_32_unaligned(m, &th->th_sum,
2760 					    &sack.end,
2761 					    htonl(ntohl(sack.end) - dst->seqdiff),
2762 					    PF_ALGNMNT(startoff),
2763 					    0);
2764 					memcpy(&opt[i], &sack, sizeof(sack));
2765 				}
2766 				copyback = 1;
2767 			}
2768 			/* FALLTHROUGH */
2769 		default:
2770 			if (olen < 2)
2771 				olen = 2;
2772 			hlen -= olen;
2773 			opt += olen;
2774 		}
2775 	}
2776 
2777 	if (copyback)
2778 		m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2779 	return (copyback);
2780 }
2781 
2782 struct mbuf *
pf_build_tcp(const struct pf_krule * 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)2783 pf_build_tcp(const struct pf_krule *r, sa_family_t af,
2784     const struct pf_addr *saddr, const struct pf_addr *daddr,
2785     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2786     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2787     u_int16_t rtag)
2788 {
2789 	struct mbuf	*m;
2790 	int		 len, tlen;
2791 #ifdef INET
2792 	struct ip	*h = NULL;
2793 #endif /* INET */
2794 #ifdef INET6
2795 	struct ip6_hdr	*h6 = NULL;
2796 #endif /* INET6 */
2797 	struct tcphdr	*th;
2798 	char		*opt;
2799 	struct pf_mtag  *pf_mtag;
2800 
2801 	len = 0;
2802 	th = NULL;
2803 
2804 	/* maximum segment size tcp option */
2805 	tlen = sizeof(struct tcphdr);
2806 	if (mss)
2807 		tlen += 4;
2808 
2809 	switch (af) {
2810 #ifdef INET
2811 	case AF_INET:
2812 		len = sizeof(struct ip) + tlen;
2813 		break;
2814 #endif /* INET */
2815 #ifdef INET6
2816 	case AF_INET6:
2817 		len = sizeof(struct ip6_hdr) + tlen;
2818 		break;
2819 #endif /* INET6 */
2820 	default:
2821 		panic("%s: unsupported af %d", __func__, af);
2822 	}
2823 
2824 	m = m_gethdr(M_NOWAIT, MT_DATA);
2825 	if (m == NULL)
2826 		return (NULL);
2827 
2828 #ifdef MAC
2829 	mac_netinet_firewall_send(m);
2830 #endif
2831 	if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2832 		m_freem(m);
2833 		return (NULL);
2834 	}
2835 	if (tag)
2836 		m->m_flags |= M_SKIP_FIREWALL;
2837 	pf_mtag->tag = rtag;
2838 
2839 	if (r != NULL && r->rtableid >= 0)
2840 		M_SETFIB(m, r->rtableid);
2841 
2842 #ifdef ALTQ
2843 	if (r != NULL && r->qid) {
2844 		pf_mtag->qid = r->qid;
2845 
2846 		/* add hints for ecn */
2847 		pf_mtag->hdr = mtod(m, struct ip *);
2848 	}
2849 #endif /* ALTQ */
2850 	m->m_data += max_linkhdr;
2851 	m->m_pkthdr.len = m->m_len = len;
2852 	/* The rest of the stack assumes a rcvif, so provide one.
2853 	 * This is a locally generated packet, so .. close enough. */
2854 	m->m_pkthdr.rcvif = V_loif;
2855 	bzero(m->m_data, len);
2856 	switch (af) {
2857 #ifdef INET
2858 	case AF_INET:
2859 		h = mtod(m, struct ip *);
2860 
2861 		/* IP header fields included in the TCP checksum */
2862 		h->ip_p = IPPROTO_TCP;
2863 		h->ip_len = htons(tlen);
2864 		h->ip_src.s_addr = saddr->v4.s_addr;
2865 		h->ip_dst.s_addr = daddr->v4.s_addr;
2866 
2867 		th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2868 		break;
2869 #endif /* INET */
2870 #ifdef INET6
2871 	case AF_INET6:
2872 		h6 = mtod(m, struct ip6_hdr *);
2873 
2874 		/* IP header fields included in the TCP checksum */
2875 		h6->ip6_nxt = IPPROTO_TCP;
2876 		h6->ip6_plen = htons(tlen);
2877 		memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2878 		memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2879 
2880 		th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2881 		break;
2882 #endif /* INET6 */
2883 	}
2884 
2885 	/* TCP header */
2886 	th->th_sport = sport;
2887 	th->th_dport = dport;
2888 	th->th_seq = htonl(seq);
2889 	th->th_ack = htonl(ack);
2890 	th->th_off = tlen >> 2;
2891 	th->th_flags = flags;
2892 	th->th_win = htons(win);
2893 
2894 	if (mss) {
2895 		opt = (char *)(th + 1);
2896 		opt[0] = TCPOPT_MAXSEG;
2897 		opt[1] = 4;
2898 		HTONS(mss);
2899 		bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2900 	}
2901 
2902 	switch (af) {
2903 #ifdef INET
2904 	case AF_INET:
2905 		/* TCP checksum */
2906 		th->th_sum = in_cksum(m, len);
2907 
2908 		/* Finish the IP header */
2909 		h->ip_v = 4;
2910 		h->ip_hl = sizeof(*h) >> 2;
2911 		h->ip_tos = IPTOS_LOWDELAY;
2912 		h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2913 		h->ip_len = htons(len);
2914 		h->ip_ttl = ttl ? ttl : V_ip_defttl;
2915 		h->ip_sum = 0;
2916 		break;
2917 #endif /* INET */
2918 #ifdef INET6
2919 	case AF_INET6:
2920 		/* TCP checksum */
2921 		th->th_sum = in6_cksum(m, IPPROTO_TCP,
2922 		    sizeof(struct ip6_hdr), tlen);
2923 
2924 		h6->ip6_vfc |= IPV6_VERSION;
2925 		h6->ip6_hlim = IPV6_DEFHLIM;
2926 		break;
2927 #endif /* INET6 */
2928 	}
2929 
2930 	return (m);
2931 }
2932 
2933 void
pf_send_tcp(const struct pf_krule * 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)2934 pf_send_tcp(const struct pf_krule *r, sa_family_t af,
2935     const struct pf_addr *saddr, const struct pf_addr *daddr,
2936     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2937     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2938     u_int16_t rtag)
2939 {
2940 	struct pf_send_entry *pfse;
2941 	struct mbuf	*m;
2942 
2943 	m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, flags,
2944 	    win, mss, ttl, tag, rtag);
2945 	if (m == NULL)
2946 		return;
2947 
2948 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
2949 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2950 	if (pfse == NULL) {
2951 		m_freem(m);
2952 		return;
2953 	}
2954 
2955 	switch (af) {
2956 #ifdef INET
2957 	case AF_INET:
2958 		pfse->pfse_type = PFSE_IP;
2959 		break;
2960 #endif /* INET */
2961 #ifdef INET6
2962 	case AF_INET6:
2963 		pfse->pfse_type = PFSE_IP6;
2964 		break;
2965 #endif /* INET6 */
2966 	}
2967 
2968 	pfse->pfse_m = m;
2969 	pf_send(pfse);
2970 }
2971 
2972 static void
pf_return(struct pf_krule * r,struct pf_krule * nr,struct pf_pdesc * pd,struct pf_state_key * sk,int off,struct mbuf * m,struct tcphdr * th,struct pfi_kkif * kif,u_int16_t bproto_sum,u_int16_t bip_sum,int hdrlen,u_short * reason)2973 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd,
2974     struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th,
2975     struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
2976     u_short *reason)
2977 {
2978 	struct pf_addr	* const saddr = pd->src;
2979 	struct pf_addr	* const daddr = pd->dst;
2980 	sa_family_t	 af = pd->af;
2981 
2982 	/* undo NAT changes, if they have taken place */
2983 	if (nr != NULL) {
2984 		PF_ACPY(saddr, &sk->addr[pd->sidx], af);
2985 		PF_ACPY(daddr, &sk->addr[pd->didx], af);
2986 		if (pd->sport)
2987 			*pd->sport = sk->port[pd->sidx];
2988 		if (pd->dport)
2989 			*pd->dport = sk->port[pd->didx];
2990 		if (pd->proto_sum)
2991 			*pd->proto_sum = bproto_sum;
2992 		if (pd->ip_sum)
2993 			*pd->ip_sum = bip_sum;
2994 		m_copyback(m, off, hdrlen, pd->hdr.any);
2995 	}
2996 	if (pd->proto == IPPROTO_TCP &&
2997 	    ((r->rule_flag & PFRULE_RETURNRST) ||
2998 	    (r->rule_flag & PFRULE_RETURN)) &&
2999 	    !(th->th_flags & TH_RST)) {
3000 		u_int32_t	 ack = ntohl(th->th_seq) + pd->p_len;
3001 		int		 len = 0;
3002 #ifdef INET
3003 		struct ip	*h4;
3004 #endif
3005 #ifdef INET6
3006 		struct ip6_hdr	*h6;
3007 #endif
3008 
3009 		switch (af) {
3010 #ifdef INET
3011 		case AF_INET:
3012 			h4 = mtod(m, struct ip *);
3013 			len = ntohs(h4->ip_len) - off;
3014 			break;
3015 #endif
3016 #ifdef INET6
3017 		case AF_INET6:
3018 			h6 = mtod(m, struct ip6_hdr *);
3019 			len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3020 			break;
3021 #endif
3022 		}
3023 
3024 		if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3025 			REASON_SET(reason, PFRES_PROTCKSUM);
3026 		else {
3027 			if (th->th_flags & TH_SYN)
3028 				ack++;
3029 			if (th->th_flags & TH_FIN)
3030 				ack++;
3031 			pf_send_tcp(r, af, pd->dst,
3032 				pd->src, th->th_dport, th->th_sport,
3033 				ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3034 				r->return_ttl, 1, 0);
3035 		}
3036 	} else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3037 		r->return_icmp)
3038 		pf_send_icmp(m, r->return_icmp >> 8,
3039 			r->return_icmp & 255, af, r);
3040 	else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3041 		r->return_icmp6)
3042 		pf_send_icmp(m, r->return_icmp6 >> 8,
3043 			r->return_icmp6 & 255, af, r);
3044 }
3045 
3046 
3047 static int
pf_match_ieee8021q_pcp(u_int8_t prio,struct mbuf * m)3048 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
3049 {
3050 	struct m_tag *mtag;
3051 	u_int8_t mpcp;
3052 
3053 	mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
3054 	if (mtag == NULL)
3055 		return (0);
3056 
3057 	if (prio == PF_PRIO_ZERO)
3058 		prio = 0;
3059 
3060 	mpcp = *(uint8_t *)(mtag + 1);
3061 
3062 	return (mpcp == prio);
3063 }
3064 
3065 static void
pf_send_icmp(struct mbuf * m,u_int8_t type,u_int8_t code,sa_family_t af,struct pf_krule * r)3066 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
3067     struct pf_krule *r)
3068 {
3069 	struct pf_send_entry *pfse;
3070 	struct mbuf *m0;
3071 	struct pf_mtag *pf_mtag;
3072 
3073 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3074 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3075 	if (pfse == NULL)
3076 		return;
3077 
3078 	if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
3079 		free(pfse, M_PFTEMP);
3080 		return;
3081 	}
3082 
3083 	if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
3084 		free(pfse, M_PFTEMP);
3085 		return;
3086 	}
3087 	/* XXX: revisit */
3088 	m0->m_flags |= M_SKIP_FIREWALL;
3089 
3090 	if (r->rtableid >= 0)
3091 		M_SETFIB(m0, r->rtableid);
3092 
3093 #ifdef ALTQ
3094 	if (r->qid) {
3095 		pf_mtag->qid = r->qid;
3096 		/* add hints for ecn */
3097 		pf_mtag->hdr = mtod(m0, struct ip *);
3098 	}
3099 #endif /* ALTQ */
3100 
3101 	switch (af) {
3102 #ifdef INET
3103 	case AF_INET:
3104 		pfse->pfse_type = PFSE_ICMP;
3105 		break;
3106 #endif /* INET */
3107 #ifdef INET6
3108 	case AF_INET6:
3109 		pfse->pfse_type = PFSE_ICMP6;
3110 		break;
3111 #endif /* INET6 */
3112 	}
3113 	pfse->pfse_m = m0;
3114 	pfse->icmpopts.type = type;
3115 	pfse->icmpopts.code = code;
3116 	pf_send(pfse);
3117 }
3118 
3119 /*
3120  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
3121  * If n is 0, they match if they are equal. If n is != 0, they match if they
3122  * are different.
3123  */
3124 int
pf_match_addr(u_int8_t n,struct pf_addr * a,struct pf_addr * m,struct pf_addr * b,sa_family_t af)3125 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
3126     struct pf_addr *b, sa_family_t af)
3127 {
3128 	int	match = 0;
3129 
3130 	switch (af) {
3131 #ifdef INET
3132 	case AF_INET:
3133 		if ((a->addr32[0] & m->addr32[0]) ==
3134 		    (b->addr32[0] & m->addr32[0]))
3135 			match++;
3136 		break;
3137 #endif /* INET */
3138 #ifdef INET6
3139 	case AF_INET6:
3140 		if (((a->addr32[0] & m->addr32[0]) ==
3141 		     (b->addr32[0] & m->addr32[0])) &&
3142 		    ((a->addr32[1] & m->addr32[1]) ==
3143 		     (b->addr32[1] & m->addr32[1])) &&
3144 		    ((a->addr32[2] & m->addr32[2]) ==
3145 		     (b->addr32[2] & m->addr32[2])) &&
3146 		    ((a->addr32[3] & m->addr32[3]) ==
3147 		     (b->addr32[3] & m->addr32[3])))
3148 			match++;
3149 		break;
3150 #endif /* INET6 */
3151 	}
3152 	if (match) {
3153 		if (n)
3154 			return (0);
3155 		else
3156 			return (1);
3157 	} else {
3158 		if (n)
3159 			return (1);
3160 		else
3161 			return (0);
3162 	}
3163 }
3164 
3165 /*
3166  * Return 1 if b <= a <= e, otherwise return 0.
3167  */
3168 int
pf_match_addr_range(struct pf_addr * b,struct pf_addr * e,struct pf_addr * a,sa_family_t af)3169 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
3170     struct pf_addr *a, sa_family_t af)
3171 {
3172 	switch (af) {
3173 #ifdef INET
3174 	case AF_INET:
3175 		if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
3176 		    (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
3177 			return (0);
3178 		break;
3179 #endif /* INET */
3180 #ifdef INET6
3181 	case AF_INET6: {
3182 		int	i;
3183 
3184 		/* check a >= b */
3185 		for (i = 0; i < 4; ++i)
3186 			if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
3187 				break;
3188 			else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
3189 				return (0);
3190 		/* check a <= e */
3191 		for (i = 0; i < 4; ++i)
3192 			if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
3193 				break;
3194 			else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
3195 				return (0);
3196 		break;
3197 	}
3198 #endif /* INET6 */
3199 	}
3200 	return (1);
3201 }
3202 
3203 static int
pf_match(u_int8_t op,u_int32_t a1,u_int32_t a2,u_int32_t p)3204 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
3205 {
3206 	switch (op) {
3207 	case PF_OP_IRG:
3208 		return ((p > a1) && (p < a2));
3209 	case PF_OP_XRG:
3210 		return ((p < a1) || (p > a2));
3211 	case PF_OP_RRG:
3212 		return ((p >= a1) && (p <= a2));
3213 	case PF_OP_EQ:
3214 		return (p == a1);
3215 	case PF_OP_NE:
3216 		return (p != a1);
3217 	case PF_OP_LT:
3218 		return (p < a1);
3219 	case PF_OP_LE:
3220 		return (p <= a1);
3221 	case PF_OP_GT:
3222 		return (p > a1);
3223 	case PF_OP_GE:
3224 		return (p >= a1);
3225 	}
3226 	return (0); /* never reached */
3227 }
3228 
3229 int
pf_match_port(u_int8_t op,u_int16_t a1,u_int16_t a2,u_int16_t p)3230 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
3231 {
3232 	NTOHS(a1);
3233 	NTOHS(a2);
3234 	NTOHS(p);
3235 	return (pf_match(op, a1, a2, p));
3236 }
3237 
3238 static int
pf_match_uid(u_int8_t op,uid_t a1,uid_t a2,uid_t u)3239 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
3240 {
3241 	if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
3242 		return (0);
3243 	return (pf_match(op, a1, a2, u));
3244 }
3245 
3246 static int
pf_match_gid(u_int8_t op,gid_t a1,gid_t a2,gid_t g)3247 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
3248 {
3249 	if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
3250 		return (0);
3251 	return (pf_match(op, a1, a2, g));
3252 }
3253 
3254 int
pf_match_tag(struct mbuf * m,struct pf_krule * r,int * tag,int mtag)3255 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag)
3256 {
3257 	if (*tag == -1)
3258 		*tag = mtag;
3259 
3260 	return ((!r->match_tag_not && r->match_tag == *tag) ||
3261 	    (r->match_tag_not && r->match_tag != *tag));
3262 }
3263 
3264 int
pf_tag_packet(struct mbuf * m,struct pf_pdesc * pd,int tag)3265 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
3266 {
3267 
3268 	KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
3269 
3270 	if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
3271 		return (ENOMEM);
3272 
3273 	pd->pf_mtag->tag = tag;
3274 
3275 	return (0);
3276 }
3277 
3278 #define	PF_ANCHOR_STACKSIZE	32
3279 struct pf_kanchor_stackframe {
3280 	struct pf_kruleset	*rs;
3281 	struct pf_krule		*r;	/* XXX: + match bit */
3282 	struct pf_kanchor	*child;
3283 };
3284 
3285 /*
3286  * XXX: We rely on malloc(9) returning pointer aligned addresses.
3287  */
3288 #define	PF_ANCHORSTACK_MATCH	0x00000001
3289 #define	PF_ANCHORSTACK_MASK	(PF_ANCHORSTACK_MATCH)
3290 
3291 #define	PF_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
3292 #define	PF_ANCHOR_RULE(f)	(struct pf_krule *)			\
3293 				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
3294 #define	PF_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 			\
3295 				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
3296 } while (0)
3297 
3298 void
pf_step_into_anchor(struct pf_kanchor_stackframe * stack,int * depth,struct pf_kruleset ** rs,int n,struct pf_krule ** r,struct pf_krule ** a,int * match)3299 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth,
3300     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
3301     int *match)
3302 {
3303 	struct pf_kanchor_stackframe	*f;
3304 
3305 	PF_RULES_RASSERT();
3306 
3307 	if (match)
3308 		*match = 0;
3309 	if (*depth >= PF_ANCHOR_STACKSIZE) {
3310 		printf("%s: anchor stack overflow on %s\n",
3311 		    __func__, (*r)->anchor->name);
3312 		*r = TAILQ_NEXT(*r, entries);
3313 		return;
3314 	} else if (*depth == 0 && a != NULL)
3315 		*a = *r;
3316 	f = stack + (*depth)++;
3317 	f->rs = *rs;
3318 	f->r = *r;
3319 	if ((*r)->anchor_wildcard) {
3320 		struct pf_kanchor_node *parent = &(*r)->anchor->children;
3321 
3322 		if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) {
3323 			*r = NULL;
3324 			return;
3325 		}
3326 		*rs = &f->child->ruleset;
3327 	} else {
3328 		f->child = NULL;
3329 		*rs = &(*r)->anchor->ruleset;
3330 	}
3331 	*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
3332 }
3333 
3334 int
pf_step_out_of_anchor(struct pf_kanchor_stackframe * stack,int * depth,struct pf_kruleset ** rs,int n,struct pf_krule ** r,struct pf_krule ** a,int * match)3335 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth,
3336     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
3337     int *match)
3338 {
3339 	struct pf_kanchor_stackframe	*f;
3340 	struct pf_krule *fr;
3341 	int quick = 0;
3342 
3343 	PF_RULES_RASSERT();
3344 
3345 	do {
3346 		if (*depth <= 0)
3347 			break;
3348 		f = stack + *depth - 1;
3349 		fr = PF_ANCHOR_RULE(f);
3350 		if (f->child != NULL) {
3351 			struct pf_kanchor_node *parent;
3352 
3353 			/*
3354 			 * This block traverses through
3355 			 * a wildcard anchor.
3356 			 */
3357 			parent = &fr->anchor->children;
3358 			if (match != NULL && *match) {
3359 				/*
3360 				 * If any of "*" matched, then
3361 				 * "foo/ *" matched, mark frame
3362 				 * appropriately.
3363 				 */
3364 				PF_ANCHOR_SET_MATCH(f);
3365 				*match = 0;
3366 			}
3367 			f->child = RB_NEXT(pf_kanchor_node, parent, f->child);
3368 			if (f->child != NULL) {
3369 				*rs = &f->child->ruleset;
3370 				*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
3371 				if (*r == NULL)
3372 					continue;
3373 				else
3374 					break;
3375 			}
3376 		}
3377 		(*depth)--;
3378 		if (*depth == 0 && a != NULL)
3379 			*a = NULL;
3380 		*rs = f->rs;
3381 		if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
3382 			quick = fr->quick;
3383 		*r = TAILQ_NEXT(fr, entries);
3384 	} while (*r == NULL);
3385 
3386 	return (quick);
3387 }
3388 
3389 #ifdef INET6
3390 void
pf_poolmask(struct pf_addr * naddr,struct pf_addr * raddr,struct pf_addr * rmask,struct pf_addr * saddr,sa_family_t af)3391 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
3392     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
3393 {
3394 	switch (af) {
3395 #ifdef INET
3396 	case AF_INET:
3397 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
3398 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
3399 		break;
3400 #endif /* INET */
3401 	case AF_INET6:
3402 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
3403 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
3404 		naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
3405 		((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
3406 		naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
3407 		((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
3408 		naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
3409 		((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
3410 		break;
3411 	}
3412 }
3413 
3414 void
pf_addr_inc(struct pf_addr * addr,sa_family_t af)3415 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
3416 {
3417 	switch (af) {
3418 #ifdef INET
3419 	case AF_INET:
3420 		addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
3421 		break;
3422 #endif /* INET */
3423 	case AF_INET6:
3424 		if (addr->addr32[3] == 0xffffffff) {
3425 			addr->addr32[3] = 0;
3426 			if (addr->addr32[2] == 0xffffffff) {
3427 				addr->addr32[2] = 0;
3428 				if (addr->addr32[1] == 0xffffffff) {
3429 					addr->addr32[1] = 0;
3430 					addr->addr32[0] =
3431 					    htonl(ntohl(addr->addr32[0]) + 1);
3432 				} else
3433 					addr->addr32[1] =
3434 					    htonl(ntohl(addr->addr32[1]) + 1);
3435 			} else
3436 				addr->addr32[2] =
3437 				    htonl(ntohl(addr->addr32[2]) + 1);
3438 		} else
3439 			addr->addr32[3] =
3440 			    htonl(ntohl(addr->addr32[3]) + 1);
3441 		break;
3442 	}
3443 }
3444 #endif /* INET6 */
3445 
3446 void
pf_rule_to_actions(struct pf_krule * r,struct pf_rule_actions * a)3447 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a)
3448 {
3449 	if (r->qid)
3450 		a->qid = r->qid;
3451 	if (r->pqid)
3452 		a->pqid = r->pqid;
3453 }
3454 
3455 int
pf_socket_lookup(int direction,struct pf_pdesc * pd,struct mbuf * m)3456 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
3457 {
3458 	struct pf_addr		*saddr, *daddr;
3459 	u_int16_t		 sport, dport;
3460 	struct inpcbinfo	*pi;
3461 	struct inpcb		*inp;
3462 
3463 	pd->lookup.uid = UID_MAX;
3464 	pd->lookup.gid = GID_MAX;
3465 
3466 	switch (pd->proto) {
3467 	case IPPROTO_TCP:
3468 		sport = pd->hdr.tcp.th_sport;
3469 		dport = pd->hdr.tcp.th_dport;
3470 		pi = &V_tcbinfo;
3471 		break;
3472 	case IPPROTO_UDP:
3473 		sport = pd->hdr.udp.uh_sport;
3474 		dport = pd->hdr.udp.uh_dport;
3475 		pi = &V_udbinfo;
3476 		break;
3477 	default:
3478 		return (-1);
3479 	}
3480 	if (direction == PF_IN) {
3481 		saddr = pd->src;
3482 		daddr = pd->dst;
3483 	} else {
3484 		u_int16_t	p;
3485 
3486 		p = sport;
3487 		sport = dport;
3488 		dport = p;
3489 		saddr = pd->dst;
3490 		daddr = pd->src;
3491 	}
3492 	switch (pd->af) {
3493 #ifdef INET
3494 	case AF_INET:
3495 		inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
3496 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
3497 		if (inp == NULL) {
3498 			inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
3499 			   daddr->v4, dport, INPLOOKUP_WILDCARD |
3500 			   INPLOOKUP_RLOCKPCB, NULL, m);
3501 			if (inp == NULL)
3502 				return (-1);
3503 		}
3504 		break;
3505 #endif /* INET */
3506 #ifdef INET6
3507 	case AF_INET6:
3508 		inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
3509 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
3510 		if (inp == NULL) {
3511 			inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
3512 			    &daddr->v6, dport, INPLOOKUP_WILDCARD |
3513 			    INPLOOKUP_RLOCKPCB, NULL, m);
3514 			if (inp == NULL)
3515 				return (-1);
3516 		}
3517 		break;
3518 #endif /* INET6 */
3519 
3520 	default:
3521 		return (-1);
3522 	}
3523 	INP_RLOCK_ASSERT(inp);
3524 	pd->lookup.uid = inp->inp_cred->cr_uid;
3525 	pd->lookup.gid = inp->inp_cred->cr_groups[0];
3526 	INP_RUNLOCK(inp);
3527 
3528 	return (1);
3529 }
3530 
3531 u_int8_t
pf_get_wscale(struct mbuf * m,int off,u_int16_t th_off,sa_family_t af)3532 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3533 {
3534 	int		 hlen;
3535 	u_int8_t	 hdr[60];
3536 	u_int8_t	*opt, optlen;
3537 	u_int8_t	 wscale = 0;
3538 
3539 	hlen = th_off << 2;		/* hlen <= sizeof(hdr) */
3540 	if (hlen <= sizeof(struct tcphdr))
3541 		return (0);
3542 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3543 		return (0);
3544 	opt = hdr + sizeof(struct tcphdr);
3545 	hlen -= sizeof(struct tcphdr);
3546 	while (hlen >= 3) {
3547 		switch (*opt) {
3548 		case TCPOPT_EOL:
3549 		case TCPOPT_NOP:
3550 			++opt;
3551 			--hlen;
3552 			break;
3553 		case TCPOPT_WINDOW:
3554 			wscale = opt[2];
3555 			if (wscale > TCP_MAX_WINSHIFT)
3556 				wscale = TCP_MAX_WINSHIFT;
3557 			wscale |= PF_WSCALE_FLAG;
3558 			/* FALLTHROUGH */
3559 		default:
3560 			optlen = opt[1];
3561 			if (optlen < 2)
3562 				optlen = 2;
3563 			hlen -= optlen;
3564 			opt += optlen;
3565 			break;
3566 		}
3567 	}
3568 	return (wscale);
3569 }
3570 
3571 u_int16_t
pf_get_mss(struct mbuf * m,int off,u_int16_t th_off,sa_family_t af)3572 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3573 {
3574 	int		 hlen;
3575 	u_int8_t	 hdr[60];
3576 	u_int8_t	*opt, optlen;
3577 	u_int16_t	 mss = V_tcp_mssdflt;
3578 
3579 	hlen = th_off << 2;	/* hlen <= sizeof(hdr) */
3580 	if (hlen <= sizeof(struct tcphdr))
3581 		return (0);
3582 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3583 		return (0);
3584 	opt = hdr + sizeof(struct tcphdr);
3585 	hlen -= sizeof(struct tcphdr);
3586 	while (hlen >= TCPOLEN_MAXSEG) {
3587 		switch (*opt) {
3588 		case TCPOPT_EOL:
3589 		case TCPOPT_NOP:
3590 			++opt;
3591 			--hlen;
3592 			break;
3593 		case TCPOPT_MAXSEG:
3594 			bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
3595 			NTOHS(mss);
3596 			/* FALLTHROUGH */
3597 		default:
3598 			optlen = opt[1];
3599 			if (optlen < 2)
3600 				optlen = 2;
3601 			hlen -= optlen;
3602 			opt += optlen;
3603 			break;
3604 		}
3605 	}
3606 	return (mss);
3607 }
3608 
3609 static u_int16_t
pf_calc_mss(struct pf_addr * addr,sa_family_t af,int rtableid,u_int16_t offer)3610 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
3611 {
3612 #ifdef INET
3613 	struct nhop4_basic	nh4;
3614 #endif /* INET */
3615 #ifdef INET6
3616 	struct nhop6_basic	nh6;
3617 	struct in6_addr		dst6;
3618 	uint32_t		scopeid;
3619 #endif /* INET6 */
3620 	int			 hlen = 0;
3621 	uint16_t		 mss = 0;
3622 
3623 	switch (af) {
3624 #ifdef INET
3625 	case AF_INET:
3626 		hlen = sizeof(struct ip);
3627 		if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) == 0)
3628 			mss = nh4.nh_mtu - hlen - sizeof(struct tcphdr);
3629 		break;
3630 #endif /* INET */
3631 #ifdef INET6
3632 	case AF_INET6:
3633 		hlen = sizeof(struct ip6_hdr);
3634 		in6_splitscope(&addr->v6, &dst6, &scopeid);
3635 		if (fib6_lookup_nh_basic(rtableid, &dst6, scopeid, 0,0,&nh6)==0)
3636 			mss = nh6.nh_mtu - hlen - sizeof(struct tcphdr);
3637 		break;
3638 #endif /* INET6 */
3639 	}
3640 
3641 	mss = max(V_tcp_mssdflt, mss);
3642 	mss = min(mss, offer);
3643 	mss = max(mss, 64);		/* sanity - at least max opt space */
3644 	return (mss);
3645 }
3646 
3647 static u_int32_t
pf_tcp_iss(struct pf_pdesc * pd)3648 pf_tcp_iss(struct pf_pdesc *pd)
3649 {
3650 	MD5_CTX ctx;
3651 	u_int32_t digest[4];
3652 
3653 	if (V_pf_tcp_secret_init == 0) {
3654 		arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
3655 		MD5Init(&V_pf_tcp_secret_ctx);
3656 		MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
3657 		    sizeof(V_pf_tcp_secret));
3658 		V_pf_tcp_secret_init = 1;
3659 	}
3660 
3661 	ctx = V_pf_tcp_secret_ctx;
3662 
3663 	MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short));
3664 	MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short));
3665 	if (pd->af == AF_INET6) {
3666 		MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3667 		MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3668 	} else {
3669 		MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3670 		MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3671 	}
3672 	MD5Final((u_char *)digest, &ctx);
3673 	V_pf_tcp_iss_off += 4096;
3674 #define	ISN_RANDOM_INCREMENT (4096 - 1)
3675 	return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
3676 	    V_pf_tcp_iss_off);
3677 #undef	ISN_RANDOM_INCREMENT
3678 }
3679 
3680 static int
pf_test_rule(struct pf_krule ** rm,struct pf_kstate ** sm,int direction,struct pfi_kkif * kif,struct mbuf * m,int off,struct pf_pdesc * pd,struct pf_krule ** am,struct pf_kruleset ** rsm,struct inpcb * inp)3681 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, int direction,
3682     struct pfi_kkif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
3683     struct pf_krule **am, struct pf_kruleset **rsm, struct inpcb *inp)
3684 {
3685 	struct pf_krule		*nr = NULL;
3686 	struct pf_addr		* const saddr = pd->src;
3687 	struct pf_addr		* const daddr = pd->dst;
3688 	sa_family_t		 af = pd->af;
3689 	struct pf_krule		*r, *a = NULL;
3690 	struct pf_kruleset	*ruleset = NULL;
3691 	struct pf_ksrc_node	*nsn = NULL;
3692 	struct tcphdr		*th = &pd->hdr.tcp;
3693 	struct pf_state_key	*sk = NULL, *nk = NULL;
3694 	u_short			 reason;
3695 	int			 rewrite = 0, hdrlen = 0;
3696 	int			 tag = -1, rtableid = -1;
3697 	int			 asd = 0;
3698 	int			 match = 0;
3699 	int			 state_icmp = 0;
3700 	u_int16_t		 sport = 0, dport = 0;
3701 	u_int16_t		 bproto_sum = 0, bip_sum = 0;
3702 	u_int8_t		 icmptype = 0, icmpcode = 0;
3703 	struct pf_kanchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
3704 
3705 	PF_RULES_RASSERT();
3706 
3707 	if (inp != NULL) {
3708 		INP_LOCK_ASSERT(inp);
3709 		pd->lookup.uid = inp->inp_cred->cr_uid;
3710 		pd->lookup.gid = inp->inp_cred->cr_groups[0];
3711 		pd->lookup.done = 1;
3712 	}
3713 
3714 	switch (pd->proto) {
3715 	case IPPROTO_TCP:
3716 		sport = th->th_sport;
3717 		dport = th->th_dport;
3718 		hdrlen = sizeof(*th);
3719 		break;
3720 	case IPPROTO_UDP:
3721 		sport = pd->hdr.udp.uh_sport;
3722 		dport = pd->hdr.udp.uh_dport;
3723 		hdrlen = sizeof(pd->hdr.udp);
3724 		break;
3725 #ifdef INET
3726 	case IPPROTO_ICMP:
3727 		if (pd->af != AF_INET)
3728 			break;
3729 		sport = dport = pd->hdr.icmp.icmp_id;
3730 		hdrlen = sizeof(pd->hdr.icmp);
3731 		icmptype = pd->hdr.icmp.icmp_type;
3732 		icmpcode = pd->hdr.icmp.icmp_code;
3733 
3734 		if (icmptype == ICMP_UNREACH ||
3735 		    icmptype == ICMP_SOURCEQUENCH ||
3736 		    icmptype == ICMP_REDIRECT ||
3737 		    icmptype == ICMP_TIMXCEED ||
3738 		    icmptype == ICMP_PARAMPROB)
3739 			state_icmp++;
3740 		break;
3741 #endif /* INET */
3742 #ifdef INET6
3743 	case IPPROTO_ICMPV6:
3744 		if (af != AF_INET6)
3745 			break;
3746 		sport = dport = pd->hdr.icmp6.icmp6_id;
3747 		hdrlen = sizeof(pd->hdr.icmp6);
3748 		icmptype = pd->hdr.icmp6.icmp6_type;
3749 		icmpcode = pd->hdr.icmp6.icmp6_code;
3750 
3751 		if (icmptype == ICMP6_DST_UNREACH ||
3752 		    icmptype == ICMP6_PACKET_TOO_BIG ||
3753 		    icmptype == ICMP6_TIME_EXCEEDED ||
3754 		    icmptype == ICMP6_PARAM_PROB)
3755 			state_icmp++;
3756 		break;
3757 #endif /* INET6 */
3758 	default:
3759 		sport = dport = hdrlen = 0;
3760 		break;
3761 	}
3762 
3763 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3764 
3765 	/* check packet for BINAT/NAT/RDR */
3766 	if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3767 	    &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3768 		KASSERT(sk != NULL, ("%s: null sk", __func__));
3769 		KASSERT(nk != NULL, ("%s: null nk", __func__));
3770 
3771 		if (pd->ip_sum)
3772 			bip_sum = *pd->ip_sum;
3773 
3774 		switch (pd->proto) {
3775 		case IPPROTO_TCP:
3776 			bproto_sum = th->th_sum;
3777 			pd->proto_sum = &th->th_sum;
3778 
3779 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3780 			    nk->port[pd->sidx] != sport) {
3781 				pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
3782 				    &th->th_sum, &nk->addr[pd->sidx],
3783 				    nk->port[pd->sidx], 0, af);
3784 				pd->sport = &th->th_sport;
3785 				sport = th->th_sport;
3786 			}
3787 
3788 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3789 			    nk->port[pd->didx] != dport) {
3790 				pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
3791 				    &th->th_sum, &nk->addr[pd->didx],
3792 				    nk->port[pd->didx], 0, af);
3793 				dport = th->th_dport;
3794 				pd->dport = &th->th_dport;
3795 			}
3796 			rewrite++;
3797 			break;
3798 		case IPPROTO_UDP:
3799 			bproto_sum = pd->hdr.udp.uh_sum;
3800 			pd->proto_sum = &pd->hdr.udp.uh_sum;
3801 
3802 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3803 			    nk->port[pd->sidx] != sport) {
3804 				pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport,
3805 				    pd->ip_sum, &pd->hdr.udp.uh_sum,
3806 				    &nk->addr[pd->sidx],
3807 				    nk->port[pd->sidx], 1, af);
3808 				sport = pd->hdr.udp.uh_sport;
3809 				pd->sport = &pd->hdr.udp.uh_sport;
3810 			}
3811 
3812 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3813 			    nk->port[pd->didx] != dport) {
3814 				pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport,
3815 				    pd->ip_sum, &pd->hdr.udp.uh_sum,
3816 				    &nk->addr[pd->didx],
3817 				    nk->port[pd->didx], 1, af);
3818 				dport = pd->hdr.udp.uh_dport;
3819 				pd->dport = &pd->hdr.udp.uh_dport;
3820 			}
3821 			rewrite++;
3822 			break;
3823 #ifdef INET
3824 		case IPPROTO_ICMP:
3825 			nk->port[0] = nk->port[1];
3826 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3827 				pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3828 				    nk->addr[pd->sidx].v4.s_addr, 0);
3829 
3830 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3831 				pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3832 				    nk->addr[pd->didx].v4.s_addr, 0);
3833 
3834 			if (nk->port[1] != pd->hdr.icmp.icmp_id) {
3835 				pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
3836 				    pd->hdr.icmp.icmp_cksum, sport,
3837 				    nk->port[1], 0);
3838 				pd->hdr.icmp.icmp_id = nk->port[1];
3839 				pd->sport = &pd->hdr.icmp.icmp_id;
3840 			}
3841 			m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
3842 			break;
3843 #endif /* INET */
3844 #ifdef INET6
3845 		case IPPROTO_ICMPV6:
3846 			nk->port[0] = nk->port[1];
3847 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3848 				pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum,
3849 				    &nk->addr[pd->sidx], 0);
3850 
3851 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3852 				pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum,
3853 				    &nk->addr[pd->didx], 0);
3854 			rewrite++;
3855 			break;
3856 #endif /* INET */
3857 		default:
3858 			switch (af) {
3859 #ifdef INET
3860 			case AF_INET:
3861 				if (PF_ANEQ(saddr,
3862 				    &nk->addr[pd->sidx], AF_INET))
3863 					pf_change_a(&saddr->v4.s_addr,
3864 					    pd->ip_sum,
3865 					    nk->addr[pd->sidx].v4.s_addr, 0);
3866 
3867 				if (PF_ANEQ(daddr,
3868 				    &nk->addr[pd->didx], AF_INET))
3869 					pf_change_a(&daddr->v4.s_addr,
3870 					    pd->ip_sum,
3871 					    nk->addr[pd->didx].v4.s_addr, 0);
3872 				break;
3873 #endif /* INET */
3874 #ifdef INET6
3875 			case AF_INET6:
3876 				if (PF_ANEQ(saddr,
3877 				    &nk->addr[pd->sidx], AF_INET6))
3878 					PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3879 
3880 				if (PF_ANEQ(daddr,
3881 				    &nk->addr[pd->didx], AF_INET6))
3882 					PF_ACPY(daddr, &nk->addr[pd->didx], af);
3883 				break;
3884 #endif /* INET */
3885 			}
3886 			break;
3887 		}
3888 		if (nr->natpass)
3889 			r = NULL;
3890 		pd->nat_rule = nr;
3891 	}
3892 
3893 	while (r != NULL) {
3894 		pf_counter_u64_add(&r->evaluations, 1);
3895 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
3896 			r = r->skip[PF_SKIP_IFP].ptr;
3897 		else if (r->direction && r->direction != direction)
3898 			r = r->skip[PF_SKIP_DIR].ptr;
3899 		else if (r->af && r->af != af)
3900 			r = r->skip[PF_SKIP_AF].ptr;
3901 		else if (r->proto && r->proto != pd->proto)
3902 			r = r->skip[PF_SKIP_PROTO].ptr;
3903 		else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3904 		    r->src.neg, kif, M_GETFIB(m)))
3905 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3906 		/* tcp/udp only. port_op always 0 in other cases */
3907 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
3908 		    r->src.port[0], r->src.port[1], sport))
3909 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
3910 		else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3911 		    r->dst.neg, NULL, M_GETFIB(m)))
3912 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
3913 		/* tcp/udp only. port_op always 0 in other cases */
3914 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3915 		    r->dst.port[0], r->dst.port[1], dport))
3916 			r = r->skip[PF_SKIP_DST_PORT].ptr;
3917 		/* icmp only. type always 0 in other cases */
3918 		else if (r->type && r->type != icmptype + 1)
3919 			r = TAILQ_NEXT(r, entries);
3920 		/* icmp only. type always 0 in other cases */
3921 		else if (r->code && r->code != icmpcode + 1)
3922 			r = TAILQ_NEXT(r, entries);
3923 		else if (r->tos && !(r->tos == pd->tos))
3924 			r = TAILQ_NEXT(r, entries);
3925 		else if (r->rule_flag & PFRULE_FRAGMENT)
3926 			r = TAILQ_NEXT(r, entries);
3927 		else if (pd->proto == IPPROTO_TCP &&
3928 		    (r->flagset & th->th_flags) != r->flags)
3929 			r = TAILQ_NEXT(r, entries);
3930 		/* tcp/udp only. uid.op always 0 in other cases */
3931 		else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3932 		    pf_socket_lookup(direction, pd, m), 1)) &&
3933 		    !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3934 		    pd->lookup.uid))
3935 			r = TAILQ_NEXT(r, entries);
3936 		/* tcp/udp only. gid.op always 0 in other cases */
3937 		else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3938 		    pf_socket_lookup(direction, pd, m), 1)) &&
3939 		    !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3940 		    pd->lookup.gid))
3941 			r = TAILQ_NEXT(r, entries);
3942 		else if (r->prio &&
3943 		    !pf_match_ieee8021q_pcp(r->prio, m))
3944 			r = TAILQ_NEXT(r, entries);
3945 		else if (r->prob &&
3946 		    r->prob <= arc4random())
3947 			r = TAILQ_NEXT(r, entries);
3948 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
3949 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
3950 			r = TAILQ_NEXT(r, entries);
3951 		else if (r->os_fingerprint != PF_OSFP_ANY &&
3952 		    (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3953 		    pf_osfp_fingerprint(pd, m, off, th),
3954 		    r->os_fingerprint)))
3955 			r = TAILQ_NEXT(r, entries);
3956 		else {
3957 			if (r->tag)
3958 				tag = r->tag;
3959 			if (r->rtableid >= 0)
3960 				rtableid = r->rtableid;
3961 			if (r->anchor == NULL) {
3962 				if (r->action == PF_MATCH) {
3963 					pf_counter_u64_critical_enter();
3964 					pf_counter_u64_add_protected(&r->packets[direction == PF_OUT], 1);
3965 					pf_counter_u64_add_protected(&r->bytes[direction == PF_OUT], pd->tot_len);
3966 					pf_counter_u64_critical_exit();
3967 					pf_rule_to_actions(r, &pd->act);
3968 					if (r->log)
3969 						PFLOG_PACKET(kif, m, af,
3970 						    direction, PFRES_MATCH, r,
3971 						    a, ruleset, pd, 1);
3972 				} else {
3973 					match = 1;
3974 					*rm = r;
3975 					*am = a;
3976 					*rsm = ruleset;
3977 				}
3978 				if ((*rm)->quick)
3979 					break;
3980 				r = TAILQ_NEXT(r, entries);
3981 			} else
3982 				pf_step_into_anchor(anchor_stack, &asd,
3983 				    &ruleset, PF_RULESET_FILTER, &r, &a,
3984 				    &match);
3985 		}
3986 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3987 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3988 			break;
3989 	}
3990 	r = *rm;
3991 	a = *am;
3992 	ruleset = *rsm;
3993 
3994 	REASON_SET(&reason, PFRES_MATCH);
3995 
3996 	/* apply actions for last matching pass/block rule */
3997 	pf_rule_to_actions(r, &pd->act);
3998 
3999 	if (r->log || (nr != NULL && nr->log)) {
4000 		if (rewrite)
4001 			m_copyback(m, off, hdrlen, pd->hdr.any);
4002 		PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
4003 		    ruleset, pd, 1);
4004 	}
4005 
4006 	if ((r->action == PF_DROP) &&
4007 	    ((r->rule_flag & PFRULE_RETURNRST) ||
4008 	    (r->rule_flag & PFRULE_RETURNICMP) ||
4009 	    (r->rule_flag & PFRULE_RETURN))) {
4010 		pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum,
4011 		    bip_sum, hdrlen, &reason);
4012 	}
4013 
4014 	if (r->action == PF_DROP)
4015 		goto cleanup;
4016 
4017 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
4018 		REASON_SET(&reason, PFRES_MEMORY);
4019 		goto cleanup;
4020 	}
4021 	if (rtableid >= 0)
4022 		M_SETFIB(m, rtableid);
4023 
4024 	if (!state_icmp && (r->keep_state || nr != NULL ||
4025 	    (pd->flags & PFDESC_TCP_NORM))) {
4026 		int action;
4027 		action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
4028 		    sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
4029 		    hdrlen);
4030 		if (action != PF_PASS) {
4031 			if (action == PF_DROP &&
4032 			    (r->rule_flag & PFRULE_RETURN))
4033 				pf_return(r, nr, pd, sk, off, m, th, kif,
4034 				    bproto_sum, bip_sum, hdrlen, &reason);
4035 			return (action);
4036 		}
4037 	} else {
4038 		if (sk != NULL)
4039 			uma_zfree(V_pf_state_key_z, sk);
4040 		if (nk != NULL)
4041 			uma_zfree(V_pf_state_key_z, nk);
4042 	}
4043 
4044 	/* copy back packet headers if we performed NAT operations */
4045 	if (rewrite)
4046 		m_copyback(m, off, hdrlen, pd->hdr.any);
4047 
4048 	if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
4049 	    direction == PF_OUT &&
4050 	    V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m))
4051 		/*
4052 		 * We want the state created, but we dont
4053 		 * want to send this in case a partner
4054 		 * firewall has to know about it to allow
4055 		 * replies through it.
4056 		 */
4057 		return (PF_DEFER);
4058 
4059 	return (PF_PASS);
4060 
4061 cleanup:
4062 	if (sk != NULL)
4063 		uma_zfree(V_pf_state_key_z, sk);
4064 	if (nk != NULL)
4065 		uma_zfree(V_pf_state_key_z, nk);
4066 	return (PF_DROP);
4067 }
4068 
4069 static int
pf_create_state(struct pf_krule * r,struct pf_krule * nr,struct pf_krule * a,struct pf_pdesc * pd,struct pf_ksrc_node * nsn,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_kkif * kif,struct pf_kstate ** sm,int tag,u_int16_t bproto_sum,u_int16_t bip_sum,int hdrlen)4070 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a,
4071     struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk,
4072     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
4073     u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm,
4074     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
4075 {
4076 	struct pf_kstate	*s = NULL;
4077 	struct pf_ksrc_node	*sn = NULL;
4078 	struct tcphdr		*th = &pd->hdr.tcp;
4079 	u_int16_t		 mss = V_tcp_mssdflt;
4080 	u_short			 reason;
4081 
4082 	/* check maximums */
4083 	if (r->max_states &&
4084 	    (counter_u64_fetch(r->states_cur) >= r->max_states)) {
4085 		counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
4086 		REASON_SET(&reason, PFRES_MAXSTATES);
4087 		goto csfailed;
4088 	}
4089 	/* src node for filter rule */
4090 	if ((r->rule_flag & PFRULE_SRCTRACK ||
4091 	    r->rpool.opts & PF_POOL_STICKYADDR) &&
4092 	    pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
4093 		REASON_SET(&reason, PFRES_SRCLIMIT);
4094 		goto csfailed;
4095 	}
4096 	/* src node for translation rule */
4097 	if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
4098 	    pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
4099 		REASON_SET(&reason, PFRES_SRCLIMIT);
4100 		goto csfailed;
4101 	}
4102 	s = pf_alloc_state(M_NOWAIT);
4103 	if (s == NULL) {
4104 		REASON_SET(&reason, PFRES_MEMORY);
4105 		goto csfailed;
4106 	}
4107 	s->rule.ptr = r;
4108 	s->nat_rule.ptr = nr;
4109 	s->anchor.ptr = a;
4110 	STATE_INC_COUNTERS(s);
4111 	if (r->allow_opts)
4112 		s->state_flags |= PFSTATE_ALLOWOPTS;
4113 	if (r->rule_flag & PFRULE_STATESLOPPY)
4114 		s->state_flags |= PFSTATE_SLOPPY;
4115 	s->log = r->log & PF_LOG_ALL;
4116 	s->sync_state = PFSYNC_S_NONE;
4117 	s->qid = pd->act.qid;
4118 	s->pqid = pd->act.pqid;
4119 	if (nr != NULL)
4120 		s->log |= nr->log & PF_LOG_ALL;
4121 	switch (pd->proto) {
4122 	case IPPROTO_TCP:
4123 		s->src.seqlo = ntohl(th->th_seq);
4124 		s->src.seqhi = s->src.seqlo + pd->p_len + 1;
4125 		if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
4126 		    r->keep_state == PF_STATE_MODULATE) {
4127 			/* Generate sequence number modulator */
4128 			if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
4129 			    0)
4130 				s->src.seqdiff = 1;
4131 			pf_change_proto_a(m, &th->th_seq, &th->th_sum,
4132 			    htonl(s->src.seqlo + s->src.seqdiff), 0);
4133 			*rewrite = 1;
4134 		} else
4135 			s->src.seqdiff = 0;
4136 		if (th->th_flags & TH_SYN) {
4137 			s->src.seqhi++;
4138 			s->src.wscale = pf_get_wscale(m, off,
4139 			    th->th_off, pd->af);
4140 		}
4141 		s->src.max_win = MAX(ntohs(th->th_win), 1);
4142 		if (s->src.wscale & PF_WSCALE_MASK) {
4143 			/* Remove scale factor from initial window */
4144 			int win = s->src.max_win;
4145 			win += 1 << (s->src.wscale & PF_WSCALE_MASK);
4146 			s->src.max_win = (win - 1) >>
4147 			    (s->src.wscale & PF_WSCALE_MASK);
4148 		}
4149 		if (th->th_flags & TH_FIN)
4150 			s->src.seqhi++;
4151 		s->dst.seqhi = 1;
4152 		s->dst.max_win = 1;
4153 		pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT);
4154 		pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED);
4155 		s->timeout = PFTM_TCP_FIRST_PACKET;
4156 		atomic_add_32(&V_pf_status.states_halfopen, 1);
4157 		break;
4158 	case IPPROTO_UDP:
4159 		pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE);
4160 		pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC);
4161 		s->timeout = PFTM_UDP_FIRST_PACKET;
4162 		break;
4163 	case IPPROTO_ICMP:
4164 #ifdef INET6
4165 	case IPPROTO_ICMPV6:
4166 #endif
4167 		s->timeout = PFTM_ICMP_FIRST_PACKET;
4168 		break;
4169 	default:
4170 		pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE);
4171 		pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC);
4172 		s->timeout = PFTM_OTHER_FIRST_PACKET;
4173 	}
4174 
4175 	if (r->rt) {
4176 		if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) {
4177 			REASON_SET(&reason, PFRES_MAPFAILED);
4178 			pf_src_tree_remove_state(s);
4179 			s->timeout = PFTM_UNLINKED;
4180 			STATE_DEC_COUNTERS(s);
4181 			pf_free_state(s);
4182 			goto csfailed;
4183 		}
4184 		s->rt_kif = r->rpool.cur->kif;
4185 	}
4186 
4187 	s->creation = time_uptime;
4188 	s->expire = time_uptime;
4189 
4190 	if (sn != NULL)
4191 		s->src_node = sn;
4192 	if (nsn != NULL) {
4193 		/* XXX We only modify one side for now. */
4194 		PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
4195 		s->nat_src_node = nsn;
4196 	}
4197 	if (pd->proto == IPPROTO_TCP) {
4198 		if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
4199 		    off, pd, th, &s->src, &s->dst)) {
4200 			REASON_SET(&reason, PFRES_MEMORY);
4201 			pf_src_tree_remove_state(s);
4202 			s->timeout = PFTM_UNLINKED;
4203 			STATE_DEC_COUNTERS(s);
4204 			pf_free_state(s);
4205 			return (PF_DROP);
4206 		}
4207 		if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
4208 		    pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
4209 		    &s->src, &s->dst, rewrite)) {
4210 			/* This really shouldn't happen!!! */
4211 			DPFPRINTF(PF_DEBUG_URGENT,
4212 			    ("pf_normalize_tcp_stateful failed on first "
4213 			     "pkt\n"));
4214 			pf_src_tree_remove_state(s);
4215 			s->timeout = PFTM_UNLINKED;
4216 			STATE_DEC_COUNTERS(s);
4217 			pf_free_state(s);
4218 			return (PF_DROP);
4219 		}
4220 	}
4221 	s->direction = pd->dir;
4222 
4223 	/*
4224 	 * sk/nk could already been setup by pf_get_translation().
4225 	 */
4226 	if (nr == NULL) {
4227 		KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
4228 		    __func__, nr, sk, nk));
4229 		sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
4230 		if (sk == NULL)
4231 			goto csfailed;
4232 		nk = sk;
4233 	} else
4234 		KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
4235 		    __func__, nr, sk, nk));
4236 
4237 	/* Swap sk/nk for PF_OUT. */
4238 	if (pf_state_insert(BOUND_IFACE(r, kif), kif,
4239 	    (pd->dir == PF_IN) ? sk : nk,
4240 	    (pd->dir == PF_IN) ? nk : sk, s)) {
4241 		REASON_SET(&reason, PFRES_STATEINS);
4242 		pf_src_tree_remove_state(s);
4243 		s->timeout = PFTM_UNLINKED;
4244 		STATE_DEC_COUNTERS(s);
4245 		pf_free_state(s);
4246 		return (PF_DROP);
4247 	} else
4248 		*sm = s;
4249 
4250 	if (tag > 0)
4251 		s->tag = tag;
4252 	if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
4253 	    TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
4254 		pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC);
4255 		/* undo NAT changes, if they have taken place */
4256 		if (nr != NULL) {
4257 			struct pf_state_key *skt = s->key[PF_SK_WIRE];
4258 			if (pd->dir == PF_OUT)
4259 				skt = s->key[PF_SK_STACK];
4260 			PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
4261 			PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
4262 			if (pd->sport)
4263 				*pd->sport = skt->port[pd->sidx];
4264 			if (pd->dport)
4265 				*pd->dport = skt->port[pd->didx];
4266 			if (pd->proto_sum)
4267 				*pd->proto_sum = bproto_sum;
4268 			if (pd->ip_sum)
4269 				*pd->ip_sum = bip_sum;
4270 			m_copyback(m, off, hdrlen, pd->hdr.any);
4271 		}
4272 		s->src.seqhi = htonl(arc4random());
4273 		/* Find mss option */
4274 		int rtid = M_GETFIB(m);
4275 		mss = pf_get_mss(m, off, th->th_off, pd->af);
4276 		mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
4277 		mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
4278 		s->src.mss = mss;
4279 		pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport,
4280 		    th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
4281 		    TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0);
4282 		REASON_SET(&reason, PFRES_SYNPROXY);
4283 		return (PF_SYNPROXY_DROP);
4284 	}
4285 
4286 	return (PF_PASS);
4287 
4288 csfailed:
4289 	if (sk != NULL)
4290 		uma_zfree(V_pf_state_key_z, sk);
4291 	if (nk != NULL)
4292 		uma_zfree(V_pf_state_key_z, nk);
4293 
4294 	if (sn != NULL) {
4295 		struct pf_srchash *sh;
4296 
4297 		sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
4298 		PF_HASHROW_LOCK(sh);
4299 		if (--sn->states == 0 && sn->expire == 0) {
4300 			pf_unlink_src_node(sn);
4301 			uma_zfree(V_pf_sources_z, sn);
4302 			counter_u64_add(
4303 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
4304 		}
4305 		PF_HASHROW_UNLOCK(sh);
4306 	}
4307 
4308 	if (nsn != sn && nsn != NULL) {
4309 		struct pf_srchash *sh;
4310 
4311 		sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)];
4312 		PF_HASHROW_LOCK(sh);
4313 		if (--nsn->states == 0 && nsn->expire == 0) {
4314 			pf_unlink_src_node(nsn);
4315 			uma_zfree(V_pf_sources_z, nsn);
4316 			counter_u64_add(
4317 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
4318 		}
4319 		PF_HASHROW_UNLOCK(sh);
4320 	}
4321 
4322 	return (PF_DROP);
4323 }
4324 
4325 static int
pf_test_fragment(struct pf_krule ** rm,int direction,struct pfi_kkif * kif,struct mbuf * m,void * h,struct pf_pdesc * pd,struct pf_krule ** am,struct pf_kruleset ** rsm)4326 pf_test_fragment(struct pf_krule **rm, int direction, struct pfi_kkif *kif,
4327     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am,
4328     struct pf_kruleset **rsm)
4329 {
4330 	struct pf_krule		*r, *a = NULL;
4331 	struct pf_kruleset	*ruleset = NULL;
4332 	sa_family_t		 af = pd->af;
4333 	u_short			 reason;
4334 	int			 tag = -1;
4335 	int			 asd = 0;
4336 	int			 match = 0;
4337 	struct pf_kanchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
4338 
4339 	PF_RULES_RASSERT();
4340 
4341 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
4342 	while (r != NULL) {
4343 		pf_counter_u64_add(&r->evaluations, 1);
4344 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
4345 			r = r->skip[PF_SKIP_IFP].ptr;
4346 		else if (r->direction && r->direction != direction)
4347 			r = r->skip[PF_SKIP_DIR].ptr;
4348 		else if (r->af && r->af != af)
4349 			r = r->skip[PF_SKIP_AF].ptr;
4350 		else if (r->proto && r->proto != pd->proto)
4351 			r = r->skip[PF_SKIP_PROTO].ptr;
4352 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
4353 		    r->src.neg, kif, M_GETFIB(m)))
4354 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
4355 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
4356 		    r->dst.neg, NULL, M_GETFIB(m)))
4357 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
4358 		else if (r->tos && !(r->tos == pd->tos))
4359 			r = TAILQ_NEXT(r, entries);
4360 		else if (r->os_fingerprint != PF_OSFP_ANY)
4361 			r = TAILQ_NEXT(r, entries);
4362 		else if (pd->proto == IPPROTO_UDP &&
4363 		    (r->src.port_op || r->dst.port_op))
4364 			r = TAILQ_NEXT(r, entries);
4365 		else if (pd->proto == IPPROTO_TCP &&
4366 		    (r->src.port_op || r->dst.port_op || r->flagset))
4367 			r = TAILQ_NEXT(r, entries);
4368 		else if ((pd->proto == IPPROTO_ICMP ||
4369 		    pd->proto == IPPROTO_ICMPV6) &&
4370 		    (r->type || r->code))
4371 			r = TAILQ_NEXT(r, entries);
4372 		else if (r->prio &&
4373 		    !pf_match_ieee8021q_pcp(r->prio, m))
4374 			r = TAILQ_NEXT(r, entries);
4375 		else if (r->prob && r->prob <=
4376 		    (arc4random() % (UINT_MAX - 1) + 1))
4377 			r = TAILQ_NEXT(r, entries);
4378 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
4379 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
4380 			r = TAILQ_NEXT(r, entries);
4381 		else {
4382 			if (r->anchor == NULL) {
4383 				if (r->action == PF_MATCH) {
4384 					pf_counter_u64_critical_enter();
4385 					pf_counter_u64_add_protected(&r->packets[direction == PF_OUT], 1);
4386 					pf_counter_u64_add_protected(&r->bytes[direction == PF_OUT], pd->tot_len);
4387 					pf_counter_u64_critical_exit();
4388 					pf_rule_to_actions(r, &pd->act);
4389 					if (r->log)
4390 						PFLOG_PACKET(kif, m, af,
4391 						    direction, PFRES_MATCH, r,
4392 						    a, ruleset, pd, 1);
4393 				} else {
4394 					match = 1;
4395 					*rm = r;
4396 					*am = a;
4397 					*rsm = ruleset;
4398 				}
4399 				if ((*rm)->quick)
4400 					break;
4401 				r = TAILQ_NEXT(r, entries);
4402 			} else
4403 				pf_step_into_anchor(anchor_stack, &asd,
4404 				    &ruleset, PF_RULESET_FILTER, &r, &a,
4405 				    &match);
4406 		}
4407 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
4408 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
4409 			break;
4410 	}
4411 	r = *rm;
4412 	a = *am;
4413 	ruleset = *rsm;
4414 
4415 	REASON_SET(&reason, PFRES_MATCH);
4416 
4417 	/* apply actions for last matching pass/block rule */
4418 	pf_rule_to_actions(r, &pd->act);
4419 
4420 	if (r->log)
4421 		PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
4422 		    1);
4423 
4424 	if (r->action != PF_PASS)
4425 		return (PF_DROP);
4426 
4427 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
4428 		REASON_SET(&reason, PFRES_MEMORY);
4429 		return (PF_DROP);
4430 	}
4431 
4432 	return (PF_PASS);
4433 }
4434 
4435 static int
pf_tcp_track_full(struct pf_kstate ** state,struct pfi_kkif * kif,struct mbuf * m,int off,struct pf_pdesc * pd,u_short * reason,int * copyback)4436 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif,
4437     struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason,
4438     int *copyback)
4439 {
4440 	struct tcphdr		*th = &pd->hdr.tcp;
4441 	struct pf_state_peer	*src, *dst;
4442 	u_int16_t		 win = ntohs(th->th_win);
4443 	u_int32_t		 ack, end, seq, orig_seq;
4444 	u_int8_t		 sws, dws, psrc, pdst;
4445 	int			 ackskew;
4446 
4447 	if (pd->dir == (*state)->direction) {
4448 		src = &(*state)->src;
4449 		dst = &(*state)->dst;
4450 		psrc = PF_PEER_SRC;
4451 		pdst = PF_PEER_DST;
4452 	} else {
4453 		src = &(*state)->dst;
4454 		dst = &(*state)->src;
4455 		psrc = PF_PEER_DST;
4456 		pdst = PF_PEER_SRC;
4457 	}
4458 
4459 	if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
4460 		sws = src->wscale & PF_WSCALE_MASK;
4461 		dws = dst->wscale & PF_WSCALE_MASK;
4462 	} else
4463 		sws = dws = 0;
4464 
4465 	/*
4466 	 * Sequence tracking algorithm from Guido van Rooij's paper:
4467 	 *   http://www.madison-gurkha.com/publications/tcp_filtering/
4468 	 *	tcp_filtering.ps
4469 	 */
4470 
4471 	orig_seq = seq = ntohl(th->th_seq);
4472 	if (src->seqlo == 0) {
4473 		/* First packet from this end. Set its state */
4474 
4475 		if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
4476 		    src->scrub == NULL) {
4477 			if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
4478 				REASON_SET(reason, PFRES_MEMORY);
4479 				return (PF_DROP);
4480 			}
4481 		}
4482 
4483 		/* Deferred generation of sequence number modulator */
4484 		if (dst->seqdiff && !src->seqdiff) {
4485 			/* use random iss for the TCP server */
4486 			while ((src->seqdiff = arc4random() - seq) == 0)
4487 				;
4488 			ack = ntohl(th->th_ack) - dst->seqdiff;
4489 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
4490 			    src->seqdiff), 0);
4491 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
4492 			*copyback = 1;
4493 		} else {
4494 			ack = ntohl(th->th_ack);
4495 		}
4496 
4497 		end = seq + pd->p_len;
4498 		if (th->th_flags & TH_SYN) {
4499 			end++;
4500 			if (dst->wscale & PF_WSCALE_FLAG) {
4501 				src->wscale = pf_get_wscale(m, off, th->th_off,
4502 				    pd->af);
4503 				if (src->wscale & PF_WSCALE_FLAG) {
4504 					/* Remove scale factor from initial
4505 					 * window */
4506 					sws = src->wscale & PF_WSCALE_MASK;
4507 					win = ((u_int32_t)win + (1 << sws) - 1)
4508 					    >> sws;
4509 					dws = dst->wscale & PF_WSCALE_MASK;
4510 				} else {
4511 					/* fixup other window */
4512 					dst->max_win <<= dst->wscale &
4513 					    PF_WSCALE_MASK;
4514 					/* in case of a retrans SYN|ACK */
4515 					dst->wscale = 0;
4516 				}
4517 			}
4518 		}
4519 		if (th->th_flags & TH_FIN)
4520 			end++;
4521 
4522 		src->seqlo = seq;
4523 		if (src->state < TCPS_SYN_SENT)
4524 			pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
4525 
4526 		/*
4527 		 * May need to slide the window (seqhi may have been set by
4528 		 * the crappy stack check or if we picked up the connection
4529 		 * after establishment)
4530 		 */
4531 		if (src->seqhi == 1 ||
4532 		    SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
4533 			src->seqhi = end + MAX(1, dst->max_win << dws);
4534 		if (win > src->max_win)
4535 			src->max_win = win;
4536 
4537 	} else {
4538 		ack = ntohl(th->th_ack) - dst->seqdiff;
4539 		if (src->seqdiff) {
4540 			/* Modulate sequence numbers */
4541 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
4542 			    src->seqdiff), 0);
4543 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
4544 			*copyback = 1;
4545 		}
4546 		end = seq + pd->p_len;
4547 		if (th->th_flags & TH_SYN)
4548 			end++;
4549 		if (th->th_flags & TH_FIN)
4550 			end++;
4551 	}
4552 
4553 	if ((th->th_flags & TH_ACK) == 0) {
4554 		/* Let it pass through the ack skew check */
4555 		ack = dst->seqlo;
4556 	} else if ((ack == 0 &&
4557 	    (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
4558 	    /* broken tcp stacks do not set ack */
4559 	    (dst->state < TCPS_SYN_SENT)) {
4560 		/*
4561 		 * Many stacks (ours included) will set the ACK number in an
4562 		 * FIN|ACK if the SYN times out -- no sequence to ACK.
4563 		 */
4564 		ack = dst->seqlo;
4565 	}
4566 
4567 	if (seq == end) {
4568 		/* Ease sequencing restrictions on no data packets */
4569 		seq = src->seqlo;
4570 		end = seq;
4571 	}
4572 
4573 	ackskew = dst->seqlo - ack;
4574 
4575 
4576 	/*
4577 	 * Need to demodulate the sequence numbers in any TCP SACK options
4578 	 * (Selective ACK). We could optionally validate the SACK values
4579 	 * against the current ACK window, either forwards or backwards, but
4580 	 * I'm not confident that SACK has been implemented properly
4581 	 * everywhere. It wouldn't surprise me if several stacks accidentally
4582 	 * SACK too far backwards of previously ACKed data. There really aren't
4583 	 * any security implications of bad SACKing unless the target stack
4584 	 * doesn't validate the option length correctly. Someone trying to
4585 	 * spoof into a TCP connection won't bother blindly sending SACK
4586 	 * options anyway.
4587 	 */
4588 	if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
4589 		if (pf_modulate_sack(m, off, pd, th, dst))
4590 			*copyback = 1;
4591 	}
4592 
4593 
4594 #define	MAXACKWINDOW (0xffff + 1500)	/* 1500 is an arbitrary fudge factor */
4595 	if (SEQ_GEQ(src->seqhi, end) &&
4596 	    /* Last octet inside other's window space */
4597 	    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
4598 	    /* Retrans: not more than one window back */
4599 	    (ackskew >= -MAXACKWINDOW) &&
4600 	    /* Acking not more than one reassembled fragment backwards */
4601 	    (ackskew <= (MAXACKWINDOW << sws)) &&
4602 	    /* Acking not more than one window forward */
4603 	    ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
4604 	    (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) {
4605 	    /* Require an exact/+1 sequence match on resets when possible */
4606 
4607 		if (dst->scrub || src->scrub) {
4608 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4609 			    *state, src, dst, copyback))
4610 				return (PF_DROP);
4611 		}
4612 
4613 		/* update max window */
4614 		if (src->max_win < win)
4615 			src->max_win = win;
4616 		/* synchronize sequencing */
4617 		if (SEQ_GT(end, src->seqlo))
4618 			src->seqlo = end;
4619 		/* slide the window of what the other end can send */
4620 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4621 			dst->seqhi = ack + MAX((win << sws), 1);
4622 
4623 
4624 		/* update states */
4625 		if (th->th_flags & TH_SYN)
4626 			if (src->state < TCPS_SYN_SENT)
4627 				pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
4628 		if (th->th_flags & TH_FIN)
4629 			if (src->state < TCPS_CLOSING)
4630 				pf_set_protostate(*state, psrc, TCPS_CLOSING);
4631 		if (th->th_flags & TH_ACK) {
4632 			if (dst->state == TCPS_SYN_SENT) {
4633 				pf_set_protostate(*state, pdst,
4634 				    TCPS_ESTABLISHED);
4635 				if (src->state == TCPS_ESTABLISHED &&
4636 				    (*state)->src_node != NULL &&
4637 				    pf_src_connlimit(state)) {
4638 					REASON_SET(reason, PFRES_SRCLIMIT);
4639 					return (PF_DROP);
4640 				}
4641 			} else if (dst->state == TCPS_CLOSING)
4642 				pf_set_protostate(*state, pdst,
4643 				    TCPS_FIN_WAIT_2);
4644 		}
4645 		if (th->th_flags & TH_RST)
4646 			pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
4647 
4648 		/* update expire time */
4649 		(*state)->expire = time_uptime;
4650 		if (src->state >= TCPS_FIN_WAIT_2 &&
4651 		    dst->state >= TCPS_FIN_WAIT_2)
4652 			(*state)->timeout = PFTM_TCP_CLOSED;
4653 		else if (src->state >= TCPS_CLOSING &&
4654 		    dst->state >= TCPS_CLOSING)
4655 			(*state)->timeout = PFTM_TCP_FIN_WAIT;
4656 		else if (src->state < TCPS_ESTABLISHED ||
4657 		    dst->state < TCPS_ESTABLISHED)
4658 			(*state)->timeout = PFTM_TCP_OPENING;
4659 		else if (src->state >= TCPS_CLOSING ||
4660 		    dst->state >= TCPS_CLOSING)
4661 			(*state)->timeout = PFTM_TCP_CLOSING;
4662 		else
4663 			(*state)->timeout = PFTM_TCP_ESTABLISHED;
4664 
4665 		/* Fall through to PASS packet */
4666 
4667 	} else if ((dst->state < TCPS_SYN_SENT ||
4668 		dst->state >= TCPS_FIN_WAIT_2 ||
4669 		src->state >= TCPS_FIN_WAIT_2) &&
4670 	    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4671 	    /* Within a window forward of the originating packet */
4672 	    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4673 	    /* Within a window backward of the originating packet */
4674 
4675 		/*
4676 		 * This currently handles three situations:
4677 		 *  1) Stupid stacks will shotgun SYNs before their peer
4678 		 *     replies.
4679 		 *  2) When PF catches an already established stream (the
4680 		 *     firewall rebooted, the state table was flushed, routes
4681 		 *     changed...)
4682 		 *  3) Packets get funky immediately after the connection
4683 		 *     closes (this should catch Solaris spurious ACK|FINs
4684 		 *     that web servers like to spew after a close)
4685 		 *
4686 		 * This must be a little more careful than the above code
4687 		 * since packet floods will also be caught here. We don't
4688 		 * update the TTL here to mitigate the damage of a packet
4689 		 * flood and so the same code can handle awkward establishment
4690 		 * and a loosened connection close.
4691 		 * In the establishment case, a correct peer response will
4692 		 * validate the connection, go through the normal state code
4693 		 * and keep updating the state TTL.
4694 		 */
4695 
4696 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
4697 			printf("pf: loose state match: ");
4698 			pf_print_state(*state);
4699 			pf_print_flags(th->th_flags);
4700 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4701 			    "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
4702 			    pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
4703 			    (unsigned long long)(*state)->packets[1],
4704 			    pd->dir == PF_IN ? "in" : "out",
4705 			    pd->dir == (*state)->direction ? "fwd" : "rev");
4706 		}
4707 
4708 		if (dst->scrub || src->scrub) {
4709 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4710 			    *state, src, dst, copyback))
4711 				return (PF_DROP);
4712 		}
4713 
4714 		/* update max window */
4715 		if (src->max_win < win)
4716 			src->max_win = win;
4717 		/* synchronize sequencing */
4718 		if (SEQ_GT(end, src->seqlo))
4719 			src->seqlo = end;
4720 		/* slide the window of what the other end can send */
4721 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4722 			dst->seqhi = ack + MAX((win << sws), 1);
4723 
4724 		/*
4725 		 * Cannot set dst->seqhi here since this could be a shotgunned
4726 		 * SYN and not an already established connection.
4727 		 */
4728 
4729 		if (th->th_flags & TH_FIN)
4730 			if (src->state < TCPS_CLOSING)
4731 				pf_set_protostate(*state, psrc, TCPS_CLOSING);
4732 		if (th->th_flags & TH_RST)
4733 			pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
4734 
4735 		/* Fall through to PASS packet */
4736 
4737 	} else {
4738 		if ((*state)->dst.state == TCPS_SYN_SENT &&
4739 		    (*state)->src.state == TCPS_SYN_SENT) {
4740 			/* Send RST for state mismatches during handshake */
4741 			if (!(th->th_flags & TH_RST))
4742 				pf_send_tcp((*state)->rule.ptr, pd->af,
4743 				    pd->dst, pd->src, th->th_dport,
4744 				    th->th_sport, ntohl(th->th_ack), 0,
4745 				    TH_RST, 0, 0,
4746 				    (*state)->rule.ptr->return_ttl, 1, 0);
4747 			src->seqlo = 0;
4748 			src->seqhi = 1;
4749 			src->max_win = 1;
4750 		} else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4751 			printf("pf: BAD state: ");
4752 			pf_print_state(*state);
4753 			pf_print_flags(th->th_flags);
4754 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4755 			    "pkts=%llu:%llu dir=%s,%s\n",
4756 			    seq, orig_seq, ack, pd->p_len, ackskew,
4757 			    (unsigned long long)(*state)->packets[0],
4758 			    (unsigned long long)(*state)->packets[1],
4759 			    pd->dir == PF_IN ? "in" : "out",
4760 			    pd->dir == (*state)->direction ? "fwd" : "rev");
4761 			printf("pf: State failure on: %c %c %c %c | %c %c\n",
4762 			    SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4763 			    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4764 			    ' ': '2',
4765 			    (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4766 			    (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4767 			    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4768 			    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4769 		}
4770 		REASON_SET(reason, PFRES_BADSTATE);
4771 		return (PF_DROP);
4772 	}
4773 
4774 	return (PF_PASS);
4775 }
4776 
4777 static int
pf_tcp_track_sloppy(struct pf_kstate ** state,struct pf_pdesc * pd,u_short * reason)4778 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason)
4779 {
4780 	struct tcphdr		*th = &pd->hdr.tcp;
4781 	struct pf_state_peer	*src, *dst;
4782 	u_int8_t		 psrc, pdst;
4783 
4784 	if (pd->dir == (*state)->direction) {
4785 		src = &(*state)->src;
4786 		dst = &(*state)->dst;
4787 		psrc = PF_PEER_SRC;
4788 		pdst = PF_PEER_DST;
4789 	} else {
4790 		src = &(*state)->dst;
4791 		dst = &(*state)->src;
4792 		psrc = PF_PEER_DST;
4793 		pdst = PF_PEER_SRC;
4794 	}
4795 
4796 	if (th->th_flags & TH_SYN)
4797 		if (src->state < TCPS_SYN_SENT)
4798 			pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
4799 	if (th->th_flags & TH_FIN)
4800 		if (src->state < TCPS_CLOSING)
4801 			pf_set_protostate(*state, psrc, TCPS_CLOSING);
4802 	if (th->th_flags & TH_ACK) {
4803 		if (dst->state == TCPS_SYN_SENT) {
4804 			pf_set_protostate(*state, pdst, TCPS_ESTABLISHED);
4805 			if (src->state == TCPS_ESTABLISHED &&
4806 			    (*state)->src_node != NULL &&
4807 			    pf_src_connlimit(state)) {
4808 				REASON_SET(reason, PFRES_SRCLIMIT);
4809 				return (PF_DROP);
4810 			}
4811 		} else if (dst->state == TCPS_CLOSING) {
4812 			pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2);
4813 		} else if (src->state == TCPS_SYN_SENT &&
4814 		    dst->state < TCPS_SYN_SENT) {
4815 			/*
4816 			 * Handle a special sloppy case where we only see one
4817 			 * half of the connection. If there is a ACK after
4818 			 * the initial SYN without ever seeing a packet from
4819 			 * the destination, set the connection to established.
4820 			 */
4821 			pf_set_protostate(*state, PF_PEER_BOTH,
4822 			    TCPS_ESTABLISHED);
4823 			dst->state = src->state = TCPS_ESTABLISHED;
4824 			if ((*state)->src_node != NULL &&
4825 			    pf_src_connlimit(state)) {
4826 				REASON_SET(reason, PFRES_SRCLIMIT);
4827 				return (PF_DROP);
4828 			}
4829 		} else if (src->state == TCPS_CLOSING &&
4830 		    dst->state == TCPS_ESTABLISHED &&
4831 		    dst->seqlo == 0) {
4832 			/*
4833 			 * Handle the closing of half connections where we
4834 			 * don't see the full bidirectional FIN/ACK+ACK
4835 			 * handshake.
4836 			 */
4837 			pf_set_protostate(*state, pdst, TCPS_CLOSING);
4838 		}
4839 	}
4840 	if (th->th_flags & TH_RST)
4841 		pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
4842 
4843 	/* update expire time */
4844 	(*state)->expire = time_uptime;
4845 	if (src->state >= TCPS_FIN_WAIT_2 &&
4846 	    dst->state >= TCPS_FIN_WAIT_2)
4847 		(*state)->timeout = PFTM_TCP_CLOSED;
4848 	else if (src->state >= TCPS_CLOSING &&
4849 	    dst->state >= TCPS_CLOSING)
4850 		(*state)->timeout = PFTM_TCP_FIN_WAIT;
4851 	else if (src->state < TCPS_ESTABLISHED ||
4852 	    dst->state < TCPS_ESTABLISHED)
4853 		(*state)->timeout = PFTM_TCP_OPENING;
4854 	else if (src->state >= TCPS_CLOSING ||
4855 	    dst->state >= TCPS_CLOSING)
4856 		(*state)->timeout = PFTM_TCP_CLOSING;
4857 	else
4858 		(*state)->timeout = PFTM_TCP_ESTABLISHED;
4859 
4860 	return (PF_PASS);
4861 }
4862 
4863 static int
pf_synproxy(struct pf_pdesc * pd,struct pf_kstate ** state,u_short * reason)4864 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason)
4865 {
4866 	struct pf_state_key	*sk = (*state)->key[pd->didx];
4867 	struct tcphdr		*th = &pd->hdr.tcp;
4868 
4869 	if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4870 		if (pd->dir != (*state)->direction) {
4871 			REASON_SET(reason, PFRES_SYNPROXY);
4872 			return (PF_SYNPROXY_DROP);
4873 		}
4874 		if (th->th_flags & TH_SYN) {
4875 			if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4876 				REASON_SET(reason, PFRES_SYNPROXY);
4877 				return (PF_DROP);
4878 			}
4879 			pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
4880 			    pd->src, th->th_dport, th->th_sport,
4881 			    (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4882 			    TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0);
4883 			REASON_SET(reason, PFRES_SYNPROXY);
4884 			return (PF_SYNPROXY_DROP);
4885 		} else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK ||
4886 		    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4887 		    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4888 			REASON_SET(reason, PFRES_SYNPROXY);
4889 			return (PF_DROP);
4890 		} else if ((*state)->src_node != NULL &&
4891 		    pf_src_connlimit(state)) {
4892 			REASON_SET(reason, PFRES_SRCLIMIT);
4893 			return (PF_DROP);
4894 		} else
4895 			pf_set_protostate(*state, PF_PEER_SRC,
4896 			    PF_TCPS_PROXY_DST);
4897 	}
4898 	if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4899 		if (pd->dir == (*state)->direction) {
4900 			if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4901 			    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4902 			    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4903 				REASON_SET(reason, PFRES_SYNPROXY);
4904 				return (PF_DROP);
4905 			}
4906 			(*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4907 			if ((*state)->dst.seqhi == 1)
4908 				(*state)->dst.seqhi = htonl(arc4random());
4909 			pf_send_tcp((*state)->rule.ptr, pd->af,
4910 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
4911 			    sk->port[pd->sidx], sk->port[pd->didx],
4912 			    (*state)->dst.seqhi, 0, TH_SYN, 0,
4913 			    (*state)->src.mss, 0, 0, (*state)->tag);
4914 			REASON_SET(reason, PFRES_SYNPROXY);
4915 			return (PF_SYNPROXY_DROP);
4916 		} else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4917 		    (TH_SYN|TH_ACK)) ||
4918 		    (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4919 			REASON_SET(reason, PFRES_SYNPROXY);
4920 			return (PF_DROP);
4921 		} else {
4922 			(*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4923 			(*state)->dst.seqlo = ntohl(th->th_seq);
4924 			pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
4925 			    pd->src, th->th_dport, th->th_sport,
4926 			    ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4927 			    TH_ACK, (*state)->src.max_win, 0, 0, 0,
4928 			    (*state)->tag);
4929 			pf_send_tcp((*state)->rule.ptr, pd->af,
4930 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
4931 			    sk->port[pd->sidx], sk->port[pd->didx],
4932 			    (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4933 			    TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0);
4934 			(*state)->src.seqdiff = (*state)->dst.seqhi -
4935 			    (*state)->src.seqlo;
4936 			(*state)->dst.seqdiff = (*state)->src.seqhi -
4937 			    (*state)->dst.seqlo;
4938 			(*state)->src.seqhi = (*state)->src.seqlo +
4939 			    (*state)->dst.max_win;
4940 			(*state)->dst.seqhi = (*state)->dst.seqlo +
4941 			    (*state)->src.max_win;
4942 			(*state)->src.wscale = (*state)->dst.wscale = 0;
4943 			pf_set_protostate(*state, PF_PEER_BOTH,
4944 			    TCPS_ESTABLISHED);
4945 			REASON_SET(reason, PFRES_SYNPROXY);
4946 			return (PF_SYNPROXY_DROP);
4947 		}
4948 	}
4949 
4950 	return (PF_PASS);
4951 }
4952 
4953 static int
pf_test_state_tcp(struct pf_kstate ** state,int direction,struct pfi_kkif * kif,struct mbuf * m,int off,void * h,struct pf_pdesc * pd,u_short * reason)4954 pf_test_state_tcp(struct pf_kstate **state, int direction, struct pfi_kkif *kif,
4955     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4956     u_short *reason)
4957 {
4958 	struct pf_state_key_cmp	 key;
4959 	struct tcphdr		*th = &pd->hdr.tcp;
4960 	int			 copyback = 0;
4961 	int			 action;
4962 	struct pf_state_peer	*src, *dst;
4963 
4964 	bzero(&key, sizeof(key));
4965 	key.af = pd->af;
4966 	key.proto = IPPROTO_TCP;
4967 	if (direction == PF_IN)	{	/* wire side, straight */
4968 		PF_ACPY(&key.addr[0], pd->src, key.af);
4969 		PF_ACPY(&key.addr[1], pd->dst, key.af);
4970 		key.port[0] = th->th_sport;
4971 		key.port[1] = th->th_dport;
4972 	} else {			/* stack side, reverse */
4973 		PF_ACPY(&key.addr[1], pd->src, key.af);
4974 		PF_ACPY(&key.addr[0], pd->dst, key.af);
4975 		key.port[1] = th->th_sport;
4976 		key.port[0] = th->th_dport;
4977 	}
4978 
4979 	STATE_LOOKUP(kif, &key, direction, *state, pd);
4980 
4981 	if (direction == (*state)->direction) {
4982 		src = &(*state)->src;
4983 		dst = &(*state)->dst;
4984 	} else {
4985 		src = &(*state)->dst;
4986 		dst = &(*state)->src;
4987 	}
4988 
4989 	if ((action = pf_synproxy(pd, state, reason)) != PF_PASS)
4990 		return (action);
4991 
4992 	if (dst->state >= TCPS_FIN_WAIT_2 &&
4993 	    src->state >= TCPS_FIN_WAIT_2 &&
4994 	    (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) ||
4995 	    ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK &&
4996 	    pf_syncookie_check(pd) && pd->dir == PF_IN))) {
4997 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
4998 			printf("pf: state reuse ");
4999 			pf_print_state(*state);
5000 			pf_print_flags(th->th_flags);
5001 			printf("\n");
5002 		}
5003 		/* XXX make sure it's the same direction ?? */
5004 		pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED);
5005 		pf_unlink_state(*state, PF_ENTER_LOCKED);
5006 		*state = NULL;
5007 		return (PF_DROP);
5008 	}
5009 
5010 	if ((*state)->state_flags & PFSTATE_SLOPPY) {
5011 		if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP)
5012 			return (PF_DROP);
5013 	} else {
5014 		if (pf_tcp_track_full(state, kif, m, off, pd, reason,
5015 		    &copyback) == PF_DROP)
5016 			return (PF_DROP);
5017 	}
5018 
5019 	/* translate source/destination address, if necessary */
5020 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5021 		struct pf_state_key *nk = (*state)->key[pd->didx];
5022 
5023 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
5024 		    nk->port[pd->sidx] != th->th_sport)
5025 			pf_change_ap(m, pd->src, &th->th_sport,
5026 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
5027 			    nk->port[pd->sidx], 0, pd->af);
5028 
5029 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
5030 		    nk->port[pd->didx] != th->th_dport)
5031 			pf_change_ap(m, pd->dst, &th->th_dport,
5032 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
5033 			    nk->port[pd->didx], 0, pd->af);
5034 		copyback = 1;
5035 	}
5036 
5037 	/* Copyback sequence modulation or stateful scrub changes if needed */
5038 	if (copyback)
5039 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
5040 
5041 	return (PF_PASS);
5042 }
5043 
5044 static int
pf_test_state_udp(struct pf_kstate ** state,int direction,struct pfi_kkif * kif,struct mbuf * m,int off,void * h,struct pf_pdesc * pd)5045 pf_test_state_udp(struct pf_kstate **state, int direction, struct pfi_kkif *kif,
5046     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
5047 {
5048 	struct pf_state_peer	*src, *dst;
5049 	struct pf_state_key_cmp	 key;
5050 	struct udphdr		*uh = &pd->hdr.udp;
5051 	uint8_t			 psrc, pdst;
5052 
5053 	bzero(&key, sizeof(key));
5054 	key.af = pd->af;
5055 	key.proto = IPPROTO_UDP;
5056 	if (direction == PF_IN)	{	/* wire side, straight */
5057 		PF_ACPY(&key.addr[0], pd->src, key.af);
5058 		PF_ACPY(&key.addr[1], pd->dst, key.af);
5059 		key.port[0] = uh->uh_sport;
5060 		key.port[1] = uh->uh_dport;
5061 	} else {			/* stack side, reverse */
5062 		PF_ACPY(&key.addr[1], pd->src, key.af);
5063 		PF_ACPY(&key.addr[0], pd->dst, key.af);
5064 		key.port[1] = uh->uh_sport;
5065 		key.port[0] = uh->uh_dport;
5066 	}
5067 
5068 	STATE_LOOKUP(kif, &key, direction, *state, pd);
5069 
5070 	if (direction == (*state)->direction) {
5071 		src = &(*state)->src;
5072 		dst = &(*state)->dst;
5073 		psrc = PF_PEER_SRC;
5074 		pdst = PF_PEER_DST;
5075 	} else {
5076 		src = &(*state)->dst;
5077 		dst = &(*state)->src;
5078 		psrc = PF_PEER_DST;
5079 		pdst = PF_PEER_SRC;
5080 	}
5081 
5082 	/* update states */
5083 	if (src->state < PFUDPS_SINGLE)
5084 		pf_set_protostate(*state, psrc, PFUDPS_SINGLE);
5085 	if (dst->state == PFUDPS_SINGLE)
5086 		pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE);
5087 
5088 	/* update expire time */
5089 	(*state)->expire = time_uptime;
5090 	if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
5091 		(*state)->timeout = PFTM_UDP_MULTIPLE;
5092 	else
5093 		(*state)->timeout = PFTM_UDP_SINGLE;
5094 
5095 	/* translate source/destination address, if necessary */
5096 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5097 		struct pf_state_key *nk = (*state)->key[pd->didx];
5098 
5099 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
5100 		    nk->port[pd->sidx] != uh->uh_sport)
5101 			pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
5102 			    &uh->uh_sum, &nk->addr[pd->sidx],
5103 			    nk->port[pd->sidx], 1, pd->af);
5104 
5105 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
5106 		    nk->port[pd->didx] != uh->uh_dport)
5107 			pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
5108 			    &uh->uh_sum, &nk->addr[pd->didx],
5109 			    nk->port[pd->didx], 1, pd->af);
5110 		m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
5111 	}
5112 
5113 	return (PF_PASS);
5114 }
5115 
5116 static int
pf_test_state_icmp(struct pf_kstate ** state,int direction,struct pfi_kkif * kif,struct mbuf * m,int off,void * h,struct pf_pdesc * pd,u_short * reason)5117 pf_test_state_icmp(struct pf_kstate **state, int direction, struct pfi_kkif *kif,
5118     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
5119 {
5120 	struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
5121 	u_int16_t	 icmpid = 0, *icmpsum;
5122 	u_int8_t	 icmptype, icmpcode;
5123 	int		 state_icmp = 0;
5124 	struct pf_state_key_cmp key;
5125 
5126 	bzero(&key, sizeof(key));
5127 	switch (pd->proto) {
5128 #ifdef INET
5129 	case IPPROTO_ICMP:
5130 		icmptype = pd->hdr.icmp.icmp_type;
5131 		icmpcode = pd->hdr.icmp.icmp_code;
5132 		icmpid = pd->hdr.icmp.icmp_id;
5133 		icmpsum = &pd->hdr.icmp.icmp_cksum;
5134 
5135 		if (icmptype == ICMP_UNREACH ||
5136 		    icmptype == ICMP_SOURCEQUENCH ||
5137 		    icmptype == ICMP_REDIRECT ||
5138 		    icmptype == ICMP_TIMXCEED ||
5139 		    icmptype == ICMP_PARAMPROB)
5140 			state_icmp++;
5141 		break;
5142 #endif /* INET */
5143 #ifdef INET6
5144 	case IPPROTO_ICMPV6:
5145 		icmptype = pd->hdr.icmp6.icmp6_type;
5146 		icmpcode = pd->hdr.icmp6.icmp6_code;
5147 		icmpid = pd->hdr.icmp6.icmp6_id;
5148 		icmpsum = &pd->hdr.icmp6.icmp6_cksum;
5149 
5150 		if (icmptype == ICMP6_DST_UNREACH ||
5151 		    icmptype == ICMP6_PACKET_TOO_BIG ||
5152 		    icmptype == ICMP6_TIME_EXCEEDED ||
5153 		    icmptype == ICMP6_PARAM_PROB)
5154 			state_icmp++;
5155 		break;
5156 #endif /* INET6 */
5157 	}
5158 
5159 	if (!state_icmp) {
5160 
5161 		/*
5162 		 * ICMP query/reply message not related to a TCP/UDP packet.
5163 		 * Search for an ICMP state.
5164 		 */
5165 		key.af = pd->af;
5166 		key.proto = pd->proto;
5167 		key.port[0] = key.port[1] = icmpid;
5168 		if (direction == PF_IN)	{	/* wire side, straight */
5169 			PF_ACPY(&key.addr[0], pd->src, key.af);
5170 			PF_ACPY(&key.addr[1], pd->dst, key.af);
5171 		} else {			/* stack side, reverse */
5172 			PF_ACPY(&key.addr[1], pd->src, key.af);
5173 			PF_ACPY(&key.addr[0], pd->dst, key.af);
5174 		}
5175 
5176 		STATE_LOOKUP(kif, &key, direction, *state, pd);
5177 
5178 		(*state)->expire = time_uptime;
5179 		(*state)->timeout = PFTM_ICMP_ERROR_REPLY;
5180 
5181 		/* translate source/destination address, if necessary */
5182 		if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5183 			struct pf_state_key *nk = (*state)->key[pd->didx];
5184 
5185 			switch (pd->af) {
5186 #ifdef INET
5187 			case AF_INET:
5188 				if (PF_ANEQ(pd->src,
5189 				    &nk->addr[pd->sidx], AF_INET))
5190 					pf_change_a(&saddr->v4.s_addr,
5191 					    pd->ip_sum,
5192 					    nk->addr[pd->sidx].v4.s_addr, 0);
5193 
5194 				if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
5195 				    AF_INET))
5196 					pf_change_a(&daddr->v4.s_addr,
5197 					    pd->ip_sum,
5198 					    nk->addr[pd->didx].v4.s_addr, 0);
5199 
5200 				if (nk->port[0] !=
5201 				    pd->hdr.icmp.icmp_id) {
5202 					pd->hdr.icmp.icmp_cksum =
5203 					    pf_cksum_fixup(
5204 					    pd->hdr.icmp.icmp_cksum, icmpid,
5205 					    nk->port[pd->sidx], 0);
5206 					pd->hdr.icmp.icmp_id =
5207 					    nk->port[pd->sidx];
5208 				}
5209 
5210 				m_copyback(m, off, ICMP_MINLEN,
5211 				    (caddr_t )&pd->hdr.icmp);
5212 				break;
5213 #endif /* INET */
5214 #ifdef INET6
5215 			case AF_INET6:
5216 				if (PF_ANEQ(pd->src,
5217 				    &nk->addr[pd->sidx], AF_INET6))
5218 					pf_change_a6(saddr,
5219 					    &pd->hdr.icmp6.icmp6_cksum,
5220 					    &nk->addr[pd->sidx], 0);
5221 
5222 				if (PF_ANEQ(pd->dst,
5223 				    &nk->addr[pd->didx], AF_INET6))
5224 					pf_change_a6(daddr,
5225 					    &pd->hdr.icmp6.icmp6_cksum,
5226 					    &nk->addr[pd->didx], 0);
5227 
5228 				m_copyback(m, off, sizeof(struct icmp6_hdr),
5229 				    (caddr_t )&pd->hdr.icmp6);
5230 				break;
5231 #endif /* INET6 */
5232 			}
5233 		}
5234 		return (PF_PASS);
5235 
5236 	} else {
5237 		/*
5238 		 * ICMP error message in response to a TCP/UDP packet.
5239 		 * Extract the inner TCP/UDP header and search for that state.
5240 		 */
5241 
5242 		struct pf_pdesc	pd2;
5243 		bzero(&pd2, sizeof pd2);
5244 #ifdef INET
5245 		struct ip	h2;
5246 #endif /* INET */
5247 #ifdef INET6
5248 		struct ip6_hdr	h2_6;
5249 		int		terminal = 0;
5250 #endif /* INET6 */
5251 		int		ipoff2 = 0;
5252 		int		off2 = 0;
5253 
5254 		pd2.af = pd->af;
5255 		/* Payload packet is from the opposite direction. */
5256 		pd2.sidx = (direction == PF_IN) ? 1 : 0;
5257 		pd2.didx = (direction == PF_IN) ? 0 : 1;
5258 		switch (pd->af) {
5259 #ifdef INET
5260 		case AF_INET:
5261 			/* offset of h2 in mbuf chain */
5262 			ipoff2 = off + ICMP_MINLEN;
5263 
5264 			if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
5265 			    NULL, reason, pd2.af)) {
5266 				DPFPRINTF(PF_DEBUG_MISC,
5267 				    ("pf: ICMP error message too short "
5268 				    "(ip)\n"));
5269 				return (PF_DROP);
5270 			}
5271 			/*
5272 			 * ICMP error messages don't refer to non-first
5273 			 * fragments
5274 			 */
5275 			if (h2.ip_off & htons(IP_OFFMASK)) {
5276 				REASON_SET(reason, PFRES_FRAG);
5277 				return (PF_DROP);
5278 			}
5279 
5280 			/* offset of protocol header that follows h2 */
5281 			off2 = ipoff2 + (h2.ip_hl << 2);
5282 
5283 			pd2.proto = h2.ip_p;
5284 			pd2.src = (struct pf_addr *)&h2.ip_src;
5285 			pd2.dst = (struct pf_addr *)&h2.ip_dst;
5286 			pd2.ip_sum = &h2.ip_sum;
5287 			break;
5288 #endif /* INET */
5289 #ifdef INET6
5290 		case AF_INET6:
5291 			ipoff2 = off + sizeof(struct icmp6_hdr);
5292 
5293 			if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
5294 			    NULL, reason, pd2.af)) {
5295 				DPFPRINTF(PF_DEBUG_MISC,
5296 				    ("pf: ICMP error message too short "
5297 				    "(ip6)\n"));
5298 				return (PF_DROP);
5299 			}
5300 			pd2.proto = h2_6.ip6_nxt;
5301 			pd2.src = (struct pf_addr *)&h2_6.ip6_src;
5302 			pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
5303 			pd2.ip_sum = NULL;
5304 			off2 = ipoff2 + sizeof(h2_6);
5305 			do {
5306 				switch (pd2.proto) {
5307 				case IPPROTO_FRAGMENT:
5308 					/*
5309 					 * ICMPv6 error messages for
5310 					 * non-first fragments
5311 					 */
5312 					REASON_SET(reason, PFRES_FRAG);
5313 					return (PF_DROP);
5314 				case IPPROTO_AH:
5315 				case IPPROTO_HOPOPTS:
5316 				case IPPROTO_ROUTING:
5317 				case IPPROTO_DSTOPTS: {
5318 					/* get next header and header length */
5319 					struct ip6_ext opt6;
5320 
5321 					if (!pf_pull_hdr(m, off2, &opt6,
5322 					    sizeof(opt6), NULL, reason,
5323 					    pd2.af)) {
5324 						DPFPRINTF(PF_DEBUG_MISC,
5325 						    ("pf: ICMPv6 short opt\n"));
5326 						return (PF_DROP);
5327 					}
5328 					if (pd2.proto == IPPROTO_AH)
5329 						off2 += (opt6.ip6e_len + 2) * 4;
5330 					else
5331 						off2 += (opt6.ip6e_len + 1) * 8;
5332 					pd2.proto = opt6.ip6e_nxt;
5333 					/* goto the next header */
5334 					break;
5335 				}
5336 				default:
5337 					terminal++;
5338 					break;
5339 				}
5340 			} while (!terminal);
5341 			break;
5342 #endif /* INET6 */
5343 		}
5344 
5345 		if (PF_ANEQ(pd->dst, pd2.src, pd->af)) {
5346 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
5347 				printf("pf: BAD ICMP %d:%d outer dst: ",
5348 				    icmptype, icmpcode);
5349 				pf_print_host(pd->src, 0, pd->af);
5350 				printf(" -> ");
5351 				pf_print_host(pd->dst, 0, pd->af);
5352 				printf(" inner src: ");
5353 				pf_print_host(pd2.src, 0, pd2.af);
5354 				printf(" -> ");
5355 				pf_print_host(pd2.dst, 0, pd2.af);
5356 				printf("\n");
5357 			}
5358 			REASON_SET(reason, PFRES_BADSTATE);
5359 			return (PF_DROP);
5360 		}
5361 
5362 		switch (pd2.proto) {
5363 		case IPPROTO_TCP: {
5364 			struct tcphdr		 th;
5365 			u_int32_t		 seq;
5366 			struct pf_state_peer	*src, *dst;
5367 			u_int8_t		 dws;
5368 			int			 copyback = 0;
5369 
5370 			/*
5371 			 * Only the first 8 bytes of the TCP header can be
5372 			 * expected. Don't access any TCP header fields after
5373 			 * th_seq, an ackskew test is not possible.
5374 			 */
5375 			if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
5376 			    pd2.af)) {
5377 				DPFPRINTF(PF_DEBUG_MISC,
5378 				    ("pf: ICMP error message too short "
5379 				    "(tcp)\n"));
5380 				return (PF_DROP);
5381 			}
5382 
5383 			key.af = pd2.af;
5384 			key.proto = IPPROTO_TCP;
5385 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5386 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5387 			key.port[pd2.sidx] = th.th_sport;
5388 			key.port[pd2.didx] = th.th_dport;
5389 
5390 			STATE_LOOKUP(kif, &key, direction, *state, pd);
5391 
5392 			if (direction == (*state)->direction) {
5393 				src = &(*state)->dst;
5394 				dst = &(*state)->src;
5395 			} else {
5396 				src = &(*state)->src;
5397 				dst = &(*state)->dst;
5398 			}
5399 
5400 			if (src->wscale && dst->wscale)
5401 				dws = dst->wscale & PF_WSCALE_MASK;
5402 			else
5403 				dws = 0;
5404 
5405 			/* Demodulate sequence number */
5406 			seq = ntohl(th.th_seq) - src->seqdiff;
5407 			if (src->seqdiff) {
5408 				pf_change_a(&th.th_seq, icmpsum,
5409 				    htonl(seq), 0);
5410 				copyback = 1;
5411 			}
5412 
5413 			if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
5414 			    (!SEQ_GEQ(src->seqhi, seq) ||
5415 			    !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
5416 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
5417 					printf("pf: BAD ICMP %d:%d ",
5418 					    icmptype, icmpcode);
5419 					pf_print_host(pd->src, 0, pd->af);
5420 					printf(" -> ");
5421 					pf_print_host(pd->dst, 0, pd->af);
5422 					printf(" state: ");
5423 					pf_print_state(*state);
5424 					printf(" seq=%u\n", seq);
5425 				}
5426 				REASON_SET(reason, PFRES_BADSTATE);
5427 				return (PF_DROP);
5428 			} else {
5429 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
5430 					printf("pf: OK ICMP %d:%d ",
5431 					    icmptype, icmpcode);
5432 					pf_print_host(pd->src, 0, pd->af);
5433 					printf(" -> ");
5434 					pf_print_host(pd->dst, 0, pd->af);
5435 					printf(" state: ");
5436 					pf_print_state(*state);
5437 					printf(" seq=%u\n", seq);
5438 				}
5439 			}
5440 
5441 			/* translate source/destination address, if necessary */
5442 			if ((*state)->key[PF_SK_WIRE] !=
5443 			    (*state)->key[PF_SK_STACK]) {
5444 				struct pf_state_key *nk =
5445 				    (*state)->key[pd->didx];
5446 
5447 				if (PF_ANEQ(pd2.src,
5448 				    &nk->addr[pd2.sidx], pd2.af) ||
5449 				    nk->port[pd2.sidx] != th.th_sport)
5450 					pf_change_icmp(pd2.src, &th.th_sport,
5451 					    daddr, &nk->addr[pd2.sidx],
5452 					    nk->port[pd2.sidx], NULL,
5453 					    pd2.ip_sum, icmpsum,
5454 					    pd->ip_sum, 0, pd2.af);
5455 
5456 				if (PF_ANEQ(pd2.dst,
5457 				    &nk->addr[pd2.didx], pd2.af) ||
5458 				    nk->port[pd2.didx] != th.th_dport)
5459 					pf_change_icmp(pd2.dst, &th.th_dport,
5460 					    saddr, &nk->addr[pd2.didx],
5461 					    nk->port[pd2.didx], NULL,
5462 					    pd2.ip_sum, icmpsum,
5463 					    pd->ip_sum, 0, pd2.af);
5464 				copyback = 1;
5465 			}
5466 
5467 			if (copyback) {
5468 				switch (pd2.af) {
5469 #ifdef INET
5470 				case AF_INET:
5471 					m_copyback(m, off, ICMP_MINLEN,
5472 					    (caddr_t )&pd->hdr.icmp);
5473 					m_copyback(m, ipoff2, sizeof(h2),
5474 					    (caddr_t )&h2);
5475 					break;
5476 #endif /* INET */
5477 #ifdef INET6
5478 				case AF_INET6:
5479 					m_copyback(m, off,
5480 					    sizeof(struct icmp6_hdr),
5481 					    (caddr_t )&pd->hdr.icmp6);
5482 					m_copyback(m, ipoff2, sizeof(h2_6),
5483 					    (caddr_t )&h2_6);
5484 					break;
5485 #endif /* INET6 */
5486 				}
5487 				m_copyback(m, off2, 8, (caddr_t)&th);
5488 			}
5489 
5490 			return (PF_PASS);
5491 			break;
5492 		}
5493 		case IPPROTO_UDP: {
5494 			struct udphdr		uh;
5495 
5496 			if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
5497 			    NULL, reason, pd2.af)) {
5498 				DPFPRINTF(PF_DEBUG_MISC,
5499 				    ("pf: ICMP error message too short "
5500 				    "(udp)\n"));
5501 				return (PF_DROP);
5502 			}
5503 
5504 			key.af = pd2.af;
5505 			key.proto = IPPROTO_UDP;
5506 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5507 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5508 			key.port[pd2.sidx] = uh.uh_sport;
5509 			key.port[pd2.didx] = uh.uh_dport;
5510 
5511 			STATE_LOOKUP(kif, &key, direction, *state, pd);
5512 
5513 			/* translate source/destination address, if necessary */
5514 			if ((*state)->key[PF_SK_WIRE] !=
5515 			    (*state)->key[PF_SK_STACK]) {
5516 				struct pf_state_key *nk =
5517 				    (*state)->key[pd->didx];
5518 
5519 				if (PF_ANEQ(pd2.src,
5520 				    &nk->addr[pd2.sidx], pd2.af) ||
5521 				    nk->port[pd2.sidx] != uh.uh_sport)
5522 					pf_change_icmp(pd2.src, &uh.uh_sport,
5523 					    daddr, &nk->addr[pd2.sidx],
5524 					    nk->port[pd2.sidx], &uh.uh_sum,
5525 					    pd2.ip_sum, icmpsum,
5526 					    pd->ip_sum, 1, pd2.af);
5527 
5528 				if (PF_ANEQ(pd2.dst,
5529 				    &nk->addr[pd2.didx], pd2.af) ||
5530 				    nk->port[pd2.didx] != uh.uh_dport)
5531 					pf_change_icmp(pd2.dst, &uh.uh_dport,
5532 					    saddr, &nk->addr[pd2.didx],
5533 					    nk->port[pd2.didx], &uh.uh_sum,
5534 					    pd2.ip_sum, icmpsum,
5535 					    pd->ip_sum, 1, pd2.af);
5536 
5537 				switch (pd2.af) {
5538 #ifdef INET
5539 				case AF_INET:
5540 					m_copyback(m, off, ICMP_MINLEN,
5541 					    (caddr_t )&pd->hdr.icmp);
5542 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5543 					break;
5544 #endif /* INET */
5545 #ifdef INET6
5546 				case AF_INET6:
5547 					m_copyback(m, off,
5548 					    sizeof(struct icmp6_hdr),
5549 					    (caddr_t )&pd->hdr.icmp6);
5550 					m_copyback(m, ipoff2, sizeof(h2_6),
5551 					    (caddr_t )&h2_6);
5552 					break;
5553 #endif /* INET6 */
5554 				}
5555 				m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
5556 			}
5557 			return (PF_PASS);
5558 			break;
5559 		}
5560 #ifdef INET
5561 		case IPPROTO_ICMP: {
5562 			struct icmp		iih;
5563 
5564 			if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
5565 			    NULL, reason, pd2.af)) {
5566 				DPFPRINTF(PF_DEBUG_MISC,
5567 				    ("pf: ICMP error message too short i"
5568 				    "(icmp)\n"));
5569 				return (PF_DROP);
5570 			}
5571 
5572 			key.af = pd2.af;
5573 			key.proto = IPPROTO_ICMP;
5574 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5575 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5576 			key.port[0] = key.port[1] = iih.icmp_id;
5577 
5578 			STATE_LOOKUP(kif, &key, direction, *state, pd);
5579 
5580 			/* translate source/destination address, if necessary */
5581 			if ((*state)->key[PF_SK_WIRE] !=
5582 			    (*state)->key[PF_SK_STACK]) {
5583 				struct pf_state_key *nk =
5584 				    (*state)->key[pd->didx];
5585 
5586 				if (PF_ANEQ(pd2.src,
5587 				    &nk->addr[pd2.sidx], pd2.af) ||
5588 				    nk->port[pd2.sidx] != iih.icmp_id)
5589 					pf_change_icmp(pd2.src, &iih.icmp_id,
5590 					    daddr, &nk->addr[pd2.sidx],
5591 					    nk->port[pd2.sidx], NULL,
5592 					    pd2.ip_sum, icmpsum,
5593 					    pd->ip_sum, 0, AF_INET);
5594 
5595 				if (PF_ANEQ(pd2.dst,
5596 				    &nk->addr[pd2.didx], pd2.af) ||
5597 				    nk->port[pd2.didx] != iih.icmp_id)
5598 					pf_change_icmp(pd2.dst, &iih.icmp_id,
5599 					    saddr, &nk->addr[pd2.didx],
5600 					    nk->port[pd2.didx], NULL,
5601 					    pd2.ip_sum, icmpsum,
5602 					    pd->ip_sum, 0, AF_INET);
5603 
5604 				m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
5605 				m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5606 				m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
5607 			}
5608 			return (PF_PASS);
5609 			break;
5610 		}
5611 #endif /* INET */
5612 #ifdef INET6
5613 		case IPPROTO_ICMPV6: {
5614 			struct icmp6_hdr	iih;
5615 
5616 			if (!pf_pull_hdr(m, off2, &iih,
5617 			    sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
5618 				DPFPRINTF(PF_DEBUG_MISC,
5619 				    ("pf: ICMP error message too short "
5620 				    "(icmp6)\n"));
5621 				return (PF_DROP);
5622 			}
5623 
5624 			key.af = pd2.af;
5625 			key.proto = IPPROTO_ICMPV6;
5626 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5627 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5628 			key.port[0] = key.port[1] = iih.icmp6_id;
5629 
5630 			STATE_LOOKUP(kif, &key, direction, *state, pd);
5631 
5632 			/* translate source/destination address, if necessary */
5633 			if ((*state)->key[PF_SK_WIRE] !=
5634 			    (*state)->key[PF_SK_STACK]) {
5635 				struct pf_state_key *nk =
5636 				    (*state)->key[pd->didx];
5637 
5638 				if (PF_ANEQ(pd2.src,
5639 				    &nk->addr[pd2.sidx], pd2.af) ||
5640 				    nk->port[pd2.sidx] != iih.icmp6_id)
5641 					pf_change_icmp(pd2.src, &iih.icmp6_id,
5642 					    daddr, &nk->addr[pd2.sidx],
5643 					    nk->port[pd2.sidx], NULL,
5644 					    pd2.ip_sum, icmpsum,
5645 					    pd->ip_sum, 0, AF_INET6);
5646 
5647 				if (PF_ANEQ(pd2.dst,
5648 				    &nk->addr[pd2.didx], pd2.af) ||
5649 				    nk->port[pd2.didx] != iih.icmp6_id)
5650 					pf_change_icmp(pd2.dst, &iih.icmp6_id,
5651 					    saddr, &nk->addr[pd2.didx],
5652 					    nk->port[pd2.didx], NULL,
5653 					    pd2.ip_sum, icmpsum,
5654 					    pd->ip_sum, 0, AF_INET6);
5655 
5656 				m_copyback(m, off, sizeof(struct icmp6_hdr),
5657 				    (caddr_t)&pd->hdr.icmp6);
5658 				m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
5659 				m_copyback(m, off2, sizeof(struct icmp6_hdr),
5660 				    (caddr_t)&iih);
5661 			}
5662 			return (PF_PASS);
5663 			break;
5664 		}
5665 #endif /* INET6 */
5666 		default: {
5667 			key.af = pd2.af;
5668 			key.proto = pd2.proto;
5669 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5670 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5671 			key.port[0] = key.port[1] = 0;
5672 
5673 			STATE_LOOKUP(kif, &key, direction, *state, pd);
5674 
5675 			/* translate source/destination address, if necessary */
5676 			if ((*state)->key[PF_SK_WIRE] !=
5677 			    (*state)->key[PF_SK_STACK]) {
5678 				struct pf_state_key *nk =
5679 				    (*state)->key[pd->didx];
5680 
5681 				if (PF_ANEQ(pd2.src,
5682 				    &nk->addr[pd2.sidx], pd2.af))
5683 					pf_change_icmp(pd2.src, NULL, daddr,
5684 					    &nk->addr[pd2.sidx], 0, NULL,
5685 					    pd2.ip_sum, icmpsum,
5686 					    pd->ip_sum, 0, pd2.af);
5687 
5688 				if (PF_ANEQ(pd2.dst,
5689 				    &nk->addr[pd2.didx], pd2.af))
5690 					pf_change_icmp(pd2.dst, NULL, saddr,
5691 					    &nk->addr[pd2.didx], 0, NULL,
5692 					    pd2.ip_sum, icmpsum,
5693 					    pd->ip_sum, 0, pd2.af);
5694 
5695 				switch (pd2.af) {
5696 #ifdef INET
5697 				case AF_INET:
5698 					m_copyback(m, off, ICMP_MINLEN,
5699 					    (caddr_t)&pd->hdr.icmp);
5700 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5701 					break;
5702 #endif /* INET */
5703 #ifdef INET6
5704 				case AF_INET6:
5705 					m_copyback(m, off,
5706 					    sizeof(struct icmp6_hdr),
5707 					    (caddr_t )&pd->hdr.icmp6);
5708 					m_copyback(m, ipoff2, sizeof(h2_6),
5709 					    (caddr_t )&h2_6);
5710 					break;
5711 #endif /* INET6 */
5712 				}
5713 			}
5714 			return (PF_PASS);
5715 			break;
5716 		}
5717 		}
5718 	}
5719 }
5720 
5721 static int
pf_test_state_other(struct pf_kstate ** state,int direction,struct pfi_kkif * kif,struct mbuf * m,struct pf_pdesc * pd)5722 pf_test_state_other(struct pf_kstate **state, int direction, struct pfi_kkif *kif,
5723     struct mbuf *m, struct pf_pdesc *pd)
5724 {
5725 	struct pf_state_peer	*src, *dst;
5726 	struct pf_state_key_cmp	 key;
5727 	uint8_t			 psrc, pdst;
5728 
5729 	bzero(&key, sizeof(key));
5730 	key.af = pd->af;
5731 	key.proto = pd->proto;
5732 	if (direction == PF_IN)	{
5733 		PF_ACPY(&key.addr[0], pd->src, key.af);
5734 		PF_ACPY(&key.addr[1], pd->dst, key.af);
5735 		key.port[0] = key.port[1] = 0;
5736 	} else {
5737 		PF_ACPY(&key.addr[1], pd->src, key.af);
5738 		PF_ACPY(&key.addr[0], pd->dst, key.af);
5739 		key.port[1] = key.port[0] = 0;
5740 	}
5741 
5742 	STATE_LOOKUP(kif, &key, direction, *state, pd);
5743 
5744 	if (direction == (*state)->direction) {
5745 		src = &(*state)->src;
5746 		dst = &(*state)->dst;
5747 		psrc = PF_PEER_SRC;
5748 		pdst = PF_PEER_DST;
5749 	} else {
5750 		src = &(*state)->dst;
5751 		dst = &(*state)->src;
5752 		psrc = PF_PEER_DST;
5753 		pdst = PF_PEER_SRC;
5754 	}
5755 
5756 	/* update states */
5757 	if (src->state < PFOTHERS_SINGLE)
5758 		pf_set_protostate(*state, psrc, PFOTHERS_SINGLE);
5759 	if (dst->state == PFOTHERS_SINGLE)
5760 		pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE);
5761 
5762 	/* update expire time */
5763 	(*state)->expire = time_uptime;
5764 	if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5765 		(*state)->timeout = PFTM_OTHER_MULTIPLE;
5766 	else
5767 		(*state)->timeout = PFTM_OTHER_SINGLE;
5768 
5769 	/* translate source/destination address, if necessary */
5770 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5771 		struct pf_state_key *nk = (*state)->key[pd->didx];
5772 
5773 		KASSERT(nk, ("%s: nk is null", __func__));
5774 		KASSERT(pd, ("%s: pd is null", __func__));
5775 		KASSERT(pd->src, ("%s: pd->src is null", __func__));
5776 		KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5777 		switch (pd->af) {
5778 #ifdef INET
5779 		case AF_INET:
5780 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5781 				pf_change_a(&pd->src->v4.s_addr,
5782 				    pd->ip_sum,
5783 				    nk->addr[pd->sidx].v4.s_addr,
5784 				    0);
5785 
5786 
5787 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5788 				pf_change_a(&pd->dst->v4.s_addr,
5789 				    pd->ip_sum,
5790 				    nk->addr[pd->didx].v4.s_addr,
5791 				    0);
5792 
5793 			break;
5794 #endif /* INET */
5795 #ifdef INET6
5796 		case AF_INET6:
5797 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5798 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5799 
5800 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5801 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5802 #endif /* INET6 */
5803 		}
5804 	}
5805 	return (PF_PASS);
5806 }
5807 
5808 /*
5809  * ipoff and off are measured from the start of the mbuf chain.
5810  * h must be at "ipoff" on the mbuf chain.
5811  */
5812 void *
pf_pull_hdr(struct mbuf * m,int off,void * p,int len,u_short * actionp,u_short * reasonp,sa_family_t af)5813 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5814     u_short *actionp, u_short *reasonp, sa_family_t af)
5815 {
5816 	switch (af) {
5817 #ifdef INET
5818 	case AF_INET: {
5819 		struct ip	*h = mtod(m, struct ip *);
5820 		u_int16_t	 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5821 
5822 		if (fragoff) {
5823 			if (fragoff >= len)
5824 				ACTION_SET(actionp, PF_PASS);
5825 			else {
5826 				ACTION_SET(actionp, PF_DROP);
5827 				REASON_SET(reasonp, PFRES_FRAG);
5828 			}
5829 			return (NULL);
5830 		}
5831 		if (m->m_pkthdr.len < off + len ||
5832 		    ntohs(h->ip_len) < off + len) {
5833 			ACTION_SET(actionp, PF_DROP);
5834 			REASON_SET(reasonp, PFRES_SHORT);
5835 			return (NULL);
5836 		}
5837 		break;
5838 	}
5839 #endif /* INET */
5840 #ifdef INET6
5841 	case AF_INET6: {
5842 		struct ip6_hdr	*h = mtod(m, struct ip6_hdr *);
5843 
5844 		if (m->m_pkthdr.len < off + len ||
5845 		    (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5846 		    (unsigned)(off + len)) {
5847 			ACTION_SET(actionp, PF_DROP);
5848 			REASON_SET(reasonp, PFRES_SHORT);
5849 			return (NULL);
5850 		}
5851 		break;
5852 	}
5853 #endif /* INET6 */
5854 	}
5855 	m_copydata(m, off, len, p);
5856 	return (p);
5857 }
5858 
5859 #ifdef RADIX_MPATH
5860 static int
pf_routable_oldmpath(struct pf_addr * addr,sa_family_t af,struct pfi_kif * kif,int rtableid)5861 pf_routable_oldmpath(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5862     int rtableid)
5863 {
5864 	struct radix_node_head	*rnh;
5865 	struct sockaddr_in	*dst;
5866 	int			 ret = 1;
5867 	int			 check_mpath;
5868 #ifdef INET6
5869 	struct sockaddr_in6	*dst6;
5870 	struct route_in6	 ro;
5871 #else
5872 	struct route		 ro;
5873 #endif
5874 	struct radix_node	*rn;
5875 	struct rtentry		*rt;
5876 	struct ifnet		*ifp;
5877 
5878 	check_mpath = 0;
5879 	/* XXX: stick to table 0 for now */
5880 	rnh = rt_tables_get_rnh(0, af);
5881 	if (rnh != NULL && rn_mpath_capable(rnh))
5882 		check_mpath = 1;
5883 	bzero(&ro, sizeof(ro));
5884 	switch (af) {
5885 	case AF_INET:
5886 		dst = satosin(&ro.ro_dst);
5887 		dst->sin_family = AF_INET;
5888 		dst->sin_len = sizeof(*dst);
5889 		dst->sin_addr = addr->v4;
5890 		break;
5891 #ifdef INET6
5892 	case AF_INET6:
5893 		/*
5894 		 * Skip check for addresses with embedded interface scope,
5895 		 * as they would always match anyway.
5896 		 */
5897 		if (IN6_IS_SCOPE_EMBED(&addr->v6))
5898 			goto out;
5899 		dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5900 		dst6->sin6_family = AF_INET6;
5901 		dst6->sin6_len = sizeof(*dst6);
5902 		dst6->sin6_addr = addr->v6;
5903 		break;
5904 #endif /* INET6 */
5905 	default:
5906 		return (0);
5907 	}
5908 
5909 	/* Skip checks for ipsec interfaces */
5910 	if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5911 		goto out;
5912 
5913 	switch (af) {
5914 #ifdef INET6
5915 	case AF_INET6:
5916 		in6_rtalloc_ign(&ro, 0, rtableid);
5917 		break;
5918 #endif
5919 #ifdef INET
5920 	case AF_INET:
5921 		in_rtalloc_ign((struct route *)&ro, 0, rtableid);
5922 		break;
5923 #endif
5924 	}
5925 
5926 	if (ro.ro_rt != NULL) {
5927 		/* No interface given, this is a no-route check */
5928 		if (kif == NULL)
5929 			goto out;
5930 
5931 		if (kif->pfik_ifp == NULL) {
5932 			ret = 0;
5933 			goto out;
5934 		}
5935 
5936 		/* Perform uRPF check if passed input interface */
5937 		ret = 0;
5938 		rn = (struct radix_node *)ro.ro_rt;
5939 		do {
5940 			rt = (struct rtentry *)rn;
5941 			ifp = rt->rt_ifp;
5942 
5943 			if (kif->pfik_ifp == ifp)
5944 				ret = 1;
5945 			rn = rn_mpath_next(rn);
5946 		} while (check_mpath == 1 && rn != NULL && ret == 0);
5947 	} else
5948 		ret = 0;
5949 out:
5950 	if (ro.ro_rt != NULL)
5951 		RTFREE(ro.ro_rt);
5952 	return (ret);
5953 }
5954 #endif
5955 
5956 int
pf_routable(struct pf_addr * addr,sa_family_t af,struct pfi_kkif * kif,int rtableid)5957 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif,
5958     int rtableid)
5959 {
5960 #ifdef INET
5961 	struct nhop4_basic	nh4;
5962 #endif
5963 #ifdef INET6
5964 	struct nhop6_basic	nh6;
5965 #endif
5966 	struct ifnet		*ifp;
5967 #ifdef RADIX_MPATH
5968 	struct radix_node_head	*rnh;
5969 
5970 	/* XXX: stick to table 0 for now */
5971 	rnh = rt_tables_get_rnh(0, af);
5972 	if (rnh != NULL && rn_mpath_capable(rnh))
5973 		return (pf_routable_oldmpath(addr, af, kif, rtableid));
5974 #endif
5975 	/*
5976 	 * Skip check for addresses with embedded interface scope,
5977 	 * as they would always match anyway.
5978 	 */
5979 	if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
5980 		return (1);
5981 
5982 	if (af != AF_INET && af != AF_INET6)
5983 		return (0);
5984 
5985 	/* Skip checks for ipsec interfaces */
5986 	if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5987 		return (1);
5988 
5989 	ifp = NULL;
5990 
5991 	switch (af) {
5992 #ifdef INET6
5993 	case AF_INET6:
5994 		if (fib6_lookup_nh_basic(rtableid, &addr->v6, 0, 0, 0, &nh6)!=0)
5995 			return (0);
5996 		ifp = nh6.nh_ifp;
5997 		break;
5998 #endif
5999 #ifdef INET
6000 	case AF_INET:
6001 		if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) != 0)
6002 			return (0);
6003 		ifp = nh4.nh_ifp;
6004 		break;
6005 #endif
6006 	}
6007 
6008 	/* No interface given, this is a no-route check */
6009 	if (kif == NULL)
6010 		return (1);
6011 
6012 	if (kif->pfik_ifp == NULL)
6013 		return (0);
6014 
6015 	/* Perform uRPF check if passed input interface */
6016 	if (kif->pfik_ifp == ifp)
6017 		return (1);
6018 	return (0);
6019 }
6020 
6021 #ifdef INET
6022 static void
pf_route(struct mbuf ** m,struct pf_krule * r,int dir,struct ifnet * oifp,struct pf_kstate * s,struct pf_pdesc * pd,struct inpcb * inp)6023 pf_route(struct mbuf **m, struct pf_krule *r, int dir, struct ifnet *oifp,
6024     struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp)
6025 {
6026 	struct mbuf		*m0, *m1;
6027 	struct sockaddr_in	dst;
6028 	struct ip		*ip;
6029 	struct ifnet		*ifp = NULL;
6030 	struct pf_addr		 naddr;
6031 	struct pf_ksrc_node	*sn = NULL;
6032 	int			 error = 0;
6033 	uint16_t		 ip_len, ip_off;
6034 
6035 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
6036 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
6037 	    __func__));
6038 
6039 	if ((pd->pf_mtag == NULL &&
6040 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
6041 	    pd->pf_mtag->routed++ > 3) {
6042 		m0 = *m;
6043 		*m = NULL;
6044 		goto bad_locked;
6045 	}
6046 
6047 	if (r->rt == PF_DUPTO) {
6048 		if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
6049 			if (s)
6050 				PF_STATE_UNLOCK(s);
6051 			return;
6052 		}
6053 	} else {
6054 		if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
6055 			if (s)
6056 				PF_STATE_UNLOCK(s);
6057 			return;
6058 		}
6059 		m0 = *m;
6060 	}
6061 
6062 	ip = mtod(m0, struct ip *);
6063 
6064 	bzero(&dst, sizeof(dst));
6065 	dst.sin_family = AF_INET;
6066 	dst.sin_len = sizeof(dst);
6067 	dst.sin_addr = ip->ip_dst;
6068 
6069 	bzero(&naddr, sizeof(naddr));
6070 
6071 	if (TAILQ_EMPTY(&r->rpool.list)) {
6072 		DPFPRINTF(PF_DEBUG_URGENT,
6073 		    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
6074 		goto bad_locked;
6075 	}
6076 	if (s == NULL) {
6077 		pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
6078 		    &naddr, NULL, &sn);
6079 		if (!PF_AZERO(&naddr, AF_INET))
6080 			dst.sin_addr.s_addr = naddr.v4.s_addr;
6081 		ifp = r->rpool.cur->kif ?
6082 		    r->rpool.cur->kif->pfik_ifp : NULL;
6083 	} else {
6084 		if (!PF_AZERO(&s->rt_addr, AF_INET))
6085 			dst.sin_addr.s_addr =
6086 			    s->rt_addr.v4.s_addr;
6087 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
6088 		PF_STATE_UNLOCK(s);
6089 	}
6090 	/* If pfsync'd */
6091 	if (ifp == NULL)
6092 		ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
6093 	if (ifp == NULL)
6094 		goto bad;
6095 
6096 	if (dir == PF_IN) {
6097 		if (pf_test(PF_OUT, 0, ifp, &m0, inp) != PF_PASS)
6098 			goto bad;
6099 		else if (m0 == NULL)
6100 			goto done;
6101 		if (m0->m_len < sizeof(struct ip)) {
6102 			DPFPRINTF(PF_DEBUG_URGENT,
6103 			    ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
6104 			goto bad;
6105 		}
6106 		ip = mtod(m0, struct ip *);
6107 	}
6108 
6109 	if (ifp->if_flags & IFF_LOOPBACK)
6110 		m0->m_flags |= M_SKIP_FIREWALL;
6111 
6112 	ip_len = ntohs(ip->ip_len);
6113 	ip_off = ntohs(ip->ip_off);
6114 
6115 	/* Copied from FreeBSD 10.0-CURRENT ip_output. */
6116 	m0->m_pkthdr.csum_flags |= CSUM_IP;
6117 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
6118 		in_delayed_cksum(m0);
6119 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
6120 	}
6121 #if defined(SCTP) || defined(SCTP_SUPPORT)
6122 	if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
6123 		sctp_delayed_cksum(m0, (uint32_t)(ip->ip_hl << 2));
6124 		m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
6125 	}
6126 #endif
6127 
6128 	/*
6129 	 * If small enough for interface, or the interface will take
6130 	 * care of the fragmentation for us, we can just send directly.
6131 	 */
6132 	if (ip_len <= ifp->if_mtu ||
6133 	    (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
6134 		ip->ip_sum = 0;
6135 		if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
6136 			ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
6137 			m0->m_pkthdr.csum_flags &= ~CSUM_IP;
6138 		}
6139 		m_clrprotoflags(m0);	/* Avoid confusing lower layers. */
6140 		error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
6141 		goto done;
6142 	}
6143 
6144 	/* Balk when DF bit is set or the interface didn't support TSO. */
6145 	if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
6146 		error = EMSGSIZE;
6147 		KMOD_IPSTAT_INC(ips_cantfrag);
6148 		if (r->rt != PF_DUPTO) {
6149 			if (s && pd->nat_rule != NULL)
6150 				PACKET_UNDO_NAT(m0, pd,
6151 				    (ip->ip_hl << 2) + (ip_off & IP_OFFMASK),
6152 				    s, dir);
6153 
6154 			icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
6155 			    ifp->if_mtu);
6156 			goto done;
6157 		} else
6158 			goto bad;
6159 	}
6160 
6161 	error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
6162 	if (error)
6163 		goto bad;
6164 
6165 	for (; m0; m0 = m1) {
6166 		m1 = m0->m_nextpkt;
6167 		m0->m_nextpkt = NULL;
6168 		if (error == 0) {
6169 			m_clrprotoflags(m0);
6170 			error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
6171 		} else
6172 			m_freem(m0);
6173 	}
6174 
6175 	if (error == 0)
6176 		KMOD_IPSTAT_INC(ips_fragmented);
6177 
6178 done:
6179 	if (r->rt != PF_DUPTO)
6180 		*m = NULL;
6181 	return;
6182 
6183 bad_locked:
6184 	if (s)
6185 		PF_STATE_UNLOCK(s);
6186 bad:
6187 	m_freem(m0);
6188 	goto done;
6189 }
6190 #endif /* INET */
6191 
6192 #ifdef INET6
6193 static void
pf_route6(struct mbuf ** m,struct pf_krule * r,int dir,struct ifnet * oifp,struct pf_kstate * s,struct pf_pdesc * pd,struct inpcb * inp)6194 pf_route6(struct mbuf **m, struct pf_krule *r, int dir, struct ifnet *oifp,
6195     struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp)
6196 {
6197 	struct mbuf		*m0;
6198 	struct sockaddr_in6	dst;
6199 	struct ip6_hdr		*ip6;
6200 	struct ifnet		*ifp = NULL;
6201 	struct pf_addr		 naddr;
6202 	struct pf_ksrc_node	*sn = NULL;
6203 
6204 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
6205 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
6206 	    __func__));
6207 
6208 	if ((pd->pf_mtag == NULL &&
6209 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
6210 	    pd->pf_mtag->routed++ > 3) {
6211 		m0 = *m;
6212 		*m = NULL;
6213 		goto bad_locked;
6214 	}
6215 
6216 	if (r->rt == PF_DUPTO) {
6217 		if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
6218 			if (s)
6219 				PF_STATE_UNLOCK(s);
6220 			return;
6221 		}
6222 	} else {
6223 		if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
6224 			if (s)
6225 				PF_STATE_UNLOCK(s);
6226 			return;
6227 		}
6228 		m0 = *m;
6229 	}
6230 
6231 	ip6 = mtod(m0, struct ip6_hdr *);
6232 
6233 	bzero(&dst, sizeof(dst));
6234 	dst.sin6_family = AF_INET6;
6235 	dst.sin6_len = sizeof(dst);
6236 	dst.sin6_addr = ip6->ip6_dst;
6237 
6238 	bzero(&naddr, sizeof(naddr));
6239 
6240 	if (TAILQ_EMPTY(&r->rpool.list)) {
6241 		DPFPRINTF(PF_DEBUG_URGENT,
6242 		    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
6243 		goto bad_locked;
6244 	}
6245 	if (s == NULL) {
6246 		pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
6247 		    &naddr, NULL, &sn);
6248 		if (!PF_AZERO(&naddr, AF_INET6))
6249 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
6250 			    &naddr, AF_INET6);
6251 		ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
6252 	} else {
6253 		if (!PF_AZERO(&s->rt_addr, AF_INET6))
6254 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
6255 			    &s->rt_addr, AF_INET6);
6256 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
6257 	}
6258 
6259 	if (s)
6260 		PF_STATE_UNLOCK(s);
6261 
6262 	/* If pfsync'd */
6263 	if (ifp == NULL)
6264 		ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
6265 	if (ifp == NULL)
6266 		goto bad;
6267 
6268 	if (dir == PF_IN) {
6269 		if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp) != PF_PASS)
6270 			goto bad;
6271 		else if (m0 == NULL)
6272 			goto done;
6273 		if (m0->m_len < sizeof(struct ip6_hdr)) {
6274 			DPFPRINTF(PF_DEBUG_URGENT,
6275 			    ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
6276 			    __func__));
6277 			goto bad;
6278 		}
6279 		ip6 = mtod(m0, struct ip6_hdr *);
6280 	}
6281 
6282 	if (ifp->if_flags & IFF_LOOPBACK)
6283 		m0->m_flags |= M_SKIP_FIREWALL;
6284 
6285 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
6286 	    ~ifp->if_hwassist) {
6287 		uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
6288 		in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
6289 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
6290 	}
6291 
6292 	/*
6293 	 * If the packet is too large for the outgoing interface,
6294 	 * send back an icmp6 error.
6295 	 */
6296 	if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
6297 		dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
6298 	if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
6299 		nd6_output_ifp(ifp, ifp, m0, &dst, NULL);
6300 	else {
6301 		in6_ifstat_inc(ifp, ifs6_in_toobig);
6302 		if (r->rt != PF_DUPTO) {
6303 			if (s && pd->nat_rule != NULL)
6304 				PACKET_UNDO_NAT(m0, pd,
6305 				    ((caddr_t)ip6 - m0->m_data) +
6306 				    sizeof(struct ip6_hdr), s, dir);
6307 
6308 			icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
6309 		} else
6310 			goto bad;
6311 	}
6312 
6313 done:
6314 	if (r->rt != PF_DUPTO)
6315 		*m = NULL;
6316 	return;
6317 
6318 bad_locked:
6319 	if (s)
6320 		PF_STATE_UNLOCK(s);
6321 bad:
6322 	m_freem(m0);
6323 	goto done;
6324 }
6325 #endif /* INET6 */
6326 
6327 /*
6328  * FreeBSD supports cksum offloads for the following drivers.
6329  *  em(4), fxp(4), lge(4), ndis(4), nge(4), re(4), ti(4), txp(4), xl(4)
6330  *
6331  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
6332  *  network driver performed cksum including pseudo header, need to verify
6333  *   csum_data
6334  * CSUM_DATA_VALID :
6335  *  network driver performed cksum, needs to additional pseudo header
6336  *  cksum computation with partial csum_data(i.e. lack of H/W support for
6337  *  pseudo header, for instance hme(4), sk(4) and possibly gem(4))
6338  *
6339  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
6340  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
6341  * TCP/UDP layer.
6342  * Also, set csum_data to 0xffff to force cksum validation.
6343  */
6344 static int
pf_check_proto_cksum(struct mbuf * m,int off,int len,u_int8_t p,sa_family_t af)6345 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
6346 {
6347 	u_int16_t sum = 0;
6348 	int hw_assist = 0;
6349 	struct ip *ip;
6350 
6351 	if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
6352 		return (1);
6353 	if (m->m_pkthdr.len < off + len)
6354 		return (1);
6355 
6356 	switch (p) {
6357 	case IPPROTO_TCP:
6358 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
6359 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
6360 				sum = m->m_pkthdr.csum_data;
6361 			} else {
6362 				ip = mtod(m, struct ip *);
6363 				sum = in_pseudo(ip->ip_src.s_addr,
6364 				ip->ip_dst.s_addr, htonl((u_short)len +
6365 				m->m_pkthdr.csum_data + IPPROTO_TCP));
6366 			}
6367 			sum ^= 0xffff;
6368 			++hw_assist;
6369 		}
6370 		break;
6371 	case IPPROTO_UDP:
6372 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
6373 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
6374 				sum = m->m_pkthdr.csum_data;
6375 			} else {
6376 				ip = mtod(m, struct ip *);
6377 				sum = in_pseudo(ip->ip_src.s_addr,
6378 				ip->ip_dst.s_addr, htonl((u_short)len +
6379 				m->m_pkthdr.csum_data + IPPROTO_UDP));
6380 			}
6381 			sum ^= 0xffff;
6382 			++hw_assist;
6383 		}
6384 		break;
6385 	case IPPROTO_ICMP:
6386 #ifdef INET6
6387 	case IPPROTO_ICMPV6:
6388 #endif /* INET6 */
6389 		break;
6390 	default:
6391 		return (1);
6392 	}
6393 
6394 	if (!hw_assist) {
6395 		switch (af) {
6396 		case AF_INET:
6397 			if (p == IPPROTO_ICMP) {
6398 				if (m->m_len < off)
6399 					return (1);
6400 				m->m_data += off;
6401 				m->m_len -= off;
6402 				sum = in_cksum(m, len);
6403 				m->m_data -= off;
6404 				m->m_len += off;
6405 			} else {
6406 				if (m->m_len < sizeof(struct ip))
6407 					return (1);
6408 				sum = in4_cksum(m, p, off, len);
6409 			}
6410 			break;
6411 #ifdef INET6
6412 		case AF_INET6:
6413 			if (m->m_len < sizeof(struct ip6_hdr))
6414 				return (1);
6415 			sum = in6_cksum(m, p, off, len);
6416 			break;
6417 #endif /* INET6 */
6418 		default:
6419 			return (1);
6420 		}
6421 	}
6422 	if (sum) {
6423 		switch (p) {
6424 		case IPPROTO_TCP:
6425 		    {
6426 			KMOD_TCPSTAT_INC(tcps_rcvbadsum);
6427 			break;
6428 		    }
6429 		case IPPROTO_UDP:
6430 		    {
6431 			KMOD_UDPSTAT_INC(udps_badsum);
6432 			break;
6433 		    }
6434 #ifdef INET
6435 		case IPPROTO_ICMP:
6436 		    {
6437 			KMOD_ICMPSTAT_INC(icps_checksum);
6438 			break;
6439 		    }
6440 #endif
6441 #ifdef INET6
6442 		case IPPROTO_ICMPV6:
6443 		    {
6444 			KMOD_ICMP6STAT_INC(icp6s_checksum);
6445 			break;
6446 		    }
6447 #endif /* INET6 */
6448 		}
6449 		return (1);
6450 	} else {
6451 		if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
6452 			m->m_pkthdr.csum_flags |=
6453 			    (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
6454 			m->m_pkthdr.csum_data = 0xffff;
6455 		}
6456 	}
6457 	return (0);
6458 }
6459 
6460 
6461 #ifdef INET
6462 int
pf_test(int dir,int pflags,struct ifnet * ifp,struct mbuf ** m0,struct inpcb * inp)6463 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6464 {
6465 	struct pfi_kkif		*kif;
6466 	u_short			 action, reason = 0, log = 0;
6467 	struct mbuf		*m = *m0;
6468 	struct ip		*h = NULL;
6469 	struct m_tag		*ipfwtag;
6470 	struct pf_krule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6471 	struct pf_kstate	*s = NULL;
6472 	struct pf_kruleset	*ruleset = NULL;
6473 	struct pf_pdesc		 pd;
6474 	int			 off, dirndx, pqid = 0;
6475 
6476 	PF_RULES_RLOCK_TRACKER;
6477 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir));
6478 	M_ASSERTPKTHDR(m);
6479 
6480 	if (!V_pf_status.running)
6481 		return (PF_PASS);
6482 
6483 	memset(&pd, 0, sizeof(pd));
6484 
6485 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
6486 
6487 	if (kif == NULL) {
6488 		DPFPRINTF(PF_DEBUG_URGENT,
6489 		    ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
6490 		return (PF_DROP);
6491 	}
6492 	if (kif->pfik_flags & PFI_IFLAG_SKIP)
6493 		return (PF_PASS);
6494 
6495 	if (m->m_flags & M_SKIP_FIREWALL)
6496 		return (PF_PASS);
6497 
6498 	pd.pf_mtag = pf_find_mtag(m);
6499 
6500 	PF_RULES_RLOCK();
6501 
6502 	if (__predict_false(ip_divert_ptr != NULL) &&
6503 	    ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
6504 		struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
6505 		if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
6506 			if (pd.pf_mtag == NULL &&
6507 			    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6508 				action = PF_DROP;
6509 				goto done;
6510 			}
6511 			pd.pf_mtag->flags |= PF_PACKET_LOOPED;
6512 			m_tag_delete(m, ipfwtag);
6513 		}
6514 		if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
6515 			m->m_flags |= M_FASTFWD_OURS;
6516 			pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
6517 		}
6518 	} else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
6519 		/* We do IP header normalization and packet reassembly here */
6520 		action = PF_DROP;
6521 		goto done;
6522 	}
6523 	m = *m0;	/* pf_normalize messes with m0 */
6524 	h = mtod(m, struct ip *);
6525 
6526 	off = h->ip_hl << 2;
6527 	if (off < (int)sizeof(struct ip)) {
6528 		action = PF_DROP;
6529 		REASON_SET(&reason, PFRES_SHORT);
6530 		log = 1;
6531 		goto done;
6532 	}
6533 
6534 	pd.src = (struct pf_addr *)&h->ip_src;
6535 	pd.dst = (struct pf_addr *)&h->ip_dst;
6536 	pd.sport = pd.dport = NULL;
6537 	pd.ip_sum = &h->ip_sum;
6538 	pd.proto_sum = NULL;
6539 	pd.proto = h->ip_p;
6540 	pd.dir = dir;
6541 	pd.sidx = (dir == PF_IN) ? 0 : 1;
6542 	pd.didx = (dir == PF_IN) ? 1 : 0;
6543 	pd.af = AF_INET;
6544 	pd.tos = h->ip_tos & ~IPTOS_ECN_MASK;
6545 	pd.tot_len = ntohs(h->ip_len);
6546 
6547 	/* handle fragments that didn't get reassembled by normalization */
6548 	if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
6549 		action = pf_test_fragment(&r, dir, kif, m, h,
6550 		    &pd, &a, &ruleset);
6551 		goto done;
6552 	}
6553 
6554 	switch (h->ip_p) {
6555 
6556 	case IPPROTO_TCP: {
6557 		if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp),
6558 		    &action, &reason, AF_INET)) {
6559 			log = action != PF_PASS;
6560 			goto done;
6561 		}
6562 		pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2);
6563 
6564 		pd.sport = &pd.hdr.tcp.th_sport;
6565 		pd.dport = &pd.hdr.tcp.th_dport;
6566 
6567 		/* Respond to SYN with a syncookie. */
6568 		if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN &&
6569 		    pd.dir == PF_IN && pf_synflood_check(&pd)) {
6570 			pf_syncookie_send(m, off, &pd);
6571 			action = PF_DROP;
6572 			break;
6573 		}
6574 
6575 		if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0)
6576 			pqid = 1;
6577 		action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6578 		if (action == PF_DROP)
6579 			goto done;
6580 		action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6581 		    &reason);
6582 		if (action == PF_PASS) {
6583 			if (V_pfsync_update_state_ptr != NULL)
6584 				V_pfsync_update_state_ptr(s);
6585 			r = s->rule.ptr;
6586 			a = s->anchor.ptr;
6587 			log = s->log;
6588 		} else if (s == NULL) {
6589 			/* Validate remote SYN|ACK, re-create original SYN if
6590 			 * valid. */
6591 			if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) ==
6592 			    TH_ACK && pf_syncookie_validate(&pd) &&
6593 			    pd.dir == PF_IN) {
6594 				struct mbuf *msyn;
6595 
6596 				msyn = pf_syncookie_recreate_syn(h->ip_ttl,
6597 				    off,&pd);
6598 				if (msyn == NULL) {
6599 					action = PF_DROP;
6600 					break;
6601 				}
6602 
6603 				action = pf_test(dir, pflags, ifp, &msyn, inp);
6604 				m_freem(msyn);
6605 
6606 				if (action == PF_PASS) {
6607 					action = pf_test_state_tcp(&s, dir,
6608 					    kif, m, off, h, &pd, &reason);
6609 					if (action != PF_PASS || s == NULL) {
6610 						action = PF_DROP;
6611 						break;
6612 					}
6613 
6614 					s->src.seqhi = ntohl(pd.hdr.tcp.th_ack)
6615 					    - 1;
6616 					s->src.seqlo = ntohl(pd.hdr.tcp.th_seq)
6617 					    - 1;
6618 					pf_set_protostate(s, PF_PEER_SRC,
6619 					    PF_TCPS_PROXY_DST);
6620 
6621 					action = pf_synproxy(&pd, &s, &reason);
6622 					if (action != PF_PASS)
6623 						break;
6624 				}
6625 				break;
6626 			}
6627 			else {
6628 				action = pf_test_rule(&r, &s, dir, kif, m, off,
6629 				    &pd, &a, &ruleset, inp);
6630 			}
6631 		}
6632 		break;
6633 	}
6634 
6635 	case IPPROTO_UDP: {
6636 		if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp),
6637 		    &action, &reason, AF_INET)) {
6638 			log = action != PF_PASS;
6639 			goto done;
6640 		}
6641 		if (pd.hdr.udp.uh_dport == 0 ||
6642 		    ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off ||
6643 		    ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) {
6644 			action = PF_DROP;
6645 			REASON_SET(&reason, PFRES_SHORT);
6646 			goto done;
6647 		}
6648 		action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6649 		if (action == PF_PASS) {
6650 			if (V_pfsync_update_state_ptr != NULL)
6651 				V_pfsync_update_state_ptr(s);
6652 			r = s->rule.ptr;
6653 			a = s->anchor.ptr;
6654 			log = s->log;
6655 		} else if (s == NULL)
6656 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6657 			    &a, &ruleset, inp);
6658 		break;
6659 	}
6660 
6661 	case IPPROTO_ICMP: {
6662 		if (!pf_pull_hdr(m, off, &pd.hdr.icmp, ICMP_MINLEN,
6663 		    &action, &reason, AF_INET)) {
6664 			log = action != PF_PASS;
6665 			goto done;
6666 		}
6667 		action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
6668 		    &reason);
6669 		if (action == PF_PASS) {
6670 			if (V_pfsync_update_state_ptr != NULL)
6671 				V_pfsync_update_state_ptr(s);
6672 			r = s->rule.ptr;
6673 			a = s->anchor.ptr;
6674 			log = s->log;
6675 		} else if (s == NULL)
6676 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6677 			    &a, &ruleset, inp);
6678 		break;
6679 	}
6680 
6681 #ifdef INET6
6682 	case IPPROTO_ICMPV6: {
6683 		action = PF_DROP;
6684 		DPFPRINTF(PF_DEBUG_MISC,
6685 		    ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
6686 		goto done;
6687 	}
6688 #endif
6689 
6690 	default:
6691 		action = pf_test_state_other(&s, dir, kif, m, &pd);
6692 		if (action == PF_PASS) {
6693 			if (V_pfsync_update_state_ptr != NULL)
6694 				V_pfsync_update_state_ptr(s);
6695 			r = s->rule.ptr;
6696 			a = s->anchor.ptr;
6697 			log = s->log;
6698 		} else if (s == NULL)
6699 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6700 			    &a, &ruleset, inp);
6701 		break;
6702 	}
6703 
6704 done:
6705 	PF_RULES_RUNLOCK();
6706 	if (action == PF_PASS && h->ip_hl > 5 &&
6707 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6708 		action = PF_DROP;
6709 		REASON_SET(&reason, PFRES_IPOPTIONS);
6710 		log = r->log;
6711 		DPFPRINTF(PF_DEBUG_MISC,
6712 		    ("pf: dropping packet with ip options\n"));
6713 	}
6714 
6715 	if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6716 		action = PF_DROP;
6717 		REASON_SET(&reason, PFRES_MEMORY);
6718 	}
6719 	if (r->rtableid >= 0)
6720 		M_SETFIB(m, r->rtableid);
6721 
6722 	if (r->scrub_flags & PFSTATE_SETPRIO) {
6723 		if (pd.tos & IPTOS_LOWDELAY)
6724 			pqid = 1;
6725 		if (vlan_set_pcp(m, r->set_prio[pqid])) {
6726 			action = PF_DROP;
6727 			REASON_SET(&reason, PFRES_MEMORY);
6728 			log = 1;
6729 			DPFPRINTF(PF_DEBUG_MISC,
6730 			    ("pf: failed to allocate 802.1q mtag\n"));
6731 		}
6732 	}
6733 
6734 #ifdef ALTQ
6735 	if (s && s->qid) {
6736 		pd.act.pqid = s->pqid;
6737 		pd.act.qid = s->qid;
6738 	} else if (r->qid) {
6739 		pd.act.pqid = r->pqid;
6740 		pd.act.qid = r->qid;
6741 	}
6742 	if (action == PF_PASS && pd.act.qid) {
6743 		if (pd.pf_mtag == NULL &&
6744 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6745 			action = PF_DROP;
6746 			REASON_SET(&reason, PFRES_MEMORY);
6747 		} else {
6748 			if (s != NULL)
6749 				pd.pf_mtag->qid_hash = pf_state_hash(s);
6750 			if (pqid || (pd.tos & IPTOS_LOWDELAY))
6751 				pd.pf_mtag->qid = pd.act.pqid;
6752 			else
6753 				pd.pf_mtag->qid = pd.act.qid;
6754 			/* Add hints for ecn. */
6755 			pd.pf_mtag->hdr = h;
6756 		}
6757 
6758 	}
6759 #endif /* ALTQ */
6760 
6761 	/*
6762 	 * connections redirected to loopback should not match sockets
6763 	 * bound specifically to loopback due to security implications,
6764 	 * see tcp_input() and in_pcblookup_listen().
6765 	 */
6766 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6767 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6768 	    (s->nat_rule.ptr->action == PF_RDR ||
6769 	    s->nat_rule.ptr->action == PF_BINAT) &&
6770 	    IN_LOOPBACK(ntohl(pd.dst->v4.s_addr)))
6771 		m->m_flags |= M_SKIP_FIREWALL;
6772 
6773 	if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS &&
6774 	    r->divert.port && !PACKET_LOOPED(&pd)) {
6775 		ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
6776 		    sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
6777 		if (ipfwtag != NULL) {
6778 			((struct ipfw_rule_ref *)(ipfwtag+1))->info =
6779 			    ntohs(r->divert.port);
6780 			((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
6781 
6782 			if (s)
6783 				PF_STATE_UNLOCK(s);
6784 
6785 			m_tag_prepend(m, ipfwtag);
6786 			if (m->m_flags & M_FASTFWD_OURS) {
6787 				if (pd.pf_mtag == NULL &&
6788 				    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6789 					action = PF_DROP;
6790 					REASON_SET(&reason, PFRES_MEMORY);
6791 					log = 1;
6792 					DPFPRINTF(PF_DEBUG_MISC,
6793 					    ("pf: failed to allocate tag\n"));
6794 				} else {
6795 					pd.pf_mtag->flags |=
6796 					    PF_FASTFWD_OURS_PRESENT;
6797 					m->m_flags &= ~M_FASTFWD_OURS;
6798 				}
6799 			}
6800 			ip_divert_ptr(*m0, dir ==  PF_IN ? DIR_IN : DIR_OUT);
6801 			*m0 = NULL;
6802 
6803 			return (action);
6804 		} else {
6805 			/* XXX: ipfw has the same behaviour! */
6806 			action = PF_DROP;
6807 			REASON_SET(&reason, PFRES_MEMORY);
6808 			log = 1;
6809 			DPFPRINTF(PF_DEBUG_MISC,
6810 			    ("pf: failed to allocate divert tag\n"));
6811 		}
6812 	}
6813 
6814 	if (log) {
6815 		struct pf_krule *lr;
6816 
6817 		if (s != NULL && s->nat_rule.ptr != NULL &&
6818 		    s->nat_rule.ptr->log & PF_LOG_ALL)
6819 			lr = s->nat_rule.ptr;
6820 		else
6821 			lr = r;
6822 		PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
6823 		    (s == NULL));
6824 	}
6825 
6826 	pf_counter_u64_critical_enter();
6827 	pf_counter_u64_add_protected(&kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS],
6828 	    pd.tot_len);
6829 	pf_counter_u64_add_protected(&kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS],
6830 	    1);
6831 
6832 	if (action == PF_PASS || r->action == PF_DROP) {
6833 		dirndx = (dir == PF_OUT);
6834 		pf_counter_u64_add_protected(&r->packets[dirndx], 1);
6835 		pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len);
6836 		if (a != NULL) {
6837 			pf_counter_u64_add_protected(&a->packets[dirndx], 1);
6838 			pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len);
6839 		}
6840 		if (s != NULL) {
6841 			if (s->nat_rule.ptr != NULL) {
6842 				pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx],
6843 				    1);
6844 				pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx],
6845 				    pd.tot_len);
6846 			}
6847 			if (s->src_node != NULL) {
6848 				counter_u64_add(s->src_node->packets[dirndx],
6849 				    1);
6850 				counter_u64_add(s->src_node->bytes[dirndx],
6851 				    pd.tot_len);
6852 			}
6853 			if (s->nat_src_node != NULL) {
6854 				counter_u64_add(s->nat_src_node->packets[dirndx],
6855 				    1);
6856 				counter_u64_add(s->nat_src_node->bytes[dirndx],
6857 				    pd.tot_len);
6858 			}
6859 			dirndx = (dir == s->direction) ? 0 : 1;
6860 			s->packets[dirndx]++;
6861 			s->bytes[dirndx] += pd.tot_len;
6862 		}
6863 		tr = r;
6864 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6865 		if (nr != NULL && r == &V_pf_default_rule)
6866 			tr = nr;
6867 		if (tr->src.addr.type == PF_ADDR_TABLE)
6868 			pfr_update_stats(tr->src.addr.p.tbl,
6869 			    (s == NULL) ? pd.src :
6870 			    &s->key[(s->direction == PF_IN)]->
6871 				addr[(s->direction == PF_OUT)],
6872 			    pd.af, pd.tot_len, dir == PF_OUT,
6873 			    r->action == PF_PASS, tr->src.neg);
6874 		if (tr->dst.addr.type == PF_ADDR_TABLE)
6875 			pfr_update_stats(tr->dst.addr.p.tbl,
6876 			    (s == NULL) ? pd.dst :
6877 			    &s->key[(s->direction == PF_IN)]->
6878 				addr[(s->direction == PF_IN)],
6879 			    pd.af, pd.tot_len, dir == PF_OUT,
6880 			    r->action == PF_PASS, tr->dst.neg);
6881 	}
6882 	pf_counter_u64_critical_exit();
6883 
6884 	switch (action) {
6885 	case PF_SYNPROXY_DROP:
6886 		m_freem(*m0);
6887 	case PF_DEFER:
6888 		*m0 = NULL;
6889 		action = PF_PASS;
6890 		break;
6891 	case PF_DROP:
6892 		m_freem(*m0);
6893 		*m0 = NULL;
6894 		break;
6895 	default:
6896 		/* pf_route() returns unlocked. */
6897 		if (r->rt) {
6898 			pf_route(m0, r, dir, kif->pfik_ifp, s, &pd, inp);
6899 			return (action);
6900 		}
6901 		break;
6902 	}
6903 
6904 	SDT_PROBE4(pf, ip, test, done, action, reason, r, s);
6905 
6906 	if (s)
6907 		PF_STATE_UNLOCK(s);
6908 
6909 	return (action);
6910 }
6911 #endif /* INET */
6912 
6913 #ifdef INET6
6914 int
pf_test6(int dir,int pflags,struct ifnet * ifp,struct mbuf ** m0,struct inpcb * inp)6915 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6916 {
6917 	struct pfi_kkif		*kif;
6918 	u_short			 action, reason = 0, log = 0;
6919 	struct mbuf		*m = *m0, *n = NULL;
6920 	struct m_tag		*mtag;
6921 	struct ip6_hdr		*h = NULL;
6922 	struct pf_krule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6923 	struct pf_kstate	*s = NULL;
6924 	struct pf_kruleset	*ruleset = NULL;
6925 	struct pf_pdesc		 pd;
6926 	int			 off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0;
6927 
6928 	PF_RULES_RLOCK_TRACKER;
6929 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir));
6930 	M_ASSERTPKTHDR(m);
6931 
6932 	if (!V_pf_status.running)
6933 		return (PF_PASS);
6934 
6935 	memset(&pd, 0, sizeof(pd));
6936 	pd.pf_mtag = pf_find_mtag(m);
6937 
6938 	if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6939 		return (PF_PASS);
6940 
6941 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
6942 	if (kif == NULL) {
6943 		DPFPRINTF(PF_DEBUG_URGENT,
6944 		    ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6945 		return (PF_DROP);
6946 	}
6947 	if (kif->pfik_flags & PFI_IFLAG_SKIP)
6948 		return (PF_PASS);
6949 
6950 	if (m->m_flags & M_SKIP_FIREWALL)
6951 		return (PF_PASS);
6952 
6953 	PF_RULES_RLOCK();
6954 
6955 	/* We do IP header normalization and packet reassembly here */
6956 	if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6957 		action = PF_DROP;
6958 		goto done;
6959 	}
6960 	m = *m0;	/* pf_normalize messes with m0 */
6961 	h = mtod(m, struct ip6_hdr *);
6962 
6963 #if 1
6964 	/*
6965 	 * we do not support jumbogram yet.  if we keep going, zero ip6_plen
6966 	 * will do something bad, so drop the packet for now.
6967 	 */
6968 	if (htons(h->ip6_plen) == 0) {
6969 		action = PF_DROP;
6970 		REASON_SET(&reason, PFRES_NORM);	/*XXX*/
6971 		goto done;
6972 	}
6973 #endif
6974 
6975 	pd.src = (struct pf_addr *)&h->ip6_src;
6976 	pd.dst = (struct pf_addr *)&h->ip6_dst;
6977 	pd.sport = pd.dport = NULL;
6978 	pd.ip_sum = NULL;
6979 	pd.proto_sum = NULL;
6980 	pd.dir = dir;
6981 	pd.sidx = (dir == PF_IN) ? 0 : 1;
6982 	pd.didx = (dir == PF_IN) ? 1 : 0;
6983 	pd.af = AF_INET6;
6984 	pd.tos = (ntohl(h->ip6_flow) >> 20) & 0xfc;
6985 	pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6986 
6987 	off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6988 	pd.proto = h->ip6_nxt;
6989 	do {
6990 		switch (pd.proto) {
6991 		case IPPROTO_FRAGMENT:
6992 			action = pf_test_fragment(&r, dir, kif, m, h,
6993 			    &pd, &a, &ruleset);
6994 			if (action == PF_DROP)
6995 				REASON_SET(&reason, PFRES_FRAG);
6996 			goto done;
6997 		case IPPROTO_ROUTING: {
6998 			struct ip6_rthdr rthdr;
6999 
7000 			if (rh_cnt++) {
7001 				DPFPRINTF(PF_DEBUG_MISC,
7002 				    ("pf: IPv6 more than one rthdr\n"));
7003 				action = PF_DROP;
7004 				REASON_SET(&reason, PFRES_IPOPTIONS);
7005 				log = 1;
7006 				goto done;
7007 			}
7008 			if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
7009 			    &reason, pd.af)) {
7010 				DPFPRINTF(PF_DEBUG_MISC,
7011 				    ("pf: IPv6 short rthdr\n"));
7012 				action = PF_DROP;
7013 				REASON_SET(&reason, PFRES_SHORT);
7014 				log = 1;
7015 				goto done;
7016 			}
7017 			if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
7018 				DPFPRINTF(PF_DEBUG_MISC,
7019 				    ("pf: IPv6 rthdr0\n"));
7020 				action = PF_DROP;
7021 				REASON_SET(&reason, PFRES_IPOPTIONS);
7022 				log = 1;
7023 				goto done;
7024 			}
7025 			/* FALLTHROUGH */
7026 		}
7027 		case IPPROTO_AH:
7028 		case IPPROTO_HOPOPTS:
7029 		case IPPROTO_DSTOPTS: {
7030 			/* get next header and header length */
7031 			struct ip6_ext	opt6;
7032 
7033 			if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
7034 			    NULL, &reason, pd.af)) {
7035 				DPFPRINTF(PF_DEBUG_MISC,
7036 				    ("pf: IPv6 short opt\n"));
7037 				action = PF_DROP;
7038 				log = 1;
7039 				goto done;
7040 			}
7041 			if (pd.proto == IPPROTO_AH)
7042 				off += (opt6.ip6e_len + 2) * 4;
7043 			else
7044 				off += (opt6.ip6e_len + 1) * 8;
7045 			pd.proto = opt6.ip6e_nxt;
7046 			/* goto the next header */
7047 			break;
7048 		}
7049 		default:
7050 			terminal++;
7051 			break;
7052 		}
7053 	} while (!terminal);
7054 
7055 	/* if there's no routing header, use unmodified mbuf for checksumming */
7056 	if (!n)
7057 		n = m;
7058 
7059 	switch (pd.proto) {
7060 
7061 	case IPPROTO_TCP: {
7062 		if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp),
7063 		    &action, &reason, AF_INET6)) {
7064 			log = action != PF_PASS;
7065 			goto done;
7066 		}
7067 		pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2);
7068 		action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
7069 		if (action == PF_DROP)
7070 			goto done;
7071 		action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
7072 		    &reason);
7073 		if (action == PF_PASS) {
7074 			if (V_pfsync_update_state_ptr != NULL)
7075 				V_pfsync_update_state_ptr(s);
7076 			r = s->rule.ptr;
7077 			a = s->anchor.ptr;
7078 			log = s->log;
7079 		} else if (s == NULL)
7080 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
7081 			    &a, &ruleset, inp);
7082 		break;
7083 	}
7084 
7085 	case IPPROTO_UDP: {
7086 		if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp),
7087 		    &action, &reason, AF_INET6)) {
7088 			log = action != PF_PASS;
7089 			goto done;
7090 		}
7091 		if (pd.hdr.udp.uh_dport == 0 ||
7092 		    ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off ||
7093 		    ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) {
7094 			action = PF_DROP;
7095 			REASON_SET(&reason, PFRES_SHORT);
7096 			goto done;
7097 		}
7098 		action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
7099 		if (action == PF_PASS) {
7100 			if (V_pfsync_update_state_ptr != NULL)
7101 				V_pfsync_update_state_ptr(s);
7102 			r = s->rule.ptr;
7103 			a = s->anchor.ptr;
7104 			log = s->log;
7105 		} else if (s == NULL)
7106 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
7107 			    &a, &ruleset, inp);
7108 		break;
7109 	}
7110 
7111 	case IPPROTO_ICMP: {
7112 		action = PF_DROP;
7113 		DPFPRINTF(PF_DEBUG_MISC,
7114 		    ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
7115 		goto done;
7116 	}
7117 
7118 	case IPPROTO_ICMPV6: {
7119 		if (!pf_pull_hdr(m, off, &pd.hdr.icmp6, sizeof(pd.hdr.icmp6),
7120 		    &action, &reason, AF_INET6)) {
7121 			log = action != PF_PASS;
7122 			goto done;
7123 		}
7124 		action = pf_test_state_icmp(&s, dir, kif,
7125 		    m, off, h, &pd, &reason);
7126 		if (action == PF_PASS) {
7127 			if (V_pfsync_update_state_ptr != NULL)
7128 				V_pfsync_update_state_ptr(s);
7129 			r = s->rule.ptr;
7130 			a = s->anchor.ptr;
7131 			log = s->log;
7132 		} else if (s == NULL)
7133 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
7134 			    &a, &ruleset, inp);
7135 		break;
7136 	}
7137 
7138 	default:
7139 		action = pf_test_state_other(&s, dir, kif, m, &pd);
7140 		if (action == PF_PASS) {
7141 			if (V_pfsync_update_state_ptr != NULL)
7142 				V_pfsync_update_state_ptr(s);
7143 			r = s->rule.ptr;
7144 			a = s->anchor.ptr;
7145 			log = s->log;
7146 		} else if (s == NULL)
7147 			action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
7148 			    &a, &ruleset, inp);
7149 		break;
7150 	}
7151 
7152 done:
7153 	PF_RULES_RUNLOCK();
7154 	if (n != m) {
7155 		m_freem(n);
7156 		n = NULL;
7157 	}
7158 
7159 	/* handle dangerous IPv6 extension headers. */
7160 	if (action == PF_PASS && rh_cnt &&
7161 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
7162 		action = PF_DROP;
7163 		REASON_SET(&reason, PFRES_IPOPTIONS);
7164 		log = r->log;
7165 		DPFPRINTF(PF_DEBUG_MISC,
7166 		    ("pf: dropping packet with dangerous v6 headers\n"));
7167 	}
7168 
7169 	if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
7170 		action = PF_DROP;
7171 		REASON_SET(&reason, PFRES_MEMORY);
7172 	}
7173 	if (r->rtableid >= 0)
7174 		M_SETFIB(m, r->rtableid);
7175 
7176 	if (r->scrub_flags & PFSTATE_SETPRIO) {
7177 		if (pd.tos & IPTOS_LOWDELAY)
7178 			pqid = 1;
7179 		if (vlan_set_pcp(m, r->set_prio[pqid])) {
7180 			action = PF_DROP;
7181 			REASON_SET(&reason, PFRES_MEMORY);
7182 			log = 1;
7183 			DPFPRINTF(PF_DEBUG_MISC,
7184 			    ("pf: failed to allocate 802.1q mtag\n"));
7185 		}
7186 	}
7187 
7188 #ifdef ALTQ
7189 	if (s && s->qid) {
7190 		pd.act.pqid = s->pqid;
7191 		pd.act.qid = s->qid;
7192 	} else if (r->qid) {
7193 		pd.act.pqid = r->pqid;
7194 		pd.act.qid = r->qid;
7195 	}
7196 	if (action == PF_PASS && pd.act.qid) {
7197 		if (pd.pf_mtag == NULL &&
7198 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
7199 			action = PF_DROP;
7200 			REASON_SET(&reason, PFRES_MEMORY);
7201 		} else {
7202 			if (s != NULL)
7203 				pd.pf_mtag->qid_hash = pf_state_hash(s);
7204 			if (pd.tos & IPTOS_LOWDELAY)
7205 				pd.pf_mtag->qid = pd.act.pqid;
7206 			else
7207 				pd.pf_mtag->qid = pd.act.qid;
7208 			/* Add hints for ecn. */
7209 			pd.pf_mtag->hdr = h;
7210 		}
7211 	}
7212 #endif /* ALTQ */
7213 
7214 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
7215 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
7216 	    (s->nat_rule.ptr->action == PF_RDR ||
7217 	    s->nat_rule.ptr->action == PF_BINAT) &&
7218 	    IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
7219 		m->m_flags |= M_SKIP_FIREWALL;
7220 
7221 	/* XXX: Anybody working on it?! */
7222 	if (r->divert.port)
7223 		printf("pf: divert(9) is not supported for IPv6\n");
7224 
7225 	if (log) {
7226 		struct pf_krule *lr;
7227 
7228 		if (s != NULL && s->nat_rule.ptr != NULL &&
7229 		    s->nat_rule.ptr->log & PF_LOG_ALL)
7230 			lr = s->nat_rule.ptr;
7231 		else
7232 			lr = r;
7233 		PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
7234 		    &pd, (s == NULL));
7235 	}
7236 
7237 	pf_counter_u64_critical_enter();
7238 	pf_counter_u64_add_protected(&kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS],
7239 	    pd.tot_len);
7240 	pf_counter_u64_add_protected(&kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS],
7241 	    1);
7242 
7243 	if (action == PF_PASS || r->action == PF_DROP) {
7244 		dirndx = (dir == PF_OUT);
7245 		pf_counter_u64_add_protected(&r->packets[dirndx], 1);
7246 		pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len);
7247 		if (a != NULL) {
7248 			pf_counter_u64_add_protected(&a->packets[dirndx], 1);
7249 			pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len);
7250 		}
7251 		if (s != NULL) {
7252 			if (s->nat_rule.ptr != NULL) {
7253 				pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx],
7254 				    1);
7255 				pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx],
7256 				    pd.tot_len);
7257 			}
7258 			if (s->src_node != NULL) {
7259 				counter_u64_add(s->src_node->packets[dirndx],
7260 				    1);
7261 				counter_u64_add(s->src_node->bytes[dirndx],
7262 				    pd.tot_len);
7263 			}
7264 			if (s->nat_src_node != NULL) {
7265 				counter_u64_add(s->nat_src_node->packets[dirndx],
7266 				    1);
7267 				counter_u64_add(s->nat_src_node->bytes[dirndx],
7268 				    pd.tot_len);
7269 			}
7270 			dirndx = (dir == s->direction) ? 0 : 1;
7271 			s->packets[dirndx]++;
7272 			s->bytes[dirndx] += pd.tot_len;
7273 		}
7274 		tr = r;
7275 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
7276 		if (nr != NULL && r == &V_pf_default_rule)
7277 			tr = nr;
7278 		if (tr->src.addr.type == PF_ADDR_TABLE)
7279 			pfr_update_stats(tr->src.addr.p.tbl,
7280 			    (s == NULL) ? pd.src :
7281 			    &s->key[(s->direction == PF_IN)]->addr[0],
7282 			    pd.af, pd.tot_len, dir == PF_OUT,
7283 			    r->action == PF_PASS, tr->src.neg);
7284 		if (tr->dst.addr.type == PF_ADDR_TABLE)
7285 			pfr_update_stats(tr->dst.addr.p.tbl,
7286 			    (s == NULL) ? pd.dst :
7287 			    &s->key[(s->direction == PF_IN)]->addr[1],
7288 			    pd.af, pd.tot_len, dir == PF_OUT,
7289 			    r->action == PF_PASS, tr->dst.neg);
7290 	}
7291 	pf_counter_u64_critical_exit();
7292 
7293 	switch (action) {
7294 	case PF_SYNPROXY_DROP:
7295 		m_freem(*m0);
7296 	case PF_DEFER:
7297 		*m0 = NULL;
7298 		action = PF_PASS;
7299 		break;
7300 	case PF_DROP:
7301 		m_freem(*m0);
7302 		*m0 = NULL;
7303 		break;
7304 	default:
7305 		/* pf_route6() returns unlocked. */
7306 		if (r->rt) {
7307 			pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd, inp);
7308 			return (action);
7309 		}
7310 		break;
7311 	}
7312 
7313 	if (s)
7314 		PF_STATE_UNLOCK(s);
7315 
7316 	/* If reassembled packet passed, create new fragments. */
7317 	if (action == PF_PASS && *m0 && (pflags & PFIL_FWD) &&
7318 	    (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL)
7319 		action = pf_refragment6(ifp, m0, mtag);
7320 
7321 	SDT_PROBE4(pf, ip, test6, done, action, reason, r, s);
7322 
7323 	return (action);
7324 }
7325 #endif /* INET6 */
7326