xref: /freebsd-14.2/sys/netpfil/pf/pf.c (revision c5a9d5ff)
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 <[email protected]>
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  *
13  *    - Redistributions of source code must retain the above copyright
14  *      notice, this list of conditions and the following disclaimer.
15  *    - Redistributions in binary form must reproduce the above
16  *      copyright notice, this list of conditions and the following
17  *      disclaimer in the documentation and/or other materials provided
18  *      with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
28  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
30  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31  * POSSIBILITY OF SUCH DAMAGE.
32  *
33  * Effort sponsored in part by the Defense Advanced Research Projects
34  * Agency (DARPA) and Air Force Research Laboratory, Air Force
35  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
36  *
37  *	$OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
38  */
39 
40 #include <sys/cdefs.h>
41 #include "opt_bpf.h"
42 #include "opt_inet.h"
43 #include "opt_inet6.h"
44 #include "opt_pf.h"
45 #include "opt_sctp.h"
46 
47 #include <sys/param.h>
48 #include <sys/bus.h>
49 #include <sys/endian.h>
50 #include <sys/gsb_crc32.h>
51 #include <sys/hash.h>
52 #include <sys/interrupt.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/limits.h>
56 #include <sys/mbuf.h>
57 #include <sys/md5.h>
58 #include <sys/random.h>
59 #include <sys/refcount.h>
60 #include <sys/sdt.h>
61 #include <sys/socket.h>
62 #include <sys/sysctl.h>
63 #include <sys/taskqueue.h>
64 #include <sys/ucred.h>
65 
66 #include <net/if.h>
67 #include <net/if_var.h>
68 #include <net/if_private.h>
69 #include <net/if_types.h>
70 #include <net/if_vlan_var.h>
71 #include <net/route.h>
72 #include <net/route/nhop.h>
73 #include <net/vnet.h>
74 
75 #include <net/pfil.h>
76 #include <net/pfvar.h>
77 #include <net/if_pflog.h>
78 #include <net/if_pfsync.h>
79 
80 #include <netinet/in_pcb.h>
81 #include <netinet/in_var.h>
82 #include <netinet/in_fib.h>
83 #include <netinet/ip.h>
84 #include <netinet/ip_fw.h>
85 #include <netinet/ip_icmp.h>
86 #include <netinet/icmp_var.h>
87 #include <netinet/ip_var.h>
88 #include <netinet/tcp.h>
89 #include <netinet/tcp_fsm.h>
90 #include <netinet/tcp_seq.h>
91 #include <netinet/tcp_timer.h>
92 #include <netinet/tcp_var.h>
93 #include <netinet/udp.h>
94 #include <netinet/udp_var.h>
95 
96 /* dummynet */
97 #include <netinet/ip_dummynet.h>
98 #include <netinet/ip_fw.h>
99 #include <netpfil/ipfw/dn_heap.h>
100 #include <netpfil/ipfw/ip_fw_private.h>
101 #include <netpfil/ipfw/ip_dn_private.h>
102 
103 #ifdef INET6
104 #include <netinet/ip6.h>
105 #include <netinet/icmp6.h>
106 #include <netinet6/nd6.h>
107 #include <netinet6/ip6_var.h>
108 #include <netinet6/in6_pcb.h>
109 #include <netinet6/in6_fib.h>
110 #include <netinet6/scope6_var.h>
111 #endif /* INET6 */
112 
113 #include <netinet/sctp_header.h>
114 #include <netinet/sctp_crc32.h>
115 
116 #include <machine/in_cksum.h>
117 #include <security/mac/mac_framework.h>
118 
119 #define	DPFPRINTF(n, x)	if (V_pf_status.debug >= (n)) printf x
120 
121 SDT_PROVIDER_DEFINE(pf);
122 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *",
123     "struct pf_kstate *");
124 SDT_PROBE_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *",
125     "struct pf_kstate *");
126 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *",
127     "struct pf_state_key_cmp *", "int", "struct pf_pdesc *",
128     "struct pf_kstate *");
129 SDT_PROBE_DEFINE4(pf, sctp, multihome, test, "struct pfi_kkif *",
130     "struct pf_krule *", "struct mbuf *", "int");
131 SDT_PROBE_DEFINE2(pf, sctp, multihome, add, "uint32_t",
132     "struct pf_sctp_source *");
133 SDT_PROBE_DEFINE3(pf, sctp, multihome, remove, "uint32_t",
134     "struct pf_kstate *", "struct pf_sctp_source *");
135 
136 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *",
137     "struct mbuf *");
138 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *");
139 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch,
140     "int", "struct pf_keth_rule *", "char *");
141 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *");
142 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match,
143     "int", "struct pf_keth_rule *");
144 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t");
145 
146 /*
147  * Global variables
148  */
149 
150 /* state tables */
151 VNET_DEFINE(struct pf_altqqueue,	 pf_altqs[4]);
152 VNET_DEFINE(struct pf_kpalist,		 pf_pabuf);
153 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_active);
154 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_active);
155 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_inactive);
156 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_inactive);
157 VNET_DEFINE(struct pf_kstatus,		 pf_status);
158 
159 VNET_DEFINE(u_int32_t,			 ticket_altqs_active);
160 VNET_DEFINE(u_int32_t,			 ticket_altqs_inactive);
161 VNET_DEFINE(int,			 altqs_inactive_open);
162 VNET_DEFINE(u_int32_t,			 ticket_pabuf);
163 
164 VNET_DEFINE(MD5_CTX,			 pf_tcp_secret_ctx);
165 #define	V_pf_tcp_secret_ctx		 VNET(pf_tcp_secret_ctx)
166 VNET_DEFINE(u_char,			 pf_tcp_secret[16]);
167 #define	V_pf_tcp_secret			 VNET(pf_tcp_secret)
168 VNET_DEFINE(int,			 pf_tcp_secret_init);
169 #define	V_pf_tcp_secret_init		 VNET(pf_tcp_secret_init)
170 VNET_DEFINE(int,			 pf_tcp_iss_off);
171 #define	V_pf_tcp_iss_off		 VNET(pf_tcp_iss_off)
172 VNET_DECLARE(int,			 pf_vnet_active);
173 #define	V_pf_vnet_active		 VNET(pf_vnet_active)
174 
175 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx);
176 #define V_pf_purge_idx	VNET(pf_purge_idx)
177 
178 #ifdef PF_WANT_32_TO_64_COUNTER
179 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter);
180 #define	V_pf_counter_periodic_iter	VNET(pf_counter_periodic_iter)
181 
182 VNET_DEFINE(struct allrulelist_head, pf_allrulelist);
183 VNET_DEFINE(size_t, pf_allrulecount);
184 VNET_DEFINE(struct pf_krule *, pf_rulemarker);
185 #endif
186 
187 struct pf_sctp_endpoint;
188 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint);
189 struct pf_sctp_source {
190 	sa_family_t			af;
191 	struct pf_addr			addr;
192 	TAILQ_ENTRY(pf_sctp_source)	entry;
193 };
194 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source);
195 struct pf_sctp_endpoint
196 {
197 	uint32_t		 v_tag;
198 	struct pf_sctp_sources	 sources;
199 	RB_ENTRY(pf_sctp_endpoint)	entry;
200 };
201 static int
pf_sctp_endpoint_compare(struct pf_sctp_endpoint * a,struct pf_sctp_endpoint * b)202 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b)
203 {
204 	return (a->v_tag - b->v_tag);
205 }
206 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare);
207 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare);
208 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints);
209 #define V_pf_sctp_endpoints	VNET(pf_sctp_endpoints)
210 static struct mtx_padalign pf_sctp_endpoints_mtx;
211 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF);
212 #define	PF_SCTP_ENDPOINTS_LOCK()	mtx_lock(&pf_sctp_endpoints_mtx)
213 #define	PF_SCTP_ENDPOINTS_UNLOCK()	mtx_unlock(&pf_sctp_endpoints_mtx)
214 
215 /*
216  * Queue for pf_intr() sends.
217  */
218 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
219 struct pf_send_entry {
220 	STAILQ_ENTRY(pf_send_entry)	pfse_next;
221 	struct mbuf			*pfse_m;
222 	enum {
223 		PFSE_IP,
224 		PFSE_IP6,
225 		PFSE_ICMP,
226 		PFSE_ICMP6,
227 	}				pfse_type;
228 	struct {
229 		int		type;
230 		int		code;
231 		int		mtu;
232 	} icmpopts;
233 };
234 
235 STAILQ_HEAD(pf_send_head, pf_send_entry);
236 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue);
237 #define	V_pf_sendqueue	VNET(pf_sendqueue)
238 
239 static struct mtx_padalign pf_sendqueue_mtx;
240 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
241 #define	PF_SENDQ_LOCK()		mtx_lock(&pf_sendqueue_mtx)
242 #define	PF_SENDQ_UNLOCK()	mtx_unlock(&pf_sendqueue_mtx)
243 
244 /*
245  * Queue for pf_overload_task() tasks.
246  */
247 struct pf_overload_entry {
248 	SLIST_ENTRY(pf_overload_entry)	next;
249 	struct pf_addr  		addr;
250 	sa_family_t			af;
251 	uint8_t				dir;
252 	struct pf_krule  		*rule;
253 };
254 
255 SLIST_HEAD(pf_overload_head, pf_overload_entry);
256 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue);
257 #define V_pf_overloadqueue	VNET(pf_overloadqueue)
258 VNET_DEFINE_STATIC(struct task, pf_overloadtask);
259 #define	V_pf_overloadtask	VNET(pf_overloadtask)
260 
261 static struct mtx_padalign pf_overloadqueue_mtx;
262 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
263     "pf overload/flush queue", MTX_DEF);
264 #define	PF_OVERLOADQ_LOCK()	mtx_lock(&pf_overloadqueue_mtx)
265 #define	PF_OVERLOADQ_UNLOCK()	mtx_unlock(&pf_overloadqueue_mtx)
266 
267 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules);
268 struct mtx_padalign pf_unlnkdrules_mtx;
269 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
270     MTX_DEF);
271 
272 struct sx pf_config_lock;
273 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config");
274 
275 struct mtx_padalign pf_table_stats_lock;
276 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats",
277     MTX_DEF);
278 
279 VNET_DEFINE_STATIC(uma_zone_t,	pf_sources_z);
280 #define	V_pf_sources_z	VNET(pf_sources_z)
281 uma_zone_t		pf_mtag_z;
282 VNET_DEFINE(uma_zone_t,	 pf_state_z);
283 VNET_DEFINE(uma_zone_t,	 pf_state_key_z);
284 
285 VNET_DEFINE(struct unrhdr64, pf_stateid);
286 
287 static void		 pf_src_tree_remove_state(struct pf_kstate *);
288 static void		 pf_init_threshold(struct pf_threshold *, u_int32_t,
289 			    u_int32_t);
290 static void		 pf_add_threshold(struct pf_threshold *);
291 static int		 pf_check_threshold(struct pf_threshold *);
292 
293 static void		 pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
294 			    u_int16_t *, u_int16_t *, struct pf_addr *,
295 			    u_int16_t, u_int8_t, sa_family_t);
296 static int		 pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
297 			    struct tcphdr *, struct pf_state_peer *);
298 int			 pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *,
299 			    int *, u_int16_t *, u_int16_t *);
300 static void		 pf_change_icmp(struct pf_addr *, u_int16_t *,
301 			    struct pf_addr *, struct pf_addr *, u_int16_t,
302 			    u_int16_t *, u_int16_t *, u_int16_t *,
303 			    u_int16_t *, u_int8_t, sa_family_t);
304 static void		 pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
305 			    sa_family_t, struct pf_krule *, int);
306 static void		 pf_detach_state(struct pf_kstate *);
307 static int		 pf_state_key_attach(struct pf_state_key *,
308 			    struct pf_state_key *, struct pf_kstate *);
309 static void		 pf_state_key_detach(struct pf_kstate *, int);
310 static int		 pf_state_key_ctor(void *, int, void *, int);
311 static u_int32_t	 pf_tcp_iss(struct pf_pdesc *);
312 static __inline void	 pf_dummynet_flag_remove(struct mbuf *m,
313 			    struct pf_mtag *pf_mtag);
314 static int		 pf_dummynet(struct pf_pdesc *, struct pf_kstate *,
315 			    struct pf_krule *, struct mbuf **);
316 static int		 pf_dummynet_route(struct pf_pdesc *,
317 			    struct pf_kstate *, struct pf_krule *,
318 			    struct ifnet *, struct sockaddr *, struct mbuf **);
319 static int		 pf_test_eth_rule(int, struct pfi_kkif *,
320 			    struct mbuf **);
321 static int		 pf_test_rule(struct pf_krule **, struct pf_kstate **,
322 			    struct pfi_kkif *, struct mbuf *, int,
323 			    struct pf_pdesc *, struct pf_krule **,
324 			    struct pf_kruleset **, struct inpcb *);
325 static int		 pf_create_state(struct pf_krule *, struct pf_krule *,
326 			    struct pf_krule *, struct pf_pdesc *,
327 			    struct pf_ksrc_node *, struct pf_state_key *,
328 			    struct pf_state_key *, struct mbuf *, int,
329 			    u_int16_t, u_int16_t, int *, struct pfi_kkif *,
330 			    struct pf_kstate **, int, u_int16_t, u_int16_t,
331 			    int, struct pf_krule_slist *);
332 static int		 pf_state_key_addr_setup(struct pf_pdesc *, struct mbuf *,
333 			    int, struct pf_state_key_cmp *, int, struct pf_addr *,
334 			    int, struct pf_addr *, int);
335 static int		 pf_test_fragment(struct pf_krule **, struct pfi_kkif *,
336 			    struct mbuf *, void *, struct pf_pdesc *,
337 			    struct pf_krule **, struct pf_kruleset **);
338 static int		 pf_tcp_track_full(struct pf_kstate **,
339 			    struct pfi_kkif *, struct mbuf *, int,
340 			    struct pf_pdesc *, u_short *, int *);
341 static int		 pf_tcp_track_sloppy(struct pf_kstate **,
342 			    struct pf_pdesc *, u_short *);
343 static int		 pf_test_state_tcp(struct pf_kstate **,
344 			    struct pfi_kkif *, struct mbuf *, int,
345 			    void *, struct pf_pdesc *, u_short *);
346 static int		 pf_test_state_udp(struct pf_kstate **,
347 			    struct pfi_kkif *, struct mbuf *, int,
348 			    void *, struct pf_pdesc *);
349 int			 pf_icmp_state_lookup(struct pf_state_key_cmp *,
350 			    struct pf_pdesc *, struct pf_kstate **, struct mbuf *,
351 			    int, int, struct pfi_kkif *, u_int16_t, u_int16_t,
352 			    int, int *, int, int);
353 static int		 pf_test_state_icmp(struct pf_kstate **,
354 			    struct pfi_kkif *, struct mbuf *, int,
355 			    void *, struct pf_pdesc *, u_short *);
356 static void		 pf_sctp_multihome_detach_addr(const struct pf_kstate *);
357 static void		 pf_sctp_multihome_delayed(struct pf_pdesc *, int,
358 			    struct pfi_kkif *, struct pf_kstate *, int);
359 static int		 pf_test_state_sctp(struct pf_kstate **,
360 			    struct pfi_kkif *, struct mbuf *, int,
361 			    void *, struct pf_pdesc *, u_short *);
362 static int		 pf_test_state_other(struct pf_kstate **,
363 			    struct pfi_kkif *, struct mbuf *, struct pf_pdesc *);
364 static u_int16_t	 pf_calc_mss(struct pf_addr *, sa_family_t,
365 				int, u_int16_t);
366 static int		 pf_check_proto_cksum(struct mbuf *, int, int,
367 			    u_int8_t, sa_family_t);
368 static void		 pf_print_state_parts(struct pf_kstate *,
369 			    struct pf_state_key *, struct pf_state_key *);
370 static void		 pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t,
371 			    bool, u_int8_t);
372 static struct pf_kstate	*pf_find_state(struct pfi_kkif *,
373 			    const struct pf_state_key_cmp *, u_int);
374 static int		 pf_src_connlimit(struct pf_kstate **);
375 static void		 pf_overload_task(void *v, int pending);
376 static u_short		 pf_insert_src_node(struct pf_ksrc_node **,
377 			    struct pf_krule *, struct pf_addr *, sa_family_t);
378 static u_int		 pf_purge_expired_states(u_int, int);
379 static void		 pf_purge_unlinked_rules(void);
380 static int		 pf_mtag_uminit(void *, int, int);
381 static void		 pf_mtag_free(struct m_tag *);
382 static void		 pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *,
383 			    int, struct pf_state_key *);
384 #ifdef INET
385 static void		 pf_route(struct mbuf **, struct pf_krule *,
386 			    struct ifnet *, struct pf_kstate *,
387 			    struct pf_pdesc *, struct inpcb *);
388 #endif /* INET */
389 #ifdef INET6
390 static void		 pf_change_a6(struct pf_addr *, u_int16_t *,
391 			    struct pf_addr *, u_int8_t);
392 static void		 pf_route6(struct mbuf **, struct pf_krule *,
393 			    struct ifnet *, struct pf_kstate *,
394 			    struct pf_pdesc *, struct inpcb *);
395 #endif /* INET6 */
396 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t);
397 
398 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
399 
400 extern int pf_end_threads;
401 extern struct proc *pf_purge_proc;
402 
403 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
404 
405 enum { PF_ICMP_MULTI_NONE, PF_ICMP_MULTI_LINK };
406 
407 #define	PACKET_UNDO_NAT(_m, _pd, _off, _s)		\
408 	do {								\
409 		struct pf_state_key *nk;				\
410 		if ((pd->dir) == PF_OUT)					\
411 			nk = (_s)->key[PF_SK_STACK];			\
412 		else							\
413 			nk = (_s)->key[PF_SK_WIRE];			\
414 		pf_packet_rework_nat(_m, _pd, _off, nk);		\
415 	} while (0)
416 
417 #define	PACKET_LOOPED(pd)	((pd)->pf_mtag &&			\
418 				 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED)
419 
420 #define	STATE_LOOKUP(i, k, s, pd)					\
421 	do {								\
422 		(s) = pf_find_state((i), (k), (pd->dir));			\
423 		SDT_PROBE5(pf, ip, state, lookup, i, k, (pd->dir), pd, (s));	\
424 		if ((s) == NULL)					\
425 			return (PF_DROP);				\
426 		if (PACKET_LOOPED(pd))					\
427 			return (PF_PASS);				\
428 	} while (0)
429 
430 #define	BOUND_IFACE(r, k) \
431 	((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
432 
433 #define	STATE_INC_COUNTERS(s)						\
434 	do {								\
435 		struct pf_krule_item *mrm;				\
436 		counter_u64_add(s->rule.ptr->states_cur, 1);		\
437 		counter_u64_add(s->rule.ptr->states_tot, 1);		\
438 		if (s->anchor.ptr != NULL) {				\
439 			counter_u64_add(s->anchor.ptr->states_cur, 1);	\
440 			counter_u64_add(s->anchor.ptr->states_tot, 1);	\
441 		}							\
442 		if (s->nat_rule.ptr != NULL) {				\
443 			counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
444 			counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
445 		}							\
446 		SLIST_FOREACH(mrm, &s->match_rules, entry) {		\
447 			counter_u64_add(mrm->r->states_cur, 1);		\
448 			counter_u64_add(mrm->r->states_tot, 1);		\
449 		}							\
450 	} while (0)
451 
452 #define	STATE_DEC_COUNTERS(s)						\
453 	do {								\
454 		struct pf_krule_item *mrm;				\
455 		if (s->nat_rule.ptr != NULL)				\
456 			counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
457 		if (s->anchor.ptr != NULL)				\
458 			counter_u64_add(s->anchor.ptr->states_cur, -1);	\
459 		counter_u64_add(s->rule.ptr->states_cur, -1);		\
460 		SLIST_FOREACH(mrm, &s->match_rules, entry)		\
461 			counter_u64_add(mrm->r->states_cur, -1);	\
462 	} while (0)
463 
464 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
465 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items");
466 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
467 VNET_DEFINE(struct pf_idhash *, pf_idhash);
468 VNET_DEFINE(struct pf_srchash *, pf_srchash);
469 
470 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
471     "pf(4)");
472 
473 VNET_DEFINE(u_long, pf_hashmask);
474 VNET_DEFINE(u_long, pf_srchashmask);
475 VNET_DEFINE_STATIC(u_long, pf_hashsize);
476 #define V_pf_hashsize	VNET(pf_hashsize)
477 VNET_DEFINE_STATIC(u_long, pf_srchashsize);
478 #define V_pf_srchashsize	VNET(pf_srchashsize)
479 u_long	pf_ioctl_maxcount = 65535;
480 
481 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN,
482     &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable");
483 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN,
484     &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable");
485 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN,
486     &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call");
487 
488 VNET_DEFINE(void *, pf_swi_cookie);
489 VNET_DEFINE(struct intr_event *, pf_swi_ie);
490 
491 VNET_DEFINE(uint32_t, pf_hashseed);
492 #define	V_pf_hashseed	VNET(pf_hashseed)
493 
494 static void
pf_sctp_checksum(struct mbuf * m,int off)495 pf_sctp_checksum(struct mbuf *m, int off)
496 {
497 	uint32_t sum = 0;
498 
499 	/* Zero out the checksum, to enable recalculation. */
500 	m_copyback(m, off + offsetof(struct sctphdr, checksum),
501 	    sizeof(sum), (caddr_t)&sum);
502 
503 	sum = sctp_calculate_cksum(m, off);
504 
505 	m_copyback(m, off + offsetof(struct sctphdr, checksum),
506 	    sizeof(sum), (caddr_t)&sum);
507 }
508 
509 int
pf_addr_cmp(struct pf_addr * a,struct pf_addr * b,sa_family_t af)510 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
511 {
512 
513 	switch (af) {
514 #ifdef INET
515 	case AF_INET:
516 		if (a->addr32[0] > b->addr32[0])
517 			return (1);
518 		if (a->addr32[0] < b->addr32[0])
519 			return (-1);
520 		break;
521 #endif /* INET */
522 #ifdef INET6
523 	case AF_INET6:
524 		if (a->addr32[3] > b->addr32[3])
525 			return (1);
526 		if (a->addr32[3] < b->addr32[3])
527 			return (-1);
528 		if (a->addr32[2] > b->addr32[2])
529 			return (1);
530 		if (a->addr32[2] < b->addr32[2])
531 			return (-1);
532 		if (a->addr32[1] > b->addr32[1])
533 			return (1);
534 		if (a->addr32[1] < b->addr32[1])
535 			return (-1);
536 		if (a->addr32[0] > b->addr32[0])
537 			return (1);
538 		if (a->addr32[0] < b->addr32[0])
539 			return (-1);
540 		break;
541 #endif /* INET6 */
542 	default:
543 		panic("%s: unknown address family %u", __func__, af);
544 	}
545 	return (0);
546 }
547 
548 static void
pf_packet_rework_nat(struct mbuf * m,struct pf_pdesc * pd,int off,struct pf_state_key * nk)549 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off,
550 	struct pf_state_key *nk)
551 {
552 
553 	switch (pd->proto) {
554 	case IPPROTO_TCP: {
555 		struct tcphdr *th = &pd->hdr.tcp;
556 
557 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af))
558 			pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum,
559 			    &th->th_sum, &nk->addr[pd->sidx],
560 			    nk->port[pd->sidx], 0, pd->af);
561 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af))
562 			pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum,
563 			    &th->th_sum, &nk->addr[pd->didx],
564 			    nk->port[pd->didx], 0, pd->af);
565 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
566 		break;
567 	}
568 	case IPPROTO_UDP: {
569 		struct udphdr *uh = &pd->hdr.udp;
570 
571 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af))
572 			pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
573 			    &uh->uh_sum, &nk->addr[pd->sidx],
574 			    nk->port[pd->sidx], 1, pd->af);
575 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af))
576 			pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
577 			    &uh->uh_sum, &nk->addr[pd->didx],
578 			    nk->port[pd->didx], 1, pd->af);
579 		m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
580 		break;
581 	}
582 	case IPPROTO_SCTP: {
583 		struct sctphdr *sh = &pd->hdr.sctp;
584 		uint16_t checksum = 0;
585 
586 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) {
587 			pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum,
588 			    &checksum, &nk->addr[pd->sidx],
589 			    nk->port[pd->sidx], 1, pd->af);
590 		}
591 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) {
592 			pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum,
593 			    &checksum, &nk->addr[pd->didx],
594 			    nk->port[pd->didx], 1, pd->af);
595 		}
596 
597 		break;
598 	}
599 	case IPPROTO_ICMP: {
600 		struct icmp *ih = &pd->hdr.icmp;
601 
602 		if (nk->port[pd->sidx] != ih->icmp_id) {
603 			pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
604 			    ih->icmp_cksum, ih->icmp_id,
605 			    nk->port[pd->sidx], 0);
606 			ih->icmp_id = nk->port[pd->sidx];
607 			pd->sport = &ih->icmp_id;
608 
609 			m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih);
610 		}
611 		/* FALLTHROUGH */
612 	}
613 	default:
614 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) {
615 			switch (pd->af) {
616 			case AF_INET:
617 				pf_change_a(&pd->src->v4.s_addr,
618 				    pd->ip_sum, nk->addr[pd->sidx].v4.s_addr,
619 				    0);
620 				break;
621 			case AF_INET6:
622 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
623 				break;
624 			}
625 		}
626 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) {
627 			switch (pd->af) {
628 			case AF_INET:
629 				pf_change_a(&pd->dst->v4.s_addr,
630 				    pd->ip_sum, nk->addr[pd->didx].v4.s_addr,
631 				    0);
632 				break;
633 			case AF_INET6:
634 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
635 				break;
636 			}
637 		}
638 		break;
639 	}
640 }
641 
642 static __inline uint32_t
pf_hashkey(const struct pf_state_key * sk)643 pf_hashkey(const struct pf_state_key *sk)
644 {
645 	uint32_t h;
646 
647 	h = murmur3_32_hash32((const uint32_t *)sk,
648 	    sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
649 	    V_pf_hashseed);
650 
651 	return (h & V_pf_hashmask);
652 }
653 
654 static __inline uint32_t
pf_hashsrc(struct pf_addr * addr,sa_family_t af)655 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
656 {
657 	uint32_t h;
658 
659 	switch (af) {
660 	case AF_INET:
661 		h = murmur3_32_hash32((uint32_t *)&addr->v4,
662 		    sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
663 		break;
664 	case AF_INET6:
665 		h = murmur3_32_hash32((uint32_t *)&addr->v6,
666 		    sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
667 		break;
668 	default:
669 		panic("%s: unknown address family %u", __func__, af);
670 	}
671 
672 	return (h & V_pf_srchashmask);
673 }
674 
675 #ifdef ALTQ
676 static int
pf_state_hash(struct pf_kstate * s)677 pf_state_hash(struct pf_kstate *s)
678 {
679 	u_int32_t hv = (intptr_t)s / sizeof(*s);
680 
681 	hv ^= crc32(&s->src, sizeof(s->src));
682 	hv ^= crc32(&s->dst, sizeof(s->dst));
683 	if (hv == 0)
684 		hv = 1;
685 	return (hv);
686 }
687 #endif
688 
689 static __inline void
pf_set_protostate(struct pf_kstate * s,int which,u_int8_t newstate)690 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate)
691 {
692 	if (which == PF_PEER_DST || which == PF_PEER_BOTH)
693 		s->dst.state = newstate;
694 	if (which == PF_PEER_DST)
695 		return;
696 	if (s->src.state == newstate)
697 		return;
698 	if (s->creatorid == V_pf_status.hostid &&
699 	    s->key[PF_SK_STACK] != NULL &&
700 	    s->key[PF_SK_STACK]->proto == IPPROTO_TCP &&
701 	    !(TCPS_HAVEESTABLISHED(s->src.state) ||
702 	    s->src.state == TCPS_CLOSED) &&
703 	    (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED))
704 		atomic_add_32(&V_pf_status.states_halfopen, -1);
705 
706 	s->src.state = newstate;
707 }
708 
709 #ifdef INET6
710 void
pf_addrcpy(struct pf_addr * dst,struct pf_addr * src,sa_family_t af)711 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
712 {
713 	switch (af) {
714 #ifdef INET
715 	case AF_INET:
716 		dst->addr32[0] = src->addr32[0];
717 		break;
718 #endif /* INET */
719 	case AF_INET6:
720 		dst->addr32[0] = src->addr32[0];
721 		dst->addr32[1] = src->addr32[1];
722 		dst->addr32[2] = src->addr32[2];
723 		dst->addr32[3] = src->addr32[3];
724 		break;
725 	}
726 }
727 #endif /* INET6 */
728 
729 static void
pf_init_threshold(struct pf_threshold * threshold,u_int32_t limit,u_int32_t seconds)730 pf_init_threshold(struct pf_threshold *threshold,
731     u_int32_t limit, u_int32_t seconds)
732 {
733 	threshold->limit = limit * PF_THRESHOLD_MULT;
734 	threshold->seconds = seconds;
735 	threshold->count = 0;
736 	threshold->last = time_uptime;
737 }
738 
739 static void
pf_add_threshold(struct pf_threshold * threshold)740 pf_add_threshold(struct pf_threshold *threshold)
741 {
742 	u_int32_t t = time_uptime, diff = t - threshold->last;
743 
744 	if (diff >= threshold->seconds)
745 		threshold->count = 0;
746 	else
747 		threshold->count -= threshold->count * diff /
748 		    threshold->seconds;
749 	threshold->count += PF_THRESHOLD_MULT;
750 	threshold->last = t;
751 }
752 
753 static int
pf_check_threshold(struct pf_threshold * threshold)754 pf_check_threshold(struct pf_threshold *threshold)
755 {
756 	return (threshold->count > threshold->limit);
757 }
758 
759 static int
pf_src_connlimit(struct pf_kstate ** state)760 pf_src_connlimit(struct pf_kstate **state)
761 {
762 	struct pf_overload_entry *pfoe;
763 	int bad = 0;
764 
765 	PF_STATE_LOCK_ASSERT(*state);
766 	/*
767 	 * XXXKS: The src node is accessed unlocked!
768 	 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node);
769 	 */
770 
771 	(*state)->src_node->conn++;
772 	(*state)->src.tcp_est = 1;
773 	pf_add_threshold(&(*state)->src_node->conn_rate);
774 
775 	if ((*state)->rule.ptr->max_src_conn &&
776 	    (*state)->rule.ptr->max_src_conn <
777 	    (*state)->src_node->conn) {
778 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
779 		bad++;
780 	}
781 
782 	if ((*state)->rule.ptr->max_src_conn_rate.limit &&
783 	    pf_check_threshold(&(*state)->src_node->conn_rate)) {
784 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
785 		bad++;
786 	}
787 
788 	if (!bad)
789 		return (0);
790 
791 	/* Kill this state. */
792 	(*state)->timeout = PFTM_PURGE;
793 	pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED);
794 
795 	if ((*state)->rule.ptr->overload_tbl == NULL)
796 		return (1);
797 
798 	/* Schedule overloading and flushing task. */
799 	pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
800 	if (pfoe == NULL)
801 		return (1);	/* too bad :( */
802 
803 	bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
804 	pfoe->af = (*state)->key[PF_SK_WIRE]->af;
805 	pfoe->rule = (*state)->rule.ptr;
806 	pfoe->dir = (*state)->direction;
807 	PF_OVERLOADQ_LOCK();
808 	SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
809 	PF_OVERLOADQ_UNLOCK();
810 	taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
811 
812 	return (1);
813 }
814 
815 static void
pf_overload_task(void * v,int pending)816 pf_overload_task(void *v, int pending)
817 {
818 	struct pf_overload_head queue;
819 	struct pfr_addr p;
820 	struct pf_overload_entry *pfoe, *pfoe1;
821 	uint32_t killed = 0;
822 
823 	CURVNET_SET((struct vnet *)v);
824 
825 	PF_OVERLOADQ_LOCK();
826 	queue = V_pf_overloadqueue;
827 	SLIST_INIT(&V_pf_overloadqueue);
828 	PF_OVERLOADQ_UNLOCK();
829 
830 	bzero(&p, sizeof(p));
831 	SLIST_FOREACH(pfoe, &queue, next) {
832 		counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
833 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
834 			printf("%s: blocking address ", __func__);
835 			pf_print_host(&pfoe->addr, 0, pfoe->af);
836 			printf("\n");
837 		}
838 
839 		p.pfra_af = pfoe->af;
840 		switch (pfoe->af) {
841 #ifdef INET
842 		case AF_INET:
843 			p.pfra_net = 32;
844 			p.pfra_ip4addr = pfoe->addr.v4;
845 			break;
846 #endif
847 #ifdef INET6
848 		case AF_INET6:
849 			p.pfra_net = 128;
850 			p.pfra_ip6addr = pfoe->addr.v6;
851 			break;
852 #endif
853 		}
854 
855 		PF_RULES_WLOCK();
856 		pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
857 		PF_RULES_WUNLOCK();
858 	}
859 
860 	/*
861 	 * Remove those entries, that don't need flushing.
862 	 */
863 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
864 		if (pfoe->rule->flush == 0) {
865 			SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
866 			free(pfoe, M_PFTEMP);
867 		} else
868 			counter_u64_add(
869 			    V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
870 
871 	/* If nothing to flush, return. */
872 	if (SLIST_EMPTY(&queue)) {
873 		CURVNET_RESTORE();
874 		return;
875 	}
876 
877 	for (int i = 0; i <= V_pf_hashmask; i++) {
878 		struct pf_idhash *ih = &V_pf_idhash[i];
879 		struct pf_state_key *sk;
880 		struct pf_kstate *s;
881 
882 		PF_HASHROW_LOCK(ih);
883 		LIST_FOREACH(s, &ih->states, entry) {
884 		    sk = s->key[PF_SK_WIRE];
885 		    SLIST_FOREACH(pfoe, &queue, next)
886 			if (sk->af == pfoe->af &&
887 			    ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
888 			    pfoe->rule == s->rule.ptr) &&
889 			    ((pfoe->dir == PF_OUT &&
890 			    PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
891 			    (pfoe->dir == PF_IN &&
892 			    PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
893 				s->timeout = PFTM_PURGE;
894 				pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED);
895 				killed++;
896 			}
897 		}
898 		PF_HASHROW_UNLOCK(ih);
899 	}
900 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
901 		free(pfoe, M_PFTEMP);
902 	if (V_pf_status.debug >= PF_DEBUG_MISC)
903 		printf("%s: %u states killed", __func__, killed);
904 
905 	CURVNET_RESTORE();
906 }
907 
908 /*
909  * Can return locked on failure, so that we can consistently
910  * allocate and insert a new one.
911  */
912 struct pf_ksrc_node *
pf_find_src_node(struct pf_addr * src,struct pf_krule * rule,sa_family_t af,struct pf_srchash ** sh,bool returnlocked)913 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af,
914 	struct pf_srchash **sh, bool returnlocked)
915 {
916 	struct pf_ksrc_node *n;
917 
918 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
919 
920 	*sh = &V_pf_srchash[pf_hashsrc(src, af)];
921 	PF_HASHROW_LOCK(*sh);
922 	LIST_FOREACH(n, &(*sh)->nodes, entry)
923 		if (n->rule.ptr == rule && n->af == af &&
924 		    ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
925 		    (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
926 			break;
927 
928 	if (n != NULL) {
929 		n->states++;
930 		PF_HASHROW_UNLOCK(*sh);
931 	} else if (returnlocked == false)
932 		PF_HASHROW_UNLOCK(*sh);
933 
934 	return (n);
935 }
936 
937 static void
pf_free_src_node(struct pf_ksrc_node * sn)938 pf_free_src_node(struct pf_ksrc_node *sn)
939 {
940 
941 	for (int i = 0; i < 2; i++) {
942 		counter_u64_free(sn->bytes[i]);
943 		counter_u64_free(sn->packets[i]);
944 	}
945 	uma_zfree(V_pf_sources_z, sn);
946 }
947 
948 static u_short
pf_insert_src_node(struct pf_ksrc_node ** sn,struct pf_krule * rule,struct pf_addr * src,sa_family_t af)949 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule,
950     struct pf_addr *src, sa_family_t af)
951 {
952 	u_short			 reason = 0;
953 	struct pf_srchash	*sh = NULL;
954 
955 	KASSERT((rule->rule_flag & PFRULE_SRCTRACK ||
956 	    rule->rpool.opts & PF_POOL_STICKYADDR),
957 	    ("%s for non-tracking rule %p", __func__, rule));
958 
959 	if (*sn == NULL)
960 		*sn = pf_find_src_node(src, rule, af, &sh, true);
961 
962 	if (*sn == NULL) {
963 		PF_HASHROW_ASSERT(sh);
964 
965 		if (rule->max_src_nodes &&
966 		    counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) {
967 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1);
968 			PF_HASHROW_UNLOCK(sh);
969 			reason = PFRES_SRCLIMIT;
970 			goto done;
971 		}
972 
973 		(*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
974 		if ((*sn) == NULL) {
975 			PF_HASHROW_UNLOCK(sh);
976 			reason = PFRES_MEMORY;
977 			goto done;
978 		}
979 
980 		for (int i = 0; i < 2; i++) {
981 			(*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT);
982 			(*sn)->packets[i] = counter_u64_alloc(M_NOWAIT);
983 
984 			if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) {
985 				pf_free_src_node(*sn);
986 				PF_HASHROW_UNLOCK(sh);
987 				reason = PFRES_MEMORY;
988 				goto done;
989 			}
990 		}
991 
992 		pf_init_threshold(&(*sn)->conn_rate,
993 		    rule->max_src_conn_rate.limit,
994 		    rule->max_src_conn_rate.seconds);
995 
996 		MPASS((*sn)->lock == NULL);
997 		(*sn)->lock = &sh->lock;
998 
999 		(*sn)->af = af;
1000 		(*sn)->rule.ptr = rule;
1001 		PF_ACPY(&(*sn)->addr, src, af);
1002 		LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
1003 		(*sn)->creation = time_uptime;
1004 		(*sn)->ruletype = rule->action;
1005 		(*sn)->states = 1;
1006 		if ((*sn)->rule.ptr != NULL)
1007 			counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
1008 		PF_HASHROW_UNLOCK(sh);
1009 		counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
1010 	} else {
1011 		if (rule->max_src_states &&
1012 		    (*sn)->states >= rule->max_src_states) {
1013 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
1014 			    1);
1015 			reason = PFRES_SRCLIMIT;
1016 			goto done;
1017 		}
1018 	}
1019 done:
1020 	return (reason);
1021 }
1022 
1023 void
pf_unlink_src_node(struct pf_ksrc_node * src)1024 pf_unlink_src_node(struct pf_ksrc_node *src)
1025 {
1026 	PF_SRC_NODE_LOCK_ASSERT(src);
1027 
1028 	LIST_REMOVE(src, entry);
1029 	if (src->rule.ptr)
1030 		counter_u64_add(src->rule.ptr->src_nodes, -1);
1031 }
1032 
1033 u_int
pf_free_src_nodes(struct pf_ksrc_node_list * head)1034 pf_free_src_nodes(struct pf_ksrc_node_list *head)
1035 {
1036 	struct pf_ksrc_node *sn, *tmp;
1037 	u_int count = 0;
1038 
1039 	LIST_FOREACH_SAFE(sn, head, entry, tmp) {
1040 		pf_free_src_node(sn);
1041 		count++;
1042 	}
1043 
1044 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
1045 
1046 	return (count);
1047 }
1048 
1049 void
pf_mtag_initialize(void)1050 pf_mtag_initialize(void)
1051 {
1052 
1053 	pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
1054 	    sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
1055 	    UMA_ALIGN_PTR, 0);
1056 }
1057 
1058 /* Per-vnet data storage structures initialization. */
1059 void
pf_initialize(void)1060 pf_initialize(void)
1061 {
1062 	struct pf_keyhash	*kh;
1063 	struct pf_idhash	*ih;
1064 	struct pf_srchash	*sh;
1065 	u_int i;
1066 
1067 	if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize))
1068 		V_pf_hashsize = PF_HASHSIZ;
1069 	if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize))
1070 		V_pf_srchashsize = PF_SRCHASHSIZ;
1071 
1072 	V_pf_hashseed = arc4random();
1073 
1074 	/* States and state keys storage. */
1075 	V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate),
1076 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1077 	V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
1078 	uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
1079 	uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
1080 
1081 	V_pf_state_key_z = uma_zcreate("pf state keys",
1082 	    sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
1083 	    UMA_ALIGN_PTR, 0);
1084 
1085 	V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash),
1086 	    M_PFHASH, M_NOWAIT | M_ZERO);
1087 	V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash),
1088 	    M_PFHASH, M_NOWAIT | M_ZERO);
1089 	if (V_pf_keyhash == NULL || V_pf_idhash == NULL) {
1090 		printf("pf: Unable to allocate memory for "
1091 		    "state_hashsize %lu.\n", V_pf_hashsize);
1092 
1093 		free(V_pf_keyhash, M_PFHASH);
1094 		free(V_pf_idhash, M_PFHASH);
1095 
1096 		V_pf_hashsize = PF_HASHSIZ;
1097 		V_pf_keyhash = mallocarray(V_pf_hashsize,
1098 		    sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO);
1099 		V_pf_idhash = mallocarray(V_pf_hashsize,
1100 		    sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO);
1101 	}
1102 
1103 	V_pf_hashmask = V_pf_hashsize - 1;
1104 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
1105 	    i++, kh++, ih++) {
1106 		mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
1107 		mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
1108 	}
1109 
1110 	/* Source nodes. */
1111 	V_pf_sources_z = uma_zcreate("pf source nodes",
1112 	    sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
1113 	    0);
1114 	V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
1115 	uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
1116 	uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
1117 
1118 	V_pf_srchash = mallocarray(V_pf_srchashsize,
1119 	    sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO);
1120 	if (V_pf_srchash == NULL) {
1121 		printf("pf: Unable to allocate memory for "
1122 		    "source_hashsize %lu.\n", V_pf_srchashsize);
1123 
1124 		V_pf_srchashsize = PF_SRCHASHSIZ;
1125 		V_pf_srchash = mallocarray(V_pf_srchashsize,
1126 		    sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO);
1127 	}
1128 
1129 	V_pf_srchashmask = V_pf_srchashsize - 1;
1130 	for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++)
1131 		mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
1132 
1133 	/* ALTQ */
1134 	TAILQ_INIT(&V_pf_altqs[0]);
1135 	TAILQ_INIT(&V_pf_altqs[1]);
1136 	TAILQ_INIT(&V_pf_altqs[2]);
1137 	TAILQ_INIT(&V_pf_altqs[3]);
1138 	TAILQ_INIT(&V_pf_pabuf);
1139 	V_pf_altqs_active = &V_pf_altqs[0];
1140 	V_pf_altq_ifs_active = &V_pf_altqs[1];
1141 	V_pf_altqs_inactive = &V_pf_altqs[2];
1142 	V_pf_altq_ifs_inactive = &V_pf_altqs[3];
1143 
1144 	/* Send & overload+flush queues. */
1145 	STAILQ_INIT(&V_pf_sendqueue);
1146 	SLIST_INIT(&V_pf_overloadqueue);
1147 	TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
1148 
1149 	/* Unlinked, but may be referenced rules. */
1150 	TAILQ_INIT(&V_pf_unlinked_rules);
1151 }
1152 
1153 void
pf_mtag_cleanup(void)1154 pf_mtag_cleanup(void)
1155 {
1156 
1157 	uma_zdestroy(pf_mtag_z);
1158 }
1159 
1160 void
pf_cleanup(void)1161 pf_cleanup(void)
1162 {
1163 	struct pf_keyhash	*kh;
1164 	struct pf_idhash	*ih;
1165 	struct pf_srchash	*sh;
1166 	struct pf_send_entry	*pfse, *next;
1167 	u_int i;
1168 
1169 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
1170 	    i++, kh++, ih++) {
1171 		KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
1172 		    __func__));
1173 		KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
1174 		    __func__));
1175 		mtx_destroy(&kh->lock);
1176 		mtx_destroy(&ih->lock);
1177 	}
1178 	free(V_pf_keyhash, M_PFHASH);
1179 	free(V_pf_idhash, M_PFHASH);
1180 
1181 	for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
1182 		KASSERT(LIST_EMPTY(&sh->nodes),
1183 		    ("%s: source node hash not empty", __func__));
1184 		mtx_destroy(&sh->lock);
1185 	}
1186 	free(V_pf_srchash, M_PFHASH);
1187 
1188 	STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
1189 		m_freem(pfse->pfse_m);
1190 		free(pfse, M_PFTEMP);
1191 	}
1192 	MPASS(RB_EMPTY(&V_pf_sctp_endpoints));
1193 
1194 	uma_zdestroy(V_pf_sources_z);
1195 	uma_zdestroy(V_pf_state_z);
1196 	uma_zdestroy(V_pf_state_key_z);
1197 }
1198 
1199 static int
pf_mtag_uminit(void * mem,int size,int how)1200 pf_mtag_uminit(void *mem, int size, int how)
1201 {
1202 	struct m_tag *t;
1203 
1204 	t = (struct m_tag *)mem;
1205 	t->m_tag_cookie = MTAG_ABI_COMPAT;
1206 	t->m_tag_id = PACKET_TAG_PF;
1207 	t->m_tag_len = sizeof(struct pf_mtag);
1208 	t->m_tag_free = pf_mtag_free;
1209 
1210 	return (0);
1211 }
1212 
1213 static void
pf_mtag_free(struct m_tag * t)1214 pf_mtag_free(struct m_tag *t)
1215 {
1216 
1217 	uma_zfree(pf_mtag_z, t);
1218 }
1219 
1220 struct pf_mtag *
pf_get_mtag(struct mbuf * m)1221 pf_get_mtag(struct mbuf *m)
1222 {
1223 	struct m_tag *mtag;
1224 
1225 	if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
1226 		return ((struct pf_mtag *)(mtag + 1));
1227 
1228 	mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
1229 	if (mtag == NULL)
1230 		return (NULL);
1231 	bzero(mtag + 1, sizeof(struct pf_mtag));
1232 	m_tag_prepend(m, mtag);
1233 
1234 	return ((struct pf_mtag *)(mtag + 1));
1235 }
1236 
1237 static int
pf_state_key_attach(struct pf_state_key * skw,struct pf_state_key * sks,struct pf_kstate * s)1238 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
1239     struct pf_kstate *s)
1240 {
1241 	struct pf_keyhash	*khs, *khw, *kh;
1242 	struct pf_state_key	*sk, *cur;
1243 	struct pf_kstate	*si, *olds = NULL;
1244 	int idx;
1245 
1246 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1247 	KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
1248 	KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
1249 
1250 	/*
1251 	 * We need to lock hash slots of both keys. To avoid deadlock
1252 	 * we always lock the slot with lower address first. Unlock order
1253 	 * isn't important.
1254 	 *
1255 	 * We also need to lock ID hash slot before dropping key
1256 	 * locks. On success we return with ID hash slot locked.
1257 	 */
1258 
1259 	if (skw == sks) {
1260 		khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
1261 		PF_HASHROW_LOCK(khs);
1262 	} else {
1263 		khs = &V_pf_keyhash[pf_hashkey(sks)];
1264 		khw = &V_pf_keyhash[pf_hashkey(skw)];
1265 		if (khs == khw) {
1266 			PF_HASHROW_LOCK(khs);
1267 		} else if (khs < khw) {
1268 			PF_HASHROW_LOCK(khs);
1269 			PF_HASHROW_LOCK(khw);
1270 		} else {
1271 			PF_HASHROW_LOCK(khw);
1272 			PF_HASHROW_LOCK(khs);
1273 		}
1274 	}
1275 
1276 #define	KEYS_UNLOCK()	do {			\
1277 	if (khs != khw) {			\
1278 		PF_HASHROW_UNLOCK(khs);		\
1279 		PF_HASHROW_UNLOCK(khw);		\
1280 	} else					\
1281 		PF_HASHROW_UNLOCK(khs);		\
1282 } while (0)
1283 
1284 	/*
1285 	 * First run: start with wire key.
1286 	 */
1287 	sk = skw;
1288 	kh = khw;
1289 	idx = PF_SK_WIRE;
1290 
1291 	MPASS(s->lock == NULL);
1292 	s->lock = &V_pf_idhash[PF_IDHASH(s)].lock;
1293 
1294 keyattach:
1295 	LIST_FOREACH(cur, &kh->keys, entry)
1296 		if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
1297 			break;
1298 
1299 	if (cur != NULL) {
1300 		/* Key exists. Check for same kif, if none, add to key. */
1301 		TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
1302 			struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
1303 
1304 			PF_HASHROW_LOCK(ih);
1305 			if (si->kif == s->kif &&
1306 			    si->direction == s->direction) {
1307 				if (sk->proto == IPPROTO_TCP &&
1308 				    si->src.state >= TCPS_FIN_WAIT_2 &&
1309 				    si->dst.state >= TCPS_FIN_WAIT_2) {
1310 					/*
1311 					 * New state matches an old >FIN_WAIT_2
1312 					 * state. We can't drop key hash locks,
1313 					 * thus we can't unlink it properly.
1314 					 *
1315 					 * As a workaround we drop it into
1316 					 * TCPS_CLOSED state, schedule purge
1317 					 * ASAP and push it into the very end
1318 					 * of the slot TAILQ, so that it won't
1319 					 * conflict with our new state.
1320 					 */
1321 					pf_set_protostate(si, PF_PEER_BOTH,
1322 					    TCPS_CLOSED);
1323 					si->timeout = PFTM_PURGE;
1324 					olds = si;
1325 				} else {
1326 					if (V_pf_status.debug >= PF_DEBUG_MISC) {
1327 						printf("pf: %s key attach "
1328 						    "failed on %s: ",
1329 						    (idx == PF_SK_WIRE) ?
1330 						    "wire" : "stack",
1331 						    s->kif->pfik_name);
1332 						pf_print_state_parts(s,
1333 						    (idx == PF_SK_WIRE) ?
1334 						    sk : NULL,
1335 						    (idx == PF_SK_STACK) ?
1336 						    sk : NULL);
1337 						printf(", existing: ");
1338 						pf_print_state_parts(si,
1339 						    (idx == PF_SK_WIRE) ?
1340 						    sk : NULL,
1341 						    (idx == PF_SK_STACK) ?
1342 						    sk : NULL);
1343 						printf("\n");
1344 					}
1345 					s->timeout = PFTM_UNLINKED;
1346 					PF_HASHROW_UNLOCK(ih);
1347 					KEYS_UNLOCK();
1348 					uma_zfree(V_pf_state_key_z, sk);
1349 					if (idx == PF_SK_STACK)
1350 						pf_detach_state(s);
1351 					return (EEXIST); /* collision! */
1352 				}
1353 			}
1354 			PF_HASHROW_UNLOCK(ih);
1355 		}
1356 		uma_zfree(V_pf_state_key_z, sk);
1357 		s->key[idx] = cur;
1358 	} else {
1359 		LIST_INSERT_HEAD(&kh->keys, sk, entry);
1360 		s->key[idx] = sk;
1361 	}
1362 
1363 stateattach:
1364 	/* List is sorted, if-bound states before floating. */
1365 	if (s->kif == V_pfi_all)
1366 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1367 	else
1368 		TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1369 
1370 	if (olds) {
1371 		TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1372 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1373 		    key_list[idx]);
1374 		olds = NULL;
1375 	}
1376 
1377 	/*
1378 	 * Attach done. See how should we (or should not?)
1379 	 * attach a second key.
1380 	 */
1381 	if (sks == skw) {
1382 		s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1383 		idx = PF_SK_STACK;
1384 		sks = NULL;
1385 		goto stateattach;
1386 	} else if (sks != NULL) {
1387 		/*
1388 		 * Continue attaching with stack key.
1389 		 */
1390 		sk = sks;
1391 		kh = khs;
1392 		idx = PF_SK_STACK;
1393 		sks = NULL;
1394 		goto keyattach;
1395 	}
1396 
1397 	PF_STATE_LOCK(s);
1398 	KEYS_UNLOCK();
1399 
1400 	KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1401 	    ("%s failure", __func__));
1402 
1403 	return (0);
1404 #undef	KEYS_UNLOCK
1405 }
1406 
1407 static void
pf_detach_state(struct pf_kstate * s)1408 pf_detach_state(struct pf_kstate *s)
1409 {
1410 	struct pf_state_key *sks = s->key[PF_SK_STACK];
1411 	struct pf_keyhash *kh;
1412 
1413 	MPASS(s->timeout >= PFTM_MAX);
1414 
1415 	pf_sctp_multihome_detach_addr(s);
1416 
1417 	if (sks != NULL) {
1418 		kh = &V_pf_keyhash[pf_hashkey(sks)];
1419 		PF_HASHROW_LOCK(kh);
1420 		if (s->key[PF_SK_STACK] != NULL)
1421 			pf_state_key_detach(s, PF_SK_STACK);
1422 		/*
1423 		 * If both point to same key, then we are done.
1424 		 */
1425 		if (sks == s->key[PF_SK_WIRE]) {
1426 			pf_state_key_detach(s, PF_SK_WIRE);
1427 			PF_HASHROW_UNLOCK(kh);
1428 			return;
1429 		}
1430 		PF_HASHROW_UNLOCK(kh);
1431 	}
1432 
1433 	if (s->key[PF_SK_WIRE] != NULL) {
1434 		kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1435 		PF_HASHROW_LOCK(kh);
1436 		if (s->key[PF_SK_WIRE] != NULL)
1437 			pf_state_key_detach(s, PF_SK_WIRE);
1438 		PF_HASHROW_UNLOCK(kh);
1439 	}
1440 }
1441 
1442 static void
pf_state_key_detach(struct pf_kstate * s,int idx)1443 pf_state_key_detach(struct pf_kstate *s, int idx)
1444 {
1445 	struct pf_state_key *sk = s->key[idx];
1446 #ifdef INVARIANTS
1447 	struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1448 
1449 	PF_HASHROW_ASSERT(kh);
1450 #endif
1451 	TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1452 	s->key[idx] = NULL;
1453 
1454 	if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1455 		LIST_REMOVE(sk, entry);
1456 		uma_zfree(V_pf_state_key_z, sk);
1457 	}
1458 }
1459 
1460 static int
pf_state_key_ctor(void * mem,int size,void * arg,int flags)1461 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1462 {
1463 	struct pf_state_key *sk = mem;
1464 
1465 	bzero(sk, sizeof(struct pf_state_key_cmp));
1466 	TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1467 	TAILQ_INIT(&sk->states[PF_SK_STACK]);
1468 
1469 	return (0);
1470 }
1471 
1472 static int
pf_state_key_addr_setup(struct pf_pdesc * pd,struct mbuf * m,int off,struct pf_state_key_cmp * key,int sidx,struct pf_addr * saddr,int didx,struct pf_addr * daddr,int multi)1473 pf_state_key_addr_setup(struct pf_pdesc *pd, struct mbuf *m, int off,
1474     struct pf_state_key_cmp *key, int sidx, struct pf_addr *saddr,
1475     int didx, struct pf_addr *daddr, int multi)
1476 {
1477 #ifdef INET6
1478 	struct nd_neighbor_solicit nd;
1479 	struct pf_addr *target;
1480 	u_short action, reason;
1481 
1482 	if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6)
1483 		goto copy;
1484 
1485 	switch (pd->hdr.icmp6.icmp6_type) {
1486 	case ND_NEIGHBOR_SOLICIT:
1487 		if (multi)
1488 			return (-1);
1489 		if (!pf_pull_hdr(m, off, &nd, sizeof(nd), &action, &reason, pd->af))
1490 			return (-1);
1491 		target = (struct pf_addr *)&nd.nd_ns_target;
1492 		daddr = target;
1493 		break;
1494 	case ND_NEIGHBOR_ADVERT:
1495 		if (multi)
1496 			return (-1);
1497 		if (!pf_pull_hdr(m, off, &nd, sizeof(nd), &action, &reason, pd->af))
1498 			return (-1);
1499 		target = (struct pf_addr *)&nd.nd_ns_target;
1500 		saddr = target;
1501 		if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) {
1502 			key->addr[didx].addr32[0] = 0;
1503 			key->addr[didx].addr32[1] = 0;
1504 			key->addr[didx].addr32[2] = 0;
1505 			key->addr[didx].addr32[3] = 0;
1506 			daddr = NULL; /* overwritten */
1507 		}
1508 		break;
1509 	default:
1510 		if (multi == PF_ICMP_MULTI_LINK) {
1511 			key->addr[sidx].addr32[0] = IPV6_ADDR_INT32_MLL;
1512 			key->addr[sidx].addr32[1] = 0;
1513 			key->addr[sidx].addr32[2] = 0;
1514 			key->addr[sidx].addr32[3] = IPV6_ADDR_INT32_ONE;
1515 			saddr = NULL; /* overwritten */
1516 		}
1517 	}
1518 copy:
1519 #endif
1520 	if (saddr)
1521 		PF_ACPY(&key->addr[sidx], saddr, pd->af);
1522 	if (daddr)
1523 		PF_ACPY(&key->addr[didx], daddr, pd->af);
1524 
1525 	return (0);
1526 }
1527 
1528 struct pf_state_key *
pf_state_key_setup(struct pf_pdesc * pd,struct mbuf * m,int off,struct pf_addr * saddr,struct pf_addr * daddr,u_int16_t sport,u_int16_t dport)1529 pf_state_key_setup(struct pf_pdesc *pd, struct mbuf *m, int off,
1530     struct pf_addr *saddr, struct pf_addr *daddr, u_int16_t sport,
1531     u_int16_t dport)
1532 {
1533 	struct pf_state_key *sk;
1534 
1535 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1536 	if (sk == NULL)
1537 		return (NULL);
1538 
1539 	if (pf_state_key_addr_setup(pd, m, off, (struct pf_state_key_cmp *)sk,
1540 	    pd->sidx, pd->src, pd->didx, pd->dst, 0)) {
1541 		uma_zfree(V_pf_state_key_z, sk);
1542 		return (NULL);
1543 	}
1544 
1545 	sk->port[pd->sidx] = sport;
1546 	sk->port[pd->didx] = dport;
1547 	sk->proto = pd->proto;
1548 	sk->af = pd->af;
1549 
1550 	return (sk);
1551 }
1552 
1553 struct pf_state_key *
pf_state_key_clone(const struct pf_state_key * orig)1554 pf_state_key_clone(const struct pf_state_key *orig)
1555 {
1556 	struct pf_state_key *sk;
1557 
1558 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1559 	if (sk == NULL)
1560 		return (NULL);
1561 
1562 	bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1563 
1564 	return (sk);
1565 }
1566 
1567 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)1568 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif,
1569     struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s)
1570 {
1571 	struct pf_idhash *ih;
1572 	struct pf_kstate *cur;
1573 	int error;
1574 
1575 	KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1576 	    ("%s: sks not pristine", __func__));
1577 	KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1578 	    ("%s: skw not pristine", __func__));
1579 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1580 
1581 	s->kif = kif;
1582 	s->orig_kif = orig_kif;
1583 
1584 	if (s->id == 0 && s->creatorid == 0) {
1585 		s->id = alloc_unr64(&V_pf_stateid);
1586 		s->id = htobe64(s->id);
1587 		s->creatorid = V_pf_status.hostid;
1588 	}
1589 
1590 	/* Returns with ID locked on success. */
1591 	if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1592 		return (error);
1593 
1594 	ih = &V_pf_idhash[PF_IDHASH(s)];
1595 	PF_HASHROW_ASSERT(ih);
1596 	LIST_FOREACH(cur, &ih->states, entry)
1597 		if (cur->id == s->id && cur->creatorid == s->creatorid)
1598 			break;
1599 
1600 	if (cur != NULL) {
1601 		s->timeout = PFTM_UNLINKED;
1602 		PF_HASHROW_UNLOCK(ih);
1603 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1604 			printf("pf: state ID collision: "
1605 			    "id: %016llx creatorid: %08x\n",
1606 			    (unsigned long long)be64toh(s->id),
1607 			    ntohl(s->creatorid));
1608 		}
1609 		pf_detach_state(s);
1610 		return (EEXIST);
1611 	}
1612 	LIST_INSERT_HEAD(&ih->states, s, entry);
1613 	/* One for keys, one for ID hash. */
1614 	refcount_init(&s->refs, 2);
1615 
1616 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1617 	if (V_pfsync_insert_state_ptr != NULL)
1618 		V_pfsync_insert_state_ptr(s);
1619 
1620 	/* Returns locked. */
1621 	return (0);
1622 }
1623 
1624 /*
1625  * Find state by ID: returns with locked row on success.
1626  */
1627 struct pf_kstate *
pf_find_state_byid(uint64_t id,uint32_t creatorid)1628 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1629 {
1630 	struct pf_idhash *ih;
1631 	struct pf_kstate *s;
1632 
1633 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1634 
1635 	ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))];
1636 
1637 	PF_HASHROW_LOCK(ih);
1638 	LIST_FOREACH(s, &ih->states, entry)
1639 		if (s->id == id && s->creatorid == creatorid)
1640 			break;
1641 
1642 	if (s == NULL)
1643 		PF_HASHROW_UNLOCK(ih);
1644 
1645 	return (s);
1646 }
1647 
1648 /*
1649  * Find state by key.
1650  * Returns with ID hash slot locked on success.
1651  */
1652 static struct pf_kstate *
pf_find_state(struct pfi_kkif * kif,const struct pf_state_key_cmp * key,u_int dir)1653 pf_find_state(struct pfi_kkif *kif, const struct pf_state_key_cmp *key,
1654     u_int dir)
1655 {
1656 	struct pf_keyhash	*kh;
1657 	struct pf_state_key	*sk;
1658 	struct pf_kstate	*s;
1659 	int idx;
1660 
1661 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1662 
1663 	kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)];
1664 
1665 	PF_HASHROW_LOCK(kh);
1666 	LIST_FOREACH(sk, &kh->keys, entry)
1667 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1668 			break;
1669 	if (sk == NULL) {
1670 		PF_HASHROW_UNLOCK(kh);
1671 		return (NULL);
1672 	}
1673 
1674 	idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1675 
1676 	/* List is sorted, if-bound states before floating ones. */
1677 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1678 		if (s->kif == V_pfi_all || s->kif == kif) {
1679 			PF_STATE_LOCK(s);
1680 			PF_HASHROW_UNLOCK(kh);
1681 			if (__predict_false(s->timeout >= PFTM_MAX)) {
1682 				/*
1683 				 * State is either being processed by
1684 				 * pf_unlink_state() in an other thread, or
1685 				 * is scheduled for immediate expiry.
1686 				 */
1687 				PF_STATE_UNLOCK(s);
1688 				return (NULL);
1689 			}
1690 			return (s);
1691 		}
1692 	PF_HASHROW_UNLOCK(kh);
1693 
1694 	return (NULL);
1695 }
1696 
1697 /*
1698  * Returns with ID hash slot locked on success.
1699  */
1700 struct pf_kstate *
pf_find_state_all(const struct pf_state_key_cmp * key,u_int dir,int * more)1701 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more)
1702 {
1703 	struct pf_keyhash	*kh;
1704 	struct pf_state_key	*sk;
1705 	struct pf_kstate	*s, *ret = NULL;
1706 	int			 idx, inout = 0;
1707 
1708 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1709 
1710 	kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)];
1711 
1712 	PF_HASHROW_LOCK(kh);
1713 	LIST_FOREACH(sk, &kh->keys, entry)
1714 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1715 			break;
1716 	if (sk == NULL) {
1717 		PF_HASHROW_UNLOCK(kh);
1718 		return (NULL);
1719 	}
1720 	switch (dir) {
1721 	case PF_IN:
1722 		idx = PF_SK_WIRE;
1723 		break;
1724 	case PF_OUT:
1725 		idx = PF_SK_STACK;
1726 		break;
1727 	case PF_INOUT:
1728 		idx = PF_SK_WIRE;
1729 		inout = 1;
1730 		break;
1731 	default:
1732 		panic("%s: dir %u", __func__, dir);
1733 	}
1734 second_run:
1735 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1736 		if (more == NULL) {
1737 			PF_STATE_LOCK(s);
1738 			PF_HASHROW_UNLOCK(kh);
1739 			return (s);
1740 		}
1741 
1742 		if (ret)
1743 			(*more)++;
1744 		else {
1745 			ret = s;
1746 			PF_STATE_LOCK(s);
1747 		}
1748 	}
1749 	if (inout == 1) {
1750 		inout = 0;
1751 		idx = PF_SK_STACK;
1752 		goto second_run;
1753 	}
1754 	PF_HASHROW_UNLOCK(kh);
1755 
1756 	return (ret);
1757 }
1758 
1759 /*
1760  * FIXME
1761  * This routine is inefficient -- locks the state only to unlock immediately on
1762  * return.
1763  * It is racy -- after the state is unlocked nothing stops other threads from
1764  * removing it.
1765  */
1766 bool
pf_find_state_all_exists(const struct pf_state_key_cmp * key,u_int dir)1767 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir)
1768 {
1769 	struct pf_kstate *s;
1770 
1771 	s = pf_find_state_all(key, dir, NULL);
1772 	if (s != NULL) {
1773 		PF_STATE_UNLOCK(s);
1774 		return (true);
1775 	}
1776 	return (false);
1777 }
1778 
1779 /* END state table stuff */
1780 
1781 static void
pf_send(struct pf_send_entry * pfse)1782 pf_send(struct pf_send_entry *pfse)
1783 {
1784 
1785 	PF_SENDQ_LOCK();
1786 	STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1787 	PF_SENDQ_UNLOCK();
1788 	swi_sched(V_pf_swi_cookie, 0);
1789 }
1790 
1791 static bool
pf_isforlocal(struct mbuf * m,int af)1792 pf_isforlocal(struct mbuf *m, int af)
1793 {
1794 	switch (af) {
1795 #ifdef INET
1796 	case AF_INET: {
1797 		struct ip *ip = mtod(m, struct ip *);
1798 
1799 		return (in_localip(ip->ip_dst));
1800 	}
1801 #endif
1802 #ifdef INET6
1803 	case AF_INET6: {
1804 		struct ip6_hdr *ip6;
1805 		struct in6_ifaddr *ia;
1806 		ip6 = mtod(m, struct ip6_hdr *);
1807 		ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1808 		if (ia == NULL)
1809 			return (false);
1810 		return (! (ia->ia6_flags & IN6_IFF_NOTREADY));
1811 	}
1812 #endif
1813 	default:
1814 		panic("Unsupported af %d", af);
1815 	}
1816 
1817 	return (false);
1818 }
1819 
1820 int
pf_icmp_mapping(struct pf_pdesc * pd,u_int8_t type,int * icmp_dir,int * multi,u_int16_t * virtual_id,u_int16_t * virtual_type)1821 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type,
1822     int *icmp_dir, int *multi, u_int16_t *virtual_id, u_int16_t *virtual_type)
1823 {
1824 	/*
1825 	 * ICMP types marked with PF_OUT are typically responses to
1826 	 * PF_IN, and will match states in the opposite direction.
1827 	 * PF_IN ICMP types need to match a state with that type.
1828 	 */
1829 	*icmp_dir = PF_OUT;
1830 	*multi = PF_ICMP_MULTI_LINK;
1831 	/* Queries (and responses) */
1832 	switch (pd->af) {
1833 #ifdef INET
1834 	case AF_INET:
1835 		switch (type) {
1836 		case ICMP_ECHO:
1837 			*icmp_dir = PF_IN;
1838 		case ICMP_ECHOREPLY:
1839 			*virtual_type = ICMP_ECHO;
1840 			*virtual_id = pd->hdr.icmp.icmp_id;
1841 			break;
1842 
1843 		case ICMP_TSTAMP:
1844 			*icmp_dir = PF_IN;
1845 		case ICMP_TSTAMPREPLY:
1846 			*virtual_type = ICMP_TSTAMP;
1847 			*virtual_id = pd->hdr.icmp.icmp_id;
1848 			break;
1849 
1850 		case ICMP_IREQ:
1851 			*icmp_dir = PF_IN;
1852 		case ICMP_IREQREPLY:
1853 			*virtual_type = ICMP_IREQ;
1854 			*virtual_id = pd->hdr.icmp.icmp_id;
1855 			break;
1856 
1857 		case ICMP_MASKREQ:
1858 			*icmp_dir = PF_IN;
1859 		case ICMP_MASKREPLY:
1860 			*virtual_type = ICMP_MASKREQ;
1861 			*virtual_id = pd->hdr.icmp.icmp_id;
1862 			break;
1863 
1864 		case ICMP_IPV6_WHEREAREYOU:
1865 			*icmp_dir = PF_IN;
1866 		case ICMP_IPV6_IAMHERE:
1867 			*virtual_type = ICMP_IPV6_WHEREAREYOU;
1868 			*virtual_id = 0; /* Nothing sane to match on! */
1869 			break;
1870 
1871 		case ICMP_MOBILE_REGREQUEST:
1872 			*icmp_dir = PF_IN;
1873 		case ICMP_MOBILE_REGREPLY:
1874 			*virtual_type = ICMP_MOBILE_REGREQUEST;
1875 			*virtual_id = 0; /* Nothing sane to match on! */
1876 			break;
1877 
1878 		case ICMP_ROUTERSOLICIT:
1879 			*icmp_dir = PF_IN;
1880 		case ICMP_ROUTERADVERT:
1881 			*virtual_type = ICMP_ROUTERSOLICIT;
1882 			*virtual_id = 0; /* Nothing sane to match on! */
1883 			break;
1884 
1885 		/* These ICMP types map to other connections */
1886 		case ICMP_UNREACH:
1887 		case ICMP_SOURCEQUENCH:
1888 		case ICMP_REDIRECT:
1889 		case ICMP_TIMXCEED:
1890 		case ICMP_PARAMPROB:
1891 			/* These will not be used, but set them anyway */
1892 			*icmp_dir = PF_IN;
1893 			*virtual_type = type;
1894 			*virtual_id = 0;
1895 			HTONS(*virtual_type);
1896 			return (1);  /* These types match to another state */
1897 
1898 		/*
1899 		 * All remaining ICMP types get their own states,
1900 		 * and will only match in one direction.
1901 		 */
1902 		default:
1903 			*icmp_dir = PF_IN;
1904 			*virtual_type = type;
1905 			*virtual_id = 0;
1906 			break;
1907 		}
1908 		break;
1909 #endif /* INET */
1910 #ifdef INET6
1911 	case AF_INET6:
1912 		switch (type) {
1913 		case ICMP6_ECHO_REQUEST:
1914 			*icmp_dir = PF_IN;
1915 		case ICMP6_ECHO_REPLY:
1916 			*virtual_type = ICMP6_ECHO_REQUEST;
1917 			*virtual_id = pd->hdr.icmp6.icmp6_id;
1918 			break;
1919 
1920 		case MLD_LISTENER_QUERY:
1921 		case MLD_LISTENER_REPORT: {
1922 			/*
1923 			 * Listener Report can be sent by clients
1924 			 * without an associated Listener Query.
1925 			 * In addition to that, when Report is sent as a
1926 			 * reply to a Query its source and destination
1927 			 * address are different.
1928 			 */
1929 			*icmp_dir = PF_IN;
1930 			*virtual_type = MLD_LISTENER_QUERY;
1931 			*virtual_id = 0;
1932 			break;
1933 		}
1934 		case MLD_MTRACE:
1935 			*icmp_dir = PF_IN;
1936 		case MLD_MTRACE_RESP:
1937 			*virtual_type = MLD_MTRACE;
1938 			*virtual_id = 0; /* Nothing sane to match on! */
1939 			break;
1940 
1941 		case ND_NEIGHBOR_SOLICIT:
1942 			*icmp_dir = PF_IN;
1943 		case ND_NEIGHBOR_ADVERT: {
1944 			*virtual_type = ND_NEIGHBOR_SOLICIT;
1945 			*virtual_id = 0;
1946 			break;
1947 		}
1948 
1949 		/*
1950 		 * These ICMP types map to other connections.
1951 		 * ND_REDIRECT can't be in this list because the triggering
1952 		 * packet header is optional.
1953 		 */
1954 		case ICMP6_DST_UNREACH:
1955 		case ICMP6_PACKET_TOO_BIG:
1956 		case ICMP6_TIME_EXCEEDED:
1957 		case ICMP6_PARAM_PROB:
1958 			/* These will not be used, but set them anyway */
1959 			*icmp_dir = PF_IN;
1960 			*virtual_type = type;
1961 			*virtual_id = 0;
1962 			HTONS(*virtual_type);
1963 			return (1);  /* These types match to another state */
1964 		/*
1965 		 * All remaining ICMP6 types get their own states,
1966 		 * and will only match in one direction.
1967 		 */
1968 		default:
1969 			*icmp_dir = PF_IN;
1970 			*virtual_type = type;
1971 			*virtual_id = 0;
1972 			break;
1973 		}
1974 		break;
1975 #endif /* INET6 */
1976 	default:
1977 		*icmp_dir = PF_IN;
1978 		*virtual_type = type;
1979 		*virtual_id = 0;
1980 		break;
1981 	}
1982 	HTONS(*virtual_type);
1983 	return (0);  /* These types match to their own state */
1984 }
1985 
1986 void
pf_intr(void * v)1987 pf_intr(void *v)
1988 {
1989 	struct epoch_tracker et;
1990 	struct pf_send_head queue;
1991 	struct pf_send_entry *pfse, *next;
1992 
1993 	CURVNET_SET((struct vnet *)v);
1994 
1995 	PF_SENDQ_LOCK();
1996 	queue = V_pf_sendqueue;
1997 	STAILQ_INIT(&V_pf_sendqueue);
1998 	PF_SENDQ_UNLOCK();
1999 
2000 	NET_EPOCH_ENTER(et);
2001 
2002 	STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
2003 		switch (pfse->pfse_type) {
2004 #ifdef INET
2005 		case PFSE_IP: {
2006 			if (pf_isforlocal(pfse->pfse_m, AF_INET)) {
2007 				pfse->pfse_m->m_flags |= M_SKIP_FIREWALL;
2008 				pfse->pfse_m->m_pkthdr.csum_flags |=
2009 				    CSUM_IP_VALID | CSUM_IP_CHECKED;
2010 				ip_input(pfse->pfse_m);
2011 			} else {
2012 				ip_output(pfse->pfse_m, NULL, NULL, 0, NULL,
2013 				    NULL);
2014 			}
2015 			break;
2016 		}
2017 		case PFSE_ICMP:
2018 			icmp_error(pfse->pfse_m, pfse->icmpopts.type,
2019 			    pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
2020 			break;
2021 #endif /* INET */
2022 #ifdef INET6
2023 		case PFSE_IP6:
2024 			if (pf_isforlocal(pfse->pfse_m, AF_INET6)) {
2025 				pfse->pfse_m->m_flags |= M_SKIP_FIREWALL;
2026 				ip6_input(pfse->pfse_m);
2027 			} else {
2028 				ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL,
2029 				    NULL, NULL);
2030 			}
2031 			break;
2032 		case PFSE_ICMP6:
2033 			icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
2034 			    pfse->icmpopts.code, pfse->icmpopts.mtu);
2035 			break;
2036 #endif /* INET6 */
2037 		default:
2038 			panic("%s: unknown type", __func__);
2039 		}
2040 		free(pfse, M_PFTEMP);
2041 	}
2042 	NET_EPOCH_EXIT(et);
2043 	CURVNET_RESTORE();
2044 }
2045 
2046 #define	pf_purge_thread_period	(hz / 10)
2047 
2048 #ifdef PF_WANT_32_TO_64_COUNTER
2049 static void
pf_status_counter_u64_periodic(void)2050 pf_status_counter_u64_periodic(void)
2051 {
2052 
2053 	PF_RULES_RASSERT();
2054 
2055 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) {
2056 		return;
2057 	}
2058 
2059 	for (int i = 0; i < FCNT_MAX; i++) {
2060 		pf_counter_u64_periodic(&V_pf_status.fcounters[i]);
2061 	}
2062 }
2063 
2064 static void
pf_kif_counter_u64_periodic(void)2065 pf_kif_counter_u64_periodic(void)
2066 {
2067 	struct pfi_kkif *kif;
2068 	size_t r, run;
2069 
2070 	PF_RULES_RASSERT();
2071 
2072 	if (__predict_false(V_pf_allkifcount == 0)) {
2073 		return;
2074 	}
2075 
2076 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) {
2077 		return;
2078 	}
2079 
2080 	run = V_pf_allkifcount / 10;
2081 	if (run < 5)
2082 		run = 5;
2083 
2084 	for (r = 0; r < run; r++) {
2085 		kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist);
2086 		if (kif == NULL) {
2087 			LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist);
2088 			LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist);
2089 			break;
2090 		}
2091 
2092 		LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist);
2093 		LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist);
2094 
2095 		for (int i = 0; i < 2; i++) {
2096 			for (int j = 0; j < 2; j++) {
2097 				for (int k = 0; k < 2; k++) {
2098 					pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]);
2099 					pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]);
2100 				}
2101 			}
2102 		}
2103 	}
2104 }
2105 
2106 static void
pf_rule_counter_u64_periodic(void)2107 pf_rule_counter_u64_periodic(void)
2108 {
2109 	struct pf_krule *rule;
2110 	size_t r, run;
2111 
2112 	PF_RULES_RASSERT();
2113 
2114 	if (__predict_false(V_pf_allrulecount == 0)) {
2115 		return;
2116 	}
2117 
2118 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) {
2119 		return;
2120 	}
2121 
2122 	run = V_pf_allrulecount / 10;
2123 	if (run < 5)
2124 		run = 5;
2125 
2126 	for (r = 0; r < run; r++) {
2127 		rule = LIST_NEXT(V_pf_rulemarker, allrulelist);
2128 		if (rule == NULL) {
2129 			LIST_REMOVE(V_pf_rulemarker, allrulelist);
2130 			LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist);
2131 			break;
2132 		}
2133 
2134 		LIST_REMOVE(V_pf_rulemarker, allrulelist);
2135 		LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist);
2136 
2137 		pf_counter_u64_periodic(&rule->evaluations);
2138 		for (int i = 0; i < 2; i++) {
2139 			pf_counter_u64_periodic(&rule->packets[i]);
2140 			pf_counter_u64_periodic(&rule->bytes[i]);
2141 		}
2142 	}
2143 }
2144 
2145 static void
pf_counter_u64_periodic_main(void)2146 pf_counter_u64_periodic_main(void)
2147 {
2148 	PF_RULES_RLOCK_TRACKER;
2149 
2150 	V_pf_counter_periodic_iter++;
2151 
2152 	PF_RULES_RLOCK();
2153 	pf_counter_u64_critical_enter();
2154 	pf_status_counter_u64_periodic();
2155 	pf_kif_counter_u64_periodic();
2156 	pf_rule_counter_u64_periodic();
2157 	pf_counter_u64_critical_exit();
2158 	PF_RULES_RUNLOCK();
2159 }
2160 #else
2161 #define	pf_counter_u64_periodic_main()	do { } while (0)
2162 #endif
2163 
2164 void
pf_purge_thread(void * unused __unused)2165 pf_purge_thread(void *unused __unused)
2166 {
2167 	VNET_ITERATOR_DECL(vnet_iter);
2168 
2169 	sx_xlock(&pf_end_lock);
2170 	while (pf_end_threads == 0) {
2171 		sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period);
2172 
2173 		VNET_LIST_RLOCK();
2174 		VNET_FOREACH(vnet_iter) {
2175 			CURVNET_SET(vnet_iter);
2176 
2177 			/* Wait until V_pf_default_rule is initialized. */
2178 			if (V_pf_vnet_active == 0) {
2179 				CURVNET_RESTORE();
2180 				continue;
2181 			}
2182 
2183 			pf_counter_u64_periodic_main();
2184 
2185 			/*
2186 			 *  Process 1/interval fraction of the state
2187 			 * table every run.
2188 			 */
2189 			V_pf_purge_idx =
2190 			    pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask /
2191 			    (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
2192 
2193 			/*
2194 			 * Purge other expired types every
2195 			 * PFTM_INTERVAL seconds.
2196 			 */
2197 			if (V_pf_purge_idx == 0) {
2198 				/*
2199 				 * Order is important:
2200 				 * - states and src nodes reference rules
2201 				 * - states and rules reference kifs
2202 				 */
2203 				pf_purge_expired_fragments();
2204 				pf_purge_expired_src_nodes();
2205 				pf_purge_unlinked_rules();
2206 				pfi_kkif_purge();
2207 			}
2208 			CURVNET_RESTORE();
2209 		}
2210 		VNET_LIST_RUNLOCK();
2211 	}
2212 
2213 	pf_end_threads++;
2214 	sx_xunlock(&pf_end_lock);
2215 	kproc_exit(0);
2216 }
2217 
2218 void
pf_unload_vnet_purge(void)2219 pf_unload_vnet_purge(void)
2220 {
2221 
2222 	/*
2223 	 * To cleanse up all kifs and rules we need
2224 	 * two runs: first one clears reference flags,
2225 	 * then pf_purge_expired_states() doesn't
2226 	 * raise them, and then second run frees.
2227 	 */
2228 	pf_purge_unlinked_rules();
2229 	pfi_kkif_purge();
2230 
2231 	/*
2232 	 * Now purge everything.
2233 	 */
2234 	pf_purge_expired_states(0, V_pf_hashmask);
2235 	pf_purge_fragments(UINT_MAX);
2236 	pf_purge_expired_src_nodes();
2237 
2238 	/*
2239 	 * Now all kifs & rules should be unreferenced,
2240 	 * thus should be successfully freed.
2241 	 */
2242 	pf_purge_unlinked_rules();
2243 	pfi_kkif_purge();
2244 }
2245 
2246 u_int32_t
pf_state_expires(const struct pf_kstate * state)2247 pf_state_expires(const struct pf_kstate *state)
2248 {
2249 	u_int32_t	timeout;
2250 	u_int32_t	start;
2251 	u_int32_t	end;
2252 	u_int32_t	states;
2253 
2254 	/* handle all PFTM_* > PFTM_MAX here */
2255 	if (state->timeout == PFTM_PURGE)
2256 		return (time_uptime);
2257 	KASSERT(state->timeout != PFTM_UNLINKED,
2258 	    ("pf_state_expires: timeout == PFTM_UNLINKED"));
2259 	KASSERT((state->timeout < PFTM_MAX),
2260 	    ("pf_state_expires: timeout > PFTM_MAX"));
2261 	timeout = state->rule.ptr->timeout[state->timeout];
2262 	if (!timeout)
2263 		timeout = V_pf_default_rule.timeout[state->timeout];
2264 	start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
2265 	if (start && state->rule.ptr != &V_pf_default_rule) {
2266 		end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
2267 		states = counter_u64_fetch(state->rule.ptr->states_cur);
2268 	} else {
2269 		start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
2270 		end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
2271 		states = V_pf_status.states;
2272 	}
2273 	if (end && states > start && start < end) {
2274 		if (states < end) {
2275 			timeout = (u_int64_t)timeout * (end - states) /
2276 			    (end - start);
2277 			return (state->expire + timeout);
2278 		}
2279 		else
2280 			return (time_uptime);
2281 	}
2282 	return (state->expire + timeout);
2283 }
2284 
2285 void
pf_purge_expired_src_nodes(void)2286 pf_purge_expired_src_nodes(void)
2287 {
2288 	struct pf_ksrc_node_list	 freelist;
2289 	struct pf_srchash	*sh;
2290 	struct pf_ksrc_node	*cur, *next;
2291 	int i;
2292 
2293 	LIST_INIT(&freelist);
2294 	for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
2295 	    PF_HASHROW_LOCK(sh);
2296 	    LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
2297 		if (cur->states == 0 && cur->expire <= time_uptime) {
2298 			pf_unlink_src_node(cur);
2299 			LIST_INSERT_HEAD(&freelist, cur, entry);
2300 		} else if (cur->rule.ptr != NULL)
2301 			cur->rule.ptr->rule_ref |= PFRULE_REFS;
2302 	    PF_HASHROW_UNLOCK(sh);
2303 	}
2304 
2305 	pf_free_src_nodes(&freelist);
2306 
2307 	V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
2308 }
2309 
2310 static void
pf_src_tree_remove_state(struct pf_kstate * s)2311 pf_src_tree_remove_state(struct pf_kstate *s)
2312 {
2313 	struct pf_ksrc_node *sn;
2314 	uint32_t timeout;
2315 
2316 	timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
2317 	    s->rule.ptr->timeout[PFTM_SRC_NODE] :
2318 	    V_pf_default_rule.timeout[PFTM_SRC_NODE];
2319 
2320 	if (s->src_node != NULL) {
2321 		sn = s->src_node;
2322 		PF_SRC_NODE_LOCK(sn);
2323 		if (s->src.tcp_est)
2324 			--sn->conn;
2325 		if (--sn->states == 0)
2326 			sn->expire = time_uptime + timeout;
2327 		PF_SRC_NODE_UNLOCK(sn);
2328 	}
2329 	if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
2330 		sn = s->nat_src_node;
2331 		PF_SRC_NODE_LOCK(sn);
2332 		if (--sn->states == 0)
2333 			sn->expire = time_uptime + timeout;
2334 		PF_SRC_NODE_UNLOCK(sn);
2335 	}
2336 	s->src_node = s->nat_src_node = NULL;
2337 }
2338 
2339 /*
2340  * Unlink and potentilly free a state. Function may be
2341  * called with ID hash row locked, but always returns
2342  * unlocked, since it needs to go through key hash locking.
2343  */
2344 int
pf_unlink_state(struct pf_kstate * s)2345 pf_unlink_state(struct pf_kstate *s)
2346 {
2347 	struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
2348 
2349 	PF_HASHROW_ASSERT(ih);
2350 
2351 	if (s->timeout == PFTM_UNLINKED) {
2352 		/*
2353 		 * State is being processed
2354 		 * by pf_unlink_state() in
2355 		 * an other thread.
2356 		 */
2357 		PF_HASHROW_UNLOCK(ih);
2358 		return (0);	/* XXXGL: undefined actually */
2359 	}
2360 
2361 	if (s->src.state == PF_TCPS_PROXY_DST) {
2362 		/* XXX wire key the right one? */
2363 		pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af,
2364 		    &s->key[PF_SK_WIRE]->addr[1],
2365 		    &s->key[PF_SK_WIRE]->addr[0],
2366 		    s->key[PF_SK_WIRE]->port[1],
2367 		    s->key[PF_SK_WIRE]->port[0],
2368 		    s->src.seqhi, s->src.seqlo + 1,
2369 		    TH_RST|TH_ACK, 0, 0, 0, true, s->tag, 0, s->act.rtableid);
2370 	}
2371 
2372 	LIST_REMOVE(s, entry);
2373 	pf_src_tree_remove_state(s);
2374 
2375 	if (V_pfsync_delete_state_ptr != NULL)
2376 		V_pfsync_delete_state_ptr(s);
2377 
2378 	STATE_DEC_COUNTERS(s);
2379 
2380 	s->timeout = PFTM_UNLINKED;
2381 
2382 	/* Ensure we remove it from the list of halfopen states, if needed. */
2383 	if (s->key[PF_SK_STACK] != NULL &&
2384 	    s->key[PF_SK_STACK]->proto == IPPROTO_TCP)
2385 		pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED);
2386 
2387 	PF_HASHROW_UNLOCK(ih);
2388 
2389 	pf_detach_state(s);
2390 	/* pf_state_insert() initialises refs to 2 */
2391 	return (pf_release_staten(s, 2));
2392 }
2393 
2394 struct pf_kstate *
pf_alloc_state(int flags)2395 pf_alloc_state(int flags)
2396 {
2397 
2398 	return (uma_zalloc(V_pf_state_z, flags | M_ZERO));
2399 }
2400 
2401 void
pf_free_state(struct pf_kstate * cur)2402 pf_free_state(struct pf_kstate *cur)
2403 {
2404 	struct pf_krule_item *ri;
2405 
2406 	KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
2407 	KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
2408 	    cur->timeout));
2409 
2410 	while ((ri = SLIST_FIRST(&cur->match_rules))) {
2411 		SLIST_REMOVE_HEAD(&cur->match_rules, entry);
2412 		free(ri, M_PF_RULE_ITEM);
2413 	}
2414 
2415 	pf_normalize_tcp_cleanup(cur);
2416 	uma_zfree(V_pf_state_z, cur);
2417 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
2418 }
2419 
2420 /*
2421  * Called only from pf_purge_thread(), thus serialized.
2422  */
2423 static u_int
pf_purge_expired_states(u_int i,int maxcheck)2424 pf_purge_expired_states(u_int i, int maxcheck)
2425 {
2426 	struct pf_idhash *ih;
2427 	struct pf_kstate *s;
2428 	struct pf_krule_item *mrm;
2429 	size_t count __unused;
2430 
2431 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2432 
2433 	/*
2434 	 * Go through hash and unlink states that expire now.
2435 	 */
2436 	while (maxcheck > 0) {
2437 		count = 0;
2438 		ih = &V_pf_idhash[i];
2439 
2440 		/* only take the lock if we expect to do work */
2441 		if (!LIST_EMPTY(&ih->states)) {
2442 relock:
2443 			PF_HASHROW_LOCK(ih);
2444 			LIST_FOREACH(s, &ih->states, entry) {
2445 				if (pf_state_expires(s) <= time_uptime) {
2446 					V_pf_status.states -=
2447 					    pf_unlink_state(s);
2448 					goto relock;
2449 				}
2450 				s->rule.ptr->rule_ref |= PFRULE_REFS;
2451 				if (s->nat_rule.ptr != NULL)
2452 					s->nat_rule.ptr->rule_ref |= PFRULE_REFS;
2453 				if (s->anchor.ptr != NULL)
2454 					s->anchor.ptr->rule_ref |= PFRULE_REFS;
2455 				s->kif->pfik_flags |= PFI_IFLAG_REFS;
2456 				SLIST_FOREACH(mrm, &s->match_rules, entry)
2457 					mrm->r->rule_ref |= PFRULE_REFS;
2458 				if (s->rt_kif)
2459 					s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
2460 				count++;
2461 			}
2462 			PF_HASHROW_UNLOCK(ih);
2463 		}
2464 
2465 		SDT_PROBE2(pf, purge, state, rowcount, i, count);
2466 
2467 		/* Return when we hit end of hash. */
2468 		if (++i > V_pf_hashmask) {
2469 			V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2470 			return (0);
2471 		}
2472 
2473 		maxcheck--;
2474 	}
2475 
2476 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2477 
2478 	return (i);
2479 }
2480 
2481 static void
pf_purge_unlinked_rules(void)2482 pf_purge_unlinked_rules(void)
2483 {
2484 	struct pf_krulequeue tmpq;
2485 	struct pf_krule *r, *r1;
2486 
2487 	/*
2488 	 * If we have overloading task pending, then we'd
2489 	 * better skip purging this time. There is a tiny
2490 	 * probability that overloading task references
2491 	 * an already unlinked rule.
2492 	 */
2493 	PF_OVERLOADQ_LOCK();
2494 	if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
2495 		PF_OVERLOADQ_UNLOCK();
2496 		return;
2497 	}
2498 	PF_OVERLOADQ_UNLOCK();
2499 
2500 	/*
2501 	 * Do naive mark-and-sweep garbage collecting of old rules.
2502 	 * Reference flag is raised by pf_purge_expired_states()
2503 	 * and pf_purge_expired_src_nodes().
2504 	 *
2505 	 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
2506 	 * use a temporary queue.
2507 	 */
2508 	TAILQ_INIT(&tmpq);
2509 	PF_UNLNKDRULES_LOCK();
2510 	TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
2511 		if (!(r->rule_ref & PFRULE_REFS)) {
2512 			TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
2513 			TAILQ_INSERT_TAIL(&tmpq, r, entries);
2514 		} else
2515 			r->rule_ref &= ~PFRULE_REFS;
2516 	}
2517 	PF_UNLNKDRULES_UNLOCK();
2518 
2519 	if (!TAILQ_EMPTY(&tmpq)) {
2520 		PF_CONFIG_LOCK();
2521 		PF_RULES_WLOCK();
2522 		TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
2523 			TAILQ_REMOVE(&tmpq, r, entries);
2524 			pf_free_rule(r);
2525 		}
2526 		PF_RULES_WUNLOCK();
2527 		PF_CONFIG_UNLOCK();
2528 	}
2529 }
2530 
2531 void
pf_print_host(struct pf_addr * addr,u_int16_t p,sa_family_t af)2532 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
2533 {
2534 	switch (af) {
2535 #ifdef INET
2536 	case AF_INET: {
2537 		u_int32_t a = ntohl(addr->addr32[0]);
2538 		printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
2539 		    (a>>8)&255, a&255);
2540 		if (p) {
2541 			p = ntohs(p);
2542 			printf(":%u", p);
2543 		}
2544 		break;
2545 	}
2546 #endif /* INET */
2547 #ifdef INET6
2548 	case AF_INET6: {
2549 		u_int16_t b;
2550 		u_int8_t i, curstart, curend, maxstart, maxend;
2551 		curstart = curend = maxstart = maxend = 255;
2552 		for (i = 0; i < 8; i++) {
2553 			if (!addr->addr16[i]) {
2554 				if (curstart == 255)
2555 					curstart = i;
2556 				curend = i;
2557 			} else {
2558 				if ((curend - curstart) >
2559 				    (maxend - maxstart)) {
2560 					maxstart = curstart;
2561 					maxend = curend;
2562 				}
2563 				curstart = curend = 255;
2564 			}
2565 		}
2566 		if ((curend - curstart) >
2567 		    (maxend - maxstart)) {
2568 			maxstart = curstart;
2569 			maxend = curend;
2570 		}
2571 		for (i = 0; i < 8; i++) {
2572 			if (i >= maxstart && i <= maxend) {
2573 				if (i == 0)
2574 					printf(":");
2575 				if (i == maxend)
2576 					printf(":");
2577 			} else {
2578 				b = ntohs(addr->addr16[i]);
2579 				printf("%x", b);
2580 				if (i < 7)
2581 					printf(":");
2582 			}
2583 		}
2584 		if (p) {
2585 			p = ntohs(p);
2586 			printf("[%u]", p);
2587 		}
2588 		break;
2589 	}
2590 #endif /* INET6 */
2591 	}
2592 }
2593 
2594 void
pf_print_state(struct pf_kstate * s)2595 pf_print_state(struct pf_kstate *s)
2596 {
2597 	pf_print_state_parts(s, NULL, NULL);
2598 }
2599 
2600 static void
pf_print_state_parts(struct pf_kstate * s,struct pf_state_key * skwp,struct pf_state_key * sksp)2601 pf_print_state_parts(struct pf_kstate *s,
2602     struct pf_state_key *skwp, struct pf_state_key *sksp)
2603 {
2604 	struct pf_state_key *skw, *sks;
2605 	u_int8_t proto, dir;
2606 
2607 	/* Do our best to fill these, but they're skipped if NULL */
2608 	skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
2609 	sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
2610 	proto = skw ? skw->proto : (sks ? sks->proto : 0);
2611 	dir = s ? s->direction : 0;
2612 
2613 	switch (proto) {
2614 	case IPPROTO_IPV4:
2615 		printf("IPv4");
2616 		break;
2617 	case IPPROTO_IPV6:
2618 		printf("IPv6");
2619 		break;
2620 	case IPPROTO_TCP:
2621 		printf("TCP");
2622 		break;
2623 	case IPPROTO_UDP:
2624 		printf("UDP");
2625 		break;
2626 	case IPPROTO_ICMP:
2627 		printf("ICMP");
2628 		break;
2629 	case IPPROTO_ICMPV6:
2630 		printf("ICMPv6");
2631 		break;
2632 	default:
2633 		printf("%u", proto);
2634 		break;
2635 	}
2636 	switch (dir) {
2637 	case PF_IN:
2638 		printf(" in");
2639 		break;
2640 	case PF_OUT:
2641 		printf(" out");
2642 		break;
2643 	}
2644 	if (skw) {
2645 		printf(" wire: ");
2646 		pf_print_host(&skw->addr[0], skw->port[0], skw->af);
2647 		printf(" ");
2648 		pf_print_host(&skw->addr[1], skw->port[1], skw->af);
2649 	}
2650 	if (sks) {
2651 		printf(" stack: ");
2652 		if (sks != skw) {
2653 			pf_print_host(&sks->addr[0], sks->port[0], sks->af);
2654 			printf(" ");
2655 			pf_print_host(&sks->addr[1], sks->port[1], sks->af);
2656 		} else
2657 			printf("-");
2658 	}
2659 	if (s) {
2660 		if (proto == IPPROTO_TCP) {
2661 			printf(" [lo=%u high=%u win=%u modulator=%u",
2662 			    s->src.seqlo, s->src.seqhi,
2663 			    s->src.max_win, s->src.seqdiff);
2664 			if (s->src.wscale && s->dst.wscale)
2665 				printf(" wscale=%u",
2666 				    s->src.wscale & PF_WSCALE_MASK);
2667 			printf("]");
2668 			printf(" [lo=%u high=%u win=%u modulator=%u",
2669 			    s->dst.seqlo, s->dst.seqhi,
2670 			    s->dst.max_win, s->dst.seqdiff);
2671 			if (s->src.wscale && s->dst.wscale)
2672 				printf(" wscale=%u",
2673 				s->dst.wscale & PF_WSCALE_MASK);
2674 			printf("]");
2675 		}
2676 		printf(" %u:%u", s->src.state, s->dst.state);
2677 	}
2678 }
2679 
2680 void
pf_print_flags(u_int8_t f)2681 pf_print_flags(u_int8_t f)
2682 {
2683 	if (f)
2684 		printf(" ");
2685 	if (f & TH_FIN)
2686 		printf("F");
2687 	if (f & TH_SYN)
2688 		printf("S");
2689 	if (f & TH_RST)
2690 		printf("R");
2691 	if (f & TH_PUSH)
2692 		printf("P");
2693 	if (f & TH_ACK)
2694 		printf("A");
2695 	if (f & TH_URG)
2696 		printf("U");
2697 	if (f & TH_ECE)
2698 		printf("E");
2699 	if (f & TH_CWR)
2700 		printf("W");
2701 }
2702 
2703 #define	PF_SET_SKIP_STEPS(i)					\
2704 	do {							\
2705 		while (head[i] != cur) {			\
2706 			head[i]->skip[i].ptr = cur;		\
2707 			head[i] = TAILQ_NEXT(head[i], entries);	\
2708 		}						\
2709 	} while (0)
2710 
2711 void
pf_calc_skip_steps(struct pf_krulequeue * rules)2712 pf_calc_skip_steps(struct pf_krulequeue *rules)
2713 {
2714 	struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT];
2715 	int i;
2716 
2717 	cur = TAILQ_FIRST(rules);
2718 	prev = cur;
2719 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2720 		head[i] = cur;
2721 	while (cur != NULL) {
2722 		if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
2723 			PF_SET_SKIP_STEPS(PF_SKIP_IFP);
2724 		if (cur->direction != prev->direction)
2725 			PF_SET_SKIP_STEPS(PF_SKIP_DIR);
2726 		if (cur->af != prev->af)
2727 			PF_SET_SKIP_STEPS(PF_SKIP_AF);
2728 		if (cur->proto != prev->proto)
2729 			PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
2730 		if (cur->src.neg != prev->src.neg ||
2731 		    pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
2732 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
2733 		if (cur->src.port[0] != prev->src.port[0] ||
2734 		    cur->src.port[1] != prev->src.port[1] ||
2735 		    cur->src.port_op != prev->src.port_op)
2736 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
2737 		if (cur->dst.neg != prev->dst.neg ||
2738 		    pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
2739 			PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
2740 		if (cur->dst.port[0] != prev->dst.port[0] ||
2741 		    cur->dst.port[1] != prev->dst.port[1] ||
2742 		    cur->dst.port_op != prev->dst.port_op)
2743 			PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
2744 
2745 		prev = cur;
2746 		cur = TAILQ_NEXT(cur, entries);
2747 	}
2748 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2749 		PF_SET_SKIP_STEPS(i);
2750 }
2751 
2752 int
pf_addr_wrap_neq(struct pf_addr_wrap * aw1,struct pf_addr_wrap * aw2)2753 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
2754 {
2755 	if (aw1->type != aw2->type)
2756 		return (1);
2757 	switch (aw1->type) {
2758 	case PF_ADDR_ADDRMASK:
2759 	case PF_ADDR_RANGE:
2760 		if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
2761 			return (1);
2762 		if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
2763 			return (1);
2764 		return (0);
2765 	case PF_ADDR_DYNIFTL:
2766 		return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
2767 	case PF_ADDR_NOROUTE:
2768 	case PF_ADDR_URPFFAILED:
2769 		return (0);
2770 	case PF_ADDR_TABLE:
2771 		return (aw1->p.tbl != aw2->p.tbl);
2772 	default:
2773 		printf("invalid address type: %d\n", aw1->type);
2774 		return (1);
2775 	}
2776 }
2777 
2778 /**
2779  * Checksum updates are a little complicated because the checksum in the TCP/UDP
2780  * header isn't always a full checksum. In some cases (i.e. output) it's a
2781  * pseudo-header checksum, which is a partial checksum over src/dst IP
2782  * addresses, protocol number and length.
2783  *
2784  * That means we have the following cases:
2785  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2786  *  	checksums, we need to update the checksum whenever we change anything.
2787  *  * Output (i.e. the checksum is a pseudo-header checksum):
2788  *  	x The field being updated is src/dst address or affects the length of
2789  *  	the packet. We need to update the pseudo-header checksum (note that this
2790  *  	checksum is not ones' complement).
2791  *  	x Some other field is being modified (e.g. src/dst port numbers): We
2792  *  	don't have to update anything.
2793  **/
2794 u_int16_t
pf_cksum_fixup(u_int16_t cksum,u_int16_t old,u_int16_t new,u_int8_t udp)2795 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2796 {
2797 	u_int32_t x;
2798 
2799 	x = cksum + old - new;
2800 	x = (x + (x >> 16)) & 0xffff;
2801 
2802 	/* optimise: eliminate a branch when not udp */
2803 	if (udp && cksum == 0x0000)
2804 		return cksum;
2805 	if (udp && x == 0x0000)
2806 		x = 0xffff;
2807 
2808 	return (u_int16_t)(x);
2809 }
2810 
2811 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)2812 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi,
2813     u_int8_t udp)
2814 {
2815 	u_int16_t old = htons(hi ? (*f << 8) : *f);
2816 	u_int16_t new = htons(hi ? ( v << 8) :  v);
2817 
2818 	if (*f == v)
2819 		return;
2820 
2821 	*f = v;
2822 
2823 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2824 		return;
2825 
2826 	*cksum = pf_cksum_fixup(*cksum, old, new, udp);
2827 }
2828 
2829 void
pf_patch_16_unaligned(struct mbuf * m,u_int16_t * cksum,void * f,u_int16_t v,bool hi,u_int8_t udp)2830 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v,
2831     bool hi, u_int8_t udp)
2832 {
2833 	u_int8_t *fb = (u_int8_t *)f;
2834 	u_int8_t *vb = (u_int8_t *)&v;
2835 
2836 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2837 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2838 }
2839 
2840 void
pf_patch_32_unaligned(struct mbuf * m,u_int16_t * cksum,void * f,u_int32_t v,bool hi,u_int8_t udp)2841 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v,
2842     bool hi, u_int8_t udp)
2843 {
2844 	u_int8_t *fb = (u_int8_t *)f;
2845 	u_int8_t *vb = (u_int8_t *)&v;
2846 
2847 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2848 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2849 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2850 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2851 }
2852 
2853 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)2854 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2855         u_int16_t new, u_int8_t udp)
2856 {
2857 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2858 		return (cksum);
2859 
2860 	return (pf_cksum_fixup(cksum, old, new, udp));
2861 }
2862 
2863 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)2864 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2865         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2866         sa_family_t af)
2867 {
2868 	struct pf_addr	ao;
2869 	u_int16_t	po = *p;
2870 
2871 	PF_ACPY(&ao, a, af);
2872 	PF_ACPY(a, an, af);
2873 
2874 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2875 		*pc = ~*pc;
2876 
2877 	*p = pn;
2878 
2879 	switch (af) {
2880 #ifdef INET
2881 	case AF_INET:
2882 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2883 		    ao.addr16[0], an->addr16[0], 0),
2884 		    ao.addr16[1], an->addr16[1], 0);
2885 		*p = pn;
2886 
2887 		*pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2888 		    ao.addr16[0], an->addr16[0], u),
2889 		    ao.addr16[1], an->addr16[1], u);
2890 
2891 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2892 		break;
2893 #endif /* INET */
2894 #ifdef INET6
2895 	case AF_INET6:
2896 		*pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2897 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2898 		    pf_cksum_fixup(pf_cksum_fixup(*pc,
2899 		    ao.addr16[0], an->addr16[0], u),
2900 		    ao.addr16[1], an->addr16[1], u),
2901 		    ao.addr16[2], an->addr16[2], u),
2902 		    ao.addr16[3], an->addr16[3], u),
2903 		    ao.addr16[4], an->addr16[4], u),
2904 		    ao.addr16[5], an->addr16[5], u),
2905 		    ao.addr16[6], an->addr16[6], u),
2906 		    ao.addr16[7], an->addr16[7], u);
2907 
2908 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2909 		break;
2910 #endif /* INET6 */
2911 	}
2912 
2913 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2914 	    CSUM_DELAY_DATA_IPV6)) {
2915 		*pc = ~*pc;
2916 		if (! *pc)
2917 			*pc = 0xffff;
2918 	}
2919 }
2920 
2921 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2922 void
pf_change_a(void * a,u_int16_t * c,u_int32_t an,u_int8_t u)2923 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2924 {
2925 	u_int32_t	ao;
2926 
2927 	memcpy(&ao, a, sizeof(ao));
2928 	memcpy(a, &an, sizeof(u_int32_t));
2929 	*c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2930 	    ao % 65536, an % 65536, u);
2931 }
2932 
2933 void
pf_change_proto_a(struct mbuf * m,void * a,u_int16_t * c,u_int32_t an,u_int8_t udp)2934 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2935 {
2936 	u_int32_t	ao;
2937 
2938 	memcpy(&ao, a, sizeof(ao));
2939 	memcpy(a, &an, sizeof(u_int32_t));
2940 
2941 	*c = pf_proto_cksum_fixup(m,
2942 	    pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2943 	    ao % 65536, an % 65536, udp);
2944 }
2945 
2946 #ifdef INET6
2947 static void
pf_change_a6(struct pf_addr * a,u_int16_t * c,struct pf_addr * an,u_int8_t u)2948 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2949 {
2950 	struct pf_addr	ao;
2951 
2952 	PF_ACPY(&ao, a, AF_INET6);
2953 	PF_ACPY(a, an, AF_INET6);
2954 
2955 	*c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2956 	    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2957 	    pf_cksum_fixup(pf_cksum_fixup(*c,
2958 	    ao.addr16[0], an->addr16[0], u),
2959 	    ao.addr16[1], an->addr16[1], u),
2960 	    ao.addr16[2], an->addr16[2], u),
2961 	    ao.addr16[3], an->addr16[3], u),
2962 	    ao.addr16[4], an->addr16[4], u),
2963 	    ao.addr16[5], an->addr16[5], u),
2964 	    ao.addr16[6], an->addr16[6], u),
2965 	    ao.addr16[7], an->addr16[7], u);
2966 }
2967 #endif /* INET6 */
2968 
2969 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)2970 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2971     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2972     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2973 {
2974 	struct pf_addr	oia, ooa;
2975 
2976 	PF_ACPY(&oia, ia, af);
2977 	if (oa)
2978 		PF_ACPY(&ooa, oa, af);
2979 
2980 	/* Change inner protocol port, fix inner protocol checksum. */
2981 	if (ip != NULL) {
2982 		u_int16_t	oip = *ip;
2983 		u_int32_t	opc;
2984 
2985 		if (pc != NULL)
2986 			opc = *pc;
2987 		*ip = np;
2988 		if (pc != NULL)
2989 			*pc = pf_cksum_fixup(*pc, oip, *ip, u);
2990 		*ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2991 		if (pc != NULL)
2992 			*ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2993 	}
2994 	/* Change inner ip address, fix inner ip and icmp checksums. */
2995 	PF_ACPY(ia, na, af);
2996 	switch (af) {
2997 #ifdef INET
2998 	case AF_INET: {
2999 		u_int32_t	 oh2c = *h2c;
3000 
3001 		*h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
3002 		    oia.addr16[0], ia->addr16[0], 0),
3003 		    oia.addr16[1], ia->addr16[1], 0);
3004 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
3005 		    oia.addr16[0], ia->addr16[0], 0),
3006 		    oia.addr16[1], ia->addr16[1], 0);
3007 		*ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
3008 		break;
3009 	}
3010 #endif /* INET */
3011 #ifdef INET6
3012 	case AF_INET6:
3013 		*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
3014 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
3015 		    pf_cksum_fixup(pf_cksum_fixup(*ic,
3016 		    oia.addr16[0], ia->addr16[0], u),
3017 		    oia.addr16[1], ia->addr16[1], u),
3018 		    oia.addr16[2], ia->addr16[2], u),
3019 		    oia.addr16[3], ia->addr16[3], u),
3020 		    oia.addr16[4], ia->addr16[4], u),
3021 		    oia.addr16[5], ia->addr16[5], u),
3022 		    oia.addr16[6], ia->addr16[6], u),
3023 		    oia.addr16[7], ia->addr16[7], u);
3024 		break;
3025 #endif /* INET6 */
3026 	}
3027 	/* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
3028 	if (oa) {
3029 		PF_ACPY(oa, na, af);
3030 		switch (af) {
3031 #ifdef INET
3032 		case AF_INET:
3033 			*hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
3034 			    ooa.addr16[0], oa->addr16[0], 0),
3035 			    ooa.addr16[1], oa->addr16[1], 0);
3036 			break;
3037 #endif /* INET */
3038 #ifdef INET6
3039 		case AF_INET6:
3040 			*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
3041 			    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
3042 			    pf_cksum_fixup(pf_cksum_fixup(*ic,
3043 			    ooa.addr16[0], oa->addr16[0], u),
3044 			    ooa.addr16[1], oa->addr16[1], u),
3045 			    ooa.addr16[2], oa->addr16[2], u),
3046 			    ooa.addr16[3], oa->addr16[3], u),
3047 			    ooa.addr16[4], oa->addr16[4], u),
3048 			    ooa.addr16[5], oa->addr16[5], u),
3049 			    ooa.addr16[6], oa->addr16[6], u),
3050 			    ooa.addr16[7], oa->addr16[7], u);
3051 			break;
3052 #endif /* INET6 */
3053 		}
3054 	}
3055 }
3056 
3057 /*
3058  * Need to modulate the sequence numbers in the TCP SACK option
3059  * (credits to Krzysztof Pfaff for report and patch)
3060  */
3061 static int
pf_modulate_sack(struct mbuf * m,int off,struct pf_pdesc * pd,struct tcphdr * th,struct pf_state_peer * dst)3062 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
3063     struct tcphdr *th, struct pf_state_peer *dst)
3064 {
3065 	int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
3066 	u_int8_t opts[TCP_MAXOLEN], *opt = opts;
3067 	int copyback = 0, i, olen;
3068 	struct sackblk sack;
3069 
3070 #define	TCPOLEN_SACKLEN	(TCPOLEN_SACK + 2)
3071 	if (hlen < TCPOLEN_SACKLEN ||
3072 	    !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
3073 		return 0;
3074 
3075 	while (hlen >= TCPOLEN_SACKLEN) {
3076 		size_t startoff = opt - opts;
3077 		olen = opt[1];
3078 		switch (*opt) {
3079 		case TCPOPT_EOL:	/* FALLTHROUGH */
3080 		case TCPOPT_NOP:
3081 			opt++;
3082 			hlen--;
3083 			break;
3084 		case TCPOPT_SACK:
3085 			if (olen > hlen)
3086 				olen = hlen;
3087 			if (olen >= TCPOLEN_SACKLEN) {
3088 				for (i = 2; i + TCPOLEN_SACK <= olen;
3089 				    i += TCPOLEN_SACK) {
3090 					memcpy(&sack, &opt[i], sizeof(sack));
3091 					pf_patch_32_unaligned(m,
3092 					    &th->th_sum, &sack.start,
3093 					    htonl(ntohl(sack.start) - dst->seqdiff),
3094 					    PF_ALGNMNT(startoff),
3095 					    0);
3096 					pf_patch_32_unaligned(m, &th->th_sum,
3097 					    &sack.end,
3098 					    htonl(ntohl(sack.end) - dst->seqdiff),
3099 					    PF_ALGNMNT(startoff),
3100 					    0);
3101 					memcpy(&opt[i], &sack, sizeof(sack));
3102 				}
3103 				copyback = 1;
3104 			}
3105 			/* FALLTHROUGH */
3106 		default:
3107 			if (olen < 2)
3108 				olen = 2;
3109 			hlen -= olen;
3110 			opt += olen;
3111 		}
3112 	}
3113 
3114 	if (copyback)
3115 		m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
3116 	return (copyback);
3117 }
3118 
3119 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 tcp_flags,u_int16_t win,u_int16_t mss,u_int8_t ttl,bool skip_firewall,u_int16_t mtag_tag,u_int16_t mtag_flags,int rtableid)3120 pf_build_tcp(const struct pf_krule *r, sa_family_t af,
3121     const struct pf_addr *saddr, const struct pf_addr *daddr,
3122     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
3123     u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl,
3124     bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid)
3125 {
3126 	struct mbuf	*m;
3127 	int		 len, tlen;
3128 #ifdef INET
3129 	struct ip	*h = NULL;
3130 #endif /* INET */
3131 #ifdef INET6
3132 	struct ip6_hdr	*h6 = NULL;
3133 #endif /* INET6 */
3134 	struct tcphdr	*th;
3135 	char		*opt;
3136 	struct pf_mtag  *pf_mtag;
3137 
3138 	len = 0;
3139 	th = NULL;
3140 
3141 	/* maximum segment size tcp option */
3142 	tlen = sizeof(struct tcphdr);
3143 	if (mss)
3144 		tlen += 4;
3145 
3146 	switch (af) {
3147 #ifdef INET
3148 	case AF_INET:
3149 		len = sizeof(struct ip) + tlen;
3150 		break;
3151 #endif /* INET */
3152 #ifdef INET6
3153 	case AF_INET6:
3154 		len = sizeof(struct ip6_hdr) + tlen;
3155 		break;
3156 #endif /* INET6 */
3157 	default:
3158 		panic("%s: unsupported af %d", __func__, af);
3159 	}
3160 
3161 	m = m_gethdr(M_NOWAIT, MT_DATA);
3162 	if (m == NULL)
3163 		return (NULL);
3164 
3165 #ifdef MAC
3166 	mac_netinet_firewall_send(m);
3167 #endif
3168 	if ((pf_mtag = pf_get_mtag(m)) == NULL) {
3169 		m_freem(m);
3170 		return (NULL);
3171 	}
3172 	if (skip_firewall)
3173 		m->m_flags |= M_SKIP_FIREWALL;
3174 	pf_mtag->tag = mtag_tag;
3175 	pf_mtag->flags = mtag_flags;
3176 
3177 	if (rtableid >= 0)
3178 		M_SETFIB(m, rtableid);
3179 
3180 #ifdef ALTQ
3181 	if (r != NULL && r->qid) {
3182 		pf_mtag->qid = r->qid;
3183 
3184 		/* add hints for ecn */
3185 		pf_mtag->hdr = mtod(m, struct ip *);
3186 	}
3187 #endif /* ALTQ */
3188 	m->m_data += max_linkhdr;
3189 	m->m_pkthdr.len = m->m_len = len;
3190 	/* The rest of the stack assumes a rcvif, so provide one.
3191 	 * This is a locally generated packet, so .. close enough. */
3192 	m->m_pkthdr.rcvif = V_loif;
3193 	bzero(m->m_data, len);
3194 	switch (af) {
3195 #ifdef INET
3196 	case AF_INET:
3197 		h = mtod(m, struct ip *);
3198 
3199 		/* IP header fields included in the TCP checksum */
3200 		h->ip_p = IPPROTO_TCP;
3201 		h->ip_len = htons(tlen);
3202 		h->ip_src.s_addr = saddr->v4.s_addr;
3203 		h->ip_dst.s_addr = daddr->v4.s_addr;
3204 
3205 		th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
3206 		break;
3207 #endif /* INET */
3208 #ifdef INET6
3209 	case AF_INET6:
3210 		h6 = mtod(m, struct ip6_hdr *);
3211 
3212 		/* IP header fields included in the TCP checksum */
3213 		h6->ip6_nxt = IPPROTO_TCP;
3214 		h6->ip6_plen = htons(tlen);
3215 		memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
3216 		memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
3217 
3218 		th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
3219 		break;
3220 #endif /* INET6 */
3221 	}
3222 
3223 	/* TCP header */
3224 	th->th_sport = sport;
3225 	th->th_dport = dport;
3226 	th->th_seq = htonl(seq);
3227 	th->th_ack = htonl(ack);
3228 	th->th_off = tlen >> 2;
3229 	th->th_flags = tcp_flags;
3230 	th->th_win = htons(win);
3231 
3232 	if (mss) {
3233 		opt = (char *)(th + 1);
3234 		opt[0] = TCPOPT_MAXSEG;
3235 		opt[1] = 4;
3236 		HTONS(mss);
3237 		bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
3238 	}
3239 
3240 	switch (af) {
3241 #ifdef INET
3242 	case AF_INET:
3243 		/* TCP checksum */
3244 		th->th_sum = in_cksum(m, len);
3245 
3246 		/* Finish the IP header */
3247 		h->ip_v = 4;
3248 		h->ip_hl = sizeof(*h) >> 2;
3249 		h->ip_tos = IPTOS_LOWDELAY;
3250 		h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
3251 		h->ip_len = htons(len);
3252 		h->ip_ttl = ttl ? ttl : V_ip_defttl;
3253 		h->ip_sum = 0;
3254 		break;
3255 #endif /* INET */
3256 #ifdef INET6
3257 	case AF_INET6:
3258 		/* TCP checksum */
3259 		th->th_sum = in6_cksum(m, IPPROTO_TCP,
3260 		    sizeof(struct ip6_hdr), tlen);
3261 
3262 		h6->ip6_vfc |= IPV6_VERSION;
3263 		h6->ip6_hlim = IPV6_DEFHLIM;
3264 		break;
3265 #endif /* INET6 */
3266 	}
3267 
3268 	return (m);
3269 }
3270 
3271 static void
pf_send_sctp_abort(sa_family_t af,struct pf_pdesc * pd,uint8_t ttl,int rtableid)3272 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd,
3273     uint8_t ttl, int rtableid)
3274 {
3275 	struct mbuf		*m;
3276 #ifdef INET
3277 	struct ip		*h = NULL;
3278 #endif /* INET */
3279 #ifdef INET6
3280 	struct ip6_hdr		*h6 = NULL;
3281 #endif /* INET6 */
3282 	struct sctphdr		*hdr;
3283 	struct sctp_chunkhdr	*chunk;
3284 	struct pf_send_entry	*pfse;
3285 	int			 off = 0;
3286 
3287 	MPASS(af == pd->af);
3288 
3289 	m = m_gethdr(M_NOWAIT, MT_DATA);
3290 	if (m == NULL)
3291 		return;
3292 
3293 	m->m_data += max_linkhdr;
3294 	m->m_flags |= M_SKIP_FIREWALL;
3295 	/* The rest of the stack assumes a rcvif, so provide one.
3296 	 * This is a locally generated packet, so .. close enough. */
3297 	m->m_pkthdr.rcvif = V_loif;
3298 
3299 	/* IPv4|6 header */
3300 	switch (af) {
3301 #ifdef INET
3302 	case AF_INET:
3303 		bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk));
3304 
3305 		h = mtod(m, struct ip *);
3306 
3307 		/* IP header fields included in the TCP checksum */
3308 
3309 		h->ip_p = IPPROTO_SCTP;
3310 		h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk));
3311 		h->ip_ttl = ttl ? ttl : V_ip_defttl;
3312 		h->ip_src = pd->dst->v4;
3313 		h->ip_dst = pd->src->v4;
3314 
3315 		off += sizeof(struct ip);
3316 		break;
3317 #endif /* INET */
3318 #ifdef INET6
3319 	case AF_INET6:
3320 		bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk));
3321 
3322 		h6 = mtod(m, struct ip6_hdr *);
3323 
3324 		/* IP header fields included in the TCP checksum */
3325 		h6->ip6_vfc |= IPV6_VERSION;
3326 		h6->ip6_nxt = IPPROTO_SCTP;
3327 		h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk));
3328 		h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim;
3329 		memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr));
3330 		memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr));
3331 
3332 		off += sizeof(struct ip6_hdr);
3333 		break;
3334 #endif /* INET6 */
3335 	}
3336 
3337 	/* SCTP header */
3338 	hdr = mtodo(m, off);
3339 
3340 	hdr->src_port = pd->hdr.sctp.dest_port;
3341 	hdr->dest_port = pd->hdr.sctp.src_port;
3342 	hdr->v_tag = pd->sctp_initiate_tag;
3343 	hdr->checksum = 0;
3344 
3345 	/* Abort chunk. */
3346 	off += sizeof(struct sctphdr);
3347 	chunk = mtodo(m, off);
3348 
3349 	chunk->chunk_type = SCTP_ABORT_ASSOCIATION;
3350 	chunk->chunk_length = htons(sizeof(*chunk));
3351 
3352 	/* SCTP checksum */
3353 	off += sizeof(*chunk);
3354 	m->m_pkthdr.len = m->m_len = off;
3355 
3356 	pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk));;
3357 
3358 	if (rtableid >= 0)
3359 		M_SETFIB(m, rtableid);
3360 
3361 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3362 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3363 	if (pfse == NULL) {
3364 		m_freem(m);
3365 		return;
3366 	}
3367 
3368 	switch (af) {
3369 #ifdef INET
3370 	case AF_INET:
3371 		pfse->pfse_type = PFSE_IP;
3372 		break;
3373 #endif /* INET */
3374 #ifdef INET6
3375 	case AF_INET6:
3376 		pfse->pfse_type = PFSE_IP6;
3377 		break;
3378 #endif /* INET6 */
3379 	}
3380 
3381 	pfse->pfse_m = m;
3382 	pf_send(pfse);
3383 }
3384 
3385 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 tcp_flags,u_int16_t win,u_int16_t mss,u_int8_t ttl,bool skip_firewall,u_int16_t mtag_tag,u_int16_t mtag_flags,int rtableid)3386 pf_send_tcp(const struct pf_krule *r, sa_family_t af,
3387     const struct pf_addr *saddr, const struct pf_addr *daddr,
3388     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
3389     u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl,
3390     bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid)
3391 {
3392 	struct pf_send_entry *pfse;
3393 	struct mbuf	*m;
3394 
3395 	m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags,
3396 	    win, mss, ttl, skip_firewall, mtag_tag, mtag_flags, rtableid);
3397 	if (m == NULL)
3398 		return;
3399 
3400 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3401 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3402 	if (pfse == NULL) {
3403 		m_freem(m);
3404 		return;
3405 	}
3406 
3407 	switch (af) {
3408 #ifdef INET
3409 	case AF_INET:
3410 		pfse->pfse_type = PFSE_IP;
3411 		break;
3412 #endif /* INET */
3413 #ifdef INET6
3414 	case AF_INET6:
3415 		pfse->pfse_type = PFSE_IP6;
3416 		break;
3417 #endif /* INET6 */
3418 	}
3419 
3420 	pfse->pfse_m = m;
3421 	pf_send(pfse);
3422 }
3423 
3424 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,int rtableid)3425 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd,
3426     struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th,
3427     struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
3428     u_short *reason, int rtableid)
3429 {
3430 	struct pf_addr	* const saddr = pd->src;
3431 	struct pf_addr	* const daddr = pd->dst;
3432 	sa_family_t	 af = pd->af;
3433 
3434 	/* undo NAT changes, if they have taken place */
3435 	if (nr != NULL) {
3436 		PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3437 		PF_ACPY(daddr, &sk->addr[pd->didx], af);
3438 		if (pd->sport)
3439 			*pd->sport = sk->port[pd->sidx];
3440 		if (pd->dport)
3441 			*pd->dport = sk->port[pd->didx];
3442 		if (pd->proto_sum)
3443 			*pd->proto_sum = bproto_sum;
3444 		if (pd->ip_sum)
3445 			*pd->ip_sum = bip_sum;
3446 		m_copyback(m, off, hdrlen, pd->hdr.any);
3447 	}
3448 	if (pd->proto == IPPROTO_TCP &&
3449 	    ((r->rule_flag & PFRULE_RETURNRST) ||
3450 	    (r->rule_flag & PFRULE_RETURN)) &&
3451 	    !(th->th_flags & TH_RST)) {
3452 		u_int32_t	 ack = ntohl(th->th_seq) + pd->p_len;
3453 		int		 len = 0;
3454 #ifdef INET
3455 		struct ip	*h4;
3456 #endif
3457 #ifdef INET6
3458 		struct ip6_hdr	*h6;
3459 #endif
3460 
3461 		switch (af) {
3462 #ifdef INET
3463 		case AF_INET:
3464 			h4 = mtod(m, struct ip *);
3465 			len = ntohs(h4->ip_len) - off;
3466 			break;
3467 #endif
3468 #ifdef INET6
3469 		case AF_INET6:
3470 			h6 = mtod(m, struct ip6_hdr *);
3471 			len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3472 			break;
3473 #endif
3474 		}
3475 
3476 		if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3477 			REASON_SET(reason, PFRES_PROTCKSUM);
3478 		else {
3479 			if (th->th_flags & TH_SYN)
3480 				ack++;
3481 			if (th->th_flags & TH_FIN)
3482 				ack++;
3483 			pf_send_tcp(r, af, pd->dst,
3484 				pd->src, th->th_dport, th->th_sport,
3485 				ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3486 				r->return_ttl, true, 0, 0, rtableid);
3487 		}
3488 	} else if (pd->proto == IPPROTO_SCTP &&
3489 	    (r->rule_flag & PFRULE_RETURN)) {
3490 		pf_send_sctp_abort(af, pd, r->return_ttl, rtableid);
3491 	} else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3492 		r->return_icmp)
3493 		pf_send_icmp(m, r->return_icmp >> 8,
3494 			r->return_icmp & 255, af, r, rtableid);
3495 	else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3496 		r->return_icmp6)
3497 		pf_send_icmp(m, r->return_icmp6 >> 8,
3498 			r->return_icmp6 & 255, af, r, rtableid);
3499 }
3500 
3501 static int
pf_match_ieee8021q_pcp(u_int8_t prio,struct mbuf * m)3502 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
3503 {
3504 	struct m_tag *mtag;
3505 	u_int8_t mpcp;
3506 
3507 	mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
3508 	if (mtag == NULL)
3509 		return (0);
3510 
3511 	if (prio == PF_PRIO_ZERO)
3512 		prio = 0;
3513 
3514 	mpcp = *(uint8_t *)(mtag + 1);
3515 
3516 	return (mpcp == prio);
3517 }
3518 
3519 static int
pf_icmp_to_bandlim(uint8_t type)3520 pf_icmp_to_bandlim(uint8_t type)
3521 {
3522 	switch (type) {
3523 		case ICMP_ECHO:
3524 		case ICMP_ECHOREPLY:
3525 			return (BANDLIM_ICMP_ECHO);
3526 		case ICMP_TSTAMP:
3527 		case ICMP_TSTAMPREPLY:
3528 			return (BANDLIM_ICMP_TSTAMP);
3529 		case ICMP_UNREACH:
3530 		default:
3531 			return (BANDLIM_ICMP_UNREACH);
3532 	}
3533 }
3534 
3535 static void
pf_send_icmp(struct mbuf * m,u_int8_t type,u_int8_t code,sa_family_t af,struct pf_krule * r,int rtableid)3536 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
3537     struct pf_krule *r, int rtableid)
3538 {
3539 	struct pf_send_entry *pfse;
3540 	struct mbuf *m0;
3541 	struct pf_mtag *pf_mtag;
3542 
3543 	/* ICMP packet rate limitation. */
3544 #ifdef INET6
3545 	if (af == AF_INET6) {
3546 		if (icmp6_ratelimit(NULL, type, code))
3547 			return;
3548 	}
3549 #endif
3550 #ifdef INET
3551 	if (af == AF_INET) {
3552 		if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0)
3553 			return;
3554 	}
3555 #endif
3556 
3557 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3558 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3559 	if (pfse == NULL)
3560 		return;
3561 
3562 	if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
3563 		free(pfse, M_PFTEMP);
3564 		return;
3565 	}
3566 
3567 	if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
3568 		free(pfse, M_PFTEMP);
3569 		return;
3570 	}
3571 	/* XXX: revisit */
3572 	m0->m_flags |= M_SKIP_FIREWALL;
3573 
3574 	if (rtableid >= 0)
3575 		M_SETFIB(m0, rtableid);
3576 
3577 #ifdef ALTQ
3578 	if (r->qid) {
3579 		pf_mtag->qid = r->qid;
3580 		/* add hints for ecn */
3581 		pf_mtag->hdr = mtod(m0, struct ip *);
3582 	}
3583 #endif /* ALTQ */
3584 
3585 	switch (af) {
3586 #ifdef INET
3587 	case AF_INET:
3588 		pfse->pfse_type = PFSE_ICMP;
3589 		break;
3590 #endif /* INET */
3591 #ifdef INET6
3592 	case AF_INET6:
3593 		pfse->pfse_type = PFSE_ICMP6;
3594 		break;
3595 #endif /* INET6 */
3596 	}
3597 	pfse->pfse_m = m0;
3598 	pfse->icmpopts.type = type;
3599 	pfse->icmpopts.code = code;
3600 	pf_send(pfse);
3601 }
3602 
3603 /*
3604  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
3605  * If n is 0, they match if they are equal. If n is != 0, they match if they
3606  * are different.
3607  */
3608 int
pf_match_addr(u_int8_t n,struct pf_addr * a,struct pf_addr * m,struct pf_addr * b,sa_family_t af)3609 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
3610     struct pf_addr *b, sa_family_t af)
3611 {
3612 	int	match = 0;
3613 
3614 	switch (af) {
3615 #ifdef INET
3616 	case AF_INET:
3617 		if ((a->addr32[0] & m->addr32[0]) ==
3618 		    (b->addr32[0] & m->addr32[0]))
3619 			match++;
3620 		break;
3621 #endif /* INET */
3622 #ifdef INET6
3623 	case AF_INET6:
3624 		if (((a->addr32[0] & m->addr32[0]) ==
3625 		     (b->addr32[0] & m->addr32[0])) &&
3626 		    ((a->addr32[1] & m->addr32[1]) ==
3627 		     (b->addr32[1] & m->addr32[1])) &&
3628 		    ((a->addr32[2] & m->addr32[2]) ==
3629 		     (b->addr32[2] & m->addr32[2])) &&
3630 		    ((a->addr32[3] & m->addr32[3]) ==
3631 		     (b->addr32[3] & m->addr32[3])))
3632 			match++;
3633 		break;
3634 #endif /* INET6 */
3635 	}
3636 	if (match) {
3637 		if (n)
3638 			return (0);
3639 		else
3640 			return (1);
3641 	} else {
3642 		if (n)
3643 			return (1);
3644 		else
3645 			return (0);
3646 	}
3647 }
3648 
3649 /*
3650  * Return 1 if b <= a <= e, otherwise return 0.
3651  */
3652 int
pf_match_addr_range(struct pf_addr * b,struct pf_addr * e,struct pf_addr * a,sa_family_t af)3653 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
3654     struct pf_addr *a, sa_family_t af)
3655 {
3656 	switch (af) {
3657 #ifdef INET
3658 	case AF_INET:
3659 		if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
3660 		    (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
3661 			return (0);
3662 		break;
3663 #endif /* INET */
3664 #ifdef INET6
3665 	case AF_INET6: {
3666 		int	i;
3667 
3668 		/* check a >= b */
3669 		for (i = 0; i < 4; ++i)
3670 			if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
3671 				break;
3672 			else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
3673 				return (0);
3674 		/* check a <= e */
3675 		for (i = 0; i < 4; ++i)
3676 			if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
3677 				break;
3678 			else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
3679 				return (0);
3680 		break;
3681 	}
3682 #endif /* INET6 */
3683 	}
3684 	return (1);
3685 }
3686 
3687 static int
pf_match(u_int8_t op,u_int32_t a1,u_int32_t a2,u_int32_t p)3688 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
3689 {
3690 	switch (op) {
3691 	case PF_OP_IRG:
3692 		return ((p > a1) && (p < a2));
3693 	case PF_OP_XRG:
3694 		return ((p < a1) || (p > a2));
3695 	case PF_OP_RRG:
3696 		return ((p >= a1) && (p <= a2));
3697 	case PF_OP_EQ:
3698 		return (p == a1);
3699 	case PF_OP_NE:
3700 		return (p != a1);
3701 	case PF_OP_LT:
3702 		return (p < a1);
3703 	case PF_OP_LE:
3704 		return (p <= a1);
3705 	case PF_OP_GT:
3706 		return (p > a1);
3707 	case PF_OP_GE:
3708 		return (p >= a1);
3709 	}
3710 	return (0); /* never reached */
3711 }
3712 
3713 int
pf_match_port(u_int8_t op,u_int16_t a1,u_int16_t a2,u_int16_t p)3714 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
3715 {
3716 	NTOHS(a1);
3717 	NTOHS(a2);
3718 	NTOHS(p);
3719 	return (pf_match(op, a1, a2, p));
3720 }
3721 
3722 static int
pf_match_uid(u_int8_t op,uid_t a1,uid_t a2,uid_t u)3723 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
3724 {
3725 	if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
3726 		return (0);
3727 	return (pf_match(op, a1, a2, u));
3728 }
3729 
3730 static int
pf_match_gid(u_int8_t op,gid_t a1,gid_t a2,gid_t g)3731 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
3732 {
3733 	if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
3734 		return (0);
3735 	return (pf_match(op, a1, a2, g));
3736 }
3737 
3738 int
pf_match_tag(struct mbuf * m,struct pf_krule * r,int * tag,int mtag)3739 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag)
3740 {
3741 	if (*tag == -1)
3742 		*tag = mtag;
3743 
3744 	return ((!r->match_tag_not && r->match_tag == *tag) ||
3745 	    (r->match_tag_not && r->match_tag != *tag));
3746 }
3747 
3748 int
pf_tag_packet(struct mbuf * m,struct pf_pdesc * pd,int tag)3749 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
3750 {
3751 
3752 	KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
3753 
3754 	if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
3755 		return (ENOMEM);
3756 
3757 	pd->pf_mtag->tag = tag;
3758 
3759 	return (0);
3760 }
3761 
3762 #define	PF_ANCHOR_STACKSIZE	32
3763 struct pf_kanchor_stackframe {
3764 	struct pf_kruleset	*rs;
3765 	struct pf_krule		*r;	/* XXX: + match bit */
3766 	struct pf_kanchor	*child;
3767 };
3768 
3769 /*
3770  * XXX: We rely on malloc(9) returning pointer aligned addresses.
3771  */
3772 #define	PF_ANCHORSTACK_MATCH	0x00000001
3773 #define	PF_ANCHORSTACK_MASK	(PF_ANCHORSTACK_MATCH)
3774 
3775 #define	PF_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
3776 #define	PF_ANCHOR_RULE(f)	(struct pf_krule *)			\
3777 				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
3778 #define	PF_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 			\
3779 				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
3780 } while (0)
3781 
3782 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)3783 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth,
3784     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
3785     int *match)
3786 {
3787 	struct pf_kanchor_stackframe	*f;
3788 
3789 	PF_RULES_RASSERT();
3790 
3791 	if (match)
3792 		*match = 0;
3793 	if (*depth >= PF_ANCHOR_STACKSIZE) {
3794 		printf("%s: anchor stack overflow on %s\n",
3795 		    __func__, (*r)->anchor->name);
3796 		*r = TAILQ_NEXT(*r, entries);
3797 		return;
3798 	} else if (*depth == 0 && a != NULL)
3799 		*a = *r;
3800 	f = stack + (*depth)++;
3801 	f->rs = *rs;
3802 	f->r = *r;
3803 	if ((*r)->anchor_wildcard) {
3804 		struct pf_kanchor_node *parent = &(*r)->anchor->children;
3805 
3806 		if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) {
3807 			*r = NULL;
3808 			return;
3809 		}
3810 		*rs = &f->child->ruleset;
3811 	} else {
3812 		f->child = NULL;
3813 		*rs = &(*r)->anchor->ruleset;
3814 	}
3815 	*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
3816 }
3817 
3818 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)3819 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth,
3820     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
3821     int *match)
3822 {
3823 	struct pf_kanchor_stackframe	*f;
3824 	struct pf_krule *fr;
3825 	int quick = 0;
3826 
3827 	PF_RULES_RASSERT();
3828 
3829 	do {
3830 		if (*depth <= 0)
3831 			break;
3832 		f = stack + *depth - 1;
3833 		fr = PF_ANCHOR_RULE(f);
3834 		if (f->child != NULL) {
3835 			/*
3836 			 * This block traverses through
3837 			 * a wildcard anchor.
3838 			 */
3839 			if (match != NULL && *match) {
3840 				/*
3841 				 * If any of "*" matched, then
3842 				 * "foo/ *" matched, mark frame
3843 				 * appropriately.
3844 				 */
3845 				PF_ANCHOR_SET_MATCH(f);
3846 				*match = 0;
3847 			}
3848 			f->child = RB_NEXT(pf_kanchor_node,
3849 			    &fr->anchor->children, f->child);
3850 			if (f->child != NULL) {
3851 				*rs = &f->child->ruleset;
3852 				*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
3853 				if (*r == NULL)
3854 					continue;
3855 				else
3856 					break;
3857 			}
3858 		}
3859 		(*depth)--;
3860 		if (*depth == 0 && a != NULL)
3861 			*a = NULL;
3862 		*rs = f->rs;
3863 		if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
3864 			quick = fr->quick;
3865 		*r = TAILQ_NEXT(fr, entries);
3866 	} while (*r == NULL);
3867 
3868 	return (quick);
3869 }
3870 
3871 struct pf_keth_anchor_stackframe {
3872 	struct pf_keth_ruleset	*rs;
3873 	struct pf_keth_rule	*r;	/* XXX: + match bit */
3874 	struct pf_keth_anchor	*child;
3875 };
3876 
3877 #define	PF_ETH_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
3878 #define	PF_ETH_ANCHOR_RULE(f)	(struct pf_keth_rule *)			\
3879 				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
3880 #define	PF_ETH_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 		\
3881 				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
3882 } while (0)
3883 
3884 void
pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe * stack,int * depth,struct pf_keth_ruleset ** rs,struct pf_keth_rule ** r,struct pf_keth_rule ** a,int * match)3885 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth,
3886     struct pf_keth_ruleset **rs, struct pf_keth_rule **r,
3887     struct pf_keth_rule **a, int *match)
3888 {
3889 	struct pf_keth_anchor_stackframe	*f;
3890 
3891 	NET_EPOCH_ASSERT();
3892 
3893 	if (match)
3894 		*match = 0;
3895 	if (*depth >= PF_ANCHOR_STACKSIZE) {
3896 		printf("%s: anchor stack overflow on %s\n",
3897 		    __func__, (*r)->anchor->name);
3898 		*r = TAILQ_NEXT(*r, entries);
3899 		return;
3900 	} else if (*depth == 0 && a != NULL)
3901 		*a = *r;
3902 	f = stack + (*depth)++;
3903 	f->rs = *rs;
3904 	f->r = *r;
3905 	if ((*r)->anchor_wildcard) {
3906 		struct pf_keth_anchor_node *parent = &(*r)->anchor->children;
3907 
3908 		if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) {
3909 			*r = NULL;
3910 			return;
3911 		}
3912 		*rs = &f->child->ruleset;
3913 	} else {
3914 		f->child = NULL;
3915 		*rs = &(*r)->anchor->ruleset;
3916 	}
3917 	*r = TAILQ_FIRST((*rs)->active.rules);
3918 }
3919 
3920 int
pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe * stack,int * depth,struct pf_keth_ruleset ** rs,struct pf_keth_rule ** r,struct pf_keth_rule ** a,int * match)3921 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth,
3922     struct pf_keth_ruleset **rs, struct pf_keth_rule **r,
3923     struct pf_keth_rule **a, int *match)
3924 {
3925 	struct pf_keth_anchor_stackframe	*f;
3926 	struct pf_keth_rule *fr;
3927 	int quick = 0;
3928 
3929 	NET_EPOCH_ASSERT();
3930 
3931 	do {
3932 		if (*depth <= 0)
3933 			break;
3934 		f = stack + *depth - 1;
3935 		fr = PF_ETH_ANCHOR_RULE(f);
3936 		if (f->child != NULL) {
3937 			/*
3938 			 * This block traverses through
3939 			 * a wildcard anchor.
3940 			 */
3941 			if (match != NULL && *match) {
3942 				/*
3943 				 * If any of "*" matched, then
3944 				 * "foo/ *" matched, mark frame
3945 				 * appropriately.
3946 				 */
3947 				PF_ETH_ANCHOR_SET_MATCH(f);
3948 				*match = 0;
3949 			}
3950 			f->child = RB_NEXT(pf_keth_anchor_node,
3951 			    &fr->anchor->children, f->child);
3952 			if (f->child != NULL) {
3953 				*rs = &f->child->ruleset;
3954 				*r = TAILQ_FIRST((*rs)->active.rules);
3955 				if (*r == NULL)
3956 					continue;
3957 				else
3958 					break;
3959 			}
3960 		}
3961 		(*depth)--;
3962 		if (*depth == 0 && a != NULL)
3963 			*a = NULL;
3964 		*rs = f->rs;
3965 		if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match))
3966 			quick = fr->quick;
3967 		*r = TAILQ_NEXT(fr, entries);
3968 	} while (*r == NULL);
3969 
3970 	return (quick);
3971 }
3972 
3973 #ifdef INET6
3974 void
pf_poolmask(struct pf_addr * naddr,struct pf_addr * raddr,struct pf_addr * rmask,struct pf_addr * saddr,sa_family_t af)3975 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
3976     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
3977 {
3978 	switch (af) {
3979 #ifdef INET
3980 	case AF_INET:
3981 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
3982 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
3983 		break;
3984 #endif /* INET */
3985 	case AF_INET6:
3986 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
3987 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
3988 		naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
3989 		((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
3990 		naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
3991 		((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
3992 		naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
3993 		((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
3994 		break;
3995 	}
3996 }
3997 
3998 void
pf_addr_inc(struct pf_addr * addr,sa_family_t af)3999 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
4000 {
4001 	switch (af) {
4002 #ifdef INET
4003 	case AF_INET:
4004 		addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
4005 		break;
4006 #endif /* INET */
4007 	case AF_INET6:
4008 		if (addr->addr32[3] == 0xffffffff) {
4009 			addr->addr32[3] = 0;
4010 			if (addr->addr32[2] == 0xffffffff) {
4011 				addr->addr32[2] = 0;
4012 				if (addr->addr32[1] == 0xffffffff) {
4013 					addr->addr32[1] = 0;
4014 					addr->addr32[0] =
4015 					    htonl(ntohl(addr->addr32[0]) + 1);
4016 				} else
4017 					addr->addr32[1] =
4018 					    htonl(ntohl(addr->addr32[1]) + 1);
4019 			} else
4020 				addr->addr32[2] =
4021 				    htonl(ntohl(addr->addr32[2]) + 1);
4022 		} else
4023 			addr->addr32[3] =
4024 			    htonl(ntohl(addr->addr32[3]) + 1);
4025 		break;
4026 	}
4027 }
4028 #endif /* INET6 */
4029 
4030 void
pf_rule_to_actions(struct pf_krule * r,struct pf_rule_actions * a)4031 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a)
4032 {
4033 	/*
4034 	 * Modern rules use the same flags in rules as they do in states.
4035 	 */
4036 	a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID|
4037 	    PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO));
4038 
4039 	/*
4040 	 * Old-style scrub rules have different flags which need to be translated.
4041 	 */
4042 	if (r->rule_flag & PFRULE_RANDOMID)
4043 		a->flags |= PFSTATE_RANDOMID;
4044 	if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) {
4045 		a->flags |= PFSTATE_SETTOS;
4046 		a->set_tos = r->set_tos;
4047 	}
4048 
4049 	if (r->qid)
4050 		a->qid = r->qid;
4051 	if (r->pqid)
4052 		a->pqid = r->pqid;
4053 	if (r->rtableid >= 0)
4054 		a->rtableid = r->rtableid;
4055 	a->log |= r->log;
4056 	if (r->min_ttl)
4057 		a->min_ttl = r->min_ttl;
4058 	if (r->max_mss)
4059 		a->max_mss = r->max_mss;
4060 	if (r->dnpipe)
4061 		a->dnpipe = r->dnpipe;
4062 	if (r->dnrpipe)
4063 		a->dnrpipe = r->dnrpipe;
4064 	if (r->dnpipe || r->dnrpipe) {
4065 		if (r->free_flags & PFRULE_DN_IS_PIPE)
4066 			a->flags |= PFSTATE_DN_IS_PIPE;
4067 		else
4068 			a->flags &= ~PFSTATE_DN_IS_PIPE;
4069 	}
4070 	if (r->scrub_flags & PFSTATE_SETPRIO) {
4071 		a->set_prio[0] = r->set_prio[0];
4072 		a->set_prio[1] = r->set_prio[1];
4073 	}
4074 }
4075 
4076 int
pf_socket_lookup(struct pf_pdesc * pd,struct mbuf * m)4077 pf_socket_lookup(struct pf_pdesc *pd, struct mbuf *m)
4078 {
4079 	struct pf_addr		*saddr, *daddr;
4080 	u_int16_t		 sport, dport;
4081 	struct inpcbinfo	*pi;
4082 	struct inpcb		*inp;
4083 
4084 	pd->lookup.uid = UID_MAX;
4085 	pd->lookup.gid = GID_MAX;
4086 
4087 	switch (pd->proto) {
4088 	case IPPROTO_TCP:
4089 		sport = pd->hdr.tcp.th_sport;
4090 		dport = pd->hdr.tcp.th_dport;
4091 		pi = &V_tcbinfo;
4092 		break;
4093 	case IPPROTO_UDP:
4094 		sport = pd->hdr.udp.uh_sport;
4095 		dport = pd->hdr.udp.uh_dport;
4096 		pi = &V_udbinfo;
4097 		break;
4098 	default:
4099 		return (-1);
4100 	}
4101 	if (pd->dir == PF_IN) {
4102 		saddr = pd->src;
4103 		daddr = pd->dst;
4104 	} else {
4105 		u_int16_t	p;
4106 
4107 		p = sport;
4108 		sport = dport;
4109 		dport = p;
4110 		saddr = pd->dst;
4111 		daddr = pd->src;
4112 	}
4113 	switch (pd->af) {
4114 #ifdef INET
4115 	case AF_INET:
4116 		inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
4117 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
4118 		if (inp == NULL) {
4119 			inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
4120 			   daddr->v4, dport, INPLOOKUP_WILDCARD |
4121 			   INPLOOKUP_RLOCKPCB, NULL, m);
4122 			if (inp == NULL)
4123 				return (-1);
4124 		}
4125 		break;
4126 #endif /* INET */
4127 #ifdef INET6
4128 	case AF_INET6:
4129 		inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
4130 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
4131 		if (inp == NULL) {
4132 			inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
4133 			    &daddr->v6, dport, INPLOOKUP_WILDCARD |
4134 			    INPLOOKUP_RLOCKPCB, NULL, m);
4135 			if (inp == NULL)
4136 				return (-1);
4137 		}
4138 		break;
4139 #endif /* INET6 */
4140 
4141 	default:
4142 		return (-1);
4143 	}
4144 	INP_RLOCK_ASSERT(inp);
4145 	pd->lookup.uid = inp->inp_cred->cr_uid;
4146 	pd->lookup.gid = inp->inp_cred->cr_groups[0];
4147 	INP_RUNLOCK(inp);
4148 
4149 	return (1);
4150 }
4151 
4152 u_int8_t
pf_get_wscale(struct mbuf * m,int off,u_int16_t th_off,sa_family_t af)4153 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
4154 {
4155 	int		 hlen;
4156 	u_int8_t	 hdr[60];
4157 	u_int8_t	*opt, optlen;
4158 	u_int8_t	 wscale = 0;
4159 
4160 	hlen = th_off << 2;		/* hlen <= sizeof(hdr) */
4161 	if (hlen <= sizeof(struct tcphdr))
4162 		return (0);
4163 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
4164 		return (0);
4165 	opt = hdr + sizeof(struct tcphdr);
4166 	hlen -= sizeof(struct tcphdr);
4167 	while (hlen >= 3) {
4168 		switch (*opt) {
4169 		case TCPOPT_EOL:
4170 		case TCPOPT_NOP:
4171 			++opt;
4172 			--hlen;
4173 			break;
4174 		case TCPOPT_WINDOW:
4175 			wscale = opt[2];
4176 			if (wscale > TCP_MAX_WINSHIFT)
4177 				wscale = TCP_MAX_WINSHIFT;
4178 			wscale |= PF_WSCALE_FLAG;
4179 			/* FALLTHROUGH */
4180 		default:
4181 			optlen = opt[1];
4182 			if (optlen < 2)
4183 				optlen = 2;
4184 			hlen -= optlen;
4185 			opt += optlen;
4186 			break;
4187 		}
4188 	}
4189 	return (wscale);
4190 }
4191 
4192 u_int16_t
pf_get_mss(struct mbuf * m,int off,u_int16_t th_off,sa_family_t af)4193 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
4194 {
4195 	int		 hlen;
4196 	u_int8_t	 hdr[60];
4197 	u_int8_t	*opt, optlen;
4198 	u_int16_t	 mss = V_tcp_mssdflt;
4199 
4200 	hlen = th_off << 2;	/* hlen <= sizeof(hdr) */
4201 	if (hlen <= sizeof(struct tcphdr))
4202 		return (0);
4203 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
4204 		return (0);
4205 	opt = hdr + sizeof(struct tcphdr);
4206 	hlen -= sizeof(struct tcphdr);
4207 	while (hlen >= TCPOLEN_MAXSEG) {
4208 		switch (*opt) {
4209 		case TCPOPT_EOL:
4210 		case TCPOPT_NOP:
4211 			++opt;
4212 			--hlen;
4213 			break;
4214 		case TCPOPT_MAXSEG:
4215 			bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
4216 			NTOHS(mss);
4217 			/* FALLTHROUGH */
4218 		default:
4219 			optlen = opt[1];
4220 			if (optlen < 2)
4221 				optlen = 2;
4222 			hlen -= optlen;
4223 			opt += optlen;
4224 			break;
4225 		}
4226 	}
4227 	return (mss);
4228 }
4229 
4230 static u_int16_t
pf_calc_mss(struct pf_addr * addr,sa_family_t af,int rtableid,u_int16_t offer)4231 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
4232 {
4233 	struct nhop_object *nh;
4234 #ifdef INET6
4235 	struct in6_addr		dst6;
4236 	uint32_t		scopeid;
4237 #endif /* INET6 */
4238 	int			 hlen = 0;
4239 	uint16_t		 mss = 0;
4240 
4241 	NET_EPOCH_ASSERT();
4242 
4243 	switch (af) {
4244 #ifdef INET
4245 	case AF_INET:
4246 		hlen = sizeof(struct ip);
4247 		nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0);
4248 		if (nh != NULL)
4249 			mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
4250 		break;
4251 #endif /* INET */
4252 #ifdef INET6
4253 	case AF_INET6:
4254 		hlen = sizeof(struct ip6_hdr);
4255 		in6_splitscope(&addr->v6, &dst6, &scopeid);
4256 		nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0);
4257 		if (nh != NULL)
4258 			mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
4259 		break;
4260 #endif /* INET6 */
4261 	}
4262 
4263 	mss = max(V_tcp_mssdflt, mss);
4264 	mss = min(mss, offer);
4265 	mss = max(mss, 64);		/* sanity - at least max opt space */
4266 	return (mss);
4267 }
4268 
4269 static u_int32_t
pf_tcp_iss(struct pf_pdesc * pd)4270 pf_tcp_iss(struct pf_pdesc *pd)
4271 {
4272 	MD5_CTX ctx;
4273 	u_int32_t digest[4];
4274 
4275 	if (V_pf_tcp_secret_init == 0) {
4276 		arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
4277 		MD5Init(&V_pf_tcp_secret_ctx);
4278 		MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
4279 		    sizeof(V_pf_tcp_secret));
4280 		V_pf_tcp_secret_init = 1;
4281 	}
4282 
4283 	ctx = V_pf_tcp_secret_ctx;
4284 
4285 	MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short));
4286 	MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short));
4287 	if (pd->af == AF_INET6) {
4288 		MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
4289 		MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
4290 	} else {
4291 		MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
4292 		MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
4293 	}
4294 	MD5Final((u_char *)digest, &ctx);
4295 	V_pf_tcp_iss_off += 4096;
4296 #define	ISN_RANDOM_INCREMENT (4096 - 1)
4297 	return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
4298 	    V_pf_tcp_iss_off);
4299 #undef	ISN_RANDOM_INCREMENT
4300 }
4301 
4302 static bool
pf_match_eth_addr(const uint8_t * a,const struct pf_keth_rule_addr * r)4303 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r)
4304 {
4305 	bool match = true;
4306 
4307 	/* Always matches if not set */
4308 	if (! r->isset)
4309 		return (!r->neg);
4310 
4311 	for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4312 		if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) {
4313 			match = false;
4314 			break;
4315 		}
4316 	}
4317 
4318 	return (match ^ r->neg);
4319 }
4320 
4321 static int
pf_match_eth_tag(struct mbuf * m,struct pf_keth_rule * r,int * tag,int mtag)4322 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag)
4323 {
4324 	if (*tag == -1)
4325 		*tag = mtag;
4326 
4327 	return ((!r->match_tag_not && r->match_tag == *tag) ||
4328 	    (r->match_tag_not && r->match_tag != *tag));
4329 }
4330 
4331 static void
pf_bridge_to(struct ifnet * ifp,struct mbuf * m)4332 pf_bridge_to(struct ifnet *ifp, struct mbuf *m)
4333 {
4334 	/* If we don't have the interface drop the packet. */
4335 	if (ifp == NULL) {
4336 		m_freem(m);
4337 		return;
4338 	}
4339 
4340 	switch (ifp->if_type) {
4341 	case IFT_ETHER:
4342 	case IFT_XETHER:
4343 	case IFT_L2VLAN:
4344 	case IFT_BRIDGE:
4345 	case IFT_IEEE8023ADLAG:
4346 		break;
4347 	default:
4348 		m_freem(m);
4349 		return;
4350 	}
4351 
4352 	ifp->if_transmit(ifp, m);
4353 }
4354 
4355 static int
pf_test_eth_rule(int dir,struct pfi_kkif * kif,struct mbuf ** m0)4356 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0)
4357 {
4358 #ifdef INET
4359 	struct ip ip;
4360 #endif
4361 #ifdef INET6
4362 	struct ip6_hdr ip6;
4363 #endif
4364 	struct mbuf *m = *m0;
4365 	struct ether_header *e;
4366 	struct pf_keth_rule *r, *rm, *a = NULL;
4367 	struct pf_keth_ruleset *ruleset = NULL;
4368 	struct pf_mtag *mtag;
4369 	struct pf_keth_ruleq *rules;
4370 	struct pf_addr *src = NULL, *dst = NULL;
4371 	struct pfi_kkif *bridge_to;
4372 	sa_family_t af = 0;
4373 	uint16_t proto;
4374 	int asd = 0, match = 0;
4375 	int tag = -1;
4376 	uint8_t action;
4377 	struct pf_keth_anchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
4378 
4379 	MPASS(kif->pfik_ifp->if_vnet == curvnet);
4380 	NET_EPOCH_ASSERT();
4381 
4382 	PF_RULES_RLOCK_TRACKER;
4383 
4384 	SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m);
4385 
4386 	mtag = pf_find_mtag(m);
4387 	if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
4388 		/* Dummynet re-injects packets after they've
4389 		 * completed their delay. We've already
4390 		 * processed them, so pass unconditionally. */
4391 
4392 		/* But only once. We may see the packet multiple times (e.g.
4393 		 * PFIL_IN/PFIL_OUT). */
4394 		pf_dummynet_flag_remove(m, mtag);
4395 
4396 		return (PF_PASS);
4397 	}
4398 
4399 	ruleset = V_pf_keth;
4400 	rules = ck_pr_load_ptr(&ruleset->active.rules);
4401 	r = TAILQ_FIRST(rules);
4402 	rm = NULL;
4403 
4404 	e = mtod(m, struct ether_header *);
4405 	proto = ntohs(e->ether_type);
4406 
4407 	switch (proto) {
4408 #ifdef INET
4409 	case ETHERTYPE_IP: {
4410 		if (m_length(m, NULL) < (sizeof(struct ether_header) +
4411 		    sizeof(ip)))
4412 			return (PF_DROP);
4413 
4414 		af = AF_INET;
4415 		m_copydata(m, sizeof(struct ether_header), sizeof(ip),
4416 		    (caddr_t)&ip);
4417 		src = (struct pf_addr *)&ip.ip_src;
4418 		dst = (struct pf_addr *)&ip.ip_dst;
4419 		break;
4420 	}
4421 #endif /* INET */
4422 #ifdef INET6
4423 	case ETHERTYPE_IPV6: {
4424 		if (m_length(m, NULL) < (sizeof(struct ether_header) +
4425 		    sizeof(ip6)))
4426 			return (PF_DROP);
4427 
4428 		af = AF_INET6;
4429 		m_copydata(m, sizeof(struct ether_header), sizeof(ip6),
4430 		    (caddr_t)&ip6);
4431 		src = (struct pf_addr *)&ip6.ip6_src;
4432 		dst = (struct pf_addr *)&ip6.ip6_dst;
4433 		break;
4434 	}
4435 #endif /* INET6 */
4436 	}
4437 
4438 	PF_RULES_RLOCK();
4439 
4440 	while (r != NULL) {
4441 		counter_u64_add(r->evaluations, 1);
4442 		SDT_PROBE2(pf, eth, test_rule, test, r->nr, r);
4443 
4444 		if (pfi_kkif_match(r->kif, kif) == r->ifnot) {
4445 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4446 			    "kif");
4447 			r = r->skip[PFE_SKIP_IFP].ptr;
4448 		}
4449 		else if (r->direction && r->direction != dir) {
4450 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4451 			    "dir");
4452 			r = r->skip[PFE_SKIP_DIR].ptr;
4453 		}
4454 		else if (r->proto && r->proto != proto) {
4455 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4456 			    "proto");
4457 			r = r->skip[PFE_SKIP_PROTO].ptr;
4458 		}
4459 		else if (! pf_match_eth_addr(e->ether_shost, &r->src)) {
4460 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4461 			    "src");
4462 			r = r->skip[PFE_SKIP_SRC_ADDR].ptr;
4463 		}
4464 		else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) {
4465 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4466 			    "dst");
4467 			r = r->skip[PFE_SKIP_DST_ADDR].ptr;
4468 		}
4469 		else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af,
4470 		    r->ipsrc.neg, kif, M_GETFIB(m))) {
4471 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4472 			    "ip_src");
4473 			r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr;
4474 		}
4475 		else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af,
4476 		    r->ipdst.neg, kif, M_GETFIB(m))) {
4477 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4478 			    "ip_dst");
4479 			r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr;
4480 		}
4481 		else if (r->match_tag && !pf_match_eth_tag(m, r, &tag,
4482 		    mtag ? mtag->tag : 0)) {
4483 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4484 			    "match_tag");
4485 			r = TAILQ_NEXT(r, entries);
4486 		}
4487 		else {
4488 			if (r->tag)
4489 				tag = r->tag;
4490 			if (r->anchor == NULL) {
4491 				/* Rule matches */
4492 				rm = r;
4493 
4494 				SDT_PROBE2(pf, eth, test_rule, match, r->nr, r);
4495 
4496 				if (r->quick)
4497 					break;
4498 
4499 				r = TAILQ_NEXT(r, entries);
4500 			} else {
4501 				pf_step_into_keth_anchor(anchor_stack, &asd,
4502 				    &ruleset, &r, &a, &match);
4503 			}
4504 		}
4505 		if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd,
4506 		    &ruleset, &r, &a, &match))
4507 			break;
4508 	}
4509 
4510 	r = rm;
4511 
4512 	SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r);
4513 
4514 	/* Default to pass. */
4515 	if (r == NULL) {
4516 		PF_RULES_RUNLOCK();
4517 		return (PF_PASS);
4518 	}
4519 
4520 	/* Execute action. */
4521 	counter_u64_add(r->packets[dir == PF_OUT], 1);
4522 	counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL));
4523 	pf_update_timestamp(r);
4524 
4525 	/* Shortcut. Don't tag if we're just going to drop anyway. */
4526 	if (r->action == PF_DROP) {
4527 		PF_RULES_RUNLOCK();
4528 		return (PF_DROP);
4529 	}
4530 
4531 	if (tag > 0) {
4532 		if (mtag == NULL)
4533 			mtag = pf_get_mtag(m);
4534 		if (mtag == NULL) {
4535 			PF_RULES_RUNLOCK();
4536 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4537 			return (PF_DROP);
4538 		}
4539 		mtag->tag = tag;
4540 	}
4541 
4542 	if (r->qid != 0) {
4543 		if (mtag == NULL)
4544 			mtag = pf_get_mtag(m);
4545 		if (mtag == NULL) {
4546 			PF_RULES_RUNLOCK();
4547 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4548 			return (PF_DROP);
4549 		}
4550 		mtag->qid = r->qid;
4551 	}
4552 
4553 	action = r->action;
4554 	bridge_to = r->bridge_to;
4555 
4556 	/* Dummynet */
4557 	if (r->dnpipe) {
4558 		struct ip_fw_args dnflow;
4559 
4560 		/* Drop packet if dummynet is not loaded. */
4561 		if (ip_dn_io_ptr == NULL) {
4562 			PF_RULES_RUNLOCK();
4563 			m_freem(m);
4564 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4565 			return (PF_DROP);
4566 		}
4567 		if (mtag == NULL)
4568 			mtag = pf_get_mtag(m);
4569 		if (mtag == NULL) {
4570 			PF_RULES_RUNLOCK();
4571 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4572 			return (PF_DROP);
4573 		}
4574 
4575 		bzero(&dnflow, sizeof(dnflow));
4576 
4577 		/* We don't have port numbers here, so we set 0.  That means
4578 		 * that we'll be somewhat limited in distinguishing flows (i.e.
4579 		 * only based on IP addresses, not based on port numbers), but
4580 		 * it's better than nothing. */
4581 		dnflow.f_id.dst_port = 0;
4582 		dnflow.f_id.src_port = 0;
4583 		dnflow.f_id.proto = 0;
4584 
4585 		dnflow.rule.info = r->dnpipe;
4586 		dnflow.rule.info |= IPFW_IS_DUMMYNET;
4587 		if (r->dnflags & PFRULE_DN_IS_PIPE)
4588 			dnflow.rule.info |= IPFW_IS_PIPE;
4589 
4590 		dnflow.f_id.extra = dnflow.rule.info;
4591 
4592 		dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT;
4593 		dnflow.flags |= IPFW_ARGS_ETHER;
4594 		dnflow.ifp = kif->pfik_ifp;
4595 
4596 		switch (af) {
4597 		case AF_INET:
4598 			dnflow.f_id.addr_type = 4;
4599 			dnflow.f_id.src_ip = src->v4.s_addr;
4600 			dnflow.f_id.dst_ip = dst->v4.s_addr;
4601 			break;
4602 		case AF_INET6:
4603 			dnflow.flags |= IPFW_ARGS_IP6;
4604 			dnflow.f_id.addr_type = 6;
4605 			dnflow.f_id.src_ip6 = src->v6;
4606 			dnflow.f_id.dst_ip6 = dst->v6;
4607 			break;
4608 		}
4609 
4610 		PF_RULES_RUNLOCK();
4611 
4612 		mtag->flags |= PF_MTAG_FLAG_DUMMYNET;
4613 		ip_dn_io_ptr(m0, &dnflow);
4614 		if (*m0 != NULL)
4615 			pf_dummynet_flag_remove(m, mtag);
4616 	} else {
4617 		PF_RULES_RUNLOCK();
4618 	}
4619 
4620 	if (action == PF_PASS && bridge_to) {
4621 		pf_bridge_to(bridge_to->pfik_ifp, *m0);
4622 		*m0 = NULL; /* We've eaten the packet. */
4623 	}
4624 
4625 	return (action);
4626 }
4627 
4628 static int
pf_test_rule(struct pf_krule ** rm,struct pf_kstate ** sm,struct pfi_kkif * kif,struct mbuf * m,int off,struct pf_pdesc * pd,struct pf_krule ** am,struct pf_kruleset ** rsm,struct inpcb * inp)4629 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, struct pfi_kkif *kif,
4630     struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_krule **am,
4631     struct pf_kruleset **rsm, struct inpcb *inp)
4632 {
4633 	struct pf_krule		*nr = NULL;
4634 	struct pf_addr		* const saddr = pd->src;
4635 	struct pf_addr		* const daddr = pd->dst;
4636 	sa_family_t		 af = pd->af;
4637 	struct pf_krule		*r, *a = NULL;
4638 	struct pf_kruleset	*ruleset = NULL;
4639 	struct pf_krule_slist	 match_rules;
4640 	struct pf_krule_item	*ri;
4641 	struct pf_ksrc_node	*nsn = NULL;
4642 	struct tcphdr		*th = &pd->hdr.tcp;
4643 	struct pf_state_key	*sk = NULL, *nk = NULL;
4644 	u_short			 reason;
4645 	int			 rewrite = 0, hdrlen = 0;
4646 	int			 tag = -1;
4647 	int			 asd = 0;
4648 	int			 match = 0;
4649 	int			 state_icmp = 0, icmp_dir, multi;
4650 	u_int16_t		 sport = 0, dport = 0, virtual_type, virtual_id;
4651 	u_int16_t		 bproto_sum = 0, bip_sum = 0;
4652 	u_int8_t		 icmptype = 0, icmpcode = 0;
4653 	struct pf_kanchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
4654 
4655 	PF_RULES_RASSERT();
4656 
4657 	if (inp != NULL) {
4658 		INP_LOCK_ASSERT(inp);
4659 		pd->lookup.uid = inp->inp_cred->cr_uid;
4660 		pd->lookup.gid = inp->inp_cred->cr_groups[0];
4661 		pd->lookup.done = 1;
4662 	}
4663 
4664 	switch (pd->proto) {
4665 	case IPPROTO_TCP:
4666 		sport = th->th_sport;
4667 		dport = th->th_dport;
4668 		hdrlen = sizeof(*th);
4669 		break;
4670 	case IPPROTO_UDP:
4671 		sport = pd->hdr.udp.uh_sport;
4672 		dport = pd->hdr.udp.uh_dport;
4673 		hdrlen = sizeof(pd->hdr.udp);
4674 		break;
4675 	case IPPROTO_SCTP:
4676 		sport = pd->hdr.sctp.src_port;
4677 		dport = pd->hdr.sctp.dest_port;
4678 		hdrlen = sizeof(pd->hdr.sctp);
4679 		break;
4680 #ifdef INET
4681 	case IPPROTO_ICMP:
4682 		if (pd->af != AF_INET)
4683 			break;
4684 		hdrlen = sizeof(pd->hdr.icmp);
4685 		icmptype = pd->hdr.icmp.icmp_type;
4686 		icmpcode = pd->hdr.icmp.icmp_code;
4687 		state_icmp = pf_icmp_mapping(pd, icmptype,
4688 		    &icmp_dir, &multi, &virtual_id, &virtual_type);
4689 		if (icmp_dir == PF_IN) {
4690 			sport = virtual_id;
4691 			dport = virtual_type;
4692 		} else {
4693 			sport = virtual_type;
4694 			dport = virtual_id;
4695 		}
4696 		break;
4697 #endif /* INET */
4698 #ifdef INET6
4699 	case IPPROTO_ICMPV6:
4700 		if (af != AF_INET6)
4701 			break;
4702 		hdrlen = sizeof(pd->hdr.icmp6);
4703 		icmptype = pd->hdr.icmp6.icmp6_type;
4704 		icmpcode = pd->hdr.icmp6.icmp6_code;
4705 		state_icmp = pf_icmp_mapping(pd, icmptype,
4706 		    &icmp_dir, &multi, &virtual_id, &virtual_type);
4707 		if (icmp_dir == PF_IN) {
4708 			sport = virtual_id;
4709 			dport = virtual_type;
4710 		} else {
4711 			sport = virtual_type;
4712 			dport = virtual_id;
4713 		}
4714 
4715 		break;
4716 #endif /* INET6 */
4717 	default:
4718 		sport = dport = hdrlen = 0;
4719 		break;
4720 	}
4721 
4722 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
4723 
4724 	/* check packet for BINAT/NAT/RDR */
4725 	if ((nr = pf_get_translation(pd, m, off, kif, &nsn, &sk,
4726 	    &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
4727 		KASSERT(sk != NULL, ("%s: null sk", __func__));
4728 		KASSERT(nk != NULL, ("%s: null nk", __func__));
4729 
4730 		if (nr->log) {
4731 			PFLOG_PACKET(kif, m, af, PFRES_MATCH, nr, a,
4732 			    ruleset, pd, 1);
4733 		}
4734 
4735 		if (pd->ip_sum)
4736 			bip_sum = *pd->ip_sum;
4737 
4738 		switch (pd->proto) {
4739 		case IPPROTO_TCP:
4740 			bproto_sum = th->th_sum;
4741 			pd->proto_sum = &th->th_sum;
4742 
4743 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
4744 			    nk->port[pd->sidx] != sport) {
4745 				pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
4746 				    &th->th_sum, &nk->addr[pd->sidx],
4747 				    nk->port[pd->sidx], 0, af);
4748 				pd->sport = &th->th_sport;
4749 				sport = th->th_sport;
4750 			}
4751 
4752 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
4753 			    nk->port[pd->didx] != dport) {
4754 				pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
4755 				    &th->th_sum, &nk->addr[pd->didx],
4756 				    nk->port[pd->didx], 0, af);
4757 				dport = th->th_dport;
4758 				pd->dport = &th->th_dport;
4759 			}
4760 			rewrite++;
4761 			break;
4762 		case IPPROTO_UDP:
4763 			bproto_sum = pd->hdr.udp.uh_sum;
4764 			pd->proto_sum = &pd->hdr.udp.uh_sum;
4765 
4766 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
4767 			    nk->port[pd->sidx] != sport) {
4768 				pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport,
4769 				    pd->ip_sum, &pd->hdr.udp.uh_sum,
4770 				    &nk->addr[pd->sidx],
4771 				    nk->port[pd->sidx], 1, af);
4772 				sport = pd->hdr.udp.uh_sport;
4773 				pd->sport = &pd->hdr.udp.uh_sport;
4774 			}
4775 
4776 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
4777 			    nk->port[pd->didx] != dport) {
4778 				pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport,
4779 				    pd->ip_sum, &pd->hdr.udp.uh_sum,
4780 				    &nk->addr[pd->didx],
4781 				    nk->port[pd->didx], 1, af);
4782 				dport = pd->hdr.udp.uh_dport;
4783 				pd->dport = &pd->hdr.udp.uh_dport;
4784 			}
4785 			rewrite++;
4786 			break;
4787 		case IPPROTO_SCTP: {
4788 			uint16_t checksum = 0;
4789 
4790 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
4791 			    nk->port[pd->sidx] != sport) {
4792 				pf_change_ap(m, saddr, &pd->hdr.sctp.src_port,
4793 				    pd->ip_sum, &checksum,
4794 				    &nk->addr[pd->sidx],
4795 				    nk->port[pd->sidx], 1, af);
4796 			}
4797 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
4798 			    nk->port[pd->didx] != dport) {
4799 				pf_change_ap(m, daddr, &pd->hdr.sctp.dest_port,
4800 				    pd->ip_sum, &checksum,
4801 				    &nk->addr[pd->didx],
4802 				    nk->port[pd->didx], 1, af);
4803 			}
4804 			break;
4805 		}
4806 #ifdef INET
4807 		case IPPROTO_ICMP:
4808 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
4809 				pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
4810 				    nk->addr[pd->sidx].v4.s_addr, 0);
4811 
4812 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
4813 				pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
4814 				    nk->addr[pd->didx].v4.s_addr, 0);
4815 
4816 			if (virtual_type == htons(ICMP_ECHO) &&
4817 			     nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) {
4818 				pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
4819 				    pd->hdr.icmp.icmp_cksum, sport,
4820 				    nk->port[pd->sidx], 0);
4821 				pd->hdr.icmp.icmp_id = nk->port[pd->sidx];
4822 				pd->sport = &pd->hdr.icmp.icmp_id;
4823 			}
4824 			m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
4825 			break;
4826 #endif /* INET */
4827 #ifdef INET6
4828 		case IPPROTO_ICMPV6:
4829 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
4830 				pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum,
4831 				    &nk->addr[pd->sidx], 0);
4832 
4833 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
4834 				pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum,
4835 				    &nk->addr[pd->didx], 0);
4836 			rewrite++;
4837 			break;
4838 #endif /* INET */
4839 		default:
4840 			switch (af) {
4841 #ifdef INET
4842 			case AF_INET:
4843 				if (PF_ANEQ(saddr,
4844 				    &nk->addr[pd->sidx], AF_INET))
4845 					pf_change_a(&saddr->v4.s_addr,
4846 					    pd->ip_sum,
4847 					    nk->addr[pd->sidx].v4.s_addr, 0);
4848 
4849 				if (PF_ANEQ(daddr,
4850 				    &nk->addr[pd->didx], AF_INET))
4851 					pf_change_a(&daddr->v4.s_addr,
4852 					    pd->ip_sum,
4853 					    nk->addr[pd->didx].v4.s_addr, 0);
4854 				break;
4855 #endif /* INET */
4856 #ifdef INET6
4857 			case AF_INET6:
4858 				if (PF_ANEQ(saddr,
4859 				    &nk->addr[pd->sidx], AF_INET6))
4860 					PF_ACPY(saddr, &nk->addr[pd->sidx], af);
4861 
4862 				if (PF_ANEQ(daddr,
4863 				    &nk->addr[pd->didx], AF_INET6))
4864 					PF_ACPY(daddr, &nk->addr[pd->didx], af);
4865 				break;
4866 #endif /* INET */
4867 			}
4868 			break;
4869 		}
4870 		if (nr->natpass)
4871 			r = NULL;
4872 		pd->nat_rule = nr;
4873 	}
4874 
4875 	SLIST_INIT(&match_rules);
4876 	while (r != NULL) {
4877 		pf_counter_u64_add(&r->evaluations, 1);
4878 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
4879 			r = r->skip[PF_SKIP_IFP].ptr;
4880 		else if (r->direction && r->direction != pd->dir)
4881 			r = r->skip[PF_SKIP_DIR].ptr;
4882 		else if (r->af && r->af != af)
4883 			r = r->skip[PF_SKIP_AF].ptr;
4884 		else if (r->proto && r->proto != pd->proto)
4885 			r = r->skip[PF_SKIP_PROTO].ptr;
4886 		else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
4887 		    r->src.neg, kif, M_GETFIB(m)))
4888 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
4889 		/* tcp/udp only. port_op always 0 in other cases */
4890 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
4891 		    r->src.port[0], r->src.port[1], sport))
4892 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
4893 		else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
4894 		    r->dst.neg, NULL, M_GETFIB(m)))
4895 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
4896 		/* tcp/udp only. port_op always 0 in other cases */
4897 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
4898 		    r->dst.port[0], r->dst.port[1], dport))
4899 			r = r->skip[PF_SKIP_DST_PORT].ptr;
4900 		/* icmp only. type always 0 in other cases */
4901 		else if (r->type && r->type != icmptype + 1)
4902 			r = TAILQ_NEXT(r, entries);
4903 		/* icmp only. type always 0 in other cases */
4904 		else if (r->code && r->code != icmpcode + 1)
4905 			r = TAILQ_NEXT(r, entries);
4906 		else if (r->tos && !(r->tos == pd->tos))
4907 			r = TAILQ_NEXT(r, entries);
4908 		else if (r->rule_flag & PFRULE_FRAGMENT)
4909 			r = TAILQ_NEXT(r, entries);
4910 		else if (pd->proto == IPPROTO_TCP &&
4911 		    (r->flagset & th->th_flags) != r->flags)
4912 			r = TAILQ_NEXT(r, entries);
4913 		/* tcp/udp only. uid.op always 0 in other cases */
4914 		else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
4915 		    pf_socket_lookup(pd, m), 1)) &&
4916 		    !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
4917 		    pd->lookup.uid))
4918 			r = TAILQ_NEXT(r, entries);
4919 		/* tcp/udp only. gid.op always 0 in other cases */
4920 		else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
4921 		    pf_socket_lookup(pd, m), 1)) &&
4922 		    !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
4923 		    pd->lookup.gid))
4924 			r = TAILQ_NEXT(r, entries);
4925 		else if (r->prio &&
4926 		    !pf_match_ieee8021q_pcp(r->prio, m))
4927 			r = TAILQ_NEXT(r, entries);
4928 		else if (r->prob &&
4929 		    r->prob <= arc4random())
4930 			r = TAILQ_NEXT(r, entries);
4931 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
4932 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
4933 			r = TAILQ_NEXT(r, entries);
4934 		else if (r->os_fingerprint != PF_OSFP_ANY &&
4935 		    (pd->proto != IPPROTO_TCP || !pf_osfp_match(
4936 		    pf_osfp_fingerprint(pd, m, off, th),
4937 		    r->os_fingerprint)))
4938 			r = TAILQ_NEXT(r, entries);
4939 		else {
4940 			if (r->tag)
4941 				tag = r->tag;
4942 			if (r->anchor == NULL) {
4943 				if (r->action == PF_MATCH) {
4944 					ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO);
4945 					if (ri == NULL) {
4946 						REASON_SET(&reason, PFRES_MEMORY);
4947 						goto cleanup;
4948 					}
4949 					ri->r = r;
4950 					SLIST_INSERT_HEAD(&match_rules, ri, entry);
4951 					pf_counter_u64_critical_enter();
4952 					pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
4953 					pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
4954 					pf_counter_u64_critical_exit();
4955 					pf_rule_to_actions(r, &pd->act);
4956 					if (r->log)
4957 						PFLOG_PACKET(kif, m, af,
4958 						    PFRES_MATCH, r,
4959 						    a, ruleset, pd, 1);
4960 				} else {
4961 					match = 1;
4962 					*rm = r;
4963 					*am = a;
4964 					*rsm = ruleset;
4965 				}
4966 				if ((*rm)->quick)
4967 					break;
4968 				r = TAILQ_NEXT(r, entries);
4969 			} else
4970 				pf_step_into_anchor(anchor_stack, &asd,
4971 				    &ruleset, PF_RULESET_FILTER, &r, &a,
4972 				    &match);
4973 		}
4974 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
4975 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
4976 			break;
4977 	}
4978 	r = *rm;
4979 	a = *am;
4980 	ruleset = *rsm;
4981 
4982 	REASON_SET(&reason, PFRES_MATCH);
4983 
4984 	/* apply actions for last matching pass/block rule */
4985 	pf_rule_to_actions(r, &pd->act);
4986 
4987 	if (r->log) {
4988 		if (rewrite)
4989 			m_copyback(m, off, hdrlen, pd->hdr.any);
4990 		PFLOG_PACKET(kif, m, af, reason, r, a, ruleset, pd, 1);
4991 	}
4992 
4993 	if ((r->action == PF_DROP) &&
4994 	    ((r->rule_flag & PFRULE_RETURNRST) ||
4995 	    (r->rule_flag & PFRULE_RETURNICMP) ||
4996 	    (r->rule_flag & PFRULE_RETURN))) {
4997 		pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum,
4998 		    bip_sum, hdrlen, &reason, r->rtableid);
4999 	}
5000 
5001 	if (r->action == PF_DROP)
5002 		goto cleanup;
5003 
5004 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
5005 		REASON_SET(&reason, PFRES_MEMORY);
5006 		goto cleanup;
5007 	}
5008 	if (pd->act.rtableid >= 0)
5009 		M_SETFIB(m, pd->act.rtableid);
5010 
5011 	if (!state_icmp && (r->keep_state || nr != NULL ||
5012 	    (pd->flags & PFDESC_TCP_NORM))) {
5013 		int action;
5014 		action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
5015 		    sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
5016 		    hdrlen, &match_rules);
5017 		if (action != PF_PASS) {
5018 			if (action == PF_DROP &&
5019 			    (r->rule_flag & PFRULE_RETURN))
5020 				pf_return(r, nr, pd, sk, off, m, th, kif,
5021 				    bproto_sum, bip_sum, hdrlen, &reason,
5022 				    pd->act.rtableid);
5023 			return (action);
5024 		}
5025 	} else {
5026 		while ((ri = SLIST_FIRST(&match_rules))) {
5027 			SLIST_REMOVE_HEAD(&match_rules, entry);
5028 			free(ri, M_PF_RULE_ITEM);
5029 		}
5030 
5031 		uma_zfree(V_pf_state_key_z, sk);
5032 		uma_zfree(V_pf_state_key_z, nk);
5033 	}
5034 
5035 	/* copy back packet headers if we performed NAT operations */
5036 	if (rewrite)
5037 		m_copyback(m, off, hdrlen, pd->hdr.any);
5038 
5039 	if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
5040 	    pd->dir == PF_OUT &&
5041 	    V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m))
5042 		/*
5043 		 * We want the state created, but we dont
5044 		 * want to send this in case a partner
5045 		 * firewall has to know about it to allow
5046 		 * replies through it.
5047 		 */
5048 		return (PF_DEFER);
5049 
5050 	return (PF_PASS);
5051 
5052 cleanup:
5053 	while ((ri = SLIST_FIRST(&match_rules))) {
5054 		SLIST_REMOVE_HEAD(&match_rules, entry);
5055 		free(ri, M_PF_RULE_ITEM);
5056 	}
5057 
5058 	uma_zfree(V_pf_state_key_z, sk);
5059 	uma_zfree(V_pf_state_key_z, nk);
5060 	return (PF_DROP);
5061 }
5062 
5063 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,struct pf_krule_slist * match_rules)5064 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a,
5065     struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk,
5066     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
5067     u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm,
5068     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
5069     struct pf_krule_slist *match_rules)
5070 {
5071 	struct pf_kstate	*s = NULL;
5072 	struct pf_ksrc_node	*sn = NULL;
5073 	struct tcphdr		*th = &pd->hdr.tcp;
5074 	u_int16_t		 mss = V_tcp_mssdflt;
5075 	u_short			 reason, sn_reason;
5076 	struct pf_krule_item	*ri;
5077 
5078 	/* check maximums */
5079 	if (r->max_states &&
5080 	    (counter_u64_fetch(r->states_cur) >= r->max_states)) {
5081 		counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
5082 		REASON_SET(&reason, PFRES_MAXSTATES);
5083 		goto csfailed;
5084 	}
5085 	/* src node for filter rule */
5086 	if ((r->rule_flag & PFRULE_SRCTRACK ||
5087 	    r->rpool.opts & PF_POOL_STICKYADDR) &&
5088 	    (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) {
5089 		REASON_SET(&reason, sn_reason);
5090 		goto csfailed;
5091 	}
5092 	/* src node for translation rule */
5093 	if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
5094 	    (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx],
5095 	    pd->af)) != 0 ) {
5096 		REASON_SET(&reason, sn_reason);
5097 		goto csfailed;
5098 	}
5099 	s = pf_alloc_state(M_NOWAIT);
5100 	if (s == NULL) {
5101 		REASON_SET(&reason, PFRES_MEMORY);
5102 		goto csfailed;
5103 	}
5104 	s->rule.ptr = r;
5105 	s->nat_rule.ptr = nr;
5106 	s->anchor.ptr = a;
5107 	bcopy(match_rules, &s->match_rules, sizeof(s->match_rules));
5108 	memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions));
5109 
5110 	STATE_INC_COUNTERS(s);
5111 	if (r->allow_opts)
5112 		s->state_flags |= PFSTATE_ALLOWOPTS;
5113 	if (r->rule_flag & PFRULE_STATESLOPPY)
5114 		s->state_flags |= PFSTATE_SLOPPY;
5115 	if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */
5116 		s->state_flags |= PFSTATE_SCRUB_TCP;
5117 
5118 	s->act.log = pd->act.log & PF_LOG_ALL;
5119 	s->sync_state = PFSYNC_S_NONE;
5120 	s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */
5121 
5122 	if (nr != NULL)
5123 		s->act.log |= nr->log & PF_LOG_ALL;
5124 	switch (pd->proto) {
5125 	case IPPROTO_TCP:
5126 		s->src.seqlo = ntohl(th->th_seq);
5127 		s->src.seqhi = s->src.seqlo + pd->p_len + 1;
5128 		if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
5129 		    r->keep_state == PF_STATE_MODULATE) {
5130 			/* Generate sequence number modulator */
5131 			if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
5132 			    0)
5133 				s->src.seqdiff = 1;
5134 			pf_change_proto_a(m, &th->th_seq, &th->th_sum,
5135 			    htonl(s->src.seqlo + s->src.seqdiff), 0);
5136 			*rewrite = 1;
5137 		} else
5138 			s->src.seqdiff = 0;
5139 		if (th->th_flags & TH_SYN) {
5140 			s->src.seqhi++;
5141 			s->src.wscale = pf_get_wscale(m, off,
5142 			    th->th_off, pd->af);
5143 		}
5144 		s->src.max_win = MAX(ntohs(th->th_win), 1);
5145 		if (s->src.wscale & PF_WSCALE_MASK) {
5146 			/* Remove scale factor from initial window */
5147 			int win = s->src.max_win;
5148 			win += 1 << (s->src.wscale & PF_WSCALE_MASK);
5149 			s->src.max_win = (win - 1) >>
5150 			    (s->src.wscale & PF_WSCALE_MASK);
5151 		}
5152 		if (th->th_flags & TH_FIN)
5153 			s->src.seqhi++;
5154 		s->dst.seqhi = 1;
5155 		s->dst.max_win = 1;
5156 		pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT);
5157 		pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED);
5158 		s->timeout = PFTM_TCP_FIRST_PACKET;
5159 		atomic_add_32(&V_pf_status.states_halfopen, 1);
5160 		break;
5161 	case IPPROTO_UDP:
5162 		pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE);
5163 		pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC);
5164 		s->timeout = PFTM_UDP_FIRST_PACKET;
5165 		break;
5166 	case IPPROTO_SCTP:
5167 		pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT);
5168 		pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED);
5169 		s->timeout = PFTM_SCTP_FIRST_PACKET;
5170 		break;
5171 	case IPPROTO_ICMP:
5172 #ifdef INET6
5173 	case IPPROTO_ICMPV6:
5174 #endif
5175 		s->timeout = PFTM_ICMP_FIRST_PACKET;
5176 		break;
5177 	default:
5178 		pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE);
5179 		pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC);
5180 		s->timeout = PFTM_OTHER_FIRST_PACKET;
5181 	}
5182 
5183 	if (r->rt) {
5184 		/* pf_map_addr increases the reason counters */
5185 		if ((reason = pf_map_addr(pd->af, r, pd->src, &s->rt_addr,
5186 		    &s->rt_kif, NULL, &sn)) != 0)
5187 			goto csfailed;
5188 		s->rt = r->rt;
5189 	}
5190 
5191 	s->creation = time_uptime;
5192 	s->expire = time_uptime;
5193 
5194 	if (sn != NULL)
5195 		s->src_node = sn;
5196 	if (nsn != NULL) {
5197 		/* XXX We only modify one side for now. */
5198 		PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
5199 		s->nat_src_node = nsn;
5200 	}
5201 	if (pd->proto == IPPROTO_TCP) {
5202 		if (s->state_flags & PFSTATE_SCRUB_TCP &&
5203 		    pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) {
5204 			REASON_SET(&reason, PFRES_MEMORY);
5205 			goto drop;
5206 		}
5207 		if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub &&
5208 		    pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
5209 		    &s->src, &s->dst, rewrite)) {
5210 			/* This really shouldn't happen!!! */
5211 			DPFPRINTF(PF_DEBUG_URGENT,
5212 			    ("pf_normalize_tcp_stateful failed on first "
5213 			     "pkt\n"));
5214 			goto drop;
5215 		}
5216 	} else if (pd->proto == IPPROTO_SCTP) {
5217 		if (pf_normalize_sctp_init(m, off, pd, &s->src, &s->dst))
5218 			goto drop;
5219 		if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP)))
5220 			goto drop;
5221 	}
5222 	s->direction = pd->dir;
5223 
5224 	/*
5225 	 * sk/nk could already been setup by pf_get_translation().
5226 	 */
5227 	if (nr == NULL) {
5228 		KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
5229 		    __func__, nr, sk, nk));
5230 		sk = pf_state_key_setup(pd, m, off, pd->src, pd->dst, sport, dport);
5231 		if (sk == NULL)
5232 			goto csfailed;
5233 		nk = sk;
5234 	} else
5235 		KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
5236 		    __func__, nr, sk, nk));
5237 
5238 	/* Swap sk/nk for PF_OUT. */
5239 	if (pf_state_insert(BOUND_IFACE(r, kif), kif,
5240 	    (pd->dir == PF_IN) ? sk : nk,
5241 	    (pd->dir == PF_IN) ? nk : sk, s)) {
5242 		REASON_SET(&reason, PFRES_STATEINS);
5243 		goto drop;
5244 	} else
5245 		*sm = s;
5246 
5247 	if (tag > 0)
5248 		s->tag = tag;
5249 	if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
5250 	    TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
5251 		pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC);
5252 		/* undo NAT changes, if they have taken place */
5253 		if (nr != NULL) {
5254 			struct pf_state_key *skt = s->key[PF_SK_WIRE];
5255 			if (pd->dir == PF_OUT)
5256 				skt = s->key[PF_SK_STACK];
5257 			PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
5258 			PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
5259 			if (pd->sport)
5260 				*pd->sport = skt->port[pd->sidx];
5261 			if (pd->dport)
5262 				*pd->dport = skt->port[pd->didx];
5263 			if (pd->proto_sum)
5264 				*pd->proto_sum = bproto_sum;
5265 			if (pd->ip_sum)
5266 				*pd->ip_sum = bip_sum;
5267 			m_copyback(m, off, hdrlen, pd->hdr.any);
5268 		}
5269 		s->src.seqhi = htonl(arc4random());
5270 		/* Find mss option */
5271 		int rtid = M_GETFIB(m);
5272 		mss = pf_get_mss(m, off, th->th_off, pd->af);
5273 		mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
5274 		mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
5275 		s->src.mss = mss;
5276 		pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport,
5277 		    th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
5278 		    TH_SYN|TH_ACK, 0, s->src.mss, 0, true, 0, 0,
5279 		    pd->act.rtableid);
5280 		REASON_SET(&reason, PFRES_SYNPROXY);
5281 		return (PF_SYNPROXY_DROP);
5282 	}
5283 
5284 	return (PF_PASS);
5285 
5286 csfailed:
5287 	while ((ri = SLIST_FIRST(match_rules))) {
5288 		SLIST_REMOVE_HEAD(match_rules, entry);
5289 		free(ri, M_PF_RULE_ITEM);
5290 	}
5291 
5292 	uma_zfree(V_pf_state_key_z, sk);
5293 	uma_zfree(V_pf_state_key_z, nk);
5294 
5295 	if (sn != NULL) {
5296 		PF_SRC_NODE_LOCK(sn);
5297 		if (--sn->states == 0 && sn->expire == 0) {
5298 			pf_unlink_src_node(sn);
5299 			uma_zfree(V_pf_sources_z, sn);
5300 			counter_u64_add(
5301 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
5302 		}
5303 		PF_SRC_NODE_UNLOCK(sn);
5304 	}
5305 
5306 	if (nsn != sn && nsn != NULL) {
5307 		PF_SRC_NODE_LOCK(nsn);
5308 		if (--nsn->states == 0 && nsn->expire == 0) {
5309 			pf_unlink_src_node(nsn);
5310 			uma_zfree(V_pf_sources_z, nsn);
5311 			counter_u64_add(
5312 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
5313 		}
5314 		PF_SRC_NODE_UNLOCK(nsn);
5315 	}
5316 
5317 drop:
5318 	if (s != NULL) {
5319 		pf_src_tree_remove_state(s);
5320 		s->timeout = PFTM_UNLINKED;
5321 		STATE_DEC_COUNTERS(s);
5322 		pf_free_state(s);
5323 	}
5324 
5325 	return (PF_DROP);
5326 }
5327 
5328 static int
pf_test_fragment(struct pf_krule ** rm,struct pfi_kkif * kif,struct mbuf * m,void * h,struct pf_pdesc * pd,struct pf_krule ** am,struct pf_kruleset ** rsm)5329 pf_test_fragment(struct pf_krule **rm, struct pfi_kkif *kif,
5330     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am,
5331     struct pf_kruleset **rsm)
5332 {
5333 	struct pf_krule		*r, *a = NULL;
5334 	struct pf_kruleset	*ruleset = NULL;
5335 	struct pf_krule_slist	 match_rules;
5336 	struct pf_krule_item	*ri;
5337 	sa_family_t		 af = pd->af;
5338 	u_short			 reason;
5339 	int			 tag = -1;
5340 	int			 asd = 0;
5341 	int			 match = 0;
5342 	struct pf_kanchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
5343 
5344 	PF_RULES_RASSERT();
5345 
5346 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
5347 	SLIST_INIT(&match_rules);
5348 	while (r != NULL) {
5349 		pf_counter_u64_add(&r->evaluations, 1);
5350 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
5351 			r = r->skip[PF_SKIP_IFP].ptr;
5352 		else if (r->direction && r->direction != pd->dir)
5353 			r = r->skip[PF_SKIP_DIR].ptr;
5354 		else if (r->af && r->af != af)
5355 			r = r->skip[PF_SKIP_AF].ptr;
5356 		else if (r->proto && r->proto != pd->proto)
5357 			r = r->skip[PF_SKIP_PROTO].ptr;
5358 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
5359 		    r->src.neg, kif, M_GETFIB(m)))
5360 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
5361 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
5362 		    r->dst.neg, NULL, M_GETFIB(m)))
5363 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
5364 		else if (r->tos && !(r->tos == pd->tos))
5365 			r = TAILQ_NEXT(r, entries);
5366 		else if (r->os_fingerprint != PF_OSFP_ANY)
5367 			r = TAILQ_NEXT(r, entries);
5368 		else if (pd->proto == IPPROTO_UDP &&
5369 		    (r->src.port_op || r->dst.port_op))
5370 			r = TAILQ_NEXT(r, entries);
5371 		else if (pd->proto == IPPROTO_TCP &&
5372 		    (r->src.port_op || r->dst.port_op || r->flagset))
5373 			r = TAILQ_NEXT(r, entries);
5374 		else if ((pd->proto == IPPROTO_ICMP ||
5375 		    pd->proto == IPPROTO_ICMPV6) &&
5376 		    (r->type || r->code))
5377 			r = TAILQ_NEXT(r, entries);
5378 		else if (r->prio &&
5379 		    !pf_match_ieee8021q_pcp(r->prio, m))
5380 			r = TAILQ_NEXT(r, entries);
5381 		else if (r->prob && r->prob <=
5382 		    (arc4random() % (UINT_MAX - 1) + 1))
5383 			r = TAILQ_NEXT(r, entries);
5384 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
5385 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
5386 			r = TAILQ_NEXT(r, entries);
5387 		else {
5388 			if (r->anchor == NULL) {
5389 				if (r->action == PF_MATCH) {
5390 					ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO);
5391 					if (ri == NULL) {
5392 						REASON_SET(&reason, PFRES_MEMORY);
5393 						goto cleanup;
5394 					}
5395 					ri->r = r;
5396 					SLIST_INSERT_HEAD(&match_rules, ri, entry);
5397 					pf_counter_u64_critical_enter();
5398 					pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
5399 					pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
5400 					pf_counter_u64_critical_exit();
5401 					pf_rule_to_actions(r, &pd->act);
5402 					if (r->log)
5403 						PFLOG_PACKET(kif, m, af,
5404 						    PFRES_MATCH, r,
5405 						    a, ruleset, pd, 1);
5406 				} else {
5407 					match = 1;
5408 					*rm = r;
5409 					*am = a;
5410 					*rsm = ruleset;
5411 				}
5412 				if ((*rm)->quick)
5413 					break;
5414 				r = TAILQ_NEXT(r, entries);
5415 			} else
5416 				pf_step_into_anchor(anchor_stack, &asd,
5417 				    &ruleset, PF_RULESET_FILTER, &r, &a,
5418 				    &match);
5419 		}
5420 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
5421 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
5422 			break;
5423 	}
5424 	r = *rm;
5425 	a = *am;
5426 	ruleset = *rsm;
5427 
5428 	REASON_SET(&reason, PFRES_MATCH);
5429 
5430 	/* apply actions for last matching pass/block rule */
5431 	pf_rule_to_actions(r, &pd->act);
5432 
5433 	if (r->log)
5434 		PFLOG_PACKET(kif, m, af, reason, r, a, ruleset, pd, 1);
5435 
5436 	if (r->action != PF_PASS)
5437 		return (PF_DROP);
5438 
5439 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
5440 		REASON_SET(&reason, PFRES_MEMORY);
5441 		goto cleanup;
5442 	}
5443 
5444 	return (PF_PASS);
5445 
5446 cleanup:
5447 	while ((ri = SLIST_FIRST(&match_rules))) {
5448 		SLIST_REMOVE_HEAD(&match_rules, entry);
5449 		free(ri, M_PF_RULE_ITEM);
5450 	}
5451 
5452 	return (PF_DROP);
5453 }
5454 
5455 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)5456 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif,
5457     struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason,
5458     int *copyback)
5459 {
5460 	struct tcphdr		*th = &pd->hdr.tcp;
5461 	struct pf_state_peer	*src, *dst;
5462 	u_int16_t		 win = ntohs(th->th_win);
5463 	u_int32_t		 ack, end, seq, orig_seq;
5464 	u_int8_t		 sws, dws, psrc, pdst;
5465 	int			 ackskew;
5466 
5467 	if (pd->dir == (*state)->direction) {
5468 		src = &(*state)->src;
5469 		dst = &(*state)->dst;
5470 		psrc = PF_PEER_SRC;
5471 		pdst = PF_PEER_DST;
5472 	} else {
5473 		src = &(*state)->dst;
5474 		dst = &(*state)->src;
5475 		psrc = PF_PEER_DST;
5476 		pdst = PF_PEER_SRC;
5477 	}
5478 
5479 	if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
5480 		sws = src->wscale & PF_WSCALE_MASK;
5481 		dws = dst->wscale & PF_WSCALE_MASK;
5482 	} else
5483 		sws = dws = 0;
5484 
5485 	/*
5486 	 * Sequence tracking algorithm from Guido van Rooij's paper:
5487 	 *   http://www.madison-gurkha.com/publications/tcp_filtering/
5488 	 *	tcp_filtering.ps
5489 	 */
5490 
5491 	orig_seq = seq = ntohl(th->th_seq);
5492 	if (src->seqlo == 0) {
5493 		/* First packet from this end. Set its state */
5494 
5495 		if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) &&
5496 		    src->scrub == NULL) {
5497 			if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
5498 				REASON_SET(reason, PFRES_MEMORY);
5499 				return (PF_DROP);
5500 			}
5501 		}
5502 
5503 		/* Deferred generation of sequence number modulator */
5504 		if (dst->seqdiff && !src->seqdiff) {
5505 			/* use random iss for the TCP server */
5506 			while ((src->seqdiff = arc4random() - seq) == 0)
5507 				;
5508 			ack = ntohl(th->th_ack) - dst->seqdiff;
5509 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
5510 			    src->seqdiff), 0);
5511 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
5512 			*copyback = 1;
5513 		} else {
5514 			ack = ntohl(th->th_ack);
5515 		}
5516 
5517 		end = seq + pd->p_len;
5518 		if (th->th_flags & TH_SYN) {
5519 			end++;
5520 			if (dst->wscale & PF_WSCALE_FLAG) {
5521 				src->wscale = pf_get_wscale(m, off, th->th_off,
5522 				    pd->af);
5523 				if (src->wscale & PF_WSCALE_FLAG) {
5524 					/* Remove scale factor from initial
5525 					 * window */
5526 					sws = src->wscale & PF_WSCALE_MASK;
5527 					win = ((u_int32_t)win + (1 << sws) - 1)
5528 					    >> sws;
5529 					dws = dst->wscale & PF_WSCALE_MASK;
5530 				} else {
5531 					/* fixup other window */
5532 					dst->max_win <<= dst->wscale &
5533 					    PF_WSCALE_MASK;
5534 					/* in case of a retrans SYN|ACK */
5535 					dst->wscale = 0;
5536 				}
5537 			}
5538 		}
5539 		if (th->th_flags & TH_FIN)
5540 			end++;
5541 
5542 		src->seqlo = seq;
5543 		if (src->state < TCPS_SYN_SENT)
5544 			pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
5545 
5546 		/*
5547 		 * May need to slide the window (seqhi may have been set by
5548 		 * the crappy stack check or if we picked up the connection
5549 		 * after establishment)
5550 		 */
5551 		if (src->seqhi == 1 ||
5552 		    SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
5553 			src->seqhi = end + MAX(1, dst->max_win << dws);
5554 		if (win > src->max_win)
5555 			src->max_win = win;
5556 
5557 	} else {
5558 		ack = ntohl(th->th_ack) - dst->seqdiff;
5559 		if (src->seqdiff) {
5560 			/* Modulate sequence numbers */
5561 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
5562 			    src->seqdiff), 0);
5563 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
5564 			*copyback = 1;
5565 		}
5566 		end = seq + pd->p_len;
5567 		if (th->th_flags & TH_SYN)
5568 			end++;
5569 		if (th->th_flags & TH_FIN)
5570 			end++;
5571 	}
5572 
5573 	if ((th->th_flags & TH_ACK) == 0) {
5574 		/* Let it pass through the ack skew check */
5575 		ack = dst->seqlo;
5576 	} else if ((ack == 0 &&
5577 	    (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
5578 	    /* broken tcp stacks do not set ack */
5579 	    (dst->state < TCPS_SYN_SENT)) {
5580 		/*
5581 		 * Many stacks (ours included) will set the ACK number in an
5582 		 * FIN|ACK if the SYN times out -- no sequence to ACK.
5583 		 */
5584 		ack = dst->seqlo;
5585 	}
5586 
5587 	if (seq == end) {
5588 		/* Ease sequencing restrictions on no data packets */
5589 		seq = src->seqlo;
5590 		end = seq;
5591 	}
5592 
5593 	ackskew = dst->seqlo - ack;
5594 
5595 	/*
5596 	 * Need to demodulate the sequence numbers in any TCP SACK options
5597 	 * (Selective ACK). We could optionally validate the SACK values
5598 	 * against the current ACK window, either forwards or backwards, but
5599 	 * I'm not confident that SACK has been implemented properly
5600 	 * everywhere. It wouldn't surprise me if several stacks accidentally
5601 	 * SACK too far backwards of previously ACKed data. There really aren't
5602 	 * any security implications of bad SACKing unless the target stack
5603 	 * doesn't validate the option length correctly. Someone trying to
5604 	 * spoof into a TCP connection won't bother blindly sending SACK
5605 	 * options anyway.
5606 	 */
5607 	if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
5608 		if (pf_modulate_sack(m, off, pd, th, dst))
5609 			*copyback = 1;
5610 	}
5611 
5612 #define	MAXACKWINDOW (0xffff + 1500)	/* 1500 is an arbitrary fudge factor */
5613 	if (SEQ_GEQ(src->seqhi, end) &&
5614 	    /* Last octet inside other's window space */
5615 	    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
5616 	    /* Retrans: not more than one window back */
5617 	    (ackskew >= -MAXACKWINDOW) &&
5618 	    /* Acking not more than one reassembled fragment backwards */
5619 	    (ackskew <= (MAXACKWINDOW << sws)) &&
5620 	    /* Acking not more than one window forward */
5621 	    ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
5622 	    (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) {
5623 	    /* Require an exact/+1 sequence match on resets when possible */
5624 
5625 		if (dst->scrub || src->scrub) {
5626 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
5627 			    *state, src, dst, copyback))
5628 				return (PF_DROP);
5629 		}
5630 
5631 		/* update max window */
5632 		if (src->max_win < win)
5633 			src->max_win = win;
5634 		/* synchronize sequencing */
5635 		if (SEQ_GT(end, src->seqlo))
5636 			src->seqlo = end;
5637 		/* slide the window of what the other end can send */
5638 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
5639 			dst->seqhi = ack + MAX((win << sws), 1);
5640 
5641 		/* update states */
5642 		if (th->th_flags & TH_SYN)
5643 			if (src->state < TCPS_SYN_SENT)
5644 				pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
5645 		if (th->th_flags & TH_FIN)
5646 			if (src->state < TCPS_CLOSING)
5647 				pf_set_protostate(*state, psrc, TCPS_CLOSING);
5648 		if (th->th_flags & TH_ACK) {
5649 			if (dst->state == TCPS_SYN_SENT) {
5650 				pf_set_protostate(*state, pdst,
5651 				    TCPS_ESTABLISHED);
5652 				if (src->state == TCPS_ESTABLISHED &&
5653 				    (*state)->src_node != NULL &&
5654 				    pf_src_connlimit(state)) {
5655 					REASON_SET(reason, PFRES_SRCLIMIT);
5656 					return (PF_DROP);
5657 				}
5658 			} else if (dst->state == TCPS_CLOSING)
5659 				pf_set_protostate(*state, pdst,
5660 				    TCPS_FIN_WAIT_2);
5661 		}
5662 		if (th->th_flags & TH_RST)
5663 			pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
5664 
5665 		/* update expire time */
5666 		(*state)->expire = time_uptime;
5667 		if (src->state >= TCPS_FIN_WAIT_2 &&
5668 		    dst->state >= TCPS_FIN_WAIT_2)
5669 			(*state)->timeout = PFTM_TCP_CLOSED;
5670 		else if (src->state >= TCPS_CLOSING &&
5671 		    dst->state >= TCPS_CLOSING)
5672 			(*state)->timeout = PFTM_TCP_FIN_WAIT;
5673 		else if (src->state < TCPS_ESTABLISHED ||
5674 		    dst->state < TCPS_ESTABLISHED)
5675 			(*state)->timeout = PFTM_TCP_OPENING;
5676 		else if (src->state >= TCPS_CLOSING ||
5677 		    dst->state >= TCPS_CLOSING)
5678 			(*state)->timeout = PFTM_TCP_CLOSING;
5679 		else
5680 			(*state)->timeout = PFTM_TCP_ESTABLISHED;
5681 
5682 		/* Fall through to PASS packet */
5683 
5684 	} else if ((dst->state < TCPS_SYN_SENT ||
5685 		dst->state >= TCPS_FIN_WAIT_2 ||
5686 		src->state >= TCPS_FIN_WAIT_2) &&
5687 	    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
5688 	    /* Within a window forward of the originating packet */
5689 	    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
5690 	    /* Within a window backward of the originating packet */
5691 
5692 		/*
5693 		 * This currently handles three situations:
5694 		 *  1) Stupid stacks will shotgun SYNs before their peer
5695 		 *     replies.
5696 		 *  2) When PF catches an already established stream (the
5697 		 *     firewall rebooted, the state table was flushed, routes
5698 		 *     changed...)
5699 		 *  3) Packets get funky immediately after the connection
5700 		 *     closes (this should catch Solaris spurious ACK|FINs
5701 		 *     that web servers like to spew after a close)
5702 		 *
5703 		 * This must be a little more careful than the above code
5704 		 * since packet floods will also be caught here. We don't
5705 		 * update the TTL here to mitigate the damage of a packet
5706 		 * flood and so the same code can handle awkward establishment
5707 		 * and a loosened connection close.
5708 		 * In the establishment case, a correct peer response will
5709 		 * validate the connection, go through the normal state code
5710 		 * and keep updating the state TTL.
5711 		 */
5712 
5713 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
5714 			printf("pf: loose state match: ");
5715 			pf_print_state(*state);
5716 			pf_print_flags(th->th_flags);
5717 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
5718 			    "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
5719 			    pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
5720 			    (unsigned long long)(*state)->packets[1],
5721 			    pd->dir == PF_IN ? "in" : "out",
5722 			    pd->dir == (*state)->direction ? "fwd" : "rev");
5723 		}
5724 
5725 		if (dst->scrub || src->scrub) {
5726 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
5727 			    *state, src, dst, copyback))
5728 				return (PF_DROP);
5729 		}
5730 
5731 		/* update max window */
5732 		if (src->max_win < win)
5733 			src->max_win = win;
5734 		/* synchronize sequencing */
5735 		if (SEQ_GT(end, src->seqlo))
5736 			src->seqlo = end;
5737 		/* slide the window of what the other end can send */
5738 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
5739 			dst->seqhi = ack + MAX((win << sws), 1);
5740 
5741 		/*
5742 		 * Cannot set dst->seqhi here since this could be a shotgunned
5743 		 * SYN and not an already established connection.
5744 		 */
5745 
5746 		if (th->th_flags & TH_FIN)
5747 			if (src->state < TCPS_CLOSING)
5748 				pf_set_protostate(*state, psrc, TCPS_CLOSING);
5749 		if (th->th_flags & TH_RST)
5750 			pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
5751 
5752 		/* Fall through to PASS packet */
5753 
5754 	} else {
5755 		if ((*state)->dst.state == TCPS_SYN_SENT &&
5756 		    (*state)->src.state == TCPS_SYN_SENT) {
5757 			/* Send RST for state mismatches during handshake */
5758 			if (!(th->th_flags & TH_RST))
5759 				pf_send_tcp((*state)->rule.ptr, pd->af,
5760 				    pd->dst, pd->src, th->th_dport,
5761 				    th->th_sport, ntohl(th->th_ack), 0,
5762 				    TH_RST, 0, 0,
5763 				    (*state)->rule.ptr->return_ttl, true, 0, 0,
5764 				    (*state)->act.rtableid);
5765 			src->seqlo = 0;
5766 			src->seqhi = 1;
5767 			src->max_win = 1;
5768 		} else if (V_pf_status.debug >= PF_DEBUG_MISC) {
5769 			printf("pf: BAD state: ");
5770 			pf_print_state(*state);
5771 			pf_print_flags(th->th_flags);
5772 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
5773 			    "pkts=%llu:%llu dir=%s,%s\n",
5774 			    seq, orig_seq, ack, pd->p_len, ackskew,
5775 			    (unsigned long long)(*state)->packets[0],
5776 			    (unsigned long long)(*state)->packets[1],
5777 			    pd->dir == PF_IN ? "in" : "out",
5778 			    pd->dir == (*state)->direction ? "fwd" : "rev");
5779 			printf("pf: State failure on: %c %c %c %c | %c %c\n",
5780 			    SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
5781 			    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
5782 			    ' ': '2',
5783 			    (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
5784 			    (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
5785 			    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
5786 			    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
5787 		}
5788 		REASON_SET(reason, PFRES_BADSTATE);
5789 		return (PF_DROP);
5790 	}
5791 
5792 	return (PF_PASS);
5793 }
5794 
5795 static int
pf_tcp_track_sloppy(struct pf_kstate ** state,struct pf_pdesc * pd,u_short * reason)5796 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason)
5797 {
5798 	struct tcphdr		*th = &pd->hdr.tcp;
5799 	struct pf_state_peer	*src, *dst;
5800 	u_int8_t		 psrc, pdst;
5801 
5802 	if (pd->dir == (*state)->direction) {
5803 		src = &(*state)->src;
5804 		dst = &(*state)->dst;
5805 		psrc = PF_PEER_SRC;
5806 		pdst = PF_PEER_DST;
5807 	} else {
5808 		src = &(*state)->dst;
5809 		dst = &(*state)->src;
5810 		psrc = PF_PEER_DST;
5811 		pdst = PF_PEER_SRC;
5812 	}
5813 
5814 	if (th->th_flags & TH_SYN)
5815 		if (src->state < TCPS_SYN_SENT)
5816 			pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
5817 	if (th->th_flags & TH_FIN)
5818 		if (src->state < TCPS_CLOSING)
5819 			pf_set_protostate(*state, psrc, TCPS_CLOSING);
5820 	if (th->th_flags & TH_ACK) {
5821 		if (dst->state == TCPS_SYN_SENT) {
5822 			pf_set_protostate(*state, pdst, TCPS_ESTABLISHED);
5823 			if (src->state == TCPS_ESTABLISHED &&
5824 			    (*state)->src_node != NULL &&
5825 			    pf_src_connlimit(state)) {
5826 				REASON_SET(reason, PFRES_SRCLIMIT);
5827 				return (PF_DROP);
5828 			}
5829 		} else if (dst->state == TCPS_CLOSING) {
5830 			pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2);
5831 		} else if (src->state == TCPS_SYN_SENT &&
5832 		    dst->state < TCPS_SYN_SENT) {
5833 			/*
5834 			 * Handle a special sloppy case where we only see one
5835 			 * half of the connection. If there is a ACK after
5836 			 * the initial SYN without ever seeing a packet from
5837 			 * the destination, set the connection to established.
5838 			 */
5839 			pf_set_protostate(*state, PF_PEER_BOTH,
5840 			    TCPS_ESTABLISHED);
5841 			dst->state = src->state = TCPS_ESTABLISHED;
5842 			if ((*state)->src_node != NULL &&
5843 			    pf_src_connlimit(state)) {
5844 				REASON_SET(reason, PFRES_SRCLIMIT);
5845 				return (PF_DROP);
5846 			}
5847 		} else if (src->state == TCPS_CLOSING &&
5848 		    dst->state == TCPS_ESTABLISHED &&
5849 		    dst->seqlo == 0) {
5850 			/*
5851 			 * Handle the closing of half connections where we
5852 			 * don't see the full bidirectional FIN/ACK+ACK
5853 			 * handshake.
5854 			 */
5855 			pf_set_protostate(*state, pdst, TCPS_CLOSING);
5856 		}
5857 	}
5858 	if (th->th_flags & TH_RST)
5859 		pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
5860 
5861 	/* update expire time */
5862 	(*state)->expire = time_uptime;
5863 	if (src->state >= TCPS_FIN_WAIT_2 &&
5864 	    dst->state >= TCPS_FIN_WAIT_2)
5865 		(*state)->timeout = PFTM_TCP_CLOSED;
5866 	else if (src->state >= TCPS_CLOSING &&
5867 	    dst->state >= TCPS_CLOSING)
5868 		(*state)->timeout = PFTM_TCP_FIN_WAIT;
5869 	else if (src->state < TCPS_ESTABLISHED ||
5870 	    dst->state < TCPS_ESTABLISHED)
5871 		(*state)->timeout = PFTM_TCP_OPENING;
5872 	else if (src->state >= TCPS_CLOSING ||
5873 	    dst->state >= TCPS_CLOSING)
5874 		(*state)->timeout = PFTM_TCP_CLOSING;
5875 	else
5876 		(*state)->timeout = PFTM_TCP_ESTABLISHED;
5877 
5878 	return (PF_PASS);
5879 }
5880 
5881 static int
pf_synproxy(struct pf_pdesc * pd,struct pf_kstate ** state,u_short * reason)5882 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason)
5883 {
5884 	struct pf_state_key	*sk = (*state)->key[pd->didx];
5885 	struct tcphdr		*th = &pd->hdr.tcp;
5886 
5887 	if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
5888 		if (pd->dir != (*state)->direction) {
5889 			REASON_SET(reason, PFRES_SYNPROXY);
5890 			return (PF_SYNPROXY_DROP);
5891 		}
5892 		if (th->th_flags & TH_SYN) {
5893 			if (ntohl(th->th_seq) != (*state)->src.seqlo) {
5894 				REASON_SET(reason, PFRES_SYNPROXY);
5895 				return (PF_DROP);
5896 			}
5897 			pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
5898 			    pd->src, th->th_dport, th->th_sport,
5899 			    (*state)->src.seqhi, ntohl(th->th_seq) + 1,
5900 			    TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, true, 0, 0,
5901 			    (*state)->act.rtableid);
5902 			REASON_SET(reason, PFRES_SYNPROXY);
5903 			return (PF_SYNPROXY_DROP);
5904 		} else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK ||
5905 		    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
5906 		    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
5907 			REASON_SET(reason, PFRES_SYNPROXY);
5908 			return (PF_DROP);
5909 		} else if ((*state)->src_node != NULL &&
5910 		    pf_src_connlimit(state)) {
5911 			REASON_SET(reason, PFRES_SRCLIMIT);
5912 			return (PF_DROP);
5913 		} else
5914 			pf_set_protostate(*state, PF_PEER_SRC,
5915 			    PF_TCPS_PROXY_DST);
5916 	}
5917 	if ((*state)->src.state == PF_TCPS_PROXY_DST) {
5918 		if (pd->dir == (*state)->direction) {
5919 			if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
5920 			    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
5921 			    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
5922 				REASON_SET(reason, PFRES_SYNPROXY);
5923 				return (PF_DROP);
5924 			}
5925 			(*state)->src.max_win = MAX(ntohs(th->th_win), 1);
5926 			if ((*state)->dst.seqhi == 1)
5927 				(*state)->dst.seqhi = htonl(arc4random());
5928 			pf_send_tcp((*state)->rule.ptr, pd->af,
5929 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
5930 			    sk->port[pd->sidx], sk->port[pd->didx],
5931 			    (*state)->dst.seqhi, 0, TH_SYN, 0,
5932 			    (*state)->src.mss, 0, false, (*state)->tag, 0,
5933 			    (*state)->act.rtableid);
5934 			REASON_SET(reason, PFRES_SYNPROXY);
5935 			return (PF_SYNPROXY_DROP);
5936 		} else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
5937 		    (TH_SYN|TH_ACK)) ||
5938 		    (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
5939 			REASON_SET(reason, PFRES_SYNPROXY);
5940 			return (PF_DROP);
5941 		} else {
5942 			(*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
5943 			(*state)->dst.seqlo = ntohl(th->th_seq);
5944 			pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
5945 			    pd->src, th->th_dport, th->th_sport,
5946 			    ntohl(th->th_ack), ntohl(th->th_seq) + 1,
5947 			    TH_ACK, (*state)->src.max_win, 0, 0, false,
5948 			    (*state)->tag, 0, (*state)->act.rtableid);
5949 			pf_send_tcp((*state)->rule.ptr, pd->af,
5950 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
5951 			    sk->port[pd->sidx], sk->port[pd->didx],
5952 			    (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
5953 			    TH_ACK, (*state)->dst.max_win, 0, 0, true, 0, 0,
5954 			    (*state)->act.rtableid);
5955 			(*state)->src.seqdiff = (*state)->dst.seqhi -
5956 			    (*state)->src.seqlo;
5957 			(*state)->dst.seqdiff = (*state)->src.seqhi -
5958 			    (*state)->dst.seqlo;
5959 			(*state)->src.seqhi = (*state)->src.seqlo +
5960 			    (*state)->dst.max_win;
5961 			(*state)->dst.seqhi = (*state)->dst.seqlo +
5962 			    (*state)->src.max_win;
5963 			(*state)->src.wscale = (*state)->dst.wscale = 0;
5964 			pf_set_protostate(*state, PF_PEER_BOTH,
5965 			    TCPS_ESTABLISHED);
5966 			REASON_SET(reason, PFRES_SYNPROXY);
5967 			return (PF_SYNPROXY_DROP);
5968 		}
5969 	}
5970 
5971 	return (PF_PASS);
5972 }
5973 
5974 static int
pf_test_state_tcp(struct pf_kstate ** state,struct pfi_kkif * kif,struct mbuf * m,int off,void * h,struct pf_pdesc * pd,u_short * reason)5975 pf_test_state_tcp(struct pf_kstate **state, struct pfi_kkif *kif,
5976     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
5977     u_short *reason)
5978 {
5979 	struct pf_state_key_cmp	 key;
5980 	struct tcphdr		*th = &pd->hdr.tcp;
5981 	int			 copyback = 0;
5982 	int			 action;
5983 	struct pf_state_peer	*src, *dst;
5984 
5985 	bzero(&key, sizeof(key));
5986 	key.af = pd->af;
5987 	key.proto = IPPROTO_TCP;
5988 	if (pd->dir == PF_IN)	{	/* wire side, straight */
5989 		PF_ACPY(&key.addr[0], pd->src, key.af);
5990 		PF_ACPY(&key.addr[1], pd->dst, key.af);
5991 		key.port[0] = th->th_sport;
5992 		key.port[1] = th->th_dport;
5993 	} else {			/* stack side, reverse */
5994 		PF_ACPY(&key.addr[1], pd->src, key.af);
5995 		PF_ACPY(&key.addr[0], pd->dst, key.af);
5996 		key.port[1] = th->th_sport;
5997 		key.port[0] = th->th_dport;
5998 	}
5999 
6000 	STATE_LOOKUP(kif, &key, *state, pd);
6001 
6002 	if (pd->dir == (*state)->direction) {
6003 		src = &(*state)->src;
6004 		dst = &(*state)->dst;
6005 	} else {
6006 		src = &(*state)->dst;
6007 		dst = &(*state)->src;
6008 	}
6009 
6010 	if ((action = pf_synproxy(pd, state, reason)) != PF_PASS)
6011 		return (action);
6012 
6013 	if (dst->state >= TCPS_FIN_WAIT_2 &&
6014 	    src->state >= TCPS_FIN_WAIT_2 &&
6015 	    (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) ||
6016 	    ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK &&
6017 	    pf_syncookie_check(pd) && pd->dir == PF_IN))) {
6018 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
6019 			printf("pf: state reuse ");
6020 			pf_print_state(*state);
6021 			pf_print_flags(th->th_flags);
6022 			printf("\n");
6023 		}
6024 		/* XXX make sure it's the same direction ?? */
6025 		pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED);
6026 		pf_unlink_state(*state);
6027 		*state = NULL;
6028 		return (PF_DROP);
6029 	}
6030 
6031 	if ((*state)->state_flags & PFSTATE_SLOPPY) {
6032 		if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP)
6033 			return (PF_DROP);
6034 	} else {
6035 		if (pf_tcp_track_full(state, kif, m, off, pd, reason,
6036 		    &copyback) == PF_DROP)
6037 			return (PF_DROP);
6038 	}
6039 
6040 	/* translate source/destination address, if necessary */
6041 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
6042 		struct pf_state_key *nk = (*state)->key[pd->didx];
6043 
6044 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
6045 		    nk->port[pd->sidx] != th->th_sport)
6046 			pf_change_ap(m, pd->src, &th->th_sport,
6047 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
6048 			    nk->port[pd->sidx], 0, pd->af);
6049 
6050 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
6051 		    nk->port[pd->didx] != th->th_dport)
6052 			pf_change_ap(m, pd->dst, &th->th_dport,
6053 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
6054 			    nk->port[pd->didx], 0, pd->af);
6055 		copyback = 1;
6056 	}
6057 
6058 	/* Copyback sequence modulation or stateful scrub changes if needed */
6059 	if (copyback)
6060 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
6061 
6062 	return (PF_PASS);
6063 }
6064 
6065 static int
pf_test_state_udp(struct pf_kstate ** state,struct pfi_kkif * kif,struct mbuf * m,int off,void * h,struct pf_pdesc * pd)6066 pf_test_state_udp(struct pf_kstate **state, struct pfi_kkif *kif,
6067     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
6068 {
6069 	struct pf_state_peer	*src, *dst;
6070 	struct pf_state_key_cmp	 key;
6071 	struct udphdr		*uh = &pd->hdr.udp;
6072 	uint8_t			 psrc, pdst;
6073 
6074 	bzero(&key, sizeof(key));
6075 	key.af = pd->af;
6076 	key.proto = IPPROTO_UDP;
6077 	if (pd->dir == PF_IN)	{	/* wire side, straight */
6078 		PF_ACPY(&key.addr[0], pd->src, key.af);
6079 		PF_ACPY(&key.addr[1], pd->dst, key.af);
6080 		key.port[0] = uh->uh_sport;
6081 		key.port[1] = uh->uh_dport;
6082 	} else {			/* stack side, reverse */
6083 		PF_ACPY(&key.addr[1], pd->src, key.af);
6084 		PF_ACPY(&key.addr[0], pd->dst, key.af);
6085 		key.port[1] = uh->uh_sport;
6086 		key.port[0] = uh->uh_dport;
6087 	}
6088 
6089 	STATE_LOOKUP(kif, &key, *state, pd);
6090 
6091 	if (pd->dir == (*state)->direction) {
6092 		src = &(*state)->src;
6093 		dst = &(*state)->dst;
6094 		psrc = PF_PEER_SRC;
6095 		pdst = PF_PEER_DST;
6096 	} else {
6097 		src = &(*state)->dst;
6098 		dst = &(*state)->src;
6099 		psrc = PF_PEER_DST;
6100 		pdst = PF_PEER_SRC;
6101 	}
6102 
6103 	/* update states */
6104 	if (src->state < PFUDPS_SINGLE)
6105 		pf_set_protostate(*state, psrc, PFUDPS_SINGLE);
6106 	if (dst->state == PFUDPS_SINGLE)
6107 		pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE);
6108 
6109 	/* update expire time */
6110 	(*state)->expire = time_uptime;
6111 	if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
6112 		(*state)->timeout = PFTM_UDP_MULTIPLE;
6113 	else
6114 		(*state)->timeout = PFTM_UDP_SINGLE;
6115 
6116 	/* translate source/destination address, if necessary */
6117 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
6118 		struct pf_state_key *nk = (*state)->key[pd->didx];
6119 
6120 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
6121 		    nk->port[pd->sidx] != uh->uh_sport)
6122 			pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
6123 			    &uh->uh_sum, &nk->addr[pd->sidx],
6124 			    nk->port[pd->sidx], 1, pd->af);
6125 
6126 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
6127 		    nk->port[pd->didx] != uh->uh_dport)
6128 			pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
6129 			    &uh->uh_sum, &nk->addr[pd->didx],
6130 			    nk->port[pd->didx], 1, pd->af);
6131 		m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
6132 	}
6133 
6134 	return (PF_PASS);
6135 }
6136 
6137 static int
pf_test_state_sctp(struct pf_kstate ** state,struct pfi_kkif * kif,struct mbuf * m,int off,void * h,struct pf_pdesc * pd,u_short * reason)6138 pf_test_state_sctp(struct pf_kstate **state, struct pfi_kkif *kif,
6139     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
6140 {
6141 	struct pf_state_key_cmp	 key;
6142 	struct pf_state_peer	*src, *dst;
6143 	struct sctphdr		*sh = &pd->hdr.sctp;
6144 	u_int8_t		 psrc; //, pdst;
6145 
6146 	bzero(&key, sizeof(key));
6147 	key.af = pd->af;
6148 	key.proto = IPPROTO_SCTP;
6149 	if (pd->dir == PF_IN)	{	/* wire side, straight */
6150 		PF_ACPY(&key.addr[0], pd->src, key.af);
6151 		PF_ACPY(&key.addr[1], pd->dst, key.af);
6152 		key.port[0] = sh->src_port;
6153 		key.port[1] = sh->dest_port;
6154 	} else {			/* stack side, reverse */
6155 		PF_ACPY(&key.addr[1], pd->src, key.af);
6156 		PF_ACPY(&key.addr[0], pd->dst, key.af);
6157 		key.port[1] = sh->src_port;
6158 		key.port[0] = sh->dest_port;
6159 	}
6160 
6161 	STATE_LOOKUP(kif, &key, *state, pd);
6162 
6163 	if (pd->dir == (*state)->direction) {
6164 		src = &(*state)->src;
6165 		dst = &(*state)->dst;
6166 		psrc = PF_PEER_SRC;
6167 	} else {
6168 		src = &(*state)->dst;
6169 		dst = &(*state)->src;
6170 		psrc = PF_PEER_DST;
6171 	}
6172 
6173 	if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) &&
6174 	    (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) &&
6175 	    pd->sctp_flags & PFDESC_SCTP_INIT) {
6176 		pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED);
6177 		pf_unlink_state(*state);
6178 		*state = NULL;
6179 		return (PF_DROP);
6180 	}
6181 
6182 	/* Track state. */
6183 	if (pd->sctp_flags & PFDESC_SCTP_INIT) {
6184 		if (src->state < SCTP_COOKIE_WAIT) {
6185 			pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT);
6186 			(*state)->timeout = PFTM_SCTP_OPENING;
6187 		}
6188 	}
6189 	if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) {
6190 		MPASS(dst->scrub != NULL);
6191 		if (dst->scrub->pfss_v_tag == 0)
6192 			dst->scrub->pfss_v_tag = pd->sctp_initiate_tag;
6193 	}
6194 
6195 	if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) {
6196 		if (src->state < SCTP_ESTABLISHED) {
6197 			pf_set_protostate(*state, psrc, SCTP_ESTABLISHED);
6198 			(*state)->timeout = PFTM_SCTP_ESTABLISHED;
6199 		}
6200 	}
6201 	if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT |
6202 	    PFDESC_SCTP_SHUTDOWN_COMPLETE)) {
6203 		if (src->state < SCTP_SHUTDOWN_PENDING) {
6204 			pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING);
6205 			(*state)->timeout = PFTM_SCTP_CLOSING;
6206 		}
6207 	}
6208 	if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) {
6209 		pf_set_protostate(*state, psrc, SCTP_CLOSED);
6210 		(*state)->timeout = PFTM_SCTP_CLOSED;
6211 	}
6212 
6213 	if (src->scrub != NULL) {
6214 		if (src->scrub->pfss_v_tag == 0) {
6215 			src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag;
6216 		} else  if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag)
6217 			return (PF_DROP);
6218 	}
6219 
6220 	(*state)->expire = time_uptime;
6221 
6222 	/* translate source/destination address, if necessary */
6223 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
6224 		uint16_t checksum = 0;
6225 		struct pf_state_key *nk = (*state)->key[pd->didx];
6226 
6227 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
6228 		    nk->port[pd->sidx] != pd->hdr.sctp.src_port) {
6229 			pf_change_ap(m, pd->src, &pd->hdr.sctp.src_port,
6230 			    pd->ip_sum, &checksum, &nk->addr[pd->sidx],
6231 			    nk->port[pd->sidx], 1, pd->af);
6232 		}
6233 
6234 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
6235 		    nk->port[pd->didx] != pd->hdr.sctp.dest_port) {
6236 			pf_change_ap(m, pd->dst, &pd->hdr.sctp.dest_port,
6237 			    pd->ip_sum, &checksum, &nk->addr[pd->didx],
6238 			    nk->port[pd->didx], 1, pd->af);
6239 		}
6240 	}
6241 
6242 	return (PF_PASS);
6243 }
6244 
6245 static void
pf_sctp_multihome_detach_addr(const struct pf_kstate * s)6246 pf_sctp_multihome_detach_addr(const struct pf_kstate *s)
6247 {
6248 	struct pf_sctp_endpoint key;
6249 	struct pf_sctp_endpoint *ep;
6250 	struct pf_state_key *sks = s->key[PF_SK_STACK];
6251 	struct pf_sctp_source *i, *tmp;
6252 
6253 	if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL)
6254 		return;
6255 
6256 	PF_SCTP_ENDPOINTS_LOCK();
6257 
6258 	key.v_tag = s->dst.scrub->pfss_v_tag;
6259 	ep  = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6260 	if (ep != NULL) {
6261 		TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) {
6262 			if (pf_addr_cmp(&i->addr,
6263 			    &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT],
6264 			    s->key[PF_SK_WIRE]->af) == 0) {
6265 				SDT_PROBE3(pf, sctp, multihome, remove,
6266 				    key.v_tag, s, i);
6267 				TAILQ_REMOVE(&ep->sources, i, entry);
6268 				free(i, M_PFTEMP);
6269 				break;
6270 			}
6271 		}
6272 
6273 		if (TAILQ_EMPTY(&ep->sources)) {
6274 			RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
6275 			free(ep, M_PFTEMP);
6276 		}
6277 	}
6278 
6279 	/* Other direction. */
6280 	key.v_tag = s->src.scrub->pfss_v_tag;
6281 	ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6282 	if (ep != NULL) {
6283 		TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) {
6284 			if (pf_addr_cmp(&i->addr,
6285 			    &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN],
6286 			    s->key[PF_SK_WIRE]->af) == 0) {
6287 				SDT_PROBE3(pf, sctp, multihome, remove,
6288 				    key.v_tag, s, i);
6289 				TAILQ_REMOVE(&ep->sources, i, entry);
6290 				free(i, M_PFTEMP);
6291 				break;
6292 			}
6293 		}
6294 
6295 		if (TAILQ_EMPTY(&ep->sources)) {
6296 			RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
6297 			free(ep, M_PFTEMP);
6298 		}
6299 	}
6300 
6301 	PF_SCTP_ENDPOINTS_UNLOCK();
6302 }
6303 
6304 static void
pf_sctp_multihome_add_addr(struct pf_pdesc * pd,struct pf_addr * a,uint32_t v_tag)6305 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag)
6306 {
6307 	struct pf_sctp_endpoint key = {
6308 		.v_tag = v_tag,
6309 	};
6310 	struct pf_sctp_source *i;
6311 	struct pf_sctp_endpoint *ep;
6312 
6313 	PF_SCTP_ENDPOINTS_LOCK();
6314 
6315 	ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6316 	if (ep == NULL) {
6317 		ep = malloc(sizeof(struct pf_sctp_endpoint),
6318 		    M_PFTEMP, M_NOWAIT);
6319 		if (ep == NULL) {
6320 			PF_SCTP_ENDPOINTS_UNLOCK();
6321 			return;
6322 		}
6323 
6324 		ep->v_tag = v_tag;
6325 		TAILQ_INIT(&ep->sources);
6326 		RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
6327 	}
6328 
6329 	/* Avoid inserting duplicates. */
6330 	TAILQ_FOREACH(i, &ep->sources, entry) {
6331 		if (pf_addr_cmp(&i->addr, a, pd->af) == 0) {
6332 			PF_SCTP_ENDPOINTS_UNLOCK();
6333 			return;
6334 		}
6335 	}
6336 
6337 	i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT);
6338 	if (i == NULL) {
6339 		PF_SCTP_ENDPOINTS_UNLOCK();
6340 		return;
6341 	}
6342 
6343 	i->af = pd->af;
6344 	memcpy(&i->addr, a, sizeof(*a));
6345 	TAILQ_INSERT_TAIL(&ep->sources, i, entry);
6346 	SDT_PROBE2(pf, sctp, multihome, add, v_tag, i);
6347 
6348 	PF_SCTP_ENDPOINTS_UNLOCK();
6349 }
6350 
6351 static void
pf_sctp_multihome_delayed(struct pf_pdesc * pd,int off,struct pfi_kkif * kif,struct pf_kstate * s,int action)6352 pf_sctp_multihome_delayed(struct pf_pdesc *pd, int off, struct pfi_kkif *kif,
6353     struct pf_kstate *s, int action)
6354 {
6355 	struct pf_sctp_multihome_job	*j, *tmp;
6356 	struct pf_sctp_source		*i;
6357 	int			 ret __unused;;
6358 	struct pf_kstate	*sm = NULL;
6359 	struct pf_krule		*ra = NULL;
6360 	struct pf_krule		*r = &V_pf_default_rule;
6361 	struct pf_kruleset	*rs = NULL;
6362 	bool do_extra = true;
6363 
6364 	PF_RULES_RLOCK_TRACKER;
6365 
6366 again:
6367 	TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) {
6368 		if (s == NULL || action != PF_PASS)
6369 			goto free;
6370 
6371 		/* Confirm we don't recurse here. */
6372 		MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP));
6373 
6374 		switch (j->op) {
6375 		case  SCTP_ADD_IP_ADDRESS: {
6376 			uint32_t v_tag = pd->sctp_initiate_tag;
6377 
6378 			if (v_tag == 0) {
6379 				if (s->direction == pd->dir)
6380 					v_tag = s->src.scrub->pfss_v_tag;
6381 				else
6382 					v_tag = s->dst.scrub->pfss_v_tag;
6383 			}
6384 
6385 			/*
6386 			 * Avoid duplicating states. We'll already have
6387 			 * created a state based on the source address of
6388 			 * the packet, but SCTP endpoints may also list this
6389 			 * address again in the INIT(_ACK) parameters.
6390 			 */
6391 			if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) {
6392 				break;
6393 			}
6394 
6395 			j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP;
6396 			PF_RULES_RLOCK();
6397 			sm = NULL;
6398 			/*
6399 			 * New connections need to be floating, because
6400 			 * we cannot know what interfaces it will use.
6401 			 * That's why we pass V_pfi_all rather than kif.
6402 			 */
6403 			ret = pf_test_rule(&r, &sm, V_pfi_all,
6404 			    j->m, off, &j->pd, &ra, &rs, NULL);
6405 			PF_RULES_RUNLOCK();
6406 			SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->m, ret);
6407 			if (ret != PF_DROP && sm != NULL) {
6408 				/* Inherit v_tag values. */
6409 				if (sm->direction == s->direction) {
6410 					sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag;
6411 					sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag;
6412 				} else {
6413 					sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag;
6414 					sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag;
6415 				}
6416 				PF_STATE_UNLOCK(sm);
6417 			} else {
6418 				/* If we try duplicate inserts? */
6419 				break;
6420 			}
6421 
6422 			/* Only add the address if we've actually allowed the state. */
6423 			pf_sctp_multihome_add_addr(pd, &j->src, v_tag);
6424 
6425 			if (! do_extra) {
6426 				break;
6427 			}
6428 			/*
6429 			 * We need to do this for each of our source addresses.
6430 			 * Find those based on the verification tag.
6431 			 */
6432 			struct pf_sctp_endpoint key = {
6433 				.v_tag = pd->hdr.sctp.v_tag,
6434 			};
6435 			struct pf_sctp_endpoint *ep;
6436 
6437 			PF_SCTP_ENDPOINTS_LOCK();
6438 			ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6439 			if (ep == NULL) {
6440 				PF_SCTP_ENDPOINTS_UNLOCK();
6441 				break;
6442 			}
6443 			MPASS(ep != NULL);
6444 
6445 			TAILQ_FOREACH(i, &ep->sources, entry) {
6446 				struct pf_sctp_multihome_job *nj;
6447 
6448 				/* SCTP can intermingle IPv4 and IPv6. */
6449 				if (i->af != pd->af)
6450 					continue;
6451 
6452 				nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO);
6453 				if (! nj) {
6454 					continue;
6455 				}
6456 				memcpy(&nj->pd, &j->pd, sizeof(j->pd));
6457 				memcpy(&nj->src, &j->src, sizeof(nj->src));
6458 				nj->pd.src = &nj->src;
6459 				// New destination address!
6460 				memcpy(&nj->dst, &i->addr, sizeof(nj->dst));
6461 				nj->pd.dst = &nj->dst;
6462 				nj->m = j->m;
6463 				nj->op = j->op;
6464 
6465 				TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next);
6466 			}
6467 			PF_SCTP_ENDPOINTS_UNLOCK();
6468 
6469 			break;
6470 		}
6471 		case SCTP_DEL_IP_ADDRESS: {
6472 			struct pf_state_key_cmp key;
6473 			uint8_t psrc;
6474 
6475 			bzero(&key, sizeof(key));
6476 			key.af = j->pd.af;
6477 			key.proto = IPPROTO_SCTP;
6478 			if (j->pd.dir == PF_IN)	{	/* wire side, straight */
6479 				PF_ACPY(&key.addr[0], j->pd.src, key.af);
6480 				PF_ACPY(&key.addr[1], j->pd.dst, key.af);
6481 				key.port[0] = j->pd.hdr.sctp.src_port;
6482 				key.port[1] = j->pd.hdr.sctp.dest_port;
6483 			} else {			/* stack side, reverse */
6484 				PF_ACPY(&key.addr[1], j->pd.src, key.af);
6485 				PF_ACPY(&key.addr[0], j->pd.dst, key.af);
6486 				key.port[1] = j->pd.hdr.sctp.src_port;
6487 				key.port[0] = j->pd.hdr.sctp.dest_port;
6488 			}
6489 
6490 			sm = pf_find_state(kif, &key, j->pd.dir);
6491 			if (sm != NULL) {
6492 				PF_STATE_LOCK_ASSERT(sm);
6493 				if (j->pd.dir == sm->direction) {
6494 					psrc = PF_PEER_SRC;
6495 				} else {
6496 					psrc = PF_PEER_DST;
6497 				}
6498 				pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING);
6499 				sm->timeout = PFTM_SCTP_CLOSING;
6500 				PF_STATE_UNLOCK(sm);
6501 			}
6502 			break;
6503 		default:
6504 			panic("Unknown op %#x", j->op);
6505 		}
6506 	}
6507 
6508 	free:
6509 		TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next);
6510 		free(j, M_PFTEMP);
6511 	}
6512 
6513 	/* We may have inserted extra work while processing the list. */
6514 	if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) {
6515 		do_extra = false;
6516 		goto again;
6517 	}
6518 }
6519 
6520 static int
pf_multihome_scan(struct mbuf * m,int start,int len,struct pf_pdesc * pd,struct pfi_kkif * kif,int op)6521 pf_multihome_scan(struct mbuf *m, int start, int len, struct pf_pdesc *pd,
6522     struct pfi_kkif *kif, int op)
6523 {
6524 	int			 off = 0;
6525 	struct pf_sctp_multihome_job	*job;
6526 
6527 	while (off < len) {
6528 		struct sctp_paramhdr h;
6529 
6530 		if (!pf_pull_hdr(m, start + off, &h, sizeof(h), NULL, NULL,
6531 		    pd->af))
6532 			return (PF_DROP);
6533 
6534 		/* Parameters are at least 4 bytes. */
6535 		if (ntohs(h.param_length) < 4)
6536 			return (PF_DROP);
6537 
6538 		switch (ntohs(h.param_type)) {
6539 		case  SCTP_IPV4_ADDRESS: {
6540 			struct in_addr t;
6541 
6542 			if (ntohs(h.param_length) !=
6543 			    (sizeof(struct sctp_paramhdr) + sizeof(t)))
6544 				return (PF_DROP);
6545 
6546 			if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t),
6547 			    NULL, NULL, pd->af))
6548 				return (PF_DROP);
6549 
6550 			if (in_nullhost(t))
6551 				t.s_addr = pd->src->v4.s_addr;
6552 
6553 			/*
6554 			 * We hold the state lock (idhash) here, which means
6555 			 * that we can't acquire the keyhash, or we'll get a
6556 			 * LOR (and potentially double-lock things too). We also
6557 			 * can't release the state lock here, so instead we'll
6558 			 * enqueue this for async handling.
6559 			 * There's a relatively small race here, in that a
6560 			 * packet using the new addresses could arrive already,
6561 			 * but that's just though luck for it.
6562 			 */
6563 			job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO);
6564 			if (! job)
6565 				return (PF_DROP);
6566 
6567 			memcpy(&job->pd, pd, sizeof(*pd));
6568 
6569 			// New source address!
6570 			memcpy(&job->src, &t, sizeof(t));
6571 			job->pd.src = &job->src;
6572 			memcpy(&job->dst, pd->dst, sizeof(job->dst));
6573 			job->pd.dst = &job->dst;
6574 			job->m = m;
6575 			job->op = op;
6576 
6577 			TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next);
6578 			break;
6579 		}
6580 #ifdef INET6
6581 		case SCTP_IPV6_ADDRESS: {
6582 			struct in6_addr t;
6583 
6584 			if (ntohs(h.param_length) !=
6585 			    (sizeof(struct sctp_paramhdr) + sizeof(t)))
6586 				return (PF_DROP);
6587 
6588 			if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t),
6589 			    NULL, NULL, pd->af))
6590 				return (PF_DROP);
6591 			if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0)
6592 				break;
6593 			if (memcmp(&t, &in6addr_any, sizeof(t)) == 0)
6594 				memcpy(&t, &pd->src->v6, sizeof(t));
6595 
6596 			job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO);
6597 			if (! job)
6598 				return (PF_DROP);
6599 
6600 			memcpy(&job->pd, pd, sizeof(*pd));
6601 			memcpy(&job->src, &t, sizeof(t));
6602 			job->pd.src = &job->src;
6603 			memcpy(&job->dst, pd->dst, sizeof(job->dst));
6604 			job->pd.dst = &job->dst;
6605 			job->m = m;
6606 			job->op = op;
6607 
6608 			TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next);
6609 			break;
6610 		}
6611 #endif
6612 		case SCTP_ADD_IP_ADDRESS: {
6613 			int ret;
6614 			struct sctp_asconf_paramhdr ah;
6615 
6616 			if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah),
6617 			    NULL, NULL, pd->af))
6618 				return (PF_DROP);
6619 
6620 			ret = pf_multihome_scan(m, start + off + sizeof(ah),
6621 			    ntohs(ah.ph.param_length) - sizeof(ah), pd, kif,
6622 			    SCTP_ADD_IP_ADDRESS);
6623 			if (ret != PF_PASS)
6624 				return (ret);
6625 			break;
6626 		}
6627 		case SCTP_DEL_IP_ADDRESS: {
6628 			int ret;
6629 			struct sctp_asconf_paramhdr ah;
6630 
6631 			if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah),
6632 			    NULL, NULL, pd->af))
6633 				return (PF_DROP);
6634 			ret = pf_multihome_scan(m, start + off + sizeof(ah),
6635 			    ntohs(ah.ph.param_length) - sizeof(ah), pd, kif,
6636 			    SCTP_DEL_IP_ADDRESS);
6637 			if (ret != PF_PASS)
6638 				return (ret);
6639 			break;
6640 		}
6641 		default:
6642 			break;
6643 		}
6644 
6645 		off += roundup(ntohs(h.param_length), 4);
6646 	}
6647 
6648 	return (PF_PASS);
6649 }
6650 int
pf_multihome_scan_init(struct mbuf * m,int start,int len,struct pf_pdesc * pd,struct pfi_kkif * kif)6651 pf_multihome_scan_init(struct mbuf *m, int start, int len, struct pf_pdesc *pd,
6652     struct pfi_kkif *kif)
6653 {
6654 	start += sizeof(struct sctp_init_chunk);
6655 	len -= sizeof(struct sctp_init_chunk);
6656 
6657 	return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS));
6658 }
6659 
6660 int
pf_multihome_scan_asconf(struct mbuf * m,int start,int len,struct pf_pdesc * pd,struct pfi_kkif * kif)6661 pf_multihome_scan_asconf(struct mbuf *m, int start, int len,
6662     struct pf_pdesc *pd, struct pfi_kkif *kif)
6663 {
6664 	start += sizeof(struct sctp_asconf_chunk);
6665 	len -= sizeof(struct sctp_asconf_chunk);
6666 
6667 	return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS));
6668 }
6669 
6670 int
pf_icmp_state_lookup(struct pf_state_key_cmp * key,struct pf_pdesc * pd,struct pf_kstate ** state,struct mbuf * m,int off,int direction,struct pfi_kkif * kif,u_int16_t icmpid,u_int16_t type,int icmp_dir,int * iidx,int multi,int inner)6671 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd,
6672     struct pf_kstate **state, struct mbuf *m, int off, int direction,
6673     struct pfi_kkif *kif, u_int16_t icmpid, u_int16_t type, int icmp_dir,
6674     int *iidx, int multi, int inner)
6675 {
6676 	key->af = pd->af;
6677 	key->proto = pd->proto;
6678 	if (icmp_dir == PF_IN) {
6679 		*iidx = pd->sidx;
6680 		key->port[pd->sidx] = icmpid;
6681 		key->port[pd->didx] = type;
6682 	} else {
6683 		*iidx = pd->didx;
6684 		key->port[pd->sidx] = type;
6685 		key->port[pd->didx] = icmpid;
6686 	}
6687 	if (pf_state_key_addr_setup(pd, m, off, key, pd->sidx, pd->src,
6688 	    pd->didx, pd->dst, multi))
6689 		return (PF_DROP);
6690 
6691 	STATE_LOOKUP(kif, key, *state, pd);
6692 
6693 	if ((*state)->state_flags & PFSTATE_SLOPPY)
6694 		return (-1);
6695 
6696 	/* Is this ICMP message flowing in right direction? */
6697 	if ((*state)->rule.ptr->type &&
6698 	    (((!inner && (*state)->direction == direction) ||
6699 	    (inner && (*state)->direction != direction)) ?
6700 	    PF_IN : PF_OUT) != icmp_dir) {
6701 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
6702 			printf("pf: icmp type %d in wrong direction (%d): ",
6703 			    ntohs(type), icmp_dir);
6704 			pf_print_state(*state);
6705 			printf("\n");
6706 		}
6707 		PF_STATE_UNLOCK(*state);
6708 		*state = NULL;
6709 		return (PF_DROP);
6710 	}
6711 	return (-1);
6712 }
6713 
6714 static int
pf_test_state_icmp(struct pf_kstate ** state,struct pfi_kkif * kif,struct mbuf * m,int off,void * h,struct pf_pdesc * pd,u_short * reason)6715 pf_test_state_icmp(struct pf_kstate **state, struct pfi_kkif *kif,
6716     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
6717 {
6718 	struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
6719 	u_int16_t	*icmpsum, virtual_id, virtual_type;
6720 	u_int8_t	 icmptype, icmpcode;
6721 	int		 icmp_dir, iidx, ret, multi;
6722 	struct pf_state_key_cmp key;
6723 #ifdef INET
6724 	u_int16_t	 icmpid;
6725 #endif
6726 
6727 	MPASS(*state == NULL);
6728 
6729 	bzero(&key, sizeof(key));
6730 	switch (pd->proto) {
6731 #ifdef INET
6732 	case IPPROTO_ICMP:
6733 		icmptype = pd->hdr.icmp.icmp_type;
6734 		icmpcode = pd->hdr.icmp.icmp_code;
6735 		icmpid = pd->hdr.icmp.icmp_id;
6736 		icmpsum = &pd->hdr.icmp.icmp_cksum;
6737 		break;
6738 #endif /* INET */
6739 #ifdef INET6
6740 	case IPPROTO_ICMPV6:
6741 		icmptype = pd->hdr.icmp6.icmp6_type;
6742 		icmpcode = pd->hdr.icmp6.icmp6_code;
6743 #ifdef INET
6744 		icmpid = pd->hdr.icmp6.icmp6_id;
6745 #endif
6746 		icmpsum = &pd->hdr.icmp6.icmp6_cksum;
6747 		break;
6748 #endif /* INET6 */
6749 	}
6750 
6751 	if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &multi,
6752 	    &virtual_id, &virtual_type) == 0) {
6753 		/*
6754 		 * ICMP query/reply message not related to a TCP/UDP packet.
6755 		 * Search for an ICMP state.
6756 		 */
6757 		ret = pf_icmp_state_lookup(&key, pd, state, m, off, pd->dir,
6758 		    kif, virtual_id, virtual_type, icmp_dir, &iidx,
6759 		    PF_ICMP_MULTI_NONE, 0);
6760 		if (ret >= 0) {
6761 			MPASS(*state == NULL);
6762 			if (ret == PF_DROP && pd->af == AF_INET6 &&
6763 			    icmp_dir == PF_OUT) {
6764 				ret = pf_icmp_state_lookup(&key, pd, state, m, off,
6765 				    pd->dir, kif, virtual_id, virtual_type,
6766 				    icmp_dir, &iidx, multi, 0);
6767 				if (ret >= 0) {
6768 					MPASS(*state == NULL);
6769 					return (ret);
6770 				}
6771 			} else
6772 				return (ret);
6773 		}
6774 
6775 		(*state)->expire = time_uptime;
6776 		(*state)->timeout = PFTM_ICMP_ERROR_REPLY;
6777 
6778 		/* translate source/destination address, if necessary */
6779 		if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
6780 			struct pf_state_key *nk = (*state)->key[pd->didx];
6781 
6782 			switch (pd->af) {
6783 #ifdef INET
6784 			case AF_INET:
6785 				if (PF_ANEQ(pd->src,
6786 				    &nk->addr[pd->sidx], AF_INET))
6787 					pf_change_a(&saddr->v4.s_addr,
6788 					    pd->ip_sum,
6789 					    nk->addr[pd->sidx].v4.s_addr, 0);
6790 
6791 				if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
6792 				    AF_INET))
6793 					pf_change_a(&daddr->v4.s_addr,
6794 					    pd->ip_sum,
6795 					    nk->addr[pd->didx].v4.s_addr, 0);
6796 
6797 				if (nk->port[iidx] !=
6798 				    pd->hdr.icmp.icmp_id) {
6799 					pd->hdr.icmp.icmp_cksum =
6800 					    pf_cksum_fixup(
6801 					    pd->hdr.icmp.icmp_cksum, icmpid,
6802 					    nk->port[iidx], 0);
6803 					pd->hdr.icmp.icmp_id =
6804 					    nk->port[iidx];
6805 				}
6806 
6807 				m_copyback(m, off, ICMP_MINLEN,
6808 				    (caddr_t )&pd->hdr.icmp);
6809 				break;
6810 #endif /* INET */
6811 #ifdef INET6
6812 			case AF_INET6:
6813 				if (PF_ANEQ(pd->src,
6814 				    &nk->addr[pd->sidx], AF_INET6))
6815 					pf_change_a6(saddr,
6816 					    &pd->hdr.icmp6.icmp6_cksum,
6817 					    &nk->addr[pd->sidx], 0);
6818 
6819 				if (PF_ANEQ(pd->dst,
6820 				    &nk->addr[pd->didx], AF_INET6))
6821 					pf_change_a6(daddr,
6822 					    &pd->hdr.icmp6.icmp6_cksum,
6823 					    &nk->addr[pd->didx], 0);
6824 
6825 				m_copyback(m, off, sizeof(struct icmp6_hdr),
6826 				    (caddr_t )&pd->hdr.icmp6);
6827 				break;
6828 #endif /* INET6 */
6829 			}
6830 		}
6831 		return (PF_PASS);
6832 
6833 	} else {
6834 		/*
6835 		 * ICMP error message in response to a TCP/UDP packet.
6836 		 * Extract the inner TCP/UDP header and search for that state.
6837 		 */
6838 
6839 		struct pf_pdesc	pd2;
6840 		bzero(&pd2, sizeof pd2);
6841 #ifdef INET
6842 		struct ip	h2;
6843 #endif /* INET */
6844 #ifdef INET6
6845 		struct ip6_hdr	h2_6;
6846 		int		terminal = 0;
6847 #endif /* INET6 */
6848 		int		ipoff2 = 0;
6849 		int		off2 = 0;
6850 
6851 		pd2.af = pd->af;
6852 		pd2.dir = pd->dir;
6853 		/* Payload packet is from the opposite direction. */
6854 		pd2.sidx = (pd->dir == PF_IN) ? 1 : 0;
6855 		pd2.didx = (pd->dir == PF_IN) ? 0 : 1;
6856 		switch (pd->af) {
6857 #ifdef INET
6858 		case AF_INET:
6859 			/* offset of h2 in mbuf chain */
6860 			ipoff2 = off + ICMP_MINLEN;
6861 
6862 			if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
6863 			    NULL, reason, pd2.af)) {
6864 				DPFPRINTF(PF_DEBUG_MISC,
6865 				    ("pf: ICMP error message too short "
6866 				    "(ip)\n"));
6867 				return (PF_DROP);
6868 			}
6869 			/*
6870 			 * ICMP error messages don't refer to non-first
6871 			 * fragments
6872 			 */
6873 			if (h2.ip_off & htons(IP_OFFMASK)) {
6874 				REASON_SET(reason, PFRES_FRAG);
6875 				return (PF_DROP);
6876 			}
6877 
6878 			/* offset of protocol header that follows h2 */
6879 			off2 = ipoff2 + (h2.ip_hl << 2);
6880 
6881 			pd2.proto = h2.ip_p;
6882 			pd2.src = (struct pf_addr *)&h2.ip_src;
6883 			pd2.dst = (struct pf_addr *)&h2.ip_dst;
6884 			pd2.ip_sum = &h2.ip_sum;
6885 			break;
6886 #endif /* INET */
6887 #ifdef INET6
6888 		case AF_INET6:
6889 			ipoff2 = off + sizeof(struct icmp6_hdr);
6890 
6891 			if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
6892 			    NULL, reason, pd2.af)) {
6893 				DPFPRINTF(PF_DEBUG_MISC,
6894 				    ("pf: ICMP error message too short "
6895 				    "(ip6)\n"));
6896 				return (PF_DROP);
6897 			}
6898 			pd2.proto = h2_6.ip6_nxt;
6899 			pd2.src = (struct pf_addr *)&h2_6.ip6_src;
6900 			pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
6901 			pd2.ip_sum = NULL;
6902 			off2 = ipoff2 + sizeof(h2_6);
6903 			do {
6904 				switch (pd2.proto) {
6905 				case IPPROTO_FRAGMENT:
6906 					/*
6907 					 * ICMPv6 error messages for
6908 					 * non-first fragments
6909 					 */
6910 					REASON_SET(reason, PFRES_FRAG);
6911 					return (PF_DROP);
6912 				case IPPROTO_AH:
6913 				case IPPROTO_HOPOPTS:
6914 				case IPPROTO_ROUTING:
6915 				case IPPROTO_DSTOPTS: {
6916 					/* get next header and header length */
6917 					struct ip6_ext opt6;
6918 
6919 					if (!pf_pull_hdr(m, off2, &opt6,
6920 					    sizeof(opt6), NULL, reason,
6921 					    pd2.af)) {
6922 						DPFPRINTF(PF_DEBUG_MISC,
6923 						    ("pf: ICMPv6 short opt\n"));
6924 						return (PF_DROP);
6925 					}
6926 					if (pd2.proto == IPPROTO_AH)
6927 						off2 += (opt6.ip6e_len + 2) * 4;
6928 					else
6929 						off2 += (opt6.ip6e_len + 1) * 8;
6930 					pd2.proto = opt6.ip6e_nxt;
6931 					/* goto the next header */
6932 					break;
6933 				}
6934 				default:
6935 					terminal++;
6936 					break;
6937 				}
6938 			} while (!terminal);
6939 			break;
6940 #endif /* INET6 */
6941 		}
6942 
6943 		if (PF_ANEQ(pd->dst, pd2.src, pd->af)) {
6944 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
6945 				printf("pf: BAD ICMP %d:%d outer dst: ",
6946 				    icmptype, icmpcode);
6947 				pf_print_host(pd->src, 0, pd->af);
6948 				printf(" -> ");
6949 				pf_print_host(pd->dst, 0, pd->af);
6950 				printf(" inner src: ");
6951 				pf_print_host(pd2.src, 0, pd2.af);
6952 				printf(" -> ");
6953 				pf_print_host(pd2.dst, 0, pd2.af);
6954 				printf("\n");
6955 			}
6956 			REASON_SET(reason, PFRES_BADSTATE);
6957 			return (PF_DROP);
6958 		}
6959 
6960 		switch (pd2.proto) {
6961 		case IPPROTO_TCP: {
6962 			struct tcphdr		 th;
6963 			u_int32_t		 seq;
6964 			struct pf_state_peer	*src, *dst;
6965 			u_int8_t		 dws;
6966 			int			 copyback = 0;
6967 
6968 			/*
6969 			 * Only the first 8 bytes of the TCP header can be
6970 			 * expected. Don't access any TCP header fields after
6971 			 * th_seq, an ackskew test is not possible.
6972 			 */
6973 			if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
6974 			    pd2.af)) {
6975 				DPFPRINTF(PF_DEBUG_MISC,
6976 				    ("pf: ICMP error message too short "
6977 				    "(tcp)\n"));
6978 				return (PF_DROP);
6979 			}
6980 
6981 			key.af = pd2.af;
6982 			key.proto = IPPROTO_TCP;
6983 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
6984 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
6985 			key.port[pd2.sidx] = th.th_sport;
6986 			key.port[pd2.didx] = th.th_dport;
6987 
6988 			STATE_LOOKUP(kif, &key, *state, pd);
6989 
6990 			if (pd->dir == (*state)->direction) {
6991 				src = &(*state)->dst;
6992 				dst = &(*state)->src;
6993 			} else {
6994 				src = &(*state)->src;
6995 				dst = &(*state)->dst;
6996 			}
6997 
6998 			if (src->wscale && dst->wscale)
6999 				dws = dst->wscale & PF_WSCALE_MASK;
7000 			else
7001 				dws = 0;
7002 
7003 			/* Demodulate sequence number */
7004 			seq = ntohl(th.th_seq) - src->seqdiff;
7005 			if (src->seqdiff) {
7006 				pf_change_a(&th.th_seq, icmpsum,
7007 				    htonl(seq), 0);
7008 				copyback = 1;
7009 			}
7010 
7011 			if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
7012 			    (!SEQ_GEQ(src->seqhi, seq) ||
7013 			    !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
7014 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
7015 					printf("pf: BAD ICMP %d:%d ",
7016 					    icmptype, icmpcode);
7017 					pf_print_host(pd->src, 0, pd->af);
7018 					printf(" -> ");
7019 					pf_print_host(pd->dst, 0, pd->af);
7020 					printf(" state: ");
7021 					pf_print_state(*state);
7022 					printf(" seq=%u\n", seq);
7023 				}
7024 				REASON_SET(reason, PFRES_BADSTATE);
7025 				return (PF_DROP);
7026 			} else {
7027 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
7028 					printf("pf: OK ICMP %d:%d ",
7029 					    icmptype, icmpcode);
7030 					pf_print_host(pd->src, 0, pd->af);
7031 					printf(" -> ");
7032 					pf_print_host(pd->dst, 0, pd->af);
7033 					printf(" state: ");
7034 					pf_print_state(*state);
7035 					printf(" seq=%u\n", seq);
7036 				}
7037 			}
7038 
7039 			/* translate source/destination address, if necessary */
7040 			if ((*state)->key[PF_SK_WIRE] !=
7041 			    (*state)->key[PF_SK_STACK]) {
7042 				struct pf_state_key *nk =
7043 				    (*state)->key[pd->didx];
7044 
7045 				if (PF_ANEQ(pd2.src,
7046 				    &nk->addr[pd2.sidx], pd2.af) ||
7047 				    nk->port[pd2.sidx] != th.th_sport)
7048 					pf_change_icmp(pd2.src, &th.th_sport,
7049 					    daddr, &nk->addr[pd2.sidx],
7050 					    nk->port[pd2.sidx], NULL,
7051 					    pd2.ip_sum, icmpsum,
7052 					    pd->ip_sum, 0, pd2.af);
7053 
7054 				if (PF_ANEQ(pd2.dst,
7055 				    &nk->addr[pd2.didx], pd2.af) ||
7056 				    nk->port[pd2.didx] != th.th_dport)
7057 					pf_change_icmp(pd2.dst, &th.th_dport,
7058 					    saddr, &nk->addr[pd2.didx],
7059 					    nk->port[pd2.didx], NULL,
7060 					    pd2.ip_sum, icmpsum,
7061 					    pd->ip_sum, 0, pd2.af);
7062 				copyback = 1;
7063 			}
7064 
7065 			if (copyback) {
7066 				switch (pd2.af) {
7067 #ifdef INET
7068 				case AF_INET:
7069 					m_copyback(m, off, ICMP_MINLEN,
7070 					    (caddr_t )&pd->hdr.icmp);
7071 					m_copyback(m, ipoff2, sizeof(h2),
7072 					    (caddr_t )&h2);
7073 					break;
7074 #endif /* INET */
7075 #ifdef INET6
7076 				case AF_INET6:
7077 					m_copyback(m, off,
7078 					    sizeof(struct icmp6_hdr),
7079 					    (caddr_t )&pd->hdr.icmp6);
7080 					m_copyback(m, ipoff2, sizeof(h2_6),
7081 					    (caddr_t )&h2_6);
7082 					break;
7083 #endif /* INET6 */
7084 				}
7085 				m_copyback(m, off2, 8, (caddr_t)&th);
7086 			}
7087 
7088 			return (PF_PASS);
7089 			break;
7090 		}
7091 		case IPPROTO_UDP: {
7092 			struct udphdr		uh;
7093 
7094 			if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
7095 			    NULL, reason, pd2.af)) {
7096 				DPFPRINTF(PF_DEBUG_MISC,
7097 				    ("pf: ICMP error message too short "
7098 				    "(udp)\n"));
7099 				return (PF_DROP);
7100 			}
7101 
7102 			key.af = pd2.af;
7103 			key.proto = IPPROTO_UDP;
7104 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
7105 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
7106 			key.port[pd2.sidx] = uh.uh_sport;
7107 			key.port[pd2.didx] = uh.uh_dport;
7108 
7109 			STATE_LOOKUP(kif, &key, *state, pd);
7110 
7111 			/* translate source/destination address, if necessary */
7112 			if ((*state)->key[PF_SK_WIRE] !=
7113 			    (*state)->key[PF_SK_STACK]) {
7114 				struct pf_state_key *nk =
7115 				    (*state)->key[pd->didx];
7116 
7117 				if (PF_ANEQ(pd2.src,
7118 				    &nk->addr[pd2.sidx], pd2.af) ||
7119 				    nk->port[pd2.sidx] != uh.uh_sport)
7120 					pf_change_icmp(pd2.src, &uh.uh_sport,
7121 					    daddr, &nk->addr[pd2.sidx],
7122 					    nk->port[pd2.sidx], &uh.uh_sum,
7123 					    pd2.ip_sum, icmpsum,
7124 					    pd->ip_sum, 1, pd2.af);
7125 
7126 				if (PF_ANEQ(pd2.dst,
7127 				    &nk->addr[pd2.didx], pd2.af) ||
7128 				    nk->port[pd2.didx] != uh.uh_dport)
7129 					pf_change_icmp(pd2.dst, &uh.uh_dport,
7130 					    saddr, &nk->addr[pd2.didx],
7131 					    nk->port[pd2.didx], &uh.uh_sum,
7132 					    pd2.ip_sum, icmpsum,
7133 					    pd->ip_sum, 1, pd2.af);
7134 
7135 				switch (pd2.af) {
7136 #ifdef INET
7137 				case AF_INET:
7138 					m_copyback(m, off, ICMP_MINLEN,
7139 					    (caddr_t )&pd->hdr.icmp);
7140 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
7141 					break;
7142 #endif /* INET */
7143 #ifdef INET6
7144 				case AF_INET6:
7145 					m_copyback(m, off,
7146 					    sizeof(struct icmp6_hdr),
7147 					    (caddr_t )&pd->hdr.icmp6);
7148 					m_copyback(m, ipoff2, sizeof(h2_6),
7149 					    (caddr_t )&h2_6);
7150 					break;
7151 #endif /* INET6 */
7152 				}
7153 				m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
7154 			}
7155 			return (PF_PASS);
7156 			break;
7157 		}
7158 #ifdef INET
7159 		case IPPROTO_ICMP: {
7160 			struct icmp	*iih = &pd2.hdr.icmp;
7161 
7162 			if (!pf_pull_hdr(m, off2, iih, ICMP_MINLEN,
7163 			    NULL, reason, pd2.af)) {
7164 				DPFPRINTF(PF_DEBUG_MISC,
7165 				    ("pf: ICMP error message too short i"
7166 				    "(icmp)\n"));
7167 				return (PF_DROP);
7168 			}
7169 
7170 			icmpid = iih->icmp_id;
7171 			pf_icmp_mapping(&pd2, iih->icmp_type,
7172 			    &icmp_dir, &multi, &virtual_id, &virtual_type);
7173 
7174 			ret = pf_icmp_state_lookup(&key, &pd2, state, m, off,
7175 			    pd2.dir, kif, virtual_id, virtual_type,
7176 			    icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1);
7177 			if (ret >= 0) {
7178 				MPASS(*state == NULL);
7179 				return (ret);
7180 			}
7181 
7182 			/* translate source/destination address, if necessary */
7183 			if ((*state)->key[PF_SK_WIRE] !=
7184 			    (*state)->key[PF_SK_STACK]) {
7185 				struct pf_state_key *nk =
7186 				    (*state)->key[pd->didx];
7187 
7188 				if (PF_ANEQ(pd2.src,
7189 				    &nk->addr[pd2.sidx], pd2.af) ||
7190 				    (virtual_type == htons(ICMP_ECHO) &&
7191 				    nk->port[iidx] != iih->icmp_id))
7192 					pf_change_icmp(pd2.src,
7193 					    (virtual_type == htons(ICMP_ECHO)) ?
7194 					    &iih->icmp_id : NULL,
7195 					    daddr, &nk->addr[pd2.sidx],
7196 					    (virtual_type == htons(ICMP_ECHO)) ?
7197 					    nk->port[iidx] : 0, NULL,
7198 					    pd2.ip_sum, icmpsum,
7199 					    pd->ip_sum, 0, AF_INET);
7200 
7201 				if (PF_ANEQ(pd2.dst,
7202 				    &nk->addr[pd2.didx], pd2.af))
7203 					pf_change_icmp(pd2.dst, NULL, NULL,
7204 					    &nk->addr[pd2.didx], 0, NULL,
7205 					    pd2.ip_sum, icmpsum, pd->ip_sum, 0,
7206 					    AF_INET);
7207 
7208 				m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
7209 				m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
7210 				m_copyback(m, off2, ICMP_MINLEN, (caddr_t)iih);
7211 			}
7212 			return (PF_PASS);
7213 			break;
7214 		}
7215 #endif /* INET */
7216 #ifdef INET6
7217 		case IPPROTO_ICMPV6: {
7218 			struct icmp6_hdr	*iih = &pd2.hdr.icmp6;
7219 
7220 			if (!pf_pull_hdr(m, off2, iih,
7221 			    sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
7222 				DPFPRINTF(PF_DEBUG_MISC,
7223 				    ("pf: ICMP error message too short "
7224 				    "(icmp6)\n"));
7225 				return (PF_DROP);
7226 			}
7227 
7228 			pf_icmp_mapping(&pd2, iih->icmp6_type,
7229 			    &icmp_dir, &multi, &virtual_id, &virtual_type);
7230 
7231 			ret = pf_icmp_state_lookup(&key, &pd2, state, m, off,
7232 			    pd->dir, kif, virtual_id, virtual_type,
7233 			    icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1);
7234 			if (ret >= 0) {
7235 				MPASS(*state == NULL);
7236 				if (ret == PF_DROP && pd2.af == AF_INET6 &&
7237 				    icmp_dir == PF_OUT) {
7238 					ret = pf_icmp_state_lookup(&key, &pd2,
7239 					    state, m, off, pd->dir, kif,
7240 					    virtual_id, virtual_type,
7241 					    icmp_dir, &iidx, multi, 1);
7242 					if (ret >= 0) {
7243 						MPASS(*state == NULL);
7244 						return (ret);
7245 					}
7246 				} else
7247 					return (ret);
7248 			}
7249 
7250 			/* translate source/destination address, if necessary */
7251 			if ((*state)->key[PF_SK_WIRE] !=
7252 			    (*state)->key[PF_SK_STACK]) {
7253 				struct pf_state_key *nk =
7254 				    (*state)->key[pd->didx];
7255 
7256 				if (PF_ANEQ(pd2.src,
7257 				    &nk->addr[pd2.sidx], pd2.af) ||
7258 				    ((virtual_type == htons(ICMP6_ECHO_REQUEST)) &&
7259 				    nk->port[pd2.sidx] != iih->icmp6_id))
7260 					pf_change_icmp(pd2.src,
7261 					    (virtual_type == htons(ICMP6_ECHO_REQUEST))
7262 					    ? &iih->icmp6_id : NULL,
7263 					    daddr, &nk->addr[pd2.sidx],
7264 					    (virtual_type == htons(ICMP6_ECHO_REQUEST))
7265 					    ? nk->port[iidx] : 0, NULL,
7266 					    pd2.ip_sum, icmpsum,
7267 					    pd->ip_sum, 0, AF_INET6);
7268 
7269 				if (PF_ANEQ(pd2.dst,
7270 				    &nk->addr[pd2.didx], pd2.af))
7271 					pf_change_icmp(pd2.dst, NULL, NULL,
7272 					    &nk->addr[pd2.didx], 0, NULL,
7273 					    pd2.ip_sum, icmpsum,
7274 					    pd->ip_sum, 0, AF_INET6);
7275 
7276 				m_copyback(m, off, sizeof(struct icmp6_hdr),
7277 				    (caddr_t)&pd->hdr.icmp6);
7278 				m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
7279 				m_copyback(m, off2, sizeof(struct icmp6_hdr),
7280 				    (caddr_t)iih);
7281 			}
7282 			return (PF_PASS);
7283 			break;
7284 		}
7285 #endif /* INET6 */
7286 		default: {
7287 			key.af = pd2.af;
7288 			key.proto = pd2.proto;
7289 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
7290 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
7291 			key.port[0] = key.port[1] = 0;
7292 
7293 			STATE_LOOKUP(kif, &key, *state, pd);
7294 
7295 			/* translate source/destination address, if necessary */
7296 			if ((*state)->key[PF_SK_WIRE] !=
7297 			    (*state)->key[PF_SK_STACK]) {
7298 				struct pf_state_key *nk =
7299 				    (*state)->key[pd->didx];
7300 
7301 				if (PF_ANEQ(pd2.src,
7302 				    &nk->addr[pd2.sidx], pd2.af))
7303 					pf_change_icmp(pd2.src, NULL, daddr,
7304 					    &nk->addr[pd2.sidx], 0, NULL,
7305 					    pd2.ip_sum, icmpsum,
7306 					    pd->ip_sum, 0, pd2.af);
7307 
7308 				if (PF_ANEQ(pd2.dst,
7309 				    &nk->addr[pd2.didx], pd2.af))
7310 					pf_change_icmp(pd2.dst, NULL, saddr,
7311 					    &nk->addr[pd2.didx], 0, NULL,
7312 					    pd2.ip_sum, icmpsum,
7313 					    pd->ip_sum, 0, pd2.af);
7314 
7315 				switch (pd2.af) {
7316 #ifdef INET
7317 				case AF_INET:
7318 					m_copyback(m, off, ICMP_MINLEN,
7319 					    (caddr_t)&pd->hdr.icmp);
7320 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
7321 					break;
7322 #endif /* INET */
7323 #ifdef INET6
7324 				case AF_INET6:
7325 					m_copyback(m, off,
7326 					    sizeof(struct icmp6_hdr),
7327 					    (caddr_t )&pd->hdr.icmp6);
7328 					m_copyback(m, ipoff2, sizeof(h2_6),
7329 					    (caddr_t )&h2_6);
7330 					break;
7331 #endif /* INET6 */
7332 				}
7333 			}
7334 			return (PF_PASS);
7335 			break;
7336 		}
7337 		}
7338 	}
7339 }
7340 
7341 static int
pf_test_state_other(struct pf_kstate ** state,struct pfi_kkif * kif,struct mbuf * m,struct pf_pdesc * pd)7342 pf_test_state_other(struct pf_kstate **state, struct pfi_kkif *kif,
7343     struct mbuf *m, struct pf_pdesc *pd)
7344 {
7345 	struct pf_state_peer	*src, *dst;
7346 	struct pf_state_key_cmp	 key;
7347 	uint8_t			 psrc, pdst;
7348 
7349 	bzero(&key, sizeof(key));
7350 	key.af = pd->af;
7351 	key.proto = pd->proto;
7352 	if (pd->dir == PF_IN)	{
7353 		PF_ACPY(&key.addr[0], pd->src, key.af);
7354 		PF_ACPY(&key.addr[1], pd->dst, key.af);
7355 		key.port[0] = key.port[1] = 0;
7356 	} else {
7357 		PF_ACPY(&key.addr[1], pd->src, key.af);
7358 		PF_ACPY(&key.addr[0], pd->dst, key.af);
7359 		key.port[1] = key.port[0] = 0;
7360 	}
7361 
7362 	STATE_LOOKUP(kif, &key, *state, pd);
7363 
7364 	if (pd->dir == (*state)->direction) {
7365 		src = &(*state)->src;
7366 		dst = &(*state)->dst;
7367 		psrc = PF_PEER_SRC;
7368 		pdst = PF_PEER_DST;
7369 	} else {
7370 		src = &(*state)->dst;
7371 		dst = &(*state)->src;
7372 		psrc = PF_PEER_DST;
7373 		pdst = PF_PEER_SRC;
7374 	}
7375 
7376 	/* update states */
7377 	if (src->state < PFOTHERS_SINGLE)
7378 		pf_set_protostate(*state, psrc, PFOTHERS_SINGLE);
7379 	if (dst->state == PFOTHERS_SINGLE)
7380 		pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE);
7381 
7382 	/* update expire time */
7383 	(*state)->expire = time_uptime;
7384 	if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
7385 		(*state)->timeout = PFTM_OTHER_MULTIPLE;
7386 	else
7387 		(*state)->timeout = PFTM_OTHER_SINGLE;
7388 
7389 	/* translate source/destination address, if necessary */
7390 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
7391 		struct pf_state_key *nk = (*state)->key[pd->didx];
7392 
7393 		KASSERT(nk, ("%s: nk is null", __func__));
7394 		KASSERT(pd, ("%s: pd is null", __func__));
7395 		KASSERT(pd->src, ("%s: pd->src is null", __func__));
7396 		KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
7397 		switch (pd->af) {
7398 #ifdef INET
7399 		case AF_INET:
7400 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
7401 				pf_change_a(&pd->src->v4.s_addr,
7402 				    pd->ip_sum,
7403 				    nk->addr[pd->sidx].v4.s_addr,
7404 				    0);
7405 
7406 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
7407 				pf_change_a(&pd->dst->v4.s_addr,
7408 				    pd->ip_sum,
7409 				    nk->addr[pd->didx].v4.s_addr,
7410 				    0);
7411 
7412 			break;
7413 #endif /* INET */
7414 #ifdef INET6
7415 		case AF_INET6:
7416 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET6))
7417 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
7418 
7419 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET6))
7420 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
7421 #endif /* INET6 */
7422 		}
7423 	}
7424 	return (PF_PASS);
7425 }
7426 
7427 /*
7428  * ipoff and off are measured from the start of the mbuf chain.
7429  * h must be at "ipoff" on the mbuf chain.
7430  */
7431 void *
pf_pull_hdr(struct mbuf * m,int off,void * p,int len,u_short * actionp,u_short * reasonp,sa_family_t af)7432 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
7433     u_short *actionp, u_short *reasonp, sa_family_t af)
7434 {
7435 	switch (af) {
7436 #ifdef INET
7437 	case AF_INET: {
7438 		struct ip	*h = mtod(m, struct ip *);
7439 		u_int16_t	 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
7440 
7441 		if (fragoff) {
7442 			if (fragoff >= len)
7443 				ACTION_SET(actionp, PF_PASS);
7444 			else {
7445 				ACTION_SET(actionp, PF_DROP);
7446 				REASON_SET(reasonp, PFRES_FRAG);
7447 			}
7448 			return (NULL);
7449 		}
7450 		if (m->m_pkthdr.len < off + len ||
7451 		    ntohs(h->ip_len) < off + len) {
7452 			ACTION_SET(actionp, PF_DROP);
7453 			REASON_SET(reasonp, PFRES_SHORT);
7454 			return (NULL);
7455 		}
7456 		break;
7457 	}
7458 #endif /* INET */
7459 #ifdef INET6
7460 	case AF_INET6: {
7461 		struct ip6_hdr	*h = mtod(m, struct ip6_hdr *);
7462 
7463 		if (m->m_pkthdr.len < off + len ||
7464 		    (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
7465 		    (unsigned)(off + len)) {
7466 			ACTION_SET(actionp, PF_DROP);
7467 			REASON_SET(reasonp, PFRES_SHORT);
7468 			return (NULL);
7469 		}
7470 		break;
7471 	}
7472 #endif /* INET6 */
7473 	}
7474 	m_copydata(m, off, len, p);
7475 	return (p);
7476 }
7477 
7478 int
pf_routable(struct pf_addr * addr,sa_family_t af,struct pfi_kkif * kif,int rtableid)7479 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif,
7480     int rtableid)
7481 {
7482 	struct ifnet		*ifp;
7483 
7484 	/*
7485 	 * Skip check for addresses with embedded interface scope,
7486 	 * as they would always match anyway.
7487 	 */
7488 	if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
7489 		return (1);
7490 
7491 	if (af != AF_INET && af != AF_INET6)
7492 		return (0);
7493 
7494 	if (kif == V_pfi_all)
7495 		return (1);
7496 
7497 	/* Skip checks for ipsec interfaces */
7498 	if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
7499 		return (1);
7500 
7501 	ifp = (kif != NULL) ? kif->pfik_ifp : NULL;
7502 
7503 	switch (af) {
7504 #ifdef INET6
7505 	case AF_INET6:
7506 		return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE,
7507 		    ifp));
7508 #endif
7509 #ifdef INET
7510 	case AF_INET:
7511 		return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE,
7512 		    ifp));
7513 #endif
7514 	}
7515 
7516 	return (0);
7517 }
7518 
7519 #ifdef INET
7520 static void
pf_route(struct mbuf ** m,struct pf_krule * r,struct ifnet * oifp,struct pf_kstate * s,struct pf_pdesc * pd,struct inpcb * inp)7521 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp,
7522     struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp)
7523 {
7524 	struct mbuf		*m0, *m1, *md;
7525 	struct sockaddr_in	dst;
7526 	struct ip		*ip;
7527 	struct pfi_kkif		*nkif = NULL;
7528 	struct ifnet		*ifp = NULL;
7529 	struct pf_addr		 naddr;
7530 	struct pf_ksrc_node	*sn = NULL;
7531 	int			 error = 0;
7532 	uint16_t		 ip_len, ip_off;
7533 	int			 r_rt, r_dir;
7534 
7535 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
7536 
7537 	if (s) {
7538 		r_rt = s->rt;
7539 		r_dir = s->direction;
7540 	} else {
7541 		r_rt = r->rt;
7542 		r_dir = r->direction;
7543 	}
7544 
7545 	KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT ||
7546 	    r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction",
7547 	    __func__));
7548 
7549 	if ((pd->pf_mtag == NULL &&
7550 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
7551 	    pd->pf_mtag->routed++ > 3) {
7552 		m0 = *m;
7553 		*m = NULL;
7554 		goto bad_locked;
7555 	}
7556 
7557 	if (r_rt == PF_DUPTO) {
7558 		if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) {
7559 			if (s == NULL) {
7560 				ifp = r->rpool.cur->kif ?
7561 				    r->rpool.cur->kif->pfik_ifp : NULL;
7562 			} else {
7563 				ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7564 				/* If pfsync'd */
7565 				if (ifp == NULL && r->rpool.cur != NULL)
7566 					ifp = r->rpool.cur->kif ?
7567 					    r->rpool.cur->kif->pfik_ifp : NULL;
7568 				PF_STATE_UNLOCK(s);
7569 			}
7570 			if (ifp == oifp) {
7571 				/* When the 2nd interface is not skipped */
7572 				return;
7573 			} else {
7574 				m0 = *m;
7575 				*m = NULL;
7576 				goto bad;
7577 			}
7578 		} else {
7579 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED;
7580 			if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) {
7581 				if (s)
7582 					PF_STATE_UNLOCK(s);
7583 				return;
7584 			}
7585 		}
7586 	} else {
7587 		if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) {
7588 			pf_dummynet(pd, s, r, m);
7589 			if (s)
7590 				PF_STATE_UNLOCK(s);
7591 			return;
7592 		}
7593 		m0 = *m;
7594 	}
7595 
7596 	ip = mtod(m0, struct ip *);
7597 
7598 	bzero(&dst, sizeof(dst));
7599 	dst.sin_family = AF_INET;
7600 	dst.sin_len = sizeof(dst);
7601 	dst.sin_addr = ip->ip_dst;
7602 
7603 	bzero(&naddr, sizeof(naddr));
7604 
7605 	if (s == NULL) {
7606 		if (TAILQ_EMPTY(&r->rpool.list)) {
7607 			DPFPRINTF(PF_DEBUG_URGENT,
7608 			    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
7609 			goto bad_locked;
7610 		}
7611 		pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
7612 		    &naddr, &nkif, NULL, &sn);
7613 		if (!PF_AZERO(&naddr, AF_INET))
7614 			dst.sin_addr.s_addr = naddr.v4.s_addr;
7615 		ifp = nkif ? nkif->pfik_ifp : NULL;
7616 	} else {
7617 		if (!PF_AZERO(&s->rt_addr, AF_INET))
7618 			dst.sin_addr.s_addr =
7619 			    s->rt_addr.v4.s_addr;
7620 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7621 		/* If pfsync'd */
7622 		if (ifp == NULL && r->rpool.cur != NULL) {
7623 			ifp = r->rpool.cur->kif ?
7624 			    r->rpool.cur->kif->pfik_ifp : NULL;
7625 		}
7626 		PF_STATE_UNLOCK(s);
7627 	}
7628 
7629 	if (ifp == NULL)
7630 		goto bad;
7631 
7632 	if (pd->dir == PF_IN) {
7633 		if (pf_test(PF_OUT, 0, ifp, &m0, inp, &pd->act) != PF_PASS)
7634 			goto bad;
7635 		else if (m0 == NULL)
7636 			goto done;
7637 		if (m0->m_len < sizeof(struct ip)) {
7638 			DPFPRINTF(PF_DEBUG_URGENT,
7639 			    ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
7640 			goto bad;
7641 		}
7642 		ip = mtod(m0, struct ip *);
7643 	}
7644 
7645 	if (ifp->if_flags & IFF_LOOPBACK)
7646 		m0->m_flags |= M_SKIP_FIREWALL;
7647 
7648 	ip_len = ntohs(ip->ip_len);
7649 	ip_off = ntohs(ip->ip_off);
7650 
7651 	/* Copied from FreeBSD 10.0-CURRENT ip_output. */
7652 	m0->m_pkthdr.csum_flags |= CSUM_IP;
7653 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
7654 		in_delayed_cksum(m0);
7655 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
7656 	}
7657 	if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
7658 		pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2));
7659 		m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
7660 	}
7661 
7662 	/*
7663 	 * If small enough for interface, or the interface will take
7664 	 * care of the fragmentation for us, we can just send directly.
7665 	 */
7666 	if (ip_len <= ifp->if_mtu ||
7667 	    (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
7668 		ip->ip_sum = 0;
7669 		if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
7670 			ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
7671 			m0->m_pkthdr.csum_flags &= ~CSUM_IP;
7672 		}
7673 		m_clrprotoflags(m0);	/* Avoid confusing lower layers. */
7674 
7675 		md = m0;
7676 		error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md);
7677 		if (md != NULL)
7678 			error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL);
7679 		goto done;
7680 	}
7681 
7682 	/* Balk when DF bit is set or the interface didn't support TSO. */
7683 	if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
7684 		error = EMSGSIZE;
7685 		KMOD_IPSTAT_INC(ips_cantfrag);
7686 		if (r_rt != PF_DUPTO) {
7687 			if (s && pd->nat_rule != NULL)
7688 				PACKET_UNDO_NAT(m0, pd,
7689 				    (ip->ip_hl << 2) + (ip_off & IP_OFFMASK),
7690 				    s);
7691 
7692 			icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
7693 			    ifp->if_mtu);
7694 			goto done;
7695 		} else
7696 			goto bad;
7697 	}
7698 
7699 	error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
7700 	if (error)
7701 		goto bad;
7702 
7703 	for (; m0; m0 = m1) {
7704 		m1 = m0->m_nextpkt;
7705 		m0->m_nextpkt = NULL;
7706 		if (error == 0) {
7707 			m_clrprotoflags(m0);
7708 			md = m0;
7709 			pd->pf_mtag = pf_find_mtag(md);
7710 			error = pf_dummynet_route(pd, s, r, ifp,
7711 			    sintosa(&dst), &md);
7712 			if (md != NULL)
7713 				error = (*ifp->if_output)(ifp, md,
7714 				    sintosa(&dst), NULL);
7715 		} else
7716 			m_freem(m0);
7717 	}
7718 
7719 	if (error == 0)
7720 		KMOD_IPSTAT_INC(ips_fragmented);
7721 
7722 done:
7723 	if (r_rt != PF_DUPTO)
7724 		*m = NULL;
7725 	return;
7726 
7727 bad_locked:
7728 	if (s)
7729 		PF_STATE_UNLOCK(s);
7730 bad:
7731 	m_freem(m0);
7732 	goto done;
7733 }
7734 #endif /* INET */
7735 
7736 #ifdef INET6
7737 static void
pf_route6(struct mbuf ** m,struct pf_krule * r,struct ifnet * oifp,struct pf_kstate * s,struct pf_pdesc * pd,struct inpcb * inp)7738 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp,
7739     struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp)
7740 {
7741 	struct mbuf		*m0, *md;
7742 	struct sockaddr_in6	dst;
7743 	struct ip6_hdr		*ip6;
7744 	struct pfi_kkif		*nkif = NULL;
7745 	struct ifnet		*ifp = NULL;
7746 	struct pf_addr		 naddr;
7747 	struct pf_ksrc_node	*sn = NULL;
7748 	int			 r_rt, r_dir;
7749 
7750 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
7751 
7752 	if (s) {
7753 		r_rt = s->rt;
7754 		r_dir = s->direction;
7755 	} else {
7756 		r_rt = r->rt;
7757 		r_dir = r->direction;
7758 	}
7759 
7760 	KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT ||
7761 	    r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction",
7762 	    __func__));
7763 
7764 	if ((pd->pf_mtag == NULL &&
7765 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
7766 	    pd->pf_mtag->routed++ > 3) {
7767 		m0 = *m;
7768 		*m = NULL;
7769 		goto bad_locked;
7770 	}
7771 
7772 	if (r_rt == PF_DUPTO) {
7773 		if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) {
7774 			if (s == NULL) {
7775 				ifp = r->rpool.cur->kif ?
7776 				    r->rpool.cur->kif->pfik_ifp : NULL;
7777 			} else {
7778 				ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7779 				/* If pfsync'd */
7780 				if (ifp == NULL && r->rpool.cur != NULL)
7781 					ifp = r->rpool.cur->kif ?
7782 					    r->rpool.cur->kif->pfik_ifp : NULL;
7783 				PF_STATE_UNLOCK(s);
7784 			}
7785 			if (ifp == oifp) {
7786 				/* When the 2nd interface is not skipped */
7787 				return;
7788 			} else {
7789 				m0 = *m;
7790 				*m = NULL;
7791 				goto bad;
7792 			}
7793 		} else {
7794 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED;
7795 			if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) {
7796 				if (s)
7797 					PF_STATE_UNLOCK(s);
7798 				return;
7799 			}
7800 		}
7801 	} else {
7802 		if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) {
7803 			pf_dummynet(pd, s, r, m);
7804 			if (s)
7805 				PF_STATE_UNLOCK(s);
7806 			return;
7807 		}
7808 		m0 = *m;
7809 	}
7810 
7811 	ip6 = mtod(m0, struct ip6_hdr *);
7812 
7813 	bzero(&dst, sizeof(dst));
7814 	dst.sin6_family = AF_INET6;
7815 	dst.sin6_len = sizeof(dst);
7816 	dst.sin6_addr = ip6->ip6_dst;
7817 
7818 	bzero(&naddr, sizeof(naddr));
7819 
7820 	if (s == NULL) {
7821 		if (TAILQ_EMPTY(&r->rpool.list)) {
7822 			DPFPRINTF(PF_DEBUG_URGENT,
7823 			    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
7824 			goto bad_locked;
7825 		}
7826 		pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
7827 		    &naddr, &nkif, NULL, &sn);
7828 		if (!PF_AZERO(&naddr, AF_INET6))
7829 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
7830 			    &naddr, AF_INET6);
7831 		ifp = nkif ? nkif->pfik_ifp : NULL;
7832 	} else {
7833 		if (!PF_AZERO(&s->rt_addr, AF_INET6))
7834 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
7835 			    &s->rt_addr, AF_INET6);
7836 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7837 		/* If pfsync'd */
7838 		if (ifp == NULL && r->rpool.cur != NULL)
7839 			ifp = r->rpool.cur->kif ?
7840 			    r->rpool.cur->kif->pfik_ifp : NULL;
7841 	}
7842 
7843 	if (s)
7844 		PF_STATE_UNLOCK(s);
7845 
7846 	if (ifp == NULL)
7847 		goto bad;
7848 
7849 	if (pd->dir == PF_IN) {
7850 		if (pf_test6(PF_OUT, 0, ifp, &m0, inp, &pd->act) != PF_PASS)
7851 			goto bad;
7852 		else if (m0 == NULL)
7853 			goto done;
7854 		if (m0->m_len < sizeof(struct ip6_hdr)) {
7855 			DPFPRINTF(PF_DEBUG_URGENT,
7856 			    ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
7857 			    __func__));
7858 			goto bad;
7859 		}
7860 		ip6 = mtod(m0, struct ip6_hdr *);
7861 	}
7862 
7863 	if (ifp->if_flags & IFF_LOOPBACK)
7864 		m0->m_flags |= M_SKIP_FIREWALL;
7865 
7866 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
7867 	    ~ifp->if_hwassist) {
7868 		uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
7869 		in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
7870 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
7871 	}
7872 
7873 	/*
7874 	 * If the packet is too large for the outgoing interface,
7875 	 * send back an icmp6 error.
7876 	 */
7877 	if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
7878 		dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
7879 	if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) {
7880 		md = m0;
7881 		pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md);
7882 		if (md != NULL)
7883 			nd6_output_ifp(ifp, ifp, md, &dst, NULL);
7884 	}
7885 	else {
7886 		in6_ifstat_inc(ifp, ifs6_in_toobig);
7887 		if (r_rt != PF_DUPTO) {
7888 			if (s && pd->nat_rule != NULL)
7889 				PACKET_UNDO_NAT(m0, pd,
7890 				    ((caddr_t)ip6 - m0->m_data) +
7891 				    sizeof(struct ip6_hdr), s);
7892 
7893 			icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
7894 		} else
7895 			goto bad;
7896 	}
7897 
7898 done:
7899 	if (r_rt != PF_DUPTO)
7900 		*m = NULL;
7901 	return;
7902 
7903 bad_locked:
7904 	if (s)
7905 		PF_STATE_UNLOCK(s);
7906 bad:
7907 	m_freem(m0);
7908 	goto done;
7909 }
7910 #endif /* INET6 */
7911 
7912 /*
7913  * FreeBSD supports cksum offloads for the following drivers.
7914  *  em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4)
7915  *
7916  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
7917  *  network driver performed cksum including pseudo header, need to verify
7918  *   csum_data
7919  * CSUM_DATA_VALID :
7920  *  network driver performed cksum, needs to additional pseudo header
7921  *  cksum computation with partial csum_data(i.e. lack of H/W support for
7922  *  pseudo header, for instance sk(4) and possibly gem(4))
7923  *
7924  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
7925  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
7926  * TCP/UDP layer.
7927  * Also, set csum_data to 0xffff to force cksum validation.
7928  */
7929 static int
pf_check_proto_cksum(struct mbuf * m,int off,int len,u_int8_t p,sa_family_t af)7930 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
7931 {
7932 	u_int16_t sum = 0;
7933 	int hw_assist = 0;
7934 	struct ip *ip;
7935 
7936 	if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
7937 		return (1);
7938 	if (m->m_pkthdr.len < off + len)
7939 		return (1);
7940 
7941 	switch (p) {
7942 	case IPPROTO_TCP:
7943 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
7944 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
7945 				sum = m->m_pkthdr.csum_data;
7946 			} else {
7947 				ip = mtod(m, struct ip *);
7948 				sum = in_pseudo(ip->ip_src.s_addr,
7949 				ip->ip_dst.s_addr, htonl((u_short)len +
7950 				m->m_pkthdr.csum_data + IPPROTO_TCP));
7951 			}
7952 			sum ^= 0xffff;
7953 			++hw_assist;
7954 		}
7955 		break;
7956 	case IPPROTO_UDP:
7957 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
7958 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
7959 				sum = m->m_pkthdr.csum_data;
7960 			} else {
7961 				ip = mtod(m, struct ip *);
7962 				sum = in_pseudo(ip->ip_src.s_addr,
7963 				ip->ip_dst.s_addr, htonl((u_short)len +
7964 				m->m_pkthdr.csum_data + IPPROTO_UDP));
7965 			}
7966 			sum ^= 0xffff;
7967 			++hw_assist;
7968 		}
7969 		break;
7970 	case IPPROTO_ICMP:
7971 #ifdef INET6
7972 	case IPPROTO_ICMPV6:
7973 #endif /* INET6 */
7974 		break;
7975 	default:
7976 		return (1);
7977 	}
7978 
7979 	if (!hw_assist) {
7980 		switch (af) {
7981 		case AF_INET:
7982 			if (p == IPPROTO_ICMP) {
7983 				if (m->m_len < off)
7984 					return (1);
7985 				m->m_data += off;
7986 				m->m_len -= off;
7987 				sum = in_cksum(m, len);
7988 				m->m_data -= off;
7989 				m->m_len += off;
7990 			} else {
7991 				if (m->m_len < sizeof(struct ip))
7992 					return (1);
7993 				sum = in4_cksum(m, p, off, len);
7994 			}
7995 			break;
7996 #ifdef INET6
7997 		case AF_INET6:
7998 			if (m->m_len < sizeof(struct ip6_hdr))
7999 				return (1);
8000 			sum = in6_cksum(m, p, off, len);
8001 			break;
8002 #endif /* INET6 */
8003 		default:
8004 			return (1);
8005 		}
8006 	}
8007 	if (sum) {
8008 		switch (p) {
8009 		case IPPROTO_TCP:
8010 		    {
8011 			KMOD_TCPSTAT_INC(tcps_rcvbadsum);
8012 			break;
8013 		    }
8014 		case IPPROTO_UDP:
8015 		    {
8016 			KMOD_UDPSTAT_INC(udps_badsum);
8017 			break;
8018 		    }
8019 #ifdef INET
8020 		case IPPROTO_ICMP:
8021 		    {
8022 			KMOD_ICMPSTAT_INC(icps_checksum);
8023 			break;
8024 		    }
8025 #endif
8026 #ifdef INET6
8027 		case IPPROTO_ICMPV6:
8028 		    {
8029 			KMOD_ICMP6STAT_INC(icp6s_checksum);
8030 			break;
8031 		    }
8032 #endif /* INET6 */
8033 		}
8034 		return (1);
8035 	} else {
8036 		if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
8037 			m->m_pkthdr.csum_flags |=
8038 			    (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
8039 			m->m_pkthdr.csum_data = 0xffff;
8040 		}
8041 	}
8042 	return (0);
8043 }
8044 
8045 static bool
pf_pdesc_to_dnflow(const struct pf_pdesc * pd,const struct pf_krule * r,const struct pf_kstate * s,struct ip_fw_args * dnflow)8046 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r,
8047     const struct pf_kstate *s, struct ip_fw_args *dnflow)
8048 {
8049 	int dndir = r->direction;
8050 
8051 	if (s && dndir == PF_INOUT) {
8052 		dndir = s->direction;
8053 	} else if (dndir == PF_INOUT) {
8054 		/* Assume primary direction. Happens when we've set dnpipe in
8055 		 * the ethernet level code. */
8056 		dndir = pd->dir;
8057 	}
8058 
8059 	memset(dnflow, 0, sizeof(*dnflow));
8060 
8061 	if (pd->dport != NULL)
8062 		dnflow->f_id.dst_port = ntohs(*pd->dport);
8063 	if (pd->sport != NULL)
8064 		dnflow->f_id.src_port = ntohs(*pd->sport);
8065 
8066 	if (pd->dir == PF_IN)
8067 		dnflow->flags |= IPFW_ARGS_IN;
8068 	else
8069 		dnflow->flags |= IPFW_ARGS_OUT;
8070 
8071 	if (pd->dir != dndir && pd->act.dnrpipe) {
8072 		dnflow->rule.info = pd->act.dnrpipe;
8073 	}
8074 	else if (pd->dir == dndir && pd->act.dnpipe) {
8075 		dnflow->rule.info = pd->act.dnpipe;
8076 	}
8077 	else {
8078 		return (false);
8079 	}
8080 
8081 	dnflow->rule.info |= IPFW_IS_DUMMYNET;
8082 	if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE)
8083 		dnflow->rule.info |= IPFW_IS_PIPE;
8084 
8085 	dnflow->f_id.proto = pd->proto;
8086 	dnflow->f_id.extra = dnflow->rule.info;
8087 	switch (pd->af) {
8088 	case AF_INET:
8089 		dnflow->f_id.addr_type = 4;
8090 		dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr);
8091 		dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr);
8092 		break;
8093 	case AF_INET6:
8094 		dnflow->flags |= IPFW_ARGS_IP6;
8095 		dnflow->f_id.addr_type = 6;
8096 		dnflow->f_id.src_ip6 = pd->src->v6;
8097 		dnflow->f_id.dst_ip6 = pd->dst->v6;
8098 		break;
8099 	default:
8100 		panic("Invalid AF");
8101 		break;
8102 	}
8103 
8104 	return (true);
8105 }
8106 
8107 int
pf_test_eth(int dir,int pflags,struct ifnet * ifp,struct mbuf ** m0,struct inpcb * inp)8108 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0,
8109     struct inpcb *inp)
8110 {
8111 	struct pfi_kkif		*kif;
8112 	struct mbuf		*m = *m0;
8113 
8114 	M_ASSERTPKTHDR(m);
8115 	MPASS(ifp->if_vnet == curvnet);
8116 	NET_EPOCH_ASSERT();
8117 
8118 	if (!V_pf_status.running)
8119 		return (PF_PASS);
8120 
8121 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
8122 
8123 	if (kif == NULL) {
8124 		DPFPRINTF(PF_DEBUG_URGENT,
8125 		    ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname));
8126 		return (PF_DROP);
8127 	}
8128 	if (kif->pfik_flags & PFI_IFLAG_SKIP)
8129 		return (PF_PASS);
8130 
8131 	if (m->m_flags & M_SKIP_FIREWALL)
8132 		return (PF_PASS);
8133 
8134 	/* Stateless! */
8135 	return (pf_test_eth_rule(dir, kif, m0));
8136 }
8137 
8138 static __inline void
pf_dummynet_flag_remove(struct mbuf * m,struct pf_mtag * pf_mtag)8139 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag)
8140 {
8141 	struct m_tag *mtag;
8142 
8143 	pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET;
8144 
8145 	/* dummynet adds this tag, but pf does not need it,
8146 	 * and keeping it creates unexpected behavior,
8147 	 * e.g. in case of divert(4) usage right after dummynet. */
8148 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
8149 	if (mtag != NULL)
8150 		m_tag_delete(m, mtag);
8151 }
8152 
8153 static int
pf_dummynet(struct pf_pdesc * pd,struct pf_kstate * s,struct pf_krule * r,struct mbuf ** m0)8154 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s,
8155     struct pf_krule *r, struct mbuf **m0)
8156 {
8157 	return (pf_dummynet_route(pd, s, r, NULL, NULL, m0));
8158 }
8159 
8160 static int
pf_dummynet_route(struct pf_pdesc * pd,struct pf_kstate * s,struct pf_krule * r,struct ifnet * ifp,struct sockaddr * sa,struct mbuf ** m0)8161 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s,
8162     struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa,
8163     struct mbuf **m0)
8164 {
8165 	NET_EPOCH_ASSERT();
8166 
8167 	if (pd->act.dnpipe || pd->act.dnrpipe) {
8168 		struct ip_fw_args dnflow;
8169 		if (ip_dn_io_ptr == NULL) {
8170 			m_freem(*m0);
8171 			*m0 = NULL;
8172 			return (ENOMEM);
8173 		}
8174 
8175 		if (pd->pf_mtag == NULL &&
8176 		    ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) {
8177 			m_freem(*m0);
8178 			*m0 = NULL;
8179 			return (ENOMEM);
8180 		}
8181 
8182 		if (ifp != NULL) {
8183 			pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO;
8184 
8185 			pd->pf_mtag->if_index = ifp->if_index;
8186 			pd->pf_mtag->if_idxgen = ifp->if_idxgen;
8187 
8188 			MPASS(sa != NULL);
8189 
8190 			if (pd->af == AF_INET)
8191 				memcpy(&pd->pf_mtag->dst, sa,
8192 				    sizeof(struct sockaddr_in));
8193 			else
8194 				memcpy(&pd->pf_mtag->dst, sa,
8195 				    sizeof(struct sockaddr_in6));
8196 		}
8197 
8198 		if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) {
8199 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET;
8200 			ip_dn_io_ptr(m0, &dnflow);
8201 			if (*m0 != NULL) {
8202 				pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
8203 				pf_dummynet_flag_remove(*m0, pd->pf_mtag);
8204 			}
8205 		}
8206 	}
8207 
8208 	return (0);
8209 }
8210 
8211 #ifdef INET
8212 int
pf_test(int dir,int pflags,struct ifnet * ifp,struct mbuf ** m0,struct inpcb * inp,struct pf_rule_actions * default_actions)8213 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0,
8214     struct inpcb *inp, struct pf_rule_actions *default_actions)
8215 {
8216 	struct pfi_kkif		*kif;
8217 	u_short			 action, reason = 0;
8218 	struct mbuf		*m = *m0;
8219 	struct ip		*h = NULL;
8220 	struct m_tag		*mtag;
8221 	struct pf_krule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
8222 	struct pf_kstate	*s = NULL;
8223 	struct pf_kruleset	*ruleset = NULL;
8224 	struct pf_pdesc		 pd;
8225 	int			 off, dirndx, use_2nd_queue = 0;
8226 	uint16_t		 tag;
8227 	uint8_t			 rt;
8228 
8229 	PF_RULES_RLOCK_TRACKER;
8230 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir));
8231 	M_ASSERTPKTHDR(m);
8232 
8233 	if (!V_pf_status.running)
8234 		return (PF_PASS);
8235 
8236 	PF_RULES_RLOCK();
8237 
8238 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
8239 
8240 	if (__predict_false(kif == NULL)) {
8241 		DPFPRINTF(PF_DEBUG_URGENT,
8242 		    ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
8243 		PF_RULES_RUNLOCK();
8244 		return (PF_DROP);
8245 	}
8246 	if (kif->pfik_flags & PFI_IFLAG_SKIP) {
8247 		PF_RULES_RUNLOCK();
8248 		return (PF_PASS);
8249 	}
8250 
8251 	if (m->m_flags & M_SKIP_FIREWALL) {
8252 		PF_RULES_RUNLOCK();
8253 		return (PF_PASS);
8254 	}
8255 
8256 	memset(&pd, 0, sizeof(pd));
8257 	TAILQ_INIT(&pd.sctp_multihome_jobs);
8258 	if (default_actions != NULL)
8259 		memcpy(&pd.act, default_actions, sizeof(pd.act));
8260 	pd.pf_mtag = pf_find_mtag(m);
8261 
8262 	if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) {
8263 		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
8264 
8265 		ifp = ifnet_byindexgen(pd.pf_mtag->if_index,
8266 		    pd.pf_mtag->if_idxgen);
8267 		if (ifp == NULL || ifp->if_flags & IFF_DYING) {
8268 			PF_RULES_RUNLOCK();
8269 			m_freem(*m0);
8270 			*m0 = NULL;
8271 			return (PF_PASS);
8272 		}
8273 		PF_RULES_RUNLOCK();
8274 		(ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL);
8275 		*m0 = NULL;
8276 		return (PF_PASS);
8277 	}
8278 
8279 	if (pd.pf_mtag && pd.pf_mtag->dnpipe) {
8280 		pd.act.dnpipe = pd.pf_mtag->dnpipe;
8281 		pd.act.flags = pd.pf_mtag->dnflags;
8282 	}
8283 
8284 	if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL &&
8285 	    pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
8286 		/* Dummynet re-injects packets after they've
8287 		 * completed their delay. We've already
8288 		 * processed them, so pass unconditionally. */
8289 
8290 		/* But only once. We may see the packet multiple times (e.g.
8291 		 * PFIL_IN/PFIL_OUT). */
8292 		pf_dummynet_flag_remove(m, pd.pf_mtag);
8293 		PF_RULES_RUNLOCK();
8294 
8295 		return (PF_PASS);
8296 	}
8297 
8298 	pd.sport = pd.dport = NULL;
8299 	pd.proto_sum = NULL;
8300 	pd.dir = dir;
8301 	pd.sidx = (dir == PF_IN) ? 0 : 1;
8302 	pd.didx = (dir == PF_IN) ? 1 : 0;
8303 	pd.af = AF_INET;
8304 	pd.act.rtableid = -1;
8305 
8306 	h = mtod(m, struct ip *);
8307 	off = h->ip_hl << 2;
8308 
8309 	if (__predict_false(ip_divert_ptr != NULL) &&
8310 	    ((mtag = m_tag_locate(m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) {
8311 		struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1);
8312 		if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) ||
8313 		    (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) {
8314 			if (pd.pf_mtag == NULL &&
8315 			    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8316 				action = PF_DROP;
8317 				goto done;
8318 			}
8319 			pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED;
8320 		}
8321 		if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) {
8322 			m->m_flags |= M_FASTFWD_OURS;
8323 			pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT;
8324 		}
8325 		m_tag_delete(m, mtag);
8326 
8327 		mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
8328 		if (mtag != NULL)
8329 			m_tag_delete(m, mtag);
8330 	} else if (pf_normalize_ip(m0, kif, &reason, &pd) != PF_PASS) {
8331 		/* We do IP header normalization and packet reassembly here */
8332 		action = PF_DROP;
8333 		goto done;
8334 	}
8335 	m = *m0;	/* pf_normalize messes with m0 */
8336 	h = mtod(m, struct ip *);
8337 
8338 	off = h->ip_hl << 2;
8339 	if (off < (int)sizeof(struct ip)) {
8340 		action = PF_DROP;
8341 		REASON_SET(&reason, PFRES_SHORT);
8342 		pd.act.log = PF_LOG_FORCE;
8343 		goto done;
8344 	}
8345 
8346 	pd.src = (struct pf_addr *)&h->ip_src;
8347 	pd.dst = (struct pf_addr *)&h->ip_dst;
8348 	pd.ip_sum = &h->ip_sum;
8349 	pd.proto = h->ip_p;
8350 	pd.tos = h->ip_tos & ~IPTOS_ECN_MASK;
8351 	pd.tot_len = ntohs(h->ip_len);
8352 
8353 	/* handle fragments that didn't get reassembled by normalization */
8354 	if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
8355 		action = pf_test_fragment(&r, kif, m, h, &pd, &a, &ruleset);
8356 		goto done;
8357 	}
8358 
8359 	switch (h->ip_p) {
8360 	case IPPROTO_TCP: {
8361 		if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp),
8362 		    &action, &reason, AF_INET)) {
8363 			if (action != PF_PASS)
8364 				pd.act.log = PF_LOG_FORCE;
8365 			goto done;
8366 		}
8367 		pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2);
8368 
8369 		pd.sport = &pd.hdr.tcp.th_sport;
8370 		pd.dport = &pd.hdr.tcp.th_dport;
8371 
8372 		/* Respond to SYN with a syncookie. */
8373 		if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN &&
8374 		    pd.dir == PF_IN && pf_synflood_check(&pd)) {
8375 			pf_syncookie_send(m, off, &pd);
8376 			action = PF_DROP;
8377 			break;
8378 		}
8379 
8380 		if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0)
8381 			use_2nd_queue = 1;
8382 		action = pf_normalize_tcp(kif, m, 0, off, h, &pd);
8383 		if (action == PF_DROP)
8384 			goto done;
8385 		action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason);
8386 		if (action == PF_PASS) {
8387 			if (V_pfsync_update_state_ptr != NULL)
8388 				V_pfsync_update_state_ptr(s);
8389 			r = s->rule.ptr;
8390 			a = s->anchor.ptr;
8391 		} else if (s == NULL) {
8392 			/* Validate remote SYN|ACK, re-create original SYN if
8393 			 * valid. */
8394 			if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) ==
8395 			    TH_ACK && pf_syncookie_validate(&pd) &&
8396 			    pd.dir == PF_IN) {
8397 				struct mbuf *msyn;
8398 
8399 				msyn = pf_syncookie_recreate_syn(h->ip_ttl, off,
8400 				    &pd);
8401 				if (msyn == NULL) {
8402 					action = PF_DROP;
8403 					break;
8404 				}
8405 
8406 				action = pf_test(dir, pflags, ifp, &msyn, inp,
8407 				    &pd.act);
8408 				m_freem(msyn);
8409 				if (action != PF_PASS)
8410 					break;
8411 
8412 				action = pf_test_state_tcp(&s, kif, m, off, h,
8413 				    &pd, &reason);
8414 				if (action != PF_PASS || s == NULL) {
8415 					action = PF_DROP;
8416 					break;
8417 				}
8418 
8419 				s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1;
8420 				s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1;
8421 				pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST);
8422 				action = pf_synproxy(&pd, &s, &reason);
8423 				break;
8424 			} else {
8425 				action = pf_test_rule(&r, &s, kif, m, off, &pd,
8426 				    &a, &ruleset, inp);
8427 			}
8428 		}
8429 		break;
8430 	}
8431 
8432 	case IPPROTO_UDP: {
8433 		if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp),
8434 		    &action, &reason, AF_INET)) {
8435 			if (action != PF_PASS)
8436 				pd.act.log = PF_LOG_FORCE;
8437 			goto done;
8438 		}
8439 		pd.sport = &pd.hdr.udp.uh_sport;
8440 		pd.dport = &pd.hdr.udp.uh_dport;
8441 		if (pd.hdr.udp.uh_dport == 0 ||
8442 		    ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off ||
8443 		    ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) {
8444 			action = PF_DROP;
8445 			REASON_SET(&reason, PFRES_SHORT);
8446 			goto done;
8447 		}
8448 		action = pf_test_state_udp(&s, kif, m, off, h, &pd);
8449 		if (action == PF_PASS) {
8450 			if (V_pfsync_update_state_ptr != NULL)
8451 				V_pfsync_update_state_ptr(s);
8452 			r = s->rule.ptr;
8453 			a = s->anchor.ptr;
8454 		} else if (s == NULL)
8455 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8456 			    &a, &ruleset, inp);
8457 		break;
8458 	}
8459 
8460 	case IPPROTO_SCTP: {
8461 		if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp),
8462 		    &action, &reason, AF_INET)) {
8463 			if (action != PF_PASS)
8464 				pd.act.log |= PF_LOG_FORCE;
8465 			goto done;
8466 		}
8467 		pd.p_len = pd.tot_len - off;
8468 
8469 		pd.sport = &pd.hdr.sctp.src_port;
8470 		pd.dport = &pd.hdr.sctp.dest_port;
8471 		if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) {
8472 			action = PF_DROP;
8473 			REASON_SET(&reason, PFRES_SHORT);
8474 			goto done;
8475 		}
8476 		action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd);
8477 		if (action == PF_DROP)
8478 			goto done;
8479 		action = pf_test_state_sctp(&s, kif, m, off, h, &pd,
8480 		    &reason);
8481 		if (action == PF_PASS) {
8482 			if (V_pfsync_update_state_ptr != NULL)
8483 				V_pfsync_update_state_ptr(s);
8484 			r = s->rule.ptr;
8485 			a = s->anchor.ptr;
8486 		} else if (s == NULL) {
8487 			action = pf_test_rule(&r, &s, kif, m, off,
8488 			    &pd, &a, &ruleset, inp);
8489 		}
8490 		break;
8491 	}
8492 
8493 	case IPPROTO_ICMP: {
8494 		if (!pf_pull_hdr(m, off, &pd.hdr.icmp, ICMP_MINLEN,
8495 		    &action, &reason, AF_INET)) {
8496 			if (action != PF_PASS)
8497 				pd.act.log = PF_LOG_FORCE;
8498 			goto done;
8499 		}
8500 		action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason);
8501 		if (action == PF_PASS) {
8502 			if (V_pfsync_update_state_ptr != NULL)
8503 				V_pfsync_update_state_ptr(s);
8504 			r = s->rule.ptr;
8505 			a = s->anchor.ptr;
8506 		} else if (s == NULL)
8507 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8508 			    &a, &ruleset, inp);
8509 		break;
8510 	}
8511 
8512 #ifdef INET6
8513 	case IPPROTO_ICMPV6: {
8514 		action = PF_DROP;
8515 		DPFPRINTF(PF_DEBUG_MISC,
8516 		    ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
8517 		goto done;
8518 	}
8519 #endif
8520 
8521 	default:
8522 		action = pf_test_state_other(&s, kif, m, &pd);
8523 		if (action == PF_PASS) {
8524 			if (V_pfsync_update_state_ptr != NULL)
8525 				V_pfsync_update_state_ptr(s);
8526 			r = s->rule.ptr;
8527 			a = s->anchor.ptr;
8528 		} else if (s == NULL)
8529 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8530 			    &a, &ruleset, inp);
8531 		break;
8532 	}
8533 
8534 done:
8535 	PF_RULES_RUNLOCK();
8536 	if (action == PF_PASS && h->ip_hl > 5 &&
8537 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
8538 		action = PF_DROP;
8539 		REASON_SET(&reason, PFRES_IPOPTIONS);
8540 		pd.act.log = PF_LOG_FORCE;
8541 		DPFPRINTF(PF_DEBUG_MISC,
8542 		    ("pf: dropping packet with ip options\n"));
8543 	}
8544 
8545 	if (s) {
8546 		memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions));
8547 		tag = s->tag;
8548 		rt = s->rt;
8549 	} else {
8550 		tag = r->tag;
8551 		rt = r->rt;
8552 	}
8553 
8554 	if (tag > 0 && pf_tag_packet(m, &pd, tag)) {
8555 		action = PF_DROP;
8556 		REASON_SET(&reason, PFRES_MEMORY);
8557 	}
8558 
8559 	pf_scrub_ip(&m, &pd);
8560 	if (pd.proto == IPPROTO_TCP && pd.act.max_mss)
8561 		pf_normalize_mss(m, off, &pd);
8562 
8563 	if (pd.act.rtableid >= 0)
8564 		M_SETFIB(m, pd.act.rtableid);
8565 
8566 	if (pd.act.flags & PFSTATE_SETPRIO) {
8567 		if (pd.tos & IPTOS_LOWDELAY)
8568 			use_2nd_queue = 1;
8569 		if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) {
8570 			action = PF_DROP;
8571 			REASON_SET(&reason, PFRES_MEMORY);
8572 			pd.act.log = PF_LOG_FORCE;
8573 			DPFPRINTF(PF_DEBUG_MISC,
8574 			    ("pf: failed to allocate 802.1q mtag\n"));
8575 		}
8576 	}
8577 
8578 #ifdef ALTQ
8579 	if (action == PF_PASS && pd.act.qid) {
8580 		if (pd.pf_mtag == NULL &&
8581 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8582 			action = PF_DROP;
8583 			REASON_SET(&reason, PFRES_MEMORY);
8584 		} else {
8585 			if (s != NULL)
8586 				pd.pf_mtag->qid_hash = pf_state_hash(s);
8587 			if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY))
8588 				pd.pf_mtag->qid = pd.act.pqid;
8589 			else
8590 				pd.pf_mtag->qid = pd.act.qid;
8591 			/* Add hints for ecn. */
8592 			pd.pf_mtag->hdr = h;
8593 		}
8594 	}
8595 #endif /* ALTQ */
8596 
8597 	/*
8598 	 * connections redirected to loopback should not match sockets
8599 	 * bound specifically to loopback due to security implications,
8600 	 * see tcp_input() and in_pcblookup_listen().
8601 	 */
8602 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
8603 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
8604 	    (s->nat_rule.ptr->action == PF_RDR ||
8605 	    s->nat_rule.ptr->action == PF_BINAT) &&
8606 	    IN_LOOPBACK(ntohl(pd.dst->v4.s_addr)))
8607 		m->m_flags |= M_SKIP_FIREWALL;
8608 
8609 	if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS &&
8610 	    r->divert.port && !PACKET_LOOPED(&pd)) {
8611 		mtag = m_tag_alloc(MTAG_PF_DIVERT, 0,
8612 		    sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO);
8613 		if (mtag != NULL) {
8614 			((struct pf_divert_mtag *)(mtag+1))->port =
8615 			    ntohs(r->divert.port);
8616 			((struct pf_divert_mtag *)(mtag+1))->idir =
8617 			    (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN :
8618 			    PF_DIVERT_MTAG_DIR_OUT;
8619 
8620 			if (s)
8621 				PF_STATE_UNLOCK(s);
8622 
8623 			m_tag_prepend(m, mtag);
8624 			if (m->m_flags & M_FASTFWD_OURS) {
8625 				if (pd.pf_mtag == NULL &&
8626 				    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8627 					action = PF_DROP;
8628 					REASON_SET(&reason, PFRES_MEMORY);
8629 					pd.act.log = PF_LOG_FORCE;
8630 					DPFPRINTF(PF_DEBUG_MISC,
8631 					    ("pf: failed to allocate tag\n"));
8632 				} else {
8633 					pd.pf_mtag->flags |=
8634 					    PF_MTAG_FLAG_FASTFWD_OURS_PRESENT;
8635 					m->m_flags &= ~M_FASTFWD_OURS;
8636 				}
8637 			}
8638 			ip_divert_ptr(*m0, dir == PF_IN);
8639 			*m0 = NULL;
8640 
8641 			return (action);
8642 		} else {
8643 			/* XXX: ipfw has the same behaviour! */
8644 			action = PF_DROP;
8645 			REASON_SET(&reason, PFRES_MEMORY);
8646 			pd.act.log = PF_LOG_FORCE;
8647 			DPFPRINTF(PF_DEBUG_MISC,
8648 			    ("pf: failed to allocate divert tag\n"));
8649 		}
8650 	}
8651 	/* this flag will need revising if the pkt is forwarded */
8652 	if (pd.pf_mtag)
8653 		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED;
8654 
8655 	if (pd.act.log) {
8656 		struct pf_krule		*lr;
8657 		struct pf_krule_item	*ri;
8658 
8659 		if (s != NULL && s->nat_rule.ptr != NULL &&
8660 		    s->nat_rule.ptr->log & PF_LOG_ALL)
8661 			lr = s->nat_rule.ptr;
8662 		else
8663 			lr = r;
8664 
8665 		if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL)
8666 			PFLOG_PACKET(kif, m, AF_INET, reason, lr, a, ruleset,
8667 			    &pd, (s == NULL));
8668 		if (s) {
8669 			SLIST_FOREACH(ri, &s->match_rules, entry)
8670 				if (ri->r->log & PF_LOG_ALL)
8671 					PFLOG_PACKET(kif, m, AF_INET, reason,
8672 					    ri->r, a, ruleset, &pd, 0);
8673 		}
8674 	}
8675 
8676 	pf_counter_u64_critical_enter();
8677 	pf_counter_u64_add_protected(&kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS],
8678 	    pd.tot_len);
8679 	pf_counter_u64_add_protected(&kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS],
8680 	    1);
8681 
8682 	if (action == PF_PASS || r->action == PF_DROP) {
8683 		dirndx = (dir == PF_OUT);
8684 		pf_counter_u64_add_protected(&r->packets[dirndx], 1);
8685 		pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len);
8686 		pf_update_timestamp(r);
8687 
8688 		if (a != NULL) {
8689 			pf_counter_u64_add_protected(&a->packets[dirndx], 1);
8690 			pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len);
8691 		}
8692 		if (s != NULL) {
8693 			struct pf_krule_item	*ri;
8694 
8695 			if (s->nat_rule.ptr != NULL) {
8696 				pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx],
8697 				    1);
8698 				pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx],
8699 				    pd.tot_len);
8700 			}
8701 			if (s->src_node != NULL) {
8702 				counter_u64_add(s->src_node->packets[dirndx],
8703 				    1);
8704 				counter_u64_add(s->src_node->bytes[dirndx],
8705 				    pd.tot_len);
8706 			}
8707 			if (s->nat_src_node != NULL) {
8708 				counter_u64_add(s->nat_src_node->packets[dirndx],
8709 				    1);
8710 				counter_u64_add(s->nat_src_node->bytes[dirndx],
8711 				    pd.tot_len);
8712 			}
8713 			dirndx = (dir == s->direction) ? 0 : 1;
8714 			s->packets[dirndx]++;
8715 			s->bytes[dirndx] += pd.tot_len;
8716 			SLIST_FOREACH(ri, &s->match_rules, entry) {
8717 				pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1);
8718 				pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd.tot_len);
8719 			}
8720 		}
8721 		tr = r;
8722 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
8723 		if (nr != NULL && r == &V_pf_default_rule)
8724 			tr = nr;
8725 		if (tr->src.addr.type == PF_ADDR_TABLE)
8726 			pfr_update_stats(tr->src.addr.p.tbl,
8727 			    (s == NULL) ? pd.src :
8728 			    &s->key[(s->direction == PF_IN)]->
8729 				addr[(s->direction == PF_OUT)],
8730 			    pd.af, pd.tot_len, dir == PF_OUT,
8731 			    r->action == PF_PASS, tr->src.neg);
8732 		if (tr->dst.addr.type == PF_ADDR_TABLE)
8733 			pfr_update_stats(tr->dst.addr.p.tbl,
8734 			    (s == NULL) ? pd.dst :
8735 			    &s->key[(s->direction == PF_IN)]->
8736 				addr[(s->direction == PF_IN)],
8737 			    pd.af, pd.tot_len, dir == PF_OUT,
8738 			    r->action == PF_PASS, tr->dst.neg);
8739 	}
8740 	pf_counter_u64_critical_exit();
8741 
8742 	switch (action) {
8743 	case PF_SYNPROXY_DROP:
8744 		m_freem(*m0);
8745 	case PF_DEFER:
8746 		*m0 = NULL;
8747 		action = PF_PASS;
8748 		break;
8749 	case PF_DROP:
8750 		m_freem(*m0);
8751 		*m0 = NULL;
8752 		break;
8753 	default:
8754 		/* pf_route() returns unlocked. */
8755 		if (rt) {
8756 			pf_route(m0, r, kif->pfik_ifp, s, &pd, inp);
8757 			goto out;
8758 		}
8759 		if (pf_dummynet(&pd, s, r, m0) != 0) {
8760 			action = PF_DROP;
8761 			REASON_SET(&reason, PFRES_MEMORY);
8762 		}
8763 		break;
8764 	}
8765 
8766 	SDT_PROBE4(pf, ip, test, done, action, reason, r, s);
8767 
8768 	if (s)
8769 		PF_STATE_UNLOCK(s);
8770 
8771 out:
8772 	pf_sctp_multihome_delayed(&pd, off, kif, s, action);
8773 
8774 	return (action);
8775 }
8776 #endif /* INET */
8777 
8778 #ifdef INET6
8779 int
pf_test6(int dir,int pflags,struct ifnet * ifp,struct mbuf ** m0,struct inpcb * inp,struct pf_rule_actions * default_actions)8780 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp,
8781     struct pf_rule_actions *default_actions)
8782 {
8783 	struct pfi_kkif		*kif;
8784 	u_short			 action, reason = 0;
8785 	struct mbuf		*m = *m0, *n = NULL;
8786 	struct m_tag		*mtag;
8787 	struct ip6_hdr		*h = NULL;
8788 	struct pf_krule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
8789 	struct pf_kstate	*s = NULL;
8790 	struct pf_kruleset	*ruleset = NULL;
8791 	struct pf_pdesc		 pd;
8792 	int			 off, terminal = 0, dirndx, rh_cnt = 0, use_2nd_queue = 0;
8793 	uint16_t		 tag;
8794 	uint8_t			 rt;
8795 
8796 	PF_RULES_RLOCK_TRACKER;
8797 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir));
8798 	M_ASSERTPKTHDR(m);
8799 
8800 	if (!V_pf_status.running)
8801 		return (PF_PASS);
8802 
8803 	PF_RULES_RLOCK();
8804 
8805 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
8806 	if (__predict_false(kif == NULL)) {
8807 		DPFPRINTF(PF_DEBUG_URGENT,
8808 		    ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
8809 		PF_RULES_RUNLOCK();
8810 		return (PF_DROP);
8811 	}
8812 	if (kif->pfik_flags & PFI_IFLAG_SKIP) {
8813 		PF_RULES_RUNLOCK();
8814 		return (PF_PASS);
8815 	}
8816 
8817 	if (m->m_flags & M_SKIP_FIREWALL) {
8818 		PF_RULES_RUNLOCK();
8819 		return (PF_PASS);
8820 	}
8821 
8822 	memset(&pd, 0, sizeof(pd));
8823 	TAILQ_INIT(&pd.sctp_multihome_jobs);
8824 	if (default_actions != NULL)
8825 		memcpy(&pd.act, default_actions, sizeof(pd.act));
8826 	pd.pf_mtag = pf_find_mtag(m);
8827 
8828 	if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) {
8829 		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
8830 
8831 		ifp = ifnet_byindexgen(pd.pf_mtag->if_index,
8832 		    pd.pf_mtag->if_idxgen);
8833 		if (ifp == NULL || ifp->if_flags & IFF_DYING) {
8834 			PF_RULES_RUNLOCK();
8835 			m_freem(*m0);
8836 			*m0 = NULL;
8837 			return (PF_PASS);
8838 		}
8839 		PF_RULES_RUNLOCK();
8840 		nd6_output_ifp(ifp, ifp, m,
8841                     (struct sockaddr_in6 *)&pd.pf_mtag->dst, NULL);
8842 		*m0 = NULL;
8843 		return (PF_PASS);
8844 	}
8845 
8846 	if (pd.pf_mtag && pd.pf_mtag->dnpipe) {
8847 		pd.act.dnpipe = pd.pf_mtag->dnpipe;
8848 		pd.act.flags = pd.pf_mtag->dnflags;
8849 	}
8850 
8851 	if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL &&
8852 	    pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
8853 		pf_dummynet_flag_remove(m, pd.pf_mtag);
8854 		/* Dummynet re-injects packets after they've
8855 		 * completed their delay. We've already
8856 		 * processed them, so pass unconditionally. */
8857 		PF_RULES_RUNLOCK();
8858 		return (PF_PASS);
8859 	}
8860 
8861 	pd.sport = pd.dport = NULL;
8862 	pd.ip_sum = NULL;
8863 	pd.proto_sum = NULL;
8864 	pd.dir = dir;
8865 	pd.sidx = (dir == PF_IN) ? 0 : 1;
8866 	pd.didx = (dir == PF_IN) ? 1 : 0;
8867 	pd.af = AF_INET6;
8868 	pd.act.rtableid = -1;
8869 
8870 	h = mtod(m, struct ip6_hdr *);
8871 	off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
8872 
8873 	/* We do IP header normalization and packet reassembly here */
8874 	if (pf_normalize_ip6(m0, kif, &reason, &pd) != PF_PASS) {
8875 		action = PF_DROP;
8876 		goto done;
8877 	}
8878 	m = *m0;	/* pf_normalize messes with m0 */
8879 	h = mtod(m, struct ip6_hdr *);
8880 	off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
8881 
8882 	/*
8883 	 * we do not support jumbogram.  if we keep going, zero ip6_plen
8884 	 * will do something bad, so drop the packet for now.
8885 	 */
8886 	if (htons(h->ip6_plen) == 0) {
8887 		action = PF_DROP;
8888 		REASON_SET(&reason, PFRES_NORM);	/*XXX*/
8889 		goto done;
8890 	}
8891 
8892 	pd.src = (struct pf_addr *)&h->ip6_src;
8893 	pd.dst = (struct pf_addr *)&h->ip6_dst;
8894 	pd.tos = IPV6_DSCP(h);
8895 	pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
8896 
8897 	pd.proto = h->ip6_nxt;
8898 	do {
8899 		switch (pd.proto) {
8900 		case IPPROTO_FRAGMENT:
8901 			action = pf_test_fragment(&r, kif, m, h, &pd, &a,
8902 			    &ruleset);
8903 			if (action == PF_DROP)
8904 				REASON_SET(&reason, PFRES_FRAG);
8905 			goto done;
8906 		case IPPROTO_ROUTING: {
8907 			struct ip6_rthdr rthdr;
8908 
8909 			if (rh_cnt++) {
8910 				DPFPRINTF(PF_DEBUG_MISC,
8911 				    ("pf: IPv6 more than one rthdr\n"));
8912 				action = PF_DROP;
8913 				REASON_SET(&reason, PFRES_IPOPTIONS);
8914 				pd.act.log = PF_LOG_FORCE;
8915 				goto done;
8916 			}
8917 			if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
8918 			    &reason, pd.af)) {
8919 				DPFPRINTF(PF_DEBUG_MISC,
8920 				    ("pf: IPv6 short rthdr\n"));
8921 				action = PF_DROP;
8922 				REASON_SET(&reason, PFRES_SHORT);
8923 				pd.act.log = PF_LOG_FORCE;
8924 				goto done;
8925 			}
8926 			if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
8927 				DPFPRINTF(PF_DEBUG_MISC,
8928 				    ("pf: IPv6 rthdr0\n"));
8929 				action = PF_DROP;
8930 				REASON_SET(&reason, PFRES_IPOPTIONS);
8931 				pd.act.log = PF_LOG_FORCE;
8932 				goto done;
8933 			}
8934 			/* FALLTHROUGH */
8935 		}
8936 		case IPPROTO_AH:
8937 		case IPPROTO_HOPOPTS:
8938 		case IPPROTO_DSTOPTS: {
8939 			/* get next header and header length */
8940 			struct ip6_ext	opt6;
8941 
8942 			if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
8943 			    NULL, &reason, pd.af)) {
8944 				DPFPRINTF(PF_DEBUG_MISC,
8945 				    ("pf: IPv6 short opt\n"));
8946 				action = PF_DROP;
8947 				pd.act.log = PF_LOG_FORCE;
8948 				goto done;
8949 			}
8950 			if (pd.proto == IPPROTO_AH)
8951 				off += (opt6.ip6e_len + 2) * 4;
8952 			else
8953 				off += (opt6.ip6e_len + 1) * 8;
8954 			pd.proto = opt6.ip6e_nxt;
8955 			/* goto the next header */
8956 			break;
8957 		}
8958 		default:
8959 			terminal++;
8960 			break;
8961 		}
8962 	} while (!terminal);
8963 
8964 	/* if there's no routing header, use unmodified mbuf for checksumming */
8965 	if (!n)
8966 		n = m;
8967 
8968 	switch (pd.proto) {
8969 	case IPPROTO_TCP: {
8970 		if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp),
8971 		    &action, &reason, AF_INET6)) {
8972 			if (action != PF_PASS)
8973 				pd.act.log |= PF_LOG_FORCE;
8974 			goto done;
8975 		}
8976 		pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2);
8977 		pd.sport = &pd.hdr.tcp.th_sport;
8978 		pd.dport = &pd.hdr.tcp.th_dport;
8979 
8980 		/* Respond to SYN with a syncookie. */
8981 		if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN &&
8982 		    pd.dir == PF_IN && pf_synflood_check(&pd)) {
8983 			pf_syncookie_send(m, off, &pd);
8984 			action = PF_DROP;
8985 			break;
8986 		}
8987 
8988 		action = pf_normalize_tcp(kif, m, 0, off, h, &pd);
8989 		if (action == PF_DROP)
8990 			goto done;
8991 		action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason);
8992 		if (action == PF_PASS) {
8993 			if (V_pfsync_update_state_ptr != NULL)
8994 				V_pfsync_update_state_ptr(s);
8995 			r = s->rule.ptr;
8996 			a = s->anchor.ptr;
8997 		} else if (s == NULL) {
8998 			/* Validate remote SYN|ACK, re-create original SYN if
8999 			 * valid. */
9000 			if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) ==
9001 			    TH_ACK && pf_syncookie_validate(&pd) &&
9002 			    pd.dir == PF_IN) {
9003 				struct mbuf *msyn;
9004 
9005 				msyn = pf_syncookie_recreate_syn(h->ip6_hlim,
9006 				    off, &pd);
9007 				if (msyn == NULL) {
9008 					action = PF_DROP;
9009 					break;
9010 				}
9011 
9012 				action = pf_test6(dir, pflags, ifp, &msyn, inp,
9013 				    &pd.act);
9014 				m_freem(msyn);
9015 				if (action != PF_PASS)
9016 					break;
9017 
9018 				action = pf_test_state_tcp(&s, kif, m, off, h,
9019 				    &pd, &reason);
9020 				if (action != PF_PASS || s == NULL) {
9021 					action = PF_DROP;
9022 					break;
9023 				}
9024 
9025 				s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1;
9026 				s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1;
9027 				pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST);
9028 
9029 				action = pf_synproxy(&pd, &s, &reason);
9030 				break;
9031 			} else {
9032 				action = pf_test_rule(&r, &s, kif, m, off, &pd,
9033 				    &a, &ruleset, inp);
9034 			}
9035 		}
9036 		break;
9037 	}
9038 
9039 	case IPPROTO_UDP: {
9040 		if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp),
9041 		    &action, &reason, AF_INET6)) {
9042 			if (action != PF_PASS)
9043 				pd.act.log |= PF_LOG_FORCE;
9044 			goto done;
9045 		}
9046 		pd.sport = &pd.hdr.udp.uh_sport;
9047 		pd.dport = &pd.hdr.udp.uh_dport;
9048 		if (pd.hdr.udp.uh_dport == 0 ||
9049 		    ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off ||
9050 		    ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) {
9051 			action = PF_DROP;
9052 			REASON_SET(&reason, PFRES_SHORT);
9053 			goto done;
9054 		}
9055 		action = pf_test_state_udp(&s, kif, m, off, h, &pd);
9056 		if (action == PF_PASS) {
9057 			if (V_pfsync_update_state_ptr != NULL)
9058 				V_pfsync_update_state_ptr(s);
9059 			r = s->rule.ptr;
9060 			a = s->anchor.ptr;
9061 		} else if (s == NULL)
9062 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
9063 			    &a, &ruleset, inp);
9064 		break;
9065 	}
9066 
9067 	case IPPROTO_SCTP: {
9068 		if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp),
9069 		    &action, &reason, AF_INET6)) {
9070 			if (action != PF_PASS)
9071 				pd.act.log |= PF_LOG_FORCE;
9072 			goto done;
9073 		}
9074 		pd.sport = &pd.hdr.sctp.src_port;
9075 		pd.dport = &pd.hdr.sctp.dest_port;
9076 		if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) {
9077 			action = PF_DROP;
9078 			REASON_SET(&reason, PFRES_SHORT);
9079 			goto done;
9080 		}
9081 		action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd);
9082 		if (action == PF_DROP)
9083 			goto done;
9084 		action = pf_test_state_sctp(&s, kif, m, off, h, &pd,
9085 		    &reason);
9086 		if (action == PF_PASS) {
9087 			if (V_pfsync_update_state_ptr != NULL)
9088 				V_pfsync_update_state_ptr(s);
9089 			r = s->rule.ptr;
9090 			a = s->anchor.ptr;
9091 		} else if (s == NULL) {
9092 			action = pf_test_rule(&r, &s, kif, m, off,
9093 			    &pd, &a, &ruleset, inp);
9094 		}
9095 		break;
9096 	}
9097 
9098 	case IPPROTO_ICMP: {
9099 		action = PF_DROP;
9100 		DPFPRINTF(PF_DEBUG_MISC,
9101 		    ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
9102 		goto done;
9103 	}
9104 
9105 	case IPPROTO_ICMPV6: {
9106 		if (!pf_pull_hdr(m, off, &pd.hdr.icmp6, sizeof(pd.hdr.icmp6),
9107 		    &action, &reason, AF_INET6)) {
9108 			if (action != PF_PASS)
9109 				pd.act.log |= PF_LOG_FORCE;
9110 			goto done;
9111 		}
9112 		action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason);
9113 		if (action == PF_PASS) {
9114 			if (V_pfsync_update_state_ptr != NULL)
9115 				V_pfsync_update_state_ptr(s);
9116 			r = s->rule.ptr;
9117 			a = s->anchor.ptr;
9118 		} else if (s == NULL)
9119 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
9120 			    &a, &ruleset, inp);
9121 		break;
9122 	}
9123 
9124 	default:
9125 		action = pf_test_state_other(&s, kif, m, &pd);
9126 		if (action == PF_PASS) {
9127 			if (V_pfsync_update_state_ptr != NULL)
9128 				V_pfsync_update_state_ptr(s);
9129 			r = s->rule.ptr;
9130 			a = s->anchor.ptr;
9131 		} else if (s == NULL)
9132 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
9133 			    &a, &ruleset, inp);
9134 		break;
9135 	}
9136 
9137 done:
9138 	PF_RULES_RUNLOCK();
9139 	if (n != m) {
9140 		m_freem(n);
9141 		n = NULL;
9142 	}
9143 
9144 	/* handle dangerous IPv6 extension headers. */
9145 	if (action == PF_PASS && rh_cnt &&
9146 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
9147 		action = PF_DROP;
9148 		REASON_SET(&reason, PFRES_IPOPTIONS);
9149 		pd.act.log = r->log;
9150 		DPFPRINTF(PF_DEBUG_MISC,
9151 		    ("pf: dropping packet with dangerous v6 headers\n"));
9152 	}
9153 
9154 	if (s) {
9155 		memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions));
9156 		tag = s->tag;
9157 		rt = s->rt;
9158 	} else {
9159 		tag = r->tag;
9160 		rt = r->rt;
9161 	}
9162 
9163 	if (tag > 0 && pf_tag_packet(m, &pd, tag)) {
9164 		action = PF_DROP;
9165 		REASON_SET(&reason, PFRES_MEMORY);
9166 	}
9167 
9168 	pf_scrub_ip6(&m, &pd);
9169 	if (pd.proto == IPPROTO_TCP && pd.act.max_mss)
9170 		pf_normalize_mss(m, off, &pd);
9171 
9172 	if (pd.act.rtableid >= 0)
9173 		M_SETFIB(m, pd.act.rtableid);
9174 
9175 	if (pd.act.flags & PFSTATE_SETPRIO) {
9176 		if (pd.tos & IPTOS_LOWDELAY)
9177 			use_2nd_queue = 1;
9178 		if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) {
9179 			action = PF_DROP;
9180 			REASON_SET(&reason, PFRES_MEMORY);
9181 			pd.act.log = PF_LOG_FORCE;
9182 			DPFPRINTF(PF_DEBUG_MISC,
9183 			    ("pf: failed to allocate 802.1q mtag\n"));
9184 		}
9185 	}
9186 
9187 #ifdef ALTQ
9188 	if (action == PF_PASS && pd.act.qid) {
9189 		if (pd.pf_mtag == NULL &&
9190 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
9191 			action = PF_DROP;
9192 			REASON_SET(&reason, PFRES_MEMORY);
9193 		} else {
9194 			if (s != NULL)
9195 				pd.pf_mtag->qid_hash = pf_state_hash(s);
9196 			if (pd.tos & IPTOS_LOWDELAY)
9197 				pd.pf_mtag->qid = pd.act.pqid;
9198 			else
9199 				pd.pf_mtag->qid = pd.act.qid;
9200 			/* Add hints for ecn. */
9201 			pd.pf_mtag->hdr = h;
9202 		}
9203 	}
9204 #endif /* ALTQ */
9205 
9206 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
9207 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
9208 	    (s->nat_rule.ptr->action == PF_RDR ||
9209 	    s->nat_rule.ptr->action == PF_BINAT) &&
9210 	    IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
9211 		m->m_flags |= M_SKIP_FIREWALL;
9212 
9213 	/* XXX: Anybody working on it?! */
9214 	if (r->divert.port)
9215 		printf("pf: divert(9) is not supported for IPv6\n");
9216 
9217 	if (pd.act.log) {
9218 		struct pf_krule		*lr;
9219 		struct pf_krule_item	*ri;
9220 
9221 		if (s != NULL && s->nat_rule.ptr != NULL &&
9222 		    s->nat_rule.ptr->log & PF_LOG_ALL)
9223 			lr = s->nat_rule.ptr;
9224 		else
9225 			lr = r;
9226 
9227 		if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL)
9228 			PFLOG_PACKET(kif, m, AF_INET6, reason, lr, a, ruleset,
9229 			    &pd, (s == NULL));
9230 		if (s) {
9231 			SLIST_FOREACH(ri, &s->match_rules, entry)
9232 				if (ri->r->log & PF_LOG_ALL)
9233 					PFLOG_PACKET(kif, m, AF_INET6, reason,
9234 					    ri->r, a, ruleset, &pd, 0);
9235 		}
9236 	}
9237 
9238 	pf_counter_u64_critical_enter();
9239 	pf_counter_u64_add_protected(&kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS],
9240 	    pd.tot_len);
9241 	pf_counter_u64_add_protected(&kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS],
9242 	    1);
9243 
9244 	if (action == PF_PASS || r->action == PF_DROP) {
9245 		dirndx = (dir == PF_OUT);
9246 		pf_counter_u64_add_protected(&r->packets[dirndx], 1);
9247 		pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len);
9248 		if (a != NULL) {
9249 			pf_counter_u64_add_protected(&a->packets[dirndx], 1);
9250 			pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len);
9251 		}
9252 		if (s != NULL) {
9253 			if (s->nat_rule.ptr != NULL) {
9254 				pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx],
9255 				    1);
9256 				pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx],
9257 				    pd.tot_len);
9258 			}
9259 			if (s->src_node != NULL) {
9260 				counter_u64_add(s->src_node->packets[dirndx],
9261 				    1);
9262 				counter_u64_add(s->src_node->bytes[dirndx],
9263 				    pd.tot_len);
9264 			}
9265 			if (s->nat_src_node != NULL) {
9266 				counter_u64_add(s->nat_src_node->packets[dirndx],
9267 				    1);
9268 				counter_u64_add(s->nat_src_node->bytes[dirndx],
9269 				    pd.tot_len);
9270 			}
9271 			dirndx = (dir == s->direction) ? 0 : 1;
9272 			s->packets[dirndx]++;
9273 			s->bytes[dirndx] += pd.tot_len;
9274 		}
9275 		tr = r;
9276 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
9277 		if (nr != NULL && r == &V_pf_default_rule)
9278 			tr = nr;
9279 		if (tr->src.addr.type == PF_ADDR_TABLE)
9280 			pfr_update_stats(tr->src.addr.p.tbl,
9281 			    (s == NULL) ? pd.src :
9282 			    &s->key[(s->direction == PF_IN)]->addr[0],
9283 			    pd.af, pd.tot_len, dir == PF_OUT,
9284 			    r->action == PF_PASS, tr->src.neg);
9285 		if (tr->dst.addr.type == PF_ADDR_TABLE)
9286 			pfr_update_stats(tr->dst.addr.p.tbl,
9287 			    (s == NULL) ? pd.dst :
9288 			    &s->key[(s->direction == PF_IN)]->addr[1],
9289 			    pd.af, pd.tot_len, dir == PF_OUT,
9290 			    r->action == PF_PASS, tr->dst.neg);
9291 	}
9292 	pf_counter_u64_critical_exit();
9293 
9294 	switch (action) {
9295 	case PF_SYNPROXY_DROP:
9296 		m_freem(*m0);
9297 	case PF_DEFER:
9298 		*m0 = NULL;
9299 		action = PF_PASS;
9300 		break;
9301 	case PF_DROP:
9302 		m_freem(*m0);
9303 		*m0 = NULL;
9304 		break;
9305 	default:
9306 		/* pf_route6() returns unlocked. */
9307 		if (rt) {
9308 			pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp);
9309 			goto out;
9310 		}
9311 		if (pf_dummynet(&pd, s, r, m0) != 0) {
9312 			action = PF_DROP;
9313 			REASON_SET(&reason, PFRES_MEMORY);
9314 		}
9315 		break;
9316 	}
9317 
9318 	if (s)
9319 		PF_STATE_UNLOCK(s);
9320 
9321 	/* If reassembled packet passed, create new fragments. */
9322 	if (action == PF_PASS && *m0 && dir == PF_OUT &&
9323 	    (mtag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL)
9324 		action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD);
9325 
9326 out:
9327 	SDT_PROBE4(pf, ip, test6, done, action, reason, r, s);
9328 
9329 	pf_sctp_multihome_delayed(&pd, off, kif, s, action);
9330 
9331 	return (action);
9332 }
9333 #endif /* INET6 */
9334