xref: /freebsd-14-stable/sys/netpfil/pf/pf.c (revision b89bc49a63483bdb0617c2fcebeb2faf2565e994)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2001 Daniel Hartmeier
5  * Copyright (c) 2002 - 2008 Henning Brauer
6  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  *
13  *    - Redistributions of source code must retain the above copyright
14  *      notice, this list of conditions and the following disclaimer.
15  *    - Redistributions in binary form must reproduce the above
16  *      copyright notice, this list of conditions and the following
17  *      disclaimer in the documentation and/or other materials provided
18  *      with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
28  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
30  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31  * POSSIBILITY OF SUCH DAMAGE.
32  *
33  * Effort sponsored in part by the Defense Advanced Research Projects
34  * Agency (DARPA) and Air Force Research Laboratory, Air Force
35  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
36  *
37  *	$OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
38  */
39 
40 #include <sys/cdefs.h>
41 #include "opt_bpf.h"
42 #include "opt_inet.h"
43 #include "opt_inet6.h"
44 #include "opt_pf.h"
45 #include "opt_sctp.h"
46 
47 #include <sys/param.h>
48 #include <sys/bus.h>
49 #include <sys/endian.h>
50 #include <sys/gsb_crc32.h>
51 #include <sys/hash.h>
52 #include <sys/interrupt.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/limits.h>
56 #include <sys/mbuf.h>
57 #include <sys/md5.h>
58 #include <sys/random.h>
59 #include <sys/refcount.h>
60 #include <sys/sdt.h>
61 #include <sys/socket.h>
62 #include <sys/sysctl.h>
63 #include <sys/taskqueue.h>
64 #include <sys/ucred.h>
65 
66 #include <net/if.h>
67 #include <net/if_var.h>
68 #include <net/if_private.h>
69 #include <net/if_types.h>
70 #include <net/if_vlan_var.h>
71 #include <net/route.h>
72 #include <net/route/nhop.h>
73 #include <net/vnet.h>
74 
75 #include <net/pfil.h>
76 #include <net/pfvar.h>
77 #include <net/if_pflog.h>
78 #include <net/if_pfsync.h>
79 
80 #include <netinet/in_pcb.h>
81 #include <netinet/in_var.h>
82 #include <netinet/in_fib.h>
83 #include <netinet/ip.h>
84 #include <netinet/ip_fw.h>
85 #include <netinet/ip_icmp.h>
86 #include <netinet/icmp_var.h>
87 #include <netinet/ip_var.h>
88 #include <netinet/tcp.h>
89 #include <netinet/tcp_fsm.h>
90 #include <netinet/tcp_seq.h>
91 #include <netinet/tcp_timer.h>
92 #include <netinet/tcp_var.h>
93 #include <netinet/udp.h>
94 #include <netinet/udp_var.h>
95 
96 /* dummynet */
97 #include <netinet/ip_dummynet.h>
98 #include <netinet/ip_fw.h>
99 #include <netpfil/ipfw/dn_heap.h>
100 #include <netpfil/ipfw/ip_fw_private.h>
101 #include <netpfil/ipfw/ip_dn_private.h>
102 
103 #ifdef INET6
104 #include <netinet/ip6.h>
105 #include <netinet/icmp6.h>
106 #include <netinet6/nd6.h>
107 #include <netinet6/ip6_var.h>
108 #include <netinet6/in6_pcb.h>
109 #include <netinet6/in6_fib.h>
110 #include <netinet6/scope6_var.h>
111 #endif /* INET6 */
112 
113 #include <netinet/sctp_header.h>
114 #include <netinet/sctp_crc32.h>
115 
116 #include <machine/in_cksum.h>
117 #include <security/mac/mac_framework.h>
118 
119 #define	DPFPRINTF(n, x)	if (V_pf_status.debug >= (n)) printf x
120 
121 SDT_PROVIDER_DEFINE(pf);
122 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *",
123     "struct pf_kstate *");
124 SDT_PROBE_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *",
125     "struct pf_kstate *");
126 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *",
127     "struct pf_state_key_cmp *", "int", "struct pf_pdesc *",
128     "struct pf_kstate *");
129 SDT_PROBE_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 SDT_PROBE_DEFINE4(pf, sctp, multihome_scan, entry, "int",
136     "int", "struct pf_pdesc *", "int");
137 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, param, "uint16_t", "uint16_t");
138 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv4, "struct in_addr *",
139     "int");
140 SDT_PROBE_DEFINE2(pf, sctp, multihome_scan, ipv6, "struct in_addr6 *",
141     "int");
142 
143 SDT_PROBE_DEFINE3(pf, eth, test_rule, entry, "int", "struct ifnet *",
144     "struct mbuf *");
145 SDT_PROBE_DEFINE2(pf, eth, test_rule, test, "int", "struct pf_keth_rule *");
146 SDT_PROBE_DEFINE3(pf, eth, test_rule, mismatch,
147     "int", "struct pf_keth_rule *", "char *");
148 SDT_PROBE_DEFINE2(pf, eth, test_rule, match, "int", "struct pf_keth_rule *");
149 SDT_PROBE_DEFINE2(pf, eth, test_rule, final_match,
150     "int", "struct pf_keth_rule *");
151 SDT_PROBE_DEFINE2(pf, purge, state, rowcount, "int", "size_t");
152 
153 /*
154  * Global variables
155  */
156 
157 /* state tables */
158 VNET_DEFINE(struct pf_altqqueue,	 pf_altqs[4]);
159 VNET_DEFINE(struct pf_kpalist,		 pf_pabuf);
160 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_active);
161 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_active);
162 VNET_DEFINE(struct pf_altqqueue *,	 pf_altqs_inactive);
163 VNET_DEFINE(struct pf_altqqueue *,	 pf_altq_ifs_inactive);
164 VNET_DEFINE(struct pf_kstatus,		 pf_status);
165 
166 VNET_DEFINE(u_int32_t,			 ticket_altqs_active);
167 VNET_DEFINE(u_int32_t,			 ticket_altqs_inactive);
168 VNET_DEFINE(int,			 altqs_inactive_open);
169 VNET_DEFINE(u_int32_t,			 ticket_pabuf);
170 
171 VNET_DEFINE(MD5_CTX,			 pf_tcp_secret_ctx);
172 #define	V_pf_tcp_secret_ctx		 VNET(pf_tcp_secret_ctx)
173 VNET_DEFINE(u_char,			 pf_tcp_secret[16]);
174 #define	V_pf_tcp_secret			 VNET(pf_tcp_secret)
175 VNET_DEFINE(int,			 pf_tcp_secret_init);
176 #define	V_pf_tcp_secret_init		 VNET(pf_tcp_secret_init)
177 VNET_DEFINE(int,			 pf_tcp_iss_off);
178 #define	V_pf_tcp_iss_off		 VNET(pf_tcp_iss_off)
179 VNET_DECLARE(int,			 pf_vnet_active);
180 #define	V_pf_vnet_active		 VNET(pf_vnet_active)
181 
182 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx);
183 #define V_pf_purge_idx	VNET(pf_purge_idx)
184 
185 #ifdef PF_WANT_32_TO_64_COUNTER
186 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter);
187 #define	V_pf_counter_periodic_iter	VNET(pf_counter_periodic_iter)
188 
189 VNET_DEFINE(struct allrulelist_head, pf_allrulelist);
190 VNET_DEFINE(size_t, pf_allrulecount);
191 VNET_DEFINE(struct pf_krule *, pf_rulemarker);
192 #endif
193 
194 struct pf_sctp_endpoint;
195 RB_HEAD(pf_sctp_endpoints, pf_sctp_endpoint);
196 struct pf_sctp_source {
197 	sa_family_t			af;
198 	struct pf_addr			addr;
199 	TAILQ_ENTRY(pf_sctp_source)	entry;
200 };
201 TAILQ_HEAD(pf_sctp_sources, pf_sctp_source);
202 struct pf_sctp_endpoint
203 {
204 	uint32_t		 v_tag;
205 	struct pf_sctp_sources	 sources;
206 	RB_ENTRY(pf_sctp_endpoint)	entry;
207 };
208 static int
pf_sctp_endpoint_compare(struct pf_sctp_endpoint * a,struct pf_sctp_endpoint * b)209 pf_sctp_endpoint_compare(struct pf_sctp_endpoint *a, struct pf_sctp_endpoint *b)
210 {
211 	return (a->v_tag - b->v_tag);
212 }
213 RB_PROTOTYPE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare);
214 RB_GENERATE(pf_sctp_endpoints, pf_sctp_endpoint, entry, pf_sctp_endpoint_compare);
215 VNET_DEFINE_STATIC(struct pf_sctp_endpoints, pf_sctp_endpoints);
216 #define V_pf_sctp_endpoints	VNET(pf_sctp_endpoints)
217 static struct mtx_padalign pf_sctp_endpoints_mtx;
218 MTX_SYSINIT(pf_sctp_endpoints_mtx, &pf_sctp_endpoints_mtx, "SCTP endpoints", MTX_DEF);
219 #define	PF_SCTP_ENDPOINTS_LOCK()	mtx_lock(&pf_sctp_endpoints_mtx)
220 #define	PF_SCTP_ENDPOINTS_UNLOCK()	mtx_unlock(&pf_sctp_endpoints_mtx)
221 
222 /*
223  * Queue for pf_intr() sends.
224  */
225 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
226 struct pf_send_entry {
227 	STAILQ_ENTRY(pf_send_entry)	pfse_next;
228 	struct mbuf			*pfse_m;
229 	enum {
230 		PFSE_IP,
231 		PFSE_IP6,
232 		PFSE_ICMP,
233 		PFSE_ICMP6,
234 	}				pfse_type;
235 	struct {
236 		int		type;
237 		int		code;
238 		int		mtu;
239 	} icmpopts;
240 };
241 
242 STAILQ_HEAD(pf_send_head, pf_send_entry);
243 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue);
244 #define	V_pf_sendqueue	VNET(pf_sendqueue)
245 
246 static struct mtx_padalign pf_sendqueue_mtx;
247 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
248 #define	PF_SENDQ_LOCK()		mtx_lock(&pf_sendqueue_mtx)
249 #define	PF_SENDQ_UNLOCK()	mtx_unlock(&pf_sendqueue_mtx)
250 
251 /*
252  * Queue for pf_overload_task() tasks.
253  */
254 struct pf_overload_entry {
255 	SLIST_ENTRY(pf_overload_entry)	next;
256 	struct pf_addr  		addr;
257 	sa_family_t			af;
258 	uint8_t				dir;
259 	struct pf_krule  		*rule;
260 };
261 
262 SLIST_HEAD(pf_overload_head, pf_overload_entry);
263 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue);
264 #define V_pf_overloadqueue	VNET(pf_overloadqueue)
265 VNET_DEFINE_STATIC(struct task, pf_overloadtask);
266 #define	V_pf_overloadtask	VNET(pf_overloadtask)
267 
268 static struct mtx_padalign pf_overloadqueue_mtx;
269 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
270     "pf overload/flush queue", MTX_DEF);
271 #define	PF_OVERLOADQ_LOCK()	mtx_lock(&pf_overloadqueue_mtx)
272 #define	PF_OVERLOADQ_UNLOCK()	mtx_unlock(&pf_overloadqueue_mtx)
273 
274 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules);
275 struct mtx_padalign pf_unlnkdrules_mtx;
276 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
277     MTX_DEF);
278 
279 struct sx pf_config_lock;
280 SX_SYSINIT(pf_config_lock, &pf_config_lock, "pf config");
281 
282 struct mtx_padalign pf_table_stats_lock;
283 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats",
284     MTX_DEF);
285 
286 VNET_DEFINE_STATIC(uma_zone_t,	pf_sources_z);
287 #define	V_pf_sources_z	VNET(pf_sources_z)
288 uma_zone_t		pf_mtag_z;
289 VNET_DEFINE(uma_zone_t,	 pf_state_z);
290 VNET_DEFINE(uma_zone_t,	 pf_state_key_z);
291 
292 VNET_DEFINE(struct unrhdr64, pf_stateid);
293 
294 static void		 pf_src_tree_remove_state(struct pf_kstate *);
295 static void		 pf_init_threshold(struct pf_threshold *, u_int32_t,
296 			    u_int32_t);
297 static void		 pf_add_threshold(struct pf_threshold *);
298 static int		 pf_check_threshold(struct pf_threshold *);
299 
300 static void		 pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
301 			    u_int16_t *, u_int16_t *, struct pf_addr *,
302 			    u_int16_t, u_int8_t, sa_family_t);
303 static int		 pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
304 			    struct tcphdr *, struct pf_state_peer *);
305 int			 pf_icmp_mapping(struct pf_pdesc *, u_int8_t, int *,
306 			    int *, u_int16_t *, u_int16_t *);
307 static void		 pf_change_icmp(struct pf_addr *, u_int16_t *,
308 			    struct pf_addr *, struct pf_addr *, u_int16_t,
309 			    u_int16_t *, u_int16_t *, u_int16_t *,
310 			    u_int16_t *, u_int8_t, sa_family_t);
311 static void		 pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
312 			    sa_family_t, struct pf_krule *, int);
313 static void		 pf_detach_state(struct pf_kstate *);
314 static int		 pf_state_key_attach(struct pf_state_key *,
315 			    struct pf_state_key *, struct pf_kstate *);
316 static void		 pf_state_key_detach(struct pf_kstate *, int);
317 static int		 pf_state_key_ctor(void *, int, void *, int);
318 static u_int32_t	 pf_tcp_iss(struct pf_pdesc *);
319 static __inline void	 pf_dummynet_flag_remove(struct mbuf *m,
320 			    struct pf_mtag *pf_mtag);
321 static int		 pf_dummynet(struct pf_pdesc *, struct pf_kstate *,
322 			    struct pf_krule *, struct mbuf **);
323 static int		 pf_dummynet_route(struct pf_pdesc *,
324 			    struct pf_kstate *, struct pf_krule *,
325 			    struct ifnet *, struct sockaddr *, struct mbuf **);
326 static int		 pf_test_eth_rule(int, struct pfi_kkif *,
327 			    struct mbuf **);
328 static int		 pf_test_rule(struct pf_krule **, struct pf_kstate **,
329 			    struct pfi_kkif *, struct mbuf *, int,
330 			    struct pf_pdesc *, struct pf_krule **,
331 			    struct pf_kruleset **, struct inpcb *);
332 static int		 pf_create_state(struct pf_krule *, struct pf_krule *,
333 			    struct pf_krule *, struct pf_pdesc *,
334 			    struct pf_ksrc_node *, struct pf_state_key *,
335 			    struct pf_state_key *, struct mbuf *, int,
336 			    u_int16_t, u_int16_t, int *, struct pfi_kkif *,
337 			    struct pf_kstate **, int, u_int16_t, u_int16_t,
338 			    int, struct pf_krule_slist *);
339 static int		 pf_state_key_addr_setup(struct pf_pdesc *, struct mbuf *,
340 			    int, struct pf_state_key_cmp *, int, struct pf_addr *,
341 			    int, struct pf_addr *, int);
342 static int		 pf_test_fragment(struct pf_krule **, struct pfi_kkif *,
343 			    struct mbuf *, void *, struct pf_pdesc *,
344 			    struct pf_krule **, struct pf_kruleset **);
345 static int		 pf_tcp_track_full(struct pf_kstate **,
346 			    struct pfi_kkif *, struct mbuf *, int,
347 			    struct pf_pdesc *, u_short *, int *);
348 static int		 pf_tcp_track_sloppy(struct pf_kstate **,
349 			    struct pf_pdesc *, u_short *);
350 static int		 pf_test_state_tcp(struct pf_kstate **,
351 			    struct pfi_kkif *, struct mbuf *, int,
352 			    void *, struct pf_pdesc *, u_short *);
353 static int		 pf_test_state_udp(struct pf_kstate **,
354 			    struct pfi_kkif *, struct mbuf *, int,
355 			    void *, struct pf_pdesc *);
356 int			 pf_icmp_state_lookup(struct pf_state_key_cmp *,
357 			    struct pf_pdesc *, struct pf_kstate **, struct mbuf *,
358 			    int, int, struct pfi_kkif *, u_int16_t, u_int16_t,
359 			    int, int *, int, int);
360 static int		 pf_test_state_icmp(struct pf_kstate **,
361 			    struct pfi_kkif *, struct mbuf *, int,
362 			    void *, struct pf_pdesc *, u_short *);
363 static void		 pf_sctp_multihome_detach_addr(const struct pf_kstate *);
364 static void		 pf_sctp_multihome_delayed(struct pf_pdesc *, int,
365 			    struct pfi_kkif *, struct pf_kstate *, int);
366 static int		 pf_test_state_sctp(struct pf_kstate **,
367 			    struct pfi_kkif *, struct mbuf *, int,
368 			    void *, struct pf_pdesc *, u_short *);
369 static int		 pf_test_state_other(struct pf_kstate **,
370 			    struct pfi_kkif *, struct mbuf *, struct pf_pdesc *);
371 static u_int16_t	 pf_calc_mss(struct pf_addr *, sa_family_t,
372 				int, u_int16_t);
373 static int		 pf_check_proto_cksum(struct mbuf *, int, int,
374 			    u_int8_t, sa_family_t);
375 static void		 pf_print_state_parts(struct pf_kstate *,
376 			    struct pf_state_key *, struct pf_state_key *);
377 static void		 pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t,
378 			    bool, u_int8_t);
379 static struct pf_kstate	*pf_find_state(struct pfi_kkif *,
380 			    const struct pf_state_key_cmp *, u_int);
381 static int		 pf_src_connlimit(struct pf_kstate **);
382 static void		 pf_overload_task(void *v, int pending);
383 static u_short		 pf_insert_src_node(struct pf_ksrc_node **,
384 			    struct pf_krule *, struct pf_addr *, sa_family_t);
385 static u_int		 pf_purge_expired_states(u_int, int);
386 static void		 pf_purge_unlinked_rules(void);
387 static int		 pf_mtag_uminit(void *, int, int);
388 static void		 pf_mtag_free(struct m_tag *);
389 static void		 pf_packet_rework_nat(struct mbuf *, struct pf_pdesc *,
390 			    int, struct pf_state_key *);
391 #ifdef INET
392 static void		 pf_route(struct mbuf **, struct pf_krule *,
393 			    struct ifnet *, struct pf_kstate *,
394 			    struct pf_pdesc *, struct inpcb *);
395 #endif /* INET */
396 #ifdef INET6
397 static void		 pf_change_a6(struct pf_addr *, u_int16_t *,
398 			    struct pf_addr *, u_int8_t);
399 static void		 pf_route6(struct mbuf **, struct pf_krule *,
400 			    struct ifnet *, struct pf_kstate *,
401 			    struct pf_pdesc *, struct inpcb *);
402 #endif /* INET6 */
403 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t);
404 
405 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
406 
407 extern int pf_end_threads;
408 extern struct proc *pf_purge_proc;
409 
410 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
411 
412 enum { PF_ICMP_MULTI_NONE, PF_ICMP_MULTI_LINK };
413 
414 #define	PACKET_UNDO_NAT(_m, _pd, _off, _s)		\
415 	do {								\
416 		struct pf_state_key *nk;				\
417 		if ((pd->dir) == PF_OUT)					\
418 			nk = (_s)->key[PF_SK_STACK];			\
419 		else							\
420 			nk = (_s)->key[PF_SK_WIRE];			\
421 		pf_packet_rework_nat(_m, _pd, _off, nk);		\
422 	} while (0)
423 
424 #define	PACKET_LOOPED(pd)	((pd)->pf_mtag &&			\
425 				 (pd)->pf_mtag->flags & PF_MTAG_FLAG_PACKET_LOOPED)
426 
427 #define	STATE_LOOKUP(i, k, s, pd)					\
428 	do {								\
429 		(s) = pf_find_state((i), (k), (pd->dir));			\
430 		SDT_PROBE5(pf, ip, state, lookup, i, k, (pd->dir), pd, (s));	\
431 		if ((s) == NULL)					\
432 			return (PF_DROP);				\
433 		if (PACKET_LOOPED(pd))					\
434 			return (PF_PASS);				\
435 	} while (0)
436 
437 static struct pfi_kkif *
BOUND_IFACE(struct pf_krule * rule,struct pfi_kkif * k,struct pf_pdesc * pd)438 BOUND_IFACE(struct pf_krule *rule, struct pfi_kkif *k, struct pf_pdesc *pd)
439 {
440 	/* Floating unless otherwise specified. */
441 	if (! (rule->rule_flag & PFRULE_IFBOUND))
442 		return (V_pfi_all);
443 
444 	/*
445 	 * If this state is created based on another state (e.g. SCTP
446 	 * multihome) always set it floating initially. We can't know for sure
447 	 * what interface the actual traffic for this state will come in on.
448 	 */
449 	if (pd->related_rule)
450 		return (V_pfi_all);
451 
452 	return (k);
453 }
454 
455 #define	STATE_INC_COUNTERS(s)						\
456 	do {								\
457 		struct pf_krule_item *mrm;				\
458 		counter_u64_add(s->rule.ptr->states_cur, 1);		\
459 		counter_u64_add(s->rule.ptr->states_tot, 1);		\
460 		if (s->anchor.ptr != NULL) {				\
461 			counter_u64_add(s->anchor.ptr->states_cur, 1);	\
462 			counter_u64_add(s->anchor.ptr->states_tot, 1);	\
463 		}							\
464 		if (s->nat_rule.ptr != NULL) {				\
465 			counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
466 			counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
467 		}							\
468 		SLIST_FOREACH(mrm, &s->match_rules, entry) {		\
469 			counter_u64_add(mrm->r->states_cur, 1);		\
470 			counter_u64_add(mrm->r->states_tot, 1);		\
471 		}							\
472 	} while (0)
473 
474 #define	STATE_DEC_COUNTERS(s)						\
475 	do {								\
476 		struct pf_krule_item *mrm;				\
477 		if (s->nat_rule.ptr != NULL)				\
478 			counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
479 		if (s->anchor.ptr != NULL)				\
480 			counter_u64_add(s->anchor.ptr->states_cur, -1);	\
481 		counter_u64_add(s->rule.ptr->states_cur, -1);		\
482 		SLIST_FOREACH(mrm, &s->match_rules, entry)		\
483 			counter_u64_add(mrm->r->states_cur, -1);	\
484 	} while (0)
485 
486 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
487 MALLOC_DEFINE(M_PF_RULE_ITEM, "pf_krule_item", "pf(4) rule items");
488 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
489 VNET_DEFINE(struct pf_idhash *, pf_idhash);
490 VNET_DEFINE(struct pf_srchash *, pf_srchash);
491 
492 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
493     "pf(4)");
494 
495 VNET_DEFINE(u_long, pf_hashmask);
496 VNET_DEFINE(u_long, pf_srchashmask);
497 VNET_DEFINE_STATIC(u_long, pf_hashsize);
498 #define V_pf_hashsize	VNET(pf_hashsize)
499 VNET_DEFINE_STATIC(u_long, pf_srchashsize);
500 #define V_pf_srchashsize	VNET(pf_srchashsize)
501 u_long	pf_ioctl_maxcount = 65535;
502 
503 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN,
504     &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable");
505 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_VNET | CTLFLAG_RDTUN,
506     &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable");
507 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN,
508     &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call");
509 
510 VNET_DEFINE(void *, pf_swi_cookie);
511 VNET_DEFINE(struct intr_event *, pf_swi_ie);
512 
513 VNET_DEFINE(uint32_t, pf_hashseed);
514 #define	V_pf_hashseed	VNET(pf_hashseed)
515 
516 static void
pf_sctp_checksum(struct mbuf * m,int off)517 pf_sctp_checksum(struct mbuf *m, int off)
518 {
519 	uint32_t sum = 0;
520 
521 	/* Zero out the checksum, to enable recalculation. */
522 	m_copyback(m, off + offsetof(struct sctphdr, checksum),
523 	    sizeof(sum), (caddr_t)&sum);
524 
525 	sum = sctp_calculate_cksum(m, off);
526 
527 	m_copyback(m, off + offsetof(struct sctphdr, checksum),
528 	    sizeof(sum), (caddr_t)&sum);
529 }
530 
531 int
pf_addr_cmp(struct pf_addr * a,struct pf_addr * b,sa_family_t af)532 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
533 {
534 
535 	switch (af) {
536 #ifdef INET
537 	case AF_INET:
538 		if (a->addr32[0] > b->addr32[0])
539 			return (1);
540 		if (a->addr32[0] < b->addr32[0])
541 			return (-1);
542 		break;
543 #endif /* INET */
544 #ifdef INET6
545 	case AF_INET6:
546 		if (a->addr32[3] > b->addr32[3])
547 			return (1);
548 		if (a->addr32[3] < b->addr32[3])
549 			return (-1);
550 		if (a->addr32[2] > b->addr32[2])
551 			return (1);
552 		if (a->addr32[2] < b->addr32[2])
553 			return (-1);
554 		if (a->addr32[1] > b->addr32[1])
555 			return (1);
556 		if (a->addr32[1] < b->addr32[1])
557 			return (-1);
558 		if (a->addr32[0] > b->addr32[0])
559 			return (1);
560 		if (a->addr32[0] < b->addr32[0])
561 			return (-1);
562 		break;
563 #endif /* INET6 */
564 	default:
565 		panic("%s: unknown address family %u", __func__, af);
566 	}
567 	return (0);
568 }
569 
570 static void
pf_packet_rework_nat(struct mbuf * m,struct pf_pdesc * pd,int off,struct pf_state_key * nk)571 pf_packet_rework_nat(struct mbuf *m, struct pf_pdesc *pd, int off,
572 	struct pf_state_key *nk)
573 {
574 
575 	switch (pd->proto) {
576 	case IPPROTO_TCP: {
577 		struct tcphdr *th = &pd->hdr.tcp;
578 
579 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af))
580 			pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum,
581 			    &th->th_sum, &nk->addr[pd->sidx],
582 			    nk->port[pd->sidx], 0, pd->af);
583 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af))
584 			pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum,
585 			    &th->th_sum, &nk->addr[pd->didx],
586 			    nk->port[pd->didx], 0, pd->af);
587 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
588 		break;
589 	}
590 	case IPPROTO_UDP: {
591 		struct udphdr *uh = &pd->hdr.udp;
592 
593 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af))
594 			pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
595 			    &uh->uh_sum, &nk->addr[pd->sidx],
596 			    nk->port[pd->sidx], 1, pd->af);
597 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af))
598 			pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
599 			    &uh->uh_sum, &nk->addr[pd->didx],
600 			    nk->port[pd->didx], 1, pd->af);
601 		m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
602 		break;
603 	}
604 	case IPPROTO_SCTP: {
605 		struct sctphdr *sh = &pd->hdr.sctp;
606 		uint16_t checksum = 0;
607 
608 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) {
609 			pf_change_ap(m, pd->src, &sh->src_port, pd->ip_sum,
610 			    &checksum, &nk->addr[pd->sidx],
611 			    nk->port[pd->sidx], 1, pd->af);
612 		}
613 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) {
614 			pf_change_ap(m, pd->dst, &sh->dest_port, pd->ip_sum,
615 			    &checksum, &nk->addr[pd->didx],
616 			    nk->port[pd->didx], 1, pd->af);
617 		}
618 
619 		break;
620 	}
621 	case IPPROTO_ICMP: {
622 		struct icmp *ih = &pd->hdr.icmp;
623 
624 		if (nk->port[pd->sidx] != ih->icmp_id) {
625 			pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
626 			    ih->icmp_cksum, ih->icmp_id,
627 			    nk->port[pd->sidx], 0);
628 			ih->icmp_id = nk->port[pd->sidx];
629 			pd->sport = &ih->icmp_id;
630 
631 			m_copyback(m, off, ICMP_MINLEN, (caddr_t)ih);
632 		}
633 		/* FALLTHROUGH */
634 	}
635 	default:
636 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af)) {
637 			switch (pd->af) {
638 			case AF_INET:
639 				pf_change_a(&pd->src->v4.s_addr,
640 				    pd->ip_sum, nk->addr[pd->sidx].v4.s_addr,
641 				    0);
642 				break;
643 			case AF_INET6:
644 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
645 				break;
646 			}
647 		}
648 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af)) {
649 			switch (pd->af) {
650 			case AF_INET:
651 				pf_change_a(&pd->dst->v4.s_addr,
652 				    pd->ip_sum, nk->addr[pd->didx].v4.s_addr,
653 				    0);
654 				break;
655 			case AF_INET6:
656 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
657 				break;
658 			}
659 		}
660 		break;
661 	}
662 }
663 
664 static __inline uint32_t
pf_hashkey(const struct pf_state_key * sk)665 pf_hashkey(const struct pf_state_key *sk)
666 {
667 	uint32_t h;
668 
669 	h = murmur3_32_hash32((const uint32_t *)sk,
670 	    sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
671 	    V_pf_hashseed);
672 
673 	return (h & V_pf_hashmask);
674 }
675 
676 static __inline uint32_t
pf_hashsrc(struct pf_addr * addr,sa_family_t af)677 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
678 {
679 	uint32_t h;
680 
681 	switch (af) {
682 	case AF_INET:
683 		h = murmur3_32_hash32((uint32_t *)&addr->v4,
684 		    sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
685 		break;
686 	case AF_INET6:
687 		h = murmur3_32_hash32((uint32_t *)&addr->v6,
688 		    sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
689 		break;
690 	default:
691 		panic("%s: unknown address family %u", __func__, af);
692 	}
693 
694 	return (h & V_pf_srchashmask);
695 }
696 
697 #ifdef ALTQ
698 static int
pf_state_hash(struct pf_kstate * s)699 pf_state_hash(struct pf_kstate *s)
700 {
701 	u_int32_t hv = (intptr_t)s / sizeof(*s);
702 
703 	hv ^= crc32(&s->src, sizeof(s->src));
704 	hv ^= crc32(&s->dst, sizeof(s->dst));
705 	if (hv == 0)
706 		hv = 1;
707 	return (hv);
708 }
709 #endif
710 
711 static __inline void
pf_set_protostate(struct pf_kstate * s,int which,u_int8_t newstate)712 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate)
713 {
714 	if (which == PF_PEER_DST || which == PF_PEER_BOTH)
715 		s->dst.state = newstate;
716 	if (which == PF_PEER_DST)
717 		return;
718 	if (s->src.state == newstate)
719 		return;
720 	if (s->creatorid == V_pf_status.hostid &&
721 	    s->key[PF_SK_STACK] != NULL &&
722 	    s->key[PF_SK_STACK]->proto == IPPROTO_TCP &&
723 	    !(TCPS_HAVEESTABLISHED(s->src.state) ||
724 	    s->src.state == TCPS_CLOSED) &&
725 	    (TCPS_HAVEESTABLISHED(newstate) || newstate == TCPS_CLOSED))
726 		atomic_add_32(&V_pf_status.states_halfopen, -1);
727 
728 	s->src.state = newstate;
729 }
730 
731 #ifdef INET6
732 void
pf_addrcpy(struct pf_addr * dst,struct pf_addr * src,sa_family_t af)733 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
734 {
735 	switch (af) {
736 #ifdef INET
737 	case AF_INET:
738 		dst->addr32[0] = src->addr32[0];
739 		break;
740 #endif /* INET */
741 	case AF_INET6:
742 		dst->addr32[0] = src->addr32[0];
743 		dst->addr32[1] = src->addr32[1];
744 		dst->addr32[2] = src->addr32[2];
745 		dst->addr32[3] = src->addr32[3];
746 		break;
747 	}
748 }
749 #endif /* INET6 */
750 
751 static void
pf_init_threshold(struct pf_threshold * threshold,u_int32_t limit,u_int32_t seconds)752 pf_init_threshold(struct pf_threshold *threshold,
753     u_int32_t limit, u_int32_t seconds)
754 {
755 	threshold->limit = limit * PF_THRESHOLD_MULT;
756 	threshold->seconds = seconds;
757 	threshold->count = 0;
758 	threshold->last = time_uptime;
759 }
760 
761 static void
pf_add_threshold(struct pf_threshold * threshold)762 pf_add_threshold(struct pf_threshold *threshold)
763 {
764 	u_int32_t t = time_uptime, diff = t - threshold->last;
765 
766 	if (diff >= threshold->seconds)
767 		threshold->count = 0;
768 	else
769 		threshold->count -= threshold->count * diff /
770 		    threshold->seconds;
771 	threshold->count += PF_THRESHOLD_MULT;
772 	threshold->last = t;
773 }
774 
775 static int
pf_check_threshold(struct pf_threshold * threshold)776 pf_check_threshold(struct pf_threshold *threshold)
777 {
778 	return (threshold->count > threshold->limit);
779 }
780 
781 static int
pf_src_connlimit(struct pf_kstate ** state)782 pf_src_connlimit(struct pf_kstate **state)
783 {
784 	struct pf_overload_entry *pfoe;
785 	int bad = 0;
786 
787 	PF_STATE_LOCK_ASSERT(*state);
788 	/*
789 	 * XXXKS: The src node is accessed unlocked!
790 	 * PF_SRC_NODE_LOCK_ASSERT((*state)->src_node);
791 	 */
792 
793 	(*state)->src_node->conn++;
794 	(*state)->src.tcp_est = 1;
795 	pf_add_threshold(&(*state)->src_node->conn_rate);
796 
797 	if ((*state)->rule.ptr->max_src_conn &&
798 	    (*state)->rule.ptr->max_src_conn <
799 	    (*state)->src_node->conn) {
800 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
801 		bad++;
802 	}
803 
804 	if ((*state)->rule.ptr->max_src_conn_rate.limit &&
805 	    pf_check_threshold(&(*state)->src_node->conn_rate)) {
806 		counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
807 		bad++;
808 	}
809 
810 	if (!bad)
811 		return (0);
812 
813 	/* Kill this state. */
814 	(*state)->timeout = PFTM_PURGE;
815 	pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED);
816 
817 	if ((*state)->rule.ptr->overload_tbl == NULL)
818 		return (1);
819 
820 	/* Schedule overloading and flushing task. */
821 	pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
822 	if (pfoe == NULL)
823 		return (1);	/* too bad :( */
824 
825 	bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
826 	pfoe->af = (*state)->key[PF_SK_WIRE]->af;
827 	pfoe->rule = (*state)->rule.ptr;
828 	pfoe->dir = (*state)->direction;
829 	PF_OVERLOADQ_LOCK();
830 	SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
831 	PF_OVERLOADQ_UNLOCK();
832 	taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
833 
834 	return (1);
835 }
836 
837 static void
pf_overload_task(void * v,int pending)838 pf_overload_task(void *v, int pending)
839 {
840 	struct pf_overload_head queue;
841 	struct pfr_addr p;
842 	struct pf_overload_entry *pfoe, *pfoe1;
843 	uint32_t killed = 0;
844 
845 	CURVNET_SET((struct vnet *)v);
846 
847 	PF_OVERLOADQ_LOCK();
848 	queue = V_pf_overloadqueue;
849 	SLIST_INIT(&V_pf_overloadqueue);
850 	PF_OVERLOADQ_UNLOCK();
851 
852 	bzero(&p, sizeof(p));
853 	SLIST_FOREACH(pfoe, &queue, next) {
854 		counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
855 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
856 			printf("%s: blocking address ", __func__);
857 			pf_print_host(&pfoe->addr, 0, pfoe->af);
858 			printf("\n");
859 		}
860 
861 		p.pfra_af = pfoe->af;
862 		switch (pfoe->af) {
863 #ifdef INET
864 		case AF_INET:
865 			p.pfra_net = 32;
866 			p.pfra_ip4addr = pfoe->addr.v4;
867 			break;
868 #endif
869 #ifdef INET6
870 		case AF_INET6:
871 			p.pfra_net = 128;
872 			p.pfra_ip6addr = pfoe->addr.v6;
873 			break;
874 #endif
875 		}
876 
877 		PF_RULES_WLOCK();
878 		pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
879 		PF_RULES_WUNLOCK();
880 	}
881 
882 	/*
883 	 * Remove those entries, that don't need flushing.
884 	 */
885 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
886 		if (pfoe->rule->flush == 0) {
887 			SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
888 			free(pfoe, M_PFTEMP);
889 		} else
890 			counter_u64_add(
891 			    V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
892 
893 	/* If nothing to flush, return. */
894 	if (SLIST_EMPTY(&queue)) {
895 		CURVNET_RESTORE();
896 		return;
897 	}
898 
899 	for (int i = 0; i <= V_pf_hashmask; i++) {
900 		struct pf_idhash *ih = &V_pf_idhash[i];
901 		struct pf_state_key *sk;
902 		struct pf_kstate *s;
903 
904 		PF_HASHROW_LOCK(ih);
905 		LIST_FOREACH(s, &ih->states, entry) {
906 		    sk = s->key[PF_SK_WIRE];
907 		    SLIST_FOREACH(pfoe, &queue, next)
908 			if (sk->af == pfoe->af &&
909 			    ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
910 			    pfoe->rule == s->rule.ptr) &&
911 			    ((pfoe->dir == PF_OUT &&
912 			    PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
913 			    (pfoe->dir == PF_IN &&
914 			    PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
915 				s->timeout = PFTM_PURGE;
916 				pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED);
917 				killed++;
918 			}
919 		}
920 		PF_HASHROW_UNLOCK(ih);
921 	}
922 	SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
923 		free(pfoe, M_PFTEMP);
924 	if (V_pf_status.debug >= PF_DEBUG_MISC)
925 		printf("%s: %u states killed", __func__, killed);
926 
927 	CURVNET_RESTORE();
928 }
929 
930 /*
931  * Can return locked on failure, so that we can consistently
932  * allocate and insert a new one.
933  */
934 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)935 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af,
936 	struct pf_srchash **sh, bool returnlocked)
937 {
938 	struct pf_ksrc_node *n;
939 
940 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
941 
942 	*sh = &V_pf_srchash[pf_hashsrc(src, af)];
943 	PF_HASHROW_LOCK(*sh);
944 	LIST_FOREACH(n, &(*sh)->nodes, entry)
945 		if (n->rule.ptr == rule && n->af == af &&
946 		    ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
947 		    (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
948 			break;
949 
950 	if (n != NULL) {
951 		n->states++;
952 		PF_HASHROW_UNLOCK(*sh);
953 	} else if (returnlocked == false)
954 		PF_HASHROW_UNLOCK(*sh);
955 
956 	return (n);
957 }
958 
959 static void
pf_free_src_node(struct pf_ksrc_node * sn)960 pf_free_src_node(struct pf_ksrc_node *sn)
961 {
962 
963 	for (int i = 0; i < 2; i++) {
964 		counter_u64_free(sn->bytes[i]);
965 		counter_u64_free(sn->packets[i]);
966 	}
967 	uma_zfree(V_pf_sources_z, sn);
968 }
969 
970 static u_short
pf_insert_src_node(struct pf_ksrc_node ** sn,struct pf_krule * rule,struct pf_addr * src,sa_family_t af)971 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule,
972     struct pf_addr *src, sa_family_t af)
973 {
974 	u_short			 reason = 0;
975 	struct pf_srchash	*sh = NULL;
976 
977 	KASSERT((rule->rule_flag & PFRULE_SRCTRACK ||
978 	    rule->rpool.opts & PF_POOL_STICKYADDR),
979 	    ("%s for non-tracking rule %p", __func__, rule));
980 
981 	if (*sn == NULL)
982 		*sn = pf_find_src_node(src, rule, af, &sh, true);
983 
984 	if (*sn == NULL) {
985 		PF_HASHROW_ASSERT(sh);
986 
987 		if (rule->max_src_nodes &&
988 		    counter_u64_fetch(rule->src_nodes) >= rule->max_src_nodes) {
989 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1);
990 			PF_HASHROW_UNLOCK(sh);
991 			reason = PFRES_SRCLIMIT;
992 			goto done;
993 		}
994 
995 		(*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
996 		if ((*sn) == NULL) {
997 			PF_HASHROW_UNLOCK(sh);
998 			reason = PFRES_MEMORY;
999 			goto done;
1000 		}
1001 
1002 		for (int i = 0; i < 2; i++) {
1003 			(*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT);
1004 			(*sn)->packets[i] = counter_u64_alloc(M_NOWAIT);
1005 
1006 			if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) {
1007 				pf_free_src_node(*sn);
1008 				PF_HASHROW_UNLOCK(sh);
1009 				reason = PFRES_MEMORY;
1010 				goto done;
1011 			}
1012 		}
1013 
1014 		pf_init_threshold(&(*sn)->conn_rate,
1015 		    rule->max_src_conn_rate.limit,
1016 		    rule->max_src_conn_rate.seconds);
1017 
1018 		MPASS((*sn)->lock == NULL);
1019 		(*sn)->lock = &sh->lock;
1020 
1021 		(*sn)->af = af;
1022 		(*sn)->rule.ptr = rule;
1023 		PF_ACPY(&(*sn)->addr, src, af);
1024 		LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
1025 		(*sn)->creation = time_uptime;
1026 		(*sn)->ruletype = rule->action;
1027 		(*sn)->states = 1;
1028 		if ((*sn)->rule.ptr != NULL)
1029 			counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
1030 		PF_HASHROW_UNLOCK(sh);
1031 		counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
1032 	} else {
1033 		if (rule->max_src_states &&
1034 		    (*sn)->states >= rule->max_src_states) {
1035 			counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
1036 			    1);
1037 			reason = PFRES_SRCLIMIT;
1038 			goto done;
1039 		}
1040 	}
1041 done:
1042 	return (reason);
1043 }
1044 
1045 void
pf_unlink_src_node(struct pf_ksrc_node * src)1046 pf_unlink_src_node(struct pf_ksrc_node *src)
1047 {
1048 	PF_SRC_NODE_LOCK_ASSERT(src);
1049 
1050 	LIST_REMOVE(src, entry);
1051 	if (src->rule.ptr)
1052 		counter_u64_add(src->rule.ptr->src_nodes, -1);
1053 }
1054 
1055 u_int
pf_free_src_nodes(struct pf_ksrc_node_list * head)1056 pf_free_src_nodes(struct pf_ksrc_node_list *head)
1057 {
1058 	struct pf_ksrc_node *sn, *tmp;
1059 	u_int count = 0;
1060 
1061 	LIST_FOREACH_SAFE(sn, head, entry, tmp) {
1062 		pf_free_src_node(sn);
1063 		count++;
1064 	}
1065 
1066 	counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
1067 
1068 	return (count);
1069 }
1070 
1071 void
pf_mtag_initialize(void)1072 pf_mtag_initialize(void)
1073 {
1074 
1075 	pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
1076 	    sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
1077 	    UMA_ALIGN_PTR, 0);
1078 }
1079 
1080 /* Per-vnet data storage structures initialization. */
1081 void
pf_initialize(void)1082 pf_initialize(void)
1083 {
1084 	struct pf_keyhash	*kh;
1085 	struct pf_idhash	*ih;
1086 	struct pf_srchash	*sh;
1087 	u_int i;
1088 
1089 	if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize))
1090 		V_pf_hashsize = PF_HASHSIZ;
1091 	if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize))
1092 		V_pf_srchashsize = PF_SRCHASHSIZ;
1093 
1094 	V_pf_hashseed = arc4random();
1095 
1096 	/* States and state keys storage. */
1097 	V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate),
1098 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1099 	V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
1100 	uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
1101 	uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
1102 
1103 	V_pf_state_key_z = uma_zcreate("pf state keys",
1104 	    sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
1105 	    UMA_ALIGN_PTR, 0);
1106 
1107 	V_pf_keyhash = mallocarray(V_pf_hashsize, sizeof(struct pf_keyhash),
1108 	    M_PFHASH, M_NOWAIT | M_ZERO);
1109 	V_pf_idhash = mallocarray(V_pf_hashsize, sizeof(struct pf_idhash),
1110 	    M_PFHASH, M_NOWAIT | M_ZERO);
1111 	if (V_pf_keyhash == NULL || V_pf_idhash == NULL) {
1112 		printf("pf: Unable to allocate memory for "
1113 		    "state_hashsize %lu.\n", V_pf_hashsize);
1114 
1115 		free(V_pf_keyhash, M_PFHASH);
1116 		free(V_pf_idhash, M_PFHASH);
1117 
1118 		V_pf_hashsize = PF_HASHSIZ;
1119 		V_pf_keyhash = mallocarray(V_pf_hashsize,
1120 		    sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO);
1121 		V_pf_idhash = mallocarray(V_pf_hashsize,
1122 		    sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO);
1123 	}
1124 
1125 	V_pf_hashmask = V_pf_hashsize - 1;
1126 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
1127 	    i++, kh++, ih++) {
1128 		mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
1129 		mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
1130 	}
1131 
1132 	/* Source nodes. */
1133 	V_pf_sources_z = uma_zcreate("pf source nodes",
1134 	    sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
1135 	    0);
1136 	V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
1137 	uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
1138 	uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
1139 
1140 	V_pf_srchash = mallocarray(V_pf_srchashsize,
1141 	    sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO);
1142 	if (V_pf_srchash == NULL) {
1143 		printf("pf: Unable to allocate memory for "
1144 		    "source_hashsize %lu.\n", V_pf_srchashsize);
1145 
1146 		V_pf_srchashsize = PF_SRCHASHSIZ;
1147 		V_pf_srchash = mallocarray(V_pf_srchashsize,
1148 		    sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO);
1149 	}
1150 
1151 	V_pf_srchashmask = V_pf_srchashsize - 1;
1152 	for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++)
1153 		mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
1154 
1155 	/* ALTQ */
1156 	TAILQ_INIT(&V_pf_altqs[0]);
1157 	TAILQ_INIT(&V_pf_altqs[1]);
1158 	TAILQ_INIT(&V_pf_altqs[2]);
1159 	TAILQ_INIT(&V_pf_altqs[3]);
1160 	TAILQ_INIT(&V_pf_pabuf);
1161 	V_pf_altqs_active = &V_pf_altqs[0];
1162 	V_pf_altq_ifs_active = &V_pf_altqs[1];
1163 	V_pf_altqs_inactive = &V_pf_altqs[2];
1164 	V_pf_altq_ifs_inactive = &V_pf_altqs[3];
1165 
1166 	/* Send & overload+flush queues. */
1167 	STAILQ_INIT(&V_pf_sendqueue);
1168 	SLIST_INIT(&V_pf_overloadqueue);
1169 	TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
1170 
1171 	/* Unlinked, but may be referenced rules. */
1172 	TAILQ_INIT(&V_pf_unlinked_rules);
1173 }
1174 
1175 void
pf_mtag_cleanup(void)1176 pf_mtag_cleanup(void)
1177 {
1178 
1179 	uma_zdestroy(pf_mtag_z);
1180 }
1181 
1182 void
pf_cleanup(void)1183 pf_cleanup(void)
1184 {
1185 	struct pf_keyhash	*kh;
1186 	struct pf_idhash	*ih;
1187 	struct pf_srchash	*sh;
1188 	struct pf_send_entry	*pfse, *next;
1189 	u_int i;
1190 
1191 	for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
1192 	    i++, kh++, ih++) {
1193 		KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
1194 		    __func__));
1195 		KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
1196 		    __func__));
1197 		mtx_destroy(&kh->lock);
1198 		mtx_destroy(&ih->lock);
1199 	}
1200 	free(V_pf_keyhash, M_PFHASH);
1201 	free(V_pf_idhash, M_PFHASH);
1202 
1203 	for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
1204 		KASSERT(LIST_EMPTY(&sh->nodes),
1205 		    ("%s: source node hash not empty", __func__));
1206 		mtx_destroy(&sh->lock);
1207 	}
1208 	free(V_pf_srchash, M_PFHASH);
1209 
1210 	STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
1211 		m_freem(pfse->pfse_m);
1212 		free(pfse, M_PFTEMP);
1213 	}
1214 	MPASS(RB_EMPTY(&V_pf_sctp_endpoints));
1215 
1216 	uma_zdestroy(V_pf_sources_z);
1217 	uma_zdestroy(V_pf_state_z);
1218 	uma_zdestroy(V_pf_state_key_z);
1219 }
1220 
1221 static int
pf_mtag_uminit(void * mem,int size,int how)1222 pf_mtag_uminit(void *mem, int size, int how)
1223 {
1224 	struct m_tag *t;
1225 
1226 	t = (struct m_tag *)mem;
1227 	t->m_tag_cookie = MTAG_ABI_COMPAT;
1228 	t->m_tag_id = PACKET_TAG_PF;
1229 	t->m_tag_len = sizeof(struct pf_mtag);
1230 	t->m_tag_free = pf_mtag_free;
1231 
1232 	return (0);
1233 }
1234 
1235 static void
pf_mtag_free(struct m_tag * t)1236 pf_mtag_free(struct m_tag *t)
1237 {
1238 
1239 	uma_zfree(pf_mtag_z, t);
1240 }
1241 
1242 struct pf_mtag *
pf_get_mtag(struct mbuf * m)1243 pf_get_mtag(struct mbuf *m)
1244 {
1245 	struct m_tag *mtag;
1246 
1247 	if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
1248 		return ((struct pf_mtag *)(mtag + 1));
1249 
1250 	mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
1251 	if (mtag == NULL)
1252 		return (NULL);
1253 	bzero(mtag + 1, sizeof(struct pf_mtag));
1254 	m_tag_prepend(m, mtag);
1255 
1256 	return ((struct pf_mtag *)(mtag + 1));
1257 }
1258 
1259 static int
pf_state_key_attach(struct pf_state_key * skw,struct pf_state_key * sks,struct pf_kstate * s)1260 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
1261     struct pf_kstate *s)
1262 {
1263 	struct pf_keyhash	*khs, *khw, *kh;
1264 	struct pf_state_key	*sk, *cur;
1265 	struct pf_kstate	*si, *olds = NULL;
1266 	int idx;
1267 
1268 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1269 	KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
1270 	KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
1271 
1272 	/*
1273 	 * We need to lock hash slots of both keys. To avoid deadlock
1274 	 * we always lock the slot with lower address first. Unlock order
1275 	 * isn't important.
1276 	 *
1277 	 * We also need to lock ID hash slot before dropping key
1278 	 * locks. On success we return with ID hash slot locked.
1279 	 */
1280 
1281 	if (skw == sks) {
1282 		khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
1283 		PF_HASHROW_LOCK(khs);
1284 	} else {
1285 		khs = &V_pf_keyhash[pf_hashkey(sks)];
1286 		khw = &V_pf_keyhash[pf_hashkey(skw)];
1287 		if (khs == khw) {
1288 			PF_HASHROW_LOCK(khs);
1289 		} else if (khs < khw) {
1290 			PF_HASHROW_LOCK(khs);
1291 			PF_HASHROW_LOCK(khw);
1292 		} else {
1293 			PF_HASHROW_LOCK(khw);
1294 			PF_HASHROW_LOCK(khs);
1295 		}
1296 	}
1297 
1298 #define	KEYS_UNLOCK()	do {			\
1299 	if (khs != khw) {			\
1300 		PF_HASHROW_UNLOCK(khs);		\
1301 		PF_HASHROW_UNLOCK(khw);		\
1302 	} else					\
1303 		PF_HASHROW_UNLOCK(khs);		\
1304 } while (0)
1305 
1306 	/*
1307 	 * First run: start with wire key.
1308 	 */
1309 	sk = skw;
1310 	kh = khw;
1311 	idx = PF_SK_WIRE;
1312 
1313 	MPASS(s->lock == NULL);
1314 	s->lock = &V_pf_idhash[PF_IDHASH(s)].lock;
1315 
1316 keyattach:
1317 	LIST_FOREACH(cur, &kh->keys, entry)
1318 		if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
1319 			break;
1320 
1321 	if (cur != NULL) {
1322 		/* Key exists. Check for same kif, if none, add to key. */
1323 		TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
1324 			struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
1325 
1326 			PF_HASHROW_LOCK(ih);
1327 			if (si->kif == s->kif &&
1328 			    si->direction == s->direction) {
1329 				if (sk->proto == IPPROTO_TCP &&
1330 				    si->src.state >= TCPS_FIN_WAIT_2 &&
1331 				    si->dst.state >= TCPS_FIN_WAIT_2) {
1332 					/*
1333 					 * New state matches an old >FIN_WAIT_2
1334 					 * state. We can't drop key hash locks,
1335 					 * thus we can't unlink it properly.
1336 					 *
1337 					 * As a workaround we drop it into
1338 					 * TCPS_CLOSED state, schedule purge
1339 					 * ASAP and push it into the very end
1340 					 * of the slot TAILQ, so that it won't
1341 					 * conflict with our new state.
1342 					 */
1343 					pf_set_protostate(si, PF_PEER_BOTH,
1344 					    TCPS_CLOSED);
1345 					si->timeout = PFTM_PURGE;
1346 					olds = si;
1347 				} else {
1348 					if (V_pf_status.debug >= PF_DEBUG_MISC) {
1349 						printf("pf: %s key attach "
1350 						    "failed on %s: ",
1351 						    (idx == PF_SK_WIRE) ?
1352 						    "wire" : "stack",
1353 						    s->kif->pfik_name);
1354 						pf_print_state_parts(s,
1355 						    (idx == PF_SK_WIRE) ?
1356 						    sk : NULL,
1357 						    (idx == PF_SK_STACK) ?
1358 						    sk : NULL);
1359 						printf(", existing: ");
1360 						pf_print_state_parts(si,
1361 						    (idx == PF_SK_WIRE) ?
1362 						    sk : NULL,
1363 						    (idx == PF_SK_STACK) ?
1364 						    sk : NULL);
1365 						printf("\n");
1366 					}
1367 					s->timeout = PFTM_UNLINKED;
1368 					if (idx == PF_SK_STACK)
1369 						/*
1370 						 * Remove the wire key from
1371 						 * the hash. Other threads
1372 						 * can't be referencing it
1373 						 * because we still hold the
1374 						 * hash lock.
1375 						 */
1376 						pf_state_key_detach(s,
1377 						    PF_SK_WIRE);
1378 					PF_HASHROW_UNLOCK(ih);
1379 					KEYS_UNLOCK();
1380 					if (idx == PF_SK_WIRE)
1381 						/*
1382 						 * We've not inserted either key.
1383 						 * Free both.
1384 						 */
1385 						uma_zfree(V_pf_state_key_z, skw);
1386 					if (skw != sks)
1387 						uma_zfree(
1388 						    V_pf_state_key_z,
1389 						    sks);
1390 					return (EEXIST); /* collision! */
1391 				}
1392 			}
1393 			PF_HASHROW_UNLOCK(ih);
1394 		}
1395 		uma_zfree(V_pf_state_key_z, sk);
1396 		s->key[idx] = cur;
1397 	} else {
1398 		LIST_INSERT_HEAD(&kh->keys, sk, entry);
1399 		s->key[idx] = sk;
1400 	}
1401 
1402 stateattach:
1403 	/* List is sorted, if-bound states before floating. */
1404 	if (s->kif == V_pfi_all)
1405 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1406 	else
1407 		TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1408 
1409 	if (olds) {
1410 		TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1411 		TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1412 		    key_list[idx]);
1413 		olds = NULL;
1414 	}
1415 
1416 	/*
1417 	 * Attach done. See how should we (or should not?)
1418 	 * attach a second key.
1419 	 */
1420 	if (sks == skw) {
1421 		s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1422 		idx = PF_SK_STACK;
1423 		sks = NULL;
1424 		goto stateattach;
1425 	} else if (sks != NULL) {
1426 		/*
1427 		 * Continue attaching with stack key.
1428 		 */
1429 		sk = sks;
1430 		kh = khs;
1431 		idx = PF_SK_STACK;
1432 		sks = NULL;
1433 		goto keyattach;
1434 	}
1435 
1436 	PF_STATE_LOCK(s);
1437 	KEYS_UNLOCK();
1438 
1439 	KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1440 	    ("%s failure", __func__));
1441 
1442 	return (0);
1443 #undef	KEYS_UNLOCK
1444 }
1445 
1446 static void
pf_detach_state(struct pf_kstate * s)1447 pf_detach_state(struct pf_kstate *s)
1448 {
1449 	struct pf_state_key *sks = s->key[PF_SK_STACK];
1450 	struct pf_keyhash *kh;
1451 
1452 	MPASS(s->timeout >= PFTM_MAX);
1453 
1454 	pf_sctp_multihome_detach_addr(s);
1455 
1456 	if (sks != NULL) {
1457 		kh = &V_pf_keyhash[pf_hashkey(sks)];
1458 		PF_HASHROW_LOCK(kh);
1459 		if (s->key[PF_SK_STACK] != NULL)
1460 			pf_state_key_detach(s, PF_SK_STACK);
1461 		/*
1462 		 * If both point to same key, then we are done.
1463 		 */
1464 		if (sks == s->key[PF_SK_WIRE]) {
1465 			pf_state_key_detach(s, PF_SK_WIRE);
1466 			PF_HASHROW_UNLOCK(kh);
1467 			return;
1468 		}
1469 		PF_HASHROW_UNLOCK(kh);
1470 	}
1471 
1472 	if (s->key[PF_SK_WIRE] != NULL) {
1473 		kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1474 		PF_HASHROW_LOCK(kh);
1475 		if (s->key[PF_SK_WIRE] != NULL)
1476 			pf_state_key_detach(s, PF_SK_WIRE);
1477 		PF_HASHROW_UNLOCK(kh);
1478 	}
1479 }
1480 
1481 static void
pf_state_key_detach(struct pf_kstate * s,int idx)1482 pf_state_key_detach(struct pf_kstate *s, int idx)
1483 {
1484 	struct pf_state_key *sk = s->key[idx];
1485 #ifdef INVARIANTS
1486 	struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1487 
1488 	PF_HASHROW_ASSERT(kh);
1489 #endif
1490 	TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1491 	s->key[idx] = NULL;
1492 
1493 	if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1494 		LIST_REMOVE(sk, entry);
1495 		uma_zfree(V_pf_state_key_z, sk);
1496 	}
1497 }
1498 
1499 static int
pf_state_key_ctor(void * mem,int size,void * arg,int flags)1500 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1501 {
1502 	struct pf_state_key *sk = mem;
1503 
1504 	bzero(sk, sizeof(struct pf_state_key_cmp));
1505 	TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1506 	TAILQ_INIT(&sk->states[PF_SK_STACK]);
1507 
1508 	return (0);
1509 }
1510 
1511 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)1512 pf_state_key_addr_setup(struct pf_pdesc *pd, struct mbuf *m, int off,
1513     struct pf_state_key_cmp *key, int sidx, struct pf_addr *saddr,
1514     int didx, struct pf_addr *daddr, int multi)
1515 {
1516 #ifdef INET6
1517 	struct nd_neighbor_solicit nd;
1518 	struct pf_addr *target;
1519 	u_short action, reason;
1520 
1521 	if (pd->af == AF_INET || pd->proto != IPPROTO_ICMPV6)
1522 		goto copy;
1523 
1524 	switch (pd->hdr.icmp6.icmp6_type) {
1525 	case ND_NEIGHBOR_SOLICIT:
1526 		if (multi)
1527 			return (-1);
1528 		if (!pf_pull_hdr(m, off, &nd, sizeof(nd), &action, &reason, pd->af))
1529 			return (-1);
1530 		target = (struct pf_addr *)&nd.nd_ns_target;
1531 		daddr = target;
1532 		break;
1533 	case ND_NEIGHBOR_ADVERT:
1534 		if (multi)
1535 			return (-1);
1536 		if (!pf_pull_hdr(m, off, &nd, sizeof(nd), &action, &reason, pd->af))
1537 			return (-1);
1538 		target = (struct pf_addr *)&nd.nd_ns_target;
1539 		saddr = target;
1540 		if (IN6_IS_ADDR_MULTICAST(&pd->dst->v6)) {
1541 			key->addr[didx].addr32[0] = 0;
1542 			key->addr[didx].addr32[1] = 0;
1543 			key->addr[didx].addr32[2] = 0;
1544 			key->addr[didx].addr32[3] = 0;
1545 			daddr = NULL; /* overwritten */
1546 		}
1547 		break;
1548 	default:
1549 		if (multi == PF_ICMP_MULTI_LINK) {
1550 			key->addr[sidx].addr32[0] = IPV6_ADDR_INT32_MLL;
1551 			key->addr[sidx].addr32[1] = 0;
1552 			key->addr[sidx].addr32[2] = 0;
1553 			key->addr[sidx].addr32[3] = IPV6_ADDR_INT32_ONE;
1554 			saddr = NULL; /* overwritten */
1555 		}
1556 	}
1557 copy:
1558 #endif
1559 	if (saddr)
1560 		PF_ACPY(&key->addr[sidx], saddr, pd->af);
1561 	if (daddr)
1562 		PF_ACPY(&key->addr[didx], daddr, pd->af);
1563 
1564 	return (0);
1565 }
1566 
1567 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)1568 pf_state_key_setup(struct pf_pdesc *pd, struct mbuf *m, int off,
1569     struct pf_addr *saddr, struct pf_addr *daddr, u_int16_t sport,
1570     u_int16_t dport)
1571 {
1572 	struct pf_state_key *sk;
1573 
1574 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1575 	if (sk == NULL)
1576 		return (NULL);
1577 
1578 	if (pf_state_key_addr_setup(pd, m, off, (struct pf_state_key_cmp *)sk,
1579 	    pd->sidx, pd->src, pd->didx, pd->dst, 0)) {
1580 		uma_zfree(V_pf_state_key_z, sk);
1581 		return (NULL);
1582 	}
1583 
1584 	sk->port[pd->sidx] = sport;
1585 	sk->port[pd->didx] = dport;
1586 	sk->proto = pd->proto;
1587 	sk->af = pd->af;
1588 
1589 	return (sk);
1590 }
1591 
1592 struct pf_state_key *
pf_state_key_clone(const struct pf_state_key * orig)1593 pf_state_key_clone(const struct pf_state_key *orig)
1594 {
1595 	struct pf_state_key *sk;
1596 
1597 	sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1598 	if (sk == NULL)
1599 		return (NULL);
1600 
1601 	bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1602 
1603 	return (sk);
1604 }
1605 
1606 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)1607 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif,
1608     struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s)
1609 {
1610 	struct pf_idhash *ih;
1611 	struct pf_kstate *cur;
1612 	int error;
1613 
1614 	KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1615 	    ("%s: sks not pristine", __func__));
1616 	KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1617 	    ("%s: skw not pristine", __func__));
1618 	KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1619 
1620 	s->kif = kif;
1621 	s->orig_kif = orig_kif;
1622 
1623 	if (s->id == 0 && s->creatorid == 0) {
1624 		s->id = alloc_unr64(&V_pf_stateid);
1625 		s->id = htobe64(s->id);
1626 		s->creatorid = V_pf_status.hostid;
1627 	}
1628 
1629 	/* Returns with ID locked on success. */
1630 	if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1631 		return (error);
1632 	skw = sks = NULL;
1633 
1634 	ih = &V_pf_idhash[PF_IDHASH(s)];
1635 	PF_HASHROW_ASSERT(ih);
1636 	LIST_FOREACH(cur, &ih->states, entry)
1637 		if (cur->id == s->id && cur->creatorid == s->creatorid)
1638 			break;
1639 
1640 	if (cur != NULL) {
1641 		s->timeout = PFTM_UNLINKED;
1642 		PF_HASHROW_UNLOCK(ih);
1643 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
1644 			printf("pf: state ID collision: "
1645 			    "id: %016llx creatorid: %08x\n",
1646 			    (unsigned long long)be64toh(s->id),
1647 			    ntohl(s->creatorid));
1648 		}
1649 		pf_detach_state(s);
1650 		return (EEXIST);
1651 	}
1652 	LIST_INSERT_HEAD(&ih->states, s, entry);
1653 	/* One for keys, one for ID hash. */
1654 	refcount_init(&s->refs, 2);
1655 
1656 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1657 	if (V_pfsync_insert_state_ptr != NULL)
1658 		V_pfsync_insert_state_ptr(s);
1659 
1660 	/* Returns locked. */
1661 	return (0);
1662 }
1663 
1664 /*
1665  * Find state by ID: returns with locked row on success.
1666  */
1667 struct pf_kstate *
pf_find_state_byid(uint64_t id,uint32_t creatorid)1668 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1669 {
1670 	struct pf_idhash *ih;
1671 	struct pf_kstate *s;
1672 
1673 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1674 
1675 	ih = &V_pf_idhash[PF_IDHASHID(id)];
1676 
1677 	PF_HASHROW_LOCK(ih);
1678 	LIST_FOREACH(s, &ih->states, entry)
1679 		if (s->id == id && s->creatorid == creatorid)
1680 			break;
1681 
1682 	if (s == NULL)
1683 		PF_HASHROW_UNLOCK(ih);
1684 
1685 	return (s);
1686 }
1687 
1688 /*
1689  * Find state by key.
1690  * Returns with ID hash slot locked on success.
1691  */
1692 static struct pf_kstate *
pf_find_state(struct pfi_kkif * kif,const struct pf_state_key_cmp * key,u_int dir)1693 pf_find_state(struct pfi_kkif *kif, const struct pf_state_key_cmp *key,
1694     u_int dir)
1695 {
1696 	struct pf_keyhash	*kh;
1697 	struct pf_state_key	*sk;
1698 	struct pf_kstate	*s;
1699 	int idx;
1700 
1701 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1702 
1703 	kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)];
1704 
1705 	PF_HASHROW_LOCK(kh);
1706 	LIST_FOREACH(sk, &kh->keys, entry)
1707 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1708 			break;
1709 	if (sk == NULL) {
1710 		PF_HASHROW_UNLOCK(kh);
1711 		return (NULL);
1712 	}
1713 
1714 	idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1715 
1716 	/* List is sorted, if-bound states before floating ones. */
1717 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1718 		if (s->kif == V_pfi_all || s->kif == kif) {
1719 			PF_STATE_LOCK(s);
1720 			PF_HASHROW_UNLOCK(kh);
1721 			if (__predict_false(s->timeout >= PFTM_MAX)) {
1722 				/*
1723 				 * State is either being processed by
1724 				 * pf_unlink_state() in an other thread, or
1725 				 * is scheduled for immediate expiry.
1726 				 */
1727 				PF_STATE_UNLOCK(s);
1728 				return (NULL);
1729 			}
1730 			return (s);
1731 		}
1732 	PF_HASHROW_UNLOCK(kh);
1733 
1734 	return (NULL);
1735 }
1736 
1737 /*
1738  * Returns with ID hash slot locked on success.
1739  */
1740 struct pf_kstate *
pf_find_state_all(const struct pf_state_key_cmp * key,u_int dir,int * more)1741 pf_find_state_all(const struct pf_state_key_cmp *key, u_int dir, int *more)
1742 {
1743 	struct pf_keyhash	*kh;
1744 	struct pf_state_key	*sk;
1745 	struct pf_kstate	*s, *ret = NULL;
1746 	int			 idx, inout = 0;
1747 
1748 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1749 
1750 	kh = &V_pf_keyhash[pf_hashkey((const struct pf_state_key *)key)];
1751 
1752 	PF_HASHROW_LOCK(kh);
1753 	LIST_FOREACH(sk, &kh->keys, entry)
1754 		if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1755 			break;
1756 	if (sk == NULL) {
1757 		PF_HASHROW_UNLOCK(kh);
1758 		return (NULL);
1759 	}
1760 	switch (dir) {
1761 	case PF_IN:
1762 		idx = PF_SK_WIRE;
1763 		break;
1764 	case PF_OUT:
1765 		idx = PF_SK_STACK;
1766 		break;
1767 	case PF_INOUT:
1768 		idx = PF_SK_WIRE;
1769 		inout = 1;
1770 		break;
1771 	default:
1772 		panic("%s: dir %u", __func__, dir);
1773 	}
1774 second_run:
1775 	TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1776 		if (more == NULL) {
1777 			PF_STATE_LOCK(s);
1778 			PF_HASHROW_UNLOCK(kh);
1779 			return (s);
1780 		}
1781 
1782 		if (ret)
1783 			(*more)++;
1784 		else {
1785 			ret = s;
1786 			PF_STATE_LOCK(s);
1787 		}
1788 	}
1789 	if (inout == 1) {
1790 		inout = 0;
1791 		idx = PF_SK_STACK;
1792 		goto second_run;
1793 	}
1794 	PF_HASHROW_UNLOCK(kh);
1795 
1796 	return (ret);
1797 }
1798 
1799 /*
1800  * FIXME
1801  * This routine is inefficient -- locks the state only to unlock immediately on
1802  * return.
1803  * It is racy -- after the state is unlocked nothing stops other threads from
1804  * removing it.
1805  */
1806 bool
pf_find_state_all_exists(const struct pf_state_key_cmp * key,u_int dir)1807 pf_find_state_all_exists(const struct pf_state_key_cmp *key, u_int dir)
1808 {
1809 	struct pf_kstate *s;
1810 
1811 	s = pf_find_state_all(key, dir, NULL);
1812 	if (s != NULL) {
1813 		PF_STATE_UNLOCK(s);
1814 		return (true);
1815 	}
1816 	return (false);
1817 }
1818 
1819 /* END state table stuff */
1820 
1821 static void
pf_send(struct pf_send_entry * pfse)1822 pf_send(struct pf_send_entry *pfse)
1823 {
1824 
1825 	PF_SENDQ_LOCK();
1826 	STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1827 	PF_SENDQ_UNLOCK();
1828 	swi_sched(V_pf_swi_cookie, 0);
1829 }
1830 
1831 static bool
pf_isforlocal(struct mbuf * m,int af)1832 pf_isforlocal(struct mbuf *m, int af)
1833 {
1834 	switch (af) {
1835 #ifdef INET
1836 	case AF_INET: {
1837 		struct ip *ip = mtod(m, struct ip *);
1838 
1839 		return (in_localip(ip->ip_dst));
1840 	}
1841 #endif
1842 #ifdef INET6
1843 	case AF_INET6: {
1844 		struct ip6_hdr *ip6;
1845 		struct in6_ifaddr *ia;
1846 		ip6 = mtod(m, struct ip6_hdr *);
1847 		ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1848 		if (ia == NULL)
1849 			return (false);
1850 		return (! (ia->ia6_flags & IN6_IFF_NOTREADY));
1851 	}
1852 #endif
1853 	default:
1854 		panic("Unsupported af %d", af);
1855 	}
1856 
1857 	return (false);
1858 }
1859 
1860 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)1861 pf_icmp_mapping(struct pf_pdesc *pd, u_int8_t type,
1862     int *icmp_dir, int *multi, u_int16_t *virtual_id, u_int16_t *virtual_type)
1863 {
1864 	/*
1865 	 * ICMP types marked with PF_OUT are typically responses to
1866 	 * PF_IN, and will match states in the opposite direction.
1867 	 * PF_IN ICMP types need to match a state with that type.
1868 	 */
1869 	*icmp_dir = PF_OUT;
1870 	*multi = PF_ICMP_MULTI_LINK;
1871 	/* Queries (and responses) */
1872 	switch (pd->af) {
1873 #ifdef INET
1874 	case AF_INET:
1875 		switch (type) {
1876 		case ICMP_ECHO:
1877 			*icmp_dir = PF_IN;
1878 		case ICMP_ECHOREPLY:
1879 			*virtual_type = ICMP_ECHO;
1880 			*virtual_id = pd->hdr.icmp.icmp_id;
1881 			break;
1882 
1883 		case ICMP_TSTAMP:
1884 			*icmp_dir = PF_IN;
1885 		case ICMP_TSTAMPREPLY:
1886 			*virtual_type = ICMP_TSTAMP;
1887 			*virtual_id = pd->hdr.icmp.icmp_id;
1888 			break;
1889 
1890 		case ICMP_IREQ:
1891 			*icmp_dir = PF_IN;
1892 		case ICMP_IREQREPLY:
1893 			*virtual_type = ICMP_IREQ;
1894 			*virtual_id = pd->hdr.icmp.icmp_id;
1895 			break;
1896 
1897 		case ICMP_MASKREQ:
1898 			*icmp_dir = PF_IN;
1899 		case ICMP_MASKREPLY:
1900 			*virtual_type = ICMP_MASKREQ;
1901 			*virtual_id = pd->hdr.icmp.icmp_id;
1902 			break;
1903 
1904 		case ICMP_IPV6_WHEREAREYOU:
1905 			*icmp_dir = PF_IN;
1906 		case ICMP_IPV6_IAMHERE:
1907 			*virtual_type = ICMP_IPV6_WHEREAREYOU;
1908 			*virtual_id = 0; /* Nothing sane to match on! */
1909 			break;
1910 
1911 		case ICMP_MOBILE_REGREQUEST:
1912 			*icmp_dir = PF_IN;
1913 		case ICMP_MOBILE_REGREPLY:
1914 			*virtual_type = ICMP_MOBILE_REGREQUEST;
1915 			*virtual_id = 0; /* Nothing sane to match on! */
1916 			break;
1917 
1918 		case ICMP_ROUTERSOLICIT:
1919 			*icmp_dir = PF_IN;
1920 		case ICMP_ROUTERADVERT:
1921 			*virtual_type = ICMP_ROUTERSOLICIT;
1922 			*virtual_id = 0; /* Nothing sane to match on! */
1923 			break;
1924 
1925 		/* These ICMP types map to other connections */
1926 		case ICMP_UNREACH:
1927 		case ICMP_SOURCEQUENCH:
1928 		case ICMP_REDIRECT:
1929 		case ICMP_TIMXCEED:
1930 		case ICMP_PARAMPROB:
1931 			/* These will not be used, but set them anyway */
1932 			*icmp_dir = PF_IN;
1933 			*virtual_type = type;
1934 			*virtual_id = 0;
1935 			HTONS(*virtual_type);
1936 			return (1);  /* These types match to another state */
1937 
1938 		/*
1939 		 * All remaining ICMP types get their own states,
1940 		 * and will only match in one direction.
1941 		 */
1942 		default:
1943 			*icmp_dir = PF_IN;
1944 			*virtual_type = type;
1945 			*virtual_id = 0;
1946 			break;
1947 		}
1948 		break;
1949 #endif /* INET */
1950 #ifdef INET6
1951 	case AF_INET6:
1952 		switch (type) {
1953 		case ICMP6_ECHO_REQUEST:
1954 			*icmp_dir = PF_IN;
1955 		case ICMP6_ECHO_REPLY:
1956 			*virtual_type = ICMP6_ECHO_REQUEST;
1957 			*virtual_id = pd->hdr.icmp6.icmp6_id;
1958 			break;
1959 
1960 		case MLD_LISTENER_QUERY:
1961 		case MLD_LISTENER_REPORT: {
1962 			/*
1963 			 * Listener Report can be sent by clients
1964 			 * without an associated Listener Query.
1965 			 * In addition to that, when Report is sent as a
1966 			 * reply to a Query its source and destination
1967 			 * address are different.
1968 			 */
1969 			*icmp_dir = PF_IN;
1970 			*virtual_type = MLD_LISTENER_QUERY;
1971 			*virtual_id = 0;
1972 			break;
1973 		}
1974 		case MLD_MTRACE:
1975 			*icmp_dir = PF_IN;
1976 		case MLD_MTRACE_RESP:
1977 			*virtual_type = MLD_MTRACE;
1978 			*virtual_id = 0; /* Nothing sane to match on! */
1979 			break;
1980 
1981 		case ND_NEIGHBOR_SOLICIT:
1982 			*icmp_dir = PF_IN;
1983 		case ND_NEIGHBOR_ADVERT: {
1984 			*virtual_type = ND_NEIGHBOR_SOLICIT;
1985 			*virtual_id = 0;
1986 			break;
1987 		}
1988 
1989 		/*
1990 		 * These ICMP types map to other connections.
1991 		 * ND_REDIRECT can't be in this list because the triggering
1992 		 * packet header is optional.
1993 		 */
1994 		case ICMP6_DST_UNREACH:
1995 		case ICMP6_PACKET_TOO_BIG:
1996 		case ICMP6_TIME_EXCEEDED:
1997 		case ICMP6_PARAM_PROB:
1998 			/* These will not be used, but set them anyway */
1999 			*icmp_dir = PF_IN;
2000 			*virtual_type = type;
2001 			*virtual_id = 0;
2002 			HTONS(*virtual_type);
2003 			return (1);  /* These types match to another state */
2004 		/*
2005 		 * All remaining ICMP6 types get their own states,
2006 		 * and will only match in one direction.
2007 		 */
2008 		default:
2009 			*icmp_dir = PF_IN;
2010 			*virtual_type = type;
2011 			*virtual_id = 0;
2012 			break;
2013 		}
2014 		break;
2015 #endif /* INET6 */
2016 	default:
2017 		*icmp_dir = PF_IN;
2018 		*virtual_type = type;
2019 		*virtual_id = 0;
2020 		break;
2021 	}
2022 	HTONS(*virtual_type);
2023 	return (0);  /* These types match to their own state */
2024 }
2025 
2026 void
pf_intr(void * v)2027 pf_intr(void *v)
2028 {
2029 	struct epoch_tracker et;
2030 	struct pf_send_head queue;
2031 	struct pf_send_entry *pfse, *next;
2032 
2033 	CURVNET_SET((struct vnet *)v);
2034 
2035 	PF_SENDQ_LOCK();
2036 	queue = V_pf_sendqueue;
2037 	STAILQ_INIT(&V_pf_sendqueue);
2038 	PF_SENDQ_UNLOCK();
2039 
2040 	NET_EPOCH_ENTER(et);
2041 
2042 	STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
2043 		switch (pfse->pfse_type) {
2044 #ifdef INET
2045 		case PFSE_IP: {
2046 			if (pf_isforlocal(pfse->pfse_m, AF_INET)) {
2047 				pfse->pfse_m->m_flags |= M_SKIP_FIREWALL;
2048 				pfse->pfse_m->m_pkthdr.csum_flags |=
2049 				    CSUM_IP_VALID | CSUM_IP_CHECKED;
2050 				ip_input(pfse->pfse_m);
2051 			} else {
2052 				ip_output(pfse->pfse_m, NULL, NULL, 0, NULL,
2053 				    NULL);
2054 			}
2055 			break;
2056 		}
2057 		case PFSE_ICMP:
2058 			icmp_error(pfse->pfse_m, pfse->icmpopts.type,
2059 			    pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
2060 			break;
2061 #endif /* INET */
2062 #ifdef INET6
2063 		case PFSE_IP6:
2064 			if (pf_isforlocal(pfse->pfse_m, AF_INET6)) {
2065 				pfse->pfse_m->m_flags |= M_SKIP_FIREWALL;
2066 				ip6_input(pfse->pfse_m);
2067 			} else {
2068 				ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL,
2069 				    NULL, NULL);
2070 			}
2071 			break;
2072 		case PFSE_ICMP6:
2073 			icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
2074 			    pfse->icmpopts.code, pfse->icmpopts.mtu);
2075 			break;
2076 #endif /* INET6 */
2077 		default:
2078 			panic("%s: unknown type", __func__);
2079 		}
2080 		free(pfse, M_PFTEMP);
2081 	}
2082 	NET_EPOCH_EXIT(et);
2083 	CURVNET_RESTORE();
2084 }
2085 
2086 #define	pf_purge_thread_period	(hz / 10)
2087 
2088 #ifdef PF_WANT_32_TO_64_COUNTER
2089 static void
pf_status_counter_u64_periodic(void)2090 pf_status_counter_u64_periodic(void)
2091 {
2092 
2093 	PF_RULES_RASSERT();
2094 
2095 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) {
2096 		return;
2097 	}
2098 
2099 	for (int i = 0; i < FCNT_MAX; i++) {
2100 		pf_counter_u64_periodic(&V_pf_status.fcounters[i]);
2101 	}
2102 }
2103 
2104 static void
pf_kif_counter_u64_periodic(void)2105 pf_kif_counter_u64_periodic(void)
2106 {
2107 	struct pfi_kkif *kif;
2108 	size_t r, run;
2109 
2110 	PF_RULES_RASSERT();
2111 
2112 	if (__predict_false(V_pf_allkifcount == 0)) {
2113 		return;
2114 	}
2115 
2116 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) {
2117 		return;
2118 	}
2119 
2120 	run = V_pf_allkifcount / 10;
2121 	if (run < 5)
2122 		run = 5;
2123 
2124 	for (r = 0; r < run; r++) {
2125 		kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist);
2126 		if (kif == NULL) {
2127 			LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist);
2128 			LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist);
2129 			break;
2130 		}
2131 
2132 		LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist);
2133 		LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist);
2134 
2135 		for (int i = 0; i < 2; i++) {
2136 			for (int j = 0; j < 2; j++) {
2137 				for (int k = 0; k < 2; k++) {
2138 					pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]);
2139 					pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]);
2140 				}
2141 			}
2142 		}
2143 	}
2144 }
2145 
2146 static void
pf_rule_counter_u64_periodic(void)2147 pf_rule_counter_u64_periodic(void)
2148 {
2149 	struct pf_krule *rule;
2150 	size_t r, run;
2151 
2152 	PF_RULES_RASSERT();
2153 
2154 	if (__predict_false(V_pf_allrulecount == 0)) {
2155 		return;
2156 	}
2157 
2158 	if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) {
2159 		return;
2160 	}
2161 
2162 	run = V_pf_allrulecount / 10;
2163 	if (run < 5)
2164 		run = 5;
2165 
2166 	for (r = 0; r < run; r++) {
2167 		rule = LIST_NEXT(V_pf_rulemarker, allrulelist);
2168 		if (rule == NULL) {
2169 			LIST_REMOVE(V_pf_rulemarker, allrulelist);
2170 			LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist);
2171 			break;
2172 		}
2173 
2174 		LIST_REMOVE(V_pf_rulemarker, allrulelist);
2175 		LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist);
2176 
2177 		pf_counter_u64_periodic(&rule->evaluations);
2178 		for (int i = 0; i < 2; i++) {
2179 			pf_counter_u64_periodic(&rule->packets[i]);
2180 			pf_counter_u64_periodic(&rule->bytes[i]);
2181 		}
2182 	}
2183 }
2184 
2185 static void
pf_counter_u64_periodic_main(void)2186 pf_counter_u64_periodic_main(void)
2187 {
2188 	PF_RULES_RLOCK_TRACKER;
2189 
2190 	V_pf_counter_periodic_iter++;
2191 
2192 	PF_RULES_RLOCK();
2193 	pf_counter_u64_critical_enter();
2194 	pf_status_counter_u64_periodic();
2195 	pf_kif_counter_u64_periodic();
2196 	pf_rule_counter_u64_periodic();
2197 	pf_counter_u64_critical_exit();
2198 	PF_RULES_RUNLOCK();
2199 }
2200 #else
2201 #define	pf_counter_u64_periodic_main()	do { } while (0)
2202 #endif
2203 
2204 void
pf_purge_thread(void * unused __unused)2205 pf_purge_thread(void *unused __unused)
2206 {
2207 	VNET_ITERATOR_DECL(vnet_iter);
2208 
2209 	sx_xlock(&pf_end_lock);
2210 	while (pf_end_threads == 0) {
2211 		sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period);
2212 
2213 		VNET_LIST_RLOCK();
2214 		VNET_FOREACH(vnet_iter) {
2215 			CURVNET_SET(vnet_iter);
2216 
2217 			/* Wait until V_pf_default_rule is initialized. */
2218 			if (V_pf_vnet_active == 0) {
2219 				CURVNET_RESTORE();
2220 				continue;
2221 			}
2222 
2223 			pf_counter_u64_periodic_main();
2224 
2225 			/*
2226 			 *  Process 1/interval fraction of the state
2227 			 * table every run.
2228 			 */
2229 			V_pf_purge_idx =
2230 			    pf_purge_expired_states(V_pf_purge_idx, V_pf_hashmask /
2231 			    (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
2232 
2233 			/*
2234 			 * Purge other expired types every
2235 			 * PFTM_INTERVAL seconds.
2236 			 */
2237 			if (V_pf_purge_idx == 0) {
2238 				/*
2239 				 * Order is important:
2240 				 * - states and src nodes reference rules
2241 				 * - states and rules reference kifs
2242 				 */
2243 				pf_purge_expired_fragments();
2244 				pf_purge_expired_src_nodes();
2245 				pf_purge_unlinked_rules();
2246 				pfi_kkif_purge();
2247 			}
2248 			CURVNET_RESTORE();
2249 		}
2250 		VNET_LIST_RUNLOCK();
2251 	}
2252 
2253 	pf_end_threads++;
2254 	sx_xunlock(&pf_end_lock);
2255 	kproc_exit(0);
2256 }
2257 
2258 void
pf_unload_vnet_purge(void)2259 pf_unload_vnet_purge(void)
2260 {
2261 
2262 	/*
2263 	 * To cleanse up all kifs and rules we need
2264 	 * two runs: first one clears reference flags,
2265 	 * then pf_purge_expired_states() doesn't
2266 	 * raise them, and then second run frees.
2267 	 */
2268 	pf_purge_unlinked_rules();
2269 	pfi_kkif_purge();
2270 
2271 	/*
2272 	 * Now purge everything.
2273 	 */
2274 	pf_purge_expired_states(0, V_pf_hashmask);
2275 	pf_purge_fragments(UINT_MAX);
2276 	pf_purge_expired_src_nodes();
2277 
2278 	/*
2279 	 * Now all kifs & rules should be unreferenced,
2280 	 * thus should be successfully freed.
2281 	 */
2282 	pf_purge_unlinked_rules();
2283 	pfi_kkif_purge();
2284 }
2285 
2286 u_int32_t
pf_state_expires(const struct pf_kstate * state)2287 pf_state_expires(const struct pf_kstate *state)
2288 {
2289 	u_int32_t	timeout;
2290 	u_int32_t	start;
2291 	u_int32_t	end;
2292 	u_int32_t	states;
2293 
2294 	/* handle all PFTM_* > PFTM_MAX here */
2295 	if (state->timeout == PFTM_PURGE)
2296 		return (time_uptime);
2297 	KASSERT(state->timeout != PFTM_UNLINKED,
2298 	    ("pf_state_expires: timeout == PFTM_UNLINKED"));
2299 	KASSERT((state->timeout < PFTM_MAX),
2300 	    ("pf_state_expires: timeout > PFTM_MAX"));
2301 	timeout = state->rule.ptr->timeout[state->timeout];
2302 	if (!timeout)
2303 		timeout = V_pf_default_rule.timeout[state->timeout];
2304 	start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
2305 	if (start && state->rule.ptr != &V_pf_default_rule) {
2306 		end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
2307 		states = counter_u64_fetch(state->rule.ptr->states_cur);
2308 	} else {
2309 		start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
2310 		end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
2311 		states = V_pf_status.states;
2312 	}
2313 	if (end && states > start && start < end) {
2314 		if (states < end) {
2315 			timeout = (u_int64_t)timeout * (end - states) /
2316 			    (end - start);
2317 			return (state->expire + timeout);
2318 		}
2319 		else
2320 			return (time_uptime);
2321 	}
2322 	return (state->expire + timeout);
2323 }
2324 
2325 void
pf_purge_expired_src_nodes(void)2326 pf_purge_expired_src_nodes(void)
2327 {
2328 	struct pf_ksrc_node_list	 freelist;
2329 	struct pf_srchash	*sh;
2330 	struct pf_ksrc_node	*cur, *next;
2331 	int i;
2332 
2333 	LIST_INIT(&freelist);
2334 	for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
2335 	    PF_HASHROW_LOCK(sh);
2336 	    LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
2337 		if (cur->states == 0 && cur->expire <= time_uptime) {
2338 			pf_unlink_src_node(cur);
2339 			LIST_INSERT_HEAD(&freelist, cur, entry);
2340 		} else if (cur->rule.ptr != NULL)
2341 			cur->rule.ptr->rule_ref |= PFRULE_REFS;
2342 	    PF_HASHROW_UNLOCK(sh);
2343 	}
2344 
2345 	pf_free_src_nodes(&freelist);
2346 
2347 	V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
2348 }
2349 
2350 static void
pf_src_tree_remove_state(struct pf_kstate * s)2351 pf_src_tree_remove_state(struct pf_kstate *s)
2352 {
2353 	struct pf_ksrc_node *sn;
2354 	uint32_t timeout;
2355 
2356 	timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
2357 	    s->rule.ptr->timeout[PFTM_SRC_NODE] :
2358 	    V_pf_default_rule.timeout[PFTM_SRC_NODE];
2359 
2360 	if (s->src_node != NULL) {
2361 		sn = s->src_node;
2362 		PF_SRC_NODE_LOCK(sn);
2363 		if (s->src.tcp_est)
2364 			--sn->conn;
2365 		if (--sn->states == 0)
2366 			sn->expire = time_uptime + timeout;
2367 		PF_SRC_NODE_UNLOCK(sn);
2368 	}
2369 	if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
2370 		sn = s->nat_src_node;
2371 		PF_SRC_NODE_LOCK(sn);
2372 		if (--sn->states == 0)
2373 			sn->expire = time_uptime + timeout;
2374 		PF_SRC_NODE_UNLOCK(sn);
2375 	}
2376 	s->src_node = s->nat_src_node = NULL;
2377 }
2378 
2379 /*
2380  * Unlink and potentilly free a state. Function may be
2381  * called with ID hash row locked, but always returns
2382  * unlocked, since it needs to go through key hash locking.
2383  */
2384 int
pf_unlink_state(struct pf_kstate * s)2385 pf_unlink_state(struct pf_kstate *s)
2386 {
2387 	struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
2388 
2389 	PF_HASHROW_ASSERT(ih);
2390 
2391 	if (s->timeout == PFTM_UNLINKED) {
2392 		/*
2393 		 * State is being processed
2394 		 * by pf_unlink_state() in
2395 		 * an other thread.
2396 		 */
2397 		PF_HASHROW_UNLOCK(ih);
2398 		return (0);	/* XXXGL: undefined actually */
2399 	}
2400 
2401 	if (s->src.state == PF_TCPS_PROXY_DST) {
2402 		/* XXX wire key the right one? */
2403 		pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af,
2404 		    &s->key[PF_SK_WIRE]->addr[1],
2405 		    &s->key[PF_SK_WIRE]->addr[0],
2406 		    s->key[PF_SK_WIRE]->port[1],
2407 		    s->key[PF_SK_WIRE]->port[0],
2408 		    s->src.seqhi, s->src.seqlo + 1,
2409 		    TH_RST|TH_ACK, 0, 0, 0, true, s->tag, 0, s->act.rtableid);
2410 	}
2411 
2412 	LIST_REMOVE(s, entry);
2413 	pf_src_tree_remove_state(s);
2414 
2415 	if (V_pfsync_delete_state_ptr != NULL)
2416 		V_pfsync_delete_state_ptr(s);
2417 
2418 	STATE_DEC_COUNTERS(s);
2419 
2420 	s->timeout = PFTM_UNLINKED;
2421 
2422 	/* Ensure we remove it from the list of halfopen states, if needed. */
2423 	if (s->key[PF_SK_STACK] != NULL &&
2424 	    s->key[PF_SK_STACK]->proto == IPPROTO_TCP)
2425 		pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED);
2426 
2427 	PF_HASHROW_UNLOCK(ih);
2428 
2429 	pf_detach_state(s);
2430 	/* pf_state_insert() initialises refs to 2 */
2431 	return (pf_release_staten(s, 2));
2432 }
2433 
2434 struct pf_kstate *
pf_alloc_state(int flags)2435 pf_alloc_state(int flags)
2436 {
2437 
2438 	return (uma_zalloc(V_pf_state_z, flags | M_ZERO));
2439 }
2440 
2441 void
pf_free_state(struct pf_kstate * cur)2442 pf_free_state(struct pf_kstate *cur)
2443 {
2444 	struct pf_krule_item *ri;
2445 
2446 	KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
2447 	KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
2448 	    cur->timeout));
2449 
2450 	while ((ri = SLIST_FIRST(&cur->match_rules))) {
2451 		SLIST_REMOVE_HEAD(&cur->match_rules, entry);
2452 		free(ri, M_PF_RULE_ITEM);
2453 	}
2454 
2455 	pf_normalize_tcp_cleanup(cur);
2456 	uma_zfree(V_pf_state_z, cur);
2457 	pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
2458 }
2459 
2460 /*
2461  * Called only from pf_purge_thread(), thus serialized.
2462  */
2463 static u_int
pf_purge_expired_states(u_int i,int maxcheck)2464 pf_purge_expired_states(u_int i, int maxcheck)
2465 {
2466 	struct pf_idhash *ih;
2467 	struct pf_kstate *s;
2468 	struct pf_krule_item *mrm;
2469 	size_t count __unused;
2470 
2471 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2472 
2473 	/*
2474 	 * Go through hash and unlink states that expire now.
2475 	 */
2476 	while (maxcheck > 0) {
2477 		count = 0;
2478 		ih = &V_pf_idhash[i];
2479 
2480 		/* only take the lock if we expect to do work */
2481 		if (!LIST_EMPTY(&ih->states)) {
2482 relock:
2483 			PF_HASHROW_LOCK(ih);
2484 			LIST_FOREACH(s, &ih->states, entry) {
2485 				if (pf_state_expires(s) <= time_uptime) {
2486 					V_pf_status.states -=
2487 					    pf_unlink_state(s);
2488 					goto relock;
2489 				}
2490 				s->rule.ptr->rule_ref |= PFRULE_REFS;
2491 				if (s->nat_rule.ptr != NULL)
2492 					s->nat_rule.ptr->rule_ref |= PFRULE_REFS;
2493 				if (s->anchor.ptr != NULL)
2494 					s->anchor.ptr->rule_ref |= PFRULE_REFS;
2495 				s->kif->pfik_flags |= PFI_IFLAG_REFS;
2496 				SLIST_FOREACH(mrm, &s->match_rules, entry)
2497 					mrm->r->rule_ref |= PFRULE_REFS;
2498 				if (s->rt_kif)
2499 					s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
2500 				count++;
2501 			}
2502 			PF_HASHROW_UNLOCK(ih);
2503 		}
2504 
2505 		SDT_PROBE2(pf, purge, state, rowcount, i, count);
2506 
2507 		/* Return when we hit end of hash. */
2508 		if (++i > V_pf_hashmask) {
2509 			V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2510 			return (0);
2511 		}
2512 
2513 		maxcheck--;
2514 	}
2515 
2516 	V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
2517 
2518 	return (i);
2519 }
2520 
2521 static void
pf_purge_unlinked_rules(void)2522 pf_purge_unlinked_rules(void)
2523 {
2524 	struct pf_krulequeue tmpq;
2525 	struct pf_krule *r, *r1;
2526 
2527 	/*
2528 	 * If we have overloading task pending, then we'd
2529 	 * better skip purging this time. There is a tiny
2530 	 * probability that overloading task references
2531 	 * an already unlinked rule.
2532 	 */
2533 	PF_OVERLOADQ_LOCK();
2534 	if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
2535 		PF_OVERLOADQ_UNLOCK();
2536 		return;
2537 	}
2538 	PF_OVERLOADQ_UNLOCK();
2539 
2540 	/*
2541 	 * Do naive mark-and-sweep garbage collecting of old rules.
2542 	 * Reference flag is raised by pf_purge_expired_states()
2543 	 * and pf_purge_expired_src_nodes().
2544 	 *
2545 	 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
2546 	 * use a temporary queue.
2547 	 */
2548 	TAILQ_INIT(&tmpq);
2549 	PF_UNLNKDRULES_LOCK();
2550 	TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
2551 		if (!(r->rule_ref & PFRULE_REFS)) {
2552 			TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
2553 			TAILQ_INSERT_TAIL(&tmpq, r, entries);
2554 		} else
2555 			r->rule_ref &= ~PFRULE_REFS;
2556 	}
2557 	PF_UNLNKDRULES_UNLOCK();
2558 
2559 	if (!TAILQ_EMPTY(&tmpq)) {
2560 		PF_CONFIG_LOCK();
2561 		PF_RULES_WLOCK();
2562 		TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
2563 			TAILQ_REMOVE(&tmpq, r, entries);
2564 			pf_free_rule(r);
2565 		}
2566 		PF_RULES_WUNLOCK();
2567 		PF_CONFIG_UNLOCK();
2568 	}
2569 }
2570 
2571 void
pf_print_host(struct pf_addr * addr,u_int16_t p,sa_family_t af)2572 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
2573 {
2574 	switch (af) {
2575 #ifdef INET
2576 	case AF_INET: {
2577 		u_int32_t a = ntohl(addr->addr32[0]);
2578 		printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
2579 		    (a>>8)&255, a&255);
2580 		if (p) {
2581 			p = ntohs(p);
2582 			printf(":%u", p);
2583 		}
2584 		break;
2585 	}
2586 #endif /* INET */
2587 #ifdef INET6
2588 	case AF_INET6: {
2589 		u_int16_t b;
2590 		u_int8_t i, curstart, curend, maxstart, maxend;
2591 		curstart = curend = maxstart = maxend = 255;
2592 		for (i = 0; i < 8; i++) {
2593 			if (!addr->addr16[i]) {
2594 				if (curstart == 255)
2595 					curstart = i;
2596 				curend = i;
2597 			} else {
2598 				if ((curend - curstart) >
2599 				    (maxend - maxstart)) {
2600 					maxstart = curstart;
2601 					maxend = curend;
2602 				}
2603 				curstart = curend = 255;
2604 			}
2605 		}
2606 		if ((curend - curstart) >
2607 		    (maxend - maxstart)) {
2608 			maxstart = curstart;
2609 			maxend = curend;
2610 		}
2611 		for (i = 0; i < 8; i++) {
2612 			if (i >= maxstart && i <= maxend) {
2613 				if (i == 0)
2614 					printf(":");
2615 				if (i == maxend)
2616 					printf(":");
2617 			} else {
2618 				b = ntohs(addr->addr16[i]);
2619 				printf("%x", b);
2620 				if (i < 7)
2621 					printf(":");
2622 			}
2623 		}
2624 		if (p) {
2625 			p = ntohs(p);
2626 			printf("[%u]", p);
2627 		}
2628 		break;
2629 	}
2630 #endif /* INET6 */
2631 	}
2632 }
2633 
2634 void
pf_print_state(struct pf_kstate * s)2635 pf_print_state(struct pf_kstate *s)
2636 {
2637 	pf_print_state_parts(s, NULL, NULL);
2638 }
2639 
2640 static void
pf_print_state_parts(struct pf_kstate * s,struct pf_state_key * skwp,struct pf_state_key * sksp)2641 pf_print_state_parts(struct pf_kstate *s,
2642     struct pf_state_key *skwp, struct pf_state_key *sksp)
2643 {
2644 	struct pf_state_key *skw, *sks;
2645 	u_int8_t proto, dir;
2646 
2647 	/* Do our best to fill these, but they're skipped if NULL */
2648 	skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
2649 	sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
2650 	proto = skw ? skw->proto : (sks ? sks->proto : 0);
2651 	dir = s ? s->direction : 0;
2652 
2653 	switch (proto) {
2654 	case IPPROTO_IPV4:
2655 		printf("IPv4");
2656 		break;
2657 	case IPPROTO_IPV6:
2658 		printf("IPv6");
2659 		break;
2660 	case IPPROTO_TCP:
2661 		printf("TCP");
2662 		break;
2663 	case IPPROTO_UDP:
2664 		printf("UDP");
2665 		break;
2666 	case IPPROTO_ICMP:
2667 		printf("ICMP");
2668 		break;
2669 	case IPPROTO_ICMPV6:
2670 		printf("ICMPv6");
2671 		break;
2672 	default:
2673 		printf("%u", proto);
2674 		break;
2675 	}
2676 	switch (dir) {
2677 	case PF_IN:
2678 		printf(" in");
2679 		break;
2680 	case PF_OUT:
2681 		printf(" out");
2682 		break;
2683 	}
2684 	if (skw) {
2685 		printf(" wire: ");
2686 		pf_print_host(&skw->addr[0], skw->port[0], skw->af);
2687 		printf(" ");
2688 		pf_print_host(&skw->addr[1], skw->port[1], skw->af);
2689 	}
2690 	if (sks) {
2691 		printf(" stack: ");
2692 		if (sks != skw) {
2693 			pf_print_host(&sks->addr[0], sks->port[0], sks->af);
2694 			printf(" ");
2695 			pf_print_host(&sks->addr[1], sks->port[1], sks->af);
2696 		} else
2697 			printf("-");
2698 	}
2699 	if (s) {
2700 		if (proto == IPPROTO_TCP) {
2701 			printf(" [lo=%u high=%u win=%u modulator=%u",
2702 			    s->src.seqlo, s->src.seqhi,
2703 			    s->src.max_win, s->src.seqdiff);
2704 			if (s->src.wscale && s->dst.wscale)
2705 				printf(" wscale=%u",
2706 				    s->src.wscale & PF_WSCALE_MASK);
2707 			printf("]");
2708 			printf(" [lo=%u high=%u win=%u modulator=%u",
2709 			    s->dst.seqlo, s->dst.seqhi,
2710 			    s->dst.max_win, s->dst.seqdiff);
2711 			if (s->src.wscale && s->dst.wscale)
2712 				printf(" wscale=%u",
2713 				s->dst.wscale & PF_WSCALE_MASK);
2714 			printf("]");
2715 		}
2716 		printf(" %u:%u", s->src.state, s->dst.state);
2717 	}
2718 }
2719 
2720 void
pf_print_flags(u_int8_t f)2721 pf_print_flags(u_int8_t f)
2722 {
2723 	if (f)
2724 		printf(" ");
2725 	if (f & TH_FIN)
2726 		printf("F");
2727 	if (f & TH_SYN)
2728 		printf("S");
2729 	if (f & TH_RST)
2730 		printf("R");
2731 	if (f & TH_PUSH)
2732 		printf("P");
2733 	if (f & TH_ACK)
2734 		printf("A");
2735 	if (f & TH_URG)
2736 		printf("U");
2737 	if (f & TH_ECE)
2738 		printf("E");
2739 	if (f & TH_CWR)
2740 		printf("W");
2741 }
2742 
2743 #define	PF_SET_SKIP_STEPS(i)					\
2744 	do {							\
2745 		while (head[i] != cur) {			\
2746 			head[i]->skip[i].ptr = cur;		\
2747 			head[i] = TAILQ_NEXT(head[i], entries);	\
2748 		}						\
2749 	} while (0)
2750 
2751 void
pf_calc_skip_steps(struct pf_krulequeue * rules)2752 pf_calc_skip_steps(struct pf_krulequeue *rules)
2753 {
2754 	struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT];
2755 	int i;
2756 
2757 	cur = TAILQ_FIRST(rules);
2758 	prev = cur;
2759 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2760 		head[i] = cur;
2761 	while (cur != NULL) {
2762 		if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
2763 			PF_SET_SKIP_STEPS(PF_SKIP_IFP);
2764 		if (cur->direction != prev->direction)
2765 			PF_SET_SKIP_STEPS(PF_SKIP_DIR);
2766 		if (cur->af != prev->af)
2767 			PF_SET_SKIP_STEPS(PF_SKIP_AF);
2768 		if (cur->proto != prev->proto)
2769 			PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
2770 		if (cur->src.neg != prev->src.neg ||
2771 		    pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
2772 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
2773 		if (cur->src.port[0] != prev->src.port[0] ||
2774 		    cur->src.port[1] != prev->src.port[1] ||
2775 		    cur->src.port_op != prev->src.port_op)
2776 			PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
2777 		if (cur->dst.neg != prev->dst.neg ||
2778 		    pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
2779 			PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
2780 		if (cur->dst.port[0] != prev->dst.port[0] ||
2781 		    cur->dst.port[1] != prev->dst.port[1] ||
2782 		    cur->dst.port_op != prev->dst.port_op)
2783 			PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
2784 
2785 		prev = cur;
2786 		cur = TAILQ_NEXT(cur, entries);
2787 	}
2788 	for (i = 0; i < PF_SKIP_COUNT; ++i)
2789 		PF_SET_SKIP_STEPS(i);
2790 }
2791 
2792 int
pf_addr_wrap_neq(struct pf_addr_wrap * aw1,struct pf_addr_wrap * aw2)2793 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
2794 {
2795 	if (aw1->type != aw2->type)
2796 		return (1);
2797 	switch (aw1->type) {
2798 	case PF_ADDR_ADDRMASK:
2799 	case PF_ADDR_RANGE:
2800 		if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
2801 			return (1);
2802 		if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
2803 			return (1);
2804 		return (0);
2805 	case PF_ADDR_DYNIFTL:
2806 		return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
2807 	case PF_ADDR_NOROUTE:
2808 	case PF_ADDR_URPFFAILED:
2809 		return (0);
2810 	case PF_ADDR_TABLE:
2811 		return (aw1->p.tbl != aw2->p.tbl);
2812 	default:
2813 		printf("invalid address type: %d\n", aw1->type);
2814 		return (1);
2815 	}
2816 }
2817 
2818 /**
2819  * Checksum updates are a little complicated because the checksum in the TCP/UDP
2820  * header isn't always a full checksum. In some cases (i.e. output) it's a
2821  * pseudo-header checksum, which is a partial checksum over src/dst IP
2822  * addresses, protocol number and length.
2823  *
2824  * That means we have the following cases:
2825  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2826  *  	checksums, we need to update the checksum whenever we change anything.
2827  *  * Output (i.e. the checksum is a pseudo-header checksum):
2828  *  	x The field being updated is src/dst address or affects the length of
2829  *  	the packet. We need to update the pseudo-header checksum (note that this
2830  *  	checksum is not ones' complement).
2831  *  	x Some other field is being modified (e.g. src/dst port numbers): We
2832  *  	don't have to update anything.
2833  **/
2834 u_int16_t
pf_cksum_fixup(u_int16_t cksum,u_int16_t old,u_int16_t new,u_int8_t udp)2835 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2836 {
2837 	u_int32_t x;
2838 
2839 	x = cksum + old - new;
2840 	x = (x + (x >> 16)) & 0xffff;
2841 
2842 	/* optimise: eliminate a branch when not udp */
2843 	if (udp && cksum == 0x0000)
2844 		return cksum;
2845 	if (udp && x == 0x0000)
2846 		x = 0xffff;
2847 
2848 	return (u_int16_t)(x);
2849 }
2850 
2851 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)2852 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi,
2853     u_int8_t udp)
2854 {
2855 	u_int16_t old = htons(hi ? (*f << 8) : *f);
2856 	u_int16_t new = htons(hi ? ( v << 8) :  v);
2857 
2858 	if (*f == v)
2859 		return;
2860 
2861 	*f = v;
2862 
2863 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2864 		return;
2865 
2866 	*cksum = pf_cksum_fixup(*cksum, old, new, udp);
2867 }
2868 
2869 void
pf_patch_16_unaligned(struct mbuf * m,u_int16_t * cksum,void * f,u_int16_t v,bool hi,u_int8_t udp)2870 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v,
2871     bool hi, u_int8_t udp)
2872 {
2873 	u_int8_t *fb = (u_int8_t *)f;
2874 	u_int8_t *vb = (u_int8_t *)&v;
2875 
2876 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2877 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2878 }
2879 
2880 void
pf_patch_32_unaligned(struct mbuf * m,u_int16_t * cksum,void * f,u_int32_t v,bool hi,u_int8_t udp)2881 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v,
2882     bool hi, u_int8_t udp)
2883 {
2884 	u_int8_t *fb = (u_int8_t *)f;
2885 	u_int8_t *vb = (u_int8_t *)&v;
2886 
2887 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2888 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2889 	pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2890 	pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2891 }
2892 
2893 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)2894 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2895         u_int16_t new, u_int8_t udp)
2896 {
2897 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2898 		return (cksum);
2899 
2900 	return (pf_cksum_fixup(cksum, old, new, udp));
2901 }
2902 
2903 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)2904 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2905         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2906         sa_family_t af)
2907 {
2908 	struct pf_addr	ao;
2909 	u_int16_t	po = *p;
2910 
2911 	PF_ACPY(&ao, a, af);
2912 	PF_ACPY(a, an, af);
2913 
2914 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2915 		*pc = ~*pc;
2916 
2917 	*p = pn;
2918 
2919 	switch (af) {
2920 #ifdef INET
2921 	case AF_INET:
2922 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2923 		    ao.addr16[0], an->addr16[0], 0),
2924 		    ao.addr16[1], an->addr16[1], 0);
2925 		*p = pn;
2926 
2927 		*pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2928 		    ao.addr16[0], an->addr16[0], u),
2929 		    ao.addr16[1], an->addr16[1], u);
2930 
2931 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2932 		break;
2933 #endif /* INET */
2934 #ifdef INET6
2935 	case AF_INET6:
2936 		*pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2937 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2938 		    pf_cksum_fixup(pf_cksum_fixup(*pc,
2939 		    ao.addr16[0], an->addr16[0], u),
2940 		    ao.addr16[1], an->addr16[1], u),
2941 		    ao.addr16[2], an->addr16[2], u),
2942 		    ao.addr16[3], an->addr16[3], u),
2943 		    ao.addr16[4], an->addr16[4], u),
2944 		    ao.addr16[5], an->addr16[5], u),
2945 		    ao.addr16[6], an->addr16[6], u),
2946 		    ao.addr16[7], an->addr16[7], u);
2947 
2948 		*pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2949 		break;
2950 #endif /* INET6 */
2951 	}
2952 
2953 	if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2954 	    CSUM_DELAY_DATA_IPV6)) {
2955 		*pc = ~*pc;
2956 		if (! *pc)
2957 			*pc = 0xffff;
2958 	}
2959 }
2960 
2961 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2962 void
pf_change_a(void * a,u_int16_t * c,u_int32_t an,u_int8_t u)2963 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2964 {
2965 	u_int32_t	ao;
2966 
2967 	memcpy(&ao, a, sizeof(ao));
2968 	memcpy(a, &an, sizeof(u_int32_t));
2969 	*c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2970 	    ao % 65536, an % 65536, u);
2971 }
2972 
2973 void
pf_change_proto_a(struct mbuf * m,void * a,u_int16_t * c,u_int32_t an,u_int8_t udp)2974 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2975 {
2976 	u_int32_t	ao;
2977 
2978 	memcpy(&ao, a, sizeof(ao));
2979 	memcpy(a, &an, sizeof(u_int32_t));
2980 
2981 	*c = pf_proto_cksum_fixup(m,
2982 	    pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2983 	    ao % 65536, an % 65536, udp);
2984 }
2985 
2986 #ifdef INET6
2987 static void
pf_change_a6(struct pf_addr * a,u_int16_t * c,struct pf_addr * an,u_int8_t u)2988 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2989 {
2990 	struct pf_addr	ao;
2991 
2992 	PF_ACPY(&ao, a, AF_INET6);
2993 	PF_ACPY(a, an, AF_INET6);
2994 
2995 	*c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2996 	    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2997 	    pf_cksum_fixup(pf_cksum_fixup(*c,
2998 	    ao.addr16[0], an->addr16[0], u),
2999 	    ao.addr16[1], an->addr16[1], u),
3000 	    ao.addr16[2], an->addr16[2], u),
3001 	    ao.addr16[3], an->addr16[3], u),
3002 	    ao.addr16[4], an->addr16[4], u),
3003 	    ao.addr16[5], an->addr16[5], u),
3004 	    ao.addr16[6], an->addr16[6], u),
3005 	    ao.addr16[7], an->addr16[7], u);
3006 }
3007 #endif /* INET6 */
3008 
3009 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)3010 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
3011     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
3012     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
3013 {
3014 	struct pf_addr	oia, ooa;
3015 
3016 	PF_ACPY(&oia, ia, af);
3017 	if (oa)
3018 		PF_ACPY(&ooa, oa, af);
3019 
3020 	/* Change inner protocol port, fix inner protocol checksum. */
3021 	if (ip != NULL) {
3022 		u_int16_t	oip = *ip;
3023 		u_int32_t	opc;
3024 
3025 		if (pc != NULL)
3026 			opc = *pc;
3027 		*ip = np;
3028 		if (pc != NULL)
3029 			*pc = pf_cksum_fixup(*pc, oip, *ip, u);
3030 		*ic = pf_cksum_fixup(*ic, oip, *ip, 0);
3031 		if (pc != NULL)
3032 			*ic = pf_cksum_fixup(*ic, opc, *pc, 0);
3033 	}
3034 	/* Change inner ip address, fix inner ip and icmp checksums. */
3035 	PF_ACPY(ia, na, af);
3036 	switch (af) {
3037 #ifdef INET
3038 	case AF_INET: {
3039 		u_int32_t	 oh2c = *h2c;
3040 
3041 		*h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
3042 		    oia.addr16[0], ia->addr16[0], 0),
3043 		    oia.addr16[1], ia->addr16[1], 0);
3044 		*ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
3045 		    oia.addr16[0], ia->addr16[0], 0),
3046 		    oia.addr16[1], ia->addr16[1], 0);
3047 		*ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
3048 		break;
3049 	}
3050 #endif /* INET */
3051 #ifdef INET6
3052 	case AF_INET6:
3053 		*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
3054 		    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
3055 		    pf_cksum_fixup(pf_cksum_fixup(*ic,
3056 		    oia.addr16[0], ia->addr16[0], u),
3057 		    oia.addr16[1], ia->addr16[1], u),
3058 		    oia.addr16[2], ia->addr16[2], u),
3059 		    oia.addr16[3], ia->addr16[3], u),
3060 		    oia.addr16[4], ia->addr16[4], u),
3061 		    oia.addr16[5], ia->addr16[5], u),
3062 		    oia.addr16[6], ia->addr16[6], u),
3063 		    oia.addr16[7], ia->addr16[7], u);
3064 		break;
3065 #endif /* INET6 */
3066 	}
3067 	/* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
3068 	if (oa) {
3069 		PF_ACPY(oa, na, af);
3070 		switch (af) {
3071 #ifdef INET
3072 		case AF_INET:
3073 			*hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
3074 			    ooa.addr16[0], oa->addr16[0], 0),
3075 			    ooa.addr16[1], oa->addr16[1], 0);
3076 			break;
3077 #endif /* INET */
3078 #ifdef INET6
3079 		case AF_INET6:
3080 			*ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
3081 			    pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
3082 			    pf_cksum_fixup(pf_cksum_fixup(*ic,
3083 			    ooa.addr16[0], oa->addr16[0], u),
3084 			    ooa.addr16[1], oa->addr16[1], u),
3085 			    ooa.addr16[2], oa->addr16[2], u),
3086 			    ooa.addr16[3], oa->addr16[3], u),
3087 			    ooa.addr16[4], oa->addr16[4], u),
3088 			    ooa.addr16[5], oa->addr16[5], u),
3089 			    ooa.addr16[6], oa->addr16[6], u),
3090 			    ooa.addr16[7], oa->addr16[7], u);
3091 			break;
3092 #endif /* INET6 */
3093 		}
3094 	}
3095 }
3096 
3097 /*
3098  * Need to modulate the sequence numbers in the TCP SACK option
3099  * (credits to Krzysztof Pfaff for report and patch)
3100  */
3101 static int
pf_modulate_sack(struct mbuf * m,int off,struct pf_pdesc * pd,struct tcphdr * th,struct pf_state_peer * dst)3102 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
3103     struct tcphdr *th, struct pf_state_peer *dst)
3104 {
3105 	int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
3106 	u_int8_t opts[TCP_MAXOLEN], *opt = opts;
3107 	int copyback = 0, i, olen;
3108 	struct sackblk sack;
3109 
3110 #define	TCPOLEN_SACKLEN	(TCPOLEN_SACK + 2)
3111 	if (hlen < TCPOLEN_SACKLEN ||
3112 	    !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
3113 		return 0;
3114 
3115 	while (hlen >= TCPOLEN_SACKLEN) {
3116 		size_t startoff = opt - opts;
3117 		olen = opt[1];
3118 		switch (*opt) {
3119 		case TCPOPT_EOL:	/* FALLTHROUGH */
3120 		case TCPOPT_NOP:
3121 			opt++;
3122 			hlen--;
3123 			break;
3124 		case TCPOPT_SACK:
3125 			if (olen > hlen)
3126 				olen = hlen;
3127 			if (olen >= TCPOLEN_SACKLEN) {
3128 				for (i = 2; i + TCPOLEN_SACK <= olen;
3129 				    i += TCPOLEN_SACK) {
3130 					memcpy(&sack, &opt[i], sizeof(sack));
3131 					pf_patch_32_unaligned(m,
3132 					    &th->th_sum, &sack.start,
3133 					    htonl(ntohl(sack.start) - dst->seqdiff),
3134 					    PF_ALGNMNT(startoff),
3135 					    0);
3136 					pf_patch_32_unaligned(m, &th->th_sum,
3137 					    &sack.end,
3138 					    htonl(ntohl(sack.end) - dst->seqdiff),
3139 					    PF_ALGNMNT(startoff),
3140 					    0);
3141 					memcpy(&opt[i], &sack, sizeof(sack));
3142 				}
3143 				copyback = 1;
3144 			}
3145 			/* FALLTHROUGH */
3146 		default:
3147 			if (olen < 2)
3148 				olen = 2;
3149 			hlen -= olen;
3150 			opt += olen;
3151 		}
3152 	}
3153 
3154 	if (copyback)
3155 		m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
3156 	return (copyback);
3157 }
3158 
3159 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)3160 pf_build_tcp(const struct pf_krule *r, sa_family_t af,
3161     const struct pf_addr *saddr, const struct pf_addr *daddr,
3162     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
3163     u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl,
3164     bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid)
3165 {
3166 	struct mbuf	*m;
3167 	int		 len, tlen;
3168 #ifdef INET
3169 	struct ip	*h = NULL;
3170 #endif /* INET */
3171 #ifdef INET6
3172 	struct ip6_hdr	*h6 = NULL;
3173 #endif /* INET6 */
3174 	struct tcphdr	*th;
3175 	char		*opt;
3176 	struct pf_mtag  *pf_mtag;
3177 
3178 	len = 0;
3179 	th = NULL;
3180 
3181 	/* maximum segment size tcp option */
3182 	tlen = sizeof(struct tcphdr);
3183 	if (mss)
3184 		tlen += 4;
3185 
3186 	switch (af) {
3187 #ifdef INET
3188 	case AF_INET:
3189 		len = sizeof(struct ip) + tlen;
3190 		break;
3191 #endif /* INET */
3192 #ifdef INET6
3193 	case AF_INET6:
3194 		len = sizeof(struct ip6_hdr) + tlen;
3195 		break;
3196 #endif /* INET6 */
3197 	default:
3198 		panic("%s: unsupported af %d", __func__, af);
3199 	}
3200 
3201 	m = m_gethdr(M_NOWAIT, MT_DATA);
3202 	if (m == NULL)
3203 		return (NULL);
3204 
3205 #ifdef MAC
3206 	mac_netinet_firewall_send(m);
3207 #endif
3208 	if ((pf_mtag = pf_get_mtag(m)) == NULL) {
3209 		m_freem(m);
3210 		return (NULL);
3211 	}
3212 	if (skip_firewall)
3213 		m->m_flags |= M_SKIP_FIREWALL;
3214 	pf_mtag->tag = mtag_tag;
3215 	pf_mtag->flags = mtag_flags;
3216 
3217 	if (rtableid >= 0)
3218 		M_SETFIB(m, rtableid);
3219 
3220 #ifdef ALTQ
3221 	if (r != NULL && r->qid) {
3222 		pf_mtag->qid = r->qid;
3223 
3224 		/* add hints for ecn */
3225 		pf_mtag->hdr = mtod(m, struct ip *);
3226 	}
3227 #endif /* ALTQ */
3228 	m->m_data += max_linkhdr;
3229 	m->m_pkthdr.len = m->m_len = len;
3230 	/* The rest of the stack assumes a rcvif, so provide one.
3231 	 * This is a locally generated packet, so .. close enough. */
3232 	m->m_pkthdr.rcvif = V_loif;
3233 	bzero(m->m_data, len);
3234 	switch (af) {
3235 #ifdef INET
3236 	case AF_INET:
3237 		h = mtod(m, struct ip *);
3238 
3239 		/* IP header fields included in the TCP checksum */
3240 		h->ip_p = IPPROTO_TCP;
3241 		h->ip_len = htons(tlen);
3242 		h->ip_src.s_addr = saddr->v4.s_addr;
3243 		h->ip_dst.s_addr = daddr->v4.s_addr;
3244 
3245 		th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
3246 		break;
3247 #endif /* INET */
3248 #ifdef INET6
3249 	case AF_INET6:
3250 		h6 = mtod(m, struct ip6_hdr *);
3251 
3252 		/* IP header fields included in the TCP checksum */
3253 		h6->ip6_nxt = IPPROTO_TCP;
3254 		h6->ip6_plen = htons(tlen);
3255 		memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
3256 		memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
3257 
3258 		th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
3259 		break;
3260 #endif /* INET6 */
3261 	}
3262 
3263 	/* TCP header */
3264 	th->th_sport = sport;
3265 	th->th_dport = dport;
3266 	th->th_seq = htonl(seq);
3267 	th->th_ack = htonl(ack);
3268 	th->th_off = tlen >> 2;
3269 	th->th_flags = tcp_flags;
3270 	th->th_win = htons(win);
3271 
3272 	if (mss) {
3273 		opt = (char *)(th + 1);
3274 		opt[0] = TCPOPT_MAXSEG;
3275 		opt[1] = 4;
3276 		HTONS(mss);
3277 		bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
3278 	}
3279 
3280 	switch (af) {
3281 #ifdef INET
3282 	case AF_INET:
3283 		/* TCP checksum */
3284 		th->th_sum = in_cksum(m, len);
3285 
3286 		/* Finish the IP header */
3287 		h->ip_v = 4;
3288 		h->ip_hl = sizeof(*h) >> 2;
3289 		h->ip_tos = IPTOS_LOWDELAY;
3290 		h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
3291 		h->ip_len = htons(len);
3292 		h->ip_ttl = ttl ? ttl : V_ip_defttl;
3293 		h->ip_sum = 0;
3294 		break;
3295 #endif /* INET */
3296 #ifdef INET6
3297 	case AF_INET6:
3298 		/* TCP checksum */
3299 		th->th_sum = in6_cksum(m, IPPROTO_TCP,
3300 		    sizeof(struct ip6_hdr), tlen);
3301 
3302 		h6->ip6_vfc |= IPV6_VERSION;
3303 		h6->ip6_hlim = IPV6_DEFHLIM;
3304 		break;
3305 #endif /* INET6 */
3306 	}
3307 
3308 	return (m);
3309 }
3310 
3311 static void
pf_send_sctp_abort(sa_family_t af,struct pf_pdesc * pd,uint8_t ttl,int rtableid)3312 pf_send_sctp_abort(sa_family_t af, struct pf_pdesc *pd,
3313     uint8_t ttl, int rtableid)
3314 {
3315 	struct mbuf		*m;
3316 #ifdef INET
3317 	struct ip		*h = NULL;
3318 #endif /* INET */
3319 #ifdef INET6
3320 	struct ip6_hdr		*h6 = NULL;
3321 #endif /* INET6 */
3322 	struct sctphdr		*hdr;
3323 	struct sctp_chunkhdr	*chunk;
3324 	struct pf_send_entry	*pfse;
3325 	int			 off = 0;
3326 
3327 	MPASS(af == pd->af);
3328 
3329 	m = m_gethdr(M_NOWAIT, MT_DATA);
3330 	if (m == NULL)
3331 		return;
3332 
3333 	m->m_data += max_linkhdr;
3334 	m->m_flags |= M_SKIP_FIREWALL;
3335 	/* The rest of the stack assumes a rcvif, so provide one.
3336 	 * This is a locally generated packet, so .. close enough. */
3337 	m->m_pkthdr.rcvif = V_loif;
3338 
3339 	/* IPv4|6 header */
3340 	switch (af) {
3341 #ifdef INET
3342 	case AF_INET:
3343 		bzero(m->m_data, sizeof(struct ip) + sizeof(*hdr) + sizeof(*chunk));
3344 
3345 		h = mtod(m, struct ip *);
3346 
3347 		/* IP header fields included in the TCP checksum */
3348 
3349 		h->ip_p = IPPROTO_SCTP;
3350 		h->ip_len = htons(sizeof(*h) + sizeof(*hdr) + sizeof(*chunk));
3351 		h->ip_ttl = ttl ? ttl : V_ip_defttl;
3352 		h->ip_src = pd->dst->v4;
3353 		h->ip_dst = pd->src->v4;
3354 
3355 		off += sizeof(struct ip);
3356 		break;
3357 #endif /* INET */
3358 #ifdef INET6
3359 	case AF_INET6:
3360 		bzero(m->m_data, sizeof(struct ip6_hdr) + sizeof(*hdr) + sizeof(*chunk));
3361 
3362 		h6 = mtod(m, struct ip6_hdr *);
3363 
3364 		/* IP header fields included in the TCP checksum */
3365 		h6->ip6_vfc |= IPV6_VERSION;
3366 		h6->ip6_nxt = IPPROTO_SCTP;
3367 		h6->ip6_plen = htons(sizeof(*h6) + sizeof(*hdr) + sizeof(*chunk));
3368 		h6->ip6_hlim = ttl ? ttl : V_ip6_defhlim;
3369 		memcpy(&h6->ip6_src, &pd->dst->v6, sizeof(struct in6_addr));
3370 		memcpy(&h6->ip6_dst, &pd->src->v6, sizeof(struct in6_addr));
3371 
3372 		off += sizeof(struct ip6_hdr);
3373 		break;
3374 #endif /* INET6 */
3375 	}
3376 
3377 	/* SCTP header */
3378 	hdr = mtodo(m, off);
3379 
3380 	hdr->src_port = pd->hdr.sctp.dest_port;
3381 	hdr->dest_port = pd->hdr.sctp.src_port;
3382 	hdr->v_tag = pd->sctp_initiate_tag;
3383 	hdr->checksum = 0;
3384 
3385 	/* Abort chunk. */
3386 	off += sizeof(struct sctphdr);
3387 	chunk = mtodo(m, off);
3388 
3389 	chunk->chunk_type = SCTP_ABORT_ASSOCIATION;
3390 	chunk->chunk_length = htons(sizeof(*chunk));
3391 
3392 	/* SCTP checksum */
3393 	off += sizeof(*chunk);
3394 	m->m_pkthdr.len = m->m_len = off;
3395 
3396 	pf_sctp_checksum(m, off - sizeof(*hdr) - sizeof(*chunk));;
3397 
3398 	if (rtableid >= 0)
3399 		M_SETFIB(m, rtableid);
3400 
3401 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3402 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3403 	if (pfse == NULL) {
3404 		m_freem(m);
3405 		return;
3406 	}
3407 
3408 	switch (af) {
3409 #ifdef INET
3410 	case AF_INET:
3411 		pfse->pfse_type = PFSE_IP;
3412 		break;
3413 #endif /* INET */
3414 #ifdef INET6
3415 	case AF_INET6:
3416 		pfse->pfse_type = PFSE_IP6;
3417 		break;
3418 #endif /* INET6 */
3419 	}
3420 
3421 	pfse->pfse_m = m;
3422 	pf_send(pfse);
3423 }
3424 
3425 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)3426 pf_send_tcp(const struct pf_krule *r, sa_family_t af,
3427     const struct pf_addr *saddr, const struct pf_addr *daddr,
3428     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
3429     u_int8_t tcp_flags, u_int16_t win, u_int16_t mss, u_int8_t ttl,
3430     bool skip_firewall, u_int16_t mtag_tag, u_int16_t mtag_flags, int rtableid)
3431 {
3432 	struct pf_send_entry *pfse;
3433 	struct mbuf	*m;
3434 
3435 	m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, tcp_flags,
3436 	    win, mss, ttl, skip_firewall, mtag_tag, mtag_flags, rtableid);
3437 	if (m == NULL)
3438 		return;
3439 
3440 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3441 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3442 	if (pfse == NULL) {
3443 		m_freem(m);
3444 		return;
3445 	}
3446 
3447 	switch (af) {
3448 #ifdef INET
3449 	case AF_INET:
3450 		pfse->pfse_type = PFSE_IP;
3451 		break;
3452 #endif /* INET */
3453 #ifdef INET6
3454 	case AF_INET6:
3455 		pfse->pfse_type = PFSE_IP6;
3456 		break;
3457 #endif /* INET6 */
3458 	}
3459 
3460 	pfse->pfse_m = m;
3461 	pf_send(pfse);
3462 }
3463 
3464 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)3465 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd,
3466     struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th,
3467     struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
3468     u_short *reason, int rtableid)
3469 {
3470 	struct pf_addr	* const saddr = pd->src;
3471 	struct pf_addr	* const daddr = pd->dst;
3472 	sa_family_t	 af = pd->af;
3473 
3474 	/* undo NAT changes, if they have taken place */
3475 	if (nr != NULL) {
3476 		PF_ACPY(saddr, &pd->osrc, pd->af);
3477 		PF_ACPY(daddr, &pd->odst, pd->af);
3478 		if (pd->sport)
3479 			*pd->sport = sk->port[pd->sidx];
3480 		if (pd->dport)
3481 			*pd->dport = sk->port[pd->didx];
3482 		if (pd->proto_sum)
3483 			*pd->proto_sum = bproto_sum;
3484 		if (pd->ip_sum)
3485 			*pd->ip_sum = bip_sum;
3486 		m_copyback(m, off, hdrlen, pd->hdr.any);
3487 	}
3488 	if (pd->proto == IPPROTO_TCP &&
3489 	    ((r->rule_flag & PFRULE_RETURNRST) ||
3490 	    (r->rule_flag & PFRULE_RETURN)) &&
3491 	    !(th->th_flags & TH_RST)) {
3492 		u_int32_t	 ack = ntohl(th->th_seq) + pd->p_len;
3493 		int		 len = 0;
3494 #ifdef INET
3495 		struct ip	*h4;
3496 #endif
3497 #ifdef INET6
3498 		struct ip6_hdr	*h6;
3499 #endif
3500 
3501 		switch (af) {
3502 #ifdef INET
3503 		case AF_INET:
3504 			h4 = mtod(m, struct ip *);
3505 			len = ntohs(h4->ip_len) - off;
3506 			break;
3507 #endif
3508 #ifdef INET6
3509 		case AF_INET6:
3510 			h6 = mtod(m, struct ip6_hdr *);
3511 			len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3512 			break;
3513 #endif
3514 		}
3515 
3516 		if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3517 			REASON_SET(reason, PFRES_PROTCKSUM);
3518 		else {
3519 			if (th->th_flags & TH_SYN)
3520 				ack++;
3521 			if (th->th_flags & TH_FIN)
3522 				ack++;
3523 			pf_send_tcp(r, af, pd->dst,
3524 				pd->src, th->th_dport, th->th_sport,
3525 				ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3526 				r->return_ttl, true, 0, 0, rtableid);
3527 		}
3528 	} else if (pd->proto == IPPROTO_SCTP &&
3529 	    (r->rule_flag & PFRULE_RETURN)) {
3530 		pf_send_sctp_abort(af, pd, r->return_ttl, rtableid);
3531 	} else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3532 		r->return_icmp)
3533 		pf_send_icmp(m, r->return_icmp >> 8,
3534 			r->return_icmp & 255, af, r, rtableid);
3535 	else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3536 		r->return_icmp6)
3537 		pf_send_icmp(m, r->return_icmp6 >> 8,
3538 			r->return_icmp6 & 255, af, r, rtableid);
3539 }
3540 
3541 static int
pf_match_ieee8021q_pcp(u_int8_t prio,struct mbuf * m)3542 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
3543 {
3544 	struct m_tag *mtag;
3545 	u_int8_t mpcp;
3546 
3547 	mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
3548 	if (mtag == NULL)
3549 		return (0);
3550 
3551 	if (prio == PF_PRIO_ZERO)
3552 		prio = 0;
3553 
3554 	mpcp = *(uint8_t *)(mtag + 1);
3555 
3556 	return (mpcp == prio);
3557 }
3558 
3559 static int
pf_icmp_to_bandlim(uint8_t type)3560 pf_icmp_to_bandlim(uint8_t type)
3561 {
3562 	switch (type) {
3563 		case ICMP_ECHO:
3564 		case ICMP_ECHOREPLY:
3565 			return (BANDLIM_ICMP_ECHO);
3566 		case ICMP_TSTAMP:
3567 		case ICMP_TSTAMPREPLY:
3568 			return (BANDLIM_ICMP_TSTAMP);
3569 		case ICMP_UNREACH:
3570 		default:
3571 			return (BANDLIM_ICMP_UNREACH);
3572 	}
3573 }
3574 
3575 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)3576 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
3577     struct pf_krule *r, int rtableid)
3578 {
3579 	struct pf_send_entry *pfse;
3580 	struct mbuf *m0;
3581 	struct pf_mtag *pf_mtag;
3582 
3583 	/* ICMP packet rate limitation. */
3584 #ifdef INET6
3585 	if (af == AF_INET6) {
3586 		if (icmp6_ratelimit(NULL, type, code))
3587 			return;
3588 	}
3589 #endif
3590 #ifdef INET
3591 	if (af == AF_INET) {
3592 		if (badport_bandlim(pf_icmp_to_bandlim(type)) != 0)
3593 			return;
3594 	}
3595 #endif
3596 
3597 	/* Allocate outgoing queue entry, mbuf and mbuf tag. */
3598 	pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
3599 	if (pfse == NULL)
3600 		return;
3601 
3602 	if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
3603 		free(pfse, M_PFTEMP);
3604 		return;
3605 	}
3606 
3607 	if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
3608 		free(pfse, M_PFTEMP);
3609 		return;
3610 	}
3611 	/* XXX: revisit */
3612 	m0->m_flags |= M_SKIP_FIREWALL;
3613 
3614 	if (rtableid >= 0)
3615 		M_SETFIB(m0, rtableid);
3616 
3617 #ifdef ALTQ
3618 	if (r->qid) {
3619 		pf_mtag->qid = r->qid;
3620 		/* add hints for ecn */
3621 		pf_mtag->hdr = mtod(m0, struct ip *);
3622 	}
3623 #endif /* ALTQ */
3624 
3625 	switch (af) {
3626 #ifdef INET
3627 	case AF_INET:
3628 		pfse->pfse_type = PFSE_ICMP;
3629 		break;
3630 #endif /* INET */
3631 #ifdef INET6
3632 	case AF_INET6:
3633 		pfse->pfse_type = PFSE_ICMP6;
3634 		break;
3635 #endif /* INET6 */
3636 	}
3637 	pfse->pfse_m = m0;
3638 	pfse->icmpopts.type = type;
3639 	pfse->icmpopts.code = code;
3640 	pf_send(pfse);
3641 }
3642 
3643 /*
3644  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
3645  * If n is 0, they match if they are equal. If n is != 0, they match if they
3646  * are different.
3647  */
3648 int
pf_match_addr(u_int8_t n,struct pf_addr * a,struct pf_addr * m,struct pf_addr * b,sa_family_t af)3649 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
3650     struct pf_addr *b, sa_family_t af)
3651 {
3652 	int	match = 0;
3653 
3654 	switch (af) {
3655 #ifdef INET
3656 	case AF_INET:
3657 		if ((a->addr32[0] & m->addr32[0]) ==
3658 		    (b->addr32[0] & m->addr32[0]))
3659 			match++;
3660 		break;
3661 #endif /* INET */
3662 #ifdef INET6
3663 	case AF_INET6:
3664 		if (((a->addr32[0] & m->addr32[0]) ==
3665 		     (b->addr32[0] & m->addr32[0])) &&
3666 		    ((a->addr32[1] & m->addr32[1]) ==
3667 		     (b->addr32[1] & m->addr32[1])) &&
3668 		    ((a->addr32[2] & m->addr32[2]) ==
3669 		     (b->addr32[2] & m->addr32[2])) &&
3670 		    ((a->addr32[3] & m->addr32[3]) ==
3671 		     (b->addr32[3] & m->addr32[3])))
3672 			match++;
3673 		break;
3674 #endif /* INET6 */
3675 	}
3676 	if (match) {
3677 		if (n)
3678 			return (0);
3679 		else
3680 			return (1);
3681 	} else {
3682 		if (n)
3683 			return (1);
3684 		else
3685 			return (0);
3686 	}
3687 }
3688 
3689 /*
3690  * Return 1 if b <= a <= e, otherwise return 0.
3691  */
3692 int
pf_match_addr_range(struct pf_addr * b,struct pf_addr * e,struct pf_addr * a,sa_family_t af)3693 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
3694     struct pf_addr *a, sa_family_t af)
3695 {
3696 	switch (af) {
3697 #ifdef INET
3698 	case AF_INET:
3699 		if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
3700 		    (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
3701 			return (0);
3702 		break;
3703 #endif /* INET */
3704 #ifdef INET6
3705 	case AF_INET6: {
3706 		int	i;
3707 
3708 		/* check a >= b */
3709 		for (i = 0; i < 4; ++i)
3710 			if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
3711 				break;
3712 			else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
3713 				return (0);
3714 		/* check a <= e */
3715 		for (i = 0; i < 4; ++i)
3716 			if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
3717 				break;
3718 			else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
3719 				return (0);
3720 		break;
3721 	}
3722 #endif /* INET6 */
3723 	}
3724 	return (1);
3725 }
3726 
3727 static int
pf_match(u_int8_t op,u_int32_t a1,u_int32_t a2,u_int32_t p)3728 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
3729 {
3730 	switch (op) {
3731 	case PF_OP_IRG:
3732 		return ((p > a1) && (p < a2));
3733 	case PF_OP_XRG:
3734 		return ((p < a1) || (p > a2));
3735 	case PF_OP_RRG:
3736 		return ((p >= a1) && (p <= a2));
3737 	case PF_OP_EQ:
3738 		return (p == a1);
3739 	case PF_OP_NE:
3740 		return (p != a1);
3741 	case PF_OP_LT:
3742 		return (p < a1);
3743 	case PF_OP_LE:
3744 		return (p <= a1);
3745 	case PF_OP_GT:
3746 		return (p > a1);
3747 	case PF_OP_GE:
3748 		return (p >= a1);
3749 	}
3750 	return (0); /* never reached */
3751 }
3752 
3753 int
pf_match_port(u_int8_t op,u_int16_t a1,u_int16_t a2,u_int16_t p)3754 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
3755 {
3756 	NTOHS(a1);
3757 	NTOHS(a2);
3758 	NTOHS(p);
3759 	return (pf_match(op, a1, a2, p));
3760 }
3761 
3762 static int
pf_match_uid(u_int8_t op,uid_t a1,uid_t a2,uid_t u)3763 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
3764 {
3765 	if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
3766 		return (0);
3767 	return (pf_match(op, a1, a2, u));
3768 }
3769 
3770 static int
pf_match_gid(u_int8_t op,gid_t a1,gid_t a2,gid_t g)3771 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
3772 {
3773 	if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
3774 		return (0);
3775 	return (pf_match(op, a1, a2, g));
3776 }
3777 
3778 int
pf_match_tag(struct mbuf * m,struct pf_krule * r,int * tag,int mtag)3779 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag)
3780 {
3781 	if (*tag == -1)
3782 		*tag = mtag;
3783 
3784 	return ((!r->match_tag_not && r->match_tag == *tag) ||
3785 	    (r->match_tag_not && r->match_tag != *tag));
3786 }
3787 
3788 int
pf_tag_packet(struct mbuf * m,struct pf_pdesc * pd,int tag)3789 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
3790 {
3791 
3792 	KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
3793 
3794 	if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
3795 		return (ENOMEM);
3796 
3797 	pd->pf_mtag->tag = tag;
3798 
3799 	return (0);
3800 }
3801 
3802 #define	PF_ANCHOR_STACKSIZE	32
3803 struct pf_kanchor_stackframe {
3804 	struct pf_kruleset	*rs;
3805 	struct pf_krule		*r;	/* XXX: + match bit */
3806 	struct pf_kanchor	*child;
3807 };
3808 
3809 /*
3810  * XXX: We rely on malloc(9) returning pointer aligned addresses.
3811  */
3812 #define	PF_ANCHORSTACK_MATCH	0x00000001
3813 #define	PF_ANCHORSTACK_MASK	(PF_ANCHORSTACK_MATCH)
3814 
3815 #define	PF_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
3816 #define	PF_ANCHOR_RULE(f)	(struct pf_krule *)			\
3817 				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
3818 #define	PF_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 			\
3819 				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
3820 } while (0)
3821 
3822 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)3823 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth,
3824     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
3825     int *match)
3826 {
3827 	struct pf_kanchor_stackframe	*f;
3828 
3829 	PF_RULES_RASSERT();
3830 
3831 	if (match)
3832 		*match = 0;
3833 	if (*depth >= PF_ANCHOR_STACKSIZE) {
3834 		printf("%s: anchor stack overflow on %s\n",
3835 		    __func__, (*r)->anchor->name);
3836 		*r = TAILQ_NEXT(*r, entries);
3837 		return;
3838 	} else if (*depth == 0 && a != NULL)
3839 		*a = *r;
3840 	f = stack + (*depth)++;
3841 	f->rs = *rs;
3842 	f->r = *r;
3843 	if ((*r)->anchor_wildcard) {
3844 		struct pf_kanchor_node *parent = &(*r)->anchor->children;
3845 
3846 		if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) {
3847 			*r = NULL;
3848 			return;
3849 		}
3850 		*rs = &f->child->ruleset;
3851 	} else {
3852 		f->child = NULL;
3853 		*rs = &(*r)->anchor->ruleset;
3854 	}
3855 	*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
3856 }
3857 
3858 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)3859 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth,
3860     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
3861     int *match)
3862 {
3863 	struct pf_kanchor_stackframe	*f;
3864 	struct pf_krule *fr;
3865 	int quick = 0;
3866 
3867 	PF_RULES_RASSERT();
3868 
3869 	do {
3870 		if (*depth <= 0)
3871 			break;
3872 		f = stack + *depth - 1;
3873 		fr = PF_ANCHOR_RULE(f);
3874 		if (f->child != NULL) {
3875 			/*
3876 			 * This block traverses through
3877 			 * a wildcard anchor.
3878 			 */
3879 			if (match != NULL && *match) {
3880 				/*
3881 				 * If any of "*" matched, then
3882 				 * "foo/ *" matched, mark frame
3883 				 * appropriately.
3884 				 */
3885 				PF_ANCHOR_SET_MATCH(f);
3886 				*match = 0;
3887 			}
3888 			f->child = RB_NEXT(pf_kanchor_node,
3889 			    &fr->anchor->children, f->child);
3890 			if (f->child != NULL) {
3891 				*rs = &f->child->ruleset;
3892 				*r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
3893 				if (*r == NULL)
3894 					continue;
3895 				else
3896 					break;
3897 			}
3898 		}
3899 		(*depth)--;
3900 		if (*depth == 0 && a != NULL)
3901 			*a = NULL;
3902 		*rs = f->rs;
3903 		if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
3904 			quick = fr->quick;
3905 		*r = TAILQ_NEXT(fr, entries);
3906 	} while (*r == NULL);
3907 
3908 	return (quick);
3909 }
3910 
3911 struct pf_keth_anchor_stackframe {
3912 	struct pf_keth_ruleset	*rs;
3913 	struct pf_keth_rule	*r;	/* XXX: + match bit */
3914 	struct pf_keth_anchor	*child;
3915 };
3916 
3917 #define	PF_ETH_ANCHOR_MATCH(f)	((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
3918 #define	PF_ETH_ANCHOR_RULE(f)	(struct pf_keth_rule *)			\
3919 				((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
3920 #define	PF_ETH_ANCHOR_SET_MATCH(f)	do { (f)->r = (void *) 		\
3921 				((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
3922 } while (0)
3923 
3924 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)3925 pf_step_into_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth,
3926     struct pf_keth_ruleset **rs, struct pf_keth_rule **r,
3927     struct pf_keth_rule **a, int *match)
3928 {
3929 	struct pf_keth_anchor_stackframe	*f;
3930 
3931 	NET_EPOCH_ASSERT();
3932 
3933 	if (match)
3934 		*match = 0;
3935 	if (*depth >= PF_ANCHOR_STACKSIZE) {
3936 		printf("%s: anchor stack overflow on %s\n",
3937 		    __func__, (*r)->anchor->name);
3938 		*r = TAILQ_NEXT(*r, entries);
3939 		return;
3940 	} else if (*depth == 0 && a != NULL)
3941 		*a = *r;
3942 	f = stack + (*depth)++;
3943 	f->rs = *rs;
3944 	f->r = *r;
3945 	if ((*r)->anchor_wildcard) {
3946 		struct pf_keth_anchor_node *parent = &(*r)->anchor->children;
3947 
3948 		if ((f->child = RB_MIN(pf_keth_anchor_node, parent)) == NULL) {
3949 			*r = NULL;
3950 			return;
3951 		}
3952 		*rs = &f->child->ruleset;
3953 	} else {
3954 		f->child = NULL;
3955 		*rs = &(*r)->anchor->ruleset;
3956 	}
3957 	*r = TAILQ_FIRST((*rs)->active.rules);
3958 }
3959 
3960 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)3961 pf_step_out_of_keth_anchor(struct pf_keth_anchor_stackframe *stack, int *depth,
3962     struct pf_keth_ruleset **rs, struct pf_keth_rule **r,
3963     struct pf_keth_rule **a, int *match)
3964 {
3965 	struct pf_keth_anchor_stackframe	*f;
3966 	struct pf_keth_rule *fr;
3967 	int quick = 0;
3968 
3969 	NET_EPOCH_ASSERT();
3970 
3971 	do {
3972 		if (*depth <= 0)
3973 			break;
3974 		f = stack + *depth - 1;
3975 		fr = PF_ETH_ANCHOR_RULE(f);
3976 		if (f->child != NULL) {
3977 			/*
3978 			 * This block traverses through
3979 			 * a wildcard anchor.
3980 			 */
3981 			if (match != NULL && *match) {
3982 				/*
3983 				 * If any of "*" matched, then
3984 				 * "foo/ *" matched, mark frame
3985 				 * appropriately.
3986 				 */
3987 				PF_ETH_ANCHOR_SET_MATCH(f);
3988 				*match = 0;
3989 			}
3990 			f->child = RB_NEXT(pf_keth_anchor_node,
3991 			    &fr->anchor->children, f->child);
3992 			if (f->child != NULL) {
3993 				*rs = &f->child->ruleset;
3994 				*r = TAILQ_FIRST((*rs)->active.rules);
3995 				if (*r == NULL)
3996 					continue;
3997 				else
3998 					break;
3999 			}
4000 		}
4001 		(*depth)--;
4002 		if (*depth == 0 && a != NULL)
4003 			*a = NULL;
4004 		*rs = f->rs;
4005 		if (PF_ETH_ANCHOR_MATCH(f) || (match != NULL && *match))
4006 			quick = fr->quick;
4007 		*r = TAILQ_NEXT(fr, entries);
4008 	} while (*r == NULL);
4009 
4010 	return (quick);
4011 }
4012 
4013 #ifdef INET6
4014 void
pf_poolmask(struct pf_addr * naddr,struct pf_addr * raddr,struct pf_addr * rmask,struct pf_addr * saddr,sa_family_t af)4015 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
4016     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
4017 {
4018 	switch (af) {
4019 #ifdef INET
4020 	case AF_INET:
4021 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
4022 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
4023 		break;
4024 #endif /* INET */
4025 	case AF_INET6:
4026 		naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
4027 		((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
4028 		naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
4029 		((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
4030 		naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
4031 		((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
4032 		naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
4033 		((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
4034 		break;
4035 	}
4036 }
4037 
4038 void
pf_addr_inc(struct pf_addr * addr,sa_family_t af)4039 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
4040 {
4041 	switch (af) {
4042 #ifdef INET
4043 	case AF_INET:
4044 		addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
4045 		break;
4046 #endif /* INET */
4047 	case AF_INET6:
4048 		if (addr->addr32[3] == 0xffffffff) {
4049 			addr->addr32[3] = 0;
4050 			if (addr->addr32[2] == 0xffffffff) {
4051 				addr->addr32[2] = 0;
4052 				if (addr->addr32[1] == 0xffffffff) {
4053 					addr->addr32[1] = 0;
4054 					addr->addr32[0] =
4055 					    htonl(ntohl(addr->addr32[0]) + 1);
4056 				} else
4057 					addr->addr32[1] =
4058 					    htonl(ntohl(addr->addr32[1]) + 1);
4059 			} else
4060 				addr->addr32[2] =
4061 				    htonl(ntohl(addr->addr32[2]) + 1);
4062 		} else
4063 			addr->addr32[3] =
4064 			    htonl(ntohl(addr->addr32[3]) + 1);
4065 		break;
4066 	}
4067 }
4068 #endif /* INET6 */
4069 
4070 void
pf_rule_to_actions(struct pf_krule * r,struct pf_rule_actions * a)4071 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a)
4072 {
4073 	/*
4074 	 * Modern rules use the same flags in rules as they do in states.
4075 	 */
4076 	a->flags |= (r->scrub_flags & (PFSTATE_NODF|PFSTATE_RANDOMID|
4077 	    PFSTATE_SCRUB_TCP|PFSTATE_SETPRIO));
4078 
4079 	/*
4080 	 * Old-style scrub rules have different flags which need to be translated.
4081 	 */
4082 	if (r->rule_flag & PFRULE_RANDOMID)
4083 		a->flags |= PFSTATE_RANDOMID;
4084 	if (r->scrub_flags & PFSTATE_SETTOS || r->rule_flag & PFRULE_SET_TOS ) {
4085 		a->flags |= PFSTATE_SETTOS;
4086 		a->set_tos = r->set_tos;
4087 	}
4088 
4089 	if (r->qid)
4090 		a->qid = r->qid;
4091 	if (r->pqid)
4092 		a->pqid = r->pqid;
4093 	if (r->rtableid >= 0)
4094 		a->rtableid = r->rtableid;
4095 	a->log |= r->log;
4096 	if (r->min_ttl)
4097 		a->min_ttl = r->min_ttl;
4098 	if (r->max_mss)
4099 		a->max_mss = r->max_mss;
4100 	if (r->dnpipe)
4101 		a->dnpipe = r->dnpipe;
4102 	if (r->dnrpipe)
4103 		a->dnrpipe = r->dnrpipe;
4104 	if (r->dnpipe || r->dnrpipe) {
4105 		if (r->free_flags & PFRULE_DN_IS_PIPE)
4106 			a->flags |= PFSTATE_DN_IS_PIPE;
4107 		else
4108 			a->flags &= ~PFSTATE_DN_IS_PIPE;
4109 	}
4110 	if (r->scrub_flags & PFSTATE_SETPRIO) {
4111 		a->set_prio[0] = r->set_prio[0];
4112 		a->set_prio[1] = r->set_prio[1];
4113 	}
4114 }
4115 
4116 int
pf_socket_lookup(struct pf_pdesc * pd,struct mbuf * m)4117 pf_socket_lookup(struct pf_pdesc *pd, struct mbuf *m)
4118 {
4119 	struct pf_addr		*saddr, *daddr;
4120 	u_int16_t		 sport, dport;
4121 	struct inpcbinfo	*pi;
4122 	struct inpcb		*inp;
4123 
4124 	pd->lookup.uid = UID_MAX;
4125 	pd->lookup.gid = GID_MAX;
4126 
4127 	switch (pd->proto) {
4128 	case IPPROTO_TCP:
4129 		sport = pd->hdr.tcp.th_sport;
4130 		dport = pd->hdr.tcp.th_dport;
4131 		pi = &V_tcbinfo;
4132 		break;
4133 	case IPPROTO_UDP:
4134 		sport = pd->hdr.udp.uh_sport;
4135 		dport = pd->hdr.udp.uh_dport;
4136 		pi = &V_udbinfo;
4137 		break;
4138 	default:
4139 		return (-1);
4140 	}
4141 	if (pd->dir == PF_IN) {
4142 		saddr = pd->src;
4143 		daddr = pd->dst;
4144 	} else {
4145 		u_int16_t	p;
4146 
4147 		p = sport;
4148 		sport = dport;
4149 		dport = p;
4150 		saddr = pd->dst;
4151 		daddr = pd->src;
4152 	}
4153 	switch (pd->af) {
4154 #ifdef INET
4155 	case AF_INET:
4156 		inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
4157 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
4158 		if (inp == NULL) {
4159 			inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
4160 			   daddr->v4, dport, INPLOOKUP_WILDCARD |
4161 			   INPLOOKUP_RLOCKPCB, NULL, m);
4162 			if (inp == NULL)
4163 				return (-1);
4164 		}
4165 		break;
4166 #endif /* INET */
4167 #ifdef INET6
4168 	case AF_INET6:
4169 		inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
4170 		    dport, INPLOOKUP_RLOCKPCB, NULL, m);
4171 		if (inp == NULL) {
4172 			inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
4173 			    &daddr->v6, dport, INPLOOKUP_WILDCARD |
4174 			    INPLOOKUP_RLOCKPCB, NULL, m);
4175 			if (inp == NULL)
4176 				return (-1);
4177 		}
4178 		break;
4179 #endif /* INET6 */
4180 
4181 	default:
4182 		return (-1);
4183 	}
4184 	INP_RLOCK_ASSERT(inp);
4185 	pd->lookup.uid = inp->inp_cred->cr_uid;
4186 	pd->lookup.gid = inp->inp_cred->cr_groups[0];
4187 	INP_RUNLOCK(inp);
4188 
4189 	return (1);
4190 }
4191 
4192 u_int8_t
pf_get_wscale(struct mbuf * m,int off,u_int16_t th_off,sa_family_t af)4193 pf_get_wscale(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_int8_t	 wscale = 0;
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 >= 3) {
4208 		switch (*opt) {
4209 		case TCPOPT_EOL:
4210 		case TCPOPT_NOP:
4211 			++opt;
4212 			--hlen;
4213 			break;
4214 		case TCPOPT_WINDOW:
4215 			wscale = opt[2];
4216 			if (wscale > TCP_MAX_WINSHIFT)
4217 				wscale = TCP_MAX_WINSHIFT;
4218 			wscale |= PF_WSCALE_FLAG;
4219 			/* FALLTHROUGH */
4220 		default:
4221 			optlen = opt[1];
4222 			if (optlen < 2)
4223 				optlen = 2;
4224 			hlen -= optlen;
4225 			opt += optlen;
4226 			break;
4227 		}
4228 	}
4229 	return (wscale);
4230 }
4231 
4232 u_int16_t
pf_get_mss(struct mbuf * m,int off,u_int16_t th_off,sa_family_t af)4233 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
4234 {
4235 	int		 hlen;
4236 	u_int8_t	 hdr[60];
4237 	u_int8_t	*opt, optlen;
4238 	u_int16_t	 mss = V_tcp_mssdflt;
4239 
4240 	hlen = th_off << 2;	/* hlen <= sizeof(hdr) */
4241 	if (hlen <= sizeof(struct tcphdr))
4242 		return (0);
4243 	if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
4244 		return (0);
4245 	opt = hdr + sizeof(struct tcphdr);
4246 	hlen -= sizeof(struct tcphdr);
4247 	while (hlen >= TCPOLEN_MAXSEG) {
4248 		switch (*opt) {
4249 		case TCPOPT_EOL:
4250 		case TCPOPT_NOP:
4251 			++opt;
4252 			--hlen;
4253 			break;
4254 		case TCPOPT_MAXSEG:
4255 			bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
4256 			NTOHS(mss);
4257 			/* FALLTHROUGH */
4258 		default:
4259 			optlen = opt[1];
4260 			if (optlen < 2)
4261 				optlen = 2;
4262 			hlen -= optlen;
4263 			opt += optlen;
4264 			break;
4265 		}
4266 	}
4267 	return (mss);
4268 }
4269 
4270 static u_int16_t
pf_calc_mss(struct pf_addr * addr,sa_family_t af,int rtableid,u_int16_t offer)4271 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
4272 {
4273 	struct nhop_object *nh;
4274 #ifdef INET6
4275 	struct in6_addr		dst6;
4276 	uint32_t		scopeid;
4277 #endif /* INET6 */
4278 	int			 hlen = 0;
4279 	uint16_t		 mss = 0;
4280 
4281 	NET_EPOCH_ASSERT();
4282 
4283 	switch (af) {
4284 #ifdef INET
4285 	case AF_INET:
4286 		hlen = sizeof(struct ip);
4287 		nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0);
4288 		if (nh != NULL)
4289 			mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
4290 		break;
4291 #endif /* INET */
4292 #ifdef INET6
4293 	case AF_INET6:
4294 		hlen = sizeof(struct ip6_hdr);
4295 		in6_splitscope(&addr->v6, &dst6, &scopeid);
4296 		nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0);
4297 		if (nh != NULL)
4298 			mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
4299 		break;
4300 #endif /* INET6 */
4301 	}
4302 
4303 	mss = max(V_tcp_mssdflt, mss);
4304 	mss = min(mss, offer);
4305 	mss = max(mss, 64);		/* sanity - at least max opt space */
4306 	return (mss);
4307 }
4308 
4309 static u_int32_t
pf_tcp_iss(struct pf_pdesc * pd)4310 pf_tcp_iss(struct pf_pdesc *pd)
4311 {
4312 	MD5_CTX ctx;
4313 	u_int32_t digest[4];
4314 
4315 	if (V_pf_tcp_secret_init == 0) {
4316 		arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
4317 		MD5Init(&V_pf_tcp_secret_ctx);
4318 		MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
4319 		    sizeof(V_pf_tcp_secret));
4320 		V_pf_tcp_secret_init = 1;
4321 	}
4322 
4323 	ctx = V_pf_tcp_secret_ctx;
4324 
4325 	MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short));
4326 	MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short));
4327 	if (pd->af == AF_INET6) {
4328 		MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
4329 		MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
4330 	} else {
4331 		MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
4332 		MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
4333 	}
4334 	MD5Final((u_char *)digest, &ctx);
4335 	V_pf_tcp_iss_off += 4096;
4336 #define	ISN_RANDOM_INCREMENT (4096 - 1)
4337 	return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
4338 	    V_pf_tcp_iss_off);
4339 #undef	ISN_RANDOM_INCREMENT
4340 }
4341 
4342 static bool
pf_match_eth_addr(const uint8_t * a,const struct pf_keth_rule_addr * r)4343 pf_match_eth_addr(const uint8_t *a, const struct pf_keth_rule_addr *r)
4344 {
4345 	bool match = true;
4346 
4347 	/* Always matches if not set */
4348 	if (! r->isset)
4349 		return (!r->neg);
4350 
4351 	for (int i = 0; i < ETHER_ADDR_LEN; i++) {
4352 		if ((a[i] & r->mask[i]) != (r->addr[i] & r->mask[i])) {
4353 			match = false;
4354 			break;
4355 		}
4356 	}
4357 
4358 	return (match ^ r->neg);
4359 }
4360 
4361 static int
pf_match_eth_tag(struct mbuf * m,struct pf_keth_rule * r,int * tag,int mtag)4362 pf_match_eth_tag(struct mbuf *m, struct pf_keth_rule *r, int *tag, int mtag)
4363 {
4364 	if (*tag == -1)
4365 		*tag = mtag;
4366 
4367 	return ((!r->match_tag_not && r->match_tag == *tag) ||
4368 	    (r->match_tag_not && r->match_tag != *tag));
4369 }
4370 
4371 static void
pf_bridge_to(struct ifnet * ifp,struct mbuf * m)4372 pf_bridge_to(struct ifnet *ifp, struct mbuf *m)
4373 {
4374 	/* If we don't have the interface drop the packet. */
4375 	if (ifp == NULL) {
4376 		m_freem(m);
4377 		return;
4378 	}
4379 
4380 	switch (ifp->if_type) {
4381 	case IFT_ETHER:
4382 	case IFT_XETHER:
4383 	case IFT_L2VLAN:
4384 	case IFT_BRIDGE:
4385 	case IFT_IEEE8023ADLAG:
4386 		break;
4387 	default:
4388 		m_freem(m);
4389 		return;
4390 	}
4391 
4392 	ifp->if_transmit(ifp, m);
4393 }
4394 
4395 static int
pf_test_eth_rule(int dir,struct pfi_kkif * kif,struct mbuf ** m0)4396 pf_test_eth_rule(int dir, struct pfi_kkif *kif, struct mbuf **m0)
4397 {
4398 #ifdef INET
4399 	struct ip ip;
4400 #endif
4401 #ifdef INET6
4402 	struct ip6_hdr ip6;
4403 #endif
4404 	struct mbuf *m = *m0;
4405 	struct ether_header *e;
4406 	struct pf_keth_rule *r, *rm, *a = NULL;
4407 	struct pf_keth_ruleset *ruleset = NULL;
4408 	struct pf_mtag *mtag;
4409 	struct pf_keth_ruleq *rules;
4410 	struct pf_addr *src = NULL, *dst = NULL;
4411 	struct pfi_kkif *bridge_to;
4412 	sa_family_t af = 0;
4413 	uint16_t proto;
4414 	int asd = 0, match = 0;
4415 	int tag = -1;
4416 	uint8_t action;
4417 	struct pf_keth_anchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
4418 
4419 	MPASS(kif->pfik_ifp->if_vnet == curvnet);
4420 	NET_EPOCH_ASSERT();
4421 
4422 	PF_RULES_RLOCK_TRACKER;
4423 
4424 	SDT_PROBE3(pf, eth, test_rule, entry, dir, kif->pfik_ifp, m);
4425 
4426 	mtag = pf_find_mtag(m);
4427 	if (mtag != NULL && mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
4428 		/* Dummynet re-injects packets after they've
4429 		 * completed their delay. We've already
4430 		 * processed them, so pass unconditionally. */
4431 
4432 		/* But only once. We may see the packet multiple times (e.g.
4433 		 * PFIL_IN/PFIL_OUT). */
4434 		pf_dummynet_flag_remove(m, mtag);
4435 
4436 		return (PF_PASS);
4437 	}
4438 
4439 	e = mtod(m, struct ether_header *);
4440 	proto = ntohs(e->ether_type);
4441 
4442 	switch (proto) {
4443 #ifdef INET
4444 	case ETHERTYPE_IP: {
4445 		if (m_length(m, NULL) < (sizeof(struct ether_header) +
4446 		    sizeof(ip)))
4447 			return (PF_DROP);
4448 
4449 		af = AF_INET;
4450 		m_copydata(m, sizeof(struct ether_header), sizeof(ip),
4451 		    (caddr_t)&ip);
4452 		src = (struct pf_addr *)&ip.ip_src;
4453 		dst = (struct pf_addr *)&ip.ip_dst;
4454 		break;
4455 	}
4456 #endif /* INET */
4457 #ifdef INET6
4458 	case ETHERTYPE_IPV6: {
4459 		if (m_length(m, NULL) < (sizeof(struct ether_header) +
4460 		    sizeof(ip6)))
4461 			return (PF_DROP);
4462 
4463 		af = AF_INET6;
4464 		m_copydata(m, sizeof(struct ether_header), sizeof(ip6),
4465 		    (caddr_t)&ip6);
4466 		src = (struct pf_addr *)&ip6.ip6_src;
4467 		dst = (struct pf_addr *)&ip6.ip6_dst;
4468 		break;
4469 	}
4470 #endif /* INET6 */
4471 	}
4472 
4473 	PF_RULES_RLOCK();
4474 
4475 	ruleset = V_pf_keth;
4476 	rules = atomic_load_ptr(&ruleset->active.rules);
4477 	for (r = TAILQ_FIRST(rules), rm = NULL; r != NULL;) {
4478 		counter_u64_add(r->evaluations, 1);
4479 		SDT_PROBE2(pf, eth, test_rule, test, r->nr, r);
4480 
4481 		if (pfi_kkif_match(r->kif, kif) == r->ifnot) {
4482 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4483 			    "kif");
4484 			r = r->skip[PFE_SKIP_IFP].ptr;
4485 		}
4486 		else if (r->direction && r->direction != dir) {
4487 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4488 			    "dir");
4489 			r = r->skip[PFE_SKIP_DIR].ptr;
4490 		}
4491 		else if (r->proto && r->proto != proto) {
4492 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4493 			    "proto");
4494 			r = r->skip[PFE_SKIP_PROTO].ptr;
4495 		}
4496 		else if (! pf_match_eth_addr(e->ether_shost, &r->src)) {
4497 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4498 			    "src");
4499 			r = r->skip[PFE_SKIP_SRC_ADDR].ptr;
4500 		}
4501 		else if (! pf_match_eth_addr(e->ether_dhost, &r->dst)) {
4502 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4503 			    "dst");
4504 			r = r->skip[PFE_SKIP_DST_ADDR].ptr;
4505 		}
4506 		else if (src != NULL && PF_MISMATCHAW(&r->ipsrc.addr, src, af,
4507 		    r->ipsrc.neg, kif, M_GETFIB(m))) {
4508 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4509 			    "ip_src");
4510 			r = r->skip[PFE_SKIP_SRC_IP_ADDR].ptr;
4511 		}
4512 		else if (dst != NULL && PF_MISMATCHAW(&r->ipdst.addr, dst, af,
4513 		    r->ipdst.neg, kif, M_GETFIB(m))) {
4514 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4515 			    "ip_dst");
4516 			r = r->skip[PFE_SKIP_DST_IP_ADDR].ptr;
4517 		}
4518 		else if (r->match_tag && !pf_match_eth_tag(m, r, &tag,
4519 		    mtag ? mtag->tag : 0)) {
4520 			SDT_PROBE3(pf, eth, test_rule, mismatch, r->nr, r,
4521 			    "match_tag");
4522 			r = TAILQ_NEXT(r, entries);
4523 		}
4524 		else {
4525 			if (r->tag)
4526 				tag = r->tag;
4527 			if (r->anchor == NULL) {
4528 				/* Rule matches */
4529 				rm = r;
4530 
4531 				SDT_PROBE2(pf, eth, test_rule, match, r->nr, r);
4532 
4533 				if (r->quick)
4534 					break;
4535 
4536 				r = TAILQ_NEXT(r, entries);
4537 			} else {
4538 				pf_step_into_keth_anchor(anchor_stack, &asd,
4539 				    &ruleset, &r, &a, &match);
4540 			}
4541 		}
4542 		if (r == NULL && pf_step_out_of_keth_anchor(anchor_stack, &asd,
4543 		    &ruleset, &r, &a, &match))
4544 			break;
4545 	}
4546 
4547 	r = rm;
4548 
4549 	SDT_PROBE2(pf, eth, test_rule, final_match, (r != NULL ? r->nr : -1), r);
4550 
4551 	/* Default to pass. */
4552 	if (r == NULL) {
4553 		PF_RULES_RUNLOCK();
4554 		return (PF_PASS);
4555 	}
4556 
4557 	/* Execute action. */
4558 	counter_u64_add(r->packets[dir == PF_OUT], 1);
4559 	counter_u64_add(r->bytes[dir == PF_OUT], m_length(m, NULL));
4560 	pf_update_timestamp(r);
4561 
4562 	/* Shortcut. Don't tag if we're just going to drop anyway. */
4563 	if (r->action == PF_DROP) {
4564 		PF_RULES_RUNLOCK();
4565 		return (PF_DROP);
4566 	}
4567 
4568 	if (tag > 0) {
4569 		if (mtag == NULL)
4570 			mtag = pf_get_mtag(m);
4571 		if (mtag == NULL) {
4572 			PF_RULES_RUNLOCK();
4573 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4574 			return (PF_DROP);
4575 		}
4576 		mtag->tag = tag;
4577 	}
4578 
4579 	if (r->qid != 0) {
4580 		if (mtag == NULL)
4581 			mtag = pf_get_mtag(m);
4582 		if (mtag == NULL) {
4583 			PF_RULES_RUNLOCK();
4584 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4585 			return (PF_DROP);
4586 		}
4587 		mtag->qid = r->qid;
4588 	}
4589 
4590 	action = r->action;
4591 	bridge_to = r->bridge_to;
4592 
4593 	/* Dummynet */
4594 	if (r->dnpipe) {
4595 		struct ip_fw_args dnflow;
4596 
4597 		/* Drop packet if dummynet is not loaded. */
4598 		if (ip_dn_io_ptr == NULL) {
4599 			PF_RULES_RUNLOCK();
4600 			m_freem(m);
4601 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4602 			return (PF_DROP);
4603 		}
4604 		if (mtag == NULL)
4605 			mtag = pf_get_mtag(m);
4606 		if (mtag == NULL) {
4607 			PF_RULES_RUNLOCK();
4608 			counter_u64_add(V_pf_status.counters[PFRES_MEMORY], 1);
4609 			return (PF_DROP);
4610 		}
4611 
4612 		bzero(&dnflow, sizeof(dnflow));
4613 
4614 		/* We don't have port numbers here, so we set 0.  That means
4615 		 * that we'll be somewhat limited in distinguishing flows (i.e.
4616 		 * only based on IP addresses, not based on port numbers), but
4617 		 * it's better than nothing. */
4618 		dnflow.f_id.dst_port = 0;
4619 		dnflow.f_id.src_port = 0;
4620 		dnflow.f_id.proto = 0;
4621 
4622 		dnflow.rule.info = r->dnpipe;
4623 		dnflow.rule.info |= IPFW_IS_DUMMYNET;
4624 		if (r->dnflags & PFRULE_DN_IS_PIPE)
4625 			dnflow.rule.info |= IPFW_IS_PIPE;
4626 
4627 		dnflow.f_id.extra = dnflow.rule.info;
4628 
4629 		dnflow.flags = dir == PF_IN ? IPFW_ARGS_IN : IPFW_ARGS_OUT;
4630 		dnflow.flags |= IPFW_ARGS_ETHER;
4631 		dnflow.ifp = kif->pfik_ifp;
4632 
4633 		switch (af) {
4634 		case AF_INET:
4635 			dnflow.f_id.addr_type = 4;
4636 			dnflow.f_id.src_ip = src->v4.s_addr;
4637 			dnflow.f_id.dst_ip = dst->v4.s_addr;
4638 			break;
4639 		case AF_INET6:
4640 			dnflow.flags |= IPFW_ARGS_IP6;
4641 			dnflow.f_id.addr_type = 6;
4642 			dnflow.f_id.src_ip6 = src->v6;
4643 			dnflow.f_id.dst_ip6 = dst->v6;
4644 			break;
4645 		}
4646 
4647 		PF_RULES_RUNLOCK();
4648 
4649 		mtag->flags |= PF_MTAG_FLAG_DUMMYNET;
4650 		ip_dn_io_ptr(m0, &dnflow);
4651 		if (*m0 != NULL)
4652 			pf_dummynet_flag_remove(m, mtag);
4653 	} else {
4654 		PF_RULES_RUNLOCK();
4655 	}
4656 
4657 	if (action == PF_PASS && bridge_to) {
4658 		pf_bridge_to(bridge_to->pfik_ifp, *m0);
4659 		*m0 = NULL; /* We've eaten the packet. */
4660 	}
4661 
4662 	return (action);
4663 }
4664 
4665 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)4666 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, struct pfi_kkif *kif,
4667     struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_krule **am,
4668     struct pf_kruleset **rsm, struct inpcb *inp)
4669 {
4670 	struct pf_krule		*nr = NULL;
4671 	struct pf_addr		* const saddr = pd->src;
4672 	struct pf_addr		* const daddr = pd->dst;
4673 	sa_family_t		 af = pd->af;
4674 	struct pf_krule		*r, *a = NULL;
4675 	struct pf_kruleset	*ruleset = NULL;
4676 	struct pf_krule_slist	 match_rules;
4677 	struct pf_krule_item	*ri;
4678 	struct pf_ksrc_node	*nsn = NULL;
4679 	struct tcphdr		*th = &pd->hdr.tcp;
4680 	struct pf_state_key	*sk = NULL, *nk = NULL;
4681 	u_short			 reason, transerror;
4682 	int			 rewrite = 0, hdrlen = 0;
4683 	int			 tag = -1;
4684 	int			 asd = 0;
4685 	int			 match = 0;
4686 	int			 state_icmp = 0, icmp_dir, multi;
4687 	u_int16_t		 sport = 0, dport = 0, virtual_type, virtual_id;
4688 	u_int16_t		 bproto_sum = 0, bip_sum = 0;
4689 	u_int8_t		 icmptype = 0, icmpcode = 0;
4690 	struct pf_kanchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
4691 
4692 	PF_RULES_RASSERT();
4693 
4694 	SLIST_INIT(&match_rules);
4695 
4696 	if (inp != NULL) {
4697 		INP_LOCK_ASSERT(inp);
4698 		pd->lookup.uid = inp->inp_cred->cr_uid;
4699 		pd->lookup.gid = inp->inp_cred->cr_groups[0];
4700 		pd->lookup.done = 1;
4701 	}
4702 
4703 	switch (pd->proto) {
4704 	case IPPROTO_TCP:
4705 		sport = th->th_sport;
4706 		dport = th->th_dport;
4707 		hdrlen = sizeof(*th);
4708 		break;
4709 	case IPPROTO_UDP:
4710 		sport = pd->hdr.udp.uh_sport;
4711 		dport = pd->hdr.udp.uh_dport;
4712 		hdrlen = sizeof(pd->hdr.udp);
4713 		break;
4714 	case IPPROTO_SCTP:
4715 		sport = pd->hdr.sctp.src_port;
4716 		dport = pd->hdr.sctp.dest_port;
4717 		hdrlen = sizeof(pd->hdr.sctp);
4718 		break;
4719 #ifdef INET
4720 	case IPPROTO_ICMP:
4721 		if (pd->af != AF_INET)
4722 			break;
4723 		hdrlen = sizeof(pd->hdr.icmp);
4724 		icmptype = pd->hdr.icmp.icmp_type;
4725 		icmpcode = pd->hdr.icmp.icmp_code;
4726 		state_icmp = pf_icmp_mapping(pd, icmptype,
4727 		    &icmp_dir, &multi, &virtual_id, &virtual_type);
4728 		if (icmp_dir == PF_IN) {
4729 			sport = virtual_id;
4730 			dport = virtual_type;
4731 		} else {
4732 			sport = virtual_type;
4733 			dport = virtual_id;
4734 		}
4735 		break;
4736 #endif /* INET */
4737 #ifdef INET6
4738 	case IPPROTO_ICMPV6:
4739 		if (af != AF_INET6)
4740 			break;
4741 		hdrlen = sizeof(pd->hdr.icmp6);
4742 		icmptype = pd->hdr.icmp6.icmp6_type;
4743 		icmpcode = pd->hdr.icmp6.icmp6_code;
4744 		state_icmp = pf_icmp_mapping(pd, icmptype,
4745 		    &icmp_dir, &multi, &virtual_id, &virtual_type);
4746 		if (icmp_dir == PF_IN) {
4747 			sport = virtual_id;
4748 			dport = virtual_type;
4749 		} else {
4750 			sport = virtual_type;
4751 			dport = virtual_id;
4752 		}
4753 
4754 		break;
4755 #endif /* INET6 */
4756 	default:
4757 		sport = dport = hdrlen = 0;
4758 		break;
4759 	}
4760 
4761 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
4762 
4763 	/* check packet for BINAT/NAT/RDR */
4764 	transerror = pf_get_translation(pd, m, off, kif, &nsn, &sk,
4765 	    &nk, saddr, daddr, sport, dport, anchor_stack, &nr);
4766 	switch (transerror) {
4767 	default:
4768 		/* A translation error occurred. */
4769 		REASON_SET(&reason, transerror);
4770 		goto cleanup;
4771 	case PFRES_MAX:
4772 		/* No match. */
4773 		break;
4774 	case PFRES_MATCH:
4775 		KASSERT(sk != NULL, ("%s: null sk", __func__));
4776 		KASSERT(nk != NULL, ("%s: null nk", __func__));
4777 
4778 		if (nr->log) {
4779 			PFLOG_PACKET(kif, m, af, PFRES_MATCH, nr, a,
4780 			    ruleset, pd, 1);
4781 		}
4782 
4783 		if (pd->ip_sum)
4784 			bip_sum = *pd->ip_sum;
4785 
4786 		switch (pd->proto) {
4787 		case IPPROTO_TCP:
4788 			bproto_sum = th->th_sum;
4789 			pd->proto_sum = &th->th_sum;
4790 
4791 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
4792 			    nk->port[pd->sidx] != sport) {
4793 				pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
4794 				    &th->th_sum, &nk->addr[pd->sidx],
4795 				    nk->port[pd->sidx], 0, af);
4796 				pd->sport = &th->th_sport;
4797 				sport = th->th_sport;
4798 			}
4799 
4800 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
4801 			    nk->port[pd->didx] != dport) {
4802 				pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
4803 				    &th->th_sum, &nk->addr[pd->didx],
4804 				    nk->port[pd->didx], 0, af);
4805 				dport = th->th_dport;
4806 				pd->dport = &th->th_dport;
4807 			}
4808 			rewrite++;
4809 			break;
4810 		case IPPROTO_UDP:
4811 			bproto_sum = pd->hdr.udp.uh_sum;
4812 			pd->proto_sum = &pd->hdr.udp.uh_sum;
4813 
4814 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
4815 			    nk->port[pd->sidx] != sport) {
4816 				pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport,
4817 				    pd->ip_sum, &pd->hdr.udp.uh_sum,
4818 				    &nk->addr[pd->sidx],
4819 				    nk->port[pd->sidx], 1, af);
4820 				sport = pd->hdr.udp.uh_sport;
4821 				pd->sport = &pd->hdr.udp.uh_sport;
4822 			}
4823 
4824 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
4825 			    nk->port[pd->didx] != dport) {
4826 				pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport,
4827 				    pd->ip_sum, &pd->hdr.udp.uh_sum,
4828 				    &nk->addr[pd->didx],
4829 				    nk->port[pd->didx], 1, af);
4830 				dport = pd->hdr.udp.uh_dport;
4831 				pd->dport = &pd->hdr.udp.uh_dport;
4832 			}
4833 			rewrite++;
4834 			break;
4835 		case IPPROTO_SCTP: {
4836 			uint16_t checksum = 0;
4837 
4838 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
4839 			    nk->port[pd->sidx] != sport) {
4840 				pf_change_ap(m, saddr, &pd->hdr.sctp.src_port,
4841 				    pd->ip_sum, &checksum,
4842 				    &nk->addr[pd->sidx],
4843 				    nk->port[pd->sidx], 1, af);
4844 			}
4845 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
4846 			    nk->port[pd->didx] != dport) {
4847 				pf_change_ap(m, daddr, &pd->hdr.sctp.dest_port,
4848 				    pd->ip_sum, &checksum,
4849 				    &nk->addr[pd->didx],
4850 				    nk->port[pd->didx], 1, af);
4851 			}
4852 			break;
4853 		}
4854 #ifdef INET
4855 		case IPPROTO_ICMP:
4856 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
4857 				pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
4858 				    nk->addr[pd->sidx].v4.s_addr, 0);
4859 
4860 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
4861 				pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
4862 				    nk->addr[pd->didx].v4.s_addr, 0);
4863 
4864 			if (virtual_type == htons(ICMP_ECHO) &&
4865 			     nk->port[pd->sidx] != pd->hdr.icmp.icmp_id) {
4866 				pd->hdr.icmp.icmp_cksum = pf_cksum_fixup(
4867 				    pd->hdr.icmp.icmp_cksum, sport,
4868 				    nk->port[pd->sidx], 0);
4869 				pd->hdr.icmp.icmp_id = nk->port[pd->sidx];
4870 				pd->sport = &pd->hdr.icmp.icmp_id;
4871 			}
4872 			m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
4873 			break;
4874 #endif /* INET */
4875 #ifdef INET6
4876 		case IPPROTO_ICMPV6:
4877 			if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
4878 				pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum,
4879 				    &nk->addr[pd->sidx], 0);
4880 
4881 			if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
4882 				pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum,
4883 				    &nk->addr[pd->didx], 0);
4884 			rewrite++;
4885 			break;
4886 #endif /* INET */
4887 		default:
4888 			switch (af) {
4889 #ifdef INET
4890 			case AF_INET:
4891 				if (PF_ANEQ(saddr,
4892 				    &nk->addr[pd->sidx], AF_INET))
4893 					pf_change_a(&saddr->v4.s_addr,
4894 					    pd->ip_sum,
4895 					    nk->addr[pd->sidx].v4.s_addr, 0);
4896 
4897 				if (PF_ANEQ(daddr,
4898 				    &nk->addr[pd->didx], AF_INET))
4899 					pf_change_a(&daddr->v4.s_addr,
4900 					    pd->ip_sum,
4901 					    nk->addr[pd->didx].v4.s_addr, 0);
4902 				break;
4903 #endif /* INET */
4904 #ifdef INET6
4905 			case AF_INET6:
4906 				if (PF_ANEQ(saddr,
4907 				    &nk->addr[pd->sidx], AF_INET6))
4908 					PF_ACPY(saddr, &nk->addr[pd->sidx], af);
4909 
4910 				if (PF_ANEQ(daddr,
4911 				    &nk->addr[pd->didx], AF_INET6))
4912 					PF_ACPY(daddr, &nk->addr[pd->didx], af);
4913 				break;
4914 #endif /* INET */
4915 			}
4916 			break;
4917 		}
4918 		if (nr->natpass)
4919 			r = NULL;
4920 		pd->nat_rule = nr;
4921 	}
4922 
4923 	while (r != NULL) {
4924 		if (pd->related_rule) {
4925 			*rm = pd->related_rule;
4926 			break;
4927 		}
4928 		pf_counter_u64_add(&r->evaluations, 1);
4929 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
4930 			r = r->skip[PF_SKIP_IFP].ptr;
4931 		else if (r->direction && r->direction != pd->dir)
4932 			r = r->skip[PF_SKIP_DIR].ptr;
4933 		else if (r->af && r->af != af)
4934 			r = r->skip[PF_SKIP_AF].ptr;
4935 		else if (r->proto && r->proto != pd->proto)
4936 			r = r->skip[PF_SKIP_PROTO].ptr;
4937 		else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
4938 		    r->src.neg, kif, M_GETFIB(m)))
4939 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
4940 		/* tcp/udp only. port_op always 0 in other cases */
4941 		else if (r->src.port_op && !pf_match_port(r->src.port_op,
4942 		    r->src.port[0], r->src.port[1], sport))
4943 			r = r->skip[PF_SKIP_SRC_PORT].ptr;
4944 		else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
4945 		    r->dst.neg, NULL, M_GETFIB(m)))
4946 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
4947 		/* tcp/udp only. port_op always 0 in other cases */
4948 		else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
4949 		    r->dst.port[0], r->dst.port[1], dport))
4950 			r = r->skip[PF_SKIP_DST_PORT].ptr;
4951 		/* icmp only. type always 0 in other cases */
4952 		else if (r->type && r->type != icmptype + 1)
4953 			r = TAILQ_NEXT(r, entries);
4954 		/* icmp only. type always 0 in other cases */
4955 		else if (r->code && r->code != icmpcode + 1)
4956 			r = TAILQ_NEXT(r, entries);
4957 		else if (r->tos && !(r->tos == pd->tos))
4958 			r = TAILQ_NEXT(r, entries);
4959 		else if (r->rule_flag & PFRULE_FRAGMENT)
4960 			r = TAILQ_NEXT(r, entries);
4961 		else if (pd->proto == IPPROTO_TCP &&
4962 		    (r->flagset & th->th_flags) != r->flags)
4963 			r = TAILQ_NEXT(r, entries);
4964 		/* tcp/udp only. uid.op always 0 in other cases */
4965 		else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
4966 		    pf_socket_lookup(pd, m), 1)) &&
4967 		    !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
4968 		    pd->lookup.uid))
4969 			r = TAILQ_NEXT(r, entries);
4970 		/* tcp/udp only. gid.op always 0 in other cases */
4971 		else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
4972 		    pf_socket_lookup(pd, m), 1)) &&
4973 		    !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
4974 		    pd->lookup.gid))
4975 			r = TAILQ_NEXT(r, entries);
4976 		else if (r->prio &&
4977 		    !pf_match_ieee8021q_pcp(r->prio, m))
4978 			r = TAILQ_NEXT(r, entries);
4979 		else if (r->prob &&
4980 		    r->prob <= arc4random())
4981 			r = TAILQ_NEXT(r, entries);
4982 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
4983 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
4984 			r = TAILQ_NEXT(r, entries);
4985 		else if (r->os_fingerprint != PF_OSFP_ANY &&
4986 		    (pd->proto != IPPROTO_TCP || !pf_osfp_match(
4987 		    pf_osfp_fingerprint(pd, m, off, th),
4988 		    r->os_fingerprint)))
4989 			r = TAILQ_NEXT(r, entries);
4990 		else {
4991 			if (r->tag)
4992 				tag = r->tag;
4993 			if (r->anchor == NULL) {
4994 				if (r->action == PF_MATCH) {
4995 					ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO);
4996 					if (ri == NULL) {
4997 						REASON_SET(&reason, PFRES_MEMORY);
4998 						goto cleanup;
4999 					}
5000 					ri->r = r;
5001 					SLIST_INSERT_HEAD(&match_rules, ri, entry);
5002 					pf_counter_u64_critical_enter();
5003 					pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
5004 					pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
5005 					pf_counter_u64_critical_exit();
5006 					pf_rule_to_actions(r, &pd->act);
5007 					if (r->log)
5008 						PFLOG_PACKET(kif, m, af,
5009 						    PFRES_MATCH, r,
5010 						    a, ruleset, pd, 1);
5011 				} else {
5012 					match = 1;
5013 					*rm = r;
5014 					*am = a;
5015 					*rsm = ruleset;
5016 				}
5017 				if ((*rm)->quick)
5018 					break;
5019 				r = TAILQ_NEXT(r, entries);
5020 			} else
5021 				pf_step_into_anchor(anchor_stack, &asd,
5022 				    &ruleset, PF_RULESET_FILTER, &r, &a,
5023 				    &match);
5024 		}
5025 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
5026 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
5027 			break;
5028 	}
5029 	r = *rm;
5030 	a = *am;
5031 	ruleset = *rsm;
5032 
5033 	REASON_SET(&reason, PFRES_MATCH);
5034 
5035 	/* apply actions for last matching pass/block rule */
5036 	pf_rule_to_actions(r, &pd->act);
5037 
5038 	if (r->log) {
5039 		if (rewrite)
5040 			m_copyback(m, off, hdrlen, pd->hdr.any);
5041 		PFLOG_PACKET(kif, m, af, reason, r, a, ruleset, pd, 1);
5042 	}
5043 
5044 	if ((r->action == PF_DROP) &&
5045 	    ((r->rule_flag & PFRULE_RETURNRST) ||
5046 	    (r->rule_flag & PFRULE_RETURNICMP) ||
5047 	    (r->rule_flag & PFRULE_RETURN))) {
5048 		pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum,
5049 		    bip_sum, hdrlen, &reason, r->rtableid);
5050 	}
5051 
5052 	if (r->action == PF_DROP)
5053 		goto cleanup;
5054 
5055 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
5056 		REASON_SET(&reason, PFRES_MEMORY);
5057 		goto cleanup;
5058 	}
5059 	if (pd->act.rtableid >= 0)
5060 		M_SETFIB(m, pd->act.rtableid);
5061 
5062 	if (!state_icmp && (r->keep_state || nr != NULL ||
5063 	    (pd->flags & PFDESC_TCP_NORM))) {
5064 		int action;
5065 		action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
5066 		    sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
5067 		    hdrlen, &match_rules);
5068 		sk = nk = NULL;
5069 		if (action != PF_PASS) {
5070 			pd->act.log |= PF_LOG_FORCE;
5071 			if (action == PF_DROP &&
5072 			    (r->rule_flag & PFRULE_RETURN))
5073 				pf_return(r, nr, pd, sk, off, m, th, kif,
5074 				    bproto_sum, bip_sum, hdrlen, &reason,
5075 				    pd->act.rtableid);
5076 			return (action);
5077 		}
5078 	} else {
5079 		while ((ri = SLIST_FIRST(&match_rules))) {
5080 			SLIST_REMOVE_HEAD(&match_rules, entry);
5081 			free(ri, M_PF_RULE_ITEM);
5082 		}
5083 
5084 		uma_zfree(V_pf_state_key_z, sk);
5085 		uma_zfree(V_pf_state_key_z, nk);
5086 		sk = nk = NULL;
5087 	}
5088 
5089 	/* copy back packet headers if we performed NAT operations */
5090 	if (rewrite)
5091 		m_copyback(m, off, hdrlen, pd->hdr.any);
5092 
5093 	if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
5094 	    pd->dir == PF_OUT &&
5095 	    V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m))
5096 		/*
5097 		 * We want the state created, but we dont
5098 		 * want to send this in case a partner
5099 		 * firewall has to know about it to allow
5100 		 * replies through it.
5101 		 */
5102 		return (PF_DEFER);
5103 
5104 	return (PF_PASS);
5105 
5106 cleanup:
5107 	while ((ri = SLIST_FIRST(&match_rules))) {
5108 		SLIST_REMOVE_HEAD(&match_rules, entry);
5109 		free(ri, M_PF_RULE_ITEM);
5110 	}
5111 
5112 	uma_zfree(V_pf_state_key_z, sk);
5113 	uma_zfree(V_pf_state_key_z, nk);
5114 	return (PF_DROP);
5115 }
5116 
5117 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)5118 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a,
5119     struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk,
5120     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
5121     u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm,
5122     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
5123     struct pf_krule_slist *match_rules)
5124 {
5125 	struct pf_kstate	*s = NULL;
5126 	struct pf_ksrc_node	*sn = NULL;
5127 	struct tcphdr		*th = &pd->hdr.tcp;
5128 	u_int16_t		 mss = V_tcp_mssdflt;
5129 	u_short			 reason, sn_reason;
5130 	struct pf_krule_item	*ri;
5131 
5132 	/* check maximums */
5133 	if (r->max_states &&
5134 	    (counter_u64_fetch(r->states_cur) >= r->max_states)) {
5135 		counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
5136 		REASON_SET(&reason, PFRES_MAXSTATES);
5137 		goto csfailed;
5138 	}
5139 	/* src node for filter rule */
5140 	if ((r->rule_flag & PFRULE_SRCTRACK ||
5141 	    r->rpool.opts & PF_POOL_STICKYADDR) &&
5142 	    (sn_reason = pf_insert_src_node(&sn, r, pd->src, pd->af)) != 0) {
5143 		REASON_SET(&reason, sn_reason);
5144 		goto csfailed;
5145 	}
5146 	/* src node for translation rule */
5147 	if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
5148 	    (sn_reason = pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx],
5149 	    pd->af)) != 0 ) {
5150 		REASON_SET(&reason, sn_reason);
5151 		goto csfailed;
5152 	}
5153 	s = pf_alloc_state(M_NOWAIT);
5154 	if (s == NULL) {
5155 		REASON_SET(&reason, PFRES_MEMORY);
5156 		goto csfailed;
5157 	}
5158 	s->rule.ptr = r;
5159 	s->nat_rule.ptr = nr;
5160 	s->anchor.ptr = a;
5161 	bcopy(match_rules, &s->match_rules, sizeof(s->match_rules));
5162 	memcpy(&s->act, &pd->act, sizeof(struct pf_rule_actions));
5163 
5164 	STATE_INC_COUNTERS(s);
5165 	if (r->allow_opts)
5166 		s->state_flags |= PFSTATE_ALLOWOPTS;
5167 	if (r->rule_flag & PFRULE_STATESLOPPY)
5168 		s->state_flags |= PFSTATE_SLOPPY;
5169 	if (pd->flags & PFDESC_TCP_NORM) /* Set by old-style scrub rules */
5170 		s->state_flags |= PFSTATE_SCRUB_TCP;
5171 
5172 	s->act.log = pd->act.log & PF_LOG_ALL;
5173 	s->sync_state = PFSYNC_S_NONE;
5174 	s->state_flags |= pd->act.flags; /* Only needed for pfsync and state export */
5175 
5176 	if (nr != NULL)
5177 		s->act.log |= nr->log & PF_LOG_ALL;
5178 	switch (pd->proto) {
5179 	case IPPROTO_TCP:
5180 		s->src.seqlo = ntohl(th->th_seq);
5181 		s->src.seqhi = s->src.seqlo + pd->p_len + 1;
5182 		if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
5183 		    r->keep_state == PF_STATE_MODULATE) {
5184 			/* Generate sequence number modulator */
5185 			if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
5186 			    0)
5187 				s->src.seqdiff = 1;
5188 			pf_change_proto_a(m, &th->th_seq, &th->th_sum,
5189 			    htonl(s->src.seqlo + s->src.seqdiff), 0);
5190 			*rewrite = 1;
5191 		} else
5192 			s->src.seqdiff = 0;
5193 		if (th->th_flags & TH_SYN) {
5194 			s->src.seqhi++;
5195 			s->src.wscale = pf_get_wscale(m, off,
5196 			    th->th_off, pd->af);
5197 		}
5198 		s->src.max_win = MAX(ntohs(th->th_win), 1);
5199 		if (s->src.wscale & PF_WSCALE_MASK) {
5200 			/* Remove scale factor from initial window */
5201 			int win = s->src.max_win;
5202 			win += 1 << (s->src.wscale & PF_WSCALE_MASK);
5203 			s->src.max_win = (win - 1) >>
5204 			    (s->src.wscale & PF_WSCALE_MASK);
5205 		}
5206 		if (th->th_flags & TH_FIN)
5207 			s->src.seqhi++;
5208 		s->dst.seqhi = 1;
5209 		s->dst.max_win = 1;
5210 		pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT);
5211 		pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED);
5212 		s->timeout = PFTM_TCP_FIRST_PACKET;
5213 		atomic_add_32(&V_pf_status.states_halfopen, 1);
5214 		break;
5215 	case IPPROTO_UDP:
5216 		pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE);
5217 		pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC);
5218 		s->timeout = PFTM_UDP_FIRST_PACKET;
5219 		break;
5220 	case IPPROTO_SCTP:
5221 		pf_set_protostate(s, PF_PEER_SRC, SCTP_COOKIE_WAIT);
5222 		pf_set_protostate(s, PF_PEER_DST, SCTP_CLOSED);
5223 		s->timeout = PFTM_SCTP_FIRST_PACKET;
5224 		break;
5225 	case IPPROTO_ICMP:
5226 #ifdef INET6
5227 	case IPPROTO_ICMPV6:
5228 #endif
5229 		s->timeout = PFTM_ICMP_FIRST_PACKET;
5230 		break;
5231 	default:
5232 		pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE);
5233 		pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC);
5234 		s->timeout = PFTM_OTHER_FIRST_PACKET;
5235 	}
5236 
5237 	if (r->rt) {
5238 		/* pf_map_addr increases the reason counters */
5239 		if ((reason = pf_map_addr(pd->af, r, pd->src, &s->rt_addr,
5240 		    &s->rt_kif, NULL, &sn)) != 0)
5241 			goto csfailed;
5242 		s->rt = r->rt;
5243 	}
5244 
5245 	s->creation = time_uptime;
5246 	s->expire = time_uptime;
5247 
5248 	if (sn != NULL)
5249 		s->src_node = sn;
5250 	if (nsn != NULL) {
5251 		/* XXX We only modify one side for now. */
5252 		PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
5253 		s->nat_src_node = nsn;
5254 	}
5255 	if (pd->proto == IPPROTO_TCP) {
5256 		if (s->state_flags & PFSTATE_SCRUB_TCP &&
5257 		    pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) {
5258 			REASON_SET(&reason, PFRES_MEMORY);
5259 			goto drop;
5260 		}
5261 		if (s->state_flags & PFSTATE_SCRUB_TCP && s->src.scrub &&
5262 		    pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
5263 		    &s->src, &s->dst, rewrite)) {
5264 			/* This really shouldn't happen!!! */
5265 			DPFPRINTF(PF_DEBUG_URGENT,
5266 			    ("pf_normalize_tcp_stateful failed on first "
5267 			     "pkt\n"));
5268 			goto drop;
5269 		}
5270 	} else if (pd->proto == IPPROTO_SCTP) {
5271 		if (pf_normalize_sctp_init(m, off, pd, &s->src, &s->dst))
5272 			goto drop;
5273 		if (! (pd->sctp_flags & (PFDESC_SCTP_INIT | PFDESC_SCTP_ADD_IP)))
5274 			goto drop;
5275 	}
5276 	s->direction = pd->dir;
5277 
5278 	/*
5279 	 * sk/nk could already been setup by pf_get_translation().
5280 	 */
5281 	if (nr == NULL) {
5282 		KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
5283 		    __func__, nr, sk, nk));
5284 		sk = pf_state_key_setup(pd, m, off, pd->src, pd->dst, sport, dport);
5285 		if (sk == NULL)
5286 			goto csfailed;
5287 		nk = sk;
5288 	} else
5289 		KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
5290 		    __func__, nr, sk, nk));
5291 
5292 	/* Swap sk/nk for PF_OUT. */
5293 	if (pf_state_insert(BOUND_IFACE(r, kif, pd), kif,
5294 	    (pd->dir == PF_IN) ? sk : nk,
5295 	    (pd->dir == PF_IN) ? nk : sk, s)) {
5296 		REASON_SET(&reason, PFRES_STATEINS);
5297 		goto drop;
5298 	} else
5299 		*sm = s;
5300 	sk = nk = NULL;
5301 
5302 	if (tag > 0)
5303 		s->tag = tag;
5304 	if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
5305 	    TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
5306 		pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC);
5307 		/* undo NAT changes, if they have taken place */
5308 		if (nr != NULL) {
5309 			struct pf_state_key *skt = s->key[PF_SK_WIRE];
5310 			if (pd->dir == PF_OUT)
5311 				skt = s->key[PF_SK_STACK];
5312 			PF_ACPY(pd->src, &pd->osrc, pd->af);
5313 			PF_ACPY(pd->dst, &pd->odst, pd->af);
5314 			if (pd->sport)
5315 				*pd->sport = skt->port[pd->sidx];
5316 			if (pd->dport)
5317 				*pd->dport = skt->port[pd->didx];
5318 			if (pd->proto_sum)
5319 				*pd->proto_sum = bproto_sum;
5320 			if (pd->ip_sum)
5321 				*pd->ip_sum = bip_sum;
5322 			m_copyback(m, off, hdrlen, pd->hdr.any);
5323 		}
5324 		s->src.seqhi = htonl(arc4random());
5325 		/* Find mss option */
5326 		int rtid = M_GETFIB(m);
5327 		mss = pf_get_mss(m, off, th->th_off, pd->af);
5328 		mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
5329 		mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
5330 		s->src.mss = mss;
5331 		pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport,
5332 		    th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
5333 		    TH_SYN|TH_ACK, 0, s->src.mss, 0, true, 0, 0,
5334 		    pd->act.rtableid);
5335 		REASON_SET(&reason, PFRES_SYNPROXY);
5336 		return (PF_SYNPROXY_DROP);
5337 	}
5338 
5339 	return (PF_PASS);
5340 
5341 csfailed:
5342 	while ((ri = SLIST_FIRST(match_rules))) {
5343 		SLIST_REMOVE_HEAD(match_rules, entry);
5344 		free(ri, M_PF_RULE_ITEM);
5345 	}
5346 
5347 	uma_zfree(V_pf_state_key_z, sk);
5348 	uma_zfree(V_pf_state_key_z, nk);
5349 
5350 	if (sn != NULL) {
5351 		PF_SRC_NODE_LOCK(sn);
5352 		if (--sn->states == 0 && sn->expire == 0) {
5353 			pf_unlink_src_node(sn);
5354 			uma_zfree(V_pf_sources_z, sn);
5355 			counter_u64_add(
5356 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
5357 		}
5358 		PF_SRC_NODE_UNLOCK(sn);
5359 	}
5360 
5361 	if (nsn != sn && nsn != NULL) {
5362 		PF_SRC_NODE_LOCK(nsn);
5363 		if (--nsn->states == 0 && nsn->expire == 0) {
5364 			pf_unlink_src_node(nsn);
5365 			uma_zfree(V_pf_sources_z, nsn);
5366 			counter_u64_add(
5367 			    V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
5368 		}
5369 		PF_SRC_NODE_UNLOCK(nsn);
5370 	}
5371 
5372 drop:
5373 	if (s != NULL) {
5374 		pf_src_tree_remove_state(s);
5375 		s->timeout = PFTM_UNLINKED;
5376 		STATE_DEC_COUNTERS(s);
5377 		pf_free_state(s);
5378 	}
5379 
5380 	return (PF_DROP);
5381 }
5382 
5383 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)5384 pf_test_fragment(struct pf_krule **rm, struct pfi_kkif *kif,
5385     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am,
5386     struct pf_kruleset **rsm)
5387 {
5388 	struct pf_krule		*r, *a = NULL;
5389 	struct pf_kruleset	*ruleset = NULL;
5390 	struct pf_krule_slist	 match_rules;
5391 	struct pf_krule_item	*ri;
5392 	sa_family_t		 af = pd->af;
5393 	u_short			 reason;
5394 	int			 tag = -1;
5395 	int			 asd = 0;
5396 	int			 match = 0;
5397 	struct pf_kanchor_stackframe	anchor_stack[PF_ANCHOR_STACKSIZE];
5398 
5399 	PF_RULES_RASSERT();
5400 
5401 	r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
5402 	SLIST_INIT(&match_rules);
5403 	while (r != NULL) {
5404 		pf_counter_u64_add(&r->evaluations, 1);
5405 		if (pfi_kkif_match(r->kif, kif) == r->ifnot)
5406 			r = r->skip[PF_SKIP_IFP].ptr;
5407 		else if (r->direction && r->direction != pd->dir)
5408 			r = r->skip[PF_SKIP_DIR].ptr;
5409 		else if (r->af && r->af != af)
5410 			r = r->skip[PF_SKIP_AF].ptr;
5411 		else if (r->proto && r->proto != pd->proto)
5412 			r = r->skip[PF_SKIP_PROTO].ptr;
5413 		else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
5414 		    r->src.neg, kif, M_GETFIB(m)))
5415 			r = r->skip[PF_SKIP_SRC_ADDR].ptr;
5416 		else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
5417 		    r->dst.neg, NULL, M_GETFIB(m)))
5418 			r = r->skip[PF_SKIP_DST_ADDR].ptr;
5419 		else if (r->tos && !(r->tos == pd->tos))
5420 			r = TAILQ_NEXT(r, entries);
5421 		else if (r->os_fingerprint != PF_OSFP_ANY)
5422 			r = TAILQ_NEXT(r, entries);
5423 		else if (pd->proto == IPPROTO_UDP &&
5424 		    (r->src.port_op || r->dst.port_op))
5425 			r = TAILQ_NEXT(r, entries);
5426 		else if (pd->proto == IPPROTO_TCP &&
5427 		    (r->src.port_op || r->dst.port_op || r->flagset))
5428 			r = TAILQ_NEXT(r, entries);
5429 		else if ((pd->proto == IPPROTO_ICMP ||
5430 		    pd->proto == IPPROTO_ICMPV6) &&
5431 		    (r->type || r->code))
5432 			r = TAILQ_NEXT(r, entries);
5433 		else if (r->prio &&
5434 		    !pf_match_ieee8021q_pcp(r->prio, m))
5435 			r = TAILQ_NEXT(r, entries);
5436 		else if (r->prob && r->prob <=
5437 		    (arc4random() % (UINT_MAX - 1) + 1))
5438 			r = TAILQ_NEXT(r, entries);
5439 		else if (r->match_tag && !pf_match_tag(m, r, &tag,
5440 		    pd->pf_mtag ? pd->pf_mtag->tag : 0))
5441 			r = TAILQ_NEXT(r, entries);
5442 		else {
5443 			if (r->anchor == NULL) {
5444 				if (r->action == PF_MATCH) {
5445 					ri = malloc(sizeof(struct pf_krule_item), M_PF_RULE_ITEM, M_NOWAIT | M_ZERO);
5446 					if (ri == NULL) {
5447 						REASON_SET(&reason, PFRES_MEMORY);
5448 						goto cleanup;
5449 					}
5450 					ri->r = r;
5451 					SLIST_INSERT_HEAD(&match_rules, ri, entry);
5452 					pf_counter_u64_critical_enter();
5453 					pf_counter_u64_add_protected(&r->packets[pd->dir == PF_OUT], 1);
5454 					pf_counter_u64_add_protected(&r->bytes[pd->dir == PF_OUT], pd->tot_len);
5455 					pf_counter_u64_critical_exit();
5456 					pf_rule_to_actions(r, &pd->act);
5457 					if (r->log)
5458 						PFLOG_PACKET(kif, m, af,
5459 						    PFRES_MATCH, r,
5460 						    a, ruleset, pd, 1);
5461 				} else {
5462 					match = 1;
5463 					*rm = r;
5464 					*am = a;
5465 					*rsm = ruleset;
5466 				}
5467 				if ((*rm)->quick)
5468 					break;
5469 				r = TAILQ_NEXT(r, entries);
5470 			} else
5471 				pf_step_into_anchor(anchor_stack, &asd,
5472 				    &ruleset, PF_RULESET_FILTER, &r, &a,
5473 				    &match);
5474 		}
5475 		if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
5476 		    &ruleset, PF_RULESET_FILTER, &r, &a, &match))
5477 			break;
5478 	}
5479 	r = *rm;
5480 	a = *am;
5481 	ruleset = *rsm;
5482 
5483 	REASON_SET(&reason, PFRES_MATCH);
5484 
5485 	/* apply actions for last matching pass/block rule */
5486 	pf_rule_to_actions(r, &pd->act);
5487 
5488 	if (r->log)
5489 		PFLOG_PACKET(kif, m, af, reason, r, a, ruleset, pd, 1);
5490 
5491 	if (r->action != PF_PASS)
5492 		return (PF_DROP);
5493 
5494 	if (tag > 0 && pf_tag_packet(m, pd, tag)) {
5495 		REASON_SET(&reason, PFRES_MEMORY);
5496 		goto cleanup;
5497 	}
5498 
5499 	return (PF_PASS);
5500 
5501 cleanup:
5502 	while ((ri = SLIST_FIRST(&match_rules))) {
5503 		SLIST_REMOVE_HEAD(&match_rules, entry);
5504 		free(ri, M_PF_RULE_ITEM);
5505 	}
5506 
5507 	return (PF_DROP);
5508 }
5509 
5510 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)5511 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif,
5512     struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason,
5513     int *copyback)
5514 {
5515 	struct tcphdr		*th = &pd->hdr.tcp;
5516 	struct pf_state_peer	*src, *dst;
5517 	u_int16_t		 win = ntohs(th->th_win);
5518 	u_int32_t		 ack, end, seq, orig_seq;
5519 	u_int8_t		 sws, dws, psrc, pdst;
5520 	int			 ackskew;
5521 
5522 	if (pd->dir == (*state)->direction) {
5523 		src = &(*state)->src;
5524 		dst = &(*state)->dst;
5525 		psrc = PF_PEER_SRC;
5526 		pdst = PF_PEER_DST;
5527 	} else {
5528 		src = &(*state)->dst;
5529 		dst = &(*state)->src;
5530 		psrc = PF_PEER_DST;
5531 		pdst = PF_PEER_SRC;
5532 	}
5533 
5534 	if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
5535 		sws = src->wscale & PF_WSCALE_MASK;
5536 		dws = dst->wscale & PF_WSCALE_MASK;
5537 	} else
5538 		sws = dws = 0;
5539 
5540 	/*
5541 	 * Sequence tracking algorithm from Guido van Rooij's paper:
5542 	 *   http://www.madison-gurkha.com/publications/tcp_filtering/
5543 	 *	tcp_filtering.ps
5544 	 */
5545 
5546 	orig_seq = seq = ntohl(th->th_seq);
5547 	if (src->seqlo == 0) {
5548 		/* First packet from this end. Set its state */
5549 
5550 		if (((*state)->state_flags & PFSTATE_SCRUB_TCP || dst->scrub) &&
5551 		    src->scrub == NULL) {
5552 			if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
5553 				REASON_SET(reason, PFRES_MEMORY);
5554 				return (PF_DROP);
5555 			}
5556 		}
5557 
5558 		/* Deferred generation of sequence number modulator */
5559 		if (dst->seqdiff && !src->seqdiff) {
5560 			/* use random iss for the TCP server */
5561 			while ((src->seqdiff = arc4random() - seq) == 0)
5562 				;
5563 			ack = ntohl(th->th_ack) - dst->seqdiff;
5564 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
5565 			    src->seqdiff), 0);
5566 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
5567 			*copyback = 1;
5568 		} else {
5569 			ack = ntohl(th->th_ack);
5570 		}
5571 
5572 		end = seq + pd->p_len;
5573 		if (th->th_flags & TH_SYN) {
5574 			end++;
5575 			if (dst->wscale & PF_WSCALE_FLAG) {
5576 				src->wscale = pf_get_wscale(m, off, th->th_off,
5577 				    pd->af);
5578 				if (src->wscale & PF_WSCALE_FLAG) {
5579 					/* Remove scale factor from initial
5580 					 * window */
5581 					sws = src->wscale & PF_WSCALE_MASK;
5582 					win = ((u_int32_t)win + (1 << sws) - 1)
5583 					    >> sws;
5584 					dws = dst->wscale & PF_WSCALE_MASK;
5585 				} else {
5586 					/* fixup other window */
5587 					dst->max_win <<= dst->wscale &
5588 					    PF_WSCALE_MASK;
5589 					/* in case of a retrans SYN|ACK */
5590 					dst->wscale = 0;
5591 				}
5592 			}
5593 		}
5594 		if (th->th_flags & TH_FIN)
5595 			end++;
5596 
5597 		src->seqlo = seq;
5598 		if (src->state < TCPS_SYN_SENT)
5599 			pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
5600 
5601 		/*
5602 		 * May need to slide the window (seqhi may have been set by
5603 		 * the crappy stack check or if we picked up the connection
5604 		 * after establishment)
5605 		 */
5606 		if (src->seqhi == 1 ||
5607 		    SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
5608 			src->seqhi = end + MAX(1, dst->max_win << dws);
5609 		if (win > src->max_win)
5610 			src->max_win = win;
5611 
5612 	} else {
5613 		ack = ntohl(th->th_ack) - dst->seqdiff;
5614 		if (src->seqdiff) {
5615 			/* Modulate sequence numbers */
5616 			pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
5617 			    src->seqdiff), 0);
5618 			pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
5619 			*copyback = 1;
5620 		}
5621 		end = seq + pd->p_len;
5622 		if (th->th_flags & TH_SYN)
5623 			end++;
5624 		if (th->th_flags & TH_FIN)
5625 			end++;
5626 	}
5627 
5628 	if ((th->th_flags & TH_ACK) == 0) {
5629 		/* Let it pass through the ack skew check */
5630 		ack = dst->seqlo;
5631 	} else if ((ack == 0 &&
5632 	    (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
5633 	    /* broken tcp stacks do not set ack */
5634 	    (dst->state < TCPS_SYN_SENT)) {
5635 		/*
5636 		 * Many stacks (ours included) will set the ACK number in an
5637 		 * FIN|ACK if the SYN times out -- no sequence to ACK.
5638 		 */
5639 		ack = dst->seqlo;
5640 	}
5641 
5642 	if (seq == end) {
5643 		/* Ease sequencing restrictions on no data packets */
5644 		seq = src->seqlo;
5645 		end = seq;
5646 	}
5647 
5648 	ackskew = dst->seqlo - ack;
5649 
5650 	/*
5651 	 * Need to demodulate the sequence numbers in any TCP SACK options
5652 	 * (Selective ACK). We could optionally validate the SACK values
5653 	 * against the current ACK window, either forwards or backwards, but
5654 	 * I'm not confident that SACK has been implemented properly
5655 	 * everywhere. It wouldn't surprise me if several stacks accidentally
5656 	 * SACK too far backwards of previously ACKed data. There really aren't
5657 	 * any security implications of bad SACKing unless the target stack
5658 	 * doesn't validate the option length correctly. Someone trying to
5659 	 * spoof into a TCP connection won't bother blindly sending SACK
5660 	 * options anyway.
5661 	 */
5662 	if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
5663 		if (pf_modulate_sack(m, off, pd, th, dst))
5664 			*copyback = 1;
5665 	}
5666 
5667 #define	MAXACKWINDOW (0xffff + 1500)	/* 1500 is an arbitrary fudge factor */
5668 	if (SEQ_GEQ(src->seqhi, end) &&
5669 	    /* Last octet inside other's window space */
5670 	    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
5671 	    /* Retrans: not more than one window back */
5672 	    (ackskew >= -MAXACKWINDOW) &&
5673 	    /* Acking not more than one reassembled fragment backwards */
5674 	    (ackskew <= (MAXACKWINDOW << sws)) &&
5675 	    /* Acking not more than one window forward */
5676 	    ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
5677 	    (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo))) {
5678 	    /* Require an exact/+1 sequence match on resets when possible */
5679 
5680 		if (dst->scrub || src->scrub) {
5681 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
5682 			    *state, src, dst, copyback))
5683 				return (PF_DROP);
5684 		}
5685 
5686 		/* update max window */
5687 		if (src->max_win < win)
5688 			src->max_win = win;
5689 		/* synchronize sequencing */
5690 		if (SEQ_GT(end, src->seqlo))
5691 			src->seqlo = end;
5692 		/* slide the window of what the other end can send */
5693 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
5694 			dst->seqhi = ack + MAX((win << sws), 1);
5695 
5696 		/* update states */
5697 		if (th->th_flags & TH_SYN)
5698 			if (src->state < TCPS_SYN_SENT)
5699 				pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
5700 		if (th->th_flags & TH_FIN)
5701 			if (src->state < TCPS_CLOSING)
5702 				pf_set_protostate(*state, psrc, TCPS_CLOSING);
5703 		if (th->th_flags & TH_ACK) {
5704 			if (dst->state == TCPS_SYN_SENT) {
5705 				pf_set_protostate(*state, pdst,
5706 				    TCPS_ESTABLISHED);
5707 				if (src->state == TCPS_ESTABLISHED &&
5708 				    (*state)->src_node != NULL &&
5709 				    pf_src_connlimit(state)) {
5710 					REASON_SET(reason, PFRES_SRCLIMIT);
5711 					return (PF_DROP);
5712 				}
5713 			} else if (dst->state == TCPS_CLOSING)
5714 				pf_set_protostate(*state, pdst,
5715 				    TCPS_FIN_WAIT_2);
5716 		}
5717 		if (th->th_flags & TH_RST)
5718 			pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
5719 
5720 		/* update expire time */
5721 		(*state)->expire = time_uptime;
5722 		if (src->state >= TCPS_FIN_WAIT_2 &&
5723 		    dst->state >= TCPS_FIN_WAIT_2)
5724 			(*state)->timeout = PFTM_TCP_CLOSED;
5725 		else if (src->state >= TCPS_CLOSING &&
5726 		    dst->state >= TCPS_CLOSING)
5727 			(*state)->timeout = PFTM_TCP_FIN_WAIT;
5728 		else if (src->state < TCPS_ESTABLISHED ||
5729 		    dst->state < TCPS_ESTABLISHED)
5730 			(*state)->timeout = PFTM_TCP_OPENING;
5731 		else if (src->state >= TCPS_CLOSING ||
5732 		    dst->state >= TCPS_CLOSING)
5733 			(*state)->timeout = PFTM_TCP_CLOSING;
5734 		else
5735 			(*state)->timeout = PFTM_TCP_ESTABLISHED;
5736 
5737 		/* Fall through to PASS packet */
5738 
5739 	} else if ((dst->state < TCPS_SYN_SENT ||
5740 		dst->state >= TCPS_FIN_WAIT_2 ||
5741 		src->state >= TCPS_FIN_WAIT_2) &&
5742 	    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
5743 	    /* Within a window forward of the originating packet */
5744 	    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
5745 	    /* Within a window backward of the originating packet */
5746 
5747 		/*
5748 		 * This currently handles three situations:
5749 		 *  1) Stupid stacks will shotgun SYNs before their peer
5750 		 *     replies.
5751 		 *  2) When PF catches an already established stream (the
5752 		 *     firewall rebooted, the state table was flushed, routes
5753 		 *     changed...)
5754 		 *  3) Packets get funky immediately after the connection
5755 		 *     closes (this should catch Solaris spurious ACK|FINs
5756 		 *     that web servers like to spew after a close)
5757 		 *
5758 		 * This must be a little more careful than the above code
5759 		 * since packet floods will also be caught here. We don't
5760 		 * update the TTL here to mitigate the damage of a packet
5761 		 * flood and so the same code can handle awkward establishment
5762 		 * and a loosened connection close.
5763 		 * In the establishment case, a correct peer response will
5764 		 * validate the connection, go through the normal state code
5765 		 * and keep updating the state TTL.
5766 		 */
5767 
5768 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
5769 			printf("pf: loose state match: ");
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", seq, orig_seq, ack,
5774 			    pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
5775 			    (unsigned long long)(*state)->packets[1],
5776 			    pd->dir == PF_IN ? "in" : "out",
5777 			    pd->dir == (*state)->direction ? "fwd" : "rev");
5778 		}
5779 
5780 		if (dst->scrub || src->scrub) {
5781 			if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
5782 			    *state, src, dst, copyback))
5783 				return (PF_DROP);
5784 		}
5785 
5786 		/* update max window */
5787 		if (src->max_win < win)
5788 			src->max_win = win;
5789 		/* synchronize sequencing */
5790 		if (SEQ_GT(end, src->seqlo))
5791 			src->seqlo = end;
5792 		/* slide the window of what the other end can send */
5793 		if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
5794 			dst->seqhi = ack + MAX((win << sws), 1);
5795 
5796 		/*
5797 		 * Cannot set dst->seqhi here since this could be a shotgunned
5798 		 * SYN and not an already established connection.
5799 		 */
5800 
5801 		if (th->th_flags & TH_FIN)
5802 			if (src->state < TCPS_CLOSING)
5803 				pf_set_protostate(*state, psrc, TCPS_CLOSING);
5804 		if (th->th_flags & TH_RST)
5805 			pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
5806 
5807 		/* Fall through to PASS packet */
5808 
5809 	} else {
5810 		if ((*state)->dst.state == TCPS_SYN_SENT &&
5811 		    (*state)->src.state == TCPS_SYN_SENT) {
5812 			/* Send RST for state mismatches during handshake */
5813 			if (!(th->th_flags & TH_RST))
5814 				pf_send_tcp((*state)->rule.ptr, pd->af,
5815 				    pd->dst, pd->src, th->th_dport,
5816 				    th->th_sport, ntohl(th->th_ack), 0,
5817 				    TH_RST, 0, 0,
5818 				    (*state)->rule.ptr->return_ttl, true, 0, 0,
5819 				    (*state)->act.rtableid);
5820 			src->seqlo = 0;
5821 			src->seqhi = 1;
5822 			src->max_win = 1;
5823 		} else if (V_pf_status.debug >= PF_DEBUG_MISC) {
5824 			printf("pf: BAD state: ");
5825 			pf_print_state(*state);
5826 			pf_print_flags(th->th_flags);
5827 			printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
5828 			    "pkts=%llu:%llu dir=%s,%s\n",
5829 			    seq, orig_seq, ack, pd->p_len, ackskew,
5830 			    (unsigned long long)(*state)->packets[0],
5831 			    (unsigned long long)(*state)->packets[1],
5832 			    pd->dir == PF_IN ? "in" : "out",
5833 			    pd->dir == (*state)->direction ? "fwd" : "rev");
5834 			printf("pf: State failure on: %c %c %c %c | %c %c\n",
5835 			    SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
5836 			    SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
5837 			    ' ': '2',
5838 			    (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
5839 			    (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
5840 			    SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
5841 			    SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
5842 		}
5843 		REASON_SET(reason, PFRES_BADSTATE);
5844 		return (PF_DROP);
5845 	}
5846 
5847 	return (PF_PASS);
5848 }
5849 
5850 static int
pf_tcp_track_sloppy(struct pf_kstate ** state,struct pf_pdesc * pd,u_short * reason)5851 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason)
5852 {
5853 	struct tcphdr		*th = &pd->hdr.tcp;
5854 	struct pf_state_peer	*src, *dst;
5855 	u_int8_t		 psrc, pdst;
5856 
5857 	if (pd->dir == (*state)->direction) {
5858 		src = &(*state)->src;
5859 		dst = &(*state)->dst;
5860 		psrc = PF_PEER_SRC;
5861 		pdst = PF_PEER_DST;
5862 	} else {
5863 		src = &(*state)->dst;
5864 		dst = &(*state)->src;
5865 		psrc = PF_PEER_DST;
5866 		pdst = PF_PEER_SRC;
5867 	}
5868 
5869 	if (th->th_flags & TH_SYN)
5870 		if (src->state < TCPS_SYN_SENT)
5871 			pf_set_protostate(*state, psrc, TCPS_SYN_SENT);
5872 	if (th->th_flags & TH_FIN)
5873 		if (src->state < TCPS_CLOSING)
5874 			pf_set_protostate(*state, psrc, TCPS_CLOSING);
5875 	if (th->th_flags & TH_ACK) {
5876 		if (dst->state == TCPS_SYN_SENT) {
5877 			pf_set_protostate(*state, pdst, TCPS_ESTABLISHED);
5878 			if (src->state == TCPS_ESTABLISHED &&
5879 			    (*state)->src_node != NULL &&
5880 			    pf_src_connlimit(state)) {
5881 				REASON_SET(reason, PFRES_SRCLIMIT);
5882 				return (PF_DROP);
5883 			}
5884 		} else if (dst->state == TCPS_CLOSING) {
5885 			pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2);
5886 		} else if (src->state == TCPS_SYN_SENT &&
5887 		    dst->state < TCPS_SYN_SENT) {
5888 			/*
5889 			 * Handle a special sloppy case where we only see one
5890 			 * half of the connection. If there is a ACK after
5891 			 * the initial SYN without ever seeing a packet from
5892 			 * the destination, set the connection to established.
5893 			 */
5894 			pf_set_protostate(*state, PF_PEER_BOTH,
5895 			    TCPS_ESTABLISHED);
5896 			dst->state = src->state = TCPS_ESTABLISHED;
5897 			if ((*state)->src_node != NULL &&
5898 			    pf_src_connlimit(state)) {
5899 				REASON_SET(reason, PFRES_SRCLIMIT);
5900 				return (PF_DROP);
5901 			}
5902 		} else if (src->state == TCPS_CLOSING &&
5903 		    dst->state == TCPS_ESTABLISHED &&
5904 		    dst->seqlo == 0) {
5905 			/*
5906 			 * Handle the closing of half connections where we
5907 			 * don't see the full bidirectional FIN/ACK+ACK
5908 			 * handshake.
5909 			 */
5910 			pf_set_protostate(*state, pdst, TCPS_CLOSING);
5911 		}
5912 	}
5913 	if (th->th_flags & TH_RST)
5914 		pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT);
5915 
5916 	/* update expire time */
5917 	(*state)->expire = time_uptime;
5918 	if (src->state >= TCPS_FIN_WAIT_2 &&
5919 	    dst->state >= TCPS_FIN_WAIT_2)
5920 		(*state)->timeout = PFTM_TCP_CLOSED;
5921 	else if (src->state >= TCPS_CLOSING &&
5922 	    dst->state >= TCPS_CLOSING)
5923 		(*state)->timeout = PFTM_TCP_FIN_WAIT;
5924 	else if (src->state < TCPS_ESTABLISHED ||
5925 	    dst->state < TCPS_ESTABLISHED)
5926 		(*state)->timeout = PFTM_TCP_OPENING;
5927 	else if (src->state >= TCPS_CLOSING ||
5928 	    dst->state >= TCPS_CLOSING)
5929 		(*state)->timeout = PFTM_TCP_CLOSING;
5930 	else
5931 		(*state)->timeout = PFTM_TCP_ESTABLISHED;
5932 
5933 	return (PF_PASS);
5934 }
5935 
5936 static int
pf_synproxy(struct pf_pdesc * pd,struct pf_kstate ** state,u_short * reason)5937 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason)
5938 {
5939 	struct pf_state_key	*sk = (*state)->key[pd->didx];
5940 	struct tcphdr		*th = &pd->hdr.tcp;
5941 
5942 	if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
5943 		if (pd->dir != (*state)->direction) {
5944 			REASON_SET(reason, PFRES_SYNPROXY);
5945 			return (PF_SYNPROXY_DROP);
5946 		}
5947 		if (th->th_flags & TH_SYN) {
5948 			if (ntohl(th->th_seq) != (*state)->src.seqlo) {
5949 				REASON_SET(reason, PFRES_SYNPROXY);
5950 				return (PF_DROP);
5951 			}
5952 			pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
5953 			    pd->src, th->th_dport, th->th_sport,
5954 			    (*state)->src.seqhi, ntohl(th->th_seq) + 1,
5955 			    TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, true, 0, 0,
5956 			    (*state)->act.rtableid);
5957 			REASON_SET(reason, PFRES_SYNPROXY);
5958 			return (PF_SYNPROXY_DROP);
5959 		} else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK ||
5960 		    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
5961 		    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
5962 			REASON_SET(reason, PFRES_SYNPROXY);
5963 			return (PF_DROP);
5964 		} else if ((*state)->src_node != NULL &&
5965 		    pf_src_connlimit(state)) {
5966 			REASON_SET(reason, PFRES_SRCLIMIT);
5967 			return (PF_DROP);
5968 		} else
5969 			pf_set_protostate(*state, PF_PEER_SRC,
5970 			    PF_TCPS_PROXY_DST);
5971 	}
5972 	if ((*state)->src.state == PF_TCPS_PROXY_DST) {
5973 		if (pd->dir == (*state)->direction) {
5974 			if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
5975 			    (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
5976 			    (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
5977 				REASON_SET(reason, PFRES_SYNPROXY);
5978 				return (PF_DROP);
5979 			}
5980 			(*state)->src.max_win = MAX(ntohs(th->th_win), 1);
5981 			if ((*state)->dst.seqhi == 1)
5982 				(*state)->dst.seqhi = htonl(arc4random());
5983 			pf_send_tcp((*state)->rule.ptr, pd->af,
5984 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
5985 			    sk->port[pd->sidx], sk->port[pd->didx],
5986 			    (*state)->dst.seqhi, 0, TH_SYN, 0,
5987 			    (*state)->src.mss, 0, false, (*state)->tag, 0,
5988 			    (*state)->act.rtableid);
5989 			REASON_SET(reason, PFRES_SYNPROXY);
5990 			return (PF_SYNPROXY_DROP);
5991 		} else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
5992 		    (TH_SYN|TH_ACK)) ||
5993 		    (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
5994 			REASON_SET(reason, PFRES_SYNPROXY);
5995 			return (PF_DROP);
5996 		} else {
5997 			(*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
5998 			(*state)->dst.seqlo = ntohl(th->th_seq);
5999 			pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
6000 			    pd->src, th->th_dport, th->th_sport,
6001 			    ntohl(th->th_ack), ntohl(th->th_seq) + 1,
6002 			    TH_ACK, (*state)->src.max_win, 0, 0, false,
6003 			    (*state)->tag, 0, (*state)->act.rtableid);
6004 			pf_send_tcp((*state)->rule.ptr, pd->af,
6005 			    &sk->addr[pd->sidx], &sk->addr[pd->didx],
6006 			    sk->port[pd->sidx], sk->port[pd->didx],
6007 			    (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
6008 			    TH_ACK, (*state)->dst.max_win, 0, 0, true, 0, 0,
6009 			    (*state)->act.rtableid);
6010 			(*state)->src.seqdiff = (*state)->dst.seqhi -
6011 			    (*state)->src.seqlo;
6012 			(*state)->dst.seqdiff = (*state)->src.seqhi -
6013 			    (*state)->dst.seqlo;
6014 			(*state)->src.seqhi = (*state)->src.seqlo +
6015 			    (*state)->dst.max_win;
6016 			(*state)->dst.seqhi = (*state)->dst.seqlo +
6017 			    (*state)->src.max_win;
6018 			(*state)->src.wscale = (*state)->dst.wscale = 0;
6019 			pf_set_protostate(*state, PF_PEER_BOTH,
6020 			    TCPS_ESTABLISHED);
6021 			REASON_SET(reason, PFRES_SYNPROXY);
6022 			return (PF_SYNPROXY_DROP);
6023 		}
6024 	}
6025 
6026 	return (PF_PASS);
6027 }
6028 
6029 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)6030 pf_test_state_tcp(struct pf_kstate **state, struct pfi_kkif *kif,
6031     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
6032     u_short *reason)
6033 {
6034 	struct pf_state_key_cmp	 key;
6035 	struct tcphdr		*th = &pd->hdr.tcp;
6036 	int			 copyback = 0;
6037 	int			 action;
6038 	struct pf_state_peer	*src, *dst;
6039 
6040 	bzero(&key, sizeof(key));
6041 	key.af = pd->af;
6042 	key.proto = IPPROTO_TCP;
6043 	if (pd->dir == PF_IN)	{	/* wire side, straight */
6044 		PF_ACPY(&key.addr[0], pd->src, key.af);
6045 		PF_ACPY(&key.addr[1], pd->dst, key.af);
6046 		key.port[0] = th->th_sport;
6047 		key.port[1] = th->th_dport;
6048 	} else {			/* stack side, reverse */
6049 		PF_ACPY(&key.addr[1], pd->src, key.af);
6050 		PF_ACPY(&key.addr[0], pd->dst, key.af);
6051 		key.port[1] = th->th_sport;
6052 		key.port[0] = th->th_dport;
6053 	}
6054 
6055 	STATE_LOOKUP(kif, &key, *state, pd);
6056 
6057 	if (pd->dir == (*state)->direction) {
6058 		src = &(*state)->src;
6059 		dst = &(*state)->dst;
6060 	} else {
6061 		src = &(*state)->dst;
6062 		dst = &(*state)->src;
6063 	}
6064 
6065 	if ((action = pf_synproxy(pd, state, reason)) != PF_PASS)
6066 		return (action);
6067 
6068 	if (dst->state >= TCPS_FIN_WAIT_2 &&
6069 	    src->state >= TCPS_FIN_WAIT_2 &&
6070 	    (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) ||
6071 	    ((th->th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_ACK &&
6072 	    pf_syncookie_check(pd) && pd->dir == PF_IN))) {
6073 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
6074 			printf("pf: state reuse ");
6075 			pf_print_state(*state);
6076 			pf_print_flags(th->th_flags);
6077 			printf("\n");
6078 		}
6079 		/* XXX make sure it's the same direction ?? */
6080 		pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED);
6081 		pf_unlink_state(*state);
6082 		*state = NULL;
6083 		return (PF_DROP);
6084 	}
6085 
6086 	if ((*state)->state_flags & PFSTATE_SLOPPY) {
6087 		if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP)
6088 			return (PF_DROP);
6089 	} else {
6090 		if (pf_tcp_track_full(state, kif, m, off, pd, reason,
6091 		    &copyback) == PF_DROP)
6092 			return (PF_DROP);
6093 	}
6094 
6095 	/* translate source/destination address, if necessary */
6096 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
6097 		struct pf_state_key *nk = (*state)->key[pd->didx];
6098 
6099 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
6100 		    nk->port[pd->sidx] != th->th_sport)
6101 			pf_change_ap(m, pd->src, &th->th_sport,
6102 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
6103 			    nk->port[pd->sidx], 0, pd->af);
6104 
6105 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
6106 		    nk->port[pd->didx] != th->th_dport)
6107 			pf_change_ap(m, pd->dst, &th->th_dport,
6108 			    pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
6109 			    nk->port[pd->didx], 0, pd->af);
6110 		copyback = 1;
6111 	}
6112 
6113 	/* Copyback sequence modulation or stateful scrub changes if needed */
6114 	if (copyback)
6115 		m_copyback(m, off, sizeof(*th), (caddr_t)th);
6116 
6117 	return (PF_PASS);
6118 }
6119 
6120 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)6121 pf_test_state_udp(struct pf_kstate **state, struct pfi_kkif *kif,
6122     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
6123 {
6124 	struct pf_state_peer	*src, *dst;
6125 	struct pf_state_key_cmp	 key;
6126 	struct udphdr		*uh = &pd->hdr.udp;
6127 	uint8_t			 psrc, pdst;
6128 
6129 	bzero(&key, sizeof(key));
6130 	key.af = pd->af;
6131 	key.proto = IPPROTO_UDP;
6132 	if (pd->dir == PF_IN)	{	/* wire side, straight */
6133 		PF_ACPY(&key.addr[0], pd->src, key.af);
6134 		PF_ACPY(&key.addr[1], pd->dst, key.af);
6135 		key.port[0] = uh->uh_sport;
6136 		key.port[1] = uh->uh_dport;
6137 	} else {			/* stack side, reverse */
6138 		PF_ACPY(&key.addr[1], pd->src, key.af);
6139 		PF_ACPY(&key.addr[0], pd->dst, key.af);
6140 		key.port[1] = uh->uh_sport;
6141 		key.port[0] = uh->uh_dport;
6142 	}
6143 
6144 	STATE_LOOKUP(kif, &key, *state, pd);
6145 
6146 	if (pd->dir == (*state)->direction) {
6147 		src = &(*state)->src;
6148 		dst = &(*state)->dst;
6149 		psrc = PF_PEER_SRC;
6150 		pdst = PF_PEER_DST;
6151 	} else {
6152 		src = &(*state)->dst;
6153 		dst = &(*state)->src;
6154 		psrc = PF_PEER_DST;
6155 		pdst = PF_PEER_SRC;
6156 	}
6157 
6158 	/* update states */
6159 	if (src->state < PFUDPS_SINGLE)
6160 		pf_set_protostate(*state, psrc, PFUDPS_SINGLE);
6161 	if (dst->state == PFUDPS_SINGLE)
6162 		pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE);
6163 
6164 	/* update expire time */
6165 	(*state)->expire = time_uptime;
6166 	if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
6167 		(*state)->timeout = PFTM_UDP_MULTIPLE;
6168 	else
6169 		(*state)->timeout = PFTM_UDP_SINGLE;
6170 
6171 	/* translate source/destination address, if necessary */
6172 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
6173 		struct pf_state_key *nk = (*state)->key[pd->didx];
6174 
6175 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
6176 		    nk->port[pd->sidx] != uh->uh_sport)
6177 			pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
6178 			    &uh->uh_sum, &nk->addr[pd->sidx],
6179 			    nk->port[pd->sidx], 1, pd->af);
6180 
6181 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
6182 		    nk->port[pd->didx] != uh->uh_dport)
6183 			pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
6184 			    &uh->uh_sum, &nk->addr[pd->didx],
6185 			    nk->port[pd->didx], 1, pd->af);
6186 		m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
6187 	}
6188 
6189 	return (PF_PASS);
6190 }
6191 
6192 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)6193 pf_test_state_sctp(struct pf_kstate **state, struct pfi_kkif *kif,
6194     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
6195 {
6196 	struct pf_state_key_cmp	 key;
6197 	struct pf_state_peer	*src, *dst;
6198 	struct sctphdr		*sh = &pd->hdr.sctp;
6199 	u_int8_t		 psrc; //, pdst;
6200 
6201 	bzero(&key, sizeof(key));
6202 	key.af = pd->af;
6203 	key.proto = IPPROTO_SCTP;
6204 	if (pd->dir == PF_IN)	{	/* wire side, straight */
6205 		PF_ACPY(&key.addr[0], pd->src, key.af);
6206 		PF_ACPY(&key.addr[1], pd->dst, key.af);
6207 		key.port[0] = sh->src_port;
6208 		key.port[1] = sh->dest_port;
6209 	} else {			/* stack side, reverse */
6210 		PF_ACPY(&key.addr[1], pd->src, key.af);
6211 		PF_ACPY(&key.addr[0], pd->dst, key.af);
6212 		key.port[1] = sh->src_port;
6213 		key.port[0] = sh->dest_port;
6214 	}
6215 
6216 	STATE_LOOKUP(kif, &key, *state, pd);
6217 
6218 	if (pd->dir == (*state)->direction) {
6219 		src = &(*state)->src;
6220 		dst = &(*state)->dst;
6221 		psrc = PF_PEER_SRC;
6222 	} else {
6223 		src = &(*state)->dst;
6224 		dst = &(*state)->src;
6225 		psrc = PF_PEER_DST;
6226 	}
6227 
6228 	if ((src->state >= SCTP_SHUTDOWN_SENT || src->state == SCTP_CLOSED) &&
6229 	    (dst->state >= SCTP_SHUTDOWN_SENT || dst->state == SCTP_CLOSED) &&
6230 	    pd->sctp_flags & PFDESC_SCTP_INIT) {
6231 		pf_set_protostate(*state, PF_PEER_BOTH, SCTP_CLOSED);
6232 		pf_unlink_state(*state);
6233 		*state = NULL;
6234 		return (PF_DROP);
6235 	}
6236 
6237 	if (src->scrub != NULL) {
6238 		if (src->scrub->pfss_v_tag == 0) {
6239 			src->scrub->pfss_v_tag = pd->hdr.sctp.v_tag;
6240 		} else  if (src->scrub->pfss_v_tag != pd->hdr.sctp.v_tag)
6241 			return (PF_DROP);
6242 	}
6243 
6244 	/* Track state. */
6245 	if (pd->sctp_flags & PFDESC_SCTP_INIT) {
6246 		if (src->state < SCTP_COOKIE_WAIT) {
6247 			pf_set_protostate(*state, psrc, SCTP_COOKIE_WAIT);
6248 			(*state)->timeout = PFTM_SCTP_OPENING;
6249 		}
6250 	}
6251 	if (pd->sctp_flags & PFDESC_SCTP_INIT_ACK) {
6252 		MPASS(dst->scrub != NULL);
6253 		if (dst->scrub->pfss_v_tag == 0)
6254 			dst->scrub->pfss_v_tag = pd->sctp_initiate_tag;
6255 	}
6256 
6257 	/*
6258 	 * Bind to the correct interface if we're if-bound. For multihomed
6259 	 * extra associations we don't know which interface that will be until
6260 	 * here, so we've inserted the state on V_pf_all. Fix that now.
6261 	 */
6262 	if ((*state)->kif == V_pfi_all &&
6263 	    (*state)->rule.ptr->rule_flag & PFRULE_IFBOUND)
6264 		(*state)->kif = kif;
6265 
6266 	if (pd->sctp_flags & (PFDESC_SCTP_COOKIE | PFDESC_SCTP_HEARTBEAT_ACK)) {
6267 		if (src->state < SCTP_ESTABLISHED) {
6268 			pf_set_protostate(*state, psrc, SCTP_ESTABLISHED);
6269 			(*state)->timeout = PFTM_SCTP_ESTABLISHED;
6270 		}
6271 	}
6272 	if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN | PFDESC_SCTP_ABORT |
6273 	    PFDESC_SCTP_SHUTDOWN_COMPLETE)) {
6274 		if (src->state < SCTP_SHUTDOWN_PENDING) {
6275 			pf_set_protostate(*state, psrc, SCTP_SHUTDOWN_PENDING);
6276 			(*state)->timeout = PFTM_SCTP_CLOSING;
6277 		}
6278 	}
6279 	if (pd->sctp_flags & (PFDESC_SCTP_SHUTDOWN_COMPLETE)) {
6280 		pf_set_protostate(*state, psrc, SCTP_CLOSED);
6281 		(*state)->timeout = PFTM_SCTP_CLOSED;
6282 	}
6283 
6284 	(*state)->expire = time_uptime;
6285 
6286 	/* translate source/destination address, if necessary */
6287 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
6288 		uint16_t checksum = 0;
6289 		struct pf_state_key *nk = (*state)->key[pd->didx];
6290 
6291 		if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
6292 		    nk->port[pd->sidx] != pd->hdr.sctp.src_port) {
6293 			pf_change_ap(m, pd->src, &pd->hdr.sctp.src_port,
6294 			    pd->ip_sum, &checksum, &nk->addr[pd->sidx],
6295 			    nk->port[pd->sidx], 1, pd->af);
6296 		}
6297 
6298 		if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
6299 		    nk->port[pd->didx] != pd->hdr.sctp.dest_port) {
6300 			pf_change_ap(m, pd->dst, &pd->hdr.sctp.dest_port,
6301 			    pd->ip_sum, &checksum, &nk->addr[pd->didx],
6302 			    nk->port[pd->didx], 1, pd->af);
6303 		}
6304 	}
6305 
6306 	return (PF_PASS);
6307 }
6308 
6309 static void
pf_sctp_multihome_detach_addr(const struct pf_kstate * s)6310 pf_sctp_multihome_detach_addr(const struct pf_kstate *s)
6311 {
6312 	struct pf_sctp_endpoint key;
6313 	struct pf_sctp_endpoint *ep;
6314 	struct pf_state_key *sks = s->key[PF_SK_STACK];
6315 	struct pf_sctp_source *i, *tmp;
6316 
6317 	if (sks == NULL || sks->proto != IPPROTO_SCTP || s->dst.scrub == NULL)
6318 		return;
6319 
6320 	PF_SCTP_ENDPOINTS_LOCK();
6321 
6322 	key.v_tag = s->dst.scrub->pfss_v_tag;
6323 	ep  = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6324 	if (ep != NULL) {
6325 		TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) {
6326 			if (pf_addr_cmp(&i->addr,
6327 			    &s->key[PF_SK_WIRE]->addr[s->direction == PF_OUT],
6328 			    s->key[PF_SK_WIRE]->af) == 0) {
6329 				SDT_PROBE3(pf, sctp, multihome, remove,
6330 				    key.v_tag, s, i);
6331 				TAILQ_REMOVE(&ep->sources, i, entry);
6332 				free(i, M_PFTEMP);
6333 				break;
6334 			}
6335 		}
6336 
6337 		if (TAILQ_EMPTY(&ep->sources)) {
6338 			RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
6339 			free(ep, M_PFTEMP);
6340 		}
6341 	}
6342 
6343 	/* Other direction. */
6344 	key.v_tag = s->src.scrub->pfss_v_tag;
6345 	ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6346 	if (ep != NULL) {
6347 		TAILQ_FOREACH_SAFE(i, &ep->sources, entry, tmp) {
6348 			if (pf_addr_cmp(&i->addr,
6349 			    &s->key[PF_SK_WIRE]->addr[s->direction == PF_IN],
6350 			    s->key[PF_SK_WIRE]->af) == 0) {
6351 				SDT_PROBE3(pf, sctp, multihome, remove,
6352 				    key.v_tag, s, i);
6353 				TAILQ_REMOVE(&ep->sources, i, entry);
6354 				free(i, M_PFTEMP);
6355 				break;
6356 			}
6357 		}
6358 
6359 		if (TAILQ_EMPTY(&ep->sources)) {
6360 			RB_REMOVE(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
6361 			free(ep, M_PFTEMP);
6362 		}
6363 	}
6364 
6365 	PF_SCTP_ENDPOINTS_UNLOCK();
6366 }
6367 
6368 static void
pf_sctp_multihome_add_addr(struct pf_pdesc * pd,struct pf_addr * a,uint32_t v_tag)6369 pf_sctp_multihome_add_addr(struct pf_pdesc *pd, struct pf_addr *a, uint32_t v_tag)
6370 {
6371 	struct pf_sctp_endpoint key = {
6372 		.v_tag = v_tag,
6373 	};
6374 	struct pf_sctp_source *i;
6375 	struct pf_sctp_endpoint *ep;
6376 
6377 	PF_SCTP_ENDPOINTS_LOCK();
6378 
6379 	ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6380 	if (ep == NULL) {
6381 		ep = malloc(sizeof(struct pf_sctp_endpoint),
6382 		    M_PFTEMP, M_NOWAIT);
6383 		if (ep == NULL) {
6384 			PF_SCTP_ENDPOINTS_UNLOCK();
6385 			return;
6386 		}
6387 
6388 		ep->v_tag = v_tag;
6389 		TAILQ_INIT(&ep->sources);
6390 		RB_INSERT(pf_sctp_endpoints, &V_pf_sctp_endpoints, ep);
6391 	}
6392 
6393 	/* Avoid inserting duplicates. */
6394 	TAILQ_FOREACH(i, &ep->sources, entry) {
6395 		if (pf_addr_cmp(&i->addr, a, pd->af) == 0) {
6396 			PF_SCTP_ENDPOINTS_UNLOCK();
6397 			return;
6398 		}
6399 	}
6400 
6401 	i = malloc(sizeof(*i), M_PFTEMP, M_NOWAIT);
6402 	if (i == NULL) {
6403 		PF_SCTP_ENDPOINTS_UNLOCK();
6404 		return;
6405 	}
6406 
6407 	i->af = pd->af;
6408 	memcpy(&i->addr, a, sizeof(*a));
6409 	TAILQ_INSERT_TAIL(&ep->sources, i, entry);
6410 	SDT_PROBE2(pf, sctp, multihome, add, v_tag, i);
6411 
6412 	PF_SCTP_ENDPOINTS_UNLOCK();
6413 }
6414 
6415 static void
pf_sctp_multihome_delayed(struct pf_pdesc * pd,int off,struct pfi_kkif * kif,struct pf_kstate * s,int action)6416 pf_sctp_multihome_delayed(struct pf_pdesc *pd, int off, struct pfi_kkif *kif,
6417     struct pf_kstate *s, int action)
6418 {
6419 	struct pf_sctp_multihome_job	*j, *tmp;
6420 	struct pf_sctp_source		*i;
6421 	int			 ret __unused;;
6422 	struct pf_kstate	*sm = NULL;
6423 	struct pf_krule		*ra = NULL;
6424 	struct pf_krule		*r = &V_pf_default_rule;
6425 	struct pf_kruleset	*rs = NULL;
6426 	bool do_extra = true;
6427 
6428 	PF_RULES_RLOCK_TRACKER;
6429 
6430 again:
6431 	TAILQ_FOREACH_SAFE(j, &pd->sctp_multihome_jobs, next, tmp) {
6432 		if (s == NULL || action != PF_PASS)
6433 			goto free;
6434 
6435 		/* Confirm we don't recurse here. */
6436 		MPASS(! (pd->sctp_flags & PFDESC_SCTP_ADD_IP));
6437 
6438 		switch (j->op) {
6439 		case  SCTP_ADD_IP_ADDRESS: {
6440 			uint32_t v_tag = pd->sctp_initiate_tag;
6441 
6442 			if (v_tag == 0) {
6443 				if (s->direction == pd->dir)
6444 					v_tag = s->src.scrub->pfss_v_tag;
6445 				else
6446 					v_tag = s->dst.scrub->pfss_v_tag;
6447 			}
6448 
6449 			/*
6450 			 * Avoid duplicating states. We'll already have
6451 			 * created a state based on the source address of
6452 			 * the packet, but SCTP endpoints may also list this
6453 			 * address again in the INIT(_ACK) parameters.
6454 			 */
6455 			if (pf_addr_cmp(&j->src, pd->src, pd->af) == 0) {
6456 				break;
6457 			}
6458 
6459 			j->pd.sctp_flags |= PFDESC_SCTP_ADD_IP;
6460 			PF_RULES_RLOCK();
6461 			sm = NULL;
6462 			if (s->rule.ptr->rule_flag & PFRULE_ALLOW_RELATED) {
6463 				j->pd.related_rule = s->rule.ptr;
6464 			}
6465 			ret = pf_test_rule(&r, &sm, kif,
6466 			    j->m, off, &j->pd, &ra, &rs, NULL);
6467 			PF_RULES_RUNLOCK();
6468 			SDT_PROBE4(pf, sctp, multihome, test, kif, r, j->m, ret);
6469 			if (ret != PF_DROP && sm != NULL) {
6470 				/* Inherit v_tag values. */
6471 				if (sm->direction == s->direction) {
6472 					sm->src.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag;
6473 					sm->dst.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag;
6474 				} else {
6475 					sm->src.scrub->pfss_v_tag = s->dst.scrub->pfss_v_tag;
6476 					sm->dst.scrub->pfss_v_tag = s->src.scrub->pfss_v_tag;
6477 				}
6478 				PF_STATE_UNLOCK(sm);
6479 			} else {
6480 				/* If we try duplicate inserts? */
6481 				break;
6482 			}
6483 
6484 			/* Only add the address if we've actually allowed the state. */
6485 			pf_sctp_multihome_add_addr(pd, &j->src, v_tag);
6486 
6487 			if (! do_extra) {
6488 				break;
6489 			}
6490 			/*
6491 			 * We need to do this for each of our source addresses.
6492 			 * Find those based on the verification tag.
6493 			 */
6494 			struct pf_sctp_endpoint key = {
6495 				.v_tag = pd->hdr.sctp.v_tag,
6496 			};
6497 			struct pf_sctp_endpoint *ep;
6498 
6499 			PF_SCTP_ENDPOINTS_LOCK();
6500 			ep = RB_FIND(pf_sctp_endpoints, &V_pf_sctp_endpoints, &key);
6501 			if (ep == NULL) {
6502 				PF_SCTP_ENDPOINTS_UNLOCK();
6503 				break;
6504 			}
6505 			MPASS(ep != NULL);
6506 
6507 			TAILQ_FOREACH(i, &ep->sources, entry) {
6508 				struct pf_sctp_multihome_job *nj;
6509 
6510 				/* SCTP can intermingle IPv4 and IPv6. */
6511 				if (i->af != pd->af)
6512 					continue;
6513 
6514 				nj = malloc(sizeof(*nj), M_PFTEMP, M_NOWAIT | M_ZERO);
6515 				if (! nj) {
6516 					continue;
6517 				}
6518 				memcpy(&nj->pd, &j->pd, sizeof(j->pd));
6519 				memcpy(&nj->src, &j->src, sizeof(nj->src));
6520 				nj->pd.src = &nj->src;
6521 				// New destination address!
6522 				memcpy(&nj->dst, &i->addr, sizeof(nj->dst));
6523 				nj->pd.dst = &nj->dst;
6524 				nj->m = j->m;
6525 				nj->op = j->op;
6526 
6527 				TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, nj, next);
6528 			}
6529 			PF_SCTP_ENDPOINTS_UNLOCK();
6530 
6531 			break;
6532 		}
6533 		case SCTP_DEL_IP_ADDRESS: {
6534 			struct pf_state_key_cmp key;
6535 			uint8_t psrc;
6536 
6537 			bzero(&key, sizeof(key));
6538 			key.af = j->pd.af;
6539 			key.proto = IPPROTO_SCTP;
6540 			if (j->pd.dir == PF_IN)	{	/* wire side, straight */
6541 				PF_ACPY(&key.addr[0], j->pd.src, key.af);
6542 				PF_ACPY(&key.addr[1], j->pd.dst, key.af);
6543 				key.port[0] = j->pd.hdr.sctp.src_port;
6544 				key.port[1] = j->pd.hdr.sctp.dest_port;
6545 			} else {			/* stack side, reverse */
6546 				PF_ACPY(&key.addr[1], j->pd.src, key.af);
6547 				PF_ACPY(&key.addr[0], j->pd.dst, key.af);
6548 				key.port[1] = j->pd.hdr.sctp.src_port;
6549 				key.port[0] = j->pd.hdr.sctp.dest_port;
6550 			}
6551 
6552 			sm = pf_find_state(kif, &key, j->pd.dir);
6553 			if (sm != NULL) {
6554 				PF_STATE_LOCK_ASSERT(sm);
6555 				if (j->pd.dir == sm->direction) {
6556 					psrc = PF_PEER_SRC;
6557 				} else {
6558 					psrc = PF_PEER_DST;
6559 				}
6560 				pf_set_protostate(sm, psrc, SCTP_SHUTDOWN_PENDING);
6561 				sm->timeout = PFTM_SCTP_CLOSING;
6562 				PF_STATE_UNLOCK(sm);
6563 			}
6564 			break;
6565 		default:
6566 			panic("Unknown op %#x", j->op);
6567 		}
6568 	}
6569 
6570 	free:
6571 		TAILQ_REMOVE(&pd->sctp_multihome_jobs, j, next);
6572 		free(j, M_PFTEMP);
6573 	}
6574 
6575 	/* We may have inserted extra work while processing the list. */
6576 	if (! TAILQ_EMPTY(&pd->sctp_multihome_jobs)) {
6577 		do_extra = false;
6578 		goto again;
6579 	}
6580 }
6581 
6582 static int
pf_multihome_scan(struct mbuf * m,int start,int len,struct pf_pdesc * pd,struct pfi_kkif * kif,int op)6583 pf_multihome_scan(struct mbuf *m, int start, int len, struct pf_pdesc *pd,
6584     struct pfi_kkif *kif, int op)
6585 {
6586 	int			 off = 0;
6587 	struct pf_sctp_multihome_job	*job;
6588 
6589 	SDT_PROBE4(pf, sctp, multihome_scan, entry, start, len, pd, op);
6590 
6591 	while (off < len) {
6592 		struct sctp_paramhdr h;
6593 
6594 		if (!pf_pull_hdr(m, start + off, &h, sizeof(h), NULL, NULL,
6595 		    pd->af))
6596 			return (PF_DROP);
6597 
6598 		/* Parameters are at least 4 bytes. */
6599 		if (ntohs(h.param_length) < 4)
6600 			return (PF_DROP);
6601 
6602 		SDT_PROBE2(pf, sctp, multihome_scan, param, ntohs(h.param_type),
6603 		    ntohs(h.param_length));
6604 
6605 		switch (ntohs(h.param_type)) {
6606 		case  SCTP_IPV4_ADDRESS: {
6607 			struct in_addr t;
6608 
6609 			if (ntohs(h.param_length) !=
6610 			    (sizeof(struct sctp_paramhdr) + sizeof(t)))
6611 				return (PF_DROP);
6612 
6613 			if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t),
6614 			    NULL, NULL, pd->af))
6615 				return (PF_DROP);
6616 
6617 			if (in_nullhost(t))
6618 				t.s_addr = pd->src->v4.s_addr;
6619 
6620 			/*
6621 			 * We hold the state lock (idhash) here, which means
6622 			 * that we can't acquire the keyhash, or we'll get a
6623 			 * LOR (and potentially double-lock things too). We also
6624 			 * can't release the state lock here, so instead we'll
6625 			 * enqueue this for async handling.
6626 			 * There's a relatively small race here, in that a
6627 			 * packet using the new addresses could arrive already,
6628 			 * but that's just though luck for it.
6629 			 */
6630 			job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO);
6631 			if (! job)
6632 				return (PF_DROP);
6633 
6634 			SDT_PROBE2(pf, sctp, multihome_scan, ipv4, &t, op);
6635 
6636 			memcpy(&job->pd, pd, sizeof(*pd));
6637 
6638 			// New source address!
6639 			memcpy(&job->src, &t, sizeof(t));
6640 			job->pd.src = &job->src;
6641 			memcpy(&job->dst, pd->dst, sizeof(job->dst));
6642 			job->pd.dst = &job->dst;
6643 			job->m = m;
6644 			job->op = op;
6645 
6646 			TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next);
6647 			break;
6648 		}
6649 #ifdef INET6
6650 		case SCTP_IPV6_ADDRESS: {
6651 			struct in6_addr t;
6652 
6653 			if (ntohs(h.param_length) !=
6654 			    (sizeof(struct sctp_paramhdr) + sizeof(t)))
6655 				return (PF_DROP);
6656 
6657 			if (!pf_pull_hdr(m, start + off + sizeof(h), &t, sizeof(t),
6658 			    NULL, NULL, pd->af))
6659 				return (PF_DROP);
6660 			if (memcmp(&t, &pd->src->v6, sizeof(t)) == 0)
6661 				break;
6662 			if (memcmp(&t, &in6addr_any, sizeof(t)) == 0)
6663 				memcpy(&t, &pd->src->v6, sizeof(t));
6664 
6665 			job = malloc(sizeof(*job), M_PFTEMP, M_NOWAIT | M_ZERO);
6666 			if (! job)
6667 				return (PF_DROP);
6668 
6669 			SDT_PROBE2(pf, sctp, multihome_scan, ipv6, &t, op);
6670 
6671 			memcpy(&job->pd, pd, sizeof(*pd));
6672 			memcpy(&job->src, &t, sizeof(t));
6673 			job->pd.src = &job->src;
6674 			memcpy(&job->dst, pd->dst, sizeof(job->dst));
6675 			job->pd.dst = &job->dst;
6676 			job->m = m;
6677 			job->op = op;
6678 
6679 			TAILQ_INSERT_TAIL(&pd->sctp_multihome_jobs, job, next);
6680 			break;
6681 		}
6682 #endif
6683 		case SCTP_ADD_IP_ADDRESS: {
6684 			int ret;
6685 			struct sctp_asconf_paramhdr ah;
6686 
6687 			if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah),
6688 			    NULL, NULL, pd->af))
6689 				return (PF_DROP);
6690 
6691 			ret = pf_multihome_scan(m, start + off + sizeof(ah),
6692 			    ntohs(ah.ph.param_length) - sizeof(ah), pd, kif,
6693 			    SCTP_ADD_IP_ADDRESS);
6694 			if (ret != PF_PASS)
6695 				return (ret);
6696 			break;
6697 		}
6698 		case SCTP_DEL_IP_ADDRESS: {
6699 			int ret;
6700 			struct sctp_asconf_paramhdr ah;
6701 
6702 			if (!pf_pull_hdr(m, start + off, &ah, sizeof(ah),
6703 			    NULL, NULL, pd->af))
6704 				return (PF_DROP);
6705 			ret = pf_multihome_scan(m, start + off + sizeof(ah),
6706 			    ntohs(ah.ph.param_length) - sizeof(ah), pd, kif,
6707 			    SCTP_DEL_IP_ADDRESS);
6708 			if (ret != PF_PASS)
6709 				return (ret);
6710 			break;
6711 		}
6712 		default:
6713 			break;
6714 		}
6715 
6716 		off += roundup(ntohs(h.param_length), 4);
6717 	}
6718 
6719 	return (PF_PASS);
6720 }
6721 
6722 int
pf_multihome_scan_init(struct mbuf * m,int start,int len,struct pf_pdesc * pd,struct pfi_kkif * kif)6723 pf_multihome_scan_init(struct mbuf *m, int start, int len, struct pf_pdesc *pd,
6724     struct pfi_kkif *kif)
6725 {
6726 	start += sizeof(struct sctp_init_chunk);
6727 	len -= sizeof(struct sctp_init_chunk);
6728 
6729 	return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS));
6730 }
6731 
6732 int
pf_multihome_scan_asconf(struct mbuf * m,int start,int len,struct pf_pdesc * pd,struct pfi_kkif * kif)6733 pf_multihome_scan_asconf(struct mbuf *m, int start, int len,
6734     struct pf_pdesc *pd, struct pfi_kkif *kif)
6735 {
6736 	start += sizeof(struct sctp_asconf_chunk);
6737 	len -= sizeof(struct sctp_asconf_chunk);
6738 
6739 	return (pf_multihome_scan(m, start, len, pd, kif, SCTP_ADD_IP_ADDRESS));
6740 }
6741 
6742 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)6743 pf_icmp_state_lookup(struct pf_state_key_cmp *key, struct pf_pdesc *pd,
6744     struct pf_kstate **state, struct mbuf *m, int off, int direction,
6745     struct pfi_kkif *kif, u_int16_t icmpid, u_int16_t type, int icmp_dir,
6746     int *iidx, int multi, int inner)
6747 {
6748 	key->af = pd->af;
6749 	key->proto = pd->proto;
6750 	if (icmp_dir == PF_IN) {
6751 		*iidx = pd->sidx;
6752 		key->port[pd->sidx] = icmpid;
6753 		key->port[pd->didx] = type;
6754 	} else {
6755 		*iidx = pd->didx;
6756 		key->port[pd->sidx] = type;
6757 		key->port[pd->didx] = icmpid;
6758 	}
6759 	if (pf_state_key_addr_setup(pd, m, off, key, pd->sidx, pd->src,
6760 	    pd->didx, pd->dst, multi))
6761 		return (PF_DROP);
6762 
6763 	STATE_LOOKUP(kif, key, *state, pd);
6764 
6765 	if ((*state)->state_flags & PFSTATE_SLOPPY)
6766 		return (-1);
6767 
6768 	/* Is this ICMP message flowing in right direction? */
6769 	if ((*state)->rule.ptr->type &&
6770 	    (((!inner && (*state)->direction == direction) ||
6771 	    (inner && (*state)->direction != direction)) ?
6772 	    PF_IN : PF_OUT) != icmp_dir) {
6773 		if (V_pf_status.debug >= PF_DEBUG_MISC) {
6774 			printf("pf: icmp type %d in wrong direction (%d): ",
6775 			    ntohs(type), icmp_dir);
6776 			pf_print_state(*state);
6777 			printf("\n");
6778 		}
6779 		PF_STATE_UNLOCK(*state);
6780 		*state = NULL;
6781 		return (PF_DROP);
6782 	}
6783 	return (-1);
6784 }
6785 
6786 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)6787 pf_test_state_icmp(struct pf_kstate **state, struct pfi_kkif *kif,
6788     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
6789 {
6790 	struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
6791 	u_int16_t	*icmpsum, virtual_id, virtual_type;
6792 	u_int8_t	 icmptype, icmpcode;
6793 	int		 icmp_dir, iidx, ret, multi;
6794 	struct pf_state_key_cmp key;
6795 #ifdef INET
6796 	u_int16_t	 icmpid;
6797 #endif
6798 
6799 	MPASS(*state == NULL);
6800 
6801 	bzero(&key, sizeof(key));
6802 	switch (pd->proto) {
6803 #ifdef INET
6804 	case IPPROTO_ICMP:
6805 		icmptype = pd->hdr.icmp.icmp_type;
6806 		icmpcode = pd->hdr.icmp.icmp_code;
6807 		icmpid = pd->hdr.icmp.icmp_id;
6808 		icmpsum = &pd->hdr.icmp.icmp_cksum;
6809 		break;
6810 #endif /* INET */
6811 #ifdef INET6
6812 	case IPPROTO_ICMPV6:
6813 		icmptype = pd->hdr.icmp6.icmp6_type;
6814 		icmpcode = pd->hdr.icmp6.icmp6_code;
6815 #ifdef INET
6816 		icmpid = pd->hdr.icmp6.icmp6_id;
6817 #endif
6818 		icmpsum = &pd->hdr.icmp6.icmp6_cksum;
6819 		break;
6820 #endif /* INET6 */
6821 	}
6822 
6823 	if (pf_icmp_mapping(pd, icmptype, &icmp_dir, &multi,
6824 	    &virtual_id, &virtual_type) == 0) {
6825 		/*
6826 		 * ICMP query/reply message not related to a TCP/UDP/SCTP
6827 		 * packet. Search for an ICMP state.
6828 		 */
6829 		ret = pf_icmp_state_lookup(&key, pd, state, m, off, pd->dir,
6830 		    kif, virtual_id, virtual_type, icmp_dir, &iidx,
6831 		    PF_ICMP_MULTI_NONE, 0);
6832 		if (ret >= 0) {
6833 			MPASS(*state == NULL);
6834 			if (ret == PF_DROP && pd->af == AF_INET6 &&
6835 			    icmp_dir == PF_OUT) {
6836 				ret = pf_icmp_state_lookup(&key, pd, state, m, off,
6837 				    pd->dir, kif, virtual_id, virtual_type,
6838 				    icmp_dir, &iidx, multi, 0);
6839 				if (ret >= 0) {
6840 					MPASS(*state == NULL);
6841 					return (ret);
6842 				}
6843 			} else
6844 				return (ret);
6845 		}
6846 
6847 		(*state)->expire = time_uptime;
6848 		(*state)->timeout = PFTM_ICMP_ERROR_REPLY;
6849 
6850 		/* translate source/destination address, if necessary */
6851 		if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
6852 			struct pf_state_key *nk = (*state)->key[pd->didx];
6853 
6854 			switch (pd->af) {
6855 #ifdef INET
6856 			case AF_INET:
6857 				if (PF_ANEQ(pd->src,
6858 				    &nk->addr[pd->sidx], AF_INET))
6859 					pf_change_a(&saddr->v4.s_addr,
6860 					    pd->ip_sum,
6861 					    nk->addr[pd->sidx].v4.s_addr, 0);
6862 
6863 				if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
6864 				    AF_INET))
6865 					pf_change_a(&daddr->v4.s_addr,
6866 					    pd->ip_sum,
6867 					    nk->addr[pd->didx].v4.s_addr, 0);
6868 
6869 				if (nk->port[iidx] !=
6870 				    pd->hdr.icmp.icmp_id) {
6871 					pd->hdr.icmp.icmp_cksum =
6872 					    pf_cksum_fixup(
6873 					    pd->hdr.icmp.icmp_cksum, icmpid,
6874 					    nk->port[iidx], 0);
6875 					pd->hdr.icmp.icmp_id =
6876 					    nk->port[iidx];
6877 				}
6878 
6879 				m_copyback(m, off, ICMP_MINLEN,
6880 				    (caddr_t )&pd->hdr.icmp);
6881 				break;
6882 #endif /* INET */
6883 #ifdef INET6
6884 			case AF_INET6:
6885 				if (PF_ANEQ(pd->src,
6886 				    &nk->addr[pd->sidx], AF_INET6))
6887 					pf_change_a6(saddr,
6888 					    &pd->hdr.icmp6.icmp6_cksum,
6889 					    &nk->addr[pd->sidx], 0);
6890 
6891 				if (PF_ANEQ(pd->dst,
6892 				    &nk->addr[pd->didx], AF_INET6))
6893 					pf_change_a6(daddr,
6894 					    &pd->hdr.icmp6.icmp6_cksum,
6895 					    &nk->addr[pd->didx], 0);
6896 
6897 				m_copyback(m, off, sizeof(struct icmp6_hdr),
6898 				    (caddr_t )&pd->hdr.icmp6);
6899 				break;
6900 #endif /* INET6 */
6901 			}
6902 		}
6903 		return (PF_PASS);
6904 
6905 	} else {
6906 		/*
6907 		 * ICMP error message in response to a TCP/UDP packet.
6908 		 * Extract the inner TCP/UDP header and search for that state.
6909 		 */
6910 
6911 		struct pf_pdesc	pd2;
6912 		bzero(&pd2, sizeof pd2);
6913 #ifdef INET
6914 		struct ip	h2;
6915 #endif /* INET */
6916 #ifdef INET6
6917 		struct ip6_hdr	h2_6;
6918 		int		terminal = 0;
6919 #endif /* INET6 */
6920 		int		ipoff2 = 0;
6921 		int		off2 = 0;
6922 
6923 		pd2.af = pd->af;
6924 		pd2.dir = pd->dir;
6925 		/* Payload packet is from the opposite direction. */
6926 		pd2.sidx = (pd->dir == PF_IN) ? 1 : 0;
6927 		pd2.didx = (pd->dir == PF_IN) ? 0 : 1;
6928 		switch (pd->af) {
6929 #ifdef INET
6930 		case AF_INET:
6931 			/* offset of h2 in mbuf chain */
6932 			ipoff2 = off + ICMP_MINLEN;
6933 
6934 			if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
6935 			    NULL, reason, pd2.af)) {
6936 				DPFPRINTF(PF_DEBUG_MISC,
6937 				    ("pf: ICMP error message too short "
6938 				    "(ip)\n"));
6939 				return (PF_DROP);
6940 			}
6941 			/*
6942 			 * ICMP error messages don't refer to non-first
6943 			 * fragments
6944 			 */
6945 			if (h2.ip_off & htons(IP_OFFMASK)) {
6946 				REASON_SET(reason, PFRES_FRAG);
6947 				return (PF_DROP);
6948 			}
6949 
6950 			/* offset of protocol header that follows h2 */
6951 			off2 = ipoff2 + (h2.ip_hl << 2);
6952 
6953 			pd2.proto = h2.ip_p;
6954 			pd2.src = (struct pf_addr *)&h2.ip_src;
6955 			pd2.dst = (struct pf_addr *)&h2.ip_dst;
6956 			pd2.ip_sum = &h2.ip_sum;
6957 			break;
6958 #endif /* INET */
6959 #ifdef INET6
6960 		case AF_INET6:
6961 			ipoff2 = off + sizeof(struct icmp6_hdr);
6962 
6963 			if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
6964 			    NULL, reason, pd2.af)) {
6965 				DPFPRINTF(PF_DEBUG_MISC,
6966 				    ("pf: ICMP error message too short "
6967 				    "(ip6)\n"));
6968 				return (PF_DROP);
6969 			}
6970 			pd2.proto = h2_6.ip6_nxt;
6971 			pd2.src = (struct pf_addr *)&h2_6.ip6_src;
6972 			pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
6973 			pd2.ip_sum = NULL;
6974 			off2 = ipoff2 + sizeof(h2_6);
6975 			do {
6976 				switch (pd2.proto) {
6977 				case IPPROTO_FRAGMENT:
6978 					/*
6979 					 * ICMPv6 error messages for
6980 					 * non-first fragments
6981 					 */
6982 					REASON_SET(reason, PFRES_FRAG);
6983 					return (PF_DROP);
6984 				case IPPROTO_AH:
6985 				case IPPROTO_HOPOPTS:
6986 				case IPPROTO_ROUTING:
6987 				case IPPROTO_DSTOPTS: {
6988 					/* get next header and header length */
6989 					struct ip6_ext opt6;
6990 
6991 					if (!pf_pull_hdr(m, off2, &opt6,
6992 					    sizeof(opt6), NULL, reason,
6993 					    pd2.af)) {
6994 						DPFPRINTF(PF_DEBUG_MISC,
6995 						    ("pf: ICMPv6 short opt\n"));
6996 						return (PF_DROP);
6997 					}
6998 					if (pd2.proto == IPPROTO_AH)
6999 						off2 += (opt6.ip6e_len + 2) * 4;
7000 					else
7001 						off2 += (opt6.ip6e_len + 1) * 8;
7002 					pd2.proto = opt6.ip6e_nxt;
7003 					/* goto the next header */
7004 					break;
7005 				}
7006 				default:
7007 					terminal++;
7008 					break;
7009 				}
7010 			} while (!terminal);
7011 			break;
7012 #endif /* INET6 */
7013 		}
7014 
7015 		if (PF_ANEQ(pd->dst, pd2.src, pd->af)) {
7016 			if (V_pf_status.debug >= PF_DEBUG_MISC) {
7017 				printf("pf: BAD ICMP %d:%d outer dst: ",
7018 				    icmptype, icmpcode);
7019 				pf_print_host(pd->src, 0, pd->af);
7020 				printf(" -> ");
7021 				pf_print_host(pd->dst, 0, pd->af);
7022 				printf(" inner src: ");
7023 				pf_print_host(pd2.src, 0, pd2.af);
7024 				printf(" -> ");
7025 				pf_print_host(pd2.dst, 0, pd2.af);
7026 				printf("\n");
7027 			}
7028 			REASON_SET(reason, PFRES_BADSTATE);
7029 			return (PF_DROP);
7030 		}
7031 
7032 		switch (pd2.proto) {
7033 		case IPPROTO_TCP: {
7034 			struct tcphdr		 th;
7035 			u_int32_t		 seq;
7036 			struct pf_state_peer	*src, *dst;
7037 			u_int8_t		 dws;
7038 			int			 copyback = 0;
7039 
7040 			/*
7041 			 * Only the first 8 bytes of the TCP header can be
7042 			 * expected. Don't access any TCP header fields after
7043 			 * th_seq, an ackskew test is not possible.
7044 			 */
7045 			if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
7046 			    pd2.af)) {
7047 				DPFPRINTF(PF_DEBUG_MISC,
7048 				    ("pf: ICMP error message too short "
7049 				    "(tcp)\n"));
7050 				return (PF_DROP);
7051 			}
7052 
7053 			key.af = pd2.af;
7054 			key.proto = IPPROTO_TCP;
7055 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
7056 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
7057 			key.port[pd2.sidx] = th.th_sport;
7058 			key.port[pd2.didx] = th.th_dport;
7059 
7060 			STATE_LOOKUP(kif, &key, *state, pd);
7061 
7062 			if (pd->dir == (*state)->direction) {
7063 				src = &(*state)->dst;
7064 				dst = &(*state)->src;
7065 			} else {
7066 				src = &(*state)->src;
7067 				dst = &(*state)->dst;
7068 			}
7069 
7070 			if (src->wscale && dst->wscale)
7071 				dws = dst->wscale & PF_WSCALE_MASK;
7072 			else
7073 				dws = 0;
7074 
7075 			/* Demodulate sequence number */
7076 			seq = ntohl(th.th_seq) - src->seqdiff;
7077 			if (src->seqdiff) {
7078 				pf_change_a(&th.th_seq, icmpsum,
7079 				    htonl(seq), 0);
7080 				copyback = 1;
7081 			}
7082 
7083 			if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
7084 			    (!SEQ_GEQ(src->seqhi, seq) ||
7085 			    !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
7086 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
7087 					printf("pf: BAD ICMP %d:%d ",
7088 					    icmptype, icmpcode);
7089 					pf_print_host(pd->src, 0, pd->af);
7090 					printf(" -> ");
7091 					pf_print_host(pd->dst, 0, pd->af);
7092 					printf(" state: ");
7093 					pf_print_state(*state);
7094 					printf(" seq=%u\n", seq);
7095 				}
7096 				REASON_SET(reason, PFRES_BADSTATE);
7097 				return (PF_DROP);
7098 			} else {
7099 				if (V_pf_status.debug >= PF_DEBUG_MISC) {
7100 					printf("pf: OK ICMP %d:%d ",
7101 					    icmptype, icmpcode);
7102 					pf_print_host(pd->src, 0, pd->af);
7103 					printf(" -> ");
7104 					pf_print_host(pd->dst, 0, pd->af);
7105 					printf(" state: ");
7106 					pf_print_state(*state);
7107 					printf(" seq=%u\n", seq);
7108 				}
7109 			}
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] != th.th_sport)
7120 					pf_change_icmp(pd2.src, &th.th_sport,
7121 					    daddr, &nk->addr[pd2.sidx],
7122 					    nk->port[pd2.sidx], NULL,
7123 					    pd2.ip_sum, icmpsum,
7124 					    pd->ip_sum, 0, pd2.af);
7125 
7126 				if (PF_ANEQ(pd2.dst,
7127 				    &nk->addr[pd2.didx], pd2.af) ||
7128 				    nk->port[pd2.didx] != th.th_dport)
7129 					pf_change_icmp(pd2.dst, &th.th_dport,
7130 					    saddr, &nk->addr[pd2.didx],
7131 					    nk->port[pd2.didx], NULL,
7132 					    pd2.ip_sum, icmpsum,
7133 					    pd->ip_sum, 0, pd2.af);
7134 				copyback = 1;
7135 			}
7136 
7137 			if (copyback) {
7138 				switch (pd2.af) {
7139 #ifdef INET
7140 				case AF_INET:
7141 					m_copyback(m, off, ICMP_MINLEN,
7142 					    (caddr_t )&pd->hdr.icmp);
7143 					m_copyback(m, ipoff2, sizeof(h2),
7144 					    (caddr_t )&h2);
7145 					break;
7146 #endif /* INET */
7147 #ifdef INET6
7148 				case AF_INET6:
7149 					m_copyback(m, off,
7150 					    sizeof(struct icmp6_hdr),
7151 					    (caddr_t )&pd->hdr.icmp6);
7152 					m_copyback(m, ipoff2, sizeof(h2_6),
7153 					    (caddr_t )&h2_6);
7154 					break;
7155 #endif /* INET6 */
7156 				}
7157 				m_copyback(m, off2, 8, (caddr_t)&th);
7158 			}
7159 
7160 			return (PF_PASS);
7161 			break;
7162 		}
7163 		case IPPROTO_UDP: {
7164 			struct udphdr		uh;
7165 
7166 			if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
7167 			    NULL, reason, pd2.af)) {
7168 				DPFPRINTF(PF_DEBUG_MISC,
7169 				    ("pf: ICMP error message too short "
7170 				    "(udp)\n"));
7171 				return (PF_DROP);
7172 			}
7173 
7174 			key.af = pd2.af;
7175 			key.proto = IPPROTO_UDP;
7176 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
7177 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
7178 			key.port[pd2.sidx] = uh.uh_sport;
7179 			key.port[pd2.didx] = uh.uh_dport;
7180 
7181 			STATE_LOOKUP(kif, &key, *state, pd);
7182 
7183 			/* translate source/destination address, if necessary */
7184 			if ((*state)->key[PF_SK_WIRE] !=
7185 			    (*state)->key[PF_SK_STACK]) {
7186 				struct pf_state_key *nk =
7187 				    (*state)->key[pd->didx];
7188 
7189 				if (PF_ANEQ(pd2.src,
7190 				    &nk->addr[pd2.sidx], pd2.af) ||
7191 				    nk->port[pd2.sidx] != uh.uh_sport)
7192 					pf_change_icmp(pd2.src, &uh.uh_sport,
7193 					    daddr, &nk->addr[pd2.sidx],
7194 					    nk->port[pd2.sidx], &uh.uh_sum,
7195 					    pd2.ip_sum, icmpsum,
7196 					    pd->ip_sum, 1, pd2.af);
7197 
7198 				if (PF_ANEQ(pd2.dst,
7199 				    &nk->addr[pd2.didx], pd2.af) ||
7200 				    nk->port[pd2.didx] != uh.uh_dport)
7201 					pf_change_icmp(pd2.dst, &uh.uh_dport,
7202 					    saddr, &nk->addr[pd2.didx],
7203 					    nk->port[pd2.didx], &uh.uh_sum,
7204 					    pd2.ip_sum, icmpsum,
7205 					    pd->ip_sum, 1, pd2.af);
7206 
7207 				switch (pd2.af) {
7208 #ifdef INET
7209 				case AF_INET:
7210 					m_copyback(m, off, ICMP_MINLEN,
7211 					    (caddr_t )&pd->hdr.icmp);
7212 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
7213 					break;
7214 #endif /* INET */
7215 #ifdef INET6
7216 				case AF_INET6:
7217 					m_copyback(m, off,
7218 					    sizeof(struct icmp6_hdr),
7219 					    (caddr_t )&pd->hdr.icmp6);
7220 					m_copyback(m, ipoff2, sizeof(h2_6),
7221 					    (caddr_t )&h2_6);
7222 					break;
7223 #endif /* INET6 */
7224 				}
7225 				m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
7226 			}
7227 			return (PF_PASS);
7228 			break;
7229 		}
7230 #ifdef INET
7231 		case IPPROTO_SCTP: {
7232 			struct sctphdr		sh;
7233 			struct pf_state_peer	*src;
7234 			int			 copyback = 0;
7235 
7236 			if (! pf_pull_hdr(m, off2, &sh, sizeof(sh), NULL, reason,
7237 			    pd2.af)) {
7238 				DPFPRINTF(PF_DEBUG_MISC,
7239 				    ("pf: ICMP error message too short "
7240 				    "(sctp)\n"));
7241 				return (PF_DROP);
7242 			}
7243 
7244 			key.af = pd2.af;
7245 			key.proto = IPPROTO_SCTP;
7246 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
7247 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
7248 			key.port[pd2.sidx] = sh.src_port;
7249 			key.port[pd2.didx] = sh.dest_port;
7250 
7251 			STATE_LOOKUP(kif, &key, *state, pd);
7252 
7253 			if (pd->dir == (*state)->direction) {
7254 				src = &(*state)->dst;
7255 			} else {
7256 				src = &(*state)->src;
7257 			}
7258 
7259 			if (src->scrub->pfss_v_tag != sh.v_tag) {
7260 				DPFPRINTF(PF_DEBUG_MISC,
7261 				    ("pf: ICMP error message has incorrect "
7262 				    "SCTP v_tag\n"));
7263 				return (PF_DROP);
7264 			}
7265 
7266 			/* translate source/destination address, if necessary */
7267 			if ((*state)->key[PF_SK_WIRE] !=
7268 			    (*state)->key[PF_SK_STACK]) {
7269 				struct pf_state_key *nk =
7270 				    (*state)->key[pd->didx];
7271 
7272 				if (PF_ANEQ(pd2.src,
7273 				    &nk->addr[pd2.sidx], pd2.af) ||
7274 				    nk->port[pd2.sidx] != sh.src_port)
7275 					pf_change_icmp(pd2.src, &sh.src_port,
7276 					    daddr, &nk->addr[pd2.sidx],
7277 					    nk->port[pd2.sidx], NULL,
7278 					    pd2.ip_sum, icmpsum,
7279 					    pd->ip_sum, 0, pd2.af);
7280 
7281 				if (PF_ANEQ(pd2.dst,
7282 				    &nk->addr[pd2.didx], pd2.af) ||
7283 				    nk->port[pd2.didx] != sh.dest_port)
7284 					pf_change_icmp(pd2.dst, &sh.dest_port,
7285 					    saddr, &nk->addr[pd2.didx],
7286 					    nk->port[pd2.didx], NULL,
7287 					    pd2.ip_sum, icmpsum,
7288 					    pd->ip_sum, 0, pd2.af);
7289 				copyback = 1;
7290 			}
7291 
7292 			if (copyback) {
7293 				switch (pd2.af) {
7294 #ifdef INET
7295 				case AF_INET:
7296 					m_copyback(m, off, ICMP_MINLEN,
7297 					    (caddr_t )&pd->hdr.icmp);
7298 					m_copyback(m, ipoff2, sizeof(h2),
7299 					    (caddr_t )&h2);
7300 					break;
7301 #endif /* INET */
7302 #ifdef INET6
7303 				case AF_INET6:
7304 					m_copyback(m, off,
7305 					    sizeof(struct icmp6_hdr),
7306 					    (caddr_t )&pd->hdr.icmp6);
7307 					m_copyback(m, ipoff2, sizeof(h2_6),
7308 					    (caddr_t )&h2_6);
7309 					break;
7310 #endif /* INET6 */
7311 				}
7312 				m_copyback(m, off2, sizeof(sh), (caddr_t)&sh);
7313 			}
7314 
7315 			return (PF_PASS);
7316 			break;
7317 		}
7318 		case IPPROTO_ICMP: {
7319 			struct icmp	*iih = &pd2.hdr.icmp;
7320 
7321 			if (!pf_pull_hdr(m, off2, iih, ICMP_MINLEN,
7322 			    NULL, reason, pd2.af)) {
7323 				DPFPRINTF(PF_DEBUG_MISC,
7324 				    ("pf: ICMP error message too short i"
7325 				    "(icmp)\n"));
7326 				return (PF_DROP);
7327 			}
7328 
7329 			icmpid = iih->icmp_id;
7330 			pf_icmp_mapping(&pd2, iih->icmp_type,
7331 			    &icmp_dir, &multi, &virtual_id, &virtual_type);
7332 
7333 			ret = pf_icmp_state_lookup(&key, &pd2, state, m, off,
7334 			    pd2.dir, kif, virtual_id, virtual_type,
7335 			    icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1);
7336 			if (ret >= 0) {
7337 				MPASS(*state == NULL);
7338 				return (ret);
7339 			}
7340 
7341 			/* translate source/destination address, if necessary */
7342 			if ((*state)->key[PF_SK_WIRE] !=
7343 			    (*state)->key[PF_SK_STACK]) {
7344 				struct pf_state_key *nk =
7345 				    (*state)->key[pd->didx];
7346 
7347 				if (PF_ANEQ(pd2.src,
7348 				    &nk->addr[pd2.sidx], pd2.af) ||
7349 				    (virtual_type == htons(ICMP_ECHO) &&
7350 				    nk->port[iidx] != iih->icmp_id))
7351 					pf_change_icmp(pd2.src,
7352 					    (virtual_type == htons(ICMP_ECHO)) ?
7353 					    &iih->icmp_id : NULL,
7354 					    daddr, &nk->addr[pd2.sidx],
7355 					    (virtual_type == htons(ICMP_ECHO)) ?
7356 					    nk->port[iidx] : 0, NULL,
7357 					    pd2.ip_sum, icmpsum,
7358 					    pd->ip_sum, 0, AF_INET);
7359 
7360 				if (PF_ANEQ(pd2.dst,
7361 				    &nk->addr[pd2.didx], pd2.af))
7362 					pf_change_icmp(pd2.dst, NULL, NULL,
7363 					    &nk->addr[pd2.didx], 0, NULL,
7364 					    pd2.ip_sum, icmpsum, pd->ip_sum, 0,
7365 					    AF_INET);
7366 
7367 				m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp);
7368 				m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
7369 				m_copyback(m, off2, ICMP_MINLEN, (caddr_t)iih);
7370 			}
7371 			return (PF_PASS);
7372 			break;
7373 		}
7374 #endif /* INET */
7375 #ifdef INET6
7376 		case IPPROTO_ICMPV6: {
7377 			struct icmp6_hdr	*iih = &pd2.hdr.icmp6;
7378 
7379 			if (!pf_pull_hdr(m, off2, iih,
7380 			    sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
7381 				DPFPRINTF(PF_DEBUG_MISC,
7382 				    ("pf: ICMP error message too short "
7383 				    "(icmp6)\n"));
7384 				return (PF_DROP);
7385 			}
7386 
7387 			pf_icmp_mapping(&pd2, iih->icmp6_type,
7388 			    &icmp_dir, &multi, &virtual_id, &virtual_type);
7389 
7390 			ret = pf_icmp_state_lookup(&key, &pd2, state, m, off,
7391 			    pd->dir, kif, virtual_id, virtual_type,
7392 			    icmp_dir, &iidx, PF_ICMP_MULTI_NONE, 1);
7393 			if (ret >= 0) {
7394 				MPASS(*state == NULL);
7395 				if (ret == PF_DROP && pd2.af == AF_INET6 &&
7396 				    icmp_dir == PF_OUT) {
7397 					ret = pf_icmp_state_lookup(&key, &pd2,
7398 					    state, m, off, pd->dir, kif,
7399 					    virtual_id, virtual_type,
7400 					    icmp_dir, &iidx, multi, 1);
7401 					if (ret >= 0) {
7402 						MPASS(*state == NULL);
7403 						return (ret);
7404 					}
7405 				} else
7406 					return (ret);
7407 			}
7408 
7409 			/* translate source/destination address, if necessary */
7410 			if ((*state)->key[PF_SK_WIRE] !=
7411 			    (*state)->key[PF_SK_STACK]) {
7412 				struct pf_state_key *nk =
7413 				    (*state)->key[pd->didx];
7414 
7415 				if (PF_ANEQ(pd2.src,
7416 				    &nk->addr[pd2.sidx], pd2.af) ||
7417 				    ((virtual_type == htons(ICMP6_ECHO_REQUEST)) &&
7418 				    nk->port[pd2.sidx] != iih->icmp6_id))
7419 					pf_change_icmp(pd2.src,
7420 					    (virtual_type == htons(ICMP6_ECHO_REQUEST))
7421 					    ? &iih->icmp6_id : NULL,
7422 					    daddr, &nk->addr[pd2.sidx],
7423 					    (virtual_type == htons(ICMP6_ECHO_REQUEST))
7424 					    ? nk->port[iidx] : 0, NULL,
7425 					    pd2.ip_sum, icmpsum,
7426 					    pd->ip_sum, 0, AF_INET6);
7427 
7428 				if (PF_ANEQ(pd2.dst,
7429 				    &nk->addr[pd2.didx], pd2.af))
7430 					pf_change_icmp(pd2.dst, NULL, NULL,
7431 					    &nk->addr[pd2.didx], 0, NULL,
7432 					    pd2.ip_sum, icmpsum,
7433 					    pd->ip_sum, 0, AF_INET6);
7434 
7435 				m_copyback(m, off, sizeof(struct icmp6_hdr),
7436 				    (caddr_t)&pd->hdr.icmp6);
7437 				m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
7438 				m_copyback(m, off2, sizeof(struct icmp6_hdr),
7439 				    (caddr_t)iih);
7440 			}
7441 			return (PF_PASS);
7442 			break;
7443 		}
7444 #endif /* INET6 */
7445 		default: {
7446 			key.af = pd2.af;
7447 			key.proto = pd2.proto;
7448 			PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
7449 			PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
7450 			key.port[0] = key.port[1] = 0;
7451 
7452 			STATE_LOOKUP(kif, &key, *state, pd);
7453 
7454 			/* translate source/destination address, if necessary */
7455 			if ((*state)->key[PF_SK_WIRE] !=
7456 			    (*state)->key[PF_SK_STACK]) {
7457 				struct pf_state_key *nk =
7458 				    (*state)->key[pd->didx];
7459 
7460 				if (PF_ANEQ(pd2.src,
7461 				    &nk->addr[pd2.sidx], pd2.af))
7462 					pf_change_icmp(pd2.src, NULL, daddr,
7463 					    &nk->addr[pd2.sidx], 0, NULL,
7464 					    pd2.ip_sum, icmpsum,
7465 					    pd->ip_sum, 0, pd2.af);
7466 
7467 				if (PF_ANEQ(pd2.dst,
7468 				    &nk->addr[pd2.didx], pd2.af))
7469 					pf_change_icmp(pd2.dst, NULL, saddr,
7470 					    &nk->addr[pd2.didx], 0, NULL,
7471 					    pd2.ip_sum, icmpsum,
7472 					    pd->ip_sum, 0, pd2.af);
7473 
7474 				switch (pd2.af) {
7475 #ifdef INET
7476 				case AF_INET:
7477 					m_copyback(m, off, ICMP_MINLEN,
7478 					    (caddr_t)&pd->hdr.icmp);
7479 					m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
7480 					break;
7481 #endif /* INET */
7482 #ifdef INET6
7483 				case AF_INET6:
7484 					m_copyback(m, off,
7485 					    sizeof(struct icmp6_hdr),
7486 					    (caddr_t )&pd->hdr.icmp6);
7487 					m_copyback(m, ipoff2, sizeof(h2_6),
7488 					    (caddr_t )&h2_6);
7489 					break;
7490 #endif /* INET6 */
7491 				}
7492 			}
7493 			return (PF_PASS);
7494 			break;
7495 		}
7496 		}
7497 	}
7498 }
7499 
7500 static int
pf_test_state_other(struct pf_kstate ** state,struct pfi_kkif * kif,struct mbuf * m,struct pf_pdesc * pd)7501 pf_test_state_other(struct pf_kstate **state, struct pfi_kkif *kif,
7502     struct mbuf *m, struct pf_pdesc *pd)
7503 {
7504 	struct pf_state_peer	*src, *dst;
7505 	struct pf_state_key_cmp	 key;
7506 	uint8_t			 psrc, pdst;
7507 
7508 	bzero(&key, sizeof(key));
7509 	key.af = pd->af;
7510 	key.proto = pd->proto;
7511 	if (pd->dir == PF_IN)	{
7512 		PF_ACPY(&key.addr[0], pd->src, key.af);
7513 		PF_ACPY(&key.addr[1], pd->dst, key.af);
7514 		key.port[0] = key.port[1] = 0;
7515 	} else {
7516 		PF_ACPY(&key.addr[1], pd->src, key.af);
7517 		PF_ACPY(&key.addr[0], pd->dst, key.af);
7518 		key.port[1] = key.port[0] = 0;
7519 	}
7520 
7521 	STATE_LOOKUP(kif, &key, *state, pd);
7522 
7523 	if (pd->dir == (*state)->direction) {
7524 		src = &(*state)->src;
7525 		dst = &(*state)->dst;
7526 		psrc = PF_PEER_SRC;
7527 		pdst = PF_PEER_DST;
7528 	} else {
7529 		src = &(*state)->dst;
7530 		dst = &(*state)->src;
7531 		psrc = PF_PEER_DST;
7532 		pdst = PF_PEER_SRC;
7533 	}
7534 
7535 	/* update states */
7536 	if (src->state < PFOTHERS_SINGLE)
7537 		pf_set_protostate(*state, psrc, PFOTHERS_SINGLE);
7538 	if (dst->state == PFOTHERS_SINGLE)
7539 		pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE);
7540 
7541 	/* update expire time */
7542 	(*state)->expire = time_uptime;
7543 	if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
7544 		(*state)->timeout = PFTM_OTHER_MULTIPLE;
7545 	else
7546 		(*state)->timeout = PFTM_OTHER_SINGLE;
7547 
7548 	/* translate source/destination address, if necessary */
7549 	if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
7550 		struct pf_state_key *nk = (*state)->key[pd->didx];
7551 
7552 		KASSERT(nk, ("%s: nk is null", __func__));
7553 		KASSERT(pd, ("%s: pd is null", __func__));
7554 		KASSERT(pd->src, ("%s: pd->src is null", __func__));
7555 		KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
7556 		switch (pd->af) {
7557 #ifdef INET
7558 		case AF_INET:
7559 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
7560 				pf_change_a(&pd->src->v4.s_addr,
7561 				    pd->ip_sum,
7562 				    nk->addr[pd->sidx].v4.s_addr,
7563 				    0);
7564 
7565 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
7566 				pf_change_a(&pd->dst->v4.s_addr,
7567 				    pd->ip_sum,
7568 				    nk->addr[pd->didx].v4.s_addr,
7569 				    0);
7570 
7571 			break;
7572 #endif /* INET */
7573 #ifdef INET6
7574 		case AF_INET6:
7575 			if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET6))
7576 				PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
7577 
7578 			if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET6))
7579 				PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
7580 #endif /* INET6 */
7581 		}
7582 	}
7583 	return (PF_PASS);
7584 }
7585 
7586 /*
7587  * ipoff and off are measured from the start of the mbuf chain.
7588  * h must be at "ipoff" on the mbuf chain.
7589  */
7590 void *
pf_pull_hdr(struct mbuf * m,int off,void * p,int len,u_short * actionp,u_short * reasonp,sa_family_t af)7591 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
7592     u_short *actionp, u_short *reasonp, sa_family_t af)
7593 {
7594 	switch (af) {
7595 #ifdef INET
7596 	case AF_INET: {
7597 		struct ip	*h = mtod(m, struct ip *);
7598 		u_int16_t	 fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
7599 
7600 		if (fragoff) {
7601 			if (fragoff >= len)
7602 				ACTION_SET(actionp, PF_PASS);
7603 			else {
7604 				ACTION_SET(actionp, PF_DROP);
7605 				REASON_SET(reasonp, PFRES_FRAG);
7606 			}
7607 			return (NULL);
7608 		}
7609 		if (m->m_pkthdr.len < off + len ||
7610 		    ntohs(h->ip_len) < off + len) {
7611 			ACTION_SET(actionp, PF_DROP);
7612 			REASON_SET(reasonp, PFRES_SHORT);
7613 			return (NULL);
7614 		}
7615 		break;
7616 	}
7617 #endif /* INET */
7618 #ifdef INET6
7619 	case AF_INET6: {
7620 		struct ip6_hdr	*h = mtod(m, struct ip6_hdr *);
7621 
7622 		if (m->m_pkthdr.len < off + len ||
7623 		    (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
7624 		    (unsigned)(off + len)) {
7625 			ACTION_SET(actionp, PF_DROP);
7626 			REASON_SET(reasonp, PFRES_SHORT);
7627 			return (NULL);
7628 		}
7629 		break;
7630 	}
7631 #endif /* INET6 */
7632 	}
7633 	m_copydata(m, off, len, p);
7634 	return (p);
7635 }
7636 
7637 int
pf_routable(struct pf_addr * addr,sa_family_t af,struct pfi_kkif * kif,int rtableid)7638 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif,
7639     int rtableid)
7640 {
7641 	struct ifnet		*ifp;
7642 
7643 	/*
7644 	 * Skip check for addresses with embedded interface scope,
7645 	 * as they would always match anyway.
7646 	 */
7647 	if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
7648 		return (1);
7649 
7650 	if (af != AF_INET && af != AF_INET6)
7651 		return (0);
7652 
7653 	if (kif == V_pfi_all)
7654 		return (1);
7655 
7656 	/* Skip checks for ipsec interfaces */
7657 	if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
7658 		return (1);
7659 
7660 	ifp = (kif != NULL) ? kif->pfik_ifp : NULL;
7661 
7662 	switch (af) {
7663 #ifdef INET6
7664 	case AF_INET6:
7665 		return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE,
7666 		    ifp));
7667 #endif
7668 #ifdef INET
7669 	case AF_INET:
7670 		return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE,
7671 		    ifp));
7672 #endif
7673 	}
7674 
7675 	return (0);
7676 }
7677 
7678 #ifdef INET
7679 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)7680 pf_route(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp,
7681     struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp)
7682 {
7683 	struct mbuf		*m0, *m1, *md;
7684 	struct sockaddr_in	dst;
7685 	struct ip		*ip;
7686 	struct pfi_kkif		*nkif = NULL;
7687 	struct ifnet		*ifp = NULL;
7688 	struct pf_addr		 naddr;
7689 	struct pf_ksrc_node	*sn = NULL;
7690 	int			 error = 0;
7691 	uint16_t		 ip_len, ip_off;
7692 	int			 r_rt, r_dir;
7693 
7694 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
7695 
7696 	if (s) {
7697 		r_rt = s->rt;
7698 		r_dir = s->direction;
7699 	} else {
7700 		r_rt = r->rt;
7701 		r_dir = r->direction;
7702 	}
7703 
7704 	KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT ||
7705 	    r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction",
7706 	    __func__));
7707 
7708 	if ((pd->pf_mtag == NULL &&
7709 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
7710 	    pd->pf_mtag->routed++ > 3) {
7711 		m0 = *m;
7712 		*m = NULL;
7713 		goto bad_locked;
7714 	}
7715 
7716 	if (r_rt == PF_DUPTO) {
7717 		if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) {
7718 			if (s == NULL) {
7719 				ifp = r->rpool.cur->kif ?
7720 				    r->rpool.cur->kif->pfik_ifp : NULL;
7721 			} else {
7722 				ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7723 				/* If pfsync'd */
7724 				if (ifp == NULL && r->rpool.cur != NULL)
7725 					ifp = r->rpool.cur->kif ?
7726 					    r->rpool.cur->kif->pfik_ifp : NULL;
7727 				PF_STATE_UNLOCK(s);
7728 			}
7729 			if (ifp == oifp) {
7730 				/* When the 2nd interface is not skipped */
7731 				return;
7732 			} else {
7733 				m0 = *m;
7734 				*m = NULL;
7735 				goto bad;
7736 			}
7737 		} else {
7738 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED;
7739 			if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) {
7740 				if (s)
7741 					PF_STATE_UNLOCK(s);
7742 				return;
7743 			}
7744 		}
7745 	} else {
7746 		if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) {
7747 			pf_dummynet(pd, s, r, m);
7748 			if (s)
7749 				PF_STATE_UNLOCK(s);
7750 			return;
7751 		}
7752 		m0 = *m;
7753 	}
7754 
7755 	ip = mtod(m0, struct ip *);
7756 
7757 	bzero(&dst, sizeof(dst));
7758 	dst.sin_family = AF_INET;
7759 	dst.sin_len = sizeof(dst);
7760 	dst.sin_addr = ip->ip_dst;
7761 
7762 	bzero(&naddr, sizeof(naddr));
7763 
7764 	if (s == NULL) {
7765 		if (TAILQ_EMPTY(&r->rpool.list)) {
7766 			DPFPRINTF(PF_DEBUG_URGENT,
7767 			    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
7768 			goto bad_locked;
7769 		}
7770 		pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
7771 		    &naddr, &nkif, NULL, &sn);
7772 		if (!PF_AZERO(&naddr, AF_INET))
7773 			dst.sin_addr.s_addr = naddr.v4.s_addr;
7774 		ifp = nkif ? nkif->pfik_ifp : NULL;
7775 	} else {
7776 		if (!PF_AZERO(&s->rt_addr, AF_INET))
7777 			dst.sin_addr.s_addr =
7778 			    s->rt_addr.v4.s_addr;
7779 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7780 		/* If pfsync'd */
7781 		if (ifp == NULL && r->rpool.cur != NULL) {
7782 			ifp = r->rpool.cur->kif ?
7783 			    r->rpool.cur->kif->pfik_ifp : NULL;
7784 		}
7785 		PF_STATE_UNLOCK(s);
7786 	}
7787 
7788 	if (ifp == NULL)
7789 		goto bad;
7790 
7791 	if (pd->dir == PF_IN) {
7792 		if (pf_test(PF_OUT, 0, ifp, &m0, inp, &pd->act) != PF_PASS)
7793 			goto bad;
7794 		else if (m0 == NULL)
7795 			goto done;
7796 		if (m0->m_len < sizeof(struct ip)) {
7797 			DPFPRINTF(PF_DEBUG_URGENT,
7798 			    ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
7799 			goto bad;
7800 		}
7801 		ip = mtod(m0, struct ip *);
7802 	}
7803 
7804 	if (ifp->if_flags & IFF_LOOPBACK)
7805 		m0->m_flags |= M_SKIP_FIREWALL;
7806 
7807 	ip_len = ntohs(ip->ip_len);
7808 	ip_off = ntohs(ip->ip_off);
7809 
7810 	/* Copied from FreeBSD 10.0-CURRENT ip_output. */
7811 	m0->m_pkthdr.csum_flags |= CSUM_IP;
7812 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
7813 		in_delayed_cksum(m0);
7814 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
7815 	}
7816 	if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
7817 		pf_sctp_checksum(m0, (uint32_t)(ip->ip_hl << 2));
7818 		m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
7819 	}
7820 
7821 	/*
7822 	 * If small enough for interface, or the interface will take
7823 	 * care of the fragmentation for us, we can just send directly.
7824 	 */
7825 	if (ip_len <= ifp->if_mtu ||
7826 	    (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
7827 		ip->ip_sum = 0;
7828 		if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
7829 			ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
7830 			m0->m_pkthdr.csum_flags &= ~CSUM_IP;
7831 		}
7832 		m_clrprotoflags(m0);	/* Avoid confusing lower layers. */
7833 
7834 		md = m0;
7835 		error = pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md);
7836 		if (md != NULL)
7837 			error = (*ifp->if_output)(ifp, md, sintosa(&dst), NULL);
7838 		goto done;
7839 	}
7840 
7841 	/* Balk when DF bit is set or the interface didn't support TSO. */
7842 	if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
7843 		error = EMSGSIZE;
7844 		KMOD_IPSTAT_INC(ips_cantfrag);
7845 		if (r_rt != PF_DUPTO) {
7846 			if (s && pd->nat_rule != NULL)
7847 				PACKET_UNDO_NAT(m0, pd,
7848 				    (ip->ip_hl << 2) + (ip_off & IP_OFFMASK),
7849 				    s);
7850 
7851 			icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
7852 			    ifp->if_mtu);
7853 			goto done;
7854 		} else
7855 			goto bad;
7856 	}
7857 
7858 	error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
7859 	if (error)
7860 		goto bad;
7861 
7862 	for (; m0; m0 = m1) {
7863 		m1 = m0->m_nextpkt;
7864 		m0->m_nextpkt = NULL;
7865 		if (error == 0) {
7866 			m_clrprotoflags(m0);
7867 			md = m0;
7868 			pd->pf_mtag = pf_find_mtag(md);
7869 			error = pf_dummynet_route(pd, s, r, ifp,
7870 			    sintosa(&dst), &md);
7871 			if (md != NULL)
7872 				error = (*ifp->if_output)(ifp, md,
7873 				    sintosa(&dst), NULL);
7874 		} else
7875 			m_freem(m0);
7876 	}
7877 
7878 	if (error == 0)
7879 		KMOD_IPSTAT_INC(ips_fragmented);
7880 
7881 done:
7882 	if (r_rt != PF_DUPTO)
7883 		*m = NULL;
7884 	return;
7885 
7886 bad_locked:
7887 	if (s)
7888 		PF_STATE_UNLOCK(s);
7889 bad:
7890 	m_freem(m0);
7891 	goto done;
7892 }
7893 #endif /* INET */
7894 
7895 #ifdef INET6
7896 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)7897 pf_route6(struct mbuf **m, struct pf_krule *r, struct ifnet *oifp,
7898     struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp)
7899 {
7900 	struct mbuf		*m0, *md;
7901 	struct sockaddr_in6	dst;
7902 	struct ip6_hdr		*ip6;
7903 	struct pfi_kkif		*nkif = NULL;
7904 	struct ifnet		*ifp = NULL;
7905 	struct pf_addr		 naddr;
7906 	struct pf_ksrc_node	*sn = NULL;
7907 	int			 r_rt, r_dir;
7908 
7909 	KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
7910 
7911 	if (s) {
7912 		r_rt = s->rt;
7913 		r_dir = s->direction;
7914 	} else {
7915 		r_rt = r->rt;
7916 		r_dir = r->direction;
7917 	}
7918 
7919 	KASSERT(pd->dir == PF_IN || pd->dir == PF_OUT ||
7920 	    r_dir == PF_IN || r_dir == PF_OUT, ("%s: invalid direction",
7921 	    __func__));
7922 
7923 	if ((pd->pf_mtag == NULL &&
7924 	    ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
7925 	    pd->pf_mtag->routed++ > 3) {
7926 		m0 = *m;
7927 		*m = NULL;
7928 		goto bad_locked;
7929 	}
7930 
7931 	if (r_rt == PF_DUPTO) {
7932 		if ((pd->pf_mtag->flags & PF_MTAG_FLAG_DUPLICATED)) {
7933 			if (s == NULL) {
7934 				ifp = r->rpool.cur->kif ?
7935 				    r->rpool.cur->kif->pfik_ifp : NULL;
7936 			} else {
7937 				ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7938 				/* If pfsync'd */
7939 				if (ifp == NULL && r->rpool.cur != NULL)
7940 					ifp = r->rpool.cur->kif ?
7941 					    r->rpool.cur->kif->pfik_ifp : NULL;
7942 				PF_STATE_UNLOCK(s);
7943 			}
7944 			if (ifp == oifp) {
7945 				/* When the 2nd interface is not skipped */
7946 				return;
7947 			} else {
7948 				m0 = *m;
7949 				*m = NULL;
7950 				goto bad;
7951 			}
7952 		} else {
7953 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUPLICATED;
7954 			if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) {
7955 				if (s)
7956 					PF_STATE_UNLOCK(s);
7957 				return;
7958 			}
7959 		}
7960 	} else {
7961 		if ((r_rt == PF_REPLYTO) == (r_dir == pd->dir)) {
7962 			pf_dummynet(pd, s, r, m);
7963 			if (s)
7964 				PF_STATE_UNLOCK(s);
7965 			return;
7966 		}
7967 		m0 = *m;
7968 	}
7969 
7970 	ip6 = mtod(m0, struct ip6_hdr *);
7971 
7972 	bzero(&dst, sizeof(dst));
7973 	dst.sin6_family = AF_INET6;
7974 	dst.sin6_len = sizeof(dst);
7975 	dst.sin6_addr = ip6->ip6_dst;
7976 
7977 	bzero(&naddr, sizeof(naddr));
7978 
7979 	if (s == NULL) {
7980 		if (TAILQ_EMPTY(&r->rpool.list)) {
7981 			DPFPRINTF(PF_DEBUG_URGENT,
7982 			    ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
7983 			goto bad_locked;
7984 		}
7985 		pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
7986 		    &naddr, &nkif, NULL, &sn);
7987 		if (!PF_AZERO(&naddr, AF_INET6))
7988 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
7989 			    &naddr, AF_INET6);
7990 		ifp = nkif ? nkif->pfik_ifp : NULL;
7991 	} else {
7992 		if (!PF_AZERO(&s->rt_addr, AF_INET6))
7993 			PF_ACPY((struct pf_addr *)&dst.sin6_addr,
7994 			    &s->rt_addr, AF_INET6);
7995 		ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
7996 		/* If pfsync'd */
7997 		if (ifp == NULL && r->rpool.cur != NULL)
7998 			ifp = r->rpool.cur->kif ?
7999 			    r->rpool.cur->kif->pfik_ifp : NULL;
8000 	}
8001 
8002 	if (s)
8003 		PF_STATE_UNLOCK(s);
8004 
8005 	if (ifp == NULL)
8006 		goto bad;
8007 
8008 	if (pd->dir == PF_IN) {
8009 		if (pf_test6(PF_OUT, 0, ifp, &m0, inp, &pd->act) != PF_PASS)
8010 			goto bad;
8011 		else if (m0 == NULL)
8012 			goto done;
8013 		if (m0->m_len < sizeof(struct ip6_hdr)) {
8014 			DPFPRINTF(PF_DEBUG_URGENT,
8015 			    ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
8016 			    __func__));
8017 			goto bad;
8018 		}
8019 		ip6 = mtod(m0, struct ip6_hdr *);
8020 	}
8021 
8022 	if (ifp->if_flags & IFF_LOOPBACK)
8023 		m0->m_flags |= M_SKIP_FIREWALL;
8024 
8025 	if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
8026 	    ~ifp->if_hwassist) {
8027 		uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
8028 		in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
8029 		m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
8030 	}
8031 
8032 	/*
8033 	 * If the packet is too large for the outgoing interface,
8034 	 * send back an icmp6 error.
8035 	 */
8036 	if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
8037 		dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
8038 	if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) {
8039 		md = m0;
8040 		pf_dummynet_route(pd, s, r, ifp, sintosa(&dst), &md);
8041 		if (md != NULL)
8042 			nd6_output_ifp(ifp, ifp, md, &dst, NULL);
8043 	}
8044 	else {
8045 		in6_ifstat_inc(ifp, ifs6_in_toobig);
8046 		if (r_rt != PF_DUPTO) {
8047 			if (s && pd->nat_rule != NULL)
8048 				PACKET_UNDO_NAT(m0, pd,
8049 				    ((caddr_t)ip6 - m0->m_data) +
8050 				    sizeof(struct ip6_hdr), s);
8051 
8052 			icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
8053 		} else
8054 			goto bad;
8055 	}
8056 
8057 done:
8058 	if (r_rt != PF_DUPTO)
8059 		*m = NULL;
8060 	return;
8061 
8062 bad_locked:
8063 	if (s)
8064 		PF_STATE_UNLOCK(s);
8065 bad:
8066 	m_freem(m0);
8067 	goto done;
8068 }
8069 #endif /* INET6 */
8070 
8071 /*
8072  * FreeBSD supports cksum offloads for the following drivers.
8073  *  em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4)
8074  *
8075  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
8076  *  network driver performed cksum including pseudo header, need to verify
8077  *   csum_data
8078  * CSUM_DATA_VALID :
8079  *  network driver performed cksum, needs to additional pseudo header
8080  *  cksum computation with partial csum_data(i.e. lack of H/W support for
8081  *  pseudo header, for instance sk(4) and possibly gem(4))
8082  *
8083  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
8084  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
8085  * TCP/UDP layer.
8086  * Also, set csum_data to 0xffff to force cksum validation.
8087  */
8088 static int
pf_check_proto_cksum(struct mbuf * m,int off,int len,u_int8_t p,sa_family_t af)8089 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
8090 {
8091 	u_int16_t sum = 0;
8092 	int hw_assist = 0;
8093 	struct ip *ip;
8094 
8095 	if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
8096 		return (1);
8097 	if (m->m_pkthdr.len < off + len)
8098 		return (1);
8099 
8100 	switch (p) {
8101 	case IPPROTO_TCP:
8102 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
8103 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
8104 				sum = m->m_pkthdr.csum_data;
8105 			} else {
8106 				ip = mtod(m, struct ip *);
8107 				sum = in_pseudo(ip->ip_src.s_addr,
8108 				ip->ip_dst.s_addr, htonl((u_short)len +
8109 				m->m_pkthdr.csum_data + IPPROTO_TCP));
8110 			}
8111 			sum ^= 0xffff;
8112 			++hw_assist;
8113 		}
8114 		break;
8115 	case IPPROTO_UDP:
8116 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
8117 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
8118 				sum = m->m_pkthdr.csum_data;
8119 			} else {
8120 				ip = mtod(m, struct ip *);
8121 				sum = in_pseudo(ip->ip_src.s_addr,
8122 				ip->ip_dst.s_addr, htonl((u_short)len +
8123 				m->m_pkthdr.csum_data + IPPROTO_UDP));
8124 			}
8125 			sum ^= 0xffff;
8126 			++hw_assist;
8127 		}
8128 		break;
8129 	case IPPROTO_ICMP:
8130 #ifdef INET6
8131 	case IPPROTO_ICMPV6:
8132 #endif /* INET6 */
8133 		break;
8134 	default:
8135 		return (1);
8136 	}
8137 
8138 	if (!hw_assist) {
8139 		switch (af) {
8140 		case AF_INET:
8141 			if (p == IPPROTO_ICMP) {
8142 				if (m->m_len < off)
8143 					return (1);
8144 				m->m_data += off;
8145 				m->m_len -= off;
8146 				sum = in_cksum(m, len);
8147 				m->m_data -= off;
8148 				m->m_len += off;
8149 			} else {
8150 				if (m->m_len < sizeof(struct ip))
8151 					return (1);
8152 				sum = in4_cksum(m, p, off, len);
8153 			}
8154 			break;
8155 #ifdef INET6
8156 		case AF_INET6:
8157 			if (m->m_len < sizeof(struct ip6_hdr))
8158 				return (1);
8159 			sum = in6_cksum(m, p, off, len);
8160 			break;
8161 #endif /* INET6 */
8162 		default:
8163 			return (1);
8164 		}
8165 	}
8166 	if (sum) {
8167 		switch (p) {
8168 		case IPPROTO_TCP:
8169 		    {
8170 			KMOD_TCPSTAT_INC(tcps_rcvbadsum);
8171 			break;
8172 		    }
8173 		case IPPROTO_UDP:
8174 		    {
8175 			KMOD_UDPSTAT_INC(udps_badsum);
8176 			break;
8177 		    }
8178 #ifdef INET
8179 		case IPPROTO_ICMP:
8180 		    {
8181 			KMOD_ICMPSTAT_INC(icps_checksum);
8182 			break;
8183 		    }
8184 #endif
8185 #ifdef INET6
8186 		case IPPROTO_ICMPV6:
8187 		    {
8188 			KMOD_ICMP6STAT_INC(icp6s_checksum);
8189 			break;
8190 		    }
8191 #endif /* INET6 */
8192 		}
8193 		return (1);
8194 	} else {
8195 		if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
8196 			m->m_pkthdr.csum_flags |=
8197 			    (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
8198 			m->m_pkthdr.csum_data = 0xffff;
8199 		}
8200 	}
8201 	return (0);
8202 }
8203 
8204 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)8205 pf_pdesc_to_dnflow(const struct pf_pdesc *pd, const struct pf_krule *r,
8206     const struct pf_kstate *s, struct ip_fw_args *dnflow)
8207 {
8208 	int dndir = r->direction;
8209 
8210 	if (s && dndir == PF_INOUT) {
8211 		dndir = s->direction;
8212 	} else if (dndir == PF_INOUT) {
8213 		/* Assume primary direction. Happens when we've set dnpipe in
8214 		 * the ethernet level code. */
8215 		dndir = pd->dir;
8216 	}
8217 
8218 	memset(dnflow, 0, sizeof(*dnflow));
8219 
8220 	if (pd->dport != NULL)
8221 		dnflow->f_id.dst_port = ntohs(*pd->dport);
8222 	if (pd->sport != NULL)
8223 		dnflow->f_id.src_port = ntohs(*pd->sport);
8224 
8225 	if (pd->dir == PF_IN)
8226 		dnflow->flags |= IPFW_ARGS_IN;
8227 	else
8228 		dnflow->flags |= IPFW_ARGS_OUT;
8229 
8230 	if (pd->dir != dndir && pd->act.dnrpipe) {
8231 		dnflow->rule.info = pd->act.dnrpipe;
8232 	}
8233 	else if (pd->dir == dndir && pd->act.dnpipe) {
8234 		dnflow->rule.info = pd->act.dnpipe;
8235 	}
8236 	else {
8237 		return (false);
8238 	}
8239 
8240 	dnflow->rule.info |= IPFW_IS_DUMMYNET;
8241 	if (r->free_flags & PFRULE_DN_IS_PIPE || pd->act.flags & PFSTATE_DN_IS_PIPE)
8242 		dnflow->rule.info |= IPFW_IS_PIPE;
8243 
8244 	dnflow->f_id.proto = pd->proto;
8245 	dnflow->f_id.extra = dnflow->rule.info;
8246 	switch (pd->af) {
8247 	case AF_INET:
8248 		dnflow->f_id.addr_type = 4;
8249 		dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr);
8250 		dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr);
8251 		break;
8252 	case AF_INET6:
8253 		dnflow->flags |= IPFW_ARGS_IP6;
8254 		dnflow->f_id.addr_type = 6;
8255 		dnflow->f_id.src_ip6 = pd->src->v6;
8256 		dnflow->f_id.dst_ip6 = pd->dst->v6;
8257 		break;
8258 	default:
8259 		panic("Invalid AF");
8260 		break;
8261 	}
8262 
8263 	return (true);
8264 }
8265 
8266 int
pf_test_eth(int dir,int pflags,struct ifnet * ifp,struct mbuf ** m0,struct inpcb * inp)8267 pf_test_eth(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0,
8268     struct inpcb *inp)
8269 {
8270 	struct pfi_kkif		*kif;
8271 	struct mbuf		*m = *m0;
8272 
8273 	M_ASSERTPKTHDR(m);
8274 	MPASS(ifp->if_vnet == curvnet);
8275 	NET_EPOCH_ASSERT();
8276 
8277 	if (!V_pf_status.running)
8278 		return (PF_PASS);
8279 
8280 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
8281 
8282 	if (kif == NULL) {
8283 		DPFPRINTF(PF_DEBUG_URGENT,
8284 		    ("%s: kif == NULL, if_xname %s\n", __func__, ifp->if_xname));
8285 		return (PF_DROP);
8286 	}
8287 	if (kif->pfik_flags & PFI_IFLAG_SKIP)
8288 		return (PF_PASS);
8289 
8290 	if (m->m_flags & M_SKIP_FIREWALL)
8291 		return (PF_PASS);
8292 
8293 	/* Stateless! */
8294 	return (pf_test_eth_rule(dir, kif, m0));
8295 }
8296 
8297 static __inline void
pf_dummynet_flag_remove(struct mbuf * m,struct pf_mtag * pf_mtag)8298 pf_dummynet_flag_remove(struct mbuf *m, struct pf_mtag *pf_mtag)
8299 {
8300 	struct m_tag *mtag;
8301 
8302 	pf_mtag->flags &= ~PF_MTAG_FLAG_DUMMYNET;
8303 
8304 	/* dummynet adds this tag, but pf does not need it,
8305 	 * and keeping it creates unexpected behavior,
8306 	 * e.g. in case of divert(4) usage right after dummynet. */
8307 	mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
8308 	if (mtag != NULL)
8309 		m_tag_delete(m, mtag);
8310 }
8311 
8312 static int
pf_dummynet(struct pf_pdesc * pd,struct pf_kstate * s,struct pf_krule * r,struct mbuf ** m0)8313 pf_dummynet(struct pf_pdesc *pd, struct pf_kstate *s,
8314     struct pf_krule *r, struct mbuf **m0)
8315 {
8316 	return (pf_dummynet_route(pd, s, r, NULL, NULL, m0));
8317 }
8318 
8319 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)8320 pf_dummynet_route(struct pf_pdesc *pd, struct pf_kstate *s,
8321     struct pf_krule *r, struct ifnet *ifp, struct sockaddr *sa,
8322     struct mbuf **m0)
8323 {
8324 	NET_EPOCH_ASSERT();
8325 
8326 	if (pd->act.dnpipe || pd->act.dnrpipe) {
8327 		struct ip_fw_args dnflow;
8328 		if (ip_dn_io_ptr == NULL) {
8329 			m_freem(*m0);
8330 			*m0 = NULL;
8331 			return (ENOMEM);
8332 		}
8333 
8334 		if (pd->pf_mtag == NULL &&
8335 		    ((pd->pf_mtag = pf_get_mtag(*m0)) == NULL)) {
8336 			m_freem(*m0);
8337 			*m0 = NULL;
8338 			return (ENOMEM);
8339 		}
8340 
8341 		if (ifp != NULL) {
8342 			pd->pf_mtag->flags |= PF_MTAG_FLAG_ROUTE_TO;
8343 
8344 			pd->pf_mtag->if_index = ifp->if_index;
8345 			pd->pf_mtag->if_idxgen = ifp->if_idxgen;
8346 
8347 			MPASS(sa != NULL);
8348 
8349 			if (pd->af == AF_INET)
8350 				memcpy(&pd->pf_mtag->dst, sa,
8351 				    sizeof(struct sockaddr_in));
8352 			else
8353 				memcpy(&pd->pf_mtag->dst, sa,
8354 				    sizeof(struct sockaddr_in6));
8355 		}
8356 
8357 		if (pf_pdesc_to_dnflow(pd, r, s, &dnflow)) {
8358 			pd->pf_mtag->flags |= PF_MTAG_FLAG_DUMMYNET;
8359 			ip_dn_io_ptr(m0, &dnflow);
8360 			if (*m0 != NULL) {
8361 				pd->pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
8362 				pf_dummynet_flag_remove(*m0, pd->pf_mtag);
8363 			}
8364 		}
8365 	}
8366 
8367 	return (0);
8368 }
8369 
8370 #ifdef INET
8371 int
pf_test(int dir,int pflags,struct ifnet * ifp,struct mbuf ** m0,struct inpcb * inp,struct pf_rule_actions * default_actions)8372 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0,
8373     struct inpcb *inp, struct pf_rule_actions *default_actions)
8374 {
8375 	struct pfi_kkif		*kif;
8376 	u_short			 action, reason = 0;
8377 	struct mbuf		*m = *m0;
8378 	struct ip		*h = NULL;
8379 	struct m_tag		*mtag;
8380 	struct pf_krule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
8381 	struct pf_kstate	*s = NULL;
8382 	struct pf_kruleset	*ruleset = NULL;
8383 	struct pf_pdesc		 pd;
8384 	int			 off, dirndx, use_2nd_queue = 0;
8385 	uint16_t		 tag;
8386 	uint8_t			 rt;
8387 
8388 	PF_RULES_RLOCK_TRACKER;
8389 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir));
8390 	M_ASSERTPKTHDR(m);
8391 
8392 	if (!V_pf_status.running)
8393 		return (PF_PASS);
8394 
8395 	PF_RULES_RLOCK();
8396 
8397 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
8398 
8399 	if (__predict_false(kif == NULL)) {
8400 		DPFPRINTF(PF_DEBUG_URGENT,
8401 		    ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
8402 		PF_RULES_RUNLOCK();
8403 		return (PF_DROP);
8404 	}
8405 	if (kif->pfik_flags & PFI_IFLAG_SKIP) {
8406 		PF_RULES_RUNLOCK();
8407 		return (PF_PASS);
8408 	}
8409 
8410 	if (m->m_flags & M_SKIP_FIREWALL) {
8411 		PF_RULES_RUNLOCK();
8412 		return (PF_PASS);
8413 	}
8414 
8415 	memset(&pd, 0, sizeof(pd));
8416 	TAILQ_INIT(&pd.sctp_multihome_jobs);
8417 	if (default_actions != NULL)
8418 		memcpy(&pd.act, default_actions, sizeof(pd.act));
8419 	pd.pf_mtag = pf_find_mtag(m);
8420 
8421 	if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) {
8422 		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
8423 
8424 		ifp = ifnet_byindexgen(pd.pf_mtag->if_index,
8425 		    pd.pf_mtag->if_idxgen);
8426 		if (ifp == NULL || ifp->if_flags & IFF_DYING) {
8427 			PF_RULES_RUNLOCK();
8428 			m_freem(*m0);
8429 			*m0 = NULL;
8430 			return (PF_PASS);
8431 		}
8432 		PF_RULES_RUNLOCK();
8433 		(ifp->if_output)(ifp, m, sintosa(&pd.pf_mtag->dst), NULL);
8434 		*m0 = NULL;
8435 		return (PF_PASS);
8436 	}
8437 
8438 	if (pd.pf_mtag && pd.pf_mtag->dnpipe) {
8439 		pd.act.dnpipe = pd.pf_mtag->dnpipe;
8440 		pd.act.flags = pd.pf_mtag->dnflags;
8441 	}
8442 
8443 	if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL &&
8444 	    pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
8445 		/* Dummynet re-injects packets after they've
8446 		 * completed their delay. We've already
8447 		 * processed them, so pass unconditionally. */
8448 
8449 		/* But only once. We may see the packet multiple times (e.g.
8450 		 * PFIL_IN/PFIL_OUT). */
8451 		pf_dummynet_flag_remove(m, pd.pf_mtag);
8452 		PF_RULES_RUNLOCK();
8453 
8454 		return (PF_PASS);
8455 	}
8456 
8457 	pd.sport = pd.dport = NULL;
8458 	pd.proto_sum = NULL;
8459 	pd.dir = dir;
8460 	pd.sidx = (dir == PF_IN) ? 0 : 1;
8461 	pd.didx = (dir == PF_IN) ? 1 : 0;
8462 	pd.af = AF_INET;
8463 	pd.act.rtableid = -1;
8464 
8465 	h = mtod(m, struct ip *);
8466 	off = h->ip_hl << 2;
8467 
8468 	if (__predict_false(ip_divert_ptr != NULL) &&
8469 	    ((mtag = m_tag_locate(m, MTAG_PF_DIVERT, 0, NULL)) != NULL)) {
8470 		struct pf_divert_mtag *dt = (struct pf_divert_mtag *)(mtag+1);
8471 		if ((dt->idir == PF_DIVERT_MTAG_DIR_IN && dir == PF_IN) ||
8472 		    (dt->idir == PF_DIVERT_MTAG_DIR_OUT && dir == PF_OUT)) {
8473 			if (pd.pf_mtag == NULL &&
8474 			    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8475 				action = PF_DROP;
8476 				goto done;
8477 			}
8478 			pd.pf_mtag->flags |= PF_MTAG_FLAG_PACKET_LOOPED;
8479 		}
8480 		if (pd.pf_mtag && pd.pf_mtag->flags & PF_MTAG_FLAG_FASTFWD_OURS_PRESENT) {
8481 			m->m_flags |= M_FASTFWD_OURS;
8482 			pd.pf_mtag->flags &= ~PF_MTAG_FLAG_FASTFWD_OURS_PRESENT;
8483 		}
8484 		m_tag_delete(m, mtag);
8485 
8486 		mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL);
8487 		if (mtag != NULL)
8488 			m_tag_delete(m, mtag);
8489 	} else if (pf_normalize_ip(m0, kif, &reason, &pd) != PF_PASS) {
8490 		m = *m0;
8491 		/* We do IP header normalization and packet reassembly here */
8492 		action = PF_DROP;
8493 		goto done;
8494 	}
8495 	m = *m0;	/* pf_normalize messes with m0 */
8496 	h = mtod(m, struct ip *);
8497 
8498 	off = h->ip_hl << 2;
8499 	if (off < (int)sizeof(struct ip)) {
8500 		action = PF_DROP;
8501 		REASON_SET(&reason, PFRES_SHORT);
8502 		pd.act.log = PF_LOG_FORCE;
8503 		goto done;
8504 	}
8505 
8506 	pd.src = (struct pf_addr *)&h->ip_src;
8507 	pd.dst = (struct pf_addr *)&h->ip_dst;
8508 	PF_ACPY(&pd.osrc, pd.src, pd.af);
8509 	PF_ACPY(&pd.odst, pd.dst, pd.af);
8510 	pd.ip_sum = &h->ip_sum;
8511 	pd.proto = h->ip_p;
8512 	pd.tos = h->ip_tos & ~IPTOS_ECN_MASK;
8513 	pd.tot_len = ntohs(h->ip_len);
8514 
8515 	/* handle fragments that didn't get reassembled by normalization */
8516 	if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
8517 		action = pf_test_fragment(&r, kif, m, h, &pd, &a, &ruleset);
8518 		goto done;
8519 	}
8520 
8521 	switch (h->ip_p) {
8522 	case IPPROTO_TCP: {
8523 		if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp),
8524 		    &action, &reason, AF_INET)) {
8525 			if (action != PF_PASS)
8526 				pd.act.log = PF_LOG_FORCE;
8527 			goto done;
8528 		}
8529 		pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2);
8530 
8531 		pd.sport = &pd.hdr.tcp.th_sport;
8532 		pd.dport = &pd.hdr.tcp.th_dport;
8533 
8534 		/* Respond to SYN with a syncookie. */
8535 		if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN &&
8536 		    pd.dir == PF_IN && pf_synflood_check(&pd)) {
8537 			pf_syncookie_send(m, off, &pd);
8538 			action = PF_DROP;
8539 			break;
8540 		}
8541 
8542 		if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0)
8543 			use_2nd_queue = 1;
8544 		action = pf_normalize_tcp(kif, m, 0, off, h, &pd);
8545 		if (action == PF_DROP)
8546 			goto done;
8547 		action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason);
8548 		if (action == PF_PASS) {
8549 			if (V_pfsync_update_state_ptr != NULL)
8550 				V_pfsync_update_state_ptr(s);
8551 			r = s->rule.ptr;
8552 			a = s->anchor.ptr;
8553 		} else if (s == NULL) {
8554 			/* Validate remote SYN|ACK, re-create original SYN if
8555 			 * valid. */
8556 			if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) ==
8557 			    TH_ACK && pf_syncookie_validate(&pd) &&
8558 			    pd.dir == PF_IN) {
8559 				struct mbuf *msyn;
8560 
8561 				msyn = pf_syncookie_recreate_syn(h->ip_ttl, off,
8562 				    &pd);
8563 				if (msyn == NULL) {
8564 					action = PF_DROP;
8565 					break;
8566 				}
8567 
8568 				action = pf_test(dir, pflags, ifp, &msyn, inp,
8569 				    &pd.act);
8570 				m_freem(msyn);
8571 				if (action != PF_PASS)
8572 					break;
8573 
8574 				action = pf_test_state_tcp(&s, kif, m, off, h,
8575 				    &pd, &reason);
8576 				if (action != PF_PASS || s == NULL) {
8577 					action = PF_DROP;
8578 					break;
8579 				}
8580 
8581 				s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1;
8582 				s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1;
8583 				pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST);
8584 				action = pf_synproxy(&pd, &s, &reason);
8585 				break;
8586 			} else {
8587 				action = pf_test_rule(&r, &s, kif, m, off, &pd,
8588 				    &a, &ruleset, inp);
8589 			}
8590 		}
8591 		break;
8592 	}
8593 
8594 	case IPPROTO_UDP: {
8595 		if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp),
8596 		    &action, &reason, AF_INET)) {
8597 			if (action != PF_PASS)
8598 				pd.act.log = PF_LOG_FORCE;
8599 			goto done;
8600 		}
8601 		pd.sport = &pd.hdr.udp.uh_sport;
8602 		pd.dport = &pd.hdr.udp.uh_dport;
8603 		if (pd.hdr.udp.uh_dport == 0 ||
8604 		    ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off ||
8605 		    ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) {
8606 			action = PF_DROP;
8607 			REASON_SET(&reason, PFRES_SHORT);
8608 			goto done;
8609 		}
8610 		action = pf_test_state_udp(&s, kif, m, off, h, &pd);
8611 		if (action == PF_PASS) {
8612 			if (V_pfsync_update_state_ptr != NULL)
8613 				V_pfsync_update_state_ptr(s);
8614 			r = s->rule.ptr;
8615 			a = s->anchor.ptr;
8616 		} else if (s == NULL)
8617 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8618 			    &a, &ruleset, inp);
8619 		break;
8620 	}
8621 
8622 	case IPPROTO_SCTP: {
8623 		if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp),
8624 		    &action, &reason, AF_INET)) {
8625 			if (action != PF_PASS)
8626 				pd.act.log |= PF_LOG_FORCE;
8627 			goto done;
8628 		}
8629 		pd.p_len = pd.tot_len - off;
8630 
8631 		pd.sport = &pd.hdr.sctp.src_port;
8632 		pd.dport = &pd.hdr.sctp.dest_port;
8633 		if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) {
8634 			action = PF_DROP;
8635 			REASON_SET(&reason, PFRES_SHORT);
8636 			goto done;
8637 		}
8638 		action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd);
8639 		if (action == PF_DROP)
8640 			goto done;
8641 		action = pf_test_state_sctp(&s, kif, m, off, h, &pd,
8642 		    &reason);
8643 		if (action == PF_PASS) {
8644 			if (V_pfsync_update_state_ptr != NULL)
8645 				V_pfsync_update_state_ptr(s);
8646 			r = s->rule.ptr;
8647 			a = s->anchor.ptr;
8648 		} else if (s == NULL) {
8649 			action = pf_test_rule(&r, &s, kif, m, off,
8650 			    &pd, &a, &ruleset, inp);
8651 		}
8652 		break;
8653 	}
8654 
8655 	case IPPROTO_ICMP: {
8656 		if (!pf_pull_hdr(m, off, &pd.hdr.icmp, ICMP_MINLEN,
8657 		    &action, &reason, AF_INET)) {
8658 			if (action != PF_PASS)
8659 				pd.act.log = PF_LOG_FORCE;
8660 			goto done;
8661 		}
8662 		action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason);
8663 		if (action == PF_PASS) {
8664 			if (V_pfsync_update_state_ptr != NULL)
8665 				V_pfsync_update_state_ptr(s);
8666 			r = s->rule.ptr;
8667 			a = s->anchor.ptr;
8668 		} else if (s == NULL)
8669 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8670 			    &a, &ruleset, inp);
8671 		break;
8672 	}
8673 
8674 #ifdef INET6
8675 	case IPPROTO_ICMPV6: {
8676 		action = PF_DROP;
8677 		DPFPRINTF(PF_DEBUG_MISC,
8678 		    ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
8679 		goto done;
8680 	}
8681 #endif
8682 
8683 	default:
8684 		action = pf_test_state_other(&s, kif, m, &pd);
8685 		if (action == PF_PASS) {
8686 			if (V_pfsync_update_state_ptr != NULL)
8687 				V_pfsync_update_state_ptr(s);
8688 			r = s->rule.ptr;
8689 			a = s->anchor.ptr;
8690 		} else if (s == NULL)
8691 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
8692 			    &a, &ruleset, inp);
8693 		break;
8694 	}
8695 
8696 done:
8697 	PF_RULES_RUNLOCK();
8698 
8699 	if (m == NULL)
8700 		goto out;
8701 
8702 	if (action == PF_PASS && h->ip_hl > 5 &&
8703 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
8704 		action = PF_DROP;
8705 		REASON_SET(&reason, PFRES_IPOPTIONS);
8706 		pd.act.log = PF_LOG_FORCE;
8707 		DPFPRINTF(PF_DEBUG_MISC,
8708 		    ("pf: dropping packet with ip options\n"));
8709 	}
8710 
8711 	if (s) {
8712 		memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions));
8713 		tag = s->tag;
8714 		rt = s->rt;
8715 	} else {
8716 		tag = r->tag;
8717 		rt = r->rt;
8718 	}
8719 
8720 	if (tag > 0 && pf_tag_packet(m, &pd, tag)) {
8721 		action = PF_DROP;
8722 		REASON_SET(&reason, PFRES_MEMORY);
8723 	}
8724 
8725 	pf_scrub_ip(&m, &pd);
8726 	if (pd.proto == IPPROTO_TCP && pd.act.max_mss)
8727 		pf_normalize_mss(m, off, &pd);
8728 
8729 	if (pd.act.rtableid >= 0)
8730 		M_SETFIB(m, pd.act.rtableid);
8731 
8732 	if (pd.act.flags & PFSTATE_SETPRIO) {
8733 		if (pd.tos & IPTOS_LOWDELAY)
8734 			use_2nd_queue = 1;
8735 		if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) {
8736 			action = PF_DROP;
8737 			REASON_SET(&reason, PFRES_MEMORY);
8738 			pd.act.log = PF_LOG_FORCE;
8739 			DPFPRINTF(PF_DEBUG_MISC,
8740 			    ("pf: failed to allocate 802.1q mtag\n"));
8741 		}
8742 	}
8743 
8744 #ifdef ALTQ
8745 	if (action == PF_PASS && pd.act.qid) {
8746 		if (pd.pf_mtag == NULL &&
8747 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8748 			action = PF_DROP;
8749 			REASON_SET(&reason, PFRES_MEMORY);
8750 		} else {
8751 			if (s != NULL)
8752 				pd.pf_mtag->qid_hash = pf_state_hash(s);
8753 			if (use_2nd_queue || (pd.tos & IPTOS_LOWDELAY))
8754 				pd.pf_mtag->qid = pd.act.pqid;
8755 			else
8756 				pd.pf_mtag->qid = pd.act.qid;
8757 			/* Add hints for ecn. */
8758 			pd.pf_mtag->hdr = h;
8759 		}
8760 	}
8761 #endif /* ALTQ */
8762 
8763 	/*
8764 	 * connections redirected to loopback should not match sockets
8765 	 * bound specifically to loopback due to security implications,
8766 	 * see tcp_input() and in_pcblookup_listen().
8767 	 */
8768 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
8769 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
8770 	    (s->nat_rule.ptr->action == PF_RDR ||
8771 	    s->nat_rule.ptr->action == PF_BINAT) &&
8772 	    IN_LOOPBACK(ntohl(pd.dst->v4.s_addr)))
8773 		m->m_flags |= M_SKIP_FIREWALL;
8774 
8775 	if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS &&
8776 	    r->divert.port && !PACKET_LOOPED(&pd)) {
8777 		mtag = m_tag_alloc(MTAG_PF_DIVERT, 0,
8778 		    sizeof(struct pf_divert_mtag), M_NOWAIT | M_ZERO);
8779 		if (mtag != NULL) {
8780 			((struct pf_divert_mtag *)(mtag+1))->port =
8781 			    ntohs(r->divert.port);
8782 			((struct pf_divert_mtag *)(mtag+1))->idir =
8783 			    (dir == PF_IN) ? PF_DIVERT_MTAG_DIR_IN :
8784 			    PF_DIVERT_MTAG_DIR_OUT;
8785 
8786 			if (s)
8787 				PF_STATE_UNLOCK(s);
8788 
8789 			m_tag_prepend(m, mtag);
8790 			if (m->m_flags & M_FASTFWD_OURS) {
8791 				if (pd.pf_mtag == NULL &&
8792 				    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
8793 					action = PF_DROP;
8794 					REASON_SET(&reason, PFRES_MEMORY);
8795 					pd.act.log = PF_LOG_FORCE;
8796 					DPFPRINTF(PF_DEBUG_MISC,
8797 					    ("pf: failed to allocate tag\n"));
8798 				} else {
8799 					pd.pf_mtag->flags |=
8800 					    PF_MTAG_FLAG_FASTFWD_OURS_PRESENT;
8801 					m->m_flags &= ~M_FASTFWD_OURS;
8802 				}
8803 			}
8804 			ip_divert_ptr(*m0, dir == PF_IN);
8805 			*m0 = NULL;
8806 
8807 			return (action);
8808 		} else {
8809 			/* XXX: ipfw has the same behaviour! */
8810 			action = PF_DROP;
8811 			REASON_SET(&reason, PFRES_MEMORY);
8812 			pd.act.log = PF_LOG_FORCE;
8813 			DPFPRINTF(PF_DEBUG_MISC,
8814 			    ("pf: failed to allocate divert tag\n"));
8815 		}
8816 	}
8817 	/* this flag will need revising if the pkt is forwarded */
8818 	if (pd.pf_mtag)
8819 		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_PACKET_LOOPED;
8820 
8821 	if (pd.act.log) {
8822 		struct pf_krule		*lr;
8823 		struct pf_krule_item	*ri;
8824 
8825 		if (s != NULL && s->nat_rule.ptr != NULL &&
8826 		    s->nat_rule.ptr->log & PF_LOG_ALL)
8827 			lr = s->nat_rule.ptr;
8828 		else
8829 			lr = r;
8830 
8831 		if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL)
8832 			PFLOG_PACKET(kif, m, AF_INET, reason, lr, a, ruleset,
8833 			    &pd, (s == NULL));
8834 		if (s) {
8835 			SLIST_FOREACH(ri, &s->match_rules, entry)
8836 				if (ri->r->log & PF_LOG_ALL)
8837 					PFLOG_PACKET(kif, m, AF_INET, reason,
8838 					    ri->r, a, ruleset, &pd, 0);
8839 		}
8840 	}
8841 
8842 	pf_counter_u64_critical_enter();
8843 	pf_counter_u64_add_protected(&kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS],
8844 	    pd.tot_len);
8845 	pf_counter_u64_add_protected(&kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS],
8846 	    1);
8847 
8848 	if (action == PF_PASS || r->action == PF_DROP) {
8849 		dirndx = (dir == PF_OUT);
8850 		pf_counter_u64_add_protected(&r->packets[dirndx], 1);
8851 		pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len);
8852 		pf_update_timestamp(r);
8853 
8854 		if (a != NULL) {
8855 			pf_counter_u64_add_protected(&a->packets[dirndx], 1);
8856 			pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len);
8857 		}
8858 		if (s != NULL) {
8859 			struct pf_krule_item	*ri;
8860 
8861 			if (s->nat_rule.ptr != NULL) {
8862 				pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx],
8863 				    1);
8864 				pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx],
8865 				    pd.tot_len);
8866 			}
8867 			if (s->src_node != NULL) {
8868 				counter_u64_add(s->src_node->packets[dirndx],
8869 				    1);
8870 				counter_u64_add(s->src_node->bytes[dirndx],
8871 				    pd.tot_len);
8872 			}
8873 			if (s->nat_src_node != NULL) {
8874 				counter_u64_add(s->nat_src_node->packets[dirndx],
8875 				    1);
8876 				counter_u64_add(s->nat_src_node->bytes[dirndx],
8877 				    pd.tot_len);
8878 			}
8879 			dirndx = (dir == s->direction) ? 0 : 1;
8880 			s->packets[dirndx]++;
8881 			s->bytes[dirndx] += pd.tot_len;
8882 			SLIST_FOREACH(ri, &s->match_rules, entry) {
8883 				pf_counter_u64_add_protected(&ri->r->packets[dirndx], 1);
8884 				pf_counter_u64_add_protected(&ri->r->bytes[dirndx], pd.tot_len);
8885 			}
8886 		}
8887 		tr = r;
8888 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
8889 		if (nr != NULL && r == &V_pf_default_rule)
8890 			tr = nr;
8891 		if (tr->src.addr.type == PF_ADDR_TABLE)
8892 			pfr_update_stats(tr->src.addr.p.tbl,
8893 			    (s == NULL) ? pd.src :
8894 			    &s->key[(s->direction == PF_IN)]->
8895 				addr[(s->direction == PF_OUT)],
8896 			    pd.af, pd.tot_len, dir == PF_OUT,
8897 			    r->action == PF_PASS, tr->src.neg);
8898 		if (tr->dst.addr.type == PF_ADDR_TABLE)
8899 			pfr_update_stats(tr->dst.addr.p.tbl,
8900 			    (s == NULL) ? pd.dst :
8901 			    &s->key[(s->direction == PF_IN)]->
8902 				addr[(s->direction == PF_IN)],
8903 			    pd.af, pd.tot_len, dir == PF_OUT,
8904 			    r->action == PF_PASS, tr->dst.neg);
8905 	}
8906 	pf_counter_u64_critical_exit();
8907 
8908 	switch (action) {
8909 	case PF_SYNPROXY_DROP:
8910 		m_freem(*m0);
8911 	case PF_DEFER:
8912 		*m0 = NULL;
8913 		action = PF_PASS;
8914 		break;
8915 	case PF_DROP:
8916 		m_freem(*m0);
8917 		*m0 = NULL;
8918 		break;
8919 	default:
8920 		/* pf_route() returns unlocked. */
8921 		if (rt) {
8922 			pf_route(m0, r, kif->pfik_ifp, s, &pd, inp);
8923 			goto out;
8924 		}
8925 		if (pf_dummynet(&pd, s, r, m0) != 0) {
8926 			action = PF_DROP;
8927 			REASON_SET(&reason, PFRES_MEMORY);
8928 		}
8929 		break;
8930 	}
8931 
8932 	SDT_PROBE4(pf, ip, test, done, action, reason, r, s);
8933 
8934 	if (s)
8935 		PF_STATE_UNLOCK(s);
8936 
8937 out:
8938 	pf_sctp_multihome_delayed(&pd, off, kif, s, action);
8939 
8940 	return (action);
8941 }
8942 #endif /* INET */
8943 
8944 #ifdef INET6
8945 int
pf_test6(int dir,int pflags,struct ifnet * ifp,struct mbuf ** m0,struct inpcb * inp,struct pf_rule_actions * default_actions)8946 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp,
8947     struct pf_rule_actions *default_actions)
8948 {
8949 	struct pfi_kkif		*kif;
8950 	u_short			 action, reason = 0;
8951 	struct mbuf		*m = *m0, *n = NULL;
8952 	struct m_tag		*mtag;
8953 	struct ip6_hdr		*h = NULL;
8954 	struct pf_krule		*a = NULL, *r = &V_pf_default_rule, *tr, *nr;
8955 	struct pf_kstate	*s = NULL;
8956 	struct pf_kruleset	*ruleset = NULL;
8957 	struct pf_pdesc		 pd;
8958 	int			 off, terminal = 0, dirndx, rh_cnt = 0, use_2nd_queue = 0;
8959 	uint16_t		 tag;
8960 	uint8_t			 rt;
8961 
8962 	PF_RULES_RLOCK_TRACKER;
8963 	KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir));
8964 	M_ASSERTPKTHDR(m);
8965 
8966 	if (!V_pf_status.running)
8967 		return (PF_PASS);
8968 
8969 	PF_RULES_RLOCK();
8970 
8971 	kif = (struct pfi_kkif *)ifp->if_pf_kif;
8972 	if (__predict_false(kif == NULL)) {
8973 		DPFPRINTF(PF_DEBUG_URGENT,
8974 		    ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
8975 		PF_RULES_RUNLOCK();
8976 		return (PF_DROP);
8977 	}
8978 	if (kif->pfik_flags & PFI_IFLAG_SKIP) {
8979 		PF_RULES_RUNLOCK();
8980 		return (PF_PASS);
8981 	}
8982 
8983 	if (m->m_flags & M_SKIP_FIREWALL) {
8984 		PF_RULES_RUNLOCK();
8985 		return (PF_PASS);
8986 	}
8987 
8988 	memset(&pd, 0, sizeof(pd));
8989 	TAILQ_INIT(&pd.sctp_multihome_jobs);
8990 	if (default_actions != NULL)
8991 		memcpy(&pd.act, default_actions, sizeof(pd.act));
8992 	pd.pf_mtag = pf_find_mtag(m);
8993 
8994 	if (pd.pf_mtag != NULL && (pd.pf_mtag->flags & PF_MTAG_FLAG_ROUTE_TO)) {
8995 		pd.pf_mtag->flags &= ~PF_MTAG_FLAG_ROUTE_TO;
8996 
8997 		ifp = ifnet_byindexgen(pd.pf_mtag->if_index,
8998 		    pd.pf_mtag->if_idxgen);
8999 		if (ifp == NULL || ifp->if_flags & IFF_DYING) {
9000 			PF_RULES_RUNLOCK();
9001 			m_freem(*m0);
9002 			*m0 = NULL;
9003 			return (PF_PASS);
9004 		}
9005 		PF_RULES_RUNLOCK();
9006 		nd6_output_ifp(ifp, ifp, m,
9007                     (struct sockaddr_in6 *)&pd.pf_mtag->dst, NULL);
9008 		*m0 = NULL;
9009 		return (PF_PASS);
9010 	}
9011 
9012 	if (pd.pf_mtag && pd.pf_mtag->dnpipe) {
9013 		pd.act.dnpipe = pd.pf_mtag->dnpipe;
9014 		pd.act.flags = pd.pf_mtag->dnflags;
9015 	}
9016 
9017 	if (ip_dn_io_ptr != NULL && pd.pf_mtag != NULL &&
9018 	    pd.pf_mtag->flags & PF_MTAG_FLAG_DUMMYNET) {
9019 		pf_dummynet_flag_remove(m, pd.pf_mtag);
9020 		/* Dummynet re-injects packets after they've
9021 		 * completed their delay. We've already
9022 		 * processed them, so pass unconditionally. */
9023 		PF_RULES_RUNLOCK();
9024 		return (PF_PASS);
9025 	}
9026 
9027 	pd.sport = pd.dport = NULL;
9028 	pd.ip_sum = NULL;
9029 	pd.proto_sum = NULL;
9030 	pd.dir = dir;
9031 	pd.sidx = (dir == PF_IN) ? 0 : 1;
9032 	pd.didx = (dir == PF_IN) ? 1 : 0;
9033 	pd.af = AF_INET6;
9034 	pd.act.rtableid = -1;
9035 
9036 	h = mtod(m, struct ip6_hdr *);
9037 	off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
9038 
9039 	/* We do IP header normalization and packet reassembly here */
9040 	if (pf_normalize_ip6(m0, kif, &reason, &pd) != PF_PASS) {
9041 		m = *m0;
9042 		action = PF_DROP;
9043 		goto done;
9044 	}
9045 	m = *m0;	/* pf_normalize messes with m0 */
9046 	h = mtod(m, struct ip6_hdr *);
9047 	off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
9048 
9049 	/*
9050 	 * we do not support jumbogram.  if we keep going, zero ip6_plen
9051 	 * will do something bad, so drop the packet for now.
9052 	 */
9053 	if (htons(h->ip6_plen) == 0) {
9054 		action = PF_DROP;
9055 		REASON_SET(&reason, PFRES_NORM);	/*XXX*/
9056 		goto done;
9057 	}
9058 
9059 	pd.src = (struct pf_addr *)&h->ip6_src;
9060 	pd.dst = (struct pf_addr *)&h->ip6_dst;
9061 	PF_ACPY(&pd.osrc, pd.src, pd.af);
9062 	PF_ACPY(&pd.odst, pd.dst, pd.af);
9063 	pd.tos = IPV6_DSCP(h);
9064 	pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
9065 
9066 	pd.proto = h->ip6_nxt;
9067 	do {
9068 		switch (pd.proto) {
9069 		case IPPROTO_FRAGMENT:
9070 			action = pf_test_fragment(&r, kif, m, h, &pd, &a,
9071 			    &ruleset);
9072 			if (action == PF_DROP)
9073 				REASON_SET(&reason, PFRES_FRAG);
9074 			goto done;
9075 		case IPPROTO_ROUTING: {
9076 			struct ip6_rthdr rthdr;
9077 
9078 			if (rh_cnt++) {
9079 				DPFPRINTF(PF_DEBUG_MISC,
9080 				    ("pf: IPv6 more than one rthdr\n"));
9081 				action = PF_DROP;
9082 				REASON_SET(&reason, PFRES_IPOPTIONS);
9083 				pd.act.log = PF_LOG_FORCE;
9084 				goto done;
9085 			}
9086 			if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
9087 			    &reason, pd.af)) {
9088 				DPFPRINTF(PF_DEBUG_MISC,
9089 				    ("pf: IPv6 short rthdr\n"));
9090 				action = PF_DROP;
9091 				REASON_SET(&reason, PFRES_SHORT);
9092 				pd.act.log = PF_LOG_FORCE;
9093 				goto done;
9094 			}
9095 			if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
9096 				DPFPRINTF(PF_DEBUG_MISC,
9097 				    ("pf: IPv6 rthdr0\n"));
9098 				action = PF_DROP;
9099 				REASON_SET(&reason, PFRES_IPOPTIONS);
9100 				pd.act.log = PF_LOG_FORCE;
9101 				goto done;
9102 			}
9103 			/* FALLTHROUGH */
9104 		}
9105 		case IPPROTO_AH:
9106 		case IPPROTO_HOPOPTS:
9107 		case IPPROTO_DSTOPTS: {
9108 			/* get next header and header length */
9109 			struct ip6_ext	opt6;
9110 
9111 			if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
9112 			    NULL, &reason, pd.af)) {
9113 				DPFPRINTF(PF_DEBUG_MISC,
9114 				    ("pf: IPv6 short opt\n"));
9115 				action = PF_DROP;
9116 				pd.act.log = PF_LOG_FORCE;
9117 				goto done;
9118 			}
9119 			if (pd.proto == IPPROTO_AH)
9120 				off += (opt6.ip6e_len + 2) * 4;
9121 			else
9122 				off += (opt6.ip6e_len + 1) * 8;
9123 			pd.proto = opt6.ip6e_nxt;
9124 			/* goto the next header */
9125 			break;
9126 		}
9127 		default:
9128 			terminal++;
9129 			break;
9130 		}
9131 	} while (!terminal);
9132 
9133 	/* if there's no routing header, use unmodified mbuf for checksumming */
9134 	if (!n)
9135 		n = m;
9136 
9137 	switch (pd.proto) {
9138 	case IPPROTO_TCP: {
9139 		if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp),
9140 		    &action, &reason, AF_INET6)) {
9141 			if (action != PF_PASS)
9142 				pd.act.log |= PF_LOG_FORCE;
9143 			goto done;
9144 		}
9145 		pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2);
9146 		pd.sport = &pd.hdr.tcp.th_sport;
9147 		pd.dport = &pd.hdr.tcp.th_dport;
9148 
9149 		/* Respond to SYN with a syncookie. */
9150 		if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN &&
9151 		    pd.dir == PF_IN && pf_synflood_check(&pd)) {
9152 			pf_syncookie_send(m, off, &pd);
9153 			action = PF_DROP;
9154 			break;
9155 		}
9156 
9157 		action = pf_normalize_tcp(kif, m, 0, off, h, &pd);
9158 		if (action == PF_DROP)
9159 			goto done;
9160 		action = pf_test_state_tcp(&s, kif, m, off, h, &pd, &reason);
9161 		if (action == PF_PASS) {
9162 			if (V_pfsync_update_state_ptr != NULL)
9163 				V_pfsync_update_state_ptr(s);
9164 			r = s->rule.ptr;
9165 			a = s->anchor.ptr;
9166 		} else if (s == NULL) {
9167 			/* Validate remote SYN|ACK, re-create original SYN if
9168 			 * valid. */
9169 			if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) ==
9170 			    TH_ACK && pf_syncookie_validate(&pd) &&
9171 			    pd.dir == PF_IN) {
9172 				struct mbuf *msyn;
9173 
9174 				msyn = pf_syncookie_recreate_syn(h->ip6_hlim,
9175 				    off, &pd);
9176 				if (msyn == NULL) {
9177 					action = PF_DROP;
9178 					break;
9179 				}
9180 
9181 				action = pf_test6(dir, pflags, ifp, &msyn, inp,
9182 				    &pd.act);
9183 				m_freem(msyn);
9184 				if (action != PF_PASS)
9185 					break;
9186 
9187 				action = pf_test_state_tcp(&s, kif, m, off, h,
9188 				    &pd, &reason);
9189 				if (action != PF_PASS || s == NULL) {
9190 					action = PF_DROP;
9191 					break;
9192 				}
9193 
9194 				s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) - 1;
9195 				s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) - 1;
9196 				pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_DST);
9197 
9198 				action = pf_synproxy(&pd, &s, &reason);
9199 				break;
9200 			} else {
9201 				action = pf_test_rule(&r, &s, kif, m, off, &pd,
9202 				    &a, &ruleset, inp);
9203 			}
9204 		}
9205 		break;
9206 	}
9207 
9208 	case IPPROTO_UDP: {
9209 		if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp),
9210 		    &action, &reason, AF_INET6)) {
9211 			if (action != PF_PASS)
9212 				pd.act.log |= PF_LOG_FORCE;
9213 			goto done;
9214 		}
9215 		pd.sport = &pd.hdr.udp.uh_sport;
9216 		pd.dport = &pd.hdr.udp.uh_dport;
9217 		if (pd.hdr.udp.uh_dport == 0 ||
9218 		    ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off ||
9219 		    ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) {
9220 			action = PF_DROP;
9221 			REASON_SET(&reason, PFRES_SHORT);
9222 			goto done;
9223 		}
9224 		action = pf_test_state_udp(&s, kif, m, off, h, &pd);
9225 		if (action == PF_PASS) {
9226 			if (V_pfsync_update_state_ptr != NULL)
9227 				V_pfsync_update_state_ptr(s);
9228 			r = s->rule.ptr;
9229 			a = s->anchor.ptr;
9230 		} else if (s == NULL)
9231 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
9232 			    &a, &ruleset, inp);
9233 		break;
9234 	}
9235 
9236 	case IPPROTO_SCTP: {
9237 		if (!pf_pull_hdr(m, off, &pd.hdr.sctp, sizeof(pd.hdr.sctp),
9238 		    &action, &reason, AF_INET6)) {
9239 			if (action != PF_PASS)
9240 				pd.act.log |= PF_LOG_FORCE;
9241 			goto done;
9242 		}
9243 		pd.sport = &pd.hdr.sctp.src_port;
9244 		pd.dport = &pd.hdr.sctp.dest_port;
9245 		if (pd.hdr.sctp.src_port == 0 || pd.hdr.sctp.dest_port == 0) {
9246 			action = PF_DROP;
9247 			REASON_SET(&reason, PFRES_SHORT);
9248 			goto done;
9249 		}
9250 		action = pf_normalize_sctp(dir, kif, m, 0, off, h, &pd);
9251 		if (action == PF_DROP)
9252 			goto done;
9253 		action = pf_test_state_sctp(&s, kif, m, off, h, &pd,
9254 		    &reason);
9255 		if (action == PF_PASS) {
9256 			if (V_pfsync_update_state_ptr != NULL)
9257 				V_pfsync_update_state_ptr(s);
9258 			r = s->rule.ptr;
9259 			a = s->anchor.ptr;
9260 		} else if (s == NULL) {
9261 			action = pf_test_rule(&r, &s, kif, m, off,
9262 			    &pd, &a, &ruleset, inp);
9263 		}
9264 		break;
9265 	}
9266 
9267 	case IPPROTO_ICMP: {
9268 		action = PF_DROP;
9269 		DPFPRINTF(PF_DEBUG_MISC,
9270 		    ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
9271 		goto done;
9272 	}
9273 
9274 	case IPPROTO_ICMPV6: {
9275 		if (!pf_pull_hdr(m, off, &pd.hdr.icmp6, sizeof(pd.hdr.icmp6),
9276 		    &action, &reason, AF_INET6)) {
9277 			if (action != PF_PASS)
9278 				pd.act.log |= PF_LOG_FORCE;
9279 			goto done;
9280 		}
9281 		action = pf_test_state_icmp(&s, kif, m, off, h, &pd, &reason);
9282 		if (action == PF_PASS) {
9283 			if (V_pfsync_update_state_ptr != NULL)
9284 				V_pfsync_update_state_ptr(s);
9285 			r = s->rule.ptr;
9286 			a = s->anchor.ptr;
9287 		} else if (s == NULL)
9288 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
9289 			    &a, &ruleset, inp);
9290 		break;
9291 	}
9292 
9293 	default:
9294 		action = pf_test_state_other(&s, kif, m, &pd);
9295 		if (action == PF_PASS) {
9296 			if (V_pfsync_update_state_ptr != NULL)
9297 				V_pfsync_update_state_ptr(s);
9298 			r = s->rule.ptr;
9299 			a = s->anchor.ptr;
9300 		} else if (s == NULL)
9301 			action = pf_test_rule(&r, &s, kif, m, off, &pd,
9302 			    &a, &ruleset, inp);
9303 		break;
9304 	}
9305 
9306 done:
9307 	PF_RULES_RUNLOCK();
9308 	if (n != m) {
9309 		m_freem(n);
9310 		n = NULL;
9311 	}
9312 
9313 	if (m == NULL)
9314 		goto out;
9315 
9316 	/* handle dangerous IPv6 extension headers. */
9317 	if (action == PF_PASS && rh_cnt &&
9318 	    !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
9319 		action = PF_DROP;
9320 		REASON_SET(&reason, PFRES_IPOPTIONS);
9321 		pd.act.log = r->log;
9322 		DPFPRINTF(PF_DEBUG_MISC,
9323 		    ("pf: dropping packet with dangerous v6 headers\n"));
9324 	}
9325 
9326 	if (s) {
9327 		memcpy(&pd.act, &s->act, sizeof(struct pf_rule_actions));
9328 		tag = s->tag;
9329 		rt = s->rt;
9330 	} else {
9331 		tag = r->tag;
9332 		rt = r->rt;
9333 	}
9334 
9335 	if (tag > 0 && pf_tag_packet(m, &pd, tag)) {
9336 		action = PF_DROP;
9337 		REASON_SET(&reason, PFRES_MEMORY);
9338 	}
9339 
9340 	pf_scrub_ip6(&m, &pd);
9341 	if (pd.proto == IPPROTO_TCP && pd.act.max_mss)
9342 		pf_normalize_mss(m, off, &pd);
9343 
9344 	if (pd.act.rtableid >= 0)
9345 		M_SETFIB(m, pd.act.rtableid);
9346 
9347 	if (pd.act.flags & PFSTATE_SETPRIO) {
9348 		if (pd.tos & IPTOS_LOWDELAY)
9349 			use_2nd_queue = 1;
9350 		if (vlan_set_pcp(m, pd.act.set_prio[use_2nd_queue])) {
9351 			action = PF_DROP;
9352 			REASON_SET(&reason, PFRES_MEMORY);
9353 			pd.act.log = PF_LOG_FORCE;
9354 			DPFPRINTF(PF_DEBUG_MISC,
9355 			    ("pf: failed to allocate 802.1q mtag\n"));
9356 		}
9357 	}
9358 
9359 #ifdef ALTQ
9360 	if (action == PF_PASS && pd.act.qid) {
9361 		if (pd.pf_mtag == NULL &&
9362 		    ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
9363 			action = PF_DROP;
9364 			REASON_SET(&reason, PFRES_MEMORY);
9365 		} else {
9366 			if (s != NULL)
9367 				pd.pf_mtag->qid_hash = pf_state_hash(s);
9368 			if (pd.tos & IPTOS_LOWDELAY)
9369 				pd.pf_mtag->qid = pd.act.pqid;
9370 			else
9371 				pd.pf_mtag->qid = pd.act.qid;
9372 			/* Add hints for ecn. */
9373 			pd.pf_mtag->hdr = h;
9374 		}
9375 	}
9376 #endif /* ALTQ */
9377 
9378 	if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
9379 	    pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
9380 	    (s->nat_rule.ptr->action == PF_RDR ||
9381 	    s->nat_rule.ptr->action == PF_BINAT) &&
9382 	    IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
9383 		m->m_flags |= M_SKIP_FIREWALL;
9384 
9385 	/* XXX: Anybody working on it?! */
9386 	if (r->divert.port)
9387 		printf("pf: divert(9) is not supported for IPv6\n");
9388 
9389 	if (pd.act.log) {
9390 		struct pf_krule		*lr;
9391 		struct pf_krule_item	*ri;
9392 
9393 		if (s != NULL && s->nat_rule.ptr != NULL &&
9394 		    s->nat_rule.ptr->log & PF_LOG_ALL)
9395 			lr = s->nat_rule.ptr;
9396 		else
9397 			lr = r;
9398 
9399 		if (pd.act.log & PF_LOG_FORCE || lr->log & PF_LOG_ALL)
9400 			PFLOG_PACKET(kif, m, AF_INET6, reason, lr, a, ruleset,
9401 			    &pd, (s == NULL));
9402 		if (s) {
9403 			SLIST_FOREACH(ri, &s->match_rules, entry)
9404 				if (ri->r->log & PF_LOG_ALL)
9405 					PFLOG_PACKET(kif, m, AF_INET6, reason,
9406 					    ri->r, a, ruleset, &pd, 0);
9407 		}
9408 	}
9409 
9410 	pf_counter_u64_critical_enter();
9411 	pf_counter_u64_add_protected(&kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS],
9412 	    pd.tot_len);
9413 	pf_counter_u64_add_protected(&kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS],
9414 	    1);
9415 
9416 	if (action == PF_PASS || r->action == PF_DROP) {
9417 		dirndx = (dir == PF_OUT);
9418 		pf_counter_u64_add_protected(&r->packets[dirndx], 1);
9419 		pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len);
9420 		if (a != NULL) {
9421 			pf_counter_u64_add_protected(&a->packets[dirndx], 1);
9422 			pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len);
9423 		}
9424 		if (s != NULL) {
9425 			if (s->nat_rule.ptr != NULL) {
9426 				pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx],
9427 				    1);
9428 				pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx],
9429 				    pd.tot_len);
9430 			}
9431 			if (s->src_node != NULL) {
9432 				counter_u64_add(s->src_node->packets[dirndx],
9433 				    1);
9434 				counter_u64_add(s->src_node->bytes[dirndx],
9435 				    pd.tot_len);
9436 			}
9437 			if (s->nat_src_node != NULL) {
9438 				counter_u64_add(s->nat_src_node->packets[dirndx],
9439 				    1);
9440 				counter_u64_add(s->nat_src_node->bytes[dirndx],
9441 				    pd.tot_len);
9442 			}
9443 			dirndx = (dir == s->direction) ? 0 : 1;
9444 			s->packets[dirndx]++;
9445 			s->bytes[dirndx] += pd.tot_len;
9446 		}
9447 		tr = r;
9448 		nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
9449 		if (nr != NULL && r == &V_pf_default_rule)
9450 			tr = nr;
9451 		if (tr->src.addr.type == PF_ADDR_TABLE)
9452 			pfr_update_stats(tr->src.addr.p.tbl,
9453 			    (s == NULL) ? pd.src :
9454 			    &s->key[(s->direction == PF_IN)]->addr[0],
9455 			    pd.af, pd.tot_len, dir == PF_OUT,
9456 			    r->action == PF_PASS, tr->src.neg);
9457 		if (tr->dst.addr.type == PF_ADDR_TABLE)
9458 			pfr_update_stats(tr->dst.addr.p.tbl,
9459 			    (s == NULL) ? pd.dst :
9460 			    &s->key[(s->direction == PF_IN)]->addr[1],
9461 			    pd.af, pd.tot_len, dir == PF_OUT,
9462 			    r->action == PF_PASS, tr->dst.neg);
9463 	}
9464 	pf_counter_u64_critical_exit();
9465 
9466 	switch (action) {
9467 	case PF_SYNPROXY_DROP:
9468 		m_freem(*m0);
9469 	case PF_DEFER:
9470 		*m0 = NULL;
9471 		action = PF_PASS;
9472 		break;
9473 	case PF_DROP:
9474 		m_freem(*m0);
9475 		*m0 = NULL;
9476 		break;
9477 	default:
9478 		/* pf_route6() returns unlocked. */
9479 		if (rt) {
9480 			pf_route6(m0, r, kif->pfik_ifp, s, &pd, inp);
9481 			goto out;
9482 		}
9483 		if (pf_dummynet(&pd, s, r, m0) != 0) {
9484 			action = PF_DROP;
9485 			REASON_SET(&reason, PFRES_MEMORY);
9486 		}
9487 		break;
9488 	}
9489 
9490 	if (s)
9491 		PF_STATE_UNLOCK(s);
9492 
9493 	/* If reassembled packet passed, create new fragments. */
9494 	if (action == PF_PASS && *m0 && dir == PF_OUT &&
9495 	    (mtag = m_tag_find(m, PACKET_TAG_PF_REASSEMBLED, NULL)) != NULL)
9496 		action = pf_refragment6(ifp, m0, mtag, pflags & PFIL_FWD);
9497 
9498 out:
9499 	SDT_PROBE4(pf, ip, test6, done, action, reason, r, s);
9500 
9501 	pf_sctp_multihome_delayed(&pd, off, kif, s, action);
9502 
9503 	return (action);
9504 }
9505 #endif /* INET6 */
9506