2 * Copyright (c) 2001 Daniel Hartmeier
3 * Copyright (c) 2002 - 2008 Henning Brauer
4 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * - Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * - Redistributions in binary form must reproduce the above
14 * copyright notice, this list of conditions and the following
15 * disclaimer in the documentation and/or other materials provided
16 * with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
21 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
22 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
24 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
25 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
26 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
28 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
31 * Effort sponsored in part by the Defense Advanced Research Projects
32 * Agency (DARPA) and Air Force Research Laboratory, Air Force
33 * Materiel Command, USAF, under agreement number F30602-01-2-0537.
35 * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
42 #include "opt_inet6.h"
46 #include <sys/param.h>
48 #include <sys/endian.h>
50 #include <sys/interrupt.h>
51 #include <sys/kernel.h>
52 #include <sys/kthread.h>
53 #include <sys/limits.h>
56 #include <sys/random.h>
57 #include <sys/refcount.h>
58 #include <sys/socket.h>
59 #include <sys/sysctl.h>
60 #include <sys/taskqueue.h>
61 #include <sys/ucred.h>
64 #include <net/if_types.h>
65 #include <net/route.h>
66 #include <net/radix_mpath.h>
69 #include <net/pfvar.h>
70 #include <net/pf_mtag.h>
71 #include <net/if_pflog.h>
72 #include <net/if_pfsync.h>
74 #include <netinet/in_pcb.h>
75 #include <netinet/in_var.h>
76 #include <netinet/ip.h>
77 #include <netinet/ip_fw.h>
78 #include <netinet/ip_icmp.h>
79 #include <netinet/icmp_var.h>
80 #include <netinet/ip_var.h>
81 #include <netinet/tcp.h>
82 #include <netinet/tcp_fsm.h>
83 #include <netinet/tcp_seq.h>
84 #include <netinet/tcp_timer.h>
85 #include <netinet/tcp_var.h>
86 #include <netinet/udp.h>
87 #include <netinet/udp_var.h>
89 #include <netpfil/ipfw/ip_fw_private.h> /* XXX: only for DIR_IN/DIR_OUT */
92 #include <netinet/ip6.h>
93 #include <netinet/icmp6.h>
94 #include <netinet6/nd6.h>
95 #include <netinet6/ip6_var.h>
96 #include <netinet6/in6_pcb.h>
99 #include <machine/in_cksum.h>
100 #include <security/mac/mac_framework.h>
102 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
109 VNET_DEFINE(struct pf_altqqueue, pf_altqs[2]);
110 VNET_DEFINE(struct pf_palist, pf_pabuf);
111 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active);
112 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive);
113 VNET_DEFINE(struct pf_status, pf_status);
115 VNET_DEFINE(u_int32_t, ticket_altqs_active);
116 VNET_DEFINE(u_int32_t, ticket_altqs_inactive);
117 VNET_DEFINE(int, altqs_inactive_open);
118 VNET_DEFINE(u_int32_t, ticket_pabuf);
120 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx);
121 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx)
122 VNET_DEFINE(u_char, pf_tcp_secret[16]);
123 #define V_pf_tcp_secret VNET(pf_tcp_secret)
124 VNET_DEFINE(int, pf_tcp_secret_init);
125 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init)
126 VNET_DEFINE(int, pf_tcp_iss_off);
127 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off)
130 * Queue for pf_intr() sends.
132 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
133 struct pf_send_entry {
134 STAILQ_ENTRY(pf_send_entry) pfse_next;
151 #define pfse_icmp_type u.icmpopts.type
152 #define pfse_icmp_code u.icmpopts.code
153 #define pfse_icmp_mtu u.icmpopts.mtu
156 STAILQ_HEAD(pf_send_head, pf_send_entry);
157 static VNET_DEFINE(struct pf_send_head, pf_sendqueue);
158 #define V_pf_sendqueue VNET(pf_sendqueue)
160 static struct mtx pf_sendqueue_mtx;
161 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx)
162 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx)
165 * Queue for pf_overload_task() tasks.
167 struct pf_overload_entry {
168 SLIST_ENTRY(pf_overload_entry) next;
172 struct pf_rule *rule;
175 SLIST_HEAD(pf_overload_head, pf_overload_entry);
176 static VNET_DEFINE(struct pf_overload_head, pf_overloadqueue);
177 #define V_pf_overloadqueue VNET(pf_overloadqueue)
178 static VNET_DEFINE(struct task, pf_overloadtask);
179 #define V_pf_overloadtask VNET(pf_overloadtask)
181 static struct mtx pf_overloadqueue_mtx;
182 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx)
183 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx)
185 VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules);
186 struct mtx pf_unlnkdrules_mtx;
188 static VNET_DEFINE(uma_zone_t, pf_sources_z);
189 #define V_pf_sources_z VNET(pf_sources_z)
190 static VNET_DEFINE(uma_zone_t, pf_mtag_z);
191 #define V_pf_mtag_z VNET(pf_mtag_z)
192 VNET_DEFINE(uma_zone_t, pf_state_z);
193 VNET_DEFINE(uma_zone_t, pf_state_key_z);
195 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
196 #define PFID_CPUBITS 8
197 #define PFID_CPUSHIFT (sizeof(uint64_t) * NBBY - PFID_CPUBITS)
198 #define PFID_CPUMASK ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT)
199 #define PFID_MAXID (~PFID_CPUMASK)
200 CTASSERT((1 << PFID_CPUBITS) > MAXCPU);
202 static void pf_src_tree_remove_state(struct pf_state *);
203 static void pf_init_threshold(struct pf_threshold *, u_int32_t,
205 static void pf_add_threshold(struct pf_threshold *);
206 static int pf_check_threshold(struct pf_threshold *);
208 static void pf_change_ap(struct pf_addr *, u_int16_t *,
209 u_int16_t *, u_int16_t *, struct pf_addr *,
210 u_int16_t, u_int8_t, sa_family_t);
211 static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
212 struct tcphdr *, struct pf_state_peer *);
213 static void pf_change_icmp(struct pf_addr *, u_int16_t *,
214 struct pf_addr *, struct pf_addr *, u_int16_t,
215 u_int16_t *, u_int16_t *, u_int16_t *,
216 u_int16_t *, u_int8_t, sa_family_t);
217 static void pf_send_tcp(struct mbuf *,
218 const struct pf_rule *, sa_family_t,
219 const struct pf_addr *, const struct pf_addr *,
220 u_int16_t, u_int16_t, u_int32_t, u_int32_t,
221 u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
222 u_int16_t, struct ifnet *);
223 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
224 sa_family_t, struct pf_rule *);
225 static void pf_detach_state(struct pf_state *);
226 static int pf_state_key_attach(struct pf_state_key *,
227 struct pf_state_key *, struct pf_state *);
228 static void pf_state_key_detach(struct pf_state *, int);
229 static int pf_state_key_ctor(void *, int, void *, int);
230 static u_int32_t pf_tcp_iss(struct pf_pdesc *);
231 static int pf_test_rule(struct pf_rule **, struct pf_state **,
232 int, struct pfi_kif *, struct mbuf *, int,
233 struct pf_pdesc *, struct pf_rule **,
234 struct pf_ruleset **, struct inpcb *);
235 static int pf_create_state(struct pf_rule *, struct pf_rule *,
236 struct pf_rule *, struct pf_pdesc *,
237 struct pf_src_node *, struct pf_state_key *,
238 struct pf_state_key *, struct mbuf *, int,
239 u_int16_t, u_int16_t, int *, struct pfi_kif *,
240 struct pf_state **, int, u_int16_t, u_int16_t,
242 static int pf_test_fragment(struct pf_rule **, int,
243 struct pfi_kif *, struct mbuf *, void *,
244 struct pf_pdesc *, struct pf_rule **,
245 struct pf_ruleset **);
246 static int pf_tcp_track_full(struct pf_state_peer *,
247 struct pf_state_peer *, struct pf_state **,
248 struct pfi_kif *, struct mbuf *, int,
249 struct pf_pdesc *, u_short *, int *);
250 static int pf_tcp_track_sloppy(struct pf_state_peer *,
251 struct pf_state_peer *, struct pf_state **,
252 struct pf_pdesc *, u_short *);
253 static int pf_test_state_tcp(struct pf_state **, int,
254 struct pfi_kif *, struct mbuf *, int,
255 void *, struct pf_pdesc *, u_short *);
256 static int pf_test_state_udp(struct pf_state **, int,
257 struct pfi_kif *, struct mbuf *, int,
258 void *, struct pf_pdesc *);
259 static int pf_test_state_icmp(struct pf_state **, int,
260 struct pfi_kif *, struct mbuf *, int,
261 void *, struct pf_pdesc *, u_short *);
262 static int pf_test_state_other(struct pf_state **, int,
263 struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
264 static u_int8_t pf_get_wscale(struct mbuf *, int, u_int16_t,
266 static u_int16_t pf_get_mss(struct mbuf *, int, u_int16_t,
268 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t,
270 static void pf_set_rt_ifp(struct pf_state *,
272 static int pf_check_proto_cksum(struct mbuf *, int, int,
273 u_int8_t, sa_family_t);
274 static void pf_print_state_parts(struct pf_state *,
275 struct pf_state_key *, struct pf_state_key *);
276 static int pf_addr_wrap_neq(struct pf_addr_wrap *,
277 struct pf_addr_wrap *);
278 static struct pf_state *pf_find_state(struct pfi_kif *,
279 struct pf_state_key_cmp *, u_int);
280 static int pf_src_connlimit(struct pf_state **);
281 static void pf_overload_task(void *c, int pending);
282 static int pf_insert_src_node(struct pf_src_node **,
283 struct pf_rule *, struct pf_addr *, sa_family_t);
284 static u_int pf_purge_expired_states(u_int, int);
285 static void pf_purge_unlinked_rules(void);
286 static int pf_mtag_init(void *, int, int);
287 static void pf_mtag_free(struct m_tag *);
289 static void pf_route(struct mbuf **, struct pf_rule *, int,
290 struct ifnet *, struct pf_state *,
294 static void pf_change_a6(struct pf_addr *, u_int16_t *,
295 struct pf_addr *, u_int8_t);
296 static void pf_route6(struct mbuf **, struct pf_rule *, int,
297 struct ifnet *, struct pf_state *,
301 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
303 VNET_DECLARE(int, pf_end_threads);
305 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
307 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \
308 (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
310 #define STATE_LOOKUP(i, k, d, s, pd) \
312 (s) = pf_find_state((i), (k), (d)); \
313 if ((s) == NULL || (s)->timeout == PFTM_PURGE) \
315 if (PACKET_LOOPED(pd)) \
317 if ((d) == PF_OUT && \
318 (((s)->rule.ptr->rt == PF_ROUTETO && \
319 (s)->rule.ptr->direction == PF_OUT) || \
320 ((s)->rule.ptr->rt == PF_REPLYTO && \
321 (s)->rule.ptr->direction == PF_IN)) && \
322 (s)->rt_kif != NULL && \
323 (s)->rt_kif != (i)) \
327 #define BOUND_IFACE(r, k) \
328 ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
330 #define STATE_INC_COUNTERS(s) \
332 s->rule.ptr->states_cur++; \
333 s->rule.ptr->states_tot++; \
334 if (s->anchor.ptr != NULL) { \
335 s->anchor.ptr->states_cur++; \
336 s->anchor.ptr->states_tot++; \
338 if (s->nat_rule.ptr != NULL) { \
339 s->nat_rule.ptr->states_cur++; \
340 s->nat_rule.ptr->states_tot++; \
344 #define STATE_DEC_COUNTERS(s) \
346 if (s->nat_rule.ptr != NULL) \
347 s->nat_rule.ptr->states_cur--; \
348 if (s->anchor.ptr != NULL) \
349 s->anchor.ptr->states_cur--; \
350 s->rule.ptr->states_cur--; \
353 static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
354 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
355 VNET_DEFINE(struct pf_idhash *, pf_idhash);
356 VNET_DEFINE(u_long, pf_hashmask);
357 VNET_DEFINE(struct pf_srchash *, pf_srchash);
358 VNET_DEFINE(u_long, pf_srchashmask);
360 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)");
362 VNET_DEFINE(u_long, pf_hashsize);
363 #define V_pf_hashsize VNET(pf_hashsize)
364 SYSCTL_VNET_UINT(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
365 &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable");
367 VNET_DEFINE(u_long, pf_srchashsize);
368 #define V_pf_srchashsize VNET(pf_srchashsize)
369 SYSCTL_VNET_UINT(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
370 &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable");
372 VNET_DEFINE(void *, pf_swi_cookie);
374 VNET_DEFINE(uint32_t, pf_hashseed);
375 #define V_pf_hashseed VNET(pf_hashseed)
377 static __inline uint32_t
378 pf_hashkey(struct pf_state_key *sk)
382 h = jenkins_hash32((uint32_t *)sk,
383 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
386 return (h & V_pf_hashmask);
389 static __inline uint32_t
390 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
396 h = jenkins_hash32((uint32_t *)&addr->v4,
397 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
400 h = jenkins_hash32((uint32_t *)&addr->v6,
401 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
404 panic("%s: unknown address family %u", __func__, af);
407 return (h & V_pf_srchashmask);
412 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
417 dst->addr32[0] = src->addr32[0];
421 dst->addr32[0] = src->addr32[0];
422 dst->addr32[1] = src->addr32[1];
423 dst->addr32[2] = src->addr32[2];
424 dst->addr32[3] = src->addr32[3];
431 pf_init_threshold(struct pf_threshold *threshold,
432 u_int32_t limit, u_int32_t seconds)
434 threshold->limit = limit * PF_THRESHOLD_MULT;
435 threshold->seconds = seconds;
436 threshold->count = 0;
437 threshold->last = time_uptime;
441 pf_add_threshold(struct pf_threshold *threshold)
443 u_int32_t t = time_uptime, diff = t - threshold->last;
445 if (diff >= threshold->seconds)
446 threshold->count = 0;
448 threshold->count -= threshold->count * diff /
450 threshold->count += PF_THRESHOLD_MULT;
455 pf_check_threshold(struct pf_threshold *threshold)
457 return (threshold->count > threshold->limit);
461 pf_src_connlimit(struct pf_state **state)
463 struct pf_overload_entry *pfoe;
466 PF_STATE_LOCK_ASSERT(*state);
468 (*state)->src_node->conn++;
469 (*state)->src.tcp_est = 1;
470 pf_add_threshold(&(*state)->src_node->conn_rate);
472 if ((*state)->rule.ptr->max_src_conn &&
473 (*state)->rule.ptr->max_src_conn <
474 (*state)->src_node->conn) {
475 V_pf_status.lcounters[LCNT_SRCCONN]++;
479 if ((*state)->rule.ptr->max_src_conn_rate.limit &&
480 pf_check_threshold(&(*state)->src_node->conn_rate)) {
481 V_pf_status.lcounters[LCNT_SRCCONNRATE]++;
488 /* Kill this state. */
489 (*state)->timeout = PFTM_PURGE;
490 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
492 if ((*state)->rule.ptr->overload_tbl == NULL)
495 /* Schedule overloading and flushing task. */
496 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
498 return (1); /* too bad :( */
500 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
501 pfoe->af = (*state)->key[PF_SK_WIRE]->af;
502 pfoe->rule = (*state)->rule.ptr;
503 pfoe->dir = (*state)->direction;
505 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
506 PF_OVERLOADQ_UNLOCK();
507 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
513 pf_overload_task(void *c, int pending)
515 struct pf_overload_head queue;
517 struct pf_overload_entry *pfoe, *pfoe1;
521 queue = *(struct pf_overload_head *)c;
522 SLIST_INIT((struct pf_overload_head *)c);
523 PF_OVERLOADQ_UNLOCK();
525 bzero(&p, sizeof(p));
526 SLIST_FOREACH(pfoe, &queue, next) {
527 V_pf_status.lcounters[LCNT_OVERLOAD_TABLE]++;
528 if (V_pf_status.debug >= PF_DEBUG_MISC) {
529 printf("%s: blocking address ", __func__);
530 pf_print_host(&pfoe->addr, 0, pfoe->af);
534 p.pfra_af = pfoe->af;
539 p.pfra_ip4addr = pfoe->addr.v4;
545 p.pfra_ip6addr = pfoe->addr.v6;
551 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
556 * Remove those entries, that don't need flushing.
558 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
559 if (pfoe->rule->flush == 0) {
560 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
561 free(pfoe, M_PFTEMP);
563 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH]++;
565 /* If nothing to flush, return. */
566 if (SLIST_EMPTY(&queue))
569 for (int i = 0; i <= V_pf_hashmask; i++) {
570 struct pf_idhash *ih = &V_pf_idhash[i];
571 struct pf_state_key *sk;
575 LIST_FOREACH(s, &ih->states, entry) {
576 sk = s->key[PF_SK_WIRE];
577 SLIST_FOREACH(pfoe, &queue, next)
578 if (sk->af == pfoe->af &&
579 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
580 pfoe->rule == s->rule.ptr) &&
581 ((pfoe->dir == PF_OUT &&
582 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
583 (pfoe->dir == PF_IN &&
584 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
585 s->timeout = PFTM_PURGE;
586 s->src.state = s->dst.state = TCPS_CLOSED;
590 PF_HASHROW_UNLOCK(ih);
592 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
593 free(pfoe, M_PFTEMP);
594 if (V_pf_status.debug >= PF_DEBUG_MISC)
595 printf("%s: %u states killed", __func__, killed);
599 * Can return locked on failure, so that we can consistently
600 * allocate and insert a new one.
603 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
606 struct pf_srchash *sh;
607 struct pf_src_node *n;
609 V_pf_status.scounters[SCNT_SRC_NODE_SEARCH]++;
611 sh = &V_pf_srchash[pf_hashsrc(src, af)];
613 LIST_FOREACH(n, &sh->nodes, entry)
614 if (n->rule.ptr == rule && n->af == af &&
615 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
616 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
618 if (n != NULL || returnlocked == 0)
619 PF_HASHROW_UNLOCK(sh);
625 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
626 struct pf_addr *src, sa_family_t af)
629 KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
630 rule->rpool.opts & PF_POOL_STICKYADDR),
631 ("%s for non-tracking rule %p", __func__, rule));
634 *sn = pf_find_src_node(src, rule, af, 1);
637 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
639 PF_HASHROW_ASSERT(sh);
641 if (!rule->max_src_nodes ||
642 rule->src_nodes < rule->max_src_nodes)
643 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
645 V_pf_status.lcounters[LCNT_SRCNODES]++;
647 PF_HASHROW_UNLOCK(sh);
651 pf_init_threshold(&(*sn)->conn_rate,
652 rule->max_src_conn_rate.limit,
653 rule->max_src_conn_rate.seconds);
656 (*sn)->rule.ptr = rule;
657 PF_ACPY(&(*sn)->addr, src, af);
658 LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
659 (*sn)->creation = time_uptime;
660 (*sn)->ruletype = rule->action;
661 if ((*sn)->rule.ptr != NULL)
662 (*sn)->rule.ptr->src_nodes++;
663 PF_HASHROW_UNLOCK(sh);
664 V_pf_status.scounters[SCNT_SRC_NODE_INSERT]++;
665 V_pf_status.src_nodes++;
667 if (rule->max_src_states &&
668 (*sn)->states >= rule->max_src_states) {
669 V_pf_status.lcounters[LCNT_SRCSTATES]++;
677 pf_remove_src_node(struct pf_src_node *src)
679 struct pf_srchash *sh;
681 sh = &V_pf_srchash[pf_hashsrc(&src->addr, src->af)];
683 LIST_REMOVE(src, entry);
684 PF_HASHROW_UNLOCK(sh);
686 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS]++;
687 V_pf_status.src_nodes--;
689 uma_zfree(V_pf_sources_z, src);
692 /* Data storage structures initialization. */
696 struct pf_keyhash *kh;
697 struct pf_idhash *ih;
698 struct pf_srchash *sh;
701 TUNABLE_ULONG_FETCH("net.pf.states_hashsize", &V_pf_hashsize);
702 if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize))
703 V_pf_hashsize = PF_HASHSIZ;
704 TUNABLE_ULONG_FETCH("net.pf.source_nodes_hashsize", &V_pf_srchashsize);
705 if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize))
706 V_pf_srchashsize = PF_HASHSIZ / 4;
708 V_pf_hashseed = arc4random();
710 /* States and state keys storage. */
711 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
712 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
713 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
714 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
715 uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
717 V_pf_state_key_z = uma_zcreate("pf state keys",
718 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
720 V_pf_keyhash = malloc(V_pf_hashsize * sizeof(struct pf_keyhash),
721 M_PFHASH, M_WAITOK | M_ZERO);
722 V_pf_idhash = malloc(V_pf_hashsize * sizeof(struct pf_idhash),
723 M_PFHASH, M_WAITOK | M_ZERO);
724 V_pf_hashmask = V_pf_hashsize - 1;
725 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
727 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
728 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
732 V_pf_sources_z = uma_zcreate("pf source nodes",
733 sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
735 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
736 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
737 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
738 V_pf_srchash = malloc(V_pf_srchashsize * sizeof(struct pf_srchash),
739 M_PFHASH, M_WAITOK|M_ZERO);
740 V_pf_srchashmask = V_pf_srchashsize - 1;
741 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++)
742 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
745 TAILQ_INIT(&V_pf_altqs[0]);
746 TAILQ_INIT(&V_pf_altqs[1]);
747 TAILQ_INIT(&V_pf_pabuf);
748 V_pf_altqs_active = &V_pf_altqs[0];
749 V_pf_altqs_inactive = &V_pf_altqs[1];
752 V_pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
753 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_init, NULL,
756 /* Send & overload+flush queues. */
757 STAILQ_INIT(&V_pf_sendqueue);
758 SLIST_INIT(&V_pf_overloadqueue);
759 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, &V_pf_overloadqueue);
760 mtx_init(&pf_sendqueue_mtx, "pf send queue", NULL, MTX_DEF);
761 mtx_init(&pf_overloadqueue_mtx, "pf overload/flush queue", NULL,
764 /* Unlinked, but may be referenced rules. */
765 TAILQ_INIT(&V_pf_unlinked_rules);
766 mtx_init(&pf_unlnkdrules_mtx, "pf unlinked rules", NULL, MTX_DEF);
772 struct pf_keyhash *kh;
773 struct pf_idhash *ih;
774 struct pf_srchash *sh;
775 struct pf_send_entry *pfse, *next;
778 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
780 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
782 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
784 mtx_destroy(&kh->lock);
785 mtx_destroy(&ih->lock);
787 free(V_pf_keyhash, M_PFHASH);
788 free(V_pf_idhash, M_PFHASH);
790 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
791 KASSERT(LIST_EMPTY(&sh->nodes),
792 ("%s: source node hash not empty", __func__));
793 mtx_destroy(&sh->lock);
795 free(V_pf_srchash, M_PFHASH);
797 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
798 m_freem(pfse->pfse_m);
799 free(pfse, M_PFTEMP);
802 mtx_destroy(&pf_sendqueue_mtx);
803 mtx_destroy(&pf_overloadqueue_mtx);
804 mtx_destroy(&pf_unlnkdrules_mtx);
806 uma_zdestroy(V_pf_mtag_z);
807 uma_zdestroy(V_pf_sources_z);
808 uma_zdestroy(V_pf_state_z);
809 uma_zdestroy(V_pf_state_key_z);
813 pf_mtag_init(void *mem, int size, int how)
817 t = (struct m_tag *)mem;
818 t->m_tag_cookie = MTAG_ABI_COMPAT;
819 t->m_tag_id = PACKET_TAG_PF;
820 t->m_tag_len = sizeof(struct pf_mtag);
821 t->m_tag_free = pf_mtag_free;
827 pf_mtag_free(struct m_tag *t)
830 uma_zfree(V_pf_mtag_z, t);
834 pf_get_mtag(struct mbuf *m)
838 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
839 return ((struct pf_mtag *)(mtag + 1));
841 mtag = uma_zalloc(V_pf_mtag_z, M_NOWAIT);
844 bzero(mtag + 1, sizeof(struct pf_mtag));
845 m_tag_prepend(m, mtag);
847 return ((struct pf_mtag *)(mtag + 1));
851 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
854 struct pf_keyhash *khs, *khw, *kh;
855 struct pf_state_key *sk, *cur;
856 struct pf_state *si, *olds = NULL;
859 KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
860 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
861 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
864 * We need to lock hash slots of both keys. To avoid deadlock
865 * we always lock the slot with lower address first. Unlock order
868 * We also need to lock ID hash slot before dropping key
869 * locks. On success we return with ID hash slot locked.
873 khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
874 PF_HASHROW_LOCK(khs);
876 khs = &V_pf_keyhash[pf_hashkey(sks)];
877 khw = &V_pf_keyhash[pf_hashkey(skw)];
879 PF_HASHROW_LOCK(khs);
880 } else if (khs < khw) {
881 PF_HASHROW_LOCK(khs);
882 PF_HASHROW_LOCK(khw);
884 PF_HASHROW_LOCK(khw);
885 PF_HASHROW_LOCK(khs);
889 #define KEYS_UNLOCK() do { \
891 PF_HASHROW_UNLOCK(khs); \
892 PF_HASHROW_UNLOCK(khw); \
894 PF_HASHROW_UNLOCK(khs); \
898 * First run: start with wire key.
905 LIST_FOREACH(cur, &kh->keys, entry)
906 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
910 /* Key exists. Check for same kif, if none, add to key. */
911 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
912 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
915 if (si->kif == s->kif &&
916 si->direction == s->direction) {
917 if (sk->proto == IPPROTO_TCP &&
918 si->src.state >= TCPS_FIN_WAIT_2 &&
919 si->dst.state >= TCPS_FIN_WAIT_2) {
921 * New state matches an old >FIN_WAIT_2
922 * state. We can't drop key hash locks,
923 * thus we can't unlink it properly.
925 * As a workaround we drop it into
926 * TCPS_CLOSED state, schedule purge
927 * ASAP and push it into the very end
928 * of the slot TAILQ, so that it won't
929 * conflict with our new state.
931 si->src.state = si->dst.state =
933 si->timeout = PFTM_PURGE;
936 if (V_pf_status.debug >= PF_DEBUG_MISC) {
937 printf("pf: %s key attach "
939 (idx == PF_SK_WIRE) ?
942 pf_print_state_parts(s,
943 (idx == PF_SK_WIRE) ?
945 (idx == PF_SK_STACK) ?
947 printf(", existing: ");
948 pf_print_state_parts(si,
949 (idx == PF_SK_WIRE) ?
951 (idx == PF_SK_STACK) ?
955 PF_HASHROW_UNLOCK(ih);
957 uma_zfree(V_pf_state_key_z, sk);
958 if (idx == PF_SK_STACK)
960 return (EEXIST); /* collision! */
963 PF_HASHROW_UNLOCK(ih);
965 uma_zfree(V_pf_state_key_z, sk);
968 LIST_INSERT_HEAD(&kh->keys, sk, entry);
973 /* List is sorted, if-bound states before floating. */
974 if (s->kif == V_pfi_all)
975 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
977 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
980 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
981 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
987 * Attach done. See how should we (or should not?)
988 * attach a second key.
991 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
995 } else if (sks != NULL) {
997 * Continue attaching with stack key.
1009 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1010 ("%s failure", __func__));
1017 pf_detach_state(struct pf_state *s)
1019 struct pf_state_key *sks = s->key[PF_SK_STACK];
1020 struct pf_keyhash *kh;
1023 kh = &V_pf_keyhash[pf_hashkey(sks)];
1024 PF_HASHROW_LOCK(kh);
1025 if (s->key[PF_SK_STACK] != NULL)
1026 pf_state_key_detach(s, PF_SK_STACK);
1028 * If both point to same key, then we are done.
1030 if (sks == s->key[PF_SK_WIRE]) {
1031 pf_state_key_detach(s, PF_SK_WIRE);
1032 PF_HASHROW_UNLOCK(kh);
1035 PF_HASHROW_UNLOCK(kh);
1038 if (s->key[PF_SK_WIRE] != NULL) {
1039 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1040 PF_HASHROW_LOCK(kh);
1041 if (s->key[PF_SK_WIRE] != NULL)
1042 pf_state_key_detach(s, PF_SK_WIRE);
1043 PF_HASHROW_UNLOCK(kh);
1048 pf_state_key_detach(struct pf_state *s, int idx)
1050 struct pf_state_key *sk = s->key[idx];
1052 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1054 PF_HASHROW_ASSERT(kh);
1056 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1059 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1060 LIST_REMOVE(sk, entry);
1061 uma_zfree(V_pf_state_key_z, sk);
1066 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1068 struct pf_state_key *sk = mem;
1070 bzero(sk, sizeof(struct pf_state_key_cmp));
1071 TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1072 TAILQ_INIT(&sk->states[PF_SK_STACK]);
1077 struct pf_state_key *
1078 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1079 struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1081 struct pf_state_key *sk;
1083 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1087 PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1088 PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1089 sk->port[pd->sidx] = sport;
1090 sk->port[pd->didx] = dport;
1091 sk->proto = pd->proto;
1097 struct pf_state_key *
1098 pf_state_key_clone(struct pf_state_key *orig)
1100 struct pf_state_key *sk;
1102 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1106 bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1112 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1113 struct pf_state_key *sks, struct pf_state *s)
1115 struct pf_idhash *ih;
1116 struct pf_state *cur;
1119 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1120 ("%s: sks not pristine", __func__));
1121 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1122 ("%s: skw not pristine", __func__));
1123 KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1127 if (s->id == 0 && s->creatorid == 0) {
1128 /* XXX: should be atomic, but probability of collision low */
1129 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1130 V_pf_stateid[curcpu] = 1;
1131 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1132 s->id = htobe64(s->id);
1133 s->creatorid = V_pf_status.hostid;
1136 /* Returns with ID locked on success. */
1137 if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1140 ih = &V_pf_idhash[PF_IDHASH(s)];
1141 PF_HASHROW_ASSERT(ih);
1142 LIST_FOREACH(cur, &ih->states, entry)
1143 if (cur->id == s->id && cur->creatorid == s->creatorid)
1147 PF_HASHROW_UNLOCK(ih);
1148 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1149 printf("pf: state ID collision: "
1150 "id: %016llx creatorid: %08x\n",
1151 (unsigned long long)be64toh(s->id),
1152 ntohl(s->creatorid));
1157 LIST_INSERT_HEAD(&ih->states, s, entry);
1158 /* One for keys, one for ID hash. */
1159 refcount_init(&s->refs, 2);
1161 V_pf_status.fcounters[FCNT_STATE_INSERT]++;
1162 if (pfsync_insert_state_ptr != NULL)
1163 pfsync_insert_state_ptr(s);
1165 /* Returns locked. */
1170 * Find state by ID: returns with locked row on success.
1173 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1175 struct pf_idhash *ih;
1178 V_pf_status.fcounters[FCNT_STATE_SEARCH]++;
1180 ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))];
1182 PF_HASHROW_LOCK(ih);
1183 LIST_FOREACH(s, &ih->states, entry)
1184 if (s->id == id && s->creatorid == creatorid)
1188 PF_HASHROW_UNLOCK(ih);
1194 * Find state by key.
1195 * Returns with ID hash slot locked on success.
1197 static struct pf_state *
1198 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1200 struct pf_keyhash *kh;
1201 struct pf_state_key *sk;
1205 V_pf_status.fcounters[FCNT_STATE_SEARCH]++;
1207 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1209 PF_HASHROW_LOCK(kh);
1210 LIST_FOREACH(sk, &kh->keys, entry)
1211 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1214 PF_HASHROW_UNLOCK(kh);
1218 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1220 /* List is sorted, if-bound states before floating ones. */
1221 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1222 if (s->kif == V_pfi_all || s->kif == kif) {
1224 PF_HASHROW_UNLOCK(kh);
1225 if (s->timeout == PFTM_UNLINKED) {
1227 * State is being processed
1228 * by pf_unlink_state() in
1236 PF_HASHROW_UNLOCK(kh);
1242 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1244 struct pf_keyhash *kh;
1245 struct pf_state_key *sk;
1246 struct pf_state *s, *ret = NULL;
1249 V_pf_status.fcounters[FCNT_STATE_SEARCH]++;
1251 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1253 PF_HASHROW_LOCK(kh);
1254 LIST_FOREACH(sk, &kh->keys, entry)
1255 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1258 PF_HASHROW_UNLOCK(kh);
1273 panic("%s: dir %u", __func__, dir);
1276 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1278 PF_HASHROW_UNLOCK(kh);
1292 PF_HASHROW_UNLOCK(kh);
1297 /* END state table stuff */
1300 pf_send(struct pf_send_entry *pfse)
1304 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1306 swi_sched(V_pf_swi_cookie, 0);
1312 struct pf_send_head queue;
1313 struct pf_send_entry *pfse, *next;
1315 CURVNET_SET((struct vnet *)v);
1318 queue = V_pf_sendqueue;
1319 STAILQ_INIT(&V_pf_sendqueue);
1322 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1323 switch (pfse->pfse_type) {
1326 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1329 icmp_error(pfse->pfse_m, pfse->pfse_icmp_type,
1330 pfse->pfse_icmp_code, 0, pfse->pfse_icmp_mtu);
1335 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1339 icmp6_error(pfse->pfse_m, pfse->pfse_icmp_type,
1340 pfse->pfse_icmp_code, pfse->pfse_icmp_mtu);
1344 panic("%s: unknown type", __func__);
1346 free(pfse, M_PFTEMP);
1352 pf_purge_thread(void *v)
1356 CURVNET_SET((struct vnet *)v);
1360 rw_sleep(pf_purge_thread, &pf_rules_lock, 0, "pftm", hz / 10);
1362 if (V_pf_end_threads) {
1364 * To cleanse up all kifs and rules we need
1365 * two runs: first one clears reference flags,
1366 * then pf_purge_expired_states() doesn't
1367 * raise them, and then second run frees.
1370 pf_purge_unlinked_rules();
1374 * Now purge everything.
1376 pf_purge_expired_states(0, V_pf_hashmask);
1377 pf_purge_expired_fragments();
1378 pf_purge_expired_src_nodes();
1381 * Now all kifs & rules should be unreferenced,
1382 * thus should be successfully freed.
1384 pf_purge_unlinked_rules();
1388 * Announce success and exit.
1393 wakeup(pf_purge_thread);
1398 /* Process 1/interval fraction of the state table every run. */
1399 idx = pf_purge_expired_states(idx, V_pf_hashmask /
1400 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1402 /* Purge other expired types every PFTM_INTERVAL seconds. */
1405 * Order is important:
1406 * - states and src nodes reference rules
1407 * - states and rules reference kifs
1409 pf_purge_expired_fragments();
1410 pf_purge_expired_src_nodes();
1411 pf_purge_unlinked_rules();
1420 pf_state_expires(const struct pf_state *state)
1427 /* handle all PFTM_* > PFTM_MAX here */
1428 if (state->timeout == PFTM_PURGE)
1429 return (time_uptime);
1430 if (state->timeout == PFTM_UNTIL_PACKET)
1432 KASSERT(state->timeout != PFTM_UNLINKED,
1433 ("pf_state_expires: timeout == PFTM_UNLINKED"));
1434 KASSERT((state->timeout < PFTM_MAX),
1435 ("pf_state_expires: timeout > PFTM_MAX"));
1436 timeout = state->rule.ptr->timeout[state->timeout];
1438 timeout = V_pf_default_rule.timeout[state->timeout];
1439 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1441 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1442 states = state->rule.ptr->states_cur; /* XXXGL */
1444 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1445 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1446 states = V_pf_status.states;
1448 if (end && states > start && start < end) {
1450 return (state->expire + timeout * (end - states) /
1453 return (time_uptime);
1455 return (state->expire + timeout);
1459 pf_purge_expired_src_nodes()
1461 struct pf_srchash *sh;
1462 struct pf_src_node *cur, *next;
1465 for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
1466 PF_HASHROW_LOCK(sh);
1467 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1468 if (cur->states <= 0 && cur->expire <= time_uptime) {
1469 if (cur->rule.ptr != NULL)
1470 cur->rule.ptr->src_nodes--;
1471 LIST_REMOVE(cur, entry);
1472 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS]++;
1473 V_pf_status.src_nodes--;
1474 uma_zfree(V_pf_sources_z, cur);
1475 } else if (cur->rule.ptr != NULL)
1476 cur->rule.ptr->rule_flag |= PFRULE_REFS;
1477 PF_HASHROW_UNLOCK(sh);
1482 pf_src_tree_remove_state(struct pf_state *s)
1486 if (s->src_node != NULL) {
1488 --s->src_node->conn;
1489 if (--s->src_node->states <= 0) {
1490 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1493 V_pf_default_rule.timeout[PFTM_SRC_NODE];
1494 s->src_node->expire = time_uptime + timeout;
1497 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1498 if (--s->nat_src_node->states <= 0) {
1499 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1502 V_pf_default_rule.timeout[PFTM_SRC_NODE];
1503 s->nat_src_node->expire = time_uptime + timeout;
1506 s->src_node = s->nat_src_node = NULL;
1510 * Unlink and potentilly free a state. Function may be
1511 * called with ID hash row locked, but always returns
1512 * unlocked, since it needs to go through key hash locking.
1515 pf_unlink_state(struct pf_state *s, u_int flags)
1517 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1519 if ((flags & PF_ENTER_LOCKED) == 0)
1520 PF_HASHROW_LOCK(ih);
1522 PF_HASHROW_ASSERT(ih);
1524 if (s->timeout == PFTM_UNLINKED) {
1526 * State is being processed
1527 * by pf_unlink_state() in
1530 PF_HASHROW_UNLOCK(ih);
1531 return (0); /* XXXGL: undefined actually */
1534 if (s->src.state == PF_TCPS_PROXY_DST) {
1535 /* XXX wire key the right one? */
1536 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1537 &s->key[PF_SK_WIRE]->addr[1],
1538 &s->key[PF_SK_WIRE]->addr[0],
1539 s->key[PF_SK_WIRE]->port[1],
1540 s->key[PF_SK_WIRE]->port[0],
1541 s->src.seqhi, s->src.seqlo + 1,
1542 TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1545 LIST_REMOVE(s, entry);
1546 pf_src_tree_remove_state(s);
1548 if (pfsync_delete_state_ptr != NULL)
1549 pfsync_delete_state_ptr(s);
1551 --s->rule.ptr->states_cur;
1552 if (s->nat_rule.ptr != NULL)
1553 --s->nat_rule.ptr->states_cur;
1554 if (s->anchor.ptr != NULL)
1555 --s->anchor.ptr->states_cur;
1557 s->timeout = PFTM_UNLINKED;
1559 PF_HASHROW_UNLOCK(ih);
1562 refcount_release(&s->refs);
1564 return (pf_release_state(s));
1568 pf_free_state(struct pf_state *cur)
1571 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1572 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1575 pf_normalize_tcp_cleanup(cur);
1576 uma_zfree(V_pf_state_z, cur);
1577 V_pf_status.fcounters[FCNT_STATE_REMOVALS]++;
1581 * Called only from pf_purge_thread(), thus serialized.
1584 pf_purge_expired_states(u_int i, int maxcheck)
1586 struct pf_idhash *ih;
1589 V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1592 * Go through hash and unlink states that expire now.
1594 while (maxcheck > 0) {
1596 ih = &V_pf_idhash[i];
1598 PF_HASHROW_LOCK(ih);
1599 LIST_FOREACH(s, &ih->states, entry) {
1600 if (pf_state_expires(s) <= time_uptime) {
1601 V_pf_status.states -=
1602 pf_unlink_state(s, PF_ENTER_LOCKED);
1605 s->rule.ptr->rule_flag |= PFRULE_REFS;
1606 if (s->nat_rule.ptr != NULL)
1607 s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1608 if (s->anchor.ptr != NULL)
1609 s->anchor.ptr->rule_flag |= PFRULE_REFS;
1610 s->kif->pfik_flags |= PFI_IFLAG_REFS;
1612 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1614 PF_HASHROW_UNLOCK(ih);
1616 /* Return when we hit end of hash. */
1617 if (++i > V_pf_hashmask) {
1618 V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1625 V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1631 pf_purge_unlinked_rules()
1633 struct pf_rulequeue tmpq;
1634 struct pf_rule *r, *r1;
1637 * If we have overloading task pending, then we'd
1638 * better skip purging this time. There is a tiny
1639 * probability that overloading task references
1640 * an already unlinked rule.
1642 PF_OVERLOADQ_LOCK();
1643 if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1644 PF_OVERLOADQ_UNLOCK();
1647 PF_OVERLOADQ_UNLOCK();
1650 * Do naive mark-and-sweep garbage collecting of old rules.
1651 * Reference flag is raised by pf_purge_expired_states()
1652 * and pf_purge_expired_src_nodes().
1654 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1655 * use a temporary queue.
1658 PF_UNLNKDRULES_LOCK();
1659 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1660 if (!(r->rule_flag & PFRULE_REFS)) {
1661 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1662 TAILQ_INSERT_TAIL(&tmpq, r, entries);
1664 r->rule_flag &= ~PFRULE_REFS;
1666 PF_UNLNKDRULES_UNLOCK();
1668 if (!TAILQ_EMPTY(&tmpq)) {
1670 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1671 TAILQ_REMOVE(&tmpq, r, entries);
1679 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1684 u_int32_t a = ntohl(addr->addr32[0]);
1685 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1697 u_int8_t i, curstart, curend, maxstart, maxend;
1698 curstart = curend = maxstart = maxend = 255;
1699 for (i = 0; i < 8; i++) {
1700 if (!addr->addr16[i]) {
1701 if (curstart == 255)
1705 if ((curend - curstart) >
1706 (maxend - maxstart)) {
1707 maxstart = curstart;
1710 curstart = curend = 255;
1713 if ((curend - curstart) >
1714 (maxend - maxstart)) {
1715 maxstart = curstart;
1718 for (i = 0; i < 8; i++) {
1719 if (i >= maxstart && i <= maxend) {
1725 b = ntohs(addr->addr16[i]);
1742 pf_print_state(struct pf_state *s)
1744 pf_print_state_parts(s, NULL, NULL);
1748 pf_print_state_parts(struct pf_state *s,
1749 struct pf_state_key *skwp, struct pf_state_key *sksp)
1751 struct pf_state_key *skw, *sks;
1752 u_int8_t proto, dir;
1754 /* Do our best to fill these, but they're skipped if NULL */
1755 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1756 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1757 proto = skw ? skw->proto : (sks ? sks->proto : 0);
1758 dir = s ? s->direction : 0;
1776 case IPPROTO_ICMPV6:
1780 printf("%u", skw->proto);
1793 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1795 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1800 pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1802 pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1807 if (proto == IPPROTO_TCP) {
1808 printf(" [lo=%u high=%u win=%u modulator=%u",
1809 s->src.seqlo, s->src.seqhi,
1810 s->src.max_win, s->src.seqdiff);
1811 if (s->src.wscale && s->dst.wscale)
1812 printf(" wscale=%u",
1813 s->src.wscale & PF_WSCALE_MASK);
1815 printf(" [lo=%u high=%u win=%u modulator=%u",
1816 s->dst.seqlo, s->dst.seqhi,
1817 s->dst.max_win, s->dst.seqdiff);
1818 if (s->src.wscale && s->dst.wscale)
1819 printf(" wscale=%u",
1820 s->dst.wscale & PF_WSCALE_MASK);
1823 printf(" %u:%u", s->src.state, s->dst.state);
1828 pf_print_flags(u_int8_t f)
1850 #define PF_SET_SKIP_STEPS(i) \
1852 while (head[i] != cur) { \
1853 head[i]->skip[i].ptr = cur; \
1854 head[i] = TAILQ_NEXT(head[i], entries); \
1859 pf_calc_skip_steps(struct pf_rulequeue *rules)
1861 struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1864 cur = TAILQ_FIRST(rules);
1866 for (i = 0; i < PF_SKIP_COUNT; ++i)
1868 while (cur != NULL) {
1870 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1871 PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1872 if (cur->direction != prev->direction)
1873 PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1874 if (cur->af != prev->af)
1875 PF_SET_SKIP_STEPS(PF_SKIP_AF);
1876 if (cur->proto != prev->proto)
1877 PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
1878 if (cur->src.neg != prev->src.neg ||
1879 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
1880 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
1881 if (cur->src.port[0] != prev->src.port[0] ||
1882 cur->src.port[1] != prev->src.port[1] ||
1883 cur->src.port_op != prev->src.port_op)
1884 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
1885 if (cur->dst.neg != prev->dst.neg ||
1886 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
1887 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
1888 if (cur->dst.port[0] != prev->dst.port[0] ||
1889 cur->dst.port[1] != prev->dst.port[1] ||
1890 cur->dst.port_op != prev->dst.port_op)
1891 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
1894 cur = TAILQ_NEXT(cur, entries);
1896 for (i = 0; i < PF_SKIP_COUNT; ++i)
1897 PF_SET_SKIP_STEPS(i);
1901 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
1903 if (aw1->type != aw2->type)
1905 switch (aw1->type) {
1906 case PF_ADDR_ADDRMASK:
1908 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0))
1910 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0))
1913 case PF_ADDR_DYNIFTL:
1914 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
1915 case PF_ADDR_NOROUTE:
1916 case PF_ADDR_URPFFAILED:
1919 return (aw1->p.tbl != aw2->p.tbl);
1921 printf("invalid address type: %d\n", aw1->type);
1927 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
1933 l = cksum + old - new;
1934 l = (l >> 16) + (l & 65535);
1942 pf_change_ap(struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc,
1943 struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af)
1948 PF_ACPY(&ao, a, af);
1956 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
1957 ao.addr16[0], an->addr16[0], 0),
1958 ao.addr16[1], an->addr16[1], 0);
1960 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
1961 ao.addr16[0], an->addr16[0], u),
1962 ao.addr16[1], an->addr16[1], u),
1968 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1969 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1970 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
1971 ao.addr16[0], an->addr16[0], u),
1972 ao.addr16[1], an->addr16[1], u),
1973 ao.addr16[2], an->addr16[2], u),
1974 ao.addr16[3], an->addr16[3], u),
1975 ao.addr16[4], an->addr16[4], u),
1976 ao.addr16[5], an->addr16[5], u),
1977 ao.addr16[6], an->addr16[6], u),
1978 ao.addr16[7], an->addr16[7], u),
1986 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */
1988 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
1992 memcpy(&ao, a, sizeof(ao));
1993 memcpy(a, &an, sizeof(u_int32_t));
1994 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
1995 ao % 65536, an % 65536, u);
2000 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2004 PF_ACPY(&ao, a, AF_INET6);
2005 PF_ACPY(a, an, AF_INET6);
2007 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2008 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2009 pf_cksum_fixup(pf_cksum_fixup(*c,
2010 ao.addr16[0], an->addr16[0], u),
2011 ao.addr16[1], an->addr16[1], u),
2012 ao.addr16[2], an->addr16[2], u),
2013 ao.addr16[3], an->addr16[3], u),
2014 ao.addr16[4], an->addr16[4], u),
2015 ao.addr16[5], an->addr16[5], u),
2016 ao.addr16[6], an->addr16[6], u),
2017 ao.addr16[7], an->addr16[7], u);
2022 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2023 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2024 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2026 struct pf_addr oia, ooa;
2028 PF_ACPY(&oia, ia, af);
2030 PF_ACPY(&ooa, oa, af);
2032 /* Change inner protocol port, fix inner protocol checksum. */
2034 u_int16_t oip = *ip;
2041 *pc = pf_cksum_fixup(*pc, oip, *ip, u);
2042 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2044 *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2046 /* Change inner ip address, fix inner ip and icmp checksums. */
2047 PF_ACPY(ia, na, af);
2051 u_int32_t oh2c = *h2c;
2053 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2054 oia.addr16[0], ia->addr16[0], 0),
2055 oia.addr16[1], ia->addr16[1], 0);
2056 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2057 oia.addr16[0], ia->addr16[0], 0),
2058 oia.addr16[1], ia->addr16[1], 0);
2059 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2065 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2066 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2067 pf_cksum_fixup(pf_cksum_fixup(*ic,
2068 oia.addr16[0], ia->addr16[0], u),
2069 oia.addr16[1], ia->addr16[1], u),
2070 oia.addr16[2], ia->addr16[2], u),
2071 oia.addr16[3], ia->addr16[3], u),
2072 oia.addr16[4], ia->addr16[4], u),
2073 oia.addr16[5], ia->addr16[5], u),
2074 oia.addr16[6], ia->addr16[6], u),
2075 oia.addr16[7], ia->addr16[7], u);
2079 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2081 PF_ACPY(oa, na, af);
2085 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2086 ooa.addr16[0], oa->addr16[0], 0),
2087 ooa.addr16[1], oa->addr16[1], 0);
2092 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2093 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2094 pf_cksum_fixup(pf_cksum_fixup(*ic,
2095 ooa.addr16[0], oa->addr16[0], u),
2096 ooa.addr16[1], oa->addr16[1], u),
2097 ooa.addr16[2], oa->addr16[2], u),
2098 ooa.addr16[3], oa->addr16[3], u),
2099 ooa.addr16[4], oa->addr16[4], u),
2100 ooa.addr16[5], oa->addr16[5], u),
2101 ooa.addr16[6], oa->addr16[6], u),
2102 ooa.addr16[7], oa->addr16[7], u);
2111 * Need to modulate the sequence numbers in the TCP SACK option
2112 * (credits to Krzysztof Pfaff for report and patch)
2115 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2116 struct tcphdr *th, struct pf_state_peer *dst)
2118 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2119 u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2120 int copyback = 0, i, olen;
2121 struct sackblk sack;
2123 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2124 if (hlen < TCPOLEN_SACKLEN ||
2125 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2128 while (hlen >= TCPOLEN_SACKLEN) {
2131 case TCPOPT_EOL: /* FALLTHROUGH */
2139 if (olen >= TCPOLEN_SACKLEN) {
2140 for (i = 2; i + TCPOLEN_SACK <= olen;
2141 i += TCPOLEN_SACK) {
2142 memcpy(&sack, &opt[i], sizeof(sack));
2143 pf_change_a(&sack.start, &th->th_sum,
2144 htonl(ntohl(sack.start) -
2146 pf_change_a(&sack.end, &th->th_sum,
2147 htonl(ntohl(sack.end) -
2149 memcpy(&opt[i], &sack, sizeof(sack));
2163 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2168 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2169 const struct pf_addr *saddr, const struct pf_addr *daddr,
2170 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2171 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2172 u_int16_t rtag, struct ifnet *ifp)
2174 struct pf_send_entry *pfse;
2178 struct ip *h = NULL;
2181 struct ip6_hdr *h6 = NULL;
2185 struct pf_mtag *pf_mtag;
2190 /* maximum segment size tcp option */
2191 tlen = sizeof(struct tcphdr);
2198 len = sizeof(struct ip) + tlen;
2203 len = sizeof(struct ip6_hdr) + tlen;
2207 panic("%s: unsupported af %d", __func__, af);
2210 /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2211 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2214 m = m_gethdr(M_NOWAIT, MT_DATA);
2216 free(pfse, M_PFTEMP);
2220 mac_netinet_firewall_send(m);
2222 if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2223 free(pfse, M_PFTEMP);
2228 m->m_flags |= M_SKIP_FIREWALL;
2229 pf_mtag->tag = rtag;
2231 if (r != NULL && r->rtableid >= 0)
2232 M_SETFIB(m, r->rtableid);
2235 if (r != NULL && r->qid) {
2236 pf_mtag->qid = r->qid;
2238 /* add hints for ecn */
2239 pf_mtag->hdr = mtod(m, struct ip *);
2242 m->m_data += max_linkhdr;
2243 m->m_pkthdr.len = m->m_len = len;
2244 m->m_pkthdr.rcvif = NULL;
2245 bzero(m->m_data, len);
2249 h = mtod(m, struct ip *);
2251 /* IP header fields included in the TCP checksum */
2252 h->ip_p = IPPROTO_TCP;
2253 h->ip_len = htons(tlen);
2254 h->ip_src.s_addr = saddr->v4.s_addr;
2255 h->ip_dst.s_addr = daddr->v4.s_addr;
2257 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2262 h6 = mtod(m, struct ip6_hdr *);
2264 /* IP header fields included in the TCP checksum */
2265 h6->ip6_nxt = IPPROTO_TCP;
2266 h6->ip6_plen = htons(tlen);
2267 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2268 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2270 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2276 th->th_sport = sport;
2277 th->th_dport = dport;
2278 th->th_seq = htonl(seq);
2279 th->th_ack = htonl(ack);
2280 th->th_off = tlen >> 2;
2281 th->th_flags = flags;
2282 th->th_win = htons(win);
2285 opt = (char *)(th + 1);
2286 opt[0] = TCPOPT_MAXSEG;
2289 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2296 th->th_sum = in_cksum(m, len);
2298 /* Finish the IP header */
2300 h->ip_hl = sizeof(*h) >> 2;
2301 h->ip_tos = IPTOS_LOWDELAY;
2302 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2303 h->ip_len = htons(len);
2304 h->ip_ttl = ttl ? ttl : V_ip_defttl;
2307 pfse->pfse_type = PFSE_IP;
2313 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2314 sizeof(struct ip6_hdr), tlen);
2316 h6->ip6_vfc |= IPV6_VERSION;
2317 h6->ip6_hlim = IPV6_DEFHLIM;
2319 pfse->pfse_type = PFSE_IP6;
2328 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2331 struct pf_send_entry *pfse;
2333 struct pf_mtag *pf_mtag;
2335 /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2336 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2340 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2341 free(pfse, M_PFTEMP);
2345 if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2346 free(pfse, M_PFTEMP);
2350 m0->m_flags |= M_SKIP_FIREWALL;
2352 if (r->rtableid >= 0)
2353 M_SETFIB(m0, r->rtableid);
2357 pf_mtag->qid = r->qid;
2358 /* add hints for ecn */
2359 pf_mtag->hdr = mtod(m0, struct ip *);
2366 pfse->pfse_type = PFSE_ICMP;
2371 pfse->pfse_type = PFSE_ICMP6;
2376 pfse->pfse_icmp_type = type;
2377 pfse->pfse_icmp_code = code;
2382 * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2383 * If n is 0, they match if they are equal. If n is != 0, they match if they
2387 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2388 struct pf_addr *b, sa_family_t af)
2395 if ((a->addr32[0] & m->addr32[0]) ==
2396 (b->addr32[0] & m->addr32[0]))
2402 if (((a->addr32[0] & m->addr32[0]) ==
2403 (b->addr32[0] & m->addr32[0])) &&
2404 ((a->addr32[1] & m->addr32[1]) ==
2405 (b->addr32[1] & m->addr32[1])) &&
2406 ((a->addr32[2] & m->addr32[2]) ==
2407 (b->addr32[2] & m->addr32[2])) &&
2408 ((a->addr32[3] & m->addr32[3]) ==
2409 (b->addr32[3] & m->addr32[3])))
2428 * Return 1 if b <= a <= e, otherwise return 0.
2431 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2432 struct pf_addr *a, sa_family_t af)
2437 if ((a->addr32[0] < b->addr32[0]) ||
2438 (a->addr32[0] > e->addr32[0]))
2447 for (i = 0; i < 4; ++i)
2448 if (a->addr32[i] > b->addr32[i])
2450 else if (a->addr32[i] < b->addr32[i])
2453 for (i = 0; i < 4; ++i)
2454 if (a->addr32[i] < e->addr32[i])
2456 else if (a->addr32[i] > e->addr32[i])
2466 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2470 return ((p > a1) && (p < a2));
2472 return ((p < a1) || (p > a2));
2474 return ((p >= a1) && (p <= a2));
2488 return (0); /* never reached */
2492 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2497 return (pf_match(op, a1, a2, p));
2501 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2503 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2505 return (pf_match(op, a1, a2, u));
2509 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2511 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2513 return (pf_match(op, a1, a2, g));
2517 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2522 return ((!r->match_tag_not && r->match_tag == *tag) ||
2523 (r->match_tag_not && r->match_tag != *tag));
2527 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2530 KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2532 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2535 pd->pf_mtag->tag = tag;
2540 #define PF_ANCHOR_STACKSIZE 32
2541 struct pf_anchor_stackframe {
2542 struct pf_ruleset *rs;
2543 struct pf_rule *r; /* XXX: + match bit */
2544 struct pf_anchor *child;
2548 * XXX: We rely on malloc(9) returning pointer aligned addresses.
2550 #define PF_ANCHORSTACK_MATCH 0x00000001
2551 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH)
2553 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2554 #define PF_ANCHOR_RULE(f) (struct pf_rule *) \
2555 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2556 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \
2557 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \
2561 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2562 struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2565 struct pf_anchor_stackframe *f;
2571 if (*depth >= PF_ANCHOR_STACKSIZE) {
2572 printf("%s: anchor stack overflow on %s\n",
2573 __func__, (*r)->anchor->name);
2574 *r = TAILQ_NEXT(*r, entries);
2576 } else if (*depth == 0 && a != NULL)
2578 f = stack + (*depth)++;
2581 if ((*r)->anchor_wildcard) {
2582 struct pf_anchor_node *parent = &(*r)->anchor->children;
2584 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2588 *rs = &f->child->ruleset;
2591 *rs = &(*r)->anchor->ruleset;
2593 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2597 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2598 struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2601 struct pf_anchor_stackframe *f;
2610 f = stack + *depth - 1;
2611 fr = PF_ANCHOR_RULE(f);
2612 if (f->child != NULL) {
2613 struct pf_anchor_node *parent;
2616 * This block traverses through
2617 * a wildcard anchor.
2619 parent = &fr->anchor->children;
2620 if (match != NULL && *match) {
2622 * If any of "*" matched, then
2623 * "foo/ *" matched, mark frame
2626 PF_ANCHOR_SET_MATCH(f);
2629 f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2630 if (f->child != NULL) {
2631 *rs = &f->child->ruleset;
2632 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2640 if (*depth == 0 && a != NULL)
2643 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2645 *r = TAILQ_NEXT(fr, entries);
2646 } while (*r == NULL);
2653 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2654 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2659 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2660 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2664 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2665 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2666 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2667 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2668 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2669 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2670 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2671 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2677 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2682 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2686 if (addr->addr32[3] == 0xffffffff) {
2687 addr->addr32[3] = 0;
2688 if (addr->addr32[2] == 0xffffffff) {
2689 addr->addr32[2] = 0;
2690 if (addr->addr32[1] == 0xffffffff) {
2691 addr->addr32[1] = 0;
2693 htonl(ntohl(addr->addr32[0]) + 1);
2696 htonl(ntohl(addr->addr32[1]) + 1);
2699 htonl(ntohl(addr->addr32[2]) + 1);
2702 htonl(ntohl(addr->addr32[3]) + 1);
2709 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
2711 struct pf_addr *saddr, *daddr;
2712 u_int16_t sport, dport;
2713 struct inpcbinfo *pi;
2716 pd->lookup.uid = UID_MAX;
2717 pd->lookup.gid = GID_MAX;
2719 switch (pd->proto) {
2721 if (pd->hdr.tcp == NULL)
2723 sport = pd->hdr.tcp->th_sport;
2724 dport = pd->hdr.tcp->th_dport;
2728 if (pd->hdr.udp == NULL)
2730 sport = pd->hdr.udp->uh_sport;
2731 dport = pd->hdr.udp->uh_dport;
2737 if (direction == PF_IN) {
2752 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
2753 dport, INPLOOKUP_RLOCKPCB, NULL, m);
2755 inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
2756 daddr->v4, dport, INPLOOKUP_WILDCARD |
2757 INPLOOKUP_RLOCKPCB, NULL, m);
2765 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
2766 dport, INPLOOKUP_RLOCKPCB, NULL, m);
2768 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
2769 &daddr->v6, dport, INPLOOKUP_WILDCARD |
2770 INPLOOKUP_RLOCKPCB, NULL, m);
2780 INP_RLOCK_ASSERT(inp);
2781 pd->lookup.uid = inp->inp_cred->cr_uid;
2782 pd->lookup.gid = inp->inp_cred->cr_groups[0];
2789 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2793 u_int8_t *opt, optlen;
2794 u_int8_t wscale = 0;
2796 hlen = th_off << 2; /* hlen <= sizeof(hdr) */
2797 if (hlen <= sizeof(struct tcphdr))
2799 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2801 opt = hdr + sizeof(struct tcphdr);
2802 hlen -= sizeof(struct tcphdr);
2812 if (wscale > TCP_MAX_WINSHIFT)
2813 wscale = TCP_MAX_WINSHIFT;
2814 wscale |= PF_WSCALE_FLAG;
2829 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2833 u_int8_t *opt, optlen;
2834 u_int16_t mss = V_tcp_mssdflt;
2836 hlen = th_off << 2; /* hlen <= sizeof(hdr) */
2837 if (hlen <= sizeof(struct tcphdr))
2839 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2841 opt = hdr + sizeof(struct tcphdr);
2842 hlen -= sizeof(struct tcphdr);
2843 while (hlen >= TCPOLEN_MAXSEG) {
2851 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
2867 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
2870 struct sockaddr_in *dst;
2874 struct sockaddr_in6 *dst6;
2875 struct route_in6 ro6;
2877 struct rtentry *rt = NULL;
2879 u_int16_t mss = V_tcp_mssdflt;
2884 hlen = sizeof(struct ip);
2885 bzero(&ro, sizeof(ro));
2886 dst = (struct sockaddr_in *)&ro.ro_dst;
2887 dst->sin_family = AF_INET;
2888 dst->sin_len = sizeof(*dst);
2889 dst->sin_addr = addr->v4;
2890 in_rtalloc_ign(&ro, 0, rtableid);
2896 hlen = sizeof(struct ip6_hdr);
2897 bzero(&ro6, sizeof(ro6));
2898 dst6 = (struct sockaddr_in6 *)&ro6.ro_dst;
2899 dst6->sin6_family = AF_INET6;
2900 dst6->sin6_len = sizeof(*dst6);
2901 dst6->sin6_addr = addr->v6;
2902 in6_rtalloc_ign(&ro6, 0, rtableid);
2908 if (rt && rt->rt_ifp) {
2909 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr);
2910 mss = max(V_tcp_mssdflt, mss);
2913 mss = min(mss, offer);
2914 mss = max(mss, 64); /* sanity - at least max opt space */
2919 pf_set_rt_ifp(struct pf_state *s, struct pf_addr *saddr)
2921 struct pf_rule *r = s->rule.ptr;
2922 struct pf_src_node *sn = NULL;
2925 if (!r->rt || r->rt == PF_FASTROUTE)
2927 switch (s->key[PF_SK_WIRE]->af) {
2930 pf_map_addr(AF_INET, r, saddr, &s->rt_addr, NULL, &sn);
2931 s->rt_kif = r->rpool.cur->kif;
2936 pf_map_addr(AF_INET6, r, saddr, &s->rt_addr, NULL, &sn);
2937 s->rt_kif = r->rpool.cur->kif;
2944 pf_tcp_iss(struct pf_pdesc *pd)
2947 u_int32_t digest[4];
2949 if (V_pf_tcp_secret_init == 0) {
2950 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
2951 MD5Init(&V_pf_tcp_secret_ctx);
2952 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
2953 sizeof(V_pf_tcp_secret));
2954 V_pf_tcp_secret_init = 1;
2957 ctx = V_pf_tcp_secret_ctx;
2959 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
2960 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
2961 if (pd->af == AF_INET6) {
2962 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
2963 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
2965 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
2966 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
2968 MD5Final((u_char *)digest, &ctx);
2969 V_pf_tcp_iss_off += 4096;
2970 #define ISN_RANDOM_INCREMENT (4096 - 1)
2971 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
2973 #undef ISN_RANDOM_INCREMENT
2977 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
2978 struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
2979 struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
2981 struct pf_rule *nr = NULL;
2982 struct pf_addr * const saddr = pd->src;
2983 struct pf_addr * const daddr = pd->dst;
2984 sa_family_t af = pd->af;
2985 struct pf_rule *r, *a = NULL;
2986 struct pf_ruleset *ruleset = NULL;
2987 struct pf_src_node *nsn = NULL;
2988 struct tcphdr *th = pd->hdr.tcp;
2989 struct pf_state_key *sk = NULL, *nk = NULL;
2991 int rewrite = 0, hdrlen = 0;
2992 int tag = -1, rtableid = -1;
2996 u_int16_t sport = 0, dport = 0;
2997 u_int16_t bproto_sum = 0, bip_sum = 0;
2998 u_int8_t icmptype = 0, icmpcode = 0;
2999 struct pf_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE];
3004 INP_LOCK_ASSERT(inp);
3005 pd->lookup.uid = inp->inp_cred->cr_uid;
3006 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3007 pd->lookup.done = 1;
3010 switch (pd->proto) {
3012 sport = th->th_sport;
3013 dport = th->th_dport;
3014 hdrlen = sizeof(*th);
3017 sport = pd->hdr.udp->uh_sport;
3018 dport = pd->hdr.udp->uh_dport;
3019 hdrlen = sizeof(*pd->hdr.udp);
3023 if (pd->af != AF_INET)
3025 sport = dport = pd->hdr.icmp->icmp_id;
3026 hdrlen = sizeof(*pd->hdr.icmp);
3027 icmptype = pd->hdr.icmp->icmp_type;
3028 icmpcode = pd->hdr.icmp->icmp_code;
3030 if (icmptype == ICMP_UNREACH ||
3031 icmptype == ICMP_SOURCEQUENCH ||
3032 icmptype == ICMP_REDIRECT ||
3033 icmptype == ICMP_TIMXCEED ||
3034 icmptype == ICMP_PARAMPROB)
3039 case IPPROTO_ICMPV6:
3042 sport = dport = pd->hdr.icmp6->icmp6_id;
3043 hdrlen = sizeof(*pd->hdr.icmp6);
3044 icmptype = pd->hdr.icmp6->icmp6_type;
3045 icmpcode = pd->hdr.icmp6->icmp6_code;
3047 if (icmptype == ICMP6_DST_UNREACH ||
3048 icmptype == ICMP6_PACKET_TOO_BIG ||
3049 icmptype == ICMP6_TIME_EXCEEDED ||
3050 icmptype == ICMP6_PARAM_PROB)
3055 sport = dport = hdrlen = 0;
3059 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3061 /* check packet for BINAT/NAT/RDR */
3062 if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3063 &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3064 KASSERT(sk != NULL, ("%s: null sk", __func__));
3065 KASSERT(nk != NULL, ("%s: null nk", __func__));
3068 bip_sum = *pd->ip_sum;
3070 switch (pd->proto) {
3072 bproto_sum = th->th_sum;
3073 pd->proto_sum = &th->th_sum;
3075 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3076 nk->port[pd->sidx] != sport) {
3077 pf_change_ap(saddr, &th->th_sport, pd->ip_sum,
3078 &th->th_sum, &nk->addr[pd->sidx],
3079 nk->port[pd->sidx], 0, af);
3080 pd->sport = &th->th_sport;
3081 sport = th->th_sport;
3084 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3085 nk->port[pd->didx] != dport) {
3086 pf_change_ap(daddr, &th->th_dport, pd->ip_sum,
3087 &th->th_sum, &nk->addr[pd->didx],
3088 nk->port[pd->didx], 0, af);
3089 dport = th->th_dport;
3090 pd->dport = &th->th_dport;
3095 bproto_sum = pd->hdr.udp->uh_sum;
3096 pd->proto_sum = &pd->hdr.udp->uh_sum;
3098 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3099 nk->port[pd->sidx] != sport) {
3100 pf_change_ap(saddr, &pd->hdr.udp->uh_sport,
3101 pd->ip_sum, &pd->hdr.udp->uh_sum,
3102 &nk->addr[pd->sidx],
3103 nk->port[pd->sidx], 1, af);
3104 sport = pd->hdr.udp->uh_sport;
3105 pd->sport = &pd->hdr.udp->uh_sport;
3108 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3109 nk->port[pd->didx] != dport) {
3110 pf_change_ap(daddr, &pd->hdr.udp->uh_dport,
3111 pd->ip_sum, &pd->hdr.udp->uh_sum,
3112 &nk->addr[pd->didx],
3113 nk->port[pd->didx], 1, af);
3114 dport = pd->hdr.udp->uh_dport;
3115 pd->dport = &pd->hdr.udp->uh_dport;
3121 nk->port[0] = nk->port[1];
3122 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3123 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3124 nk->addr[pd->sidx].v4.s_addr, 0);
3126 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3127 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3128 nk->addr[pd->didx].v4.s_addr, 0);
3130 if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3131 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3132 pd->hdr.icmp->icmp_cksum, sport,
3134 pd->hdr.icmp->icmp_id = nk->port[1];
3135 pd->sport = &pd->hdr.icmp->icmp_id;
3137 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3141 case IPPROTO_ICMPV6:
3142 nk->port[0] = nk->port[1];
3143 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3144 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3145 &nk->addr[pd->sidx], 0);
3147 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3148 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3149 &nk->addr[pd->didx], 0);
3158 &nk->addr[pd->sidx], AF_INET))
3159 pf_change_a(&saddr->v4.s_addr,
3161 nk->addr[pd->sidx].v4.s_addr, 0);
3164 &nk->addr[pd->didx], AF_INET))
3165 pf_change_a(&daddr->v4.s_addr,
3167 nk->addr[pd->didx].v4.s_addr, 0);
3173 &nk->addr[pd->sidx], AF_INET6))
3174 PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3177 &nk->addr[pd->didx], AF_INET6))
3178 PF_ACPY(saddr, &nk->addr[pd->didx], af);
3191 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3192 r = r->skip[PF_SKIP_IFP].ptr;
3193 else if (r->direction && r->direction != direction)
3194 r = r->skip[PF_SKIP_DIR].ptr;
3195 else if (r->af && r->af != af)
3196 r = r->skip[PF_SKIP_AF].ptr;
3197 else if (r->proto && r->proto != pd->proto)
3198 r = r->skip[PF_SKIP_PROTO].ptr;
3199 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3200 r->src.neg, kif, M_GETFIB(m)))
3201 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3202 /* tcp/udp only. port_op always 0 in other cases */
3203 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3204 r->src.port[0], r->src.port[1], sport))
3205 r = r->skip[PF_SKIP_SRC_PORT].ptr;
3206 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3207 r->dst.neg, NULL, M_GETFIB(m)))
3208 r = r->skip[PF_SKIP_DST_ADDR].ptr;
3209 /* tcp/udp only. port_op always 0 in other cases */
3210 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3211 r->dst.port[0], r->dst.port[1], dport))
3212 r = r->skip[PF_SKIP_DST_PORT].ptr;
3213 /* icmp only. type always 0 in other cases */
3214 else if (r->type && r->type != icmptype + 1)
3215 r = TAILQ_NEXT(r, entries);
3216 /* icmp only. type always 0 in other cases */
3217 else if (r->code && r->code != icmpcode + 1)
3218 r = TAILQ_NEXT(r, entries);
3219 else if (r->tos && !(r->tos == pd->tos))
3220 r = TAILQ_NEXT(r, entries);
3221 else if (r->rule_flag & PFRULE_FRAGMENT)
3222 r = TAILQ_NEXT(r, entries);
3223 else if (pd->proto == IPPROTO_TCP &&
3224 (r->flagset & th->th_flags) != r->flags)
3225 r = TAILQ_NEXT(r, entries);
3226 /* tcp/udp only. uid.op always 0 in other cases */
3227 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3228 pf_socket_lookup(direction, pd, m), 1)) &&
3229 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3231 r = TAILQ_NEXT(r, entries);
3232 /* tcp/udp only. gid.op always 0 in other cases */
3233 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3234 pf_socket_lookup(direction, pd, m), 1)) &&
3235 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3237 r = TAILQ_NEXT(r, entries);
3239 r->prob <= arc4random())
3240 r = TAILQ_NEXT(r, entries);
3241 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3242 pd->pf_mtag ? pd->pf_mtag->tag : 0))
3243 r = TAILQ_NEXT(r, entries);
3244 else if (r->os_fingerprint != PF_OSFP_ANY &&
3245 (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3246 pf_osfp_fingerprint(pd, m, off, th),
3247 r->os_fingerprint)))
3248 r = TAILQ_NEXT(r, entries);
3252 if (r->rtableid >= 0)
3253 rtableid = r->rtableid;
3254 if (r->anchor == NULL) {
3261 r = TAILQ_NEXT(r, entries);
3263 pf_step_into_anchor(anchor_stack, &asd,
3264 &ruleset, PF_RULESET_FILTER, &r, &a,
3267 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3268 &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3275 REASON_SET(&reason, PFRES_MATCH);
3277 if (r->log || (nr != NULL && nr->log)) {
3279 m_copyback(m, off, hdrlen, pd->hdr.any);
3280 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3284 if ((r->action == PF_DROP) &&
3285 ((r->rule_flag & PFRULE_RETURNRST) ||
3286 (r->rule_flag & PFRULE_RETURNICMP) ||
3287 (r->rule_flag & PFRULE_RETURN))) {
3288 /* undo NAT changes, if they have taken place */
3290 PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3291 PF_ACPY(daddr, &sk->addr[pd->didx], af);
3293 *pd->sport = sk->port[pd->sidx];
3295 *pd->dport = sk->port[pd->didx];
3297 *pd->proto_sum = bproto_sum;
3299 *pd->ip_sum = bip_sum;
3300 m_copyback(m, off, hdrlen, pd->hdr.any);
3302 if (pd->proto == IPPROTO_TCP &&
3303 ((r->rule_flag & PFRULE_RETURNRST) ||
3304 (r->rule_flag & PFRULE_RETURN)) &&
3305 !(th->th_flags & TH_RST)) {
3306 u_int32_t ack = ntohl(th->th_seq) + pd->p_len;
3318 h4 = mtod(m, struct ip *);
3319 len = ntohs(h4->ip_len) - off;
3324 h6 = mtod(m, struct ip6_hdr *);
3325 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3330 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3331 REASON_SET(&reason, PFRES_PROTCKSUM);
3333 if (th->th_flags & TH_SYN)
3335 if (th->th_flags & TH_FIN)
3337 pf_send_tcp(m, r, af, pd->dst,
3338 pd->src, th->th_dport, th->th_sport,
3339 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3340 r->return_ttl, 1, 0, kif->pfik_ifp);
3342 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3344 pf_send_icmp(m, r->return_icmp >> 8,
3345 r->return_icmp & 255, af, r);
3346 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3348 pf_send_icmp(m, r->return_icmp6 >> 8,
3349 r->return_icmp6 & 255, af, r);
3352 if (r->action == PF_DROP)
3355 if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3356 REASON_SET(&reason, PFRES_MEMORY);
3360 M_SETFIB(m, rtableid);
3362 if (!state_icmp && (r->keep_state || nr != NULL ||
3363 (pd->flags & PFDESC_TCP_NORM))) {
3365 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3366 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3368 if (action != PF_PASS)
3372 uma_zfree(V_pf_state_key_z, sk);
3374 uma_zfree(V_pf_state_key_z, nk);
3377 /* copy back packet headers if we performed NAT operations */
3379 m_copyback(m, off, hdrlen, pd->hdr.any);
3381 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3382 direction == PF_OUT &&
3383 pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m))
3385 * We want the state created, but we dont
3386 * want to send this in case a partner
3387 * firewall has to know about it to allow
3388 * replies through it.
3396 uma_zfree(V_pf_state_key_z, sk);
3398 uma_zfree(V_pf_state_key_z, nk);
3403 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3404 struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3405 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3406 u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3407 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3409 struct pf_state *s = NULL;
3410 struct pf_src_node *sn = NULL;
3411 struct tcphdr *th = pd->hdr.tcp;
3412 u_int16_t mss = V_tcp_mssdflt;
3415 /* check maximums */
3416 if (r->max_states && (r->states_cur >= r->max_states)) {
3417 V_pf_status.lcounters[LCNT_STATES]++;
3418 REASON_SET(&reason, PFRES_MAXSTATES);
3421 /* src node for filter rule */
3422 if ((r->rule_flag & PFRULE_SRCTRACK ||
3423 r->rpool.opts & PF_POOL_STICKYADDR) &&
3424 pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3425 REASON_SET(&reason, PFRES_SRCLIMIT);
3428 /* src node for translation rule */
3429 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3430 pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3431 REASON_SET(&reason, PFRES_SRCLIMIT);
3434 s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3436 REASON_SET(&reason, PFRES_MEMORY);
3440 s->nat_rule.ptr = nr;
3442 STATE_INC_COUNTERS(s);
3444 s->state_flags |= PFSTATE_ALLOWOPTS;
3445 if (r->rule_flag & PFRULE_STATESLOPPY)
3446 s->state_flags |= PFSTATE_SLOPPY;
3447 s->log = r->log & PF_LOG_ALL;
3448 s->sync_state = PFSYNC_S_NONE;
3450 s->log |= nr->log & PF_LOG_ALL;
3451 switch (pd->proto) {
3453 s->src.seqlo = ntohl(th->th_seq);
3454 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3455 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3456 r->keep_state == PF_STATE_MODULATE) {
3457 /* Generate sequence number modulator */
3458 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3461 pf_change_a(&th->th_seq, &th->th_sum,
3462 htonl(s->src.seqlo + s->src.seqdiff), 0);
3466 if (th->th_flags & TH_SYN) {
3468 s->src.wscale = pf_get_wscale(m, off,
3469 th->th_off, pd->af);
3471 s->src.max_win = MAX(ntohs(th->th_win), 1);
3472 if (s->src.wscale & PF_WSCALE_MASK) {
3473 /* Remove scale factor from initial window */
3474 int win = s->src.max_win;
3475 win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3476 s->src.max_win = (win - 1) >>
3477 (s->src.wscale & PF_WSCALE_MASK);
3479 if (th->th_flags & TH_FIN)
3483 s->src.state = TCPS_SYN_SENT;
3484 s->dst.state = TCPS_CLOSED;
3485 s->timeout = PFTM_TCP_FIRST_PACKET;
3488 s->src.state = PFUDPS_SINGLE;
3489 s->dst.state = PFUDPS_NO_TRAFFIC;
3490 s->timeout = PFTM_UDP_FIRST_PACKET;
3494 case IPPROTO_ICMPV6:
3496 s->timeout = PFTM_ICMP_FIRST_PACKET;
3499 s->src.state = PFOTHERS_SINGLE;
3500 s->dst.state = PFOTHERS_NO_TRAFFIC;
3501 s->timeout = PFTM_OTHER_FIRST_PACKET;
3504 s->creation = time_uptime;
3505 s->expire = time_uptime;
3509 s->src_node->states++;
3512 /* XXX We only modify one side for now. */
3513 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3514 s->nat_src_node = nsn;
3515 s->nat_src_node->states++;
3517 if (pd->proto == IPPROTO_TCP) {
3518 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3519 off, pd, th, &s->src, &s->dst)) {
3520 REASON_SET(&reason, PFRES_MEMORY);
3521 pf_src_tree_remove_state(s);
3522 STATE_DEC_COUNTERS(s);
3523 uma_zfree(V_pf_state_z, s);
3526 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3527 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3528 &s->src, &s->dst, rewrite)) {
3529 /* This really shouldn't happen!!! */
3530 DPFPRINTF(PF_DEBUG_URGENT,
3531 ("pf_normalize_tcp_stateful failed on first pkt"));
3532 pf_normalize_tcp_cleanup(s);
3533 pf_src_tree_remove_state(s);
3534 STATE_DEC_COUNTERS(s);
3535 uma_zfree(V_pf_state_z, s);
3539 s->direction = pd->dir;
3542 * sk/nk could already been setup by pf_get_translation().
3545 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3546 __func__, nr, sk, nk));
3547 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3552 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3553 __func__, nr, sk, nk));
3555 /* Swap sk/nk for PF_OUT. */
3556 if (pf_state_insert(BOUND_IFACE(r, kif),
3557 (pd->dir == PF_IN) ? sk : nk,
3558 (pd->dir == PF_IN) ? nk : sk, s)) {
3559 if (pd->proto == IPPROTO_TCP)
3560 pf_normalize_tcp_cleanup(s);
3561 REASON_SET(&reason, PFRES_STATEINS);
3562 pf_src_tree_remove_state(s);
3563 STATE_DEC_COUNTERS(s);
3564 uma_zfree(V_pf_state_z, s);
3569 pf_set_rt_ifp(s, pd->src); /* needs s->state_key set */
3572 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3573 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3574 s->src.state = PF_TCPS_PROXY_SRC;
3575 /* undo NAT changes, if they have taken place */
3577 struct pf_state_key *skt = s->key[PF_SK_WIRE];
3578 if (pd->dir == PF_OUT)
3579 skt = s->key[PF_SK_STACK];
3580 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3581 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3583 *pd->sport = skt->port[pd->sidx];
3585 *pd->dport = skt->port[pd->didx];
3587 *pd->proto_sum = bproto_sum;
3589 *pd->ip_sum = bip_sum;
3590 m_copyback(m, off, hdrlen, pd->hdr.any);
3592 s->src.seqhi = htonl(arc4random());
3593 /* Find mss option */
3594 int rtid = M_GETFIB(m);
3595 mss = pf_get_mss(m, off, th->th_off, pd->af);
3596 mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3597 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3599 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3600 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3601 TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3602 REASON_SET(&reason, PFRES_SYNPROXY);
3603 return (PF_SYNPROXY_DROP);
3610 uma_zfree(V_pf_state_key_z, sk);
3612 uma_zfree(V_pf_state_key_z, nk);
3614 if (sn != NULL && sn->states == 0 && sn->expire == 0)
3615 pf_remove_src_node(sn);
3617 if (nsn != sn && nsn != NULL && nsn->states == 0 && nsn->expire == 0)
3618 pf_remove_src_node(nsn);
3624 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
3625 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
3626 struct pf_ruleset **rsm)
3628 struct pf_rule *r, *a = NULL;
3629 struct pf_ruleset *ruleset = NULL;
3630 sa_family_t af = pd->af;
3635 struct pf_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE];
3639 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3642 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3643 r = r->skip[PF_SKIP_IFP].ptr;
3644 else if (r->direction && r->direction != direction)
3645 r = r->skip[PF_SKIP_DIR].ptr;
3646 else if (r->af && r->af != af)
3647 r = r->skip[PF_SKIP_AF].ptr;
3648 else if (r->proto && r->proto != pd->proto)
3649 r = r->skip[PF_SKIP_PROTO].ptr;
3650 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3651 r->src.neg, kif, M_GETFIB(m)))
3652 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3653 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3654 r->dst.neg, NULL, M_GETFIB(m)))
3655 r = r->skip[PF_SKIP_DST_ADDR].ptr;
3656 else if (r->tos && !(r->tos == pd->tos))
3657 r = TAILQ_NEXT(r, entries);
3658 else if (r->os_fingerprint != PF_OSFP_ANY)
3659 r = TAILQ_NEXT(r, entries);
3660 else if (pd->proto == IPPROTO_UDP &&
3661 (r->src.port_op || r->dst.port_op))
3662 r = TAILQ_NEXT(r, entries);
3663 else if (pd->proto == IPPROTO_TCP &&
3664 (r->src.port_op || r->dst.port_op || r->flagset))
3665 r = TAILQ_NEXT(r, entries);
3666 else if ((pd->proto == IPPROTO_ICMP ||
3667 pd->proto == IPPROTO_ICMPV6) &&
3668 (r->type || r->code))
3669 r = TAILQ_NEXT(r, entries);
3670 else if (r->prob && r->prob <=
3671 (arc4random() % (UINT_MAX - 1) + 1))
3672 r = TAILQ_NEXT(r, entries);
3673 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3674 pd->pf_mtag ? pd->pf_mtag->tag : 0))
3675 r = TAILQ_NEXT(r, entries);
3677 if (r->anchor == NULL) {
3684 r = TAILQ_NEXT(r, entries);
3686 pf_step_into_anchor(anchor_stack, &asd,
3687 &ruleset, PF_RULESET_FILTER, &r, &a,
3690 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3691 &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3698 REASON_SET(&reason, PFRES_MATCH);
3701 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
3704 if (r->action != PF_PASS)
3707 if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3708 REASON_SET(&reason, PFRES_MEMORY);
3716 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
3717 struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
3718 struct pf_pdesc *pd, u_short *reason, int *copyback)
3720 struct tcphdr *th = pd->hdr.tcp;
3721 u_int16_t win = ntohs(th->th_win);
3722 u_int32_t ack, end, seq, orig_seq;
3726 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
3727 sws = src->wscale & PF_WSCALE_MASK;
3728 dws = dst->wscale & PF_WSCALE_MASK;
3733 * Sequence tracking algorithm from Guido van Rooij's paper:
3734 * http://www.madison-gurkha.com/publications/tcp_filtering/
3738 orig_seq = seq = ntohl(th->th_seq);
3739 if (src->seqlo == 0) {
3740 /* First packet from this end. Set its state */
3742 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
3743 src->scrub == NULL) {
3744 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
3745 REASON_SET(reason, PFRES_MEMORY);
3750 /* Deferred generation of sequence number modulator */
3751 if (dst->seqdiff && !src->seqdiff) {
3752 /* use random iss for the TCP server */
3753 while ((src->seqdiff = arc4random() - seq) == 0)
3755 ack = ntohl(th->th_ack) - dst->seqdiff;
3756 pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
3758 pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
3761 ack = ntohl(th->th_ack);
3764 end = seq + pd->p_len;
3765 if (th->th_flags & TH_SYN) {
3767 if (dst->wscale & PF_WSCALE_FLAG) {
3768 src->wscale = pf_get_wscale(m, off, th->th_off,
3770 if (src->wscale & PF_WSCALE_FLAG) {
3771 /* Remove scale factor from initial
3773 sws = src->wscale & PF_WSCALE_MASK;
3774 win = ((u_int32_t)win + (1 << sws) - 1)
3776 dws = dst->wscale & PF_WSCALE_MASK;
3778 /* fixup other window */
3779 dst->max_win <<= dst->wscale &
3781 /* in case of a retrans SYN|ACK */
3786 if (th->th_flags & TH_FIN)
3790 if (src->state < TCPS_SYN_SENT)
3791 src->state = TCPS_SYN_SENT;
3794 * May need to slide the window (seqhi may have been set by
3795 * the crappy stack check or if we picked up the connection
3796 * after establishment)
3798 if (src->seqhi == 1 ||
3799 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
3800 src->seqhi = end + MAX(1, dst->max_win << dws);
3801 if (win > src->max_win)
3805 ack = ntohl(th->th_ack) - dst->seqdiff;
3807 /* Modulate sequence numbers */
3808 pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
3810 pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
3813 end = seq + pd->p_len;
3814 if (th->th_flags & TH_SYN)
3816 if (th->th_flags & TH_FIN)
3820 if ((th->th_flags & TH_ACK) == 0) {
3821 /* Let it pass through the ack skew check */
3823 } else if ((ack == 0 &&
3824 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
3825 /* broken tcp stacks do not set ack */
3826 (dst->state < TCPS_SYN_SENT)) {
3828 * Many stacks (ours included) will set the ACK number in an
3829 * FIN|ACK if the SYN times out -- no sequence to ACK.
3835 /* Ease sequencing restrictions on no data packets */
3840 ackskew = dst->seqlo - ack;
3844 * Need to demodulate the sequence numbers in any TCP SACK options
3845 * (Selective ACK). We could optionally validate the SACK values
3846 * against the current ACK window, either forwards or backwards, but
3847 * I'm not confident that SACK has been implemented properly
3848 * everywhere. It wouldn't surprise me if several stacks accidently
3849 * SACK too far backwards of previously ACKed data. There really aren't
3850 * any security implications of bad SACKing unless the target stack
3851 * doesn't validate the option length correctly. Someone trying to
3852 * spoof into a TCP connection won't bother blindly sending SACK
3855 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
3856 if (pf_modulate_sack(m, off, pd, th, dst))
3861 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */
3862 if (SEQ_GEQ(src->seqhi, end) &&
3863 /* Last octet inside other's window space */
3864 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
3865 /* Retrans: not more than one window back */
3866 (ackskew >= -MAXACKWINDOW) &&
3867 /* Acking not more than one reassembled fragment backwards */
3868 (ackskew <= (MAXACKWINDOW << sws)) &&
3869 /* Acking not more than one window forward */
3870 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
3871 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
3872 (pd->flags & PFDESC_IP_REAS) == 0)) {
3873 /* Require an exact/+1 sequence match on resets when possible */
3875 if (dst->scrub || src->scrub) {
3876 if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
3877 *state, src, dst, copyback))
3881 /* update max window */
3882 if (src->max_win < win)
3884 /* synchronize sequencing */
3885 if (SEQ_GT(end, src->seqlo))
3887 /* slide the window of what the other end can send */
3888 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
3889 dst->seqhi = ack + MAX((win << sws), 1);
3893 if (th->th_flags & TH_SYN)
3894 if (src->state < TCPS_SYN_SENT)
3895 src->state = TCPS_SYN_SENT;
3896 if (th->th_flags & TH_FIN)
3897 if (src->state < TCPS_CLOSING)
3898 src->state = TCPS_CLOSING;
3899 if (th->th_flags & TH_ACK) {
3900 if (dst->state == TCPS_SYN_SENT) {
3901 dst->state = TCPS_ESTABLISHED;
3902 if (src->state == TCPS_ESTABLISHED &&
3903 (*state)->src_node != NULL &&
3904 pf_src_connlimit(state)) {
3905 REASON_SET(reason, PFRES_SRCLIMIT);
3908 } else if (dst->state == TCPS_CLOSING)
3909 dst->state = TCPS_FIN_WAIT_2;
3911 if (th->th_flags & TH_RST)
3912 src->state = dst->state = TCPS_TIME_WAIT;
3914 /* update expire time */
3915 (*state)->expire = time_uptime;
3916 if (src->state >= TCPS_FIN_WAIT_2 &&
3917 dst->state >= TCPS_FIN_WAIT_2)
3918 (*state)->timeout = PFTM_TCP_CLOSED;
3919 else if (src->state >= TCPS_CLOSING &&
3920 dst->state >= TCPS_CLOSING)
3921 (*state)->timeout = PFTM_TCP_FIN_WAIT;
3922 else if (src->state < TCPS_ESTABLISHED ||
3923 dst->state < TCPS_ESTABLISHED)
3924 (*state)->timeout = PFTM_TCP_OPENING;
3925 else if (src->state >= TCPS_CLOSING ||
3926 dst->state >= TCPS_CLOSING)
3927 (*state)->timeout = PFTM_TCP_CLOSING;
3929 (*state)->timeout = PFTM_TCP_ESTABLISHED;
3931 /* Fall through to PASS packet */
3933 } else if ((dst->state < TCPS_SYN_SENT ||
3934 dst->state >= TCPS_FIN_WAIT_2 ||
3935 src->state >= TCPS_FIN_WAIT_2) &&
3936 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
3937 /* Within a window forward of the originating packet */
3938 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
3939 /* Within a window backward of the originating packet */
3942 * This currently handles three situations:
3943 * 1) Stupid stacks will shotgun SYNs before their peer
3945 * 2) When PF catches an already established stream (the
3946 * firewall rebooted, the state table was flushed, routes
3948 * 3) Packets get funky immediately after the connection
3949 * closes (this should catch Solaris spurious ACK|FINs
3950 * that web servers like to spew after a close)
3952 * This must be a little more careful than the above code
3953 * since packet floods will also be caught here. We don't
3954 * update the TTL here to mitigate the damage of a packet
3955 * flood and so the same code can handle awkward establishment
3956 * and a loosened connection close.
3957 * In the establishment case, a correct peer response will
3958 * validate the connection, go through the normal state code
3959 * and keep updating the state TTL.
3962 if (V_pf_status.debug >= PF_DEBUG_MISC) {
3963 printf("pf: loose state match: ");
3964 pf_print_state(*state);
3965 pf_print_flags(th->th_flags);
3966 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
3967 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
3968 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
3969 (unsigned long long)(*state)->packets[1],
3970 pd->dir == PF_IN ? "in" : "out",
3971 pd->dir == (*state)->direction ? "fwd" : "rev");
3974 if (dst->scrub || src->scrub) {
3975 if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
3976 *state, src, dst, copyback))
3980 /* update max window */
3981 if (src->max_win < win)
3983 /* synchronize sequencing */
3984 if (SEQ_GT(end, src->seqlo))
3986 /* slide the window of what the other end can send */
3987 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
3988 dst->seqhi = ack + MAX((win << sws), 1);
3991 * Cannot set dst->seqhi here since this could be a shotgunned
3992 * SYN and not an already established connection.
3995 if (th->th_flags & TH_FIN)
3996 if (src->state < TCPS_CLOSING)
3997 src->state = TCPS_CLOSING;
3998 if (th->th_flags & TH_RST)
3999 src->state = dst->state = TCPS_TIME_WAIT;
4001 /* Fall through to PASS packet */
4004 if ((*state)->dst.state == TCPS_SYN_SENT &&
4005 (*state)->src.state == TCPS_SYN_SENT) {
4006 /* Send RST for state mismatches during handshake */
4007 if (!(th->th_flags & TH_RST))
4008 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4009 pd->dst, pd->src, th->th_dport,
4010 th->th_sport, ntohl(th->th_ack), 0,
4012 (*state)->rule.ptr->return_ttl, 1, 0,
4017 } else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4018 printf("pf: BAD state: ");
4019 pf_print_state(*state);
4020 pf_print_flags(th->th_flags);
4021 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4022 "pkts=%llu:%llu dir=%s,%s\n",
4023 seq, orig_seq, ack, pd->p_len, ackskew,
4024 (unsigned long long)(*state)->packets[0],
4025 (unsigned long long)(*state)->packets[1],
4026 pd->dir == PF_IN ? "in" : "out",
4027 pd->dir == (*state)->direction ? "fwd" : "rev");
4028 printf("pf: State failure on: %c %c %c %c | %c %c\n",
4029 SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4030 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4032 (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4033 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4034 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4035 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4037 REASON_SET(reason, PFRES_BADSTATE);
4045 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4046 struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4048 struct tcphdr *th = pd->hdr.tcp;
4050 if (th->th_flags & TH_SYN)
4051 if (src->state < TCPS_SYN_SENT)
4052 src->state = TCPS_SYN_SENT;
4053 if (th->th_flags & TH_FIN)
4054 if (src->state < TCPS_CLOSING)
4055 src->state = TCPS_CLOSING;
4056 if (th->th_flags & TH_ACK) {
4057 if (dst->state == TCPS_SYN_SENT) {
4058 dst->state = TCPS_ESTABLISHED;
4059 if (src->state == TCPS_ESTABLISHED &&
4060 (*state)->src_node != NULL &&
4061 pf_src_connlimit(state)) {
4062 REASON_SET(reason, PFRES_SRCLIMIT);
4065 } else if (dst->state == TCPS_CLOSING) {
4066 dst->state = TCPS_FIN_WAIT_2;
4067 } else if (src->state == TCPS_SYN_SENT &&
4068 dst->state < TCPS_SYN_SENT) {
4070 * Handle a special sloppy case where we only see one
4071 * half of the connection. If there is a ACK after
4072 * the initial SYN without ever seeing a packet from
4073 * the destination, set the connection to established.
4075 dst->state = src->state = TCPS_ESTABLISHED;
4076 if ((*state)->src_node != NULL &&
4077 pf_src_connlimit(state)) {
4078 REASON_SET(reason, PFRES_SRCLIMIT);
4081 } else if (src->state == TCPS_CLOSING &&
4082 dst->state == TCPS_ESTABLISHED &&
4085 * Handle the closing of half connections where we
4086 * don't see the full bidirectional FIN/ACK+ACK
4089 dst->state = TCPS_CLOSING;
4092 if (th->th_flags & TH_RST)
4093 src->state = dst->state = TCPS_TIME_WAIT;
4095 /* update expire time */
4096 (*state)->expire = time_uptime;
4097 if (src->state >= TCPS_FIN_WAIT_2 &&
4098 dst->state >= TCPS_FIN_WAIT_2)
4099 (*state)->timeout = PFTM_TCP_CLOSED;
4100 else if (src->state >= TCPS_CLOSING &&
4101 dst->state >= TCPS_CLOSING)
4102 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4103 else if (src->state < TCPS_ESTABLISHED ||
4104 dst->state < TCPS_ESTABLISHED)
4105 (*state)->timeout = PFTM_TCP_OPENING;
4106 else if (src->state >= TCPS_CLOSING ||
4107 dst->state >= TCPS_CLOSING)
4108 (*state)->timeout = PFTM_TCP_CLOSING;
4110 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4116 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4117 struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4120 struct pf_state_key_cmp key;
4121 struct tcphdr *th = pd->hdr.tcp;
4123 struct pf_state_peer *src, *dst;
4124 struct pf_state_key *sk;
4126 bzero(&key, sizeof(key));
4128 key.proto = IPPROTO_TCP;
4129 if (direction == PF_IN) { /* wire side, straight */
4130 PF_ACPY(&key.addr[0], pd->src, key.af);
4131 PF_ACPY(&key.addr[1], pd->dst, key.af);
4132 key.port[0] = th->th_sport;
4133 key.port[1] = th->th_dport;
4134 } else { /* stack side, reverse */
4135 PF_ACPY(&key.addr[1], pd->src, key.af);
4136 PF_ACPY(&key.addr[0], pd->dst, key.af);
4137 key.port[1] = th->th_sport;
4138 key.port[0] = th->th_dport;
4141 STATE_LOOKUP(kif, &key, direction, *state, pd);
4143 if (direction == (*state)->direction) {
4144 src = &(*state)->src;
4145 dst = &(*state)->dst;
4147 src = &(*state)->dst;
4148 dst = &(*state)->src;
4151 sk = (*state)->key[pd->didx];
4153 if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4154 if (direction != (*state)->direction) {
4155 REASON_SET(reason, PFRES_SYNPROXY);
4156 return (PF_SYNPROXY_DROP);
4158 if (th->th_flags & TH_SYN) {
4159 if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4160 REASON_SET(reason, PFRES_SYNPROXY);
4163 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4164 pd->src, th->th_dport, th->th_sport,
4165 (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4166 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4167 REASON_SET(reason, PFRES_SYNPROXY);
4168 return (PF_SYNPROXY_DROP);
4169 } else if (!(th->th_flags & TH_ACK) ||
4170 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4171 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4172 REASON_SET(reason, PFRES_SYNPROXY);
4174 } else if ((*state)->src_node != NULL &&
4175 pf_src_connlimit(state)) {
4176 REASON_SET(reason, PFRES_SRCLIMIT);
4179 (*state)->src.state = PF_TCPS_PROXY_DST;
4181 if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4182 if (direction == (*state)->direction) {
4183 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4184 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4185 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4186 REASON_SET(reason, PFRES_SYNPROXY);
4189 (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4190 if ((*state)->dst.seqhi == 1)
4191 (*state)->dst.seqhi = htonl(arc4random());
4192 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4193 &sk->addr[pd->sidx], &sk->addr[pd->didx],
4194 sk->port[pd->sidx], sk->port[pd->didx],
4195 (*state)->dst.seqhi, 0, TH_SYN, 0,
4196 (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4197 REASON_SET(reason, PFRES_SYNPROXY);
4198 return (PF_SYNPROXY_DROP);
4199 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4201 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4202 REASON_SET(reason, PFRES_SYNPROXY);
4205 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4206 (*state)->dst.seqlo = ntohl(th->th_seq);
4207 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4208 pd->src, th->th_dport, th->th_sport,
4209 ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4210 TH_ACK, (*state)->src.max_win, 0, 0, 0,
4211 (*state)->tag, NULL);
4212 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4213 &sk->addr[pd->sidx], &sk->addr[pd->didx],
4214 sk->port[pd->sidx], sk->port[pd->didx],
4215 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4216 TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4217 (*state)->src.seqdiff = (*state)->dst.seqhi -
4218 (*state)->src.seqlo;
4219 (*state)->dst.seqdiff = (*state)->src.seqhi -
4220 (*state)->dst.seqlo;
4221 (*state)->src.seqhi = (*state)->src.seqlo +
4222 (*state)->dst.max_win;
4223 (*state)->dst.seqhi = (*state)->dst.seqlo +
4224 (*state)->src.max_win;
4225 (*state)->src.wscale = (*state)->dst.wscale = 0;
4226 (*state)->src.state = (*state)->dst.state =
4228 REASON_SET(reason, PFRES_SYNPROXY);
4229 return (PF_SYNPROXY_DROP);
4233 if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4234 dst->state >= TCPS_FIN_WAIT_2 &&
4235 src->state >= TCPS_FIN_WAIT_2) {
4236 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4237 printf("pf: state reuse ");
4238 pf_print_state(*state);
4239 pf_print_flags(th->th_flags);
4242 /* XXX make sure it's the same direction ?? */
4243 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4244 pf_unlink_state(*state, PF_ENTER_LOCKED);
4249 if ((*state)->state_flags & PFSTATE_SLOPPY) {
4250 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4253 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4254 ©back) == PF_DROP)
4258 /* translate source/destination address, if necessary */
4259 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4260 struct pf_state_key *nk = (*state)->key[pd->didx];
4262 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4263 nk->port[pd->sidx] != th->th_sport)
4264 pf_change_ap(pd->src, &th->th_sport, pd->ip_sum,
4265 &th->th_sum, &nk->addr[pd->sidx],
4266 nk->port[pd->sidx], 0, pd->af);
4268 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4269 nk->port[pd->didx] != th->th_dport)
4270 pf_change_ap(pd->dst, &th->th_dport, pd->ip_sum,
4271 &th->th_sum, &nk->addr[pd->didx],
4272 nk->port[pd->didx], 0, pd->af);
4276 /* Copyback sequence modulation or stateful scrub changes if needed */
4278 m_copyback(m, off, sizeof(*th), (caddr_t)th);
4284 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
4285 struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4287 struct pf_state_peer *src, *dst;
4288 struct pf_state_key_cmp key;
4289 struct udphdr *uh = pd->hdr.udp;
4291 bzero(&key, sizeof(key));
4293 key.proto = IPPROTO_UDP;
4294 if (direction == PF_IN) { /* wire side, straight */
4295 PF_ACPY(&key.addr[0], pd->src, key.af);
4296 PF_ACPY(&key.addr[1], pd->dst, key.af);
4297 key.port[0] = uh->uh_sport;
4298 key.port[1] = uh->uh_dport;
4299 } else { /* stack side, reverse */
4300 PF_ACPY(&key.addr[1], pd->src, key.af);
4301 PF_ACPY(&key.addr[0], pd->dst, key.af);
4302 key.port[1] = uh->uh_sport;
4303 key.port[0] = uh->uh_dport;
4306 STATE_LOOKUP(kif, &key, direction, *state, pd);
4308 if (direction == (*state)->direction) {
4309 src = &(*state)->src;
4310 dst = &(*state)->dst;
4312 src = &(*state)->dst;
4313 dst = &(*state)->src;
4317 if (src->state < PFUDPS_SINGLE)
4318 src->state = PFUDPS_SINGLE;
4319 if (dst->state == PFUDPS_SINGLE)
4320 dst->state = PFUDPS_MULTIPLE;
4322 /* update expire time */
4323 (*state)->expire = time_uptime;
4324 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4325 (*state)->timeout = PFTM_UDP_MULTIPLE;
4327 (*state)->timeout = PFTM_UDP_SINGLE;
4329 /* translate source/destination address, if necessary */
4330 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4331 struct pf_state_key *nk = (*state)->key[pd->didx];
4333 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4334 nk->port[pd->sidx] != uh->uh_sport)
4335 pf_change_ap(pd->src, &uh->uh_sport, pd->ip_sum,
4336 &uh->uh_sum, &nk->addr[pd->sidx],
4337 nk->port[pd->sidx], 1, pd->af);
4339 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4340 nk->port[pd->didx] != uh->uh_dport)
4341 pf_change_ap(pd->dst, &uh->uh_dport, pd->ip_sum,
4342 &uh->uh_sum, &nk->addr[pd->didx],
4343 nk->port[pd->didx], 1, pd->af);
4344 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4351 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
4352 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4354 struct pf_addr *saddr = pd->src, *daddr = pd->dst;
4355 u_int16_t icmpid = 0, *icmpsum;
4358 struct pf_state_key_cmp key;
4360 bzero(&key, sizeof(key));
4361 switch (pd->proto) {
4364 icmptype = pd->hdr.icmp->icmp_type;
4365 icmpid = pd->hdr.icmp->icmp_id;
4366 icmpsum = &pd->hdr.icmp->icmp_cksum;
4368 if (icmptype == ICMP_UNREACH ||
4369 icmptype == ICMP_SOURCEQUENCH ||
4370 icmptype == ICMP_REDIRECT ||
4371 icmptype == ICMP_TIMXCEED ||
4372 icmptype == ICMP_PARAMPROB)
4377 case IPPROTO_ICMPV6:
4378 icmptype = pd->hdr.icmp6->icmp6_type;
4379 icmpid = pd->hdr.icmp6->icmp6_id;
4380 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4382 if (icmptype == ICMP6_DST_UNREACH ||
4383 icmptype == ICMP6_PACKET_TOO_BIG ||
4384 icmptype == ICMP6_TIME_EXCEEDED ||
4385 icmptype == ICMP6_PARAM_PROB)
4394 * ICMP query/reply message not related to a TCP/UDP packet.
4395 * Search for an ICMP state.
4398 key.proto = pd->proto;
4399 key.port[0] = key.port[1] = icmpid;
4400 if (direction == PF_IN) { /* wire side, straight */
4401 PF_ACPY(&key.addr[0], pd->src, key.af);
4402 PF_ACPY(&key.addr[1], pd->dst, key.af);
4403 } else { /* stack side, reverse */
4404 PF_ACPY(&key.addr[1], pd->src, key.af);
4405 PF_ACPY(&key.addr[0], pd->dst, key.af);
4408 STATE_LOOKUP(kif, &key, direction, *state, pd);
4410 (*state)->expire = time_uptime;
4411 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4413 /* translate source/destination address, if necessary */
4414 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4415 struct pf_state_key *nk = (*state)->key[pd->didx];
4420 if (PF_ANEQ(pd->src,
4421 &nk->addr[pd->sidx], AF_INET))
4422 pf_change_a(&saddr->v4.s_addr,
4424 nk->addr[pd->sidx].v4.s_addr, 0);
4426 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4428 pf_change_a(&daddr->v4.s_addr,
4430 nk->addr[pd->didx].v4.s_addr, 0);
4433 pd->hdr.icmp->icmp_id) {
4434 pd->hdr.icmp->icmp_cksum =
4436 pd->hdr.icmp->icmp_cksum, icmpid,
4437 nk->port[pd->sidx], 0);
4438 pd->hdr.icmp->icmp_id =
4442 m_copyback(m, off, ICMP_MINLEN,
4443 (caddr_t )pd->hdr.icmp);
4448 if (PF_ANEQ(pd->src,
4449 &nk->addr[pd->sidx], AF_INET6))
4451 &pd->hdr.icmp6->icmp6_cksum,
4452 &nk->addr[pd->sidx], 0);
4454 if (PF_ANEQ(pd->dst,
4455 &nk->addr[pd->didx], AF_INET6))
4457 &pd->hdr.icmp6->icmp6_cksum,
4458 &nk->addr[pd->didx], 0);
4460 m_copyback(m, off, sizeof(struct icmp6_hdr),
4461 (caddr_t )pd->hdr.icmp6);
4470 * ICMP error message in response to a TCP/UDP packet.
4471 * Extract the inner TCP/UDP header and search for that state.
4474 struct pf_pdesc pd2;
4475 bzero(&pd2, sizeof pd2);
4480 struct ip6_hdr h2_6;
4487 /* Payload packet is from the opposite direction. */
4488 pd2.sidx = (direction == PF_IN) ? 1 : 0;
4489 pd2.didx = (direction == PF_IN) ? 0 : 1;
4493 /* offset of h2 in mbuf chain */
4494 ipoff2 = off + ICMP_MINLEN;
4496 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4497 NULL, reason, pd2.af)) {
4498 DPFPRINTF(PF_DEBUG_MISC,
4499 ("pf: ICMP error message too short "
4504 * ICMP error messages don't refer to non-first
4507 if (h2.ip_off & htons(IP_OFFMASK)) {
4508 REASON_SET(reason, PFRES_FRAG);
4512 /* offset of protocol header that follows h2 */
4513 off2 = ipoff2 + (h2.ip_hl << 2);
4515 pd2.proto = h2.ip_p;
4516 pd2.src = (struct pf_addr *)&h2.ip_src;
4517 pd2.dst = (struct pf_addr *)&h2.ip_dst;
4518 pd2.ip_sum = &h2.ip_sum;
4523 ipoff2 = off + sizeof(struct icmp6_hdr);
4525 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4526 NULL, reason, pd2.af)) {
4527 DPFPRINTF(PF_DEBUG_MISC,
4528 ("pf: ICMP error message too short "
4532 pd2.proto = h2_6.ip6_nxt;
4533 pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4534 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4536 off2 = ipoff2 + sizeof(h2_6);
4538 switch (pd2.proto) {
4539 case IPPROTO_FRAGMENT:
4541 * ICMPv6 error messages for
4542 * non-first fragments
4544 REASON_SET(reason, PFRES_FRAG);
4547 case IPPROTO_HOPOPTS:
4548 case IPPROTO_ROUTING:
4549 case IPPROTO_DSTOPTS: {
4550 /* get next header and header length */
4551 struct ip6_ext opt6;
4553 if (!pf_pull_hdr(m, off2, &opt6,
4554 sizeof(opt6), NULL, reason,
4556 DPFPRINTF(PF_DEBUG_MISC,
4557 ("pf: ICMPv6 short opt\n"));
4560 if (pd2.proto == IPPROTO_AH)
4561 off2 += (opt6.ip6e_len + 2) * 4;
4563 off2 += (opt6.ip6e_len + 1) * 8;
4564 pd2.proto = opt6.ip6e_nxt;
4565 /* goto the next header */
4572 } while (!terminal);
4577 switch (pd2.proto) {
4581 struct pf_state_peer *src, *dst;
4586 * Only the first 8 bytes of the TCP header can be
4587 * expected. Don't access any TCP header fields after
4588 * th_seq, an ackskew test is not possible.
4590 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4592 DPFPRINTF(PF_DEBUG_MISC,
4593 ("pf: ICMP error message too short "
4599 key.proto = IPPROTO_TCP;
4600 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4601 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4602 key.port[pd2.sidx] = th.th_sport;
4603 key.port[pd2.didx] = th.th_dport;
4605 STATE_LOOKUP(kif, &key, direction, *state, pd);
4607 if (direction == (*state)->direction) {
4608 src = &(*state)->dst;
4609 dst = &(*state)->src;
4611 src = &(*state)->src;
4612 dst = &(*state)->dst;
4615 if (src->wscale && dst->wscale)
4616 dws = dst->wscale & PF_WSCALE_MASK;
4620 /* Demodulate sequence number */
4621 seq = ntohl(th.th_seq) - src->seqdiff;
4623 pf_change_a(&th.th_seq, icmpsum,
4628 if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4629 (!SEQ_GEQ(src->seqhi, seq) ||
4630 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4631 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4632 printf("pf: BAD ICMP %d:%d ",
4633 icmptype, pd->hdr.icmp->icmp_code);
4634 pf_print_host(pd->src, 0, pd->af);
4636 pf_print_host(pd->dst, 0, pd->af);
4638 pf_print_state(*state);
4639 printf(" seq=%u\n", seq);
4641 REASON_SET(reason, PFRES_BADSTATE);
4644 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4645 printf("pf: OK ICMP %d:%d ",
4646 icmptype, pd->hdr.icmp->icmp_code);
4647 pf_print_host(pd->src, 0, pd->af);
4649 pf_print_host(pd->dst, 0, pd->af);
4651 pf_print_state(*state);
4652 printf(" seq=%u\n", seq);
4656 /* translate source/destination address, if necessary */
4657 if ((*state)->key[PF_SK_WIRE] !=
4658 (*state)->key[PF_SK_STACK]) {
4659 struct pf_state_key *nk =
4660 (*state)->key[pd->didx];
4662 if (PF_ANEQ(pd2.src,
4663 &nk->addr[pd2.sidx], pd2.af) ||
4664 nk->port[pd2.sidx] != th.th_sport)
4665 pf_change_icmp(pd2.src, &th.th_sport,
4666 daddr, &nk->addr[pd2.sidx],
4667 nk->port[pd2.sidx], NULL,
4668 pd2.ip_sum, icmpsum,
4669 pd->ip_sum, 0, pd2.af);
4671 if (PF_ANEQ(pd2.dst,
4672 &nk->addr[pd2.didx], pd2.af) ||
4673 nk->port[pd2.didx] != th.th_dport)
4674 pf_change_icmp(pd2.dst, &th.th_dport,
4675 NULL, /* XXX Inbound NAT? */
4676 &nk->addr[pd2.didx],
4677 nk->port[pd2.didx], NULL,
4678 pd2.ip_sum, icmpsum,
4679 pd->ip_sum, 0, pd2.af);
4687 m_copyback(m, off, ICMP_MINLEN,
4688 (caddr_t )pd->hdr.icmp);
4689 m_copyback(m, ipoff2, sizeof(h2),
4696 sizeof(struct icmp6_hdr),
4697 (caddr_t )pd->hdr.icmp6);
4698 m_copyback(m, ipoff2, sizeof(h2_6),
4703 m_copyback(m, off2, 8, (caddr_t)&th);
4712 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
4713 NULL, reason, pd2.af)) {
4714 DPFPRINTF(PF_DEBUG_MISC,
4715 ("pf: ICMP error message too short "
4721 key.proto = IPPROTO_UDP;
4722 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4723 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4724 key.port[pd2.sidx] = uh.uh_sport;
4725 key.port[pd2.didx] = uh.uh_dport;
4727 STATE_LOOKUP(kif, &key, direction, *state, pd);
4729 /* translate source/destination address, if necessary */
4730 if ((*state)->key[PF_SK_WIRE] !=
4731 (*state)->key[PF_SK_STACK]) {
4732 struct pf_state_key *nk =
4733 (*state)->key[pd->didx];
4735 if (PF_ANEQ(pd2.src,
4736 &nk->addr[pd2.sidx], pd2.af) ||
4737 nk->port[pd2.sidx] != uh.uh_sport)
4738 pf_change_icmp(pd2.src, &uh.uh_sport,
4739 daddr, &nk->addr[pd2.sidx],
4740 nk->port[pd2.sidx], &uh.uh_sum,
4741 pd2.ip_sum, icmpsum,
4742 pd->ip_sum, 1, pd2.af);
4744 if (PF_ANEQ(pd2.dst,
4745 &nk->addr[pd2.didx], pd2.af) ||
4746 nk->port[pd2.didx] != uh.uh_dport)
4747 pf_change_icmp(pd2.dst, &uh.uh_dport,
4748 NULL, /* XXX Inbound NAT? */
4749 &nk->addr[pd2.didx],
4750 nk->port[pd2.didx], &uh.uh_sum,
4751 pd2.ip_sum, icmpsum,
4752 pd->ip_sum, 1, pd2.af);
4757 m_copyback(m, off, ICMP_MINLEN,
4758 (caddr_t )pd->hdr.icmp);
4759 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4765 sizeof(struct icmp6_hdr),
4766 (caddr_t )pd->hdr.icmp6);
4767 m_copyback(m, ipoff2, sizeof(h2_6),
4772 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
4778 case IPPROTO_ICMP: {
4781 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
4782 NULL, reason, pd2.af)) {
4783 DPFPRINTF(PF_DEBUG_MISC,
4784 ("pf: ICMP error message too short i"
4790 key.proto = IPPROTO_ICMP;
4791 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4792 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4793 key.port[0] = key.port[1] = iih.icmp_id;
4795 STATE_LOOKUP(kif, &key, direction, *state, pd);
4797 /* translate source/destination address, if necessary */
4798 if ((*state)->key[PF_SK_WIRE] !=
4799 (*state)->key[PF_SK_STACK]) {
4800 struct pf_state_key *nk =
4801 (*state)->key[pd->didx];
4803 if (PF_ANEQ(pd2.src,
4804 &nk->addr[pd2.sidx], pd2.af) ||
4805 nk->port[pd2.sidx] != iih.icmp_id)
4806 pf_change_icmp(pd2.src, &iih.icmp_id,
4807 daddr, &nk->addr[pd2.sidx],
4808 nk->port[pd2.sidx], NULL,
4809 pd2.ip_sum, icmpsum,
4810 pd->ip_sum, 0, AF_INET);
4812 if (PF_ANEQ(pd2.dst,
4813 &nk->addr[pd2.didx], pd2.af) ||
4814 nk->port[pd2.didx] != iih.icmp_id)
4815 pf_change_icmp(pd2.dst, &iih.icmp_id,
4816 NULL, /* XXX Inbound NAT? */
4817 &nk->addr[pd2.didx],
4818 nk->port[pd2.didx], NULL,
4819 pd2.ip_sum, icmpsum,
4820 pd->ip_sum, 0, AF_INET);
4822 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
4823 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4824 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
4831 case IPPROTO_ICMPV6: {
4832 struct icmp6_hdr iih;
4834 if (!pf_pull_hdr(m, off2, &iih,
4835 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
4836 DPFPRINTF(PF_DEBUG_MISC,
4837 ("pf: ICMP error message too short "
4843 key.proto = IPPROTO_ICMPV6;
4844 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4845 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4846 key.port[0] = key.port[1] = iih.icmp6_id;
4848 STATE_LOOKUP(kif, &key, direction, *state, pd);
4850 /* translate source/destination address, if necessary */
4851 if ((*state)->key[PF_SK_WIRE] !=
4852 (*state)->key[PF_SK_STACK]) {
4853 struct pf_state_key *nk =
4854 (*state)->key[pd->didx];
4856 if (PF_ANEQ(pd2.src,
4857 &nk->addr[pd2.sidx], pd2.af) ||
4858 nk->port[pd2.sidx] != iih.icmp6_id)
4859 pf_change_icmp(pd2.src, &iih.icmp6_id,
4860 daddr, &nk->addr[pd2.sidx],
4861 nk->port[pd2.sidx], NULL,
4862 pd2.ip_sum, icmpsum,
4863 pd->ip_sum, 0, AF_INET6);
4865 if (PF_ANEQ(pd2.dst,
4866 &nk->addr[pd2.didx], pd2.af) ||
4867 nk->port[pd2.didx] != iih.icmp6_id)
4868 pf_change_icmp(pd2.dst, &iih.icmp6_id,
4869 NULL, /* XXX Inbound NAT? */
4870 &nk->addr[pd2.didx],
4871 nk->port[pd2.didx], NULL,
4872 pd2.ip_sum, icmpsum,
4873 pd->ip_sum, 0, AF_INET6);
4875 m_copyback(m, off, sizeof(struct icmp6_hdr),
4876 (caddr_t)pd->hdr.icmp6);
4877 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
4878 m_copyback(m, off2, sizeof(struct icmp6_hdr),
4887 key.proto = pd2.proto;
4888 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4889 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4890 key.port[0] = key.port[1] = 0;
4892 STATE_LOOKUP(kif, &key, direction, *state, pd);
4894 /* translate source/destination address, if necessary */
4895 if ((*state)->key[PF_SK_WIRE] !=
4896 (*state)->key[PF_SK_STACK]) {
4897 struct pf_state_key *nk =
4898 (*state)->key[pd->didx];
4900 if (PF_ANEQ(pd2.src,
4901 &nk->addr[pd2.sidx], pd2.af))
4902 pf_change_icmp(pd2.src, NULL, daddr,
4903 &nk->addr[pd2.sidx], 0, NULL,
4904 pd2.ip_sum, icmpsum,
4905 pd->ip_sum, 0, pd2.af);
4907 if (PF_ANEQ(pd2.dst,
4908 &nk->addr[pd2.didx], pd2.af))
4909 pf_change_icmp(pd2.src, NULL,
4910 NULL, /* XXX Inbound NAT? */
4911 &nk->addr[pd2.didx], 0, NULL,
4912 pd2.ip_sum, icmpsum,
4913 pd->ip_sum, 0, pd2.af);
4918 m_copyback(m, off, ICMP_MINLEN,
4919 (caddr_t)pd->hdr.icmp);
4920 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4926 sizeof(struct icmp6_hdr),
4927 (caddr_t )pd->hdr.icmp6);
4928 m_copyback(m, ipoff2, sizeof(h2_6),
4942 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
4943 struct mbuf *m, struct pf_pdesc *pd)
4945 struct pf_state_peer *src, *dst;
4946 struct pf_state_key_cmp key;
4948 bzero(&key, sizeof(key));
4950 key.proto = pd->proto;
4951 if (direction == PF_IN) {
4952 PF_ACPY(&key.addr[0], pd->src, key.af);
4953 PF_ACPY(&key.addr[1], pd->dst, key.af);
4954 key.port[0] = key.port[1] = 0;
4956 PF_ACPY(&key.addr[1], pd->src, key.af);
4957 PF_ACPY(&key.addr[0], pd->dst, key.af);
4958 key.port[1] = key.port[0] = 0;
4961 STATE_LOOKUP(kif, &key, direction, *state, pd);
4963 if (direction == (*state)->direction) {
4964 src = &(*state)->src;
4965 dst = &(*state)->dst;
4967 src = &(*state)->dst;
4968 dst = &(*state)->src;
4972 if (src->state < PFOTHERS_SINGLE)
4973 src->state = PFOTHERS_SINGLE;
4974 if (dst->state == PFOTHERS_SINGLE)
4975 dst->state = PFOTHERS_MULTIPLE;
4977 /* update expire time */
4978 (*state)->expire = time_uptime;
4979 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
4980 (*state)->timeout = PFTM_OTHER_MULTIPLE;
4982 (*state)->timeout = PFTM_OTHER_SINGLE;
4984 /* translate source/destination address, if necessary */
4985 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4986 struct pf_state_key *nk = (*state)->key[pd->didx];
4988 KASSERT(nk, ("%s: nk is null", __func__));
4989 KASSERT(pd, ("%s: pd is null", __func__));
4990 KASSERT(pd->src, ("%s: pd->src is null", __func__));
4991 KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
4995 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
4996 pf_change_a(&pd->src->v4.s_addr,
4998 nk->addr[pd->sidx].v4.s_addr,
5002 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5003 pf_change_a(&pd->dst->v4.s_addr,
5005 nk->addr[pd->didx].v4.s_addr,
5012 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5013 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5015 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5016 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5024 * ipoff and off are measured from the start of the mbuf chain.
5025 * h must be at "ipoff" on the mbuf chain.
5028 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5029 u_short *actionp, u_short *reasonp, sa_family_t af)
5034 struct ip *h = mtod(m, struct ip *);
5035 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5039 ACTION_SET(actionp, PF_PASS);
5041 ACTION_SET(actionp, PF_DROP);
5042 REASON_SET(reasonp, PFRES_FRAG);
5046 if (m->m_pkthdr.len < off + len ||
5047 ntohs(h->ip_len) < off + len) {
5048 ACTION_SET(actionp, PF_DROP);
5049 REASON_SET(reasonp, PFRES_SHORT);
5057 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
5059 if (m->m_pkthdr.len < off + len ||
5060 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5061 (unsigned)(off + len)) {
5062 ACTION_SET(actionp, PF_DROP);
5063 REASON_SET(reasonp, PFRES_SHORT);
5070 m_copydata(m, off, len, p);
5075 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5079 struct radix_node_head *rnh;
5081 struct sockaddr_in *dst;
5085 struct sockaddr_in6 *dst6;
5086 struct route_in6 ro;
5090 struct radix_node *rn;
5096 /* XXX: stick to table 0 for now */
5097 rnh = rt_tables_get_rnh(0, af);
5098 if (rnh != NULL && rn_mpath_capable(rnh))
5101 bzero(&ro, sizeof(ro));
5104 dst = satosin(&ro.ro_dst);
5105 dst->sin_family = AF_INET;
5106 dst->sin_len = sizeof(*dst);
5107 dst->sin_addr = addr->v4;
5112 * Skip check for addresses with embedded interface scope,
5113 * as they would always match anyway.
5115 if (IN6_IS_SCOPE_EMBED(&addr->v6))
5117 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5118 dst6->sin6_family = AF_INET6;
5119 dst6->sin6_len = sizeof(*dst6);
5120 dst6->sin6_addr = addr->v6;
5127 /* Skip checks for ipsec interfaces */
5128 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5134 in6_rtalloc_ign(&ro, 0, rtableid);
5139 in_rtalloc_ign((struct route *)&ro, 0, rtableid);
5143 rtalloc_ign((struct route *)&ro, 0); /* No/default FIB. */
5147 if (ro.ro_rt != NULL) {
5148 /* No interface given, this is a no-route check */
5152 if (kif->pfik_ifp == NULL) {
5157 /* Perform uRPF check if passed input interface */
5159 rn = (struct radix_node *)ro.ro_rt;
5161 rt = (struct rtentry *)rn;
5164 if (kif->pfik_ifp == ifp)
5167 rn = rn_mpath_next(rn);
5169 } while (check_mpath == 1 && rn != NULL && ret == 0);
5173 if (ro.ro_rt != NULL)
5180 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5181 struct pf_state *s, struct pf_pdesc *pd)
5183 struct mbuf *m0, *m1;
5184 struct sockaddr_in dst;
5186 struct ifnet *ifp = NULL;
5187 struct pf_addr naddr;
5188 struct pf_src_node *sn = NULL;
5190 uint16_t ip_len, ip_off;
5192 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5193 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5196 if ((pd->pf_mtag == NULL &&
5197 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5198 pd->pf_mtag->routed++ > 3) {
5204 if (r->rt == PF_DUPTO) {
5205 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5211 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5219 ip = mtod(m0, struct ip *);
5221 bzero(&dst, sizeof(dst));
5222 dst.sin_family = AF_INET;
5223 dst.sin_len = sizeof(dst);
5224 dst.sin_addr = ip->ip_dst;
5226 if (r->rt == PF_FASTROUTE) {
5231 rt = rtalloc1_fib(sintosa(&dst), 0, 0, M_GETFIB(m0));
5234 KMOD_IPSTAT_INC(ips_noroute);
5235 error = EHOSTUNREACH;
5240 rt->rt_rmx.rmx_pksent++;
5242 if (rt->rt_flags & RTF_GATEWAY)
5243 bcopy(satosin(rt->rt_gateway), &dst, sizeof(dst));
5246 if (TAILQ_EMPTY(&r->rpool.list)) {
5247 DPFPRINTF(PF_DEBUG_URGENT,
5248 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5252 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5254 if (!PF_AZERO(&naddr, AF_INET))
5255 dst.sin_addr.s_addr = naddr.v4.s_addr;
5256 ifp = r->rpool.cur->kif ?
5257 r->rpool.cur->kif->pfik_ifp : NULL;
5259 if (!PF_AZERO(&s->rt_addr, AF_INET))
5260 dst.sin_addr.s_addr =
5261 s->rt_addr.v4.s_addr;
5262 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5270 if (pf_test(PF_OUT, ifp, &m0, NULL) != PF_PASS)
5272 else if (m0 == NULL)
5274 if (m0->m_len < sizeof(struct ip)) {
5275 DPFPRINTF(PF_DEBUG_URGENT,
5276 ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5279 ip = mtod(m0, struct ip *);
5282 if (ifp->if_flags & IFF_LOOPBACK)
5283 m0->m_flags |= M_SKIP_FIREWALL;
5285 ip_len = ntohs(ip->ip_len);
5286 ip_off = ntohs(ip->ip_off);
5288 /* Copied from FreeBSD 10.0-CURRENT ip_output. */
5289 m0->m_pkthdr.csum_flags |= CSUM_IP;
5290 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5291 in_delayed_cksum(m0);
5292 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5295 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5296 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
5297 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5302 * If small enough for interface, or the interface will take
5303 * care of the fragmentation for us, we can just send directly.
5305 if (ip_len <= ifp->if_mtu ||
5306 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 ||
5307 ((ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) {
5309 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5310 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5311 m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5313 m_clrprotoflags(m0); /* Avoid confusing lower layers. */
5314 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5318 /* Balk when DF bit is set or the interface didn't support TSO. */
5319 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5321 KMOD_IPSTAT_INC(ips_cantfrag);
5322 if (r->rt != PF_DUPTO) {
5323 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5330 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5334 for (; m0; m0 = m1) {
5336 m0->m_nextpkt = NULL;
5338 m_clrprotoflags(m0);
5339 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5345 KMOD_IPSTAT_INC(ips_fragmented);
5348 if (r->rt != PF_DUPTO)
5363 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5364 struct pf_state *s, struct pf_pdesc *pd)
5367 struct sockaddr_in6 dst;
5368 struct ip6_hdr *ip6;
5369 struct ifnet *ifp = NULL;
5370 struct pf_addr naddr;
5371 struct pf_src_node *sn = NULL;
5373 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5374 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5377 if ((pd->pf_mtag == NULL &&
5378 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5379 pd->pf_mtag->routed++ > 3) {
5385 if (r->rt == PF_DUPTO) {
5386 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5392 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5400 ip6 = mtod(m0, struct ip6_hdr *);
5402 bzero(&dst, sizeof(dst));
5403 dst.sin6_family = AF_INET6;
5404 dst.sin6_len = sizeof(dst);
5405 dst.sin6_addr = ip6->ip6_dst;
5407 /* Cheat. XXX why only in the v6 case??? */
5408 if (r->rt == PF_FASTROUTE) {
5411 m0->m_flags |= M_SKIP_FIREWALL;
5412 ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL);
5416 if (TAILQ_EMPTY(&r->rpool.list)) {
5417 DPFPRINTF(PF_DEBUG_URGENT,
5418 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5422 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5424 if (!PF_AZERO(&naddr, AF_INET6))
5425 PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5427 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5429 if (!PF_AZERO(&s->rt_addr, AF_INET6))
5430 PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5431 &s->rt_addr, AF_INET6);
5432 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5442 if (pf_test6(PF_OUT, ifp, &m0, NULL) != PF_PASS)
5444 else if (m0 == NULL)
5446 if (m0->m_len < sizeof(struct ip6_hdr)) {
5447 DPFPRINTF(PF_DEBUG_URGENT,
5448 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5452 ip6 = mtod(m0, struct ip6_hdr *);
5455 if (ifp->if_flags & IFF_LOOPBACK)
5456 m0->m_flags |= M_SKIP_FIREWALL;
5459 * If the packet is too large for the outgoing interface,
5460 * send back an icmp6 error.
5462 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5463 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5464 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5465 nd6_output(ifp, ifp, m0, &dst, NULL);
5467 in6_ifstat_inc(ifp, ifs6_in_toobig);
5468 if (r->rt != PF_DUPTO)
5469 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5475 if (r->rt != PF_DUPTO)
5489 * FreeBSD supports cksum offloads for the following drivers.
5490 * em(4), fxp(4), ixgb(4), lge(4), ndis(4), nge(4), re(4),
5491 * ti(4), txp(4), xl(4)
5493 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5494 * network driver performed cksum including pseudo header, need to verify
5497 * network driver performed cksum, needs to additional pseudo header
5498 * cksum computation with partial csum_data(i.e. lack of H/W support for
5499 * pseudo header, for instance hme(4), sk(4) and possibly gem(4))
5501 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5502 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5504 * Also, set csum_data to 0xffff to force cksum validation.
5507 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5513 if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5515 if (m->m_pkthdr.len < off + len)
5520 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5521 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5522 sum = m->m_pkthdr.csum_data;
5524 ip = mtod(m, struct ip *);
5525 sum = in_pseudo(ip->ip_src.s_addr,
5526 ip->ip_dst.s_addr, htonl((u_short)len +
5527 m->m_pkthdr.csum_data + IPPROTO_TCP));
5534 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5535 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5536 sum = m->m_pkthdr.csum_data;
5538 ip = mtod(m, struct ip *);
5539 sum = in_pseudo(ip->ip_src.s_addr,
5540 ip->ip_dst.s_addr, htonl((u_short)len +
5541 m->m_pkthdr.csum_data + IPPROTO_UDP));
5549 case IPPROTO_ICMPV6:
5559 if (p == IPPROTO_ICMP) {
5564 sum = in_cksum(m, len);
5568 if (m->m_len < sizeof(struct ip))
5570 sum = in4_cksum(m, p, off, len);
5575 if (m->m_len < sizeof(struct ip6_hdr))
5577 sum = in6_cksum(m, p, off, len);
5588 KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5593 KMOD_UDPSTAT_INC(udps_badsum);
5599 KMOD_ICMPSTAT_INC(icps_checksum);
5604 case IPPROTO_ICMPV6:
5606 KMOD_ICMP6STAT_INC(icp6s_checksum);
5613 if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5614 m->m_pkthdr.csum_flags |=
5615 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5616 m->m_pkthdr.csum_data = 0xffff;
5625 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5627 struct pfi_kif *kif;
5628 u_short action, reason = 0, log = 0;
5629 struct mbuf *m = *m0;
5630 struct ip *h = NULL;
5631 struct m_tag *ipfwtag;
5632 struct pf_rule *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5633 struct pf_state *s = NULL;
5634 struct pf_ruleset *ruleset = NULL;
5636 int off, dirndx, pqid = 0;
5640 if (!V_pf_status.running)
5643 memset(&pd, 0, sizeof(pd));
5645 kif = (struct pfi_kif *)ifp->if_pf_kif;
5648 DPFPRINTF(PF_DEBUG_URGENT,
5649 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5652 if (kif->pfik_flags & PFI_IFLAG_SKIP)
5655 if (m->m_flags & M_SKIP_FIREWALL)
5658 pd.pf_mtag = pf_find_mtag(m);
5662 if (ip_divert_ptr != NULL &&
5663 ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5664 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5665 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5666 if (pd.pf_mtag == NULL &&
5667 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5671 pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5672 m_tag_delete(m, ipfwtag);
5674 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5675 m->m_flags |= M_FASTFWD_OURS;
5676 pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5678 } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5679 /* We do IP header normalization and packet reassembly here */
5683 m = *m0; /* pf_normalize messes with m0 */
5684 h = mtod(m, struct ip *);
5686 off = h->ip_hl << 2;
5687 if (off < (int)sizeof(struct ip)) {
5689 REASON_SET(&reason, PFRES_SHORT);
5694 pd.src = (struct pf_addr *)&h->ip_src;
5695 pd.dst = (struct pf_addr *)&h->ip_dst;
5696 pd.sport = pd.dport = NULL;
5697 pd.ip_sum = &h->ip_sum;
5698 pd.proto_sum = NULL;
5701 pd.sidx = (dir == PF_IN) ? 0 : 1;
5702 pd.didx = (dir == PF_IN) ? 1 : 0;
5705 pd.tot_len = ntohs(h->ip_len);
5707 /* handle fragments that didn't get reassembled by normalization */
5708 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
5709 action = pf_test_fragment(&r, dir, kif, m, h,
5720 if (!pf_pull_hdr(m, off, &th, sizeof(th),
5721 &action, &reason, AF_INET)) {
5722 log = action != PF_PASS;
5725 pd.p_len = pd.tot_len - off - (th.th_off << 2);
5726 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
5728 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
5729 if (action == PF_DROP)
5731 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
5733 if (action == PF_PASS) {
5734 if (pfsync_update_state_ptr != NULL)
5735 pfsync_update_state_ptr(s);
5739 } else if (s == NULL)
5740 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5749 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
5750 &action, &reason, AF_INET)) {
5751 log = action != PF_PASS;
5754 if (uh.uh_dport == 0 ||
5755 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
5756 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
5758 REASON_SET(&reason, PFRES_SHORT);
5761 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
5762 if (action == PF_PASS) {
5763 if (pfsync_update_state_ptr != NULL)
5764 pfsync_update_state_ptr(s);
5768 } else if (s == NULL)
5769 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5774 case IPPROTO_ICMP: {
5778 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
5779 &action, &reason, AF_INET)) {
5780 log = action != PF_PASS;
5783 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
5785 if (action == PF_PASS) {
5786 if (pfsync_update_state_ptr != NULL)
5787 pfsync_update_state_ptr(s);
5791 } else if (s == NULL)
5792 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5798 case IPPROTO_ICMPV6: {
5800 DPFPRINTF(PF_DEBUG_MISC,
5801 ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
5807 action = pf_test_state_other(&s, dir, kif, m, &pd);
5808 if (action == PF_PASS) {
5809 if (pfsync_update_state_ptr != NULL)
5810 pfsync_update_state_ptr(s);
5814 } else if (s == NULL)
5815 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5822 if (action == PF_PASS && h->ip_hl > 5 &&
5823 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
5825 REASON_SET(&reason, PFRES_IPOPTIONS);
5827 DPFPRINTF(PF_DEBUG_MISC,
5828 ("pf: dropping packet with ip options\n"));
5831 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
5833 REASON_SET(&reason, PFRES_MEMORY);
5835 if (r->rtableid >= 0)
5836 M_SETFIB(m, r->rtableid);
5839 if (action == PF_PASS && r->qid) {
5840 if (pd.pf_mtag == NULL &&
5841 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5843 REASON_SET(&reason, PFRES_MEMORY);
5845 if (pqid || (pd.tos & IPTOS_LOWDELAY))
5846 pd.pf_mtag->qid = r->pqid;
5848 pd.pf_mtag->qid = r->qid;
5849 /* add hints for ecn */
5850 pd.pf_mtag->hdr = h;
5856 * connections redirected to loopback should not match sockets
5857 * bound specifically to loopback due to security implications,
5858 * see tcp_input() and in_pcblookup_listen().
5860 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
5861 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
5862 (s->nat_rule.ptr->action == PF_RDR ||
5863 s->nat_rule.ptr->action == PF_BINAT) &&
5864 (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
5865 m->m_flags |= M_SKIP_FIREWALL;
5867 if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
5868 !PACKET_LOOPED(&pd)) {
5870 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
5871 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
5872 if (ipfwtag != NULL) {
5873 ((struct ipfw_rule_ref *)(ipfwtag+1))->info =
5874 ntohs(r->divert.port);
5875 ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
5880 m_tag_prepend(m, ipfwtag);
5881 if (m->m_flags & M_FASTFWD_OURS) {
5882 if (pd.pf_mtag == NULL &&
5883 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5885 REASON_SET(&reason, PFRES_MEMORY);
5887 DPFPRINTF(PF_DEBUG_MISC,
5888 ("pf: failed to allocate tag\n"));
5890 pd.pf_mtag->flags |= PF_FASTFWD_OURS_PRESENT;
5891 m->m_flags &= ~M_FASTFWD_OURS;
5893 ip_divert_ptr(*m0, dir == PF_IN ? DIR_IN : DIR_OUT);
5898 /* XXX: ipfw has the same behaviour! */
5900 REASON_SET(&reason, PFRES_MEMORY);
5902 DPFPRINTF(PF_DEBUG_MISC,
5903 ("pf: failed to allocate divert tag\n"));
5910 if (s != NULL && s->nat_rule.ptr != NULL &&
5911 s->nat_rule.ptr->log & PF_LOG_ALL)
5912 lr = s->nat_rule.ptr;
5915 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
5919 kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
5920 kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
5922 if (action == PF_PASS || r->action == PF_DROP) {
5923 dirndx = (dir == PF_OUT);
5924 r->packets[dirndx]++;
5925 r->bytes[dirndx] += pd.tot_len;
5927 a->packets[dirndx]++;
5928 a->bytes[dirndx] += pd.tot_len;
5931 if (s->nat_rule.ptr != NULL) {
5932 s->nat_rule.ptr->packets[dirndx]++;
5933 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
5935 if (s->src_node != NULL) {
5936 s->src_node->packets[dirndx]++;
5937 s->src_node->bytes[dirndx] += pd.tot_len;
5939 if (s->nat_src_node != NULL) {
5940 s->nat_src_node->packets[dirndx]++;
5941 s->nat_src_node->bytes[dirndx] += pd.tot_len;
5943 dirndx = (dir == s->direction) ? 0 : 1;
5944 s->packets[dirndx]++;
5945 s->bytes[dirndx] += pd.tot_len;
5948 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
5949 if (nr != NULL && r == &V_pf_default_rule)
5951 if (tr->src.addr.type == PF_ADDR_TABLE)
5952 pfr_update_stats(tr->src.addr.p.tbl,
5953 (s == NULL) ? pd.src :
5954 &s->key[(s->direction == PF_IN)]->
5955 addr[(s->direction == PF_OUT)],
5956 pd.af, pd.tot_len, dir == PF_OUT,
5957 r->action == PF_PASS, tr->src.neg);
5958 if (tr->dst.addr.type == PF_ADDR_TABLE)
5959 pfr_update_stats(tr->dst.addr.p.tbl,
5960 (s == NULL) ? pd.dst :
5961 &s->key[(s->direction == PF_IN)]->
5962 addr[(s->direction == PF_IN)],
5963 pd.af, pd.tot_len, dir == PF_OUT,
5964 r->action == PF_PASS, tr->dst.neg);
5968 case PF_SYNPROXY_DROP:
5975 /* pf_route() returns unlocked. */
5977 pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
5991 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5993 struct pfi_kif *kif;
5994 u_short action, reason = 0, log = 0;
5995 struct mbuf *m = *m0, *n = NULL;
5996 struct ip6_hdr *h = NULL;
5997 struct pf_rule *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5998 struct pf_state *s = NULL;
5999 struct pf_ruleset *ruleset = NULL;
6001 int off, terminal = 0, dirndx, rh_cnt = 0;
6005 if (!V_pf_status.running)
6008 memset(&pd, 0, sizeof(pd));
6009 pd.pf_mtag = pf_find_mtag(m);
6011 if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6014 kif = (struct pfi_kif *)ifp->if_pf_kif;
6016 DPFPRINTF(PF_DEBUG_URGENT,
6017 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6020 if (kif->pfik_flags & PFI_IFLAG_SKIP)
6025 /* We do IP header normalization and packet reassembly here */
6026 if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6030 m = *m0; /* pf_normalize messes with m0 */
6031 h = mtod(m, struct ip6_hdr *);
6035 * we do not support jumbogram yet. if we keep going, zero ip6_plen
6036 * will do something bad, so drop the packet for now.
6038 if (htons(h->ip6_plen) == 0) {
6040 REASON_SET(&reason, PFRES_NORM); /*XXX*/
6045 pd.src = (struct pf_addr *)&h->ip6_src;
6046 pd.dst = (struct pf_addr *)&h->ip6_dst;
6047 pd.sport = pd.dport = NULL;
6049 pd.proto_sum = NULL;
6051 pd.sidx = (dir == PF_IN) ? 0 : 1;
6052 pd.didx = (dir == PF_IN) ? 1 : 0;
6055 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6057 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6058 pd.proto = h->ip6_nxt;
6061 case IPPROTO_FRAGMENT:
6062 action = pf_test_fragment(&r, dir, kif, m, h,
6064 if (action == PF_DROP)
6065 REASON_SET(&reason, PFRES_FRAG);
6067 case IPPROTO_ROUTING: {
6068 struct ip6_rthdr rthdr;
6071 DPFPRINTF(PF_DEBUG_MISC,
6072 ("pf: IPv6 more than one rthdr\n"));
6074 REASON_SET(&reason, PFRES_IPOPTIONS);
6078 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6080 DPFPRINTF(PF_DEBUG_MISC,
6081 ("pf: IPv6 short rthdr\n"));
6083 REASON_SET(&reason, PFRES_SHORT);
6087 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6088 DPFPRINTF(PF_DEBUG_MISC,
6089 ("pf: IPv6 rthdr0\n"));
6091 REASON_SET(&reason, PFRES_IPOPTIONS);
6098 case IPPROTO_HOPOPTS:
6099 case IPPROTO_DSTOPTS: {
6100 /* get next header and header length */
6101 struct ip6_ext opt6;
6103 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6104 NULL, &reason, pd.af)) {
6105 DPFPRINTF(PF_DEBUG_MISC,
6106 ("pf: IPv6 short opt\n"));
6111 if (pd.proto == IPPROTO_AH)
6112 off += (opt6.ip6e_len + 2) * 4;
6114 off += (opt6.ip6e_len + 1) * 8;
6115 pd.proto = opt6.ip6e_nxt;
6116 /* goto the next header */
6123 } while (!terminal);
6125 /* if there's no routing header, use unmodified mbuf for checksumming */
6135 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6136 &action, &reason, AF_INET6)) {
6137 log = action != PF_PASS;
6140 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6141 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6142 if (action == PF_DROP)
6144 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6146 if (action == PF_PASS) {
6147 if (pfsync_update_state_ptr != NULL)
6148 pfsync_update_state_ptr(s);
6152 } else if (s == NULL)
6153 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6162 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6163 &action, &reason, AF_INET6)) {
6164 log = action != PF_PASS;
6167 if (uh.uh_dport == 0 ||
6168 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6169 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6171 REASON_SET(&reason, PFRES_SHORT);
6174 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6175 if (action == PF_PASS) {
6176 if (pfsync_update_state_ptr != NULL)
6177 pfsync_update_state_ptr(s);
6181 } else if (s == NULL)
6182 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6187 case IPPROTO_ICMP: {
6189 DPFPRINTF(PF_DEBUG_MISC,
6190 ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6194 case IPPROTO_ICMPV6: {
6195 struct icmp6_hdr ih;
6198 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6199 &action, &reason, AF_INET6)) {
6200 log = action != PF_PASS;
6203 action = pf_test_state_icmp(&s, dir, kif,
6204 m, off, h, &pd, &reason);
6205 if (action == PF_PASS) {
6206 if (pfsync_update_state_ptr != NULL)
6207 pfsync_update_state_ptr(s);
6211 } else if (s == NULL)
6212 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6218 action = pf_test_state_other(&s, dir, kif, m, &pd);
6219 if (action == PF_PASS) {
6220 if (pfsync_update_state_ptr != NULL)
6221 pfsync_update_state_ptr(s);
6225 } else if (s == NULL)
6226 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6238 /* handle dangerous IPv6 extension headers. */
6239 if (action == PF_PASS && rh_cnt &&
6240 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6242 REASON_SET(&reason, PFRES_IPOPTIONS);
6244 DPFPRINTF(PF_DEBUG_MISC,
6245 ("pf: dropping packet with dangerous v6 headers\n"));
6248 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6250 REASON_SET(&reason, PFRES_MEMORY);
6252 if (r->rtableid >= 0)
6253 M_SETFIB(m, r->rtableid);
6256 if (action == PF_PASS && r->qid) {
6257 if (pd.pf_mtag == NULL &&
6258 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6260 REASON_SET(&reason, PFRES_MEMORY);
6262 if (pd.tos & IPTOS_LOWDELAY)
6263 pd.pf_mtag->qid = r->pqid;
6265 pd.pf_mtag->qid = r->qid;
6266 /* add hints for ecn */
6267 pd.pf_mtag->hdr = h;
6271 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6272 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6273 (s->nat_rule.ptr->action == PF_RDR ||
6274 s->nat_rule.ptr->action == PF_BINAT) &&
6275 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6276 m->m_flags |= M_SKIP_FIREWALL;
6278 /* XXX: Anybody working on it?! */
6280 printf("pf: divert(9) is not supported for IPv6\n");
6285 if (s != NULL && s->nat_rule.ptr != NULL &&
6286 s->nat_rule.ptr->log & PF_LOG_ALL)
6287 lr = s->nat_rule.ptr;
6290 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6294 kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6295 kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
6297 if (action == PF_PASS || r->action == PF_DROP) {
6298 dirndx = (dir == PF_OUT);
6299 r->packets[dirndx]++;
6300 r->bytes[dirndx] += pd.tot_len;
6302 a->packets[dirndx]++;
6303 a->bytes[dirndx] += pd.tot_len;
6306 if (s->nat_rule.ptr != NULL) {
6307 s->nat_rule.ptr->packets[dirndx]++;
6308 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6310 if (s->src_node != NULL) {
6311 s->src_node->packets[dirndx]++;
6312 s->src_node->bytes[dirndx] += pd.tot_len;
6314 if (s->nat_src_node != NULL) {
6315 s->nat_src_node->packets[dirndx]++;
6316 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6318 dirndx = (dir == s->direction) ? 0 : 1;
6319 s->packets[dirndx]++;
6320 s->bytes[dirndx] += pd.tot_len;
6323 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6324 if (nr != NULL && r == &V_pf_default_rule)
6326 if (tr->src.addr.type == PF_ADDR_TABLE)
6327 pfr_update_stats(tr->src.addr.p.tbl,
6328 (s == NULL) ? pd.src :
6329 &s->key[(s->direction == PF_IN)]->addr[0],
6330 pd.af, pd.tot_len, dir == PF_OUT,
6331 r->action == PF_PASS, tr->src.neg);
6332 if (tr->dst.addr.type == PF_ADDR_TABLE)
6333 pfr_update_stats(tr->dst.addr.p.tbl,
6334 (s == NULL) ? pd.dst :
6335 &s->key[(s->direction == PF_IN)]->addr[1],
6336 pd.af, pd.tot_len, dir == PF_OUT,
6337 r->action == PF_PASS, tr->dst.neg);
6341 case PF_SYNPROXY_DROP:
6348 /* pf_route6() returns unlocked. */
6350 pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);