2 * SPDX-License-Identifier: BSD-2-Clause
4 * Copyright (c) 2001 Daniel Hartmeier
5 * Copyright (c) 2002 - 2008 Henning Brauer
6 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
13 * - Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * - Redistributions in binary form must reproduce the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer in the documentation and/or other materials provided
18 * with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
28 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
30 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31 * POSSIBILITY OF SUCH DAMAGE.
33 * Effort sponsored in part by the Defense Advanced Research Projects
34 * Agency (DARPA) and Air Force Research Laboratory, Air Force
35 * Materiel Command, USAF, under agreement number F30602-01-2-0537.
37 * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
44 #include "opt_inet6.h"
48 #include <sys/param.h>
50 #include <sys/endian.h>
52 #include <sys/interrupt.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/limits.h>
58 #include <sys/random.h>
59 #include <sys/refcount.h>
60 #include <sys/socket.h>
61 #include <sys/sysctl.h>
62 #include <sys/taskqueue.h>
63 #include <sys/ucred.h>
66 #include <net/if_var.h>
67 #include <net/if_types.h>
68 #include <net/if_vlan_var.h>
69 #include <net/route.h>
70 #include <net/radix_mpath.h>
74 #include <net/pfvar.h>
75 #include <net/if_pflog.h>
76 #include <net/if_pfsync.h>
78 #include <netinet/in_pcb.h>
79 #include <netinet/in_var.h>
80 #include <netinet/in_fib.h>
81 #include <netinet/ip.h>
82 #include <netinet/ip_fw.h>
83 #include <netinet/ip_icmp.h>
84 #include <netinet/icmp_var.h>
85 #include <netinet/ip_var.h>
86 #include <netinet/tcp.h>
87 #include <netinet/tcp_fsm.h>
88 #include <netinet/tcp_seq.h>
89 #include <netinet/tcp_timer.h>
90 #include <netinet/tcp_var.h>
91 #include <netinet/udp.h>
92 #include <netinet/udp_var.h>
94 #include <netpfil/ipfw/ip_fw_private.h> /* XXX: only for DIR_IN/DIR_OUT */
97 #include <netinet/ip6.h>
98 #include <netinet/icmp6.h>
99 #include <netinet6/nd6.h>
100 #include <netinet6/ip6_var.h>
101 #include <netinet6/in6_pcb.h>
102 #include <netinet6/in6_fib.h>
103 #include <netinet6/scope6_var.h>
106 #include <machine/in_cksum.h>
107 #include <security/mac/mac_framework.h>
109 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
116 VNET_DEFINE(struct pf_altqqueue, pf_altqs[2]);
117 VNET_DEFINE(struct pf_palist, pf_pabuf);
118 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active);
119 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive);
120 VNET_DEFINE(struct pf_kstatus, pf_status);
122 VNET_DEFINE(u_int32_t, ticket_altqs_active);
123 VNET_DEFINE(u_int32_t, ticket_altqs_inactive);
124 VNET_DEFINE(int, altqs_inactive_open);
125 VNET_DEFINE(u_int32_t, ticket_pabuf);
127 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx);
128 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx)
129 VNET_DEFINE(u_char, pf_tcp_secret[16]);
130 #define V_pf_tcp_secret VNET(pf_tcp_secret)
131 VNET_DEFINE(int, pf_tcp_secret_init);
132 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init)
133 VNET_DEFINE(int, pf_tcp_iss_off);
134 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off)
135 VNET_DECLARE(int, pf_vnet_active);
136 #define V_pf_vnet_active VNET(pf_vnet_active)
138 static VNET_DEFINE(uint32_t, pf_purge_idx);
139 #define V_pf_purge_idx VNET(pf_purge_idx)
142 * Queue for pf_intr() sends.
144 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
145 struct pf_send_entry {
146 STAILQ_ENTRY(pf_send_entry) pfse_next;
161 STAILQ_HEAD(pf_send_head, pf_send_entry);
162 static VNET_DEFINE(struct pf_send_head, pf_sendqueue);
163 #define V_pf_sendqueue VNET(pf_sendqueue)
165 static struct mtx pf_sendqueue_mtx;
166 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
167 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx)
168 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx)
171 * Queue for pf_overload_task() tasks.
173 struct pf_overload_entry {
174 SLIST_ENTRY(pf_overload_entry) next;
178 struct pf_rule *rule;
181 SLIST_HEAD(pf_overload_head, pf_overload_entry);
182 static VNET_DEFINE(struct pf_overload_head, pf_overloadqueue);
183 #define V_pf_overloadqueue VNET(pf_overloadqueue)
184 static VNET_DEFINE(struct task, pf_overloadtask);
185 #define V_pf_overloadtask VNET(pf_overloadtask)
187 static struct mtx pf_overloadqueue_mtx;
188 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
189 "pf overload/flush queue", MTX_DEF);
190 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx)
191 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx)
193 VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules);
194 struct mtx pf_unlnkdrules_mtx;
195 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
198 static VNET_DEFINE(uma_zone_t, pf_sources_z);
199 #define V_pf_sources_z VNET(pf_sources_z)
200 uma_zone_t pf_mtag_z;
201 VNET_DEFINE(uma_zone_t, pf_state_z);
202 VNET_DEFINE(uma_zone_t, pf_state_key_z);
204 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
205 #define PFID_CPUBITS 8
206 #define PFID_CPUSHIFT (sizeof(uint64_t) * NBBY - PFID_CPUBITS)
207 #define PFID_CPUMASK ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT)
208 #define PFID_MAXID (~PFID_CPUMASK)
209 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU);
211 static void pf_src_tree_remove_state(struct pf_state *);
212 static void pf_init_threshold(struct pf_threshold *, u_int32_t,
214 static void pf_add_threshold(struct pf_threshold *);
215 static int pf_check_threshold(struct pf_threshold *);
217 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
218 u_int16_t *, u_int16_t *, struct pf_addr *,
219 u_int16_t, u_int8_t, sa_family_t);
220 static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
221 struct tcphdr *, struct pf_state_peer *);
222 static void pf_change_icmp(struct pf_addr *, u_int16_t *,
223 struct pf_addr *, struct pf_addr *, u_int16_t,
224 u_int16_t *, u_int16_t *, u_int16_t *,
225 u_int16_t *, u_int8_t, sa_family_t);
226 static void pf_send_tcp(struct mbuf *,
227 const struct pf_rule *, sa_family_t,
228 const struct pf_addr *, const struct pf_addr *,
229 u_int16_t, u_int16_t, u_int32_t, u_int32_t,
230 u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
231 u_int16_t, struct ifnet *);
232 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
233 sa_family_t, struct pf_rule *);
234 static void pf_detach_state(struct pf_state *);
235 static int pf_state_key_attach(struct pf_state_key *,
236 struct pf_state_key *, struct pf_state *);
237 static void pf_state_key_detach(struct pf_state *, int);
238 static int pf_state_key_ctor(void *, int, void *, int);
239 static u_int32_t pf_tcp_iss(struct pf_pdesc *);
240 static int pf_test_rule(struct pf_rule **, struct pf_state **,
241 int, struct pfi_kif *, struct mbuf *, int,
242 struct pf_pdesc *, struct pf_rule **,
243 struct pf_ruleset **, struct inpcb *);
244 static int pf_create_state(struct pf_rule *, struct pf_rule *,
245 struct pf_rule *, struct pf_pdesc *,
246 struct pf_src_node *, struct pf_state_key *,
247 struct pf_state_key *, struct mbuf *, int,
248 u_int16_t, u_int16_t, int *, struct pfi_kif *,
249 struct pf_state **, int, u_int16_t, u_int16_t,
251 static int pf_test_fragment(struct pf_rule **, int,
252 struct pfi_kif *, struct mbuf *, void *,
253 struct pf_pdesc *, struct pf_rule **,
254 struct pf_ruleset **);
255 static int pf_tcp_track_full(struct pf_state_peer *,
256 struct pf_state_peer *, struct pf_state **,
257 struct pfi_kif *, struct mbuf *, int,
258 struct pf_pdesc *, u_short *, int *);
259 static int pf_tcp_track_sloppy(struct pf_state_peer *,
260 struct pf_state_peer *, struct pf_state **,
261 struct pf_pdesc *, u_short *);
262 static int pf_test_state_tcp(struct pf_state **, int,
263 struct pfi_kif *, struct mbuf *, int,
264 void *, struct pf_pdesc *, u_short *);
265 static int pf_test_state_udp(struct pf_state **, int,
266 struct pfi_kif *, struct mbuf *, int,
267 void *, struct pf_pdesc *);
268 static int pf_test_state_icmp(struct pf_state **, int,
269 struct pfi_kif *, struct mbuf *, int,
270 void *, struct pf_pdesc *, u_short *);
271 static int pf_test_state_other(struct pf_state **, int,
272 struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
273 static u_int8_t pf_get_wscale(struct mbuf *, int, u_int16_t,
275 static u_int16_t pf_get_mss(struct mbuf *, int, u_int16_t,
277 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t,
279 static int pf_check_proto_cksum(struct mbuf *, int, int,
280 u_int8_t, sa_family_t);
281 static void pf_print_state_parts(struct pf_state *,
282 struct pf_state_key *, struct pf_state_key *);
283 static int pf_addr_wrap_neq(struct pf_addr_wrap *,
284 struct pf_addr_wrap *);
285 static struct pf_state *pf_find_state(struct pfi_kif *,
286 struct pf_state_key_cmp *, u_int);
287 static int pf_src_connlimit(struct pf_state **);
288 static void pf_overload_task(void *v, int pending);
289 static int pf_insert_src_node(struct pf_src_node **,
290 struct pf_rule *, struct pf_addr *, sa_family_t);
291 static u_int pf_purge_expired_states(u_int, int);
292 static void pf_purge_unlinked_rules(void);
293 static int pf_mtag_uminit(void *, int, int);
294 static void pf_mtag_free(struct m_tag *);
296 static void pf_route(struct mbuf **, struct pf_rule *, int,
297 struct ifnet *, struct pf_state *,
301 static void pf_change_a6(struct pf_addr *, u_int16_t *,
302 struct pf_addr *, u_int8_t);
303 static void pf_route6(struct mbuf **, struct pf_rule *, int,
304 struct ifnet *, struct pf_state *,
308 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
310 extern int pf_end_threads;
311 extern struct proc *pf_purge_proc;
313 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
315 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \
316 (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
318 #define STATE_LOOKUP(i, k, d, s, pd) \
320 (s) = pf_find_state((i), (k), (d)); \
323 if (PACKET_LOOPED(pd)) \
325 if ((d) == PF_OUT && \
326 (((s)->rule.ptr->rt == PF_ROUTETO && \
327 (s)->rule.ptr->direction == PF_OUT) || \
328 ((s)->rule.ptr->rt == PF_REPLYTO && \
329 (s)->rule.ptr->direction == PF_IN)) && \
330 (s)->rt_kif != NULL && \
331 (s)->rt_kif != (i)) \
335 #define BOUND_IFACE(r, k) \
336 ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
338 #define STATE_INC_COUNTERS(s) \
340 counter_u64_add(s->rule.ptr->states_cur, 1); \
341 counter_u64_add(s->rule.ptr->states_tot, 1); \
342 if (s->anchor.ptr != NULL) { \
343 counter_u64_add(s->anchor.ptr->states_cur, 1); \
344 counter_u64_add(s->anchor.ptr->states_tot, 1); \
346 if (s->nat_rule.ptr != NULL) { \
347 counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
348 counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
352 #define STATE_DEC_COUNTERS(s) \
354 if (s->nat_rule.ptr != NULL) \
355 counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
356 if (s->anchor.ptr != NULL) \
357 counter_u64_add(s->anchor.ptr->states_cur, -1); \
358 counter_u64_add(s->rule.ptr->states_cur, -1); \
361 static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
362 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
363 VNET_DEFINE(struct pf_idhash *, pf_idhash);
364 VNET_DEFINE(struct pf_srchash *, pf_srchash);
366 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)");
369 u_long pf_srchashmask;
370 static u_long pf_hashsize;
371 static u_long pf_srchashsize;
372 u_long pf_ioctl_maxcount = 65535;
374 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
375 &pf_hashsize, 0, "Size of pf(4) states hashtable");
376 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
377 &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
378 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RDTUN,
379 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call");
381 VNET_DEFINE(void *, pf_swi_cookie);
383 VNET_DEFINE(uint32_t, pf_hashseed);
384 #define V_pf_hashseed VNET(pf_hashseed)
387 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
393 if (a->addr32[0] > b->addr32[0])
395 if (a->addr32[0] < b->addr32[0])
401 if (a->addr32[3] > b->addr32[3])
403 if (a->addr32[3] < b->addr32[3])
405 if (a->addr32[2] > b->addr32[2])
407 if (a->addr32[2] < b->addr32[2])
409 if (a->addr32[1] > b->addr32[1])
411 if (a->addr32[1] < b->addr32[1])
413 if (a->addr32[0] > b->addr32[0])
415 if (a->addr32[0] < b->addr32[0])
420 panic("%s: unknown address family %u", __func__, af);
425 static __inline uint32_t
426 pf_hashkey(struct pf_state_key *sk)
430 h = murmur3_32_hash32((uint32_t *)sk,
431 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
434 return (h & pf_hashmask);
437 static __inline uint32_t
438 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
444 h = murmur3_32_hash32((uint32_t *)&addr->v4,
445 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
448 h = murmur3_32_hash32((uint32_t *)&addr->v6,
449 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
452 panic("%s: unknown address family %u", __func__, af);
455 return (h & pf_srchashmask);
460 pf_state_hash(struct pf_state *s)
462 u_int32_t hv = (intptr_t)s / sizeof(*s);
464 hv ^= crc32(&s->src, sizeof(s->src));
465 hv ^= crc32(&s->dst, sizeof(s->dst));
474 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
479 dst->addr32[0] = src->addr32[0];
483 dst->addr32[0] = src->addr32[0];
484 dst->addr32[1] = src->addr32[1];
485 dst->addr32[2] = src->addr32[2];
486 dst->addr32[3] = src->addr32[3];
493 pf_init_threshold(struct pf_threshold *threshold,
494 u_int32_t limit, u_int32_t seconds)
496 threshold->limit = limit * PF_THRESHOLD_MULT;
497 threshold->seconds = seconds;
498 threshold->count = 0;
499 threshold->last = time_uptime;
503 pf_add_threshold(struct pf_threshold *threshold)
505 u_int32_t t = time_uptime, diff = t - threshold->last;
507 if (diff >= threshold->seconds)
508 threshold->count = 0;
510 threshold->count -= threshold->count * diff /
512 threshold->count += PF_THRESHOLD_MULT;
517 pf_check_threshold(struct pf_threshold *threshold)
519 return (threshold->count > threshold->limit);
523 pf_src_connlimit(struct pf_state **state)
525 struct pf_overload_entry *pfoe;
528 PF_STATE_LOCK_ASSERT(*state);
530 (*state)->src_node->conn++;
531 (*state)->src.tcp_est = 1;
532 pf_add_threshold(&(*state)->src_node->conn_rate);
534 if ((*state)->rule.ptr->max_src_conn &&
535 (*state)->rule.ptr->max_src_conn <
536 (*state)->src_node->conn) {
537 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
541 if ((*state)->rule.ptr->max_src_conn_rate.limit &&
542 pf_check_threshold(&(*state)->src_node->conn_rate)) {
543 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
550 /* Kill this state. */
551 (*state)->timeout = PFTM_PURGE;
552 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
554 if ((*state)->rule.ptr->overload_tbl == NULL)
557 /* Schedule overloading and flushing task. */
558 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
560 return (1); /* too bad :( */
562 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
563 pfoe->af = (*state)->key[PF_SK_WIRE]->af;
564 pfoe->rule = (*state)->rule.ptr;
565 pfoe->dir = (*state)->direction;
567 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
568 PF_OVERLOADQ_UNLOCK();
569 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
575 pf_overload_task(void *v, int pending)
577 struct pf_overload_head queue;
579 struct pf_overload_entry *pfoe, *pfoe1;
582 CURVNET_SET((struct vnet *)v);
585 queue = V_pf_overloadqueue;
586 SLIST_INIT(&V_pf_overloadqueue);
587 PF_OVERLOADQ_UNLOCK();
589 bzero(&p, sizeof(p));
590 SLIST_FOREACH(pfoe, &queue, next) {
591 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
592 if (V_pf_status.debug >= PF_DEBUG_MISC) {
593 printf("%s: blocking address ", __func__);
594 pf_print_host(&pfoe->addr, 0, pfoe->af);
598 p.pfra_af = pfoe->af;
603 p.pfra_ip4addr = pfoe->addr.v4;
609 p.pfra_ip6addr = pfoe->addr.v6;
615 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
620 * Remove those entries, that don't need flushing.
622 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
623 if (pfoe->rule->flush == 0) {
624 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
625 free(pfoe, M_PFTEMP);
628 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
630 /* If nothing to flush, return. */
631 if (SLIST_EMPTY(&queue)) {
636 for (int i = 0; i <= pf_hashmask; i++) {
637 struct pf_idhash *ih = &V_pf_idhash[i];
638 struct pf_state_key *sk;
642 LIST_FOREACH(s, &ih->states, entry) {
643 sk = s->key[PF_SK_WIRE];
644 SLIST_FOREACH(pfoe, &queue, next)
645 if (sk->af == pfoe->af &&
646 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
647 pfoe->rule == s->rule.ptr) &&
648 ((pfoe->dir == PF_OUT &&
649 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
650 (pfoe->dir == PF_IN &&
651 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
652 s->timeout = PFTM_PURGE;
653 s->src.state = s->dst.state = TCPS_CLOSED;
657 PF_HASHROW_UNLOCK(ih);
659 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
660 free(pfoe, M_PFTEMP);
661 if (V_pf_status.debug >= PF_DEBUG_MISC)
662 printf("%s: %u states killed", __func__, killed);
668 * Can return locked on failure, so that we can consistently
669 * allocate and insert a new one.
672 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
675 struct pf_srchash *sh;
676 struct pf_src_node *n;
678 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
680 sh = &V_pf_srchash[pf_hashsrc(src, af)];
682 LIST_FOREACH(n, &sh->nodes, entry)
683 if (n->rule.ptr == rule && n->af == af &&
684 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
685 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
689 PF_HASHROW_UNLOCK(sh);
690 } else if (returnlocked == 0)
691 PF_HASHROW_UNLOCK(sh);
697 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
698 struct pf_addr *src, sa_family_t af)
701 KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
702 rule->rpool.opts & PF_POOL_STICKYADDR),
703 ("%s for non-tracking rule %p", __func__, rule));
706 *sn = pf_find_src_node(src, rule, af, 1);
709 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
711 PF_HASHROW_ASSERT(sh);
713 if (!rule->max_src_nodes ||
714 counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
715 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
717 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
720 PF_HASHROW_UNLOCK(sh);
724 pf_init_threshold(&(*sn)->conn_rate,
725 rule->max_src_conn_rate.limit,
726 rule->max_src_conn_rate.seconds);
729 (*sn)->rule.ptr = rule;
730 PF_ACPY(&(*sn)->addr, src, af);
731 LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
732 (*sn)->creation = time_uptime;
733 (*sn)->ruletype = rule->action;
735 if ((*sn)->rule.ptr != NULL)
736 counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
737 PF_HASHROW_UNLOCK(sh);
738 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
740 if (rule->max_src_states &&
741 (*sn)->states >= rule->max_src_states) {
742 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
751 pf_unlink_src_node(struct pf_src_node *src)
754 PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]);
755 LIST_REMOVE(src, entry);
757 counter_u64_add(src->rule.ptr->src_nodes, -1);
761 pf_free_src_nodes(struct pf_src_node_list *head)
763 struct pf_src_node *sn, *tmp;
766 LIST_FOREACH_SAFE(sn, head, entry, tmp) {
767 uma_zfree(V_pf_sources_z, sn);
771 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
780 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
781 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
785 /* Per-vnet data storage structures initialization. */
789 struct pf_keyhash *kh;
790 struct pf_idhash *ih;
791 struct pf_srchash *sh;
794 if (pf_hashsize == 0 || !powerof2(pf_hashsize))
795 pf_hashsize = PF_HASHSIZ;
796 if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
797 pf_srchashsize = PF_SRCHASHSIZ;
799 V_pf_hashseed = arc4random();
801 /* States and state keys storage. */
802 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
803 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
804 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
805 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
806 uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
808 V_pf_state_key_z = uma_zcreate("pf state keys",
809 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
812 V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash),
813 M_PFHASH, M_NOWAIT | M_ZERO);
814 V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash),
815 M_PFHASH, M_NOWAIT | M_ZERO);
816 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) {
817 printf("pf: Unable to allocate memory for "
818 "state_hashsize %lu.\n", pf_hashsize);
820 free(V_pf_keyhash, M_PFHASH);
821 free(V_pf_idhash, M_PFHASH);
823 pf_hashsize = PF_HASHSIZ;
824 V_pf_keyhash = mallocarray(pf_hashsize,
825 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO);
826 V_pf_idhash = mallocarray(pf_hashsize,
827 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO);
830 pf_hashmask = pf_hashsize - 1;
831 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
833 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
834 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
838 V_pf_sources_z = uma_zcreate("pf source nodes",
839 sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
841 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
842 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
843 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
845 V_pf_srchash = mallocarray(pf_srchashsize,
846 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO);
847 if (V_pf_srchash == NULL) {
848 printf("pf: Unable to allocate memory for "
849 "source_hashsize %lu.\n", pf_srchashsize);
851 pf_srchashsize = PF_SRCHASHSIZ;
852 V_pf_srchash = mallocarray(pf_srchashsize,
853 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO);
856 pf_srchashmask = pf_srchashsize - 1;
857 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
858 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
861 TAILQ_INIT(&V_pf_altqs[0]);
862 TAILQ_INIT(&V_pf_altqs[1]);
863 TAILQ_INIT(&V_pf_pabuf);
864 V_pf_altqs_active = &V_pf_altqs[0];
865 V_pf_altqs_inactive = &V_pf_altqs[1];
867 /* Send & overload+flush queues. */
868 STAILQ_INIT(&V_pf_sendqueue);
869 SLIST_INIT(&V_pf_overloadqueue);
870 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
872 /* Unlinked, but may be referenced rules. */
873 TAILQ_INIT(&V_pf_unlinked_rules);
880 uma_zdestroy(pf_mtag_z);
886 struct pf_keyhash *kh;
887 struct pf_idhash *ih;
888 struct pf_srchash *sh;
889 struct pf_send_entry *pfse, *next;
892 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
894 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
896 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
898 mtx_destroy(&kh->lock);
899 mtx_destroy(&ih->lock);
901 free(V_pf_keyhash, M_PFHASH);
902 free(V_pf_idhash, M_PFHASH);
904 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
905 KASSERT(LIST_EMPTY(&sh->nodes),
906 ("%s: source node hash not empty", __func__));
907 mtx_destroy(&sh->lock);
909 free(V_pf_srchash, M_PFHASH);
911 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
912 m_freem(pfse->pfse_m);
913 free(pfse, M_PFTEMP);
916 uma_zdestroy(V_pf_sources_z);
917 uma_zdestroy(V_pf_state_z);
918 uma_zdestroy(V_pf_state_key_z);
922 pf_mtag_uminit(void *mem, int size, int how)
926 t = (struct m_tag *)mem;
927 t->m_tag_cookie = MTAG_ABI_COMPAT;
928 t->m_tag_id = PACKET_TAG_PF;
929 t->m_tag_len = sizeof(struct pf_mtag);
930 t->m_tag_free = pf_mtag_free;
936 pf_mtag_free(struct m_tag *t)
939 uma_zfree(pf_mtag_z, t);
943 pf_get_mtag(struct mbuf *m)
947 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
948 return ((struct pf_mtag *)(mtag + 1));
950 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
953 bzero(mtag + 1, sizeof(struct pf_mtag));
954 m_tag_prepend(m, mtag);
956 return ((struct pf_mtag *)(mtag + 1));
960 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
963 struct pf_keyhash *khs, *khw, *kh;
964 struct pf_state_key *sk, *cur;
965 struct pf_state *si, *olds = NULL;
968 KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
969 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
970 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
973 * We need to lock hash slots of both keys. To avoid deadlock
974 * we always lock the slot with lower address first. Unlock order
977 * We also need to lock ID hash slot before dropping key
978 * locks. On success we return with ID hash slot locked.
982 khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
983 PF_HASHROW_LOCK(khs);
985 khs = &V_pf_keyhash[pf_hashkey(sks)];
986 khw = &V_pf_keyhash[pf_hashkey(skw)];
988 PF_HASHROW_LOCK(khs);
989 } else if (khs < khw) {
990 PF_HASHROW_LOCK(khs);
991 PF_HASHROW_LOCK(khw);
993 PF_HASHROW_LOCK(khw);
994 PF_HASHROW_LOCK(khs);
998 #define KEYS_UNLOCK() do { \
1000 PF_HASHROW_UNLOCK(khs); \
1001 PF_HASHROW_UNLOCK(khw); \
1003 PF_HASHROW_UNLOCK(khs); \
1007 * First run: start with wire key.
1014 LIST_FOREACH(cur, &kh->keys, entry)
1015 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
1019 /* Key exists. Check for same kif, if none, add to key. */
1020 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
1021 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
1023 PF_HASHROW_LOCK(ih);
1024 if (si->kif == s->kif &&
1025 si->direction == s->direction) {
1026 if (sk->proto == IPPROTO_TCP &&
1027 si->src.state >= TCPS_FIN_WAIT_2 &&
1028 si->dst.state >= TCPS_FIN_WAIT_2) {
1030 * New state matches an old >FIN_WAIT_2
1031 * state. We can't drop key hash locks,
1032 * thus we can't unlink it properly.
1034 * As a workaround we drop it into
1035 * TCPS_CLOSED state, schedule purge
1036 * ASAP and push it into the very end
1037 * of the slot TAILQ, so that it won't
1038 * conflict with our new state.
1040 si->src.state = si->dst.state =
1042 si->timeout = PFTM_PURGE;
1045 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1046 printf("pf: %s key attach "
1048 (idx == PF_SK_WIRE) ?
1051 pf_print_state_parts(s,
1052 (idx == PF_SK_WIRE) ?
1054 (idx == PF_SK_STACK) ?
1056 printf(", existing: ");
1057 pf_print_state_parts(si,
1058 (idx == PF_SK_WIRE) ?
1060 (idx == PF_SK_STACK) ?
1064 PF_HASHROW_UNLOCK(ih);
1066 uma_zfree(V_pf_state_key_z, sk);
1067 if (idx == PF_SK_STACK)
1069 return (EEXIST); /* collision! */
1072 PF_HASHROW_UNLOCK(ih);
1074 uma_zfree(V_pf_state_key_z, sk);
1077 LIST_INSERT_HEAD(&kh->keys, sk, entry);
1082 /* List is sorted, if-bound states before floating. */
1083 if (s->kif == V_pfi_all)
1084 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1086 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1089 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1090 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1096 * Attach done. See how should we (or should not?)
1097 * attach a second key.
1100 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1104 } else if (sks != NULL) {
1106 * Continue attaching with stack key.
1118 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1119 ("%s failure", __func__));
1126 pf_detach_state(struct pf_state *s)
1128 struct pf_state_key *sks = s->key[PF_SK_STACK];
1129 struct pf_keyhash *kh;
1132 kh = &V_pf_keyhash[pf_hashkey(sks)];
1133 PF_HASHROW_LOCK(kh);
1134 if (s->key[PF_SK_STACK] != NULL)
1135 pf_state_key_detach(s, PF_SK_STACK);
1137 * If both point to same key, then we are done.
1139 if (sks == s->key[PF_SK_WIRE]) {
1140 pf_state_key_detach(s, PF_SK_WIRE);
1141 PF_HASHROW_UNLOCK(kh);
1144 PF_HASHROW_UNLOCK(kh);
1147 if (s->key[PF_SK_WIRE] != NULL) {
1148 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1149 PF_HASHROW_LOCK(kh);
1150 if (s->key[PF_SK_WIRE] != NULL)
1151 pf_state_key_detach(s, PF_SK_WIRE);
1152 PF_HASHROW_UNLOCK(kh);
1157 pf_state_key_detach(struct pf_state *s, int idx)
1159 struct pf_state_key *sk = s->key[idx];
1161 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1163 PF_HASHROW_ASSERT(kh);
1165 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1168 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1169 LIST_REMOVE(sk, entry);
1170 uma_zfree(V_pf_state_key_z, sk);
1175 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1177 struct pf_state_key *sk = mem;
1179 bzero(sk, sizeof(struct pf_state_key_cmp));
1180 TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1181 TAILQ_INIT(&sk->states[PF_SK_STACK]);
1186 struct pf_state_key *
1187 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1188 struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1190 struct pf_state_key *sk;
1192 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1196 PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1197 PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1198 sk->port[pd->sidx] = sport;
1199 sk->port[pd->didx] = dport;
1200 sk->proto = pd->proto;
1206 struct pf_state_key *
1207 pf_state_key_clone(struct pf_state_key *orig)
1209 struct pf_state_key *sk;
1211 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1215 bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1221 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1222 struct pf_state_key *sks, struct pf_state *s)
1224 struct pf_idhash *ih;
1225 struct pf_state *cur;
1228 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1229 ("%s: sks not pristine", __func__));
1230 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1231 ("%s: skw not pristine", __func__));
1232 KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1236 if (s->id == 0 && s->creatorid == 0) {
1237 /* XXX: should be atomic, but probability of collision low */
1238 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1239 V_pf_stateid[curcpu] = 1;
1240 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1241 s->id = htobe64(s->id);
1242 s->creatorid = V_pf_status.hostid;
1245 /* Returns with ID locked on success. */
1246 if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1249 ih = &V_pf_idhash[PF_IDHASH(s)];
1250 PF_HASHROW_ASSERT(ih);
1251 LIST_FOREACH(cur, &ih->states, entry)
1252 if (cur->id == s->id && cur->creatorid == s->creatorid)
1256 PF_HASHROW_UNLOCK(ih);
1257 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1258 printf("pf: state ID collision: "
1259 "id: %016llx creatorid: %08x\n",
1260 (unsigned long long)be64toh(s->id),
1261 ntohl(s->creatorid));
1266 LIST_INSERT_HEAD(&ih->states, s, entry);
1267 /* One for keys, one for ID hash. */
1268 refcount_init(&s->refs, 2);
1270 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1271 if (pfsync_insert_state_ptr != NULL)
1272 pfsync_insert_state_ptr(s);
1274 /* Returns locked. */
1279 * Find state by ID: returns with locked row on success.
1282 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1284 struct pf_idhash *ih;
1287 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1289 ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1291 PF_HASHROW_LOCK(ih);
1292 LIST_FOREACH(s, &ih->states, entry)
1293 if (s->id == id && s->creatorid == creatorid)
1297 PF_HASHROW_UNLOCK(ih);
1303 * Find state by key.
1304 * Returns with ID hash slot locked on success.
1306 static struct pf_state *
1307 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1309 struct pf_keyhash *kh;
1310 struct pf_state_key *sk;
1314 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1316 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1318 PF_HASHROW_LOCK(kh);
1319 LIST_FOREACH(sk, &kh->keys, entry)
1320 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1323 PF_HASHROW_UNLOCK(kh);
1327 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1329 /* List is sorted, if-bound states before floating ones. */
1330 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1331 if (s->kif == V_pfi_all || s->kif == kif) {
1333 PF_HASHROW_UNLOCK(kh);
1334 if (s->timeout >= PFTM_MAX) {
1336 * State is either being processed by
1337 * pf_unlink_state() in an other thread, or
1338 * is scheduled for immediate expiry.
1345 PF_HASHROW_UNLOCK(kh);
1351 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1353 struct pf_keyhash *kh;
1354 struct pf_state_key *sk;
1355 struct pf_state *s, *ret = NULL;
1358 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1360 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1362 PF_HASHROW_LOCK(kh);
1363 LIST_FOREACH(sk, &kh->keys, entry)
1364 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1367 PF_HASHROW_UNLOCK(kh);
1382 panic("%s: dir %u", __func__, dir);
1385 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1387 PF_HASHROW_UNLOCK(kh);
1401 PF_HASHROW_UNLOCK(kh);
1406 /* END state table stuff */
1409 pf_send(struct pf_send_entry *pfse)
1413 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1415 swi_sched(V_pf_swi_cookie, 0);
1421 struct pf_send_head queue;
1422 struct pf_send_entry *pfse, *next;
1424 CURVNET_SET((struct vnet *)v);
1427 queue = V_pf_sendqueue;
1428 STAILQ_INIT(&V_pf_sendqueue);
1431 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1432 switch (pfse->pfse_type) {
1435 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1438 icmp_error(pfse->pfse_m, pfse->icmpopts.type,
1439 pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
1444 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1448 icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
1449 pfse->icmpopts.code, pfse->icmpopts.mtu);
1453 panic("%s: unknown type", __func__);
1455 free(pfse, M_PFTEMP);
1461 pf_purge_thread(void *unused __unused)
1463 VNET_ITERATOR_DECL(vnet_iter);
1465 sx_xlock(&pf_end_lock);
1466 while (pf_end_threads == 0) {
1467 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", hz / 10);
1470 VNET_FOREACH(vnet_iter) {
1471 CURVNET_SET(vnet_iter);
1474 /* Wait until V_pf_default_rule is initialized. */
1475 if (V_pf_vnet_active == 0) {
1481 * Process 1/interval fraction of the state
1485 pf_purge_expired_states(V_pf_purge_idx, pf_hashmask /
1486 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1489 * Purge other expired types every
1490 * PFTM_INTERVAL seconds.
1492 if (V_pf_purge_idx == 0) {
1494 * Order is important:
1495 * - states and src nodes reference rules
1496 * - states and rules reference kifs
1498 pf_purge_expired_fragments();
1499 pf_purge_expired_src_nodes();
1500 pf_purge_unlinked_rules();
1505 VNET_LIST_RUNLOCK();
1509 sx_xunlock(&pf_end_lock);
1514 pf_unload_vnet_purge(void)
1518 * To cleanse up all kifs and rules we need
1519 * two runs: first one clears reference flags,
1520 * then pf_purge_expired_states() doesn't
1521 * raise them, and then second run frees.
1523 pf_purge_unlinked_rules();
1527 * Now purge everything.
1529 pf_purge_expired_states(0, pf_hashmask);
1530 pf_purge_fragments(UINT_MAX);
1531 pf_purge_expired_src_nodes();
1534 * Now all kifs & rules should be unreferenced,
1535 * thus should be successfully freed.
1537 pf_purge_unlinked_rules();
1543 pf_state_expires(const struct pf_state *state)
1550 /* handle all PFTM_* > PFTM_MAX here */
1551 if (state->timeout == PFTM_PURGE)
1552 return (time_uptime);
1553 KASSERT(state->timeout != PFTM_UNLINKED,
1554 ("pf_state_expires: timeout == PFTM_UNLINKED"));
1555 KASSERT((state->timeout < PFTM_MAX),
1556 ("pf_state_expires: timeout > PFTM_MAX"));
1557 timeout = state->rule.ptr->timeout[state->timeout];
1559 timeout = V_pf_default_rule.timeout[state->timeout];
1560 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1562 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1563 states = counter_u64_fetch(state->rule.ptr->states_cur);
1565 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1566 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1567 states = V_pf_status.states;
1569 if (end && states > start && start < end) {
1571 return (state->expire + timeout * (end - states) /
1574 return (time_uptime);
1576 return (state->expire + timeout);
1580 pf_purge_expired_src_nodes()
1582 struct pf_src_node_list freelist;
1583 struct pf_srchash *sh;
1584 struct pf_src_node *cur, *next;
1587 LIST_INIT(&freelist);
1588 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1589 PF_HASHROW_LOCK(sh);
1590 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1591 if (cur->states == 0 && cur->expire <= time_uptime) {
1592 pf_unlink_src_node(cur);
1593 LIST_INSERT_HEAD(&freelist, cur, entry);
1594 } else if (cur->rule.ptr != NULL)
1595 cur->rule.ptr->rule_flag |= PFRULE_REFS;
1596 PF_HASHROW_UNLOCK(sh);
1599 pf_free_src_nodes(&freelist);
1601 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1605 pf_src_tree_remove_state(struct pf_state *s)
1607 struct pf_src_node *sn;
1608 struct pf_srchash *sh;
1611 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
1612 s->rule.ptr->timeout[PFTM_SRC_NODE] :
1613 V_pf_default_rule.timeout[PFTM_SRC_NODE];
1615 if (s->src_node != NULL) {
1617 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1618 PF_HASHROW_LOCK(sh);
1621 if (--sn->states == 0)
1622 sn->expire = time_uptime + timeout;
1623 PF_HASHROW_UNLOCK(sh);
1625 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1626 sn = s->nat_src_node;
1627 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1628 PF_HASHROW_LOCK(sh);
1629 if (--sn->states == 0)
1630 sn->expire = time_uptime + timeout;
1631 PF_HASHROW_UNLOCK(sh);
1633 s->src_node = s->nat_src_node = NULL;
1637 * Unlink and potentilly free a state. Function may be
1638 * called with ID hash row locked, but always returns
1639 * unlocked, since it needs to go through key hash locking.
1642 pf_unlink_state(struct pf_state *s, u_int flags)
1644 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1646 if ((flags & PF_ENTER_LOCKED) == 0)
1647 PF_HASHROW_LOCK(ih);
1649 PF_HASHROW_ASSERT(ih);
1651 if (s->timeout == PFTM_UNLINKED) {
1653 * State is being processed
1654 * by pf_unlink_state() in
1657 PF_HASHROW_UNLOCK(ih);
1658 return (0); /* XXXGL: undefined actually */
1661 if (s->src.state == PF_TCPS_PROXY_DST) {
1662 /* XXX wire key the right one? */
1663 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1664 &s->key[PF_SK_WIRE]->addr[1],
1665 &s->key[PF_SK_WIRE]->addr[0],
1666 s->key[PF_SK_WIRE]->port[1],
1667 s->key[PF_SK_WIRE]->port[0],
1668 s->src.seqhi, s->src.seqlo + 1,
1669 TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1672 LIST_REMOVE(s, entry);
1673 pf_src_tree_remove_state(s);
1675 if (pfsync_delete_state_ptr != NULL)
1676 pfsync_delete_state_ptr(s);
1678 STATE_DEC_COUNTERS(s);
1680 s->timeout = PFTM_UNLINKED;
1682 PF_HASHROW_UNLOCK(ih);
1685 /* pf_state_insert() initialises refs to 2, so we can never release the
1686 * last reference here, only in pf_release_state(). */
1687 (void)refcount_release(&s->refs);
1689 return (pf_release_state(s));
1693 pf_free_state(struct pf_state *cur)
1696 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1697 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1700 pf_normalize_tcp_cleanup(cur);
1701 uma_zfree(V_pf_state_z, cur);
1702 counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
1706 * Called only from pf_purge_thread(), thus serialized.
1709 pf_purge_expired_states(u_int i, int maxcheck)
1711 struct pf_idhash *ih;
1714 V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1717 * Go through hash and unlink states that expire now.
1719 while (maxcheck > 0) {
1721 ih = &V_pf_idhash[i];
1723 PF_HASHROW_LOCK(ih);
1724 LIST_FOREACH(s, &ih->states, entry) {
1725 if (pf_state_expires(s) <= time_uptime) {
1726 V_pf_status.states -=
1727 pf_unlink_state(s, PF_ENTER_LOCKED);
1730 s->rule.ptr->rule_flag |= PFRULE_REFS;
1731 if (s->nat_rule.ptr != NULL)
1732 s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1733 if (s->anchor.ptr != NULL)
1734 s->anchor.ptr->rule_flag |= PFRULE_REFS;
1735 s->kif->pfik_flags |= PFI_IFLAG_REFS;
1737 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1739 PF_HASHROW_UNLOCK(ih);
1741 /* Return when we hit end of hash. */
1742 if (++i > pf_hashmask) {
1743 V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1750 V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1756 pf_purge_unlinked_rules()
1758 struct pf_rulequeue tmpq;
1759 struct pf_rule *r, *r1;
1762 * If we have overloading task pending, then we'd
1763 * better skip purging this time. There is a tiny
1764 * probability that overloading task references
1765 * an already unlinked rule.
1767 PF_OVERLOADQ_LOCK();
1768 if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1769 PF_OVERLOADQ_UNLOCK();
1772 PF_OVERLOADQ_UNLOCK();
1775 * Do naive mark-and-sweep garbage collecting of old rules.
1776 * Reference flag is raised by pf_purge_expired_states()
1777 * and pf_purge_expired_src_nodes().
1779 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1780 * use a temporary queue.
1783 PF_UNLNKDRULES_LOCK();
1784 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1785 if (!(r->rule_flag & PFRULE_REFS)) {
1786 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1787 TAILQ_INSERT_TAIL(&tmpq, r, entries);
1789 r->rule_flag &= ~PFRULE_REFS;
1791 PF_UNLNKDRULES_UNLOCK();
1793 if (!TAILQ_EMPTY(&tmpq)) {
1795 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1796 TAILQ_REMOVE(&tmpq, r, entries);
1804 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1809 u_int32_t a = ntohl(addr->addr32[0]);
1810 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1822 u_int8_t i, curstart, curend, maxstart, maxend;
1823 curstart = curend = maxstart = maxend = 255;
1824 for (i = 0; i < 8; i++) {
1825 if (!addr->addr16[i]) {
1826 if (curstart == 255)
1830 if ((curend - curstart) >
1831 (maxend - maxstart)) {
1832 maxstart = curstart;
1835 curstart = curend = 255;
1838 if ((curend - curstart) >
1839 (maxend - maxstart)) {
1840 maxstart = curstart;
1843 for (i = 0; i < 8; i++) {
1844 if (i >= maxstart && i <= maxend) {
1850 b = ntohs(addr->addr16[i]);
1867 pf_print_state(struct pf_state *s)
1869 pf_print_state_parts(s, NULL, NULL);
1873 pf_print_state_parts(struct pf_state *s,
1874 struct pf_state_key *skwp, struct pf_state_key *sksp)
1876 struct pf_state_key *skw, *sks;
1877 u_int8_t proto, dir;
1879 /* Do our best to fill these, but they're skipped if NULL */
1880 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1881 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1882 proto = skw ? skw->proto : (sks ? sks->proto : 0);
1883 dir = s ? s->direction : 0;
1901 case IPPROTO_ICMPV6:
1905 printf("%u", proto);
1918 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1920 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1925 pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1927 pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1932 if (proto == IPPROTO_TCP) {
1933 printf(" [lo=%u high=%u win=%u modulator=%u",
1934 s->src.seqlo, s->src.seqhi,
1935 s->src.max_win, s->src.seqdiff);
1936 if (s->src.wscale && s->dst.wscale)
1937 printf(" wscale=%u",
1938 s->src.wscale & PF_WSCALE_MASK);
1940 printf(" [lo=%u high=%u win=%u modulator=%u",
1941 s->dst.seqlo, s->dst.seqhi,
1942 s->dst.max_win, s->dst.seqdiff);
1943 if (s->src.wscale && s->dst.wscale)
1944 printf(" wscale=%u",
1945 s->dst.wscale & PF_WSCALE_MASK);
1948 printf(" %u:%u", s->src.state, s->dst.state);
1953 pf_print_flags(u_int8_t f)
1975 #define PF_SET_SKIP_STEPS(i) \
1977 while (head[i] != cur) { \
1978 head[i]->skip[i].ptr = cur; \
1979 head[i] = TAILQ_NEXT(head[i], entries); \
1984 pf_calc_skip_steps(struct pf_rulequeue *rules)
1986 struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1989 cur = TAILQ_FIRST(rules);
1991 for (i = 0; i < PF_SKIP_COUNT; ++i)
1993 while (cur != NULL) {
1995 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1996 PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1997 if (cur->direction != prev->direction)
1998 PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1999 if (cur->af != prev->af)
2000 PF_SET_SKIP_STEPS(PF_SKIP_AF);
2001 if (cur->proto != prev->proto)
2002 PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
2003 if (cur->src.neg != prev->src.neg ||
2004 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
2005 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
2006 if (cur->src.port[0] != prev->src.port[0] ||
2007 cur->src.port[1] != prev->src.port[1] ||
2008 cur->src.port_op != prev->src.port_op)
2009 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
2010 if (cur->dst.neg != prev->dst.neg ||
2011 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
2012 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
2013 if (cur->dst.port[0] != prev->dst.port[0] ||
2014 cur->dst.port[1] != prev->dst.port[1] ||
2015 cur->dst.port_op != prev->dst.port_op)
2016 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
2019 cur = TAILQ_NEXT(cur, entries);
2021 for (i = 0; i < PF_SKIP_COUNT; ++i)
2022 PF_SET_SKIP_STEPS(i);
2026 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
2028 if (aw1->type != aw2->type)
2030 switch (aw1->type) {
2031 case PF_ADDR_ADDRMASK:
2033 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
2035 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
2038 case PF_ADDR_DYNIFTL:
2039 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
2040 case PF_ADDR_NOROUTE:
2041 case PF_ADDR_URPFFAILED:
2044 return (aw1->p.tbl != aw2->p.tbl);
2046 printf("invalid address type: %d\n", aw1->type);
2052 * Checksum updates are a little complicated because the checksum in the TCP/UDP
2053 * header isn't always a full checksum. In some cases (i.e. output) it's a
2054 * pseudo-header checksum, which is a partial checksum over src/dst IP
2055 * addresses, protocol number and length.
2057 * That means we have the following cases:
2058 * * Input or forwarding: we don't have TSO, the checksum fields are full
2059 * checksums, we need to update the checksum whenever we change anything.
2060 * * Output (i.e. the checksum is a pseudo-header checksum):
2061 * x The field being updated is src/dst address or affects the length of
2062 * the packet. We need to update the pseudo-header checksum (note that this
2063 * checksum is not ones' complement).
2064 * x Some other field is being modified (e.g. src/dst port numbers): We
2065 * don't have to update anything.
2068 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2074 l = cksum + old - new;
2075 l = (l >> 16) + (l & 65535);
2083 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2084 u_int16_t new, u_int8_t udp)
2086 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2089 return (pf_cksum_fixup(cksum, old, new, udp));
2093 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2094 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2100 PF_ACPY(&ao, a, af);
2103 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2111 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2112 ao.addr16[0], an->addr16[0], 0),
2113 ao.addr16[1], an->addr16[1], 0);
2116 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2117 ao.addr16[0], an->addr16[0], u),
2118 ao.addr16[1], an->addr16[1], u);
2120 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2125 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2126 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2127 pf_cksum_fixup(pf_cksum_fixup(*pc,
2128 ao.addr16[0], an->addr16[0], u),
2129 ao.addr16[1], an->addr16[1], u),
2130 ao.addr16[2], an->addr16[2], u),
2131 ao.addr16[3], an->addr16[3], u),
2132 ao.addr16[4], an->addr16[4], u),
2133 ao.addr16[5], an->addr16[5], u),
2134 ao.addr16[6], an->addr16[6], u),
2135 ao.addr16[7], an->addr16[7], u);
2137 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2142 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2143 CSUM_DELAY_DATA_IPV6)) {
2150 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */
2152 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2156 memcpy(&ao, a, sizeof(ao));
2157 memcpy(a, &an, sizeof(u_int32_t));
2158 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2159 ao % 65536, an % 65536, u);
2163 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2167 memcpy(&ao, a, sizeof(ao));
2168 memcpy(a, &an, sizeof(u_int32_t));
2170 *c = pf_proto_cksum_fixup(m,
2171 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2172 ao % 65536, an % 65536, udp);
2177 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2181 PF_ACPY(&ao, a, AF_INET6);
2182 PF_ACPY(a, an, AF_INET6);
2184 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2185 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2186 pf_cksum_fixup(pf_cksum_fixup(*c,
2187 ao.addr16[0], an->addr16[0], u),
2188 ao.addr16[1], an->addr16[1], u),
2189 ao.addr16[2], an->addr16[2], u),
2190 ao.addr16[3], an->addr16[3], u),
2191 ao.addr16[4], an->addr16[4], u),
2192 ao.addr16[5], an->addr16[5], u),
2193 ao.addr16[6], an->addr16[6], u),
2194 ao.addr16[7], an->addr16[7], u);
2199 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2200 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2201 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2203 struct pf_addr oia, ooa;
2205 PF_ACPY(&oia, ia, af);
2207 PF_ACPY(&ooa, oa, af);
2209 /* Change inner protocol port, fix inner protocol checksum. */
2211 u_int16_t oip = *ip;
2218 *pc = pf_cksum_fixup(*pc, oip, *ip, u);
2219 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2221 *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2223 /* Change inner ip address, fix inner ip and icmp checksums. */
2224 PF_ACPY(ia, na, af);
2228 u_int32_t oh2c = *h2c;
2230 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2231 oia.addr16[0], ia->addr16[0], 0),
2232 oia.addr16[1], ia->addr16[1], 0);
2233 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2234 oia.addr16[0], ia->addr16[0], 0),
2235 oia.addr16[1], ia->addr16[1], 0);
2236 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2242 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2243 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2244 pf_cksum_fixup(pf_cksum_fixup(*ic,
2245 oia.addr16[0], ia->addr16[0], u),
2246 oia.addr16[1], ia->addr16[1], u),
2247 oia.addr16[2], ia->addr16[2], u),
2248 oia.addr16[3], ia->addr16[3], u),
2249 oia.addr16[4], ia->addr16[4], u),
2250 oia.addr16[5], ia->addr16[5], u),
2251 oia.addr16[6], ia->addr16[6], u),
2252 oia.addr16[7], ia->addr16[7], u);
2256 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2258 PF_ACPY(oa, na, af);
2262 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2263 ooa.addr16[0], oa->addr16[0], 0),
2264 ooa.addr16[1], oa->addr16[1], 0);
2269 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2270 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2271 pf_cksum_fixup(pf_cksum_fixup(*ic,
2272 ooa.addr16[0], oa->addr16[0], u),
2273 ooa.addr16[1], oa->addr16[1], u),
2274 ooa.addr16[2], oa->addr16[2], u),
2275 ooa.addr16[3], oa->addr16[3], u),
2276 ooa.addr16[4], oa->addr16[4], u),
2277 ooa.addr16[5], oa->addr16[5], u),
2278 ooa.addr16[6], oa->addr16[6], u),
2279 ooa.addr16[7], oa->addr16[7], u);
2288 * Need to modulate the sequence numbers in the TCP SACK option
2289 * (credits to Krzysztof Pfaff for report and patch)
2292 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2293 struct tcphdr *th, struct pf_state_peer *dst)
2295 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2296 u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2297 int copyback = 0, i, olen;
2298 struct sackblk sack;
2300 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2301 if (hlen < TCPOLEN_SACKLEN ||
2302 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2305 while (hlen >= TCPOLEN_SACKLEN) {
2308 case TCPOPT_EOL: /* FALLTHROUGH */
2316 if (olen >= TCPOLEN_SACKLEN) {
2317 for (i = 2; i + TCPOLEN_SACK <= olen;
2318 i += TCPOLEN_SACK) {
2319 memcpy(&sack, &opt[i], sizeof(sack));
2320 pf_change_proto_a(m, &sack.start, &th->th_sum,
2321 htonl(ntohl(sack.start) - dst->seqdiff), 0);
2322 pf_change_proto_a(m, &sack.end, &th->th_sum,
2323 htonl(ntohl(sack.end) - dst->seqdiff), 0);
2324 memcpy(&opt[i], &sack, sizeof(sack));
2338 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2343 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2344 const struct pf_addr *saddr, const struct pf_addr *daddr,
2345 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2346 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2347 u_int16_t rtag, struct ifnet *ifp)
2349 struct pf_send_entry *pfse;
2353 struct ip *h = NULL;
2356 struct ip6_hdr *h6 = NULL;
2360 struct pf_mtag *pf_mtag;
2365 /* maximum segment size tcp option */
2366 tlen = sizeof(struct tcphdr);
2373 len = sizeof(struct ip) + tlen;
2378 len = sizeof(struct ip6_hdr) + tlen;
2382 panic("%s: unsupported af %d", __func__, af);
2385 /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2386 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2389 m = m_gethdr(M_NOWAIT, MT_DATA);
2391 free(pfse, M_PFTEMP);
2395 mac_netinet_firewall_send(m);
2397 if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2398 free(pfse, M_PFTEMP);
2403 m->m_flags |= M_SKIP_FIREWALL;
2404 pf_mtag->tag = rtag;
2406 if (r != NULL && r->rtableid >= 0)
2407 M_SETFIB(m, r->rtableid);
2410 if (r != NULL && r->qid) {
2411 pf_mtag->qid = r->qid;
2413 /* add hints for ecn */
2414 pf_mtag->hdr = mtod(m, struct ip *);
2417 m->m_data += max_linkhdr;
2418 m->m_pkthdr.len = m->m_len = len;
2419 m->m_pkthdr.rcvif = NULL;
2420 bzero(m->m_data, len);
2424 h = mtod(m, struct ip *);
2426 /* IP header fields included in the TCP checksum */
2427 h->ip_p = IPPROTO_TCP;
2428 h->ip_len = htons(tlen);
2429 h->ip_src.s_addr = saddr->v4.s_addr;
2430 h->ip_dst.s_addr = daddr->v4.s_addr;
2432 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2437 h6 = mtod(m, struct ip6_hdr *);
2439 /* IP header fields included in the TCP checksum */
2440 h6->ip6_nxt = IPPROTO_TCP;
2441 h6->ip6_plen = htons(tlen);
2442 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2443 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2445 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2451 th->th_sport = sport;
2452 th->th_dport = dport;
2453 th->th_seq = htonl(seq);
2454 th->th_ack = htonl(ack);
2455 th->th_off = tlen >> 2;
2456 th->th_flags = flags;
2457 th->th_win = htons(win);
2460 opt = (char *)(th + 1);
2461 opt[0] = TCPOPT_MAXSEG;
2464 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2471 th->th_sum = in_cksum(m, len);
2473 /* Finish the IP header */
2475 h->ip_hl = sizeof(*h) >> 2;
2476 h->ip_tos = IPTOS_LOWDELAY;
2477 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2478 h->ip_len = htons(len);
2479 h->ip_ttl = ttl ? ttl : V_ip_defttl;
2482 pfse->pfse_type = PFSE_IP;
2488 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2489 sizeof(struct ip6_hdr), tlen);
2491 h6->ip6_vfc |= IPV6_VERSION;
2492 h6->ip6_hlim = IPV6_DEFHLIM;
2494 pfse->pfse_type = PFSE_IP6;
2503 pf_ieee8021q_setpcp(struct mbuf *m, u_int8_t prio)
2507 KASSERT(prio <= PF_PRIO_MAX,
2508 ("%s with invalid pcp", __func__));
2510 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_OUT, NULL);
2512 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_OUT,
2513 sizeof(uint8_t), M_NOWAIT);
2516 m_tag_prepend(m, mtag);
2519 *(uint8_t *)(mtag + 1) = prio;
2524 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
2529 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
2533 if (prio == PF_PRIO_ZERO)
2536 mpcp = *(uint8_t *)(mtag + 1);
2538 return (mpcp == prio);
2542 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2545 struct pf_send_entry *pfse;
2547 struct pf_mtag *pf_mtag;
2549 /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2550 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2554 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2555 free(pfse, M_PFTEMP);
2559 if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2560 free(pfse, M_PFTEMP);
2564 m0->m_flags |= M_SKIP_FIREWALL;
2566 if (r->rtableid >= 0)
2567 M_SETFIB(m0, r->rtableid);
2571 pf_mtag->qid = r->qid;
2572 /* add hints for ecn */
2573 pf_mtag->hdr = mtod(m0, struct ip *);
2580 pfse->pfse_type = PFSE_ICMP;
2585 pfse->pfse_type = PFSE_ICMP6;
2590 pfse->icmpopts.type = type;
2591 pfse->icmpopts.code = code;
2596 * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2597 * If n is 0, they match if they are equal. If n is != 0, they match if they
2601 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2602 struct pf_addr *b, sa_family_t af)
2609 if ((a->addr32[0] & m->addr32[0]) ==
2610 (b->addr32[0] & m->addr32[0]))
2616 if (((a->addr32[0] & m->addr32[0]) ==
2617 (b->addr32[0] & m->addr32[0])) &&
2618 ((a->addr32[1] & m->addr32[1]) ==
2619 (b->addr32[1] & m->addr32[1])) &&
2620 ((a->addr32[2] & m->addr32[2]) ==
2621 (b->addr32[2] & m->addr32[2])) &&
2622 ((a->addr32[3] & m->addr32[3]) ==
2623 (b->addr32[3] & m->addr32[3])))
2642 * Return 1 if b <= a <= e, otherwise return 0.
2645 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2646 struct pf_addr *a, sa_family_t af)
2651 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
2652 (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
2661 for (i = 0; i < 4; ++i)
2662 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
2664 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
2667 for (i = 0; i < 4; ++i)
2668 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
2670 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
2680 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2684 return ((p > a1) && (p < a2));
2686 return ((p < a1) || (p > a2));
2688 return ((p >= a1) && (p <= a2));
2702 return (0); /* never reached */
2706 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2711 return (pf_match(op, a1, a2, p));
2715 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2717 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2719 return (pf_match(op, a1, a2, u));
2723 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2725 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2727 return (pf_match(op, a1, a2, g));
2731 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2736 return ((!r->match_tag_not && r->match_tag == *tag) ||
2737 (r->match_tag_not && r->match_tag != *tag));
2741 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2744 KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2746 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2749 pd->pf_mtag->tag = tag;
2754 #define PF_ANCHOR_STACKSIZE 32
2755 struct pf_anchor_stackframe {
2756 struct pf_ruleset *rs;
2757 struct pf_rule *r; /* XXX: + match bit */
2758 struct pf_anchor *child;
2762 * XXX: We rely on malloc(9) returning pointer aligned addresses.
2764 #define PF_ANCHORSTACK_MATCH 0x00000001
2765 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH)
2767 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2768 #define PF_ANCHOR_RULE(f) (struct pf_rule *) \
2769 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2770 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \
2771 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \
2775 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2776 struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2779 struct pf_anchor_stackframe *f;
2785 if (*depth >= PF_ANCHOR_STACKSIZE) {
2786 printf("%s: anchor stack overflow on %s\n",
2787 __func__, (*r)->anchor->name);
2788 *r = TAILQ_NEXT(*r, entries);
2790 } else if (*depth == 0 && a != NULL)
2792 f = stack + (*depth)++;
2795 if ((*r)->anchor_wildcard) {
2796 struct pf_anchor_node *parent = &(*r)->anchor->children;
2798 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2802 *rs = &f->child->ruleset;
2805 *rs = &(*r)->anchor->ruleset;
2807 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2811 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2812 struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2815 struct pf_anchor_stackframe *f;
2824 f = stack + *depth - 1;
2825 fr = PF_ANCHOR_RULE(f);
2826 if (f->child != NULL) {
2827 struct pf_anchor_node *parent;
2830 * This block traverses through
2831 * a wildcard anchor.
2833 parent = &fr->anchor->children;
2834 if (match != NULL && *match) {
2836 * If any of "*" matched, then
2837 * "foo/ *" matched, mark frame
2840 PF_ANCHOR_SET_MATCH(f);
2843 f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2844 if (f->child != NULL) {
2845 *rs = &f->child->ruleset;
2846 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2854 if (*depth == 0 && a != NULL)
2857 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2859 *r = TAILQ_NEXT(fr, entries);
2860 } while (*r == NULL);
2867 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2868 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2873 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2874 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2878 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2879 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2880 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2881 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2882 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2883 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2884 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2885 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2891 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2896 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2900 if (addr->addr32[3] == 0xffffffff) {
2901 addr->addr32[3] = 0;
2902 if (addr->addr32[2] == 0xffffffff) {
2903 addr->addr32[2] = 0;
2904 if (addr->addr32[1] == 0xffffffff) {
2905 addr->addr32[1] = 0;
2907 htonl(ntohl(addr->addr32[0]) + 1);
2910 htonl(ntohl(addr->addr32[1]) + 1);
2913 htonl(ntohl(addr->addr32[2]) + 1);
2916 htonl(ntohl(addr->addr32[3]) + 1);
2923 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
2925 struct pf_addr *saddr, *daddr;
2926 u_int16_t sport, dport;
2927 struct inpcbinfo *pi;
2930 pd->lookup.uid = UID_MAX;
2931 pd->lookup.gid = GID_MAX;
2933 switch (pd->proto) {
2935 if (pd->hdr.tcp == NULL)
2937 sport = pd->hdr.tcp->th_sport;
2938 dport = pd->hdr.tcp->th_dport;
2942 if (pd->hdr.udp == NULL)
2944 sport = pd->hdr.udp->uh_sport;
2945 dport = pd->hdr.udp->uh_dport;
2951 if (direction == PF_IN) {
2966 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
2967 dport, INPLOOKUP_RLOCKPCB, NULL, m);
2969 inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
2970 daddr->v4, dport, INPLOOKUP_WILDCARD |
2971 INPLOOKUP_RLOCKPCB, NULL, m);
2979 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
2980 dport, INPLOOKUP_RLOCKPCB, NULL, m);
2982 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
2983 &daddr->v6, dport, INPLOOKUP_WILDCARD |
2984 INPLOOKUP_RLOCKPCB, NULL, m);
2994 INP_RLOCK_ASSERT(inp);
2995 pd->lookup.uid = inp->inp_cred->cr_uid;
2996 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3003 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3007 u_int8_t *opt, optlen;
3008 u_int8_t wscale = 0;
3010 hlen = th_off << 2; /* hlen <= sizeof(hdr) */
3011 if (hlen <= sizeof(struct tcphdr))
3013 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3015 opt = hdr + sizeof(struct tcphdr);
3016 hlen -= sizeof(struct tcphdr);
3026 if (wscale > TCP_MAX_WINSHIFT)
3027 wscale = TCP_MAX_WINSHIFT;
3028 wscale |= PF_WSCALE_FLAG;
3043 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3047 u_int8_t *opt, optlen;
3048 u_int16_t mss = V_tcp_mssdflt;
3050 hlen = th_off << 2; /* hlen <= sizeof(hdr) */
3051 if (hlen <= sizeof(struct tcphdr))
3053 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3055 opt = hdr + sizeof(struct tcphdr);
3056 hlen -= sizeof(struct tcphdr);
3057 while (hlen >= TCPOLEN_MAXSEG) {
3065 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
3081 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
3084 struct nhop4_basic nh4;
3087 struct nhop6_basic nh6;
3088 struct in6_addr dst6;
3097 hlen = sizeof(struct ip);
3098 if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) == 0)
3099 mss = nh4.nh_mtu - hlen - sizeof(struct tcphdr);
3104 hlen = sizeof(struct ip6_hdr);
3105 in6_splitscope(&addr->v6, &dst6, &scopeid);
3106 if (fib6_lookup_nh_basic(rtableid, &dst6, scopeid, 0,0,&nh6)==0)
3107 mss = nh6.nh_mtu - hlen - sizeof(struct tcphdr);
3112 mss = max(V_tcp_mssdflt, mss);
3113 mss = min(mss, offer);
3114 mss = max(mss, 64); /* sanity - at least max opt space */
3119 pf_tcp_iss(struct pf_pdesc *pd)
3122 u_int32_t digest[4];
3124 if (V_pf_tcp_secret_init == 0) {
3125 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
3126 MD5Init(&V_pf_tcp_secret_ctx);
3127 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
3128 sizeof(V_pf_tcp_secret));
3129 V_pf_tcp_secret_init = 1;
3132 ctx = V_pf_tcp_secret_ctx;
3134 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
3135 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
3136 if (pd->af == AF_INET6) {
3137 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3138 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3140 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3141 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3143 MD5Final((u_char *)digest, &ctx);
3144 V_pf_tcp_iss_off += 4096;
3145 #define ISN_RANDOM_INCREMENT (4096 - 1)
3146 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
3148 #undef ISN_RANDOM_INCREMENT
3152 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
3153 struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
3154 struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
3156 struct pf_rule *nr = NULL;
3157 struct pf_addr * const saddr = pd->src;
3158 struct pf_addr * const daddr = pd->dst;
3159 sa_family_t af = pd->af;
3160 struct pf_rule *r, *a = NULL;
3161 struct pf_ruleset *ruleset = NULL;
3162 struct pf_src_node *nsn = NULL;
3163 struct tcphdr *th = pd->hdr.tcp;
3164 struct pf_state_key *sk = NULL, *nk = NULL;
3166 int rewrite = 0, hdrlen = 0;
3167 int tag = -1, rtableid = -1;
3171 u_int16_t sport = 0, dport = 0;
3172 u_int16_t bproto_sum = 0, bip_sum = 0;
3173 u_int8_t icmptype = 0, icmpcode = 0;
3174 struct pf_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE];
3179 INP_LOCK_ASSERT(inp);
3180 pd->lookup.uid = inp->inp_cred->cr_uid;
3181 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3182 pd->lookup.done = 1;
3185 switch (pd->proto) {
3187 sport = th->th_sport;
3188 dport = th->th_dport;
3189 hdrlen = sizeof(*th);
3192 sport = pd->hdr.udp->uh_sport;
3193 dport = pd->hdr.udp->uh_dport;
3194 hdrlen = sizeof(*pd->hdr.udp);
3198 if (pd->af != AF_INET)
3200 sport = dport = pd->hdr.icmp->icmp_id;
3201 hdrlen = sizeof(*pd->hdr.icmp);
3202 icmptype = pd->hdr.icmp->icmp_type;
3203 icmpcode = pd->hdr.icmp->icmp_code;
3205 if (icmptype == ICMP_UNREACH ||
3206 icmptype == ICMP_SOURCEQUENCH ||
3207 icmptype == ICMP_REDIRECT ||
3208 icmptype == ICMP_TIMXCEED ||
3209 icmptype == ICMP_PARAMPROB)
3214 case IPPROTO_ICMPV6:
3217 sport = dport = pd->hdr.icmp6->icmp6_id;
3218 hdrlen = sizeof(*pd->hdr.icmp6);
3219 icmptype = pd->hdr.icmp6->icmp6_type;
3220 icmpcode = pd->hdr.icmp6->icmp6_code;
3222 if (icmptype == ICMP6_DST_UNREACH ||
3223 icmptype == ICMP6_PACKET_TOO_BIG ||
3224 icmptype == ICMP6_TIME_EXCEEDED ||
3225 icmptype == ICMP6_PARAM_PROB)
3230 sport = dport = hdrlen = 0;
3234 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3236 /* check packet for BINAT/NAT/RDR */
3237 if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3238 &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3239 KASSERT(sk != NULL, ("%s: null sk", __func__));
3240 KASSERT(nk != NULL, ("%s: null nk", __func__));
3243 bip_sum = *pd->ip_sum;
3245 switch (pd->proto) {
3247 bproto_sum = th->th_sum;
3248 pd->proto_sum = &th->th_sum;
3250 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3251 nk->port[pd->sidx] != sport) {
3252 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
3253 &th->th_sum, &nk->addr[pd->sidx],
3254 nk->port[pd->sidx], 0, af);
3255 pd->sport = &th->th_sport;
3256 sport = th->th_sport;
3259 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3260 nk->port[pd->didx] != dport) {
3261 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
3262 &th->th_sum, &nk->addr[pd->didx],
3263 nk->port[pd->didx], 0, af);
3264 dport = th->th_dport;
3265 pd->dport = &th->th_dport;
3270 bproto_sum = pd->hdr.udp->uh_sum;
3271 pd->proto_sum = &pd->hdr.udp->uh_sum;
3273 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3274 nk->port[pd->sidx] != sport) {
3275 pf_change_ap(m, saddr, &pd->hdr.udp->uh_sport,
3276 pd->ip_sum, &pd->hdr.udp->uh_sum,
3277 &nk->addr[pd->sidx],
3278 nk->port[pd->sidx], 1, af);
3279 sport = pd->hdr.udp->uh_sport;
3280 pd->sport = &pd->hdr.udp->uh_sport;
3283 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3284 nk->port[pd->didx] != dport) {
3285 pf_change_ap(m, daddr, &pd->hdr.udp->uh_dport,
3286 pd->ip_sum, &pd->hdr.udp->uh_sum,
3287 &nk->addr[pd->didx],
3288 nk->port[pd->didx], 1, af);
3289 dport = pd->hdr.udp->uh_dport;
3290 pd->dport = &pd->hdr.udp->uh_dport;
3296 nk->port[0] = nk->port[1];
3297 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3298 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3299 nk->addr[pd->sidx].v4.s_addr, 0);
3301 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3302 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3303 nk->addr[pd->didx].v4.s_addr, 0);
3305 if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3306 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3307 pd->hdr.icmp->icmp_cksum, sport,
3309 pd->hdr.icmp->icmp_id = nk->port[1];
3310 pd->sport = &pd->hdr.icmp->icmp_id;
3312 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3316 case IPPROTO_ICMPV6:
3317 nk->port[0] = nk->port[1];
3318 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3319 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3320 &nk->addr[pd->sidx], 0);
3322 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3323 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3324 &nk->addr[pd->didx], 0);
3333 &nk->addr[pd->sidx], AF_INET))
3334 pf_change_a(&saddr->v4.s_addr,
3336 nk->addr[pd->sidx].v4.s_addr, 0);
3339 &nk->addr[pd->didx], AF_INET))
3340 pf_change_a(&daddr->v4.s_addr,
3342 nk->addr[pd->didx].v4.s_addr, 0);
3348 &nk->addr[pd->sidx], AF_INET6))
3349 PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3352 &nk->addr[pd->didx], AF_INET6))
3353 PF_ACPY(saddr, &nk->addr[pd->didx], af);
3366 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3367 r = r->skip[PF_SKIP_IFP].ptr;
3368 else if (r->direction && r->direction != direction)
3369 r = r->skip[PF_SKIP_DIR].ptr;
3370 else if (r->af && r->af != af)
3371 r = r->skip[PF_SKIP_AF].ptr;
3372 else if (r->proto && r->proto != pd->proto)
3373 r = r->skip[PF_SKIP_PROTO].ptr;
3374 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3375 r->src.neg, kif, M_GETFIB(m)))
3376 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3377 /* tcp/udp only. port_op always 0 in other cases */
3378 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3379 r->src.port[0], r->src.port[1], sport))
3380 r = r->skip[PF_SKIP_SRC_PORT].ptr;
3381 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3382 r->dst.neg, NULL, M_GETFIB(m)))
3383 r = r->skip[PF_SKIP_DST_ADDR].ptr;
3384 /* tcp/udp only. port_op always 0 in other cases */
3385 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3386 r->dst.port[0], r->dst.port[1], dport))
3387 r = r->skip[PF_SKIP_DST_PORT].ptr;
3388 /* icmp only. type always 0 in other cases */
3389 else if (r->type && r->type != icmptype + 1)
3390 r = TAILQ_NEXT(r, entries);
3391 /* icmp only. type always 0 in other cases */
3392 else if (r->code && r->code != icmpcode + 1)
3393 r = TAILQ_NEXT(r, entries);
3394 else if (r->tos && !(r->tos == pd->tos))
3395 r = TAILQ_NEXT(r, entries);
3396 else if (r->rule_flag & PFRULE_FRAGMENT)
3397 r = TAILQ_NEXT(r, entries);
3398 else if (pd->proto == IPPROTO_TCP &&
3399 (r->flagset & th->th_flags) != r->flags)
3400 r = TAILQ_NEXT(r, entries);
3401 /* tcp/udp only. uid.op always 0 in other cases */
3402 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3403 pf_socket_lookup(direction, pd, m), 1)) &&
3404 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3406 r = TAILQ_NEXT(r, entries);
3407 /* tcp/udp only. gid.op always 0 in other cases */
3408 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3409 pf_socket_lookup(direction, pd, m), 1)) &&
3410 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3412 r = TAILQ_NEXT(r, entries);
3414 !pf_match_ieee8021q_pcp(r->prio, m))
3415 r = TAILQ_NEXT(r, entries);
3417 r->prob <= arc4random())
3418 r = TAILQ_NEXT(r, entries);
3419 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3420 pd->pf_mtag ? pd->pf_mtag->tag : 0))
3421 r = TAILQ_NEXT(r, entries);
3422 else if (r->os_fingerprint != PF_OSFP_ANY &&
3423 (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3424 pf_osfp_fingerprint(pd, m, off, th),
3425 r->os_fingerprint)))
3426 r = TAILQ_NEXT(r, entries);
3430 if (r->rtableid >= 0)
3431 rtableid = r->rtableid;
3432 if (r->anchor == NULL) {
3439 r = TAILQ_NEXT(r, entries);
3441 pf_step_into_anchor(anchor_stack, &asd,
3442 &ruleset, PF_RULESET_FILTER, &r, &a,
3445 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3446 &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3453 REASON_SET(&reason, PFRES_MATCH);
3455 if (r->log || (nr != NULL && nr->log)) {
3457 m_copyback(m, off, hdrlen, pd->hdr.any);
3458 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3462 if ((r->action == PF_DROP) &&
3463 ((r->rule_flag & PFRULE_RETURNRST) ||
3464 (r->rule_flag & PFRULE_RETURNICMP) ||
3465 (r->rule_flag & PFRULE_RETURN))) {
3466 /* undo NAT changes, if they have taken place */
3468 PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3469 PF_ACPY(daddr, &sk->addr[pd->didx], af);
3471 *pd->sport = sk->port[pd->sidx];
3473 *pd->dport = sk->port[pd->didx];
3475 *pd->proto_sum = bproto_sum;
3477 *pd->ip_sum = bip_sum;
3478 m_copyback(m, off, hdrlen, pd->hdr.any);
3480 if (pd->proto == IPPROTO_TCP &&
3481 ((r->rule_flag & PFRULE_RETURNRST) ||
3482 (r->rule_flag & PFRULE_RETURN)) &&
3483 !(th->th_flags & TH_RST)) {
3484 u_int32_t ack = ntohl(th->th_seq) + pd->p_len;
3496 h4 = mtod(m, struct ip *);
3497 len = ntohs(h4->ip_len) - off;
3502 h6 = mtod(m, struct ip6_hdr *);
3503 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3508 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3509 REASON_SET(&reason, PFRES_PROTCKSUM);
3511 if (th->th_flags & TH_SYN)
3513 if (th->th_flags & TH_FIN)
3515 pf_send_tcp(m, r, af, pd->dst,
3516 pd->src, th->th_dport, th->th_sport,
3517 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3518 r->return_ttl, 1, 0, kif->pfik_ifp);
3520 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3522 pf_send_icmp(m, r->return_icmp >> 8,
3523 r->return_icmp & 255, af, r);
3524 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3526 pf_send_icmp(m, r->return_icmp6 >> 8,
3527 r->return_icmp6 & 255, af, r);
3530 if (r->action == PF_DROP)
3533 if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3534 REASON_SET(&reason, PFRES_MEMORY);
3538 M_SETFIB(m, rtableid);
3540 if (!state_icmp && (r->keep_state || nr != NULL ||
3541 (pd->flags & PFDESC_TCP_NORM))) {
3543 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3544 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3546 if (action != PF_PASS)
3550 uma_zfree(V_pf_state_key_z, sk);
3552 uma_zfree(V_pf_state_key_z, nk);
3555 /* copy back packet headers if we performed NAT operations */
3557 m_copyback(m, off, hdrlen, pd->hdr.any);
3559 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3560 direction == PF_OUT &&
3561 pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m))
3563 * We want the state created, but we dont
3564 * want to send this in case a partner
3565 * firewall has to know about it to allow
3566 * replies through it.
3574 uma_zfree(V_pf_state_key_z, sk);
3576 uma_zfree(V_pf_state_key_z, nk);
3581 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3582 struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3583 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3584 u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3585 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3587 struct pf_state *s = NULL;
3588 struct pf_src_node *sn = NULL;
3589 struct tcphdr *th = pd->hdr.tcp;
3590 u_int16_t mss = V_tcp_mssdflt;
3593 /* check maximums */
3594 if (r->max_states &&
3595 (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3596 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
3597 REASON_SET(&reason, PFRES_MAXSTATES);
3600 /* src node for filter rule */
3601 if ((r->rule_flag & PFRULE_SRCTRACK ||
3602 r->rpool.opts & PF_POOL_STICKYADDR) &&
3603 pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3604 REASON_SET(&reason, PFRES_SRCLIMIT);
3607 /* src node for translation rule */
3608 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3609 pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3610 REASON_SET(&reason, PFRES_SRCLIMIT);
3613 s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3615 REASON_SET(&reason, PFRES_MEMORY);
3619 s->nat_rule.ptr = nr;
3621 STATE_INC_COUNTERS(s);
3623 s->state_flags |= PFSTATE_ALLOWOPTS;
3624 if (r->rule_flag & PFRULE_STATESLOPPY)
3625 s->state_flags |= PFSTATE_SLOPPY;
3626 s->log = r->log & PF_LOG_ALL;
3627 s->sync_state = PFSYNC_S_NONE;
3629 s->log |= nr->log & PF_LOG_ALL;
3630 switch (pd->proto) {
3632 s->src.seqlo = ntohl(th->th_seq);
3633 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3634 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3635 r->keep_state == PF_STATE_MODULATE) {
3636 /* Generate sequence number modulator */
3637 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3640 pf_change_proto_a(m, &th->th_seq, &th->th_sum,
3641 htonl(s->src.seqlo + s->src.seqdiff), 0);
3645 if (th->th_flags & TH_SYN) {
3647 s->src.wscale = pf_get_wscale(m, off,
3648 th->th_off, pd->af);
3650 s->src.max_win = MAX(ntohs(th->th_win), 1);
3651 if (s->src.wscale & PF_WSCALE_MASK) {
3652 /* Remove scale factor from initial window */
3653 int win = s->src.max_win;
3654 win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3655 s->src.max_win = (win - 1) >>
3656 (s->src.wscale & PF_WSCALE_MASK);
3658 if (th->th_flags & TH_FIN)
3662 s->src.state = TCPS_SYN_SENT;
3663 s->dst.state = TCPS_CLOSED;
3664 s->timeout = PFTM_TCP_FIRST_PACKET;
3667 s->src.state = PFUDPS_SINGLE;
3668 s->dst.state = PFUDPS_NO_TRAFFIC;
3669 s->timeout = PFTM_UDP_FIRST_PACKET;
3673 case IPPROTO_ICMPV6:
3675 s->timeout = PFTM_ICMP_FIRST_PACKET;
3678 s->src.state = PFOTHERS_SINGLE;
3679 s->dst.state = PFOTHERS_NO_TRAFFIC;
3680 s->timeout = PFTM_OTHER_FIRST_PACKET;
3684 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) {
3685 REASON_SET(&reason, PFRES_MAPFAILED);
3686 pf_src_tree_remove_state(s);
3687 STATE_DEC_COUNTERS(s);
3688 uma_zfree(V_pf_state_z, s);
3691 s->rt_kif = r->rpool.cur->kif;
3694 s->creation = time_uptime;
3695 s->expire = time_uptime;
3700 /* XXX We only modify one side for now. */
3701 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3702 s->nat_src_node = nsn;
3704 if (pd->proto == IPPROTO_TCP) {
3705 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3706 off, pd, th, &s->src, &s->dst)) {
3707 REASON_SET(&reason, PFRES_MEMORY);
3708 pf_src_tree_remove_state(s);
3709 STATE_DEC_COUNTERS(s);
3710 uma_zfree(V_pf_state_z, s);
3713 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3714 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3715 &s->src, &s->dst, rewrite)) {
3716 /* This really shouldn't happen!!! */
3717 DPFPRINTF(PF_DEBUG_URGENT,
3718 ("pf_normalize_tcp_stateful failed on first pkt"));
3719 pf_normalize_tcp_cleanup(s);
3720 pf_src_tree_remove_state(s);
3721 STATE_DEC_COUNTERS(s);
3722 uma_zfree(V_pf_state_z, s);
3726 s->direction = pd->dir;
3729 * sk/nk could already been setup by pf_get_translation().
3732 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3733 __func__, nr, sk, nk));
3734 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3739 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3740 __func__, nr, sk, nk));
3742 /* Swap sk/nk for PF_OUT. */
3743 if (pf_state_insert(BOUND_IFACE(r, kif),
3744 (pd->dir == PF_IN) ? sk : nk,
3745 (pd->dir == PF_IN) ? nk : sk, s)) {
3746 if (pd->proto == IPPROTO_TCP)
3747 pf_normalize_tcp_cleanup(s);
3748 REASON_SET(&reason, PFRES_STATEINS);
3749 pf_src_tree_remove_state(s);
3750 STATE_DEC_COUNTERS(s);
3751 uma_zfree(V_pf_state_z, s);
3758 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3759 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3760 s->src.state = PF_TCPS_PROXY_SRC;
3761 /* undo NAT changes, if they have taken place */
3763 struct pf_state_key *skt = s->key[PF_SK_WIRE];
3764 if (pd->dir == PF_OUT)
3765 skt = s->key[PF_SK_STACK];
3766 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3767 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3769 *pd->sport = skt->port[pd->sidx];
3771 *pd->dport = skt->port[pd->didx];
3773 *pd->proto_sum = bproto_sum;
3775 *pd->ip_sum = bip_sum;
3776 m_copyback(m, off, hdrlen, pd->hdr.any);
3778 s->src.seqhi = htonl(arc4random());
3779 /* Find mss option */
3780 int rtid = M_GETFIB(m);
3781 mss = pf_get_mss(m, off, th->th_off, pd->af);
3782 mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3783 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3785 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3786 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3787 TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3788 REASON_SET(&reason, PFRES_SYNPROXY);
3789 return (PF_SYNPROXY_DROP);
3796 uma_zfree(V_pf_state_key_z, sk);
3798 uma_zfree(V_pf_state_key_z, nk);
3801 struct pf_srchash *sh;
3803 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
3804 PF_HASHROW_LOCK(sh);
3805 if (--sn->states == 0 && sn->expire == 0) {
3806 pf_unlink_src_node(sn);
3807 uma_zfree(V_pf_sources_z, sn);
3809 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3811 PF_HASHROW_UNLOCK(sh);
3814 if (nsn != sn && nsn != NULL) {
3815 struct pf_srchash *sh;
3817 sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)];
3818 PF_HASHROW_LOCK(sh);
3819 if (--nsn->states == 0 && nsn->expire == 0) {
3820 pf_unlink_src_node(nsn);
3821 uma_zfree(V_pf_sources_z, nsn);
3823 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3825 PF_HASHROW_UNLOCK(sh);
3832 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
3833 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
3834 struct pf_ruleset **rsm)
3836 struct pf_rule *r, *a = NULL;
3837 struct pf_ruleset *ruleset = NULL;
3838 sa_family_t af = pd->af;
3843 struct pf_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE];
3847 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3850 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3851 r = r->skip[PF_SKIP_IFP].ptr;
3852 else if (r->direction && r->direction != direction)
3853 r = r->skip[PF_SKIP_DIR].ptr;
3854 else if (r->af && r->af != af)
3855 r = r->skip[PF_SKIP_AF].ptr;
3856 else if (r->proto && r->proto != pd->proto)
3857 r = r->skip[PF_SKIP_PROTO].ptr;
3858 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3859 r->src.neg, kif, M_GETFIB(m)))
3860 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3861 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3862 r->dst.neg, NULL, M_GETFIB(m)))
3863 r = r->skip[PF_SKIP_DST_ADDR].ptr;
3864 else if (r->tos && !(r->tos == pd->tos))
3865 r = TAILQ_NEXT(r, entries);
3866 else if (r->os_fingerprint != PF_OSFP_ANY)
3867 r = TAILQ_NEXT(r, entries);
3868 else if (pd->proto == IPPROTO_UDP &&
3869 (r->src.port_op || r->dst.port_op))
3870 r = TAILQ_NEXT(r, entries);
3871 else if (pd->proto == IPPROTO_TCP &&
3872 (r->src.port_op || r->dst.port_op || r->flagset))
3873 r = TAILQ_NEXT(r, entries);
3874 else if ((pd->proto == IPPROTO_ICMP ||
3875 pd->proto == IPPROTO_ICMPV6) &&
3876 (r->type || r->code))
3877 r = TAILQ_NEXT(r, entries);
3879 !pf_match_ieee8021q_pcp(r->prio, m))
3880 r = TAILQ_NEXT(r, entries);
3881 else if (r->prob && r->prob <=
3882 (arc4random() % (UINT_MAX - 1) + 1))
3883 r = TAILQ_NEXT(r, entries);
3884 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3885 pd->pf_mtag ? pd->pf_mtag->tag : 0))
3886 r = TAILQ_NEXT(r, entries);
3888 if (r->anchor == NULL) {
3895 r = TAILQ_NEXT(r, entries);
3897 pf_step_into_anchor(anchor_stack, &asd,
3898 &ruleset, PF_RULESET_FILTER, &r, &a,
3901 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3902 &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3909 REASON_SET(&reason, PFRES_MATCH);
3912 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
3915 if (r->action != PF_PASS)
3918 if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3919 REASON_SET(&reason, PFRES_MEMORY);
3927 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
3928 struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
3929 struct pf_pdesc *pd, u_short *reason, int *copyback)
3931 struct tcphdr *th = pd->hdr.tcp;
3932 u_int16_t win = ntohs(th->th_win);
3933 u_int32_t ack, end, seq, orig_seq;
3937 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
3938 sws = src->wscale & PF_WSCALE_MASK;
3939 dws = dst->wscale & PF_WSCALE_MASK;
3944 * Sequence tracking algorithm from Guido van Rooij's paper:
3945 * http://www.madison-gurkha.com/publications/tcp_filtering/
3949 orig_seq = seq = ntohl(th->th_seq);
3950 if (src->seqlo == 0) {
3951 /* First packet from this end. Set its state */
3953 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
3954 src->scrub == NULL) {
3955 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
3956 REASON_SET(reason, PFRES_MEMORY);
3961 /* Deferred generation of sequence number modulator */
3962 if (dst->seqdiff && !src->seqdiff) {
3963 /* use random iss for the TCP server */
3964 while ((src->seqdiff = arc4random() - seq) == 0)
3966 ack = ntohl(th->th_ack) - dst->seqdiff;
3967 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
3969 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
3972 ack = ntohl(th->th_ack);
3975 end = seq + pd->p_len;
3976 if (th->th_flags & TH_SYN) {
3978 if (dst->wscale & PF_WSCALE_FLAG) {
3979 src->wscale = pf_get_wscale(m, off, th->th_off,
3981 if (src->wscale & PF_WSCALE_FLAG) {
3982 /* Remove scale factor from initial
3984 sws = src->wscale & PF_WSCALE_MASK;
3985 win = ((u_int32_t)win + (1 << sws) - 1)
3987 dws = dst->wscale & PF_WSCALE_MASK;
3989 /* fixup other window */
3990 dst->max_win <<= dst->wscale &
3992 /* in case of a retrans SYN|ACK */
3997 if (th->th_flags & TH_FIN)
4001 if (src->state < TCPS_SYN_SENT)
4002 src->state = TCPS_SYN_SENT;
4005 * May need to slide the window (seqhi may have been set by
4006 * the crappy stack check or if we picked up the connection
4007 * after establishment)
4009 if (src->seqhi == 1 ||
4010 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
4011 src->seqhi = end + MAX(1, dst->max_win << dws);
4012 if (win > src->max_win)
4016 ack = ntohl(th->th_ack) - dst->seqdiff;
4018 /* Modulate sequence numbers */
4019 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
4021 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
4024 end = seq + pd->p_len;
4025 if (th->th_flags & TH_SYN)
4027 if (th->th_flags & TH_FIN)
4031 if ((th->th_flags & TH_ACK) == 0) {
4032 /* Let it pass through the ack skew check */
4034 } else if ((ack == 0 &&
4035 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
4036 /* broken tcp stacks do not set ack */
4037 (dst->state < TCPS_SYN_SENT)) {
4039 * Many stacks (ours included) will set the ACK number in an
4040 * FIN|ACK if the SYN times out -- no sequence to ACK.
4046 /* Ease sequencing restrictions on no data packets */
4051 ackskew = dst->seqlo - ack;
4055 * Need to demodulate the sequence numbers in any TCP SACK options
4056 * (Selective ACK). We could optionally validate the SACK values
4057 * against the current ACK window, either forwards or backwards, but
4058 * I'm not confident that SACK has been implemented properly
4059 * everywhere. It wouldn't surprise me if several stacks accidentally
4060 * SACK too far backwards of previously ACKed data. There really aren't
4061 * any security implications of bad SACKing unless the target stack
4062 * doesn't validate the option length correctly. Someone trying to
4063 * spoof into a TCP connection won't bother blindly sending SACK
4066 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
4067 if (pf_modulate_sack(m, off, pd, th, dst))
4072 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */
4073 if (SEQ_GEQ(src->seqhi, end) &&
4074 /* Last octet inside other's window space */
4075 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
4076 /* Retrans: not more than one window back */
4077 (ackskew >= -MAXACKWINDOW) &&
4078 /* Acking not more than one reassembled fragment backwards */
4079 (ackskew <= (MAXACKWINDOW << sws)) &&
4080 /* Acking not more than one window forward */
4081 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
4082 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
4083 (pd->flags & PFDESC_IP_REAS) == 0)) {
4084 /* Require an exact/+1 sequence match on resets when possible */
4086 if (dst->scrub || src->scrub) {
4087 if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4088 *state, src, dst, copyback))
4092 /* update max window */
4093 if (src->max_win < win)
4095 /* synchronize sequencing */
4096 if (SEQ_GT(end, src->seqlo))
4098 /* slide the window of what the other end can send */
4099 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4100 dst->seqhi = ack + MAX((win << sws), 1);
4104 if (th->th_flags & TH_SYN)
4105 if (src->state < TCPS_SYN_SENT)
4106 src->state = TCPS_SYN_SENT;
4107 if (th->th_flags & TH_FIN)
4108 if (src->state < TCPS_CLOSING)
4109 src->state = TCPS_CLOSING;
4110 if (th->th_flags & TH_ACK) {
4111 if (dst->state == TCPS_SYN_SENT) {
4112 dst->state = TCPS_ESTABLISHED;
4113 if (src->state == TCPS_ESTABLISHED &&
4114 (*state)->src_node != NULL &&
4115 pf_src_connlimit(state)) {
4116 REASON_SET(reason, PFRES_SRCLIMIT);
4119 } else if (dst->state == TCPS_CLOSING)
4120 dst->state = TCPS_FIN_WAIT_2;
4122 if (th->th_flags & TH_RST)
4123 src->state = dst->state = TCPS_TIME_WAIT;
4125 /* update expire time */
4126 (*state)->expire = time_uptime;
4127 if (src->state >= TCPS_FIN_WAIT_2 &&
4128 dst->state >= TCPS_FIN_WAIT_2)
4129 (*state)->timeout = PFTM_TCP_CLOSED;
4130 else if (src->state >= TCPS_CLOSING &&
4131 dst->state >= TCPS_CLOSING)
4132 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4133 else if (src->state < TCPS_ESTABLISHED ||
4134 dst->state < TCPS_ESTABLISHED)
4135 (*state)->timeout = PFTM_TCP_OPENING;
4136 else if (src->state >= TCPS_CLOSING ||
4137 dst->state >= TCPS_CLOSING)
4138 (*state)->timeout = PFTM_TCP_CLOSING;
4140 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4142 /* Fall through to PASS packet */
4144 } else if ((dst->state < TCPS_SYN_SENT ||
4145 dst->state >= TCPS_FIN_WAIT_2 ||
4146 src->state >= TCPS_FIN_WAIT_2) &&
4147 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4148 /* Within a window forward of the originating packet */
4149 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4150 /* Within a window backward of the originating packet */
4153 * This currently handles three situations:
4154 * 1) Stupid stacks will shotgun SYNs before their peer
4156 * 2) When PF catches an already established stream (the
4157 * firewall rebooted, the state table was flushed, routes
4159 * 3) Packets get funky immediately after the connection
4160 * closes (this should catch Solaris spurious ACK|FINs
4161 * that web servers like to spew after a close)
4163 * This must be a little more careful than the above code
4164 * since packet floods will also be caught here. We don't
4165 * update the TTL here to mitigate the damage of a packet
4166 * flood and so the same code can handle awkward establishment
4167 * and a loosened connection close.
4168 * In the establishment case, a correct peer response will
4169 * validate the connection, go through the normal state code
4170 * and keep updating the state TTL.
4173 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4174 printf("pf: loose state match: ");
4175 pf_print_state(*state);
4176 pf_print_flags(th->th_flags);
4177 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4178 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
4179 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
4180 (unsigned long long)(*state)->packets[1],
4181 pd->dir == PF_IN ? "in" : "out",
4182 pd->dir == (*state)->direction ? "fwd" : "rev");
4185 if (dst->scrub || src->scrub) {
4186 if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4187 *state, src, dst, copyback))
4191 /* update max window */
4192 if (src->max_win < win)
4194 /* synchronize sequencing */
4195 if (SEQ_GT(end, src->seqlo))
4197 /* slide the window of what the other end can send */
4198 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4199 dst->seqhi = ack + MAX((win << sws), 1);
4202 * Cannot set dst->seqhi here since this could be a shotgunned
4203 * SYN and not an already established connection.
4206 if (th->th_flags & TH_FIN)
4207 if (src->state < TCPS_CLOSING)
4208 src->state = TCPS_CLOSING;
4209 if (th->th_flags & TH_RST)
4210 src->state = dst->state = TCPS_TIME_WAIT;
4212 /* Fall through to PASS packet */
4215 if ((*state)->dst.state == TCPS_SYN_SENT &&
4216 (*state)->src.state == TCPS_SYN_SENT) {
4217 /* Send RST for state mismatches during handshake */
4218 if (!(th->th_flags & TH_RST))
4219 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4220 pd->dst, pd->src, th->th_dport,
4221 th->th_sport, ntohl(th->th_ack), 0,
4223 (*state)->rule.ptr->return_ttl, 1, 0,
4228 } else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4229 printf("pf: BAD state: ");
4230 pf_print_state(*state);
4231 pf_print_flags(th->th_flags);
4232 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4233 "pkts=%llu:%llu dir=%s,%s\n",
4234 seq, orig_seq, ack, pd->p_len, ackskew,
4235 (unsigned long long)(*state)->packets[0],
4236 (unsigned long long)(*state)->packets[1],
4237 pd->dir == PF_IN ? "in" : "out",
4238 pd->dir == (*state)->direction ? "fwd" : "rev");
4239 printf("pf: State failure on: %c %c %c %c | %c %c\n",
4240 SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4241 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4243 (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4244 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4245 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4246 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4248 REASON_SET(reason, PFRES_BADSTATE);
4256 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4257 struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4259 struct tcphdr *th = pd->hdr.tcp;
4261 if (th->th_flags & TH_SYN)
4262 if (src->state < TCPS_SYN_SENT)
4263 src->state = TCPS_SYN_SENT;
4264 if (th->th_flags & TH_FIN)
4265 if (src->state < TCPS_CLOSING)
4266 src->state = TCPS_CLOSING;
4267 if (th->th_flags & TH_ACK) {
4268 if (dst->state == TCPS_SYN_SENT) {
4269 dst->state = TCPS_ESTABLISHED;
4270 if (src->state == TCPS_ESTABLISHED &&
4271 (*state)->src_node != NULL &&
4272 pf_src_connlimit(state)) {
4273 REASON_SET(reason, PFRES_SRCLIMIT);
4276 } else if (dst->state == TCPS_CLOSING) {
4277 dst->state = TCPS_FIN_WAIT_2;
4278 } else if (src->state == TCPS_SYN_SENT &&
4279 dst->state < TCPS_SYN_SENT) {
4281 * Handle a special sloppy case where we only see one
4282 * half of the connection. If there is a ACK after
4283 * the initial SYN without ever seeing a packet from
4284 * the destination, set the connection to established.
4286 dst->state = src->state = TCPS_ESTABLISHED;
4287 if ((*state)->src_node != NULL &&
4288 pf_src_connlimit(state)) {
4289 REASON_SET(reason, PFRES_SRCLIMIT);
4292 } else if (src->state == TCPS_CLOSING &&
4293 dst->state == TCPS_ESTABLISHED &&
4296 * Handle the closing of half connections where we
4297 * don't see the full bidirectional FIN/ACK+ACK
4300 dst->state = TCPS_CLOSING;
4303 if (th->th_flags & TH_RST)
4304 src->state = dst->state = TCPS_TIME_WAIT;
4306 /* update expire time */
4307 (*state)->expire = time_uptime;
4308 if (src->state >= TCPS_FIN_WAIT_2 &&
4309 dst->state >= TCPS_FIN_WAIT_2)
4310 (*state)->timeout = PFTM_TCP_CLOSED;
4311 else if (src->state >= TCPS_CLOSING &&
4312 dst->state >= TCPS_CLOSING)
4313 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4314 else if (src->state < TCPS_ESTABLISHED ||
4315 dst->state < TCPS_ESTABLISHED)
4316 (*state)->timeout = PFTM_TCP_OPENING;
4317 else if (src->state >= TCPS_CLOSING ||
4318 dst->state >= TCPS_CLOSING)
4319 (*state)->timeout = PFTM_TCP_CLOSING;
4321 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4327 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4328 struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4331 struct pf_state_key_cmp key;
4332 struct tcphdr *th = pd->hdr.tcp;
4334 struct pf_state_peer *src, *dst;
4335 struct pf_state_key *sk;
4337 bzero(&key, sizeof(key));
4339 key.proto = IPPROTO_TCP;
4340 if (direction == PF_IN) { /* wire side, straight */
4341 PF_ACPY(&key.addr[0], pd->src, key.af);
4342 PF_ACPY(&key.addr[1], pd->dst, key.af);
4343 key.port[0] = th->th_sport;
4344 key.port[1] = th->th_dport;
4345 } else { /* stack side, reverse */
4346 PF_ACPY(&key.addr[1], pd->src, key.af);
4347 PF_ACPY(&key.addr[0], pd->dst, key.af);
4348 key.port[1] = th->th_sport;
4349 key.port[0] = th->th_dport;
4352 STATE_LOOKUP(kif, &key, direction, *state, pd);
4354 if (direction == (*state)->direction) {
4355 src = &(*state)->src;
4356 dst = &(*state)->dst;
4358 src = &(*state)->dst;
4359 dst = &(*state)->src;
4362 sk = (*state)->key[pd->didx];
4364 if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4365 if (direction != (*state)->direction) {
4366 REASON_SET(reason, PFRES_SYNPROXY);
4367 return (PF_SYNPROXY_DROP);
4369 if (th->th_flags & TH_SYN) {
4370 if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4371 REASON_SET(reason, PFRES_SYNPROXY);
4374 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4375 pd->src, th->th_dport, th->th_sport,
4376 (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4377 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4378 REASON_SET(reason, PFRES_SYNPROXY);
4379 return (PF_SYNPROXY_DROP);
4380 } else if (!(th->th_flags & TH_ACK) ||
4381 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4382 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4383 REASON_SET(reason, PFRES_SYNPROXY);
4385 } else if ((*state)->src_node != NULL &&
4386 pf_src_connlimit(state)) {
4387 REASON_SET(reason, PFRES_SRCLIMIT);
4390 (*state)->src.state = PF_TCPS_PROXY_DST;
4392 if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4393 if (direction == (*state)->direction) {
4394 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4395 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4396 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4397 REASON_SET(reason, PFRES_SYNPROXY);
4400 (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4401 if ((*state)->dst.seqhi == 1)
4402 (*state)->dst.seqhi = htonl(arc4random());
4403 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4404 &sk->addr[pd->sidx], &sk->addr[pd->didx],
4405 sk->port[pd->sidx], sk->port[pd->didx],
4406 (*state)->dst.seqhi, 0, TH_SYN, 0,
4407 (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4408 REASON_SET(reason, PFRES_SYNPROXY);
4409 return (PF_SYNPROXY_DROP);
4410 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4412 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4413 REASON_SET(reason, PFRES_SYNPROXY);
4416 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4417 (*state)->dst.seqlo = ntohl(th->th_seq);
4418 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4419 pd->src, th->th_dport, th->th_sport,
4420 ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4421 TH_ACK, (*state)->src.max_win, 0, 0, 0,
4422 (*state)->tag, NULL);
4423 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4424 &sk->addr[pd->sidx], &sk->addr[pd->didx],
4425 sk->port[pd->sidx], sk->port[pd->didx],
4426 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4427 TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4428 (*state)->src.seqdiff = (*state)->dst.seqhi -
4429 (*state)->src.seqlo;
4430 (*state)->dst.seqdiff = (*state)->src.seqhi -
4431 (*state)->dst.seqlo;
4432 (*state)->src.seqhi = (*state)->src.seqlo +
4433 (*state)->dst.max_win;
4434 (*state)->dst.seqhi = (*state)->dst.seqlo +
4435 (*state)->src.max_win;
4436 (*state)->src.wscale = (*state)->dst.wscale = 0;
4437 (*state)->src.state = (*state)->dst.state =
4439 REASON_SET(reason, PFRES_SYNPROXY);
4440 return (PF_SYNPROXY_DROP);
4444 if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4445 dst->state >= TCPS_FIN_WAIT_2 &&
4446 src->state >= TCPS_FIN_WAIT_2) {
4447 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4448 printf("pf: state reuse ");
4449 pf_print_state(*state);
4450 pf_print_flags(th->th_flags);
4453 /* XXX make sure it's the same direction ?? */
4454 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4455 pf_unlink_state(*state, PF_ENTER_LOCKED);
4460 if ((*state)->state_flags & PFSTATE_SLOPPY) {
4461 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4464 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4465 ©back) == PF_DROP)
4469 /* translate source/destination address, if necessary */
4470 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4471 struct pf_state_key *nk = (*state)->key[pd->didx];
4473 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4474 nk->port[pd->sidx] != th->th_sport)
4475 pf_change_ap(m, pd->src, &th->th_sport,
4476 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
4477 nk->port[pd->sidx], 0, pd->af);
4479 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4480 nk->port[pd->didx] != th->th_dport)
4481 pf_change_ap(m, pd->dst, &th->th_dport,
4482 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
4483 nk->port[pd->didx], 0, pd->af);
4487 /* Copyback sequence modulation or stateful scrub changes if needed */
4489 m_copyback(m, off, sizeof(*th), (caddr_t)th);
4495 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
4496 struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4498 struct pf_state_peer *src, *dst;
4499 struct pf_state_key_cmp key;
4500 struct udphdr *uh = pd->hdr.udp;
4502 bzero(&key, sizeof(key));
4504 key.proto = IPPROTO_UDP;
4505 if (direction == PF_IN) { /* wire side, straight */
4506 PF_ACPY(&key.addr[0], pd->src, key.af);
4507 PF_ACPY(&key.addr[1], pd->dst, key.af);
4508 key.port[0] = uh->uh_sport;
4509 key.port[1] = uh->uh_dport;
4510 } else { /* stack side, reverse */
4511 PF_ACPY(&key.addr[1], pd->src, key.af);
4512 PF_ACPY(&key.addr[0], pd->dst, key.af);
4513 key.port[1] = uh->uh_sport;
4514 key.port[0] = uh->uh_dport;
4517 STATE_LOOKUP(kif, &key, direction, *state, pd);
4519 if (direction == (*state)->direction) {
4520 src = &(*state)->src;
4521 dst = &(*state)->dst;
4523 src = &(*state)->dst;
4524 dst = &(*state)->src;
4528 if (src->state < PFUDPS_SINGLE)
4529 src->state = PFUDPS_SINGLE;
4530 if (dst->state == PFUDPS_SINGLE)
4531 dst->state = PFUDPS_MULTIPLE;
4533 /* update expire time */
4534 (*state)->expire = time_uptime;
4535 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4536 (*state)->timeout = PFTM_UDP_MULTIPLE;
4538 (*state)->timeout = PFTM_UDP_SINGLE;
4540 /* translate source/destination address, if necessary */
4541 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4542 struct pf_state_key *nk = (*state)->key[pd->didx];
4544 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4545 nk->port[pd->sidx] != uh->uh_sport)
4546 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
4547 &uh->uh_sum, &nk->addr[pd->sidx],
4548 nk->port[pd->sidx], 1, pd->af);
4550 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4551 nk->port[pd->didx] != uh->uh_dport)
4552 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
4553 &uh->uh_sum, &nk->addr[pd->didx],
4554 nk->port[pd->didx], 1, pd->af);
4555 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4562 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
4563 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4565 struct pf_addr *saddr = pd->src, *daddr = pd->dst;
4566 u_int16_t icmpid = 0, *icmpsum;
4569 struct pf_state_key_cmp key;
4571 bzero(&key, sizeof(key));
4572 switch (pd->proto) {
4575 icmptype = pd->hdr.icmp->icmp_type;
4576 icmpid = pd->hdr.icmp->icmp_id;
4577 icmpsum = &pd->hdr.icmp->icmp_cksum;
4579 if (icmptype == ICMP_UNREACH ||
4580 icmptype == ICMP_SOURCEQUENCH ||
4581 icmptype == ICMP_REDIRECT ||
4582 icmptype == ICMP_TIMXCEED ||
4583 icmptype == ICMP_PARAMPROB)
4588 case IPPROTO_ICMPV6:
4589 icmptype = pd->hdr.icmp6->icmp6_type;
4590 icmpid = pd->hdr.icmp6->icmp6_id;
4591 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4593 if (icmptype == ICMP6_DST_UNREACH ||
4594 icmptype == ICMP6_PACKET_TOO_BIG ||
4595 icmptype == ICMP6_TIME_EXCEEDED ||
4596 icmptype == ICMP6_PARAM_PROB)
4605 * ICMP query/reply message not related to a TCP/UDP packet.
4606 * Search for an ICMP state.
4609 key.proto = pd->proto;
4610 key.port[0] = key.port[1] = icmpid;
4611 if (direction == PF_IN) { /* wire side, straight */
4612 PF_ACPY(&key.addr[0], pd->src, key.af);
4613 PF_ACPY(&key.addr[1], pd->dst, key.af);
4614 } else { /* stack side, reverse */
4615 PF_ACPY(&key.addr[1], pd->src, key.af);
4616 PF_ACPY(&key.addr[0], pd->dst, key.af);
4619 STATE_LOOKUP(kif, &key, direction, *state, pd);
4621 (*state)->expire = time_uptime;
4622 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4624 /* translate source/destination address, if necessary */
4625 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4626 struct pf_state_key *nk = (*state)->key[pd->didx];
4631 if (PF_ANEQ(pd->src,
4632 &nk->addr[pd->sidx], AF_INET))
4633 pf_change_a(&saddr->v4.s_addr,
4635 nk->addr[pd->sidx].v4.s_addr, 0);
4637 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4639 pf_change_a(&daddr->v4.s_addr,
4641 nk->addr[pd->didx].v4.s_addr, 0);
4644 pd->hdr.icmp->icmp_id) {
4645 pd->hdr.icmp->icmp_cksum =
4647 pd->hdr.icmp->icmp_cksum, icmpid,
4648 nk->port[pd->sidx], 0);
4649 pd->hdr.icmp->icmp_id =
4653 m_copyback(m, off, ICMP_MINLEN,
4654 (caddr_t )pd->hdr.icmp);
4659 if (PF_ANEQ(pd->src,
4660 &nk->addr[pd->sidx], AF_INET6))
4662 &pd->hdr.icmp6->icmp6_cksum,
4663 &nk->addr[pd->sidx], 0);
4665 if (PF_ANEQ(pd->dst,
4666 &nk->addr[pd->didx], AF_INET6))
4668 &pd->hdr.icmp6->icmp6_cksum,
4669 &nk->addr[pd->didx], 0);
4671 m_copyback(m, off, sizeof(struct icmp6_hdr),
4672 (caddr_t )pd->hdr.icmp6);
4681 * ICMP error message in response to a TCP/UDP packet.
4682 * Extract the inner TCP/UDP header and search for that state.
4685 struct pf_pdesc pd2;
4686 bzero(&pd2, sizeof pd2);
4691 struct ip6_hdr h2_6;
4698 /* Payload packet is from the opposite direction. */
4699 pd2.sidx = (direction == PF_IN) ? 1 : 0;
4700 pd2.didx = (direction == PF_IN) ? 0 : 1;
4704 /* offset of h2 in mbuf chain */
4705 ipoff2 = off + ICMP_MINLEN;
4707 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4708 NULL, reason, pd2.af)) {
4709 DPFPRINTF(PF_DEBUG_MISC,
4710 ("pf: ICMP error message too short "
4715 * ICMP error messages don't refer to non-first
4718 if (h2.ip_off & htons(IP_OFFMASK)) {
4719 REASON_SET(reason, PFRES_FRAG);
4723 /* offset of protocol header that follows h2 */
4724 off2 = ipoff2 + (h2.ip_hl << 2);
4726 pd2.proto = h2.ip_p;
4727 pd2.src = (struct pf_addr *)&h2.ip_src;
4728 pd2.dst = (struct pf_addr *)&h2.ip_dst;
4729 pd2.ip_sum = &h2.ip_sum;
4734 ipoff2 = off + sizeof(struct icmp6_hdr);
4736 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4737 NULL, reason, pd2.af)) {
4738 DPFPRINTF(PF_DEBUG_MISC,
4739 ("pf: ICMP error message too short "
4743 pd2.proto = h2_6.ip6_nxt;
4744 pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4745 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4747 off2 = ipoff2 + sizeof(h2_6);
4749 switch (pd2.proto) {
4750 case IPPROTO_FRAGMENT:
4752 * ICMPv6 error messages for
4753 * non-first fragments
4755 REASON_SET(reason, PFRES_FRAG);
4758 case IPPROTO_HOPOPTS:
4759 case IPPROTO_ROUTING:
4760 case IPPROTO_DSTOPTS: {
4761 /* get next header and header length */
4762 struct ip6_ext opt6;
4764 if (!pf_pull_hdr(m, off2, &opt6,
4765 sizeof(opt6), NULL, reason,
4767 DPFPRINTF(PF_DEBUG_MISC,
4768 ("pf: ICMPv6 short opt\n"));
4771 if (pd2.proto == IPPROTO_AH)
4772 off2 += (opt6.ip6e_len + 2) * 4;
4774 off2 += (opt6.ip6e_len + 1) * 8;
4775 pd2.proto = opt6.ip6e_nxt;
4776 /* goto the next header */
4783 } while (!terminal);
4788 switch (pd2.proto) {
4792 struct pf_state_peer *src, *dst;
4797 * Only the first 8 bytes of the TCP header can be
4798 * expected. Don't access any TCP header fields after
4799 * th_seq, an ackskew test is not possible.
4801 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4803 DPFPRINTF(PF_DEBUG_MISC,
4804 ("pf: ICMP error message too short "
4810 key.proto = IPPROTO_TCP;
4811 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4812 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4813 key.port[pd2.sidx] = th.th_sport;
4814 key.port[pd2.didx] = th.th_dport;
4816 STATE_LOOKUP(kif, &key, direction, *state, pd);
4818 if (direction == (*state)->direction) {
4819 src = &(*state)->dst;
4820 dst = &(*state)->src;
4822 src = &(*state)->src;
4823 dst = &(*state)->dst;
4826 if (src->wscale && dst->wscale)
4827 dws = dst->wscale & PF_WSCALE_MASK;
4831 /* Demodulate sequence number */
4832 seq = ntohl(th.th_seq) - src->seqdiff;
4834 pf_change_a(&th.th_seq, icmpsum,
4839 if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4840 (!SEQ_GEQ(src->seqhi, seq) ||
4841 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4842 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4843 printf("pf: BAD ICMP %d:%d ",
4844 icmptype, pd->hdr.icmp->icmp_code);
4845 pf_print_host(pd->src, 0, pd->af);
4847 pf_print_host(pd->dst, 0, pd->af);
4849 pf_print_state(*state);
4850 printf(" seq=%u\n", seq);
4852 REASON_SET(reason, PFRES_BADSTATE);
4855 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4856 printf("pf: OK ICMP %d:%d ",
4857 icmptype, pd->hdr.icmp->icmp_code);
4858 pf_print_host(pd->src, 0, pd->af);
4860 pf_print_host(pd->dst, 0, pd->af);
4862 pf_print_state(*state);
4863 printf(" seq=%u\n", seq);
4867 /* translate source/destination address, if necessary */
4868 if ((*state)->key[PF_SK_WIRE] !=
4869 (*state)->key[PF_SK_STACK]) {
4870 struct pf_state_key *nk =
4871 (*state)->key[pd->didx];
4873 if (PF_ANEQ(pd2.src,
4874 &nk->addr[pd2.sidx], pd2.af) ||
4875 nk->port[pd2.sidx] != th.th_sport)
4876 pf_change_icmp(pd2.src, &th.th_sport,
4877 daddr, &nk->addr[pd2.sidx],
4878 nk->port[pd2.sidx], NULL,
4879 pd2.ip_sum, icmpsum,
4880 pd->ip_sum, 0, pd2.af);
4882 if (PF_ANEQ(pd2.dst,
4883 &nk->addr[pd2.didx], pd2.af) ||
4884 nk->port[pd2.didx] != th.th_dport)
4885 pf_change_icmp(pd2.dst, &th.th_dport,
4886 saddr, &nk->addr[pd2.didx],
4887 nk->port[pd2.didx], NULL,
4888 pd2.ip_sum, icmpsum,
4889 pd->ip_sum, 0, pd2.af);
4897 m_copyback(m, off, ICMP_MINLEN,
4898 (caddr_t )pd->hdr.icmp);
4899 m_copyback(m, ipoff2, sizeof(h2),
4906 sizeof(struct icmp6_hdr),
4907 (caddr_t )pd->hdr.icmp6);
4908 m_copyback(m, ipoff2, sizeof(h2_6),
4913 m_copyback(m, off2, 8, (caddr_t)&th);
4922 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
4923 NULL, reason, pd2.af)) {
4924 DPFPRINTF(PF_DEBUG_MISC,
4925 ("pf: ICMP error message too short "
4931 key.proto = IPPROTO_UDP;
4932 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4933 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4934 key.port[pd2.sidx] = uh.uh_sport;
4935 key.port[pd2.didx] = uh.uh_dport;
4937 STATE_LOOKUP(kif, &key, direction, *state, pd);
4939 /* translate source/destination address, if necessary */
4940 if ((*state)->key[PF_SK_WIRE] !=
4941 (*state)->key[PF_SK_STACK]) {
4942 struct pf_state_key *nk =
4943 (*state)->key[pd->didx];
4945 if (PF_ANEQ(pd2.src,
4946 &nk->addr[pd2.sidx], pd2.af) ||
4947 nk->port[pd2.sidx] != uh.uh_sport)
4948 pf_change_icmp(pd2.src, &uh.uh_sport,
4949 daddr, &nk->addr[pd2.sidx],
4950 nk->port[pd2.sidx], &uh.uh_sum,
4951 pd2.ip_sum, icmpsum,
4952 pd->ip_sum, 1, pd2.af);
4954 if (PF_ANEQ(pd2.dst,
4955 &nk->addr[pd2.didx], pd2.af) ||
4956 nk->port[pd2.didx] != uh.uh_dport)
4957 pf_change_icmp(pd2.dst, &uh.uh_dport,
4958 saddr, &nk->addr[pd2.didx],
4959 nk->port[pd2.didx], &uh.uh_sum,
4960 pd2.ip_sum, icmpsum,
4961 pd->ip_sum, 1, pd2.af);
4966 m_copyback(m, off, ICMP_MINLEN,
4967 (caddr_t )pd->hdr.icmp);
4968 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4974 sizeof(struct icmp6_hdr),
4975 (caddr_t )pd->hdr.icmp6);
4976 m_copyback(m, ipoff2, sizeof(h2_6),
4981 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
4987 case IPPROTO_ICMP: {
4990 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
4991 NULL, reason, pd2.af)) {
4992 DPFPRINTF(PF_DEBUG_MISC,
4993 ("pf: ICMP error message too short i"
4999 key.proto = IPPROTO_ICMP;
5000 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5001 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5002 key.port[0] = key.port[1] = iih.icmp_id;
5004 STATE_LOOKUP(kif, &key, direction, *state, pd);
5006 /* translate source/destination address, if necessary */
5007 if ((*state)->key[PF_SK_WIRE] !=
5008 (*state)->key[PF_SK_STACK]) {
5009 struct pf_state_key *nk =
5010 (*state)->key[pd->didx];
5012 if (PF_ANEQ(pd2.src,
5013 &nk->addr[pd2.sidx], pd2.af) ||
5014 nk->port[pd2.sidx] != iih.icmp_id)
5015 pf_change_icmp(pd2.src, &iih.icmp_id,
5016 daddr, &nk->addr[pd2.sidx],
5017 nk->port[pd2.sidx], NULL,
5018 pd2.ip_sum, icmpsum,
5019 pd->ip_sum, 0, AF_INET);
5021 if (PF_ANEQ(pd2.dst,
5022 &nk->addr[pd2.didx], pd2.af) ||
5023 nk->port[pd2.didx] != iih.icmp_id)
5024 pf_change_icmp(pd2.dst, &iih.icmp_id,
5025 saddr, &nk->addr[pd2.didx],
5026 nk->port[pd2.didx], NULL,
5027 pd2.ip_sum, icmpsum,
5028 pd->ip_sum, 0, AF_INET);
5030 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
5031 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5032 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
5039 case IPPROTO_ICMPV6: {
5040 struct icmp6_hdr iih;
5042 if (!pf_pull_hdr(m, off2, &iih,
5043 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
5044 DPFPRINTF(PF_DEBUG_MISC,
5045 ("pf: ICMP error message too short "
5051 key.proto = IPPROTO_ICMPV6;
5052 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5053 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5054 key.port[0] = key.port[1] = iih.icmp6_id;
5056 STATE_LOOKUP(kif, &key, direction, *state, pd);
5058 /* translate source/destination address, if necessary */
5059 if ((*state)->key[PF_SK_WIRE] !=
5060 (*state)->key[PF_SK_STACK]) {
5061 struct pf_state_key *nk =
5062 (*state)->key[pd->didx];
5064 if (PF_ANEQ(pd2.src,
5065 &nk->addr[pd2.sidx], pd2.af) ||
5066 nk->port[pd2.sidx] != iih.icmp6_id)
5067 pf_change_icmp(pd2.src, &iih.icmp6_id,
5068 daddr, &nk->addr[pd2.sidx],
5069 nk->port[pd2.sidx], NULL,
5070 pd2.ip_sum, icmpsum,
5071 pd->ip_sum, 0, AF_INET6);
5073 if (PF_ANEQ(pd2.dst,
5074 &nk->addr[pd2.didx], pd2.af) ||
5075 nk->port[pd2.didx] != iih.icmp6_id)
5076 pf_change_icmp(pd2.dst, &iih.icmp6_id,
5077 saddr, &nk->addr[pd2.didx],
5078 nk->port[pd2.didx], NULL,
5079 pd2.ip_sum, icmpsum,
5080 pd->ip_sum, 0, AF_INET6);
5082 m_copyback(m, off, sizeof(struct icmp6_hdr),
5083 (caddr_t)pd->hdr.icmp6);
5084 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
5085 m_copyback(m, off2, sizeof(struct icmp6_hdr),
5094 key.proto = pd2.proto;
5095 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5096 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5097 key.port[0] = key.port[1] = 0;
5099 STATE_LOOKUP(kif, &key, direction, *state, pd);
5101 /* translate source/destination address, if necessary */
5102 if ((*state)->key[PF_SK_WIRE] !=
5103 (*state)->key[PF_SK_STACK]) {
5104 struct pf_state_key *nk =
5105 (*state)->key[pd->didx];
5107 if (PF_ANEQ(pd2.src,
5108 &nk->addr[pd2.sidx], pd2.af))
5109 pf_change_icmp(pd2.src, NULL, daddr,
5110 &nk->addr[pd2.sidx], 0, NULL,
5111 pd2.ip_sum, icmpsum,
5112 pd->ip_sum, 0, pd2.af);
5114 if (PF_ANEQ(pd2.dst,
5115 &nk->addr[pd2.didx], pd2.af))
5116 pf_change_icmp(pd2.dst, NULL, saddr,
5117 &nk->addr[pd2.didx], 0, NULL,
5118 pd2.ip_sum, icmpsum,
5119 pd->ip_sum, 0, pd2.af);
5124 m_copyback(m, off, ICMP_MINLEN,
5125 (caddr_t)pd->hdr.icmp);
5126 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5132 sizeof(struct icmp6_hdr),
5133 (caddr_t )pd->hdr.icmp6);
5134 m_copyback(m, ipoff2, sizeof(h2_6),
5148 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
5149 struct mbuf *m, struct pf_pdesc *pd)
5151 struct pf_state_peer *src, *dst;
5152 struct pf_state_key_cmp key;
5154 bzero(&key, sizeof(key));
5156 key.proto = pd->proto;
5157 if (direction == PF_IN) {
5158 PF_ACPY(&key.addr[0], pd->src, key.af);
5159 PF_ACPY(&key.addr[1], pd->dst, key.af);
5160 key.port[0] = key.port[1] = 0;
5162 PF_ACPY(&key.addr[1], pd->src, key.af);
5163 PF_ACPY(&key.addr[0], pd->dst, key.af);
5164 key.port[1] = key.port[0] = 0;
5167 STATE_LOOKUP(kif, &key, direction, *state, pd);
5169 if (direction == (*state)->direction) {
5170 src = &(*state)->src;
5171 dst = &(*state)->dst;
5173 src = &(*state)->dst;
5174 dst = &(*state)->src;
5178 if (src->state < PFOTHERS_SINGLE)
5179 src->state = PFOTHERS_SINGLE;
5180 if (dst->state == PFOTHERS_SINGLE)
5181 dst->state = PFOTHERS_MULTIPLE;
5183 /* update expire time */
5184 (*state)->expire = time_uptime;
5185 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5186 (*state)->timeout = PFTM_OTHER_MULTIPLE;
5188 (*state)->timeout = PFTM_OTHER_SINGLE;
5190 /* translate source/destination address, if necessary */
5191 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5192 struct pf_state_key *nk = (*state)->key[pd->didx];
5194 KASSERT(nk, ("%s: nk is null", __func__));
5195 KASSERT(pd, ("%s: pd is null", __func__));
5196 KASSERT(pd->src, ("%s: pd->src is null", __func__));
5197 KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5201 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5202 pf_change_a(&pd->src->v4.s_addr,
5204 nk->addr[pd->sidx].v4.s_addr,
5208 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5209 pf_change_a(&pd->dst->v4.s_addr,
5211 nk->addr[pd->didx].v4.s_addr,
5218 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5219 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5221 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5222 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5230 * ipoff and off are measured from the start of the mbuf chain.
5231 * h must be at "ipoff" on the mbuf chain.
5234 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5235 u_short *actionp, u_short *reasonp, sa_family_t af)
5240 struct ip *h = mtod(m, struct ip *);
5241 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5245 ACTION_SET(actionp, PF_PASS);
5247 ACTION_SET(actionp, PF_DROP);
5248 REASON_SET(reasonp, PFRES_FRAG);
5252 if (m->m_pkthdr.len < off + len ||
5253 ntohs(h->ip_len) < off + len) {
5254 ACTION_SET(actionp, PF_DROP);
5255 REASON_SET(reasonp, PFRES_SHORT);
5263 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
5265 if (m->m_pkthdr.len < off + len ||
5266 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5267 (unsigned)(off + len)) {
5268 ACTION_SET(actionp, PF_DROP);
5269 REASON_SET(reasonp, PFRES_SHORT);
5276 m_copydata(m, off, len, p);
5282 pf_routable_oldmpath(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5285 struct radix_node_head *rnh;
5286 struct sockaddr_in *dst;
5290 struct sockaddr_in6 *dst6;
5291 struct route_in6 ro;
5295 struct radix_node *rn;
5300 /* XXX: stick to table 0 for now */
5301 rnh = rt_tables_get_rnh(0, af);
5302 if (rnh != NULL && rn_mpath_capable(rnh))
5304 bzero(&ro, sizeof(ro));
5307 dst = satosin(&ro.ro_dst);
5308 dst->sin_family = AF_INET;
5309 dst->sin_len = sizeof(*dst);
5310 dst->sin_addr = addr->v4;
5315 * Skip check for addresses with embedded interface scope,
5316 * as they would always match anyway.
5318 if (IN6_IS_SCOPE_EMBED(&addr->v6))
5320 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5321 dst6->sin6_family = AF_INET6;
5322 dst6->sin6_len = sizeof(*dst6);
5323 dst6->sin6_addr = addr->v6;
5330 /* Skip checks for ipsec interfaces */
5331 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5337 in6_rtalloc_ign(&ro, 0, rtableid);
5342 in_rtalloc_ign((struct route *)&ro, 0, rtableid);
5347 if (ro.ro_rt != NULL) {
5348 /* No interface given, this is a no-route check */
5352 if (kif->pfik_ifp == NULL) {
5357 /* Perform uRPF check if passed input interface */
5359 rn = (struct radix_node *)ro.ro_rt;
5361 rt = (struct rtentry *)rn;
5364 if (kif->pfik_ifp == ifp)
5366 rn = rn_mpath_next(rn);
5367 } while (check_mpath == 1 && rn != NULL && ret == 0);
5371 if (ro.ro_rt != NULL)
5378 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5382 struct nhop4_basic nh4;
5385 struct nhop6_basic nh6;
5389 struct radix_node_head *rnh;
5391 /* XXX: stick to table 0 for now */
5392 rnh = rt_tables_get_rnh(0, af);
5393 if (rnh != NULL && rn_mpath_capable(rnh))
5394 return (pf_routable_oldmpath(addr, af, kif, rtableid));
5397 * Skip check for addresses with embedded interface scope,
5398 * as they would always match anyway.
5400 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
5403 if (af != AF_INET && af != AF_INET6)
5406 /* Skip checks for ipsec interfaces */
5407 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5415 if (fib6_lookup_nh_basic(rtableid, &addr->v6, 0, 0, 0, &nh6)!=0)
5422 if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) != 0)
5429 /* No interface given, this is a no-route check */
5433 if (kif->pfik_ifp == NULL)
5436 /* Perform uRPF check if passed input interface */
5437 if (kif->pfik_ifp == ifp)
5444 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5445 struct pf_state *s, struct pf_pdesc *pd)
5447 struct mbuf *m0, *m1;
5448 struct sockaddr_in dst;
5450 struct ifnet *ifp = NULL;
5451 struct pf_addr naddr;
5452 struct pf_src_node *sn = NULL;
5454 uint16_t ip_len, ip_off;
5456 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5457 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5460 if ((pd->pf_mtag == NULL &&
5461 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5462 pd->pf_mtag->routed++ > 3) {
5468 if (r->rt == PF_DUPTO) {
5469 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5475 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5483 ip = mtod(m0, struct ip *);
5485 bzero(&dst, sizeof(dst));
5486 dst.sin_family = AF_INET;
5487 dst.sin_len = sizeof(dst);
5488 dst.sin_addr = ip->ip_dst;
5490 if (TAILQ_EMPTY(&r->rpool.list)) {
5491 DPFPRINTF(PF_DEBUG_URGENT,
5492 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5496 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5498 if (!PF_AZERO(&naddr, AF_INET))
5499 dst.sin_addr.s_addr = naddr.v4.s_addr;
5500 ifp = r->rpool.cur->kif ?
5501 r->rpool.cur->kif->pfik_ifp : NULL;
5503 if (!PF_AZERO(&s->rt_addr, AF_INET))
5504 dst.sin_addr.s_addr =
5505 s->rt_addr.v4.s_addr;
5506 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5513 if (pf_test(PF_OUT, 0, ifp, &m0, NULL) != PF_PASS)
5515 else if (m0 == NULL)
5517 if (m0->m_len < sizeof(struct ip)) {
5518 DPFPRINTF(PF_DEBUG_URGENT,
5519 ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5522 ip = mtod(m0, struct ip *);
5525 if (ifp->if_flags & IFF_LOOPBACK)
5526 m0->m_flags |= M_SKIP_FIREWALL;
5528 ip_len = ntohs(ip->ip_len);
5529 ip_off = ntohs(ip->ip_off);
5531 /* Copied from FreeBSD 10.0-CURRENT ip_output. */
5532 m0->m_pkthdr.csum_flags |= CSUM_IP;
5533 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5534 in_delayed_cksum(m0);
5535 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5538 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5539 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
5540 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5545 * If small enough for interface, or the interface will take
5546 * care of the fragmentation for us, we can just send directly.
5548 if (ip_len <= ifp->if_mtu ||
5549 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
5551 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5552 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5553 m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5555 m_clrprotoflags(m0); /* Avoid confusing lower layers. */
5556 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5560 /* Balk when DF bit is set or the interface didn't support TSO. */
5561 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5563 KMOD_IPSTAT_INC(ips_cantfrag);
5564 if (r->rt != PF_DUPTO) {
5565 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5572 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5576 for (; m0; m0 = m1) {
5578 m0->m_nextpkt = NULL;
5580 m_clrprotoflags(m0);
5581 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5587 KMOD_IPSTAT_INC(ips_fragmented);
5590 if (r->rt != PF_DUPTO)
5605 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5606 struct pf_state *s, struct pf_pdesc *pd)
5609 struct sockaddr_in6 dst;
5610 struct ip6_hdr *ip6;
5611 struct ifnet *ifp = NULL;
5612 struct pf_addr naddr;
5613 struct pf_src_node *sn = NULL;
5615 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5616 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5619 if ((pd->pf_mtag == NULL &&
5620 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5621 pd->pf_mtag->routed++ > 3) {
5627 if (r->rt == PF_DUPTO) {
5628 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5634 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5642 ip6 = mtod(m0, struct ip6_hdr *);
5644 bzero(&dst, sizeof(dst));
5645 dst.sin6_family = AF_INET6;
5646 dst.sin6_len = sizeof(dst);
5647 dst.sin6_addr = ip6->ip6_dst;
5649 if (TAILQ_EMPTY(&r->rpool.list)) {
5650 DPFPRINTF(PF_DEBUG_URGENT,
5651 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5655 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5657 if (!PF_AZERO(&naddr, AF_INET6))
5658 PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5660 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5662 if (!PF_AZERO(&s->rt_addr, AF_INET6))
5663 PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5664 &s->rt_addr, AF_INET6);
5665 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5675 if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, NULL) != PF_PASS)
5677 else if (m0 == NULL)
5679 if (m0->m_len < sizeof(struct ip6_hdr)) {
5680 DPFPRINTF(PF_DEBUG_URGENT,
5681 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5685 ip6 = mtod(m0, struct ip6_hdr *);
5688 if (ifp->if_flags & IFF_LOOPBACK)
5689 m0->m_flags |= M_SKIP_FIREWALL;
5691 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
5692 ~ifp->if_hwassist) {
5693 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
5694 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
5695 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
5699 * If the packet is too large for the outgoing interface,
5700 * send back an icmp6 error.
5702 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5703 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5704 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5705 nd6_output_ifp(ifp, ifp, m0, &dst, NULL);
5707 in6_ifstat_inc(ifp, ifs6_in_toobig);
5708 if (r->rt != PF_DUPTO)
5709 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5715 if (r->rt != PF_DUPTO)
5729 * FreeBSD supports cksum offloads for the following drivers.
5730 * em(4), fxp(4), lge(4), ndis(4), nge(4), re(4), ti(4), txp(4), xl(4)
5732 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5733 * network driver performed cksum including pseudo header, need to verify
5736 * network driver performed cksum, needs to additional pseudo header
5737 * cksum computation with partial csum_data(i.e. lack of H/W support for
5738 * pseudo header, for instance hme(4), sk(4) and possibly gem(4))
5740 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5741 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5743 * Also, set csum_data to 0xffff to force cksum validation.
5746 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5752 if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5754 if (m->m_pkthdr.len < off + len)
5759 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5760 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5761 sum = m->m_pkthdr.csum_data;
5763 ip = mtod(m, struct ip *);
5764 sum = in_pseudo(ip->ip_src.s_addr,
5765 ip->ip_dst.s_addr, htonl((u_short)len +
5766 m->m_pkthdr.csum_data + IPPROTO_TCP));
5773 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5774 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5775 sum = m->m_pkthdr.csum_data;
5777 ip = mtod(m, struct ip *);
5778 sum = in_pseudo(ip->ip_src.s_addr,
5779 ip->ip_dst.s_addr, htonl((u_short)len +
5780 m->m_pkthdr.csum_data + IPPROTO_UDP));
5788 case IPPROTO_ICMPV6:
5798 if (p == IPPROTO_ICMP) {
5803 sum = in_cksum(m, len);
5807 if (m->m_len < sizeof(struct ip))
5809 sum = in4_cksum(m, p, off, len);
5814 if (m->m_len < sizeof(struct ip6_hdr))
5816 sum = in6_cksum(m, p, off, len);
5827 KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5832 KMOD_UDPSTAT_INC(udps_badsum);
5838 KMOD_ICMPSTAT_INC(icps_checksum);
5843 case IPPROTO_ICMPV6:
5845 KMOD_ICMP6STAT_INC(icp6s_checksum);
5852 if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5853 m->m_pkthdr.csum_flags |=
5854 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5855 m->m_pkthdr.csum_data = 0xffff;
5864 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5866 struct pfi_kif *kif;
5867 u_short action, reason = 0, log = 0;
5868 struct mbuf *m = *m0;
5869 struct ip *h = NULL;
5870 struct m_tag *ipfwtag;
5871 struct pf_rule *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5872 struct pf_state *s = NULL;
5873 struct pf_ruleset *ruleset = NULL;
5875 int off, dirndx, pqid = 0;
5879 if (!V_pf_status.running)
5882 memset(&pd, 0, sizeof(pd));
5884 kif = (struct pfi_kif *)ifp->if_pf_kif;
5887 DPFPRINTF(PF_DEBUG_URGENT,
5888 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5891 if (kif->pfik_flags & PFI_IFLAG_SKIP)
5894 if (m->m_flags & M_SKIP_FIREWALL)
5897 pd.pf_mtag = pf_find_mtag(m);
5901 if (ip_divert_ptr != NULL &&
5902 ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5903 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5904 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5905 if (pd.pf_mtag == NULL &&
5906 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5910 pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5911 m_tag_delete(m, ipfwtag);
5913 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5914 m->m_flags |= M_FASTFWD_OURS;
5915 pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5917 } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5918 /* We do IP header normalization and packet reassembly here */
5922 m = *m0; /* pf_normalize messes with m0 */
5923 h = mtod(m, struct ip *);
5925 off = h->ip_hl << 2;
5926 if (off < (int)sizeof(struct ip)) {
5928 REASON_SET(&reason, PFRES_SHORT);
5933 pd.src = (struct pf_addr *)&h->ip_src;
5934 pd.dst = (struct pf_addr *)&h->ip_dst;
5935 pd.sport = pd.dport = NULL;
5936 pd.ip_sum = &h->ip_sum;
5937 pd.proto_sum = NULL;
5940 pd.sidx = (dir == PF_IN) ? 0 : 1;
5941 pd.didx = (dir == PF_IN) ? 1 : 0;
5943 pd.tos = h->ip_tos & ~IPTOS_ECN_MASK;
5944 pd.tot_len = ntohs(h->ip_len);
5946 /* handle fragments that didn't get reassembled by normalization */
5947 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
5948 action = pf_test_fragment(&r, dir, kif, m, h,
5959 if (!pf_pull_hdr(m, off, &th, sizeof(th),
5960 &action, &reason, AF_INET)) {
5961 log = action != PF_PASS;
5964 pd.p_len = pd.tot_len - off - (th.th_off << 2);
5965 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
5967 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
5968 if (action == PF_DROP)
5970 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
5972 if (action == PF_PASS) {
5973 if (pfsync_update_state_ptr != NULL)
5974 pfsync_update_state_ptr(s);
5978 } else if (s == NULL)
5979 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5988 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
5989 &action, &reason, AF_INET)) {
5990 log = action != PF_PASS;
5993 if (uh.uh_dport == 0 ||
5994 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
5995 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
5997 REASON_SET(&reason, PFRES_SHORT);
6000 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6001 if (action == PF_PASS) {
6002 if (pfsync_update_state_ptr != NULL)
6003 pfsync_update_state_ptr(s);
6007 } else if (s == NULL)
6008 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6013 case IPPROTO_ICMP: {
6017 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
6018 &action, &reason, AF_INET)) {
6019 log = action != PF_PASS;
6022 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
6024 if (action == PF_PASS) {
6025 if (pfsync_update_state_ptr != NULL)
6026 pfsync_update_state_ptr(s);
6030 } else if (s == NULL)
6031 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6037 case IPPROTO_ICMPV6: {
6039 DPFPRINTF(PF_DEBUG_MISC,
6040 ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
6046 action = pf_test_state_other(&s, dir, kif, m, &pd);
6047 if (action == PF_PASS) {
6048 if (pfsync_update_state_ptr != NULL)
6049 pfsync_update_state_ptr(s);
6053 } else if (s == NULL)
6054 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6061 if (action == PF_PASS && h->ip_hl > 5 &&
6062 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6064 REASON_SET(&reason, PFRES_IPOPTIONS);
6066 DPFPRINTF(PF_DEBUG_MISC,
6067 ("pf: dropping packet with ip options\n"));
6070 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6072 REASON_SET(&reason, PFRES_MEMORY);
6074 if (r->rtableid >= 0)
6075 M_SETFIB(m, r->rtableid);
6077 if (r->scrub_flags & PFSTATE_SETPRIO) {
6078 if (pd.tos & IPTOS_LOWDELAY)
6080 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6082 REASON_SET(&reason, PFRES_MEMORY);
6084 DPFPRINTF(PF_DEBUG_MISC,
6085 ("pf: failed to allocate 802.1q mtag\n"));
6090 if (action == PF_PASS && r->qid) {
6091 if (pd.pf_mtag == NULL &&
6092 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6094 REASON_SET(&reason, PFRES_MEMORY);
6097 pd.pf_mtag->qid_hash = pf_state_hash(s);
6098 if (pqid || (pd.tos & IPTOS_LOWDELAY))
6099 pd.pf_mtag->qid = r->pqid;
6101 pd.pf_mtag->qid = r->qid;
6102 /* Add hints for ecn. */
6103 pd.pf_mtag->hdr = h;
6110 * connections redirected to loopback should not match sockets
6111 * bound specifically to loopback due to security implications,
6112 * see tcp_input() and in_pcblookup_listen().
6114 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6115 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6116 (s->nat_rule.ptr->action == PF_RDR ||
6117 s->nat_rule.ptr->action == PF_BINAT) &&
6118 (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
6119 m->m_flags |= M_SKIP_FIREWALL;
6121 if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
6122 !PACKET_LOOPED(&pd)) {
6124 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
6125 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
6126 if (ipfwtag != NULL) {
6127 ((struct ipfw_rule_ref *)(ipfwtag+1))->info =
6128 ntohs(r->divert.port);
6129 ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
6134 m_tag_prepend(m, ipfwtag);
6135 if (m->m_flags & M_FASTFWD_OURS) {
6136 if (pd.pf_mtag == NULL &&
6137 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6139 REASON_SET(&reason, PFRES_MEMORY);
6141 DPFPRINTF(PF_DEBUG_MISC,
6142 ("pf: failed to allocate tag\n"));
6144 pd.pf_mtag->flags |=
6145 PF_FASTFWD_OURS_PRESENT;
6146 m->m_flags &= ~M_FASTFWD_OURS;
6149 ip_divert_ptr(*m0, dir == PF_IN ? DIR_IN : DIR_OUT);
6154 /* XXX: ipfw has the same behaviour! */
6156 REASON_SET(&reason, PFRES_MEMORY);
6158 DPFPRINTF(PF_DEBUG_MISC,
6159 ("pf: failed to allocate divert tag\n"));
6166 if (s != NULL && s->nat_rule.ptr != NULL &&
6167 s->nat_rule.ptr->log & PF_LOG_ALL)
6168 lr = s->nat_rule.ptr;
6171 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
6175 kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6176 kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
6178 if (action == PF_PASS || r->action == PF_DROP) {
6179 dirndx = (dir == PF_OUT);
6180 r->packets[dirndx]++;
6181 r->bytes[dirndx] += pd.tot_len;
6183 a->packets[dirndx]++;
6184 a->bytes[dirndx] += pd.tot_len;
6187 if (s->nat_rule.ptr != NULL) {
6188 s->nat_rule.ptr->packets[dirndx]++;
6189 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6191 if (s->src_node != NULL) {
6192 s->src_node->packets[dirndx]++;
6193 s->src_node->bytes[dirndx] += pd.tot_len;
6195 if (s->nat_src_node != NULL) {
6196 s->nat_src_node->packets[dirndx]++;
6197 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6199 dirndx = (dir == s->direction) ? 0 : 1;
6200 s->packets[dirndx]++;
6201 s->bytes[dirndx] += pd.tot_len;
6204 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6205 if (nr != NULL && r == &V_pf_default_rule)
6207 if (tr->src.addr.type == PF_ADDR_TABLE)
6208 pfr_update_stats(tr->src.addr.p.tbl,
6209 (s == NULL) ? pd.src :
6210 &s->key[(s->direction == PF_IN)]->
6211 addr[(s->direction == PF_OUT)],
6212 pd.af, pd.tot_len, dir == PF_OUT,
6213 r->action == PF_PASS, tr->src.neg);
6214 if (tr->dst.addr.type == PF_ADDR_TABLE)
6215 pfr_update_stats(tr->dst.addr.p.tbl,
6216 (s == NULL) ? pd.dst :
6217 &s->key[(s->direction == PF_IN)]->
6218 addr[(s->direction == PF_IN)],
6219 pd.af, pd.tot_len, dir == PF_OUT,
6220 r->action == PF_PASS, tr->dst.neg);
6224 case PF_SYNPROXY_DROP:
6235 /* pf_route() returns unlocked. */
6237 pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
6251 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6253 struct pfi_kif *kif;
6254 u_short action, reason = 0, log = 0;
6255 struct mbuf *m = *m0, *n = NULL;
6257 struct ip6_hdr *h = NULL;
6258 struct pf_rule *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6259 struct pf_state *s = NULL;
6260 struct pf_ruleset *ruleset = NULL;
6262 int off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0;
6266 if (!V_pf_status.running)
6269 memset(&pd, 0, sizeof(pd));
6270 pd.pf_mtag = pf_find_mtag(m);
6272 if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6275 kif = (struct pfi_kif *)ifp->if_pf_kif;
6277 DPFPRINTF(PF_DEBUG_URGENT,
6278 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6281 if (kif->pfik_flags & PFI_IFLAG_SKIP)
6284 if (m->m_flags & M_SKIP_FIREWALL)
6289 /* We do IP header normalization and packet reassembly here */
6290 if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6294 m = *m0; /* pf_normalize messes with m0 */
6295 h = mtod(m, struct ip6_hdr *);
6299 * we do not support jumbogram yet. if we keep going, zero ip6_plen
6300 * will do something bad, so drop the packet for now.
6302 if (htons(h->ip6_plen) == 0) {
6304 REASON_SET(&reason, PFRES_NORM); /*XXX*/
6309 pd.src = (struct pf_addr *)&h->ip6_src;
6310 pd.dst = (struct pf_addr *)&h->ip6_dst;
6311 pd.sport = pd.dport = NULL;
6313 pd.proto_sum = NULL;
6315 pd.sidx = (dir == PF_IN) ? 0 : 1;
6316 pd.didx = (dir == PF_IN) ? 1 : 0;
6319 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6321 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6322 pd.proto = h->ip6_nxt;
6325 case IPPROTO_FRAGMENT:
6326 action = pf_test_fragment(&r, dir, kif, m, h,
6328 if (action == PF_DROP)
6329 REASON_SET(&reason, PFRES_FRAG);
6331 case IPPROTO_ROUTING: {
6332 struct ip6_rthdr rthdr;
6335 DPFPRINTF(PF_DEBUG_MISC,
6336 ("pf: IPv6 more than one rthdr\n"));
6338 REASON_SET(&reason, PFRES_IPOPTIONS);
6342 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6344 DPFPRINTF(PF_DEBUG_MISC,
6345 ("pf: IPv6 short rthdr\n"));
6347 REASON_SET(&reason, PFRES_SHORT);
6351 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6352 DPFPRINTF(PF_DEBUG_MISC,
6353 ("pf: IPv6 rthdr0\n"));
6355 REASON_SET(&reason, PFRES_IPOPTIONS);
6362 case IPPROTO_HOPOPTS:
6363 case IPPROTO_DSTOPTS: {
6364 /* get next header and header length */
6365 struct ip6_ext opt6;
6367 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6368 NULL, &reason, pd.af)) {
6369 DPFPRINTF(PF_DEBUG_MISC,
6370 ("pf: IPv6 short opt\n"));
6375 if (pd.proto == IPPROTO_AH)
6376 off += (opt6.ip6e_len + 2) * 4;
6378 off += (opt6.ip6e_len + 1) * 8;
6379 pd.proto = opt6.ip6e_nxt;
6380 /* goto the next header */
6387 } while (!terminal);
6389 /* if there's no routing header, use unmodified mbuf for checksumming */
6399 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6400 &action, &reason, AF_INET6)) {
6401 log = action != PF_PASS;
6404 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6405 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6406 if (action == PF_DROP)
6408 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6410 if (action == PF_PASS) {
6411 if (pfsync_update_state_ptr != NULL)
6412 pfsync_update_state_ptr(s);
6416 } else if (s == NULL)
6417 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6426 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6427 &action, &reason, AF_INET6)) {
6428 log = action != PF_PASS;
6431 if (uh.uh_dport == 0 ||
6432 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6433 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6435 REASON_SET(&reason, PFRES_SHORT);
6438 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6439 if (action == PF_PASS) {
6440 if (pfsync_update_state_ptr != NULL)
6441 pfsync_update_state_ptr(s);
6445 } else if (s == NULL)
6446 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6451 case IPPROTO_ICMP: {
6453 DPFPRINTF(PF_DEBUG_MISC,
6454 ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6458 case IPPROTO_ICMPV6: {
6459 struct icmp6_hdr ih;
6462 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6463 &action, &reason, AF_INET6)) {
6464 log = action != PF_PASS;
6467 action = pf_test_state_icmp(&s, dir, kif,
6468 m, off, h, &pd, &reason);
6469 if (action == PF_PASS) {
6470 if (pfsync_update_state_ptr != NULL)
6471 pfsync_update_state_ptr(s);
6475 } else if (s == NULL)
6476 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6482 action = pf_test_state_other(&s, dir, kif, m, &pd);
6483 if (action == PF_PASS) {
6484 if (pfsync_update_state_ptr != NULL)
6485 pfsync_update_state_ptr(s);
6489 } else if (s == NULL)
6490 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6502 /* handle dangerous IPv6 extension headers. */
6503 if (action == PF_PASS && rh_cnt &&
6504 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6506 REASON_SET(&reason, PFRES_IPOPTIONS);
6508 DPFPRINTF(PF_DEBUG_MISC,
6509 ("pf: dropping packet with dangerous v6 headers\n"));
6512 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6514 REASON_SET(&reason, PFRES_MEMORY);
6516 if (r->rtableid >= 0)
6517 M_SETFIB(m, r->rtableid);
6519 if (r->scrub_flags & PFSTATE_SETPRIO) {
6520 if (pd.tos & IPTOS_LOWDELAY)
6522 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6524 REASON_SET(&reason, PFRES_MEMORY);
6526 DPFPRINTF(PF_DEBUG_MISC,
6527 ("pf: failed to allocate 802.1q mtag\n"));
6532 if (action == PF_PASS && r->qid) {
6533 if (pd.pf_mtag == NULL &&
6534 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6536 REASON_SET(&reason, PFRES_MEMORY);
6539 pd.pf_mtag->qid_hash = pf_state_hash(s);
6540 if (pd.tos & IPTOS_LOWDELAY)
6541 pd.pf_mtag->qid = r->pqid;
6543 pd.pf_mtag->qid = r->qid;
6544 /* Add hints for ecn. */
6545 pd.pf_mtag->hdr = h;
6550 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6551 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6552 (s->nat_rule.ptr->action == PF_RDR ||
6553 s->nat_rule.ptr->action == PF_BINAT) &&
6554 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6555 m->m_flags |= M_SKIP_FIREWALL;
6557 /* XXX: Anybody working on it?! */
6559 printf("pf: divert(9) is not supported for IPv6\n");
6564 if (s != NULL && s->nat_rule.ptr != NULL &&
6565 s->nat_rule.ptr->log & PF_LOG_ALL)
6566 lr = s->nat_rule.ptr;
6569 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6573 kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6574 kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
6576 if (action == PF_PASS || r->action == PF_DROP) {
6577 dirndx = (dir == PF_OUT);
6578 r->packets[dirndx]++;
6579 r->bytes[dirndx] += pd.tot_len;
6581 a->packets[dirndx]++;
6582 a->bytes[dirndx] += pd.tot_len;
6585 if (s->nat_rule.ptr != NULL) {
6586 s->nat_rule.ptr->packets[dirndx]++;
6587 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6589 if (s->src_node != NULL) {
6590 s->src_node->packets[dirndx]++;
6591 s->src_node->bytes[dirndx] += pd.tot_len;
6593 if (s->nat_src_node != NULL) {
6594 s->nat_src_node->packets[dirndx]++;
6595 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6597 dirndx = (dir == s->direction) ? 0 : 1;
6598 s->packets[dirndx]++;
6599 s->bytes[dirndx] += pd.tot_len;
6602 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6603 if (nr != NULL && r == &V_pf_default_rule)
6605 if (tr->src.addr.type == PF_ADDR_TABLE)
6606 pfr_update_stats(tr->src.addr.p.tbl,
6607 (s == NULL) ? pd.src :
6608 &s->key[(s->direction == PF_IN)]->addr[0],
6609 pd.af, pd.tot_len, dir == PF_OUT,
6610 r->action == PF_PASS, tr->src.neg);
6611 if (tr->dst.addr.type == PF_ADDR_TABLE)
6612 pfr_update_stats(tr->dst.addr.p.tbl,
6613 (s == NULL) ? pd.dst :
6614 &s->key[(s->direction == PF_IN)]->addr[1],
6615 pd.af, pd.tot_len, dir == PF_OUT,
6616 r->action == PF_PASS, tr->dst.neg);
6620 case PF_SYNPROXY_DROP:
6631 /* pf_route6() returns unlocked. */
6633 pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
6642 /* If reassembled packet passed, create new fragments. */
6643 if (action == PF_PASS && *m0 && (pflags & PFIL_FWD) &&
6644 (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL)
6645 action = pf_refragment6(ifp, m0, mtag);