2 * Copyright (c) 2004 The DragonFly Project. All rights reserved.
4 * Copyright (c) 2001 Daniel Hartmeier
5 * Copyright (c) 2002 - 2008 Henning Brauer
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * - Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * - Redistributions in binary form must reproduce the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer in the documentation and/or other materials provided
17 * with the distribution.
19 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
22 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
23 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
24 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
25 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
26 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
27 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
29 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
32 * Effort sponsored in part by the Defense Advanced Research Projects
33 * Agency (DARPA) and Air Force Research Laboratory, Air Force
34 * Materiel Command, USAF, under agreement number F30602-01-2-0537.
39 #include "opt_inet6.h"
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/malloc.h>
45 #include <sys/filio.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/kernel.h>
50 #include <sys/sysctl.h>
51 #include <sys/endian.h>
53 #include <sys/kthread.h>
54 #include <sys/spinlock.h>
56 #include <machine/inttypes.h>
61 #include <net/if_types.h>
63 #include <net/netisr2.h>
64 #include <net/route.h>
66 #include <netinet/in.h>
67 #include <netinet/in_var.h>
68 #include <netinet/in_systm.h>
69 #include <netinet/ip.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/tcp.h>
72 #include <netinet/tcp_seq.h>
73 #include <netinet/udp.h>
74 #include <netinet/ip_icmp.h>
75 #include <netinet/in_pcb.h>
76 #include <netinet/tcp_timer.h>
77 #include <netinet/tcp_var.h>
78 #include <netinet/udp_var.h>
79 #include <netinet/icmp_var.h>
80 #include <netinet/if_ether.h>
82 #include <net/pf/pfvar.h>
83 #include <net/pf/if_pflog.h>
85 #include <net/pf/if_pfsync.h>
88 #include <netinet/ip6.h>
89 #include <netinet/icmp6.h>
90 #include <netinet6/nd6.h>
91 #include <netinet6/ip6_var.h>
92 #include <netinet6/in6_pcb.h>
95 #include <sys/in_cksum.h>
96 #include <sys/ucred.h>
97 #include <machine/limits.h>
98 #include <sys/msgport2.h>
99 #include <sys/spinlock2.h>
100 #include <net/netmsg2.h>
101 #include <net/toeplitz2.h>
103 extern int ip_optcopy(struct ip *, struct ip *);
104 extern int debug_pfugidhack;
107 * pf_token - shared lock for cpu-localized operations,
108 * exclusive lock otherwise.
110 * pf_gtoken- exclusive lock used for initialization.
112 * pf_spin - only used to atomically fetch and increment stateid
115 struct lwkt_token pf_token = LWKT_TOKEN_INITIALIZER(pf_token);
116 struct lwkt_token pf_gtoken = LWKT_TOKEN_INITIALIZER(pf_gtoken);
117 #if __SIZEOF_LONG__ != 8
118 struct spinlock pf_spin = SPINLOCK_INITIALIZER(pf_spin, "pf_spin");
121 #define DPFPRINTF(n, x) if (pf_status.debug >= (n)) kprintf x
123 #define FAIL(code) { error = (code); goto done; }
129 /* mask radix tree */
130 struct radix_node_head *pf_maskhead;
133 struct pf_state_tree pf_statetbl[MAXCPU+1]; /* incls one global table */
135 struct pf_altqqueue pf_altqs[2];
136 struct pf_palist pf_pabuf;
137 struct pf_altqqueue *pf_altqs_active;
138 struct pf_altqqueue *pf_altqs_inactive;
139 struct pf_status pf_status;
141 u_int32_t ticket_altqs_active;
142 u_int32_t ticket_altqs_inactive;
143 int altqs_inactive_open;
144 u_int32_t ticket_pabuf;
146 MD5_CTX pf_tcp_secret_ctx;
147 u_char pf_tcp_secret[16];
148 int pf_tcp_secret_init;
151 struct pf_anchor_stackframe {
152 struct pf_ruleset *rs;
154 struct pf_anchor_node *parent;
155 struct pf_anchor *child;
156 } pf_anchor_stack[64];
158 struct malloc_type *pf_src_tree_pl, *pf_rule_pl, *pf_pooladdr_pl;
159 struct malloc_type *pf_state_pl, *pf_state_key_pl, *pf_state_item_pl;
160 struct malloc_type *pf_altq_pl;
162 void pf_print_host(struct pf_addr *, u_int16_t, u_int8_t);
164 void pf_init_threshold(struct pf_threshold *, u_int32_t,
166 void pf_add_threshold(struct pf_threshold *);
167 int pf_check_threshold(struct pf_threshold *);
169 void pf_change_ap(struct pf_addr *, u_int16_t *,
170 u_int16_t *, u_int16_t *, struct pf_addr *,
171 u_int16_t, u_int8_t, sa_family_t);
172 int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
173 struct tcphdr *, struct pf_state_peer *);
175 void pf_change_a6(struct pf_addr *, u_int16_t *,
176 struct pf_addr *, u_int8_t);
178 void pf_change_icmp(struct pf_addr *, u_int16_t *,
179 struct pf_addr *, struct pf_addr *, u_int16_t,
180 u_int16_t *, u_int16_t *, u_int16_t *,
181 u_int16_t *, u_int8_t, sa_family_t);
182 void pf_send_tcp(const struct pf_rule *, sa_family_t,
183 const struct pf_addr *, const struct pf_addr *,
184 u_int16_t, u_int16_t, u_int32_t, u_int32_t,
185 u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
186 u_int16_t, struct ether_header *, struct ifnet *);
187 void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
188 sa_family_t, struct pf_rule *);
189 struct pf_rule *pf_match_translation(struct pf_pdesc *, struct mbuf *,
190 int, int, struct pfi_kif *,
191 struct pf_addr *, u_int16_t, struct pf_addr *,
193 struct pf_rule *pf_get_translation(struct pf_pdesc *, struct mbuf *,
194 int, int, struct pfi_kif *, struct pf_src_node **,
195 struct pf_state_key **, struct pf_state_key **,
196 struct pf_state_key **, struct pf_state_key **,
197 struct pf_addr *, struct pf_addr *,
198 u_int16_t, u_int16_t);
199 void pf_detach_state(struct pf_state *);
200 int pf_state_key_setup(struct pf_pdesc *, struct pf_rule *,
201 struct pf_state_key **, struct pf_state_key **,
202 struct pf_state_key **, struct pf_state_key **,
203 struct pf_addr *, struct pf_addr *,
204 u_int16_t, u_int16_t);
205 void pf_state_key_detach(struct pf_state *, int);
206 u_int32_t pf_tcp_iss(struct pf_pdesc *);
207 int pf_test_rule(struct pf_rule **, struct pf_state **,
208 int, struct pfi_kif *, struct mbuf *, int,
209 void *, struct pf_pdesc *, struct pf_rule **,
210 struct pf_ruleset **, struct ifqueue *, struct inpcb *);
211 static __inline int pf_create_state(struct pf_rule *, struct pf_rule *,
212 struct pf_rule *, struct pf_pdesc *,
213 struct pf_src_node *, struct pf_state_key *,
214 struct pf_state_key *, struct pf_state_key *,
215 struct pf_state_key *, struct mbuf *, int,
216 u_int16_t, u_int16_t, int *, struct pfi_kif *,
217 struct pf_state **, int, u_int16_t, u_int16_t,
219 int pf_test_fragment(struct pf_rule **, int,
220 struct pfi_kif *, struct mbuf *, void *,
221 struct pf_pdesc *, struct pf_rule **,
222 struct pf_ruleset **);
223 int pf_tcp_track_full(struct pf_state_peer *,
224 struct pf_state_peer *, struct pf_state **,
225 struct pfi_kif *, struct mbuf *, int,
226 struct pf_pdesc *, u_short *, int *);
227 int pf_tcp_track_sloppy(struct pf_state_peer *,
228 struct pf_state_peer *, struct pf_state **,
229 struct pf_pdesc *, u_short *);
230 int pf_test_state_tcp(struct pf_state **, int,
231 struct pfi_kif *, struct mbuf *, int,
232 void *, struct pf_pdesc *, u_short *);
233 int pf_test_state_udp(struct pf_state **, int,
234 struct pfi_kif *, struct mbuf *, int,
235 void *, struct pf_pdesc *);
236 int pf_test_state_icmp(struct pf_state **, int,
237 struct pfi_kif *, struct mbuf *, int,
238 void *, struct pf_pdesc *, u_short *);
239 int pf_test_state_other(struct pf_state **, int,
240 struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
241 void pf_step_into_anchor(int *, struct pf_ruleset **, int,
242 struct pf_rule **, struct pf_rule **, int *);
243 int pf_step_out_of_anchor(int *, struct pf_ruleset **,
244 int, struct pf_rule **, struct pf_rule **,
246 void pf_hash(struct pf_addr *, struct pf_addr *,
247 struct pf_poolhashkey *, sa_family_t);
248 int pf_map_addr(u_int8_t, struct pf_rule *,
249 struct pf_addr *, struct pf_addr *,
250 struct pf_addr *, struct pf_src_node **);
251 int pf_get_sport(struct pf_pdesc *,
252 sa_family_t, u_int8_t, struct pf_rule *,
253 struct pf_addr *, struct pf_addr *,
254 u_int16_t, u_int16_t,
255 struct pf_addr *, u_int16_t *,
256 u_int16_t, u_int16_t,
257 struct pf_src_node **);
258 void pf_route(struct mbuf **, struct pf_rule *, int,
259 struct ifnet *, struct pf_state *,
261 void pf_route6(struct mbuf **, struct pf_rule *, int,
262 struct ifnet *, struct pf_state *,
264 u_int8_t pf_get_wscale(struct mbuf *, int, u_int16_t,
266 u_int16_t pf_get_mss(struct mbuf *, int, u_int16_t,
268 u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t,
270 void pf_set_rt_ifp(struct pf_state *,
272 int pf_check_proto_cksum(struct mbuf *, int, int,
273 u_int8_t, sa_family_t);
274 struct pf_divert *pf_get_divert(struct mbuf *);
275 void pf_print_state_parts(struct pf_state *,
276 struct pf_state_key *, struct pf_state_key *);
277 int pf_addr_wrap_neq(struct pf_addr_wrap *,
278 struct pf_addr_wrap *);
279 struct pf_state *pf_find_state(struct pfi_kif *,
280 struct pf_state_key_cmp *, u_int, struct mbuf *);
281 int pf_src_connlimit(struct pf_state *);
282 int pf_check_congestion(struct ifqueue *);
284 extern int pf_end_threads;
286 struct pf_pool_limit pf_pool_limits[PF_LIMIT_MAX] = {
287 { &pf_state_pl, PFSTATE_HIWAT },
288 { &pf_src_tree_pl, PFSNODE_HIWAT },
289 { &pf_frent_pl, PFFRAG_FRENT_HIWAT },
290 { &pfr_ktable_pl, PFR_KTABLE_HIWAT },
291 { &pfr_kentry_pl, PFR_KENTRY_HIWAT }
295 * If route-to and direction is out we match with no further processing
296 * (rt_kif must be assigned and not equal to the out interface)
297 * If reply-to and direction is in we match with no further processing
298 * (rt_kif must be assigned and not equal to the in interface)
300 #define STATE_LOOKUP(i, k, d, s, m) \
302 s = pf_find_state(i, k, d, m); \
303 if (s == NULL || (s)->timeout == PFTM_PURGE) \
306 (((s)->rule.ptr->rt == PF_ROUTETO && \
307 (s)->rule.ptr->direction == PF_OUT) || \
308 ((s)->rule.ptr->rt == PF_REPLYTO && \
309 (s)->rule.ptr->direction == PF_IN)) && \
310 (s)->rt_kif != NULL && \
315 #define BOUND_IFACE(r, k) \
316 ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : pfi_all
318 #define STATE_INC_COUNTERS(s) \
320 atomic_add_int(&s->rule.ptr->states_cur, 1); \
321 s->rule.ptr->states_tot++; \
322 if (s->anchor.ptr != NULL) { \
323 atomic_add_int(&s->anchor.ptr->states_cur, 1); \
324 s->anchor.ptr->states_tot++; \
326 if (s->nat_rule.ptr != NULL) { \
327 atomic_add_int(&s->nat_rule.ptr->states_cur, 1); \
328 s->nat_rule.ptr->states_tot++; \
332 #define STATE_DEC_COUNTERS(s) \
334 if (s->nat_rule.ptr != NULL) \
335 atomic_add_int(&s->nat_rule.ptr->states_cur, -1); \
336 if (s->anchor.ptr != NULL) \
337 atomic_add_int(&s->anchor.ptr->states_cur, -1); \
338 atomic_add_int(&s->rule.ptr->states_cur, -1); \
341 static MALLOC_DEFINE(M_PFSTATEPL, "pfstatepl", "pf state pool list");
342 static MALLOC_DEFINE(M_PFSRCTREEPL, "pfsrctpl", "pf source tree pool list");
343 static MALLOC_DEFINE(M_PFSTATEKEYPL, "pfstatekeypl", "pf state key pool list");
344 static MALLOC_DEFINE(M_PFSTATEITEMPL, "pfstateitempl", "pf state item pool list");
346 static __inline int pf_src_compare(struct pf_src_node *, struct pf_src_node *);
347 static __inline int pf_state_compare_key(struct pf_state_key *,
348 struct pf_state_key *);
349 static __inline int pf_state_compare_rkey(struct pf_state_key *,
350 struct pf_state_key *);
351 static __inline int pf_state_compare_id(struct pf_state *,
354 struct pf_src_tree tree_src_tracking[MAXCPU];
355 struct pf_state_tree_id tree_id[MAXCPU];
356 struct pf_state_queue state_list[MAXCPU];
358 RB_GENERATE(pf_src_tree, pf_src_node, entry, pf_src_compare);
359 RB_GENERATE(pf_state_tree, pf_state_key, entry, pf_state_compare_key);
360 RB_GENERATE(pf_state_rtree, pf_state_key, entry, pf_state_compare_rkey);
361 RB_GENERATE(pf_state_tree_id, pf_state, entry_id, pf_state_compare_id);
364 pf_src_compare(struct pf_src_node *a, struct pf_src_node *b)
368 if (a->rule.ptr > b->rule.ptr)
370 if (a->rule.ptr < b->rule.ptr)
372 if ((diff = a->af - b->af) != 0)
377 if (a->addr.addr32[0] > b->addr.addr32[0])
379 if (a->addr.addr32[0] < b->addr.addr32[0])
385 if (a->addr.addr32[3] > b->addr.addr32[3])
387 if (a->addr.addr32[3] < b->addr.addr32[3])
389 if (a->addr.addr32[2] > b->addr.addr32[2])
391 if (a->addr.addr32[2] < b->addr.addr32[2])
393 if (a->addr.addr32[1] > b->addr.addr32[1])
395 if (a->addr.addr32[1] < b->addr.addr32[1])
397 if (a->addr.addr32[0] > b->addr.addr32[0])
399 if (a->addr.addr32[0] < b->addr.addr32[0])
408 pf_state_hash(struct pf_state_key *sk)
410 u_int32_t hv = (u_int32_t)(((intptr_t)sk >> 6) ^ ((intptr_t)sk >> 15));
411 if (hv == 0) /* disallow 0 */
418 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
423 dst->addr32[0] = src->addr32[0];
427 dst->addr32[0] = src->addr32[0];
428 dst->addr32[1] = src->addr32[1];
429 dst->addr32[2] = src->addr32[2];
430 dst->addr32[3] = src->addr32[3];
437 pf_init_threshold(struct pf_threshold *threshold,
438 u_int32_t limit, u_int32_t seconds)
440 threshold->limit = limit * PF_THRESHOLD_MULT;
441 threshold->seconds = seconds;
442 threshold->count = 0;
443 threshold->last = time_second;
447 pf_add_threshold(struct pf_threshold *threshold)
449 u_int32_t t = time_second, diff = t - threshold->last;
451 if (diff >= threshold->seconds)
452 threshold->count = 0;
454 threshold->count -= threshold->count * diff /
456 threshold->count += PF_THRESHOLD_MULT;
461 pf_check_threshold(struct pf_threshold *threshold)
463 return (threshold->count > threshold->limit);
467 pf_src_connlimit(struct pf_state *state)
470 int cpu = mycpu->gd_cpuid;
472 state->src_node->conn++;
473 state->src.tcp_est = 1;
474 pf_add_threshold(&state->src_node->conn_rate);
476 if (state->rule.ptr->max_src_conn &&
477 state->rule.ptr->max_src_conn <
478 state->src_node->conn) {
479 pf_status.lcounters[LCNT_SRCCONN]++;
483 if (state->rule.ptr->max_src_conn_rate.limit &&
484 pf_check_threshold(&state->src_node->conn_rate)) {
485 pf_status.lcounters[LCNT_SRCCONNRATE]++;
492 if (state->rule.ptr->overload_tbl) {
494 u_int32_t killed = 0;
496 pf_status.lcounters[LCNT_OVERLOAD_TABLE]++;
497 if (pf_status.debug >= PF_DEBUG_MISC) {
498 kprintf("pf_src_connlimit: blocking address ");
499 pf_print_host(&state->src_node->addr, 0,
500 state->key[PF_SK_WIRE]->af);
503 bzero(&p, sizeof(p));
504 p.pfra_af = state->key[PF_SK_WIRE]->af;
505 switch (state->key[PF_SK_WIRE]->af) {
509 p.pfra_ip4addr = state->src_node->addr.v4;
515 p.pfra_ip6addr = state->src_node->addr.v6;
520 pfr_insert_kentry(state->rule.ptr->overload_tbl,
523 /* kill existing states if that's required. */
524 if (state->rule.ptr->flush) {
525 struct pf_state_key *sk;
528 pf_status.lcounters[LCNT_OVERLOAD_FLUSH]++;
529 RB_FOREACH(st, pf_state_tree_id, &tree_id[cpu]) {
530 sk = st->key[PF_SK_WIRE];
532 * Kill states from this source. (Only those
533 * from the same rule if PF_FLUSH_GLOBAL is not
534 * set). (Only on current cpu).
537 state->key[PF_SK_WIRE]->af &&
538 ((state->direction == PF_OUT &&
539 PF_AEQ(&state->src_node->addr,
540 &sk->addr[0], sk->af)) ||
541 (state->direction == PF_IN &&
542 PF_AEQ(&state->src_node->addr,
543 &sk->addr[1], sk->af))) &&
544 (state->rule.ptr->flush &
546 state->rule.ptr == st->rule.ptr)) {
547 st->timeout = PFTM_PURGE;
548 st->src.state = st->dst.state =
553 if (pf_status.debug >= PF_DEBUG_MISC)
554 kprintf(", %u states killed", killed);
556 if (pf_status.debug >= PF_DEBUG_MISC)
560 /* kill this state */
561 state->timeout = PFTM_PURGE;
562 state->src.state = state->dst.state = TCPS_CLOSED;
568 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
569 struct pf_addr *src, sa_family_t af)
571 struct pf_src_node k;
572 int cpu = mycpu->gd_cpuid;
574 bzero(&k, sizeof(k)); /* avoid gcc warnings */
577 PF_ACPY(&k.addr, src, af);
578 if (rule->rule_flag & PFRULE_RULESRCTRACK ||
579 rule->rpool.opts & PF_POOL_STICKYADDR)
583 pf_status.scounters[SCNT_SRC_NODE_SEARCH]++;
584 *sn = RB_FIND(pf_src_tree, &tree_src_tracking[cpu], &k);
587 if (!rule->max_src_nodes ||
588 rule->src_nodes < rule->max_src_nodes)
589 (*sn) = kmalloc(sizeof(struct pf_src_node),
590 M_PFSRCTREEPL, M_NOWAIT|M_ZERO);
592 pf_status.lcounters[LCNT_SRCNODES]++;
596 pf_init_threshold(&(*sn)->conn_rate,
597 rule->max_src_conn_rate.limit,
598 rule->max_src_conn_rate.seconds);
601 if (rule->rule_flag & PFRULE_RULESRCTRACK ||
602 rule->rpool.opts & PF_POOL_STICKYADDR)
603 (*sn)->rule.ptr = rule;
605 (*sn)->rule.ptr = NULL;
606 PF_ACPY(&(*sn)->addr, src, af);
607 if (RB_INSERT(pf_src_tree,
608 &tree_src_tracking[cpu], *sn) != NULL) {
609 if (pf_status.debug >= PF_DEBUG_MISC) {
610 kprintf("pf: src_tree insert failed: ");
611 pf_print_host(&(*sn)->addr, 0, af);
614 kfree(*sn, M_PFSRCTREEPL);
619 * Atomic op required to increment src_nodes in the rule
620 * because we hold a shared token here (decrements will use
621 * an exclusive token).
623 (*sn)->creation = time_second;
624 (*sn)->ruletype = rule->action;
625 if ((*sn)->rule.ptr != NULL)
626 atomic_add_int(&(*sn)->rule.ptr->src_nodes, 1);
627 pf_status.scounters[SCNT_SRC_NODE_INSERT]++;
628 atomic_add_int(&pf_status.src_nodes, 1);
630 if (rule->max_src_states &&
631 (*sn)->states >= rule->max_src_states) {
632 pf_status.lcounters[LCNT_SRCSTATES]++;
640 * state table (indexed by the pf_state_key structure), normal RBTREE
644 pf_state_compare_key(struct pf_state_key *a, struct pf_state_key *b)
648 if ((diff = a->proto - b->proto) != 0)
650 if ((diff = a->af - b->af) != 0)
655 if (a->addr[0].addr32[0] > b->addr[0].addr32[0])
657 if (a->addr[0].addr32[0] < b->addr[0].addr32[0])
659 if (a->addr[1].addr32[0] > b->addr[1].addr32[0])
661 if (a->addr[1].addr32[0] < b->addr[1].addr32[0])
667 if (a->addr[0].addr32[3] > b->addr[0].addr32[3])
669 if (a->addr[0].addr32[3] < b->addr[0].addr32[3])
671 if (a->addr[1].addr32[3] > b->addr[1].addr32[3])
673 if (a->addr[1].addr32[3] < b->addr[1].addr32[3])
675 if (a->addr[0].addr32[2] > b->addr[0].addr32[2])
677 if (a->addr[0].addr32[2] < b->addr[0].addr32[2])
679 if (a->addr[1].addr32[2] > b->addr[1].addr32[2])
681 if (a->addr[1].addr32[2] < b->addr[1].addr32[2])
683 if (a->addr[0].addr32[1] > b->addr[0].addr32[1])
685 if (a->addr[0].addr32[1] < b->addr[0].addr32[1])
687 if (a->addr[1].addr32[1] > b->addr[1].addr32[1])
689 if (a->addr[1].addr32[1] < b->addr[1].addr32[1])
691 if (a->addr[0].addr32[0] > b->addr[0].addr32[0])
693 if (a->addr[0].addr32[0] < b->addr[0].addr32[0])
695 if (a->addr[1].addr32[0] > b->addr[1].addr32[0])
697 if (a->addr[1].addr32[0] < b->addr[1].addr32[0])
703 if ((diff = a->port[0] - b->port[0]) != 0)
705 if ((diff = a->port[1] - b->port[1]) != 0)
712 * Used for RB_FIND only, compare in the reverse direction. The
713 * element to be reversed is always (a), since we obviously can't
714 * reverse the state tree depicted by (b).
717 pf_state_compare_rkey(struct pf_state_key *a, struct pf_state_key *b)
721 if ((diff = a->proto - b->proto) != 0)
723 if ((diff = a->af - b->af) != 0)
728 if (a->addr[1].addr32[0] > b->addr[0].addr32[0])
730 if (a->addr[1].addr32[0] < b->addr[0].addr32[0])
732 if (a->addr[0].addr32[0] > b->addr[1].addr32[0])
734 if (a->addr[0].addr32[0] < b->addr[1].addr32[0])
740 if (a->addr[1].addr32[3] > b->addr[0].addr32[3])
742 if (a->addr[1].addr32[3] < b->addr[0].addr32[3])
744 if (a->addr[0].addr32[3] > b->addr[1].addr32[3])
746 if (a->addr[0].addr32[3] < b->addr[1].addr32[3])
748 if (a->addr[1].addr32[2] > b->addr[0].addr32[2])
750 if (a->addr[1].addr32[2] < b->addr[0].addr32[2])
752 if (a->addr[0].addr32[2] > b->addr[1].addr32[2])
754 if (a->addr[0].addr32[2] < b->addr[1].addr32[2])
756 if (a->addr[1].addr32[1] > b->addr[0].addr32[1])
758 if (a->addr[1].addr32[1] < b->addr[0].addr32[1])
760 if (a->addr[0].addr32[1] > b->addr[1].addr32[1])
762 if (a->addr[0].addr32[1] < b->addr[1].addr32[1])
764 if (a->addr[1].addr32[0] > b->addr[0].addr32[0])
766 if (a->addr[1].addr32[0] < b->addr[0].addr32[0])
768 if (a->addr[0].addr32[0] > b->addr[1].addr32[0])
770 if (a->addr[0].addr32[0] < b->addr[1].addr32[0])
776 if ((diff = a->port[1] - b->port[0]) != 0)
778 if ((diff = a->port[0] - b->port[1]) != 0)
785 pf_state_compare_id(struct pf_state *a, struct pf_state *b)
791 if (a->creatorid > b->creatorid)
793 if (a->creatorid < b->creatorid)
800 pf_state_key_attach(struct pf_state_key *sk, struct pf_state *s, int idx)
802 struct pf_state_item *si;
803 struct pf_state_key *cur;
808 * PFSTATE_STACK_GLOBAL is set when the state might not hash to the
809 * current cpu. The keys are managed on the global statetbl tree
810 * for this case. Only translations (RDR, NAT) can cause this.
812 * When this flag is not set we must still check the global statetbl
813 * for a collision, and if we find one we set the HALF_DUPLEX flag
816 if (s->state_flags & PFSTATE_STACK_GLOBAL) {
818 lockmgr(&pf_global_statetbl_lock, LK_EXCLUSIVE);
820 cpu = mycpu->gd_cpuid;
821 lockmgr(&pf_global_statetbl_lock, LK_SHARED);
823 KKASSERT(s->key[idx] == NULL); /* XXX handle this? */
825 if (pf_status.debug >= PF_DEBUG_MISC) {
826 kprintf("state_key attach cpu %d (%08x:%d) %s (%08x:%d)\n",
828 ntohl(sk->addr[0].addr32[0]), ntohs(sk->port[0]),
829 (idx == PF_SK_WIRE ? "->" : "<-"),
830 ntohl(sk->addr[1].addr32[0]), ntohs(sk->port[1]));
834 * Check whether (e.g.) a PASS rule being put on a per-cpu tree
835 * collides with a translation rule on the global tree. This is
836 * NOT an error. We *WANT* to establish state for this case so the
837 * packet path is short-cutted and doesn't need to scan the ruleset
838 * on every packet. But the established state will only see one
839 * side of a two-way packet conversation. To prevent this from
840 * causing problems (e.g. generating a RST), we force PFSTATE_SLOPPY
841 * to be set on the established state.
843 * A collision against RDR state can only occur with a PASS IN in the
844 * opposite direction or a PASS OUT in the forwards direction. This
845 * is because RDRs are processed on the input side.
847 * A collision against NAT state can only occur with a PASS IN in the
848 * forwards direction or a PASS OUT in the opposite direction. This
849 * is because NATs are processed on the output side.
851 * In both situations we need to do a reverse addr/port test because
852 * the PASS IN or PASS OUT only establishes if it doesn't match the
853 * established RDR state in the forwards direction. The direction
854 * flag has to be ignored (it will be one way for a PASS IN and the
855 * other way for a PASS OUT).
857 * pf_global_statetbl_lock will be locked shared when testing and
858 * not entering into the global state table.
861 (cur = RB_FIND(pf_state_rtree,
862 (struct pf_state_rtree *)&pf_statetbl[MAXCPU],
864 TAILQ_FOREACH(si, &cur->states, entry) {
866 * NOTE: We must ignore direction mismatches.
868 if (si->s->kif == s->kif) {
869 s->state_flags |= PFSTATE_HALF_DUPLEX |
871 if (pf_status.debug >= PF_DEBUG_MISC) {
873 "pf: %s key attach collision "
875 (idx == PF_SK_WIRE) ?
878 pf_print_state_parts(s,
879 (idx == PF_SK_WIRE) ? sk : NULL,
880 (idx == PF_SK_STACK) ? sk : NULL);
889 * Enter into either the per-cpu or the global state table.
891 * pf_global_statetbl_lock will be locked exclusively when entering
892 * into the global state table.
894 if ((cur = RB_INSERT(pf_state_tree, &pf_statetbl[cpu], sk)) != NULL) {
895 /* key exists. check for same kif, if none, add to key */
896 TAILQ_FOREACH(si, &cur->states, entry) {
897 if (si->s->kif == s->kif &&
898 si->s->direction == s->direction) {
899 if (pf_status.debug >= PF_DEBUG_MISC) {
901 "pf: %s key attach failed on %s: ",
902 (idx == PF_SK_WIRE) ?
905 pf_print_state_parts(s,
906 (idx == PF_SK_WIRE) ? sk : NULL,
907 (idx == PF_SK_STACK) ? sk : NULL);
910 kfree(sk, M_PFSTATEKEYPL);
912 goto failed; /* collision! */
915 kfree(sk, M_PFSTATEKEYPL);
922 if ((si = kmalloc(sizeof(struct pf_state_item),
923 M_PFSTATEITEMPL, M_NOWAIT)) == NULL) {
924 pf_state_key_detach(s, idx);
926 goto failed; /* collision! */
930 /* list is sorted, if-bound states before floating */
931 if (s->kif == pfi_all)
932 TAILQ_INSERT_TAIL(&s->key[idx]->states, si, entry);
934 TAILQ_INSERT_HEAD(&s->key[idx]->states, si, entry);
938 lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
943 * NOTE: Can only be called indirectly via the purge thread with pf_token
944 * exclusively locked.
947 pf_detach_state(struct pf_state *s)
949 if (s->key[PF_SK_WIRE] == s->key[PF_SK_STACK])
950 s->key[PF_SK_WIRE] = NULL;
952 if (s->key[PF_SK_STACK] != NULL)
953 pf_state_key_detach(s, PF_SK_STACK);
955 if (s->key[PF_SK_WIRE] != NULL)
956 pf_state_key_detach(s, PF_SK_WIRE);
960 * NOTE: Can only be called indirectly via the purge thread with pf_token
961 * exclusively locked.
964 pf_state_key_detach(struct pf_state *s, int idx)
966 struct pf_state_item *si;
970 * PFSTATE_STACK_GLOBAL is set for translations when the translated
971 * address/port is not localized to the same cpu that the untranslated
972 * address/port is on. The wire pf_state_key is managed on the global
973 * statetbl tree for this case.
975 if (s->state_flags & PFSTATE_STACK_GLOBAL) {
977 lockmgr(&pf_global_statetbl_lock, LK_EXCLUSIVE);
979 cpu = mycpu->gd_cpuid;
982 si = TAILQ_FIRST(&s->key[idx]->states);
983 while (si && si->s != s)
984 si = TAILQ_NEXT(si, entry);
987 TAILQ_REMOVE(&s->key[idx]->states, si, entry);
988 kfree(si, M_PFSTATEITEMPL);
991 if (TAILQ_EMPTY(&s->key[idx]->states)) {
992 RB_REMOVE(pf_state_tree, &pf_statetbl[cpu], s->key[idx]);
993 if (s->key[idx]->reverse)
994 s->key[idx]->reverse->reverse = NULL;
995 if (s->key[idx]->inp)
996 s->key[idx]->inp->inp_pf_sk = NULL;
997 kfree(s->key[idx], M_PFSTATEKEYPL);
1001 if (s->state_flags & PFSTATE_STACK_GLOBAL)
1002 lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
1005 struct pf_state_key *
1006 pf_alloc_state_key(int pool_flags)
1008 struct pf_state_key *sk;
1010 sk = kmalloc(sizeof(struct pf_state_key), M_PFSTATEKEYPL, pool_flags);
1012 TAILQ_INIT(&sk->states);
1018 pf_state_key_setup(struct pf_pdesc *pd, struct pf_rule *nr,
1019 struct pf_state_key **skw, struct pf_state_key **sks,
1020 struct pf_state_key **skp, struct pf_state_key **nkp,
1021 struct pf_addr *saddr, struct pf_addr *daddr,
1022 u_int16_t sport, u_int16_t dport)
1024 KKASSERT((*skp == NULL && *nkp == NULL));
1026 if ((*skp = pf_alloc_state_key(M_NOWAIT | M_ZERO)) == NULL)
1029 PF_ACPY(&(*skp)->addr[pd->sidx], saddr, pd->af);
1030 PF_ACPY(&(*skp)->addr[pd->didx], daddr, pd->af);
1031 (*skp)->port[pd->sidx] = sport;
1032 (*skp)->port[pd->didx] = dport;
1033 (*skp)->proto = pd->proto;
1034 (*skp)->af = pd->af;
1037 if ((*nkp = pf_alloc_state_key(M_NOWAIT | M_ZERO)) == NULL)
1038 return (ENOMEM); /* caller must handle cleanup */
1040 /* XXX maybe just bcopy and TAILQ_INIT(&(*nkp)->states) */
1041 PF_ACPY(&(*nkp)->addr[0], &(*skp)->addr[0], pd->af);
1042 PF_ACPY(&(*nkp)->addr[1], &(*skp)->addr[1], pd->af);
1043 (*nkp)->port[0] = (*skp)->port[0];
1044 (*nkp)->port[1] = (*skp)->port[1];
1045 (*nkp)->proto = pd->proto;
1046 (*nkp)->af = pd->af;
1051 if (pd->dir == PF_IN) {
1062 * Insert pf_state with one or two state keys (allowing a reverse path lookup
1063 * which is used by NAT). In the NAT case skw is the initiator (?) and
1064 * sks is the target.
1067 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1068 struct pf_state_key *sks, struct pf_state *s)
1070 int cpu = mycpu->gd_cpuid;
1076 if (pf_state_key_attach(skw, s, PF_SK_WIRE))
1078 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1084 if (pf_state_key_attach(skw, s, PF_SK_WIRE)) {
1085 kfree(sks, M_PFSTATEKEYPL);
1088 if (pf_state_key_attach(sks, s, PF_SK_STACK)) {
1089 pf_state_key_detach(s, PF_SK_WIRE);
1094 if (s->id == 0 && s->creatorid == 0) {
1097 #if __SIZEOF_LONG__ == 8
1098 sid = atomic_fetchadd_long(&pf_status.stateid, 1);
1100 spin_lock(&pf_spin);
1101 sid = pf_status.stateid++;
1102 spin_unlock(&pf_spin);
1104 s->id = htobe64(sid);
1105 s->creatorid = pf_status.hostid;
1109 * Calculate hash code for altq
1111 s->hash = crc32(s->key[PF_SK_WIRE], PF_STATE_KEY_HASH_LENGTH);
1113 if (RB_INSERT(pf_state_tree_id, &tree_id[cpu], s) != NULL) {
1114 if (pf_status.debug >= PF_DEBUG_MISC) {
1115 kprintf("pf: state insert failed: "
1116 "id: %016jx creatorid: %08x",
1117 (uintmax_t)be64toh(s->id), ntohl(s->creatorid));
1118 if (s->sync_flags & PFSTATE_FROMSYNC)
1119 kprintf(" (from sync)");
1125 TAILQ_INSERT_TAIL(&state_list[cpu], s, entry_list);
1126 pf_status.fcounters[FCNT_STATE_INSERT]++;
1127 atomic_add_int(&pf_status.states, 1);
1128 pfi_kif_ref(kif, PFI_KIF_REF_STATE);
1129 pfsync_insert_state(s);
1134 pf_find_state_byid(struct pf_state_cmp *key)
1136 int cpu = mycpu->gd_cpuid;
1138 pf_status.fcounters[FCNT_STATE_SEARCH]++;
1140 return (RB_FIND(pf_state_tree_id, &tree_id[cpu],
1141 (struct pf_state *)key));
1145 * WARNING! May return a state structure that was localized to another cpu,
1146 * destruction is typically protected by the callers pf_token.
1147 * The element can only be destroyed
1150 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir,
1153 struct pf_state_key *skey = (void *)key;
1154 struct pf_state_key *sk;
1155 struct pf_state_item *si;
1157 int cpu = mycpu->gd_cpuid;
1160 pf_status.fcounters[FCNT_STATE_SEARCH]++;
1162 if (dir == PF_OUT && m->m_pkthdr.pf.statekey &&
1163 ((struct pf_state_key *)m->m_pkthdr.pf.statekey)->reverse) {
1164 sk = ((struct pf_state_key *)m->m_pkthdr.pf.statekey)->reverse;
1166 sk = RB_FIND(pf_state_tree, &pf_statetbl[cpu], skey);
1168 lockmgr(&pf_global_statetbl_lock, LK_SHARED);
1169 sk = RB_FIND(pf_state_tree, &pf_statetbl[MAXCPU], skey);
1171 lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
1176 if (dir == PF_OUT && m->m_pkthdr.pf.statekey) {
1177 ((struct pf_state_key *)
1178 m->m_pkthdr.pf.statekey)->reverse = sk;
1179 sk->reverse = m->m_pkthdr.pf.statekey;
1183 m->m_pkthdr.pf.statekey = NULL;
1185 /* list is sorted, if-bound states before floating ones */
1186 TAILQ_FOREACH(si, &sk->states, entry) {
1187 if ((si->s->kif == pfi_all || si->s->kif == kif) &&
1188 sk == (dir == PF_IN ? si->s->key[PF_SK_WIRE] :
1189 si->s->key[PF_SK_STACK])) {
1195 * Extract state before potentially releasing the global statetbl
1196 * lock. Ignore the state if the create is still in-progress as
1197 * it can be deleted out from under us by the owning localized cpu.
1198 * However, if CREATEINPROG is not set, state can only be deleted
1199 * by the purge thread which we are protected from via our shared
1204 if (s && (s->state_flags & PFSTATE_CREATEINPROG))
1210 lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
1215 * WARNING! May return a state structure that was localized to another cpu,
1216 * destruction is typically protected by the callers pf_token.
1219 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1221 struct pf_state_key *skey = (void *)key;
1222 struct pf_state_key *sk;
1223 struct pf_state_item *si, *ret = NULL;
1225 int cpu = mycpu->gd_cpuid;
1228 pf_status.fcounters[FCNT_STATE_SEARCH]++;
1230 sk = RB_FIND(pf_state_tree, &pf_statetbl[cpu], skey);
1232 lockmgr(&pf_global_statetbl_lock, LK_SHARED);
1233 sk = RB_FIND(pf_state_tree, &pf_statetbl[MAXCPU], skey);
1237 TAILQ_FOREACH(si, &sk->states, entry)
1238 if (dir == PF_INOUT ||
1239 (sk == (dir == PF_IN ? si->s->key[PF_SK_WIRE] :
1240 si->s->key[PF_SK_STACK]))) {
1253 * Extract state before potentially releasing the global statetbl
1254 * lock. Ignore the state if the create is still in-progress as
1255 * it can be deleted out from under us by the owning localized cpu.
1256 * However, if CREATEINPROG is not set, state can only be deleted
1257 * by the purge thread which we are protected from via our shared
1262 if (s && (s->state_flags & PFSTATE_CREATEINPROG))
1268 lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
1272 /* END state table stuff */
1275 pf_purge_thread(void *v)
1277 globaldata_t save_gd = mycpu;
1284 tsleep(pf_purge_thread, PWAIT, "pftm", 1 * hz);
1286 endingit = pf_end_threads;
1288 for (nn = 0; nn < ncpus; ++nn) {
1289 lwkt_setcpu_self(globaldata_find(nn));
1291 lwkt_gettoken(&pf_token);
1292 lockmgr(&pf_consistency_lock, LK_EXCLUSIVE);
1296 * process a fraction of the state table every second
1298 if(!pf_purge_expired_states(
1299 1 + (pf_status.states /
1300 pf_default_rule.timeout[
1301 PFTM_INTERVAL]), 0)) {
1302 pf_purge_expired_states(
1303 1 + (pf_status.states /
1304 pf_default_rule.timeout[
1305 PFTM_INTERVAL]), 1);
1309 * purge other expired types every PFTM_INTERVAL
1313 pf_default_rule.timeout[PFTM_INTERVAL]) {
1314 pf_purge_expired_fragments();
1315 if (!pf_purge_expired_src_nodes(locked)) {
1316 pf_purge_expired_src_nodes(1);
1322 * If terminating the thread, clean everything out
1326 pf_purge_expired_states(pf_status.states, 0);
1327 pf_purge_expired_fragments();
1328 pf_purge_expired_src_nodes(1);
1332 lockmgr(&pf_consistency_lock, LK_RELEASE);
1333 lwkt_reltoken(&pf_token);
1335 lwkt_setcpu_self(save_gd);
1341 * Thread termination
1344 wakeup(pf_purge_thread);
1349 pf_state_expires(const struct pf_state *state)
1356 /* handle all PFTM_* > PFTM_MAX here */
1357 if (state->timeout == PFTM_PURGE)
1358 return (time_second);
1359 if (state->timeout == PFTM_UNTIL_PACKET)
1361 KKASSERT(state->timeout != PFTM_UNLINKED);
1362 KKASSERT(state->timeout < PFTM_MAX);
1363 timeout = state->rule.ptr->timeout[state->timeout];
1365 timeout = pf_default_rule.timeout[state->timeout];
1366 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1368 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1369 states = state->rule.ptr->states_cur;
1371 start = pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1372 end = pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1373 states = pf_status.states;
1375 if (end && states > start && start < end) {
1377 return (state->expire + timeout * (end - states) /
1380 return (time_second);
1382 return (state->expire + timeout);
1386 * (called with exclusive pf_token)
1389 pf_purge_expired_src_nodes(int waslocked)
1391 struct pf_src_node *cur, *next;
1392 int locked = waslocked;
1393 int cpu = mycpu->gd_cpuid;
1395 for (cur = RB_MIN(pf_src_tree, &tree_src_tracking[cpu]);
1398 next = RB_NEXT(pf_src_tree, &tree_src_tracking[cpu], cur);
1400 if (cur->states <= 0 && cur->expire <= time_second) {
1402 lockmgr(&pf_consistency_lock, LK_EXCLUSIVE);
1403 next = RB_NEXT(pf_src_tree,
1404 &tree_src_tracking[cpu], cur);
1407 if (cur->rule.ptr != NULL) {
1409 * decrements in rule should be ok, token is
1410 * held exclusively in this code path.
1412 cur->rule.ptr->src_nodes--;
1413 if (cur->rule.ptr->states_cur <= 0 &&
1414 cur->rule.ptr->max_src_nodes <= 0)
1415 pf_rm_rule(NULL, cur->rule.ptr);
1417 RB_REMOVE(pf_src_tree, &tree_src_tracking[cpu], cur);
1418 pf_status.scounters[SCNT_SRC_NODE_REMOVALS]++;
1419 atomic_add_int(&pf_status.src_nodes, -1);
1420 kfree(cur, M_PFSRCTREEPL);
1423 if (locked && !waslocked)
1424 lockmgr(&pf_consistency_lock, LK_RELEASE);
1429 pf_src_tree_remove_state(struct pf_state *s)
1433 if (s->src_node != NULL) {
1435 --s->src_node->conn;
1436 if (--s->src_node->states <= 0) {
1437 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1440 pf_default_rule.timeout[PFTM_SRC_NODE];
1442 s->src_node->expire = time_second + timeout;
1445 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1446 if (--s->nat_src_node->states <= 0) {
1447 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1450 pf_default_rule.timeout[PFTM_SRC_NODE];
1451 s->nat_src_node->expire = time_second + timeout;
1454 s->src_node = s->nat_src_node = NULL;
1457 /* callers should be at crit_enter() */
1459 pf_unlink_state(struct pf_state *cur)
1461 int cpu = mycpu->gd_cpuid;
1463 if (cur->src.state == PF_TCPS_PROXY_DST) {
1464 /* XXX wire key the right one? */
1465 pf_send_tcp(cur->rule.ptr, cur->key[PF_SK_WIRE]->af,
1466 &cur->key[PF_SK_WIRE]->addr[1],
1467 &cur->key[PF_SK_WIRE]->addr[0],
1468 cur->key[PF_SK_WIRE]->port[1],
1469 cur->key[PF_SK_WIRE]->port[0],
1470 cur->src.seqhi, cur->src.seqlo + 1,
1471 TH_RST|TH_ACK, 0, 0, 0, 1, cur->tag, NULL, NULL);
1473 RB_REMOVE(pf_state_tree_id, &tree_id[cpu], cur);
1474 if (cur->creatorid == pf_status.hostid)
1475 pfsync_delete_state(cur);
1476 cur->timeout = PFTM_UNLINKED;
1477 pf_src_tree_remove_state(cur);
1478 pf_detach_state(cur);
1481 static struct pf_state *purge_cur[MAXCPU];
1484 * callers should be at crit_enter() and hold pf_consistency_lock exclusively.
1485 * pf_token must also be held exclusively.
1488 pf_free_state(struct pf_state *cur)
1490 int cpu = mycpu->gd_cpuid;
1492 KKASSERT(cur->cpuid == cpu);
1494 if (pfsyncif != NULL &&
1495 (pfsyncif->sc_bulk_send_next == cur ||
1496 pfsyncif->sc_bulk_terminator == cur))
1498 KKASSERT(cur->timeout == PFTM_UNLINKED);
1500 * decrements in rule should be ok, token is
1501 * held exclusively in this code path.
1503 if (--cur->rule.ptr->states_cur <= 0 &&
1504 cur->rule.ptr->src_nodes <= 0)
1505 pf_rm_rule(NULL, cur->rule.ptr);
1506 if (cur->nat_rule.ptr != NULL) {
1507 if (--cur->nat_rule.ptr->states_cur <= 0 &&
1508 cur->nat_rule.ptr->src_nodes <= 0) {
1509 pf_rm_rule(NULL, cur->nat_rule.ptr);
1512 if (cur->anchor.ptr != NULL) {
1513 if (--cur->anchor.ptr->states_cur <= 0)
1514 pf_rm_rule(NULL, cur->anchor.ptr);
1516 pf_normalize_tcp_cleanup(cur);
1517 pfi_kif_unref(cur->kif, PFI_KIF_REF_STATE);
1520 * We may be freeing pf_purge_expired_states()'s saved scan entry,
1521 * adjust it if necessary.
1523 if (purge_cur[cpu] == cur) {
1524 kprintf("PURGE CONFLICT\n");
1525 purge_cur[cpu] = TAILQ_NEXT(purge_cur[cpu], entry_list);
1527 TAILQ_REMOVE(&state_list[cpu], cur, entry_list);
1529 pf_tag_unref(cur->tag);
1530 kfree(cur, M_PFSTATEPL);
1531 pf_status.fcounters[FCNT_STATE_REMOVALS]++;
1532 atomic_add_int(&pf_status.states, -1);
1536 pf_purge_expired_states(u_int32_t maxcheck, int waslocked)
1538 struct pf_state *cur;
1539 int locked = waslocked;
1540 int cpu = mycpu->gd_cpuid;
1542 while (maxcheck--) {
1544 * Wrap to start of list when we hit the end
1546 cur = purge_cur[cpu];
1548 cur = TAILQ_FIRST(&state_list[cpu]);
1550 break; /* list empty */
1554 * Setup next (purge_cur) while we process this one. If
1555 * we block and something else deletes purge_cur,
1556 * pf_free_state() will adjust it further ahead.
1558 purge_cur[cpu] = TAILQ_NEXT(cur, entry_list);
1560 if (cur->timeout == PFTM_UNLINKED) {
1561 /* free unlinked state */
1563 lockmgr(&pf_consistency_lock, LK_EXCLUSIVE);
1567 } else if (pf_state_expires(cur) <= time_second) {
1568 /* unlink and free expired state */
1569 pf_unlink_state(cur);
1571 if (!lockmgr(&pf_consistency_lock, LK_EXCLUSIVE))
1580 lockmgr(&pf_consistency_lock, LK_RELEASE);
1585 pf_tbladdr_setup(struct pf_ruleset *rs, struct pf_addr_wrap *aw)
1587 if (aw->type != PF_ADDR_TABLE)
1589 if ((aw->p.tbl = pfr_attach_table(rs, aw->v.tblname)) == NULL)
1595 pf_tbladdr_remove(struct pf_addr_wrap *aw)
1597 if (aw->type != PF_ADDR_TABLE || aw->p.tbl == NULL)
1599 pfr_detach_table(aw->p.tbl);
1604 pf_tbladdr_copyout(struct pf_addr_wrap *aw)
1606 struct pfr_ktable *kt = aw->p.tbl;
1608 if (aw->type != PF_ADDR_TABLE || kt == NULL)
1610 if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL)
1611 kt = kt->pfrkt_root;
1613 aw->p.tblcnt = (kt->pfrkt_flags & PFR_TFLAG_ACTIVE) ?
1618 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1623 u_int32_t a = ntohl(addr->addr32[0]);
1624 kprintf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1636 u_int8_t i, curstart = 255, curend = 0,
1637 maxstart = 0, maxend = 0;
1638 for (i = 0; i < 8; i++) {
1639 if (!addr->addr16[i]) {
1640 if (curstart == 255)
1646 if ((curend - curstart) >
1647 (maxend - maxstart)) {
1648 maxstart = curstart;
1655 for (i = 0; i < 8; i++) {
1656 if (i >= maxstart && i <= maxend) {
1665 b = ntohs(addr->addr16[i]);
1682 pf_print_state(struct pf_state *s)
1684 pf_print_state_parts(s, NULL, NULL);
1688 pf_print_state_parts(struct pf_state *s,
1689 struct pf_state_key *skwp, struct pf_state_key *sksp)
1691 struct pf_state_key *skw, *sks;
1692 u_int8_t proto, dir;
1694 /* Do our best to fill these, but they're skipped if NULL */
1695 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1696 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1697 proto = skw ? skw->proto : (sks ? sks->proto : 0);
1698 dir = s ? s->direction : 0;
1710 case IPPROTO_ICMPV6:
1714 kprintf("%u ", skw->proto);
1727 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1729 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1732 kprintf(" stack: ");
1734 pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1736 pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1741 if (proto == IPPROTO_TCP) {
1742 kprintf(" [lo=%u high=%u win=%u modulator=%u",
1743 s->src.seqlo, s->src.seqhi,
1744 s->src.max_win, s->src.seqdiff);
1745 if (s->src.wscale && s->dst.wscale)
1746 kprintf(" wscale=%u",
1747 s->src.wscale & PF_WSCALE_MASK);
1749 kprintf(" [lo=%u high=%u win=%u modulator=%u",
1750 s->dst.seqlo, s->dst.seqhi,
1751 s->dst.max_win, s->dst.seqdiff);
1752 if (s->src.wscale && s->dst.wscale)
1753 kprintf(" wscale=%u",
1754 s->dst.wscale & PF_WSCALE_MASK);
1757 kprintf(" %u:%u", s->src.state, s->dst.state);
1762 pf_print_flags(u_int8_t f)
1784 #define PF_SET_SKIP_STEPS(i) \
1786 while (head[i] != cur) { \
1787 head[i]->skip[i].ptr = cur; \
1788 head[i] = TAILQ_NEXT(head[i], entries); \
1793 pf_calc_skip_steps(struct pf_rulequeue *rules)
1795 struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1798 cur = TAILQ_FIRST(rules);
1800 for (i = 0; i < PF_SKIP_COUNT; ++i)
1802 while (cur != NULL) {
1804 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1805 PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1806 if (cur->direction != prev->direction)
1807 PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1808 if (cur->af != prev->af)
1809 PF_SET_SKIP_STEPS(PF_SKIP_AF);
1810 if (cur->proto != prev->proto)
1811 PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
1812 if (cur->src.neg != prev->src.neg ||
1813 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
1814 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
1815 if (cur->src.port[0] != prev->src.port[0] ||
1816 cur->src.port[1] != prev->src.port[1] ||
1817 cur->src.port_op != prev->src.port_op)
1818 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
1819 if (cur->dst.neg != prev->dst.neg ||
1820 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
1821 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
1822 if (cur->dst.port[0] != prev->dst.port[0] ||
1823 cur->dst.port[1] != prev->dst.port[1] ||
1824 cur->dst.port_op != prev->dst.port_op)
1825 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
1828 cur = TAILQ_NEXT(cur, entries);
1830 for (i = 0; i < PF_SKIP_COUNT; ++i)
1831 PF_SET_SKIP_STEPS(i);
1835 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
1837 if (aw1->type != aw2->type)
1839 switch (aw1->type) {
1840 case PF_ADDR_ADDRMASK:
1842 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0))
1844 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0))
1847 case PF_ADDR_DYNIFTL:
1848 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
1849 case PF_ADDR_NOROUTE:
1850 case PF_ADDR_URPFFAILED:
1853 return (aw1->p.tbl != aw2->p.tbl);
1854 case PF_ADDR_RTLABEL:
1855 return (aw1->v.rtlabel != aw2->v.rtlabel);
1857 kprintf("invalid address type: %d\n", aw1->type);
1863 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
1869 l = cksum + old - new;
1870 l = (l >> 16) + (l & 65535);
1878 pf_change_ap(struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc,
1879 struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af)
1884 PF_ACPY(&ao, a, af);
1892 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
1893 ao.addr16[0], an->addr16[0], 0),
1894 ao.addr16[1], an->addr16[1], 0);
1896 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
1897 ao.addr16[0], an->addr16[0], u),
1898 ao.addr16[1], an->addr16[1], u),
1904 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1905 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1906 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
1907 ao.addr16[0], an->addr16[0], u),
1908 ao.addr16[1], an->addr16[1], u),
1909 ao.addr16[2], an->addr16[2], u),
1910 ao.addr16[3], an->addr16[3], u),
1911 ao.addr16[4], an->addr16[4], u),
1912 ao.addr16[5], an->addr16[5], u),
1913 ao.addr16[6], an->addr16[6], u),
1914 ao.addr16[7], an->addr16[7], u),
1922 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */
1924 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
1928 memcpy(&ao, a, sizeof(ao));
1929 memcpy(a, &an, sizeof(u_int32_t));
1930 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
1931 ao % 65536, an % 65536, u);
1936 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
1940 PF_ACPY(&ao, a, AF_INET6);
1941 PF_ACPY(a, an, AF_INET6);
1943 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1944 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1945 pf_cksum_fixup(pf_cksum_fixup(*c,
1946 ao.addr16[0], an->addr16[0], u),
1947 ao.addr16[1], an->addr16[1], u),
1948 ao.addr16[2], an->addr16[2], u),
1949 ao.addr16[3], an->addr16[3], u),
1950 ao.addr16[4], an->addr16[4], u),
1951 ao.addr16[5], an->addr16[5], u),
1952 ao.addr16[6], an->addr16[6], u),
1953 ao.addr16[7], an->addr16[7], u);
1958 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
1959 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
1960 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
1962 struct pf_addr oia, ooa;
1964 PF_ACPY(&oia, ia, af);
1966 PF_ACPY(&ooa, oa, af);
1968 /* Change inner protocol port, fix inner protocol checksum. */
1970 u_int16_t oip = *ip;
1977 *pc = pf_cksum_fixup(*pc, oip, *ip, u);
1978 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
1980 *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
1982 /* Change inner ip address, fix inner ip and icmp checksums. */
1983 PF_ACPY(ia, na, af);
1987 u_int32_t oh2c = *h2c;
1989 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
1990 oia.addr16[0], ia->addr16[0], 0),
1991 oia.addr16[1], ia->addr16[1], 0);
1992 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
1993 oia.addr16[0], ia->addr16[0], 0),
1994 oia.addr16[1], ia->addr16[1], 0);
1995 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2001 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2002 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2003 pf_cksum_fixup(pf_cksum_fixup(*ic,
2004 oia.addr16[0], ia->addr16[0], u),
2005 oia.addr16[1], ia->addr16[1], u),
2006 oia.addr16[2], ia->addr16[2], u),
2007 oia.addr16[3], ia->addr16[3], u),
2008 oia.addr16[4], ia->addr16[4], u),
2009 oia.addr16[5], ia->addr16[5], u),
2010 oia.addr16[6], ia->addr16[6], u),
2011 oia.addr16[7], ia->addr16[7], u);
2015 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2017 PF_ACPY(oa, na, af);
2021 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2022 ooa.addr16[0], oa->addr16[0], 0),
2023 ooa.addr16[1], oa->addr16[1], 0);
2028 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2029 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2030 pf_cksum_fixup(pf_cksum_fixup(*ic,
2031 ooa.addr16[0], oa->addr16[0], u),
2032 ooa.addr16[1], oa->addr16[1], u),
2033 ooa.addr16[2], oa->addr16[2], u),
2034 ooa.addr16[3], oa->addr16[3], u),
2035 ooa.addr16[4], oa->addr16[4], u),
2036 ooa.addr16[5], oa->addr16[5], u),
2037 ooa.addr16[6], oa->addr16[6], u),
2038 ooa.addr16[7], oa->addr16[7], u);
2047 * Need to modulate the sequence numbers in the TCP SACK option
2048 * (credits to Krzysztof Pfaff for report and patch)
2051 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2052 struct tcphdr *th, struct pf_state_peer *dst)
2054 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2055 u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2056 int copyback = 0, i, olen;
2057 struct raw_sackblock sack;
2059 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2060 if (hlen < TCPOLEN_SACKLEN ||
2061 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2064 while (hlen >= TCPOLEN_SACKLEN) {
2067 case TCPOPT_EOL: /* FALLTHROUGH */
2075 if (olen >= TCPOLEN_SACKLEN) {
2076 for (i = 2; i + TCPOLEN_SACK <= olen;
2077 i += TCPOLEN_SACK) {
2078 memcpy(&sack, &opt[i], sizeof(sack));
2079 pf_change_a(&sack.rblk_start, &th->th_sum,
2080 htonl(ntohl(sack.rblk_start) -
2082 pf_change_a(&sack.rblk_end, &th->th_sum,
2083 htonl(ntohl(sack.rblk_end) -
2085 memcpy(&opt[i], &sack, sizeof(sack));
2099 m_copyback(m, off + sizeof(*th), thoptlen, opts);
2104 pf_send_tcp(const struct pf_rule *r, sa_family_t af,
2105 const struct pf_addr *saddr, const struct pf_addr *daddr,
2106 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2107 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2108 u_int16_t rtag, struct ether_header *eh, struct ifnet *ifp)
2113 struct ip *h = NULL;
2116 struct ip6_hdr *h6 = NULL;
2118 struct tcphdr *th = NULL;
2121 ASSERT_LWKT_TOKEN_HELD(&pf_token);
2123 /* maximum segment size tcp option */
2124 tlen = sizeof(struct tcphdr);
2131 len = sizeof(struct ip) + tlen;
2136 len = sizeof(struct ip6_hdr) + tlen;
2142 * Create outgoing mbuf.
2144 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
2145 * so make sure pf.flags is clear.
2147 m = m_gethdr(M_NOWAIT, MT_HEADER);
2152 m->m_pkthdr.fw_flags |= PF_MBUF_TAGGED;
2153 m->m_pkthdr.pf.flags = 0;
2154 m->m_pkthdr.pf.tag = rtag;
2155 /* XXX Recheck when upgrading to > 4.4 */
2156 m->m_pkthdr.pf.statekey = NULL;
2157 if (r != NULL && r->rtableid >= 0)
2158 m->m_pkthdr.pf.rtableid = r->rtableid;
2161 if (r != NULL && r->qid) {
2162 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE;
2163 m->m_pkthdr.pf.qid = r->qid;
2164 m->m_pkthdr.pf.ecn_af = af;
2165 m->m_pkthdr.pf.hdr = mtod(m, struct ip *);
2168 m->m_data += max_linkhdr;
2169 m->m_pkthdr.len = m->m_len = len;
2170 m->m_pkthdr.rcvif = NULL;
2171 bzero(m->m_data, len);
2175 h = mtod(m, struct ip *);
2177 /* IP header fields included in the TCP checksum */
2178 h->ip_p = IPPROTO_TCP;
2180 h->ip_src.s_addr = saddr->v4.s_addr;
2181 h->ip_dst.s_addr = daddr->v4.s_addr;
2183 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2188 h6 = mtod(m, struct ip6_hdr *);
2190 /* IP header fields included in the TCP checksum */
2191 h6->ip6_nxt = IPPROTO_TCP;
2192 h6->ip6_plen = htons(tlen);
2193 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2194 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2196 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2202 th->th_sport = sport;
2203 th->th_dport = dport;
2204 th->th_seq = htonl(seq);
2205 th->th_ack = htonl(ack);
2206 th->th_off = tlen >> 2;
2207 th->th_flags = flags;
2208 th->th_win = htons(win);
2211 opt = (char *)(th + 1);
2212 opt[0] = TCPOPT_MAXSEG;
2215 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2222 th->th_sum = in_cksum(m, len);
2224 /* Finish the IP header */
2226 h->ip_hl = sizeof(*h) >> 2;
2227 h->ip_tos = IPTOS_LOWDELAY;
2229 h->ip_off = path_mtu_discovery ? IP_DF : 0;
2230 h->ip_ttl = ttl ? ttl : ip_defttl;
2233 lwkt_reltoken(&pf_token);
2234 ip_output(m, NULL, NULL, 0, NULL, NULL);
2235 lwkt_gettoken(&pf_token);
2239 struct ether_header *e = (void *)ro.ro_dst.sa_data;
2247 ro.ro_dst.sa_len = sizeof(ro.ro_dst);
2248 ro.ro_dst.sa_family = pseudo_AF_HDRCMPLT;
2249 bcopy(eh->ether_dhost, e->ether_shost, ETHER_ADDR_LEN);
2250 bcopy(eh->ether_shost, e->ether_dhost, ETHER_ADDR_LEN);
2251 e->ether_type = eh->ether_type;
2252 /* XXX_IMPORT: later */
2253 lwkt_reltoken(&pf_token);
2254 ip_output(m, NULL, &ro, 0, NULL, NULL);
2255 lwkt_gettoken(&pf_token);
2262 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2263 sizeof(struct ip6_hdr), tlen);
2265 h6->ip6_vfc |= IPV6_VERSION;
2266 h6->ip6_hlim = IPV6_DEFHLIM;
2268 lwkt_reltoken(&pf_token);
2269 ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
2270 lwkt_gettoken(&pf_token);
2277 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2283 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
2284 * so make sure pf.flags is clear.
2286 if ((m0 = m_copy(m, 0, M_COPYALL)) == NULL)
2289 m0->m_pkthdr.fw_flags |= PF_MBUF_TAGGED;
2290 m0->m_pkthdr.pf.flags = 0;
2291 /* XXX Re-Check when Upgrading to > 4.4 */
2292 m0->m_pkthdr.pf.statekey = NULL;
2294 if (r->rtableid >= 0)
2295 m0->m_pkthdr.pf.rtableid = r->rtableid;
2299 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE;
2300 m0->m_pkthdr.pf.qid = r->qid;
2301 m0->m_pkthdr.pf.ecn_af = af;
2302 m0->m_pkthdr.pf.hdr = mtod(m0, struct ip *);
2309 icmp_error(m0, type, code, 0, 0);
2314 icmp6_error(m0, type, code, 0);
2321 * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2322 * If n is 0, they match if they are equal. If n is != 0, they match if they
2326 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2327 struct pf_addr *b, sa_family_t af)
2334 if ((a->addr32[0] & m->addr32[0]) ==
2335 (b->addr32[0] & m->addr32[0]))
2341 if (((a->addr32[0] & m->addr32[0]) ==
2342 (b->addr32[0] & m->addr32[0])) &&
2343 ((a->addr32[1] & m->addr32[1]) ==
2344 (b->addr32[1] & m->addr32[1])) &&
2345 ((a->addr32[2] & m->addr32[2]) ==
2346 (b->addr32[2] & m->addr32[2])) &&
2347 ((a->addr32[3] & m->addr32[3]) ==
2348 (b->addr32[3] & m->addr32[3])))
2367 * Return 1 if b <= a <= e, otherwise return 0.
2370 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2371 struct pf_addr *a, sa_family_t af)
2376 if ((a->addr32[0] < b->addr32[0]) ||
2377 (a->addr32[0] > e->addr32[0]))
2386 for (i = 0; i < 4; ++i)
2387 if (a->addr32[i] > b->addr32[i])
2389 else if (a->addr32[i] < b->addr32[i])
2392 for (i = 0; i < 4; ++i)
2393 if (a->addr32[i] < e->addr32[i])
2395 else if (a->addr32[i] > e->addr32[i])
2405 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2409 return ((p > a1) && (p < a2));
2411 return ((p < a1) || (p > a2));
2413 return ((p >= a1) && (p <= a2));
2427 return (0); /* never reached */
2431 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2436 return (pf_match(op, a1, a2, p));
2440 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2442 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2444 return (pf_match(op, a1, a2, u));
2448 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2450 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2452 return (pf_match(op, a1, a2, g));
2456 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag)
2459 *tag = m->m_pkthdr.pf.tag;
2461 return ((!r->match_tag_not && r->match_tag == *tag) ||
2462 (r->match_tag_not && r->match_tag != *tag));
2466 pf_tag_packet(struct mbuf *m, int tag, int rtableid)
2468 if (tag <= 0 && rtableid < 0)
2472 m->m_pkthdr.pf.tag = tag;
2474 m->m_pkthdr.pf.rtableid = rtableid;
2480 pf_step_into_anchor(int *depth, struct pf_ruleset **rs, int n,
2481 struct pf_rule **r, struct pf_rule **a, int *match)
2483 struct pf_anchor_stackframe *f;
2485 (*r)->anchor->match = 0;
2488 if (*depth >= NELEM(pf_anchor_stack)) {
2489 kprintf("pf_step_into_anchor: stack overflow\n");
2490 *r = TAILQ_NEXT(*r, entries);
2492 } else if (*depth == 0 && a != NULL)
2494 f = pf_anchor_stack + (*depth)++;
2497 if ((*r)->anchor_wildcard) {
2498 f->parent = &(*r)->anchor->children;
2499 if ((f->child = RB_MIN(pf_anchor_node, f->parent)) ==
2504 *rs = &f->child->ruleset;
2508 *rs = &(*r)->anchor->ruleset;
2510 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2514 pf_step_out_of_anchor(int *depth, struct pf_ruleset **rs, int n,
2515 struct pf_rule **r, struct pf_rule **a, int *match)
2517 struct pf_anchor_stackframe *f;
2523 f = pf_anchor_stack + *depth - 1;
2524 if (f->parent != NULL && f->child != NULL) {
2525 if (f->child->match ||
2526 (match != NULL && *match)) {
2527 f->r->anchor->match = 1;
2530 f->child = RB_NEXT(pf_anchor_node, f->parent, f->child);
2531 if (f->child != NULL) {
2532 *rs = &f->child->ruleset;
2533 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2541 if (*depth == 0 && a != NULL)
2544 if (f->r->anchor->match || (match != NULL && *match))
2545 quick = f->r->quick;
2546 *r = TAILQ_NEXT(f->r, entries);
2547 } while (*r == NULL);
2554 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2555 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2560 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2561 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2565 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2566 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2567 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2568 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2569 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2570 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2571 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2572 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2578 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2583 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2587 if (addr->addr32[3] == 0xffffffff) {
2588 addr->addr32[3] = 0;
2589 if (addr->addr32[2] == 0xffffffff) {
2590 addr->addr32[2] = 0;
2591 if (addr->addr32[1] == 0xffffffff) {
2592 addr->addr32[1] = 0;
2594 htonl(ntohl(addr->addr32[0]) + 1);
2597 htonl(ntohl(addr->addr32[1]) + 1);
2600 htonl(ntohl(addr->addr32[2]) + 1);
2603 htonl(ntohl(addr->addr32[3]) + 1);
2609 #define mix(a,b,c) \
2611 a -= b; a -= c; a ^= (c >> 13); \
2612 b -= c; b -= a; b ^= (a << 8); \
2613 c -= a; c -= b; c ^= (b >> 13); \
2614 a -= b; a -= c; a ^= (c >> 12); \
2615 b -= c; b -= a; b ^= (a << 16); \
2616 c -= a; c -= b; c ^= (b >> 5); \
2617 a -= b; a -= c; a ^= (c >> 3); \
2618 b -= c; b -= a; b ^= (a << 10); \
2619 c -= a; c -= b; c ^= (b >> 15); \
2623 * hash function based on bridge_hash in if_bridge.c
2626 pf_hash(struct pf_addr *inaddr, struct pf_addr *hash,
2627 struct pf_poolhashkey *key, sa_family_t af)
2629 u_int32_t a = 0x9e3779b9, b = 0x9e3779b9, c = key->key32[0];
2634 a += inaddr->addr32[0];
2637 hash->addr32[0] = c + key->key32[2];
2642 a += inaddr->addr32[0];
2643 b += inaddr->addr32[2];
2645 hash->addr32[0] = c;
2646 a += inaddr->addr32[1];
2647 b += inaddr->addr32[3];
2650 hash->addr32[1] = c;
2651 a += inaddr->addr32[2];
2652 b += inaddr->addr32[1];
2655 hash->addr32[2] = c;
2656 a += inaddr->addr32[3];
2657 b += inaddr->addr32[0];
2660 hash->addr32[3] = c;
2667 pf_map_addr(sa_family_t af, struct pf_rule *r, struct pf_addr *saddr,
2668 struct pf_addr *naddr, struct pf_addr *init_addr, struct pf_src_node **sn)
2670 unsigned char hash[16];
2671 struct pf_pool *rpool = &r->rpool;
2672 struct pf_pooladdr *acur = rpool->cur;
2673 struct pf_pooladdr *cur;
2674 struct pf_addr *raddr;
2675 struct pf_addr *rmask;
2676 struct pf_addr counter;
2677 struct pf_src_node k;
2678 int cpu = mycpu->gd_cpuid;
2681 bzero(hash, sizeof(hash)); /* avoid gcc warnings */
2684 * NOTE! rpool->cur and rpool->tblidx can be iterators and thus
2685 * may represent a SMP race due to the shared nature of the
2686 * rpool structure. We allow the race and ensure that updates
2687 * do not create a fatal condition.
2691 raddr = &cur->addr.v.a.addr;
2692 rmask = &cur->addr.v.a.mask;
2694 if (*sn == NULL && r->rpool.opts & PF_POOL_STICKYADDR &&
2695 (r->rpool.opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) {
2697 PF_ACPY(&k.addr, saddr, af);
2698 if (r->rule_flag & PFRULE_RULESRCTRACK ||
2699 r->rpool.opts & PF_POOL_STICKYADDR)
2703 pf_status.scounters[SCNT_SRC_NODE_SEARCH]++;
2704 *sn = RB_FIND(pf_src_tree, &tree_src_tracking[cpu], &k);
2705 if (*sn != NULL && !PF_AZERO(&(*sn)->raddr, af)) {
2706 PF_ACPY(naddr, &(*sn)->raddr, af);
2707 if (pf_status.debug >= PF_DEBUG_MISC) {
2708 kprintf("pf_map_addr: src tracking maps ");
2709 pf_print_host(&k.addr, 0, af);
2711 pf_print_host(naddr, 0, af);
2718 if (cur->addr.type == PF_ADDR_NOROUTE)
2720 if (cur->addr.type == PF_ADDR_DYNIFTL) {
2724 if (cur->addr.p.dyn->pfid_acnt4 < 1 &&
2725 (rpool->opts & PF_POOL_TYPEMASK) !=
2728 raddr = &cur->addr.p.dyn->pfid_addr4;
2729 rmask = &cur->addr.p.dyn->pfid_mask4;
2734 if (cur->addr.p.dyn->pfid_acnt6 < 1 &&
2735 (rpool->opts & PF_POOL_TYPEMASK) !=
2738 raddr = &cur->addr.p.dyn->pfid_addr6;
2739 rmask = &cur->addr.p.dyn->pfid_mask6;
2743 } else if (cur->addr.type == PF_ADDR_TABLE) {
2744 if ((rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_ROUNDROBIN)
2745 return (1); /* unsupported */
2747 raddr = &cur->addr.v.a.addr;
2748 rmask = &cur->addr.v.a.mask;
2751 switch (rpool->opts & PF_POOL_TYPEMASK) {
2753 PF_ACPY(naddr, raddr, af);
2755 case PF_POOL_BITMASK:
2756 PF_POOLMASK(naddr, raddr, rmask, saddr, af);
2758 case PF_POOL_RANDOM:
2759 if (init_addr != NULL && PF_AZERO(init_addr, af)) {
2763 counter.addr32[0] = htonl(karc4random());
2768 if (rmask->addr32[3] != 0xffffffff)
2770 htonl(karc4random());
2773 if (rmask->addr32[2] != 0xffffffff)
2775 htonl(karc4random());
2778 if (rmask->addr32[1] != 0xffffffff)
2780 htonl(karc4random());
2783 if (rmask->addr32[0] != 0xffffffff)
2785 htonl(karc4random());
2789 PF_POOLMASK(naddr, raddr, rmask, &counter, af);
2790 PF_ACPY(init_addr, naddr, af);
2793 counter = rpool->counter;
2795 PF_AINC(&counter, af);
2796 PF_POOLMASK(naddr, raddr, rmask, &counter, af);
2797 rpool->counter = counter;
2800 case PF_POOL_SRCHASH:
2801 pf_hash(saddr, (struct pf_addr *)&hash, &rpool->key, af);
2802 PF_POOLMASK(naddr, raddr, rmask, (struct pf_addr *)&hash, af);
2804 case PF_POOL_ROUNDROBIN:
2805 tblidx = rpool->tblidx;
2806 counter = rpool->counter;
2807 if (cur->addr.type == PF_ADDR_TABLE) {
2808 if (!pfr_pool_get(cur->addr.p.tbl,
2810 &raddr, &rmask, af)) {
2813 } else if (cur->addr.type == PF_ADDR_DYNIFTL) {
2814 if (!pfr_pool_get(cur->addr.p.dyn->pfid_kt,
2816 &raddr, &rmask, af)) {
2819 } else if (pf_match_addr(0, raddr, rmask,
2825 if ((cur = TAILQ_NEXT(cur, entries)) == NULL)
2826 cur = TAILQ_FIRST(&rpool->list);
2827 if (cur->addr.type == PF_ADDR_TABLE) {
2829 if (pfr_pool_get(cur->addr.p.tbl,
2831 &raddr, &rmask, af)) {
2832 /* table contains no address of type 'af' */
2837 } else if (cur->addr.type == PF_ADDR_DYNIFTL) {
2839 if (pfr_pool_get(cur->addr.p.dyn->pfid_kt,
2841 &raddr, &rmask, af)) {
2842 /* table contains no address of type 'af' */
2848 raddr = &cur->addr.v.a.addr;
2849 rmask = &cur->addr.v.a.mask;
2850 PF_ACPY(&counter, raddr, af);
2855 rpool->tblidx = tblidx;
2856 PF_ACPY(naddr, &counter, af);
2857 if (init_addr != NULL && PF_AZERO(init_addr, af))
2858 PF_ACPY(init_addr, naddr, af);
2859 PF_AINC(&counter, af);
2860 rpool->counter = counter;
2864 PF_ACPY(&(*sn)->raddr, naddr, af);
2866 if (pf_status.debug >= PF_DEBUG_MISC &&
2867 (rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) {
2868 kprintf("pf_map_addr: selected address ");
2869 pf_print_host(naddr, 0, af);
2877 pf_get_sport(struct pf_pdesc *pd, sa_family_t af,
2878 u_int8_t proto, struct pf_rule *r,
2879 struct pf_addr *saddr, struct pf_addr *daddr,
2880 u_int16_t sport, u_int16_t dport,
2881 struct pf_addr *naddr, u_int16_t *nport,
2882 u_int16_t low, u_int16_t high, struct pf_src_node **sn)
2884 struct pf_state_key_cmp key;
2885 struct pf_addr init_addr;
2887 u_int32_t hash_base = 0;
2890 bzero(&init_addr, sizeof(init_addr));
2891 if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn))
2894 if (proto == IPPROTO_ICMP) {
2899 bzero(&key, sizeof(key));
2902 key.port[0] = dport;
2903 PF_ACPY(&key.addr[0], daddr, key.af);
2906 PF_ACPY(&key.addr[1], naddr, key.af);
2909 * We want to select a port that calculates to a toeplitz hash
2910 * that masks to the same cpu, otherwise the response may
2911 * not see the new state.
2913 * We can still do this even if the kernel is disregarding
2914 * the hash and vectoring the packets to a specific cpu,
2915 * but it will reduce the number of ports we can use.
2919 if (proto == IPPROTO_TCP) {
2921 hash_base = toeplitz_piecemeal_port(dport) ^
2922 toeplitz_piecemeal_addr(daddr->v4.s_addr) ^
2923 toeplitz_piecemeal_addr(naddr->v4.s_addr);
2934 * port search; start random, step;
2935 * similar 2 portloop in in_pcbbind
2937 * WARNING! We try to match such that the kernel will
2938 * dispatch the translated host/port to the same
2939 * cpu, but this might not be possible.
2941 * In the case where the port is fixed, or for the
2942 * UDP case (whos toeplitz does not incorporate the
2943 * port), we set not_cpu_localized which ultimately
2944 * causes the pf_state_tree element
2946 * XXX fixed ports present a problem for cpu localization.
2948 if (!(proto == IPPROTO_TCP ||
2949 proto == IPPROTO_UDP ||
2950 proto == IPPROTO_ICMP)) {
2952 * non-specific protocol, leave port intact.
2954 key.port[1] = sport;
2955 if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
2957 pd->not_cpu_localized = 1;
2960 } else if (low == 0 && high == 0) {
2962 * static-port same as originator.
2964 key.port[1] = sport;
2965 if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
2967 pd->not_cpu_localized = 1;
2970 } else if (low == high) {
2972 * specific port as specified.
2974 key.port[1] = htons(low);
2975 if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
2976 *nport = htons(low);
2977 pd->not_cpu_localized = 1;
2982 * normal dynamic port
2992 cut = htonl(karc4random()) % (1 + high - low) + low;
2993 /* low <= cut <= high */
2994 for (tmp = cut; tmp <= high; ++(tmp)) {
2995 key.port[1] = htons(tmp);
3000 toeplitz_piecemeal_port(key.port[1]);
3001 if (netisr_hashcpu(hash) != mycpuid)
3004 if (pf_find_state_all(&key, PF_IN, NULL) ==
3005 NULL && !in_baddynamic(tmp, proto)) {
3006 if (proto == IPPROTO_UDP)
3007 pd->not_cpu_localized = 1;
3008 *nport = htons(tmp);
3012 for (tmp = cut - 1; tmp >= low; --(tmp)) {
3013 key.port[1] = htons(tmp);
3018 toeplitz_piecemeal_port(key.port[1]);
3019 if (netisr_hashcpu(hash) != mycpuid)
3022 if (pf_find_state_all(&key, PF_IN, NULL) ==
3023 NULL && !in_baddynamic(tmp, proto)) {
3024 if (proto == IPPROTO_UDP)
3025 pd->not_cpu_localized = 1;
3026 *nport = htons(tmp);
3035 switch (r->rpool.opts & PF_POOL_TYPEMASK) {
3036 case PF_POOL_RANDOM:
3037 case PF_POOL_ROUNDROBIN:
3038 if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn))
3042 case PF_POOL_SRCHASH:
3043 case PF_POOL_BITMASK:
3047 } while (! PF_AEQ(&init_addr, naddr, af) );
3048 return (1); /* none available */
3052 pf_match_translation(struct pf_pdesc *pd, struct mbuf *m, int off,
3053 int direction, struct pfi_kif *kif, struct pf_addr *saddr, u_int16_t sport,
3054 struct pf_addr *daddr, u_int16_t dport, int rs_num)
3056 struct pf_rule *r, *rm = NULL;
3057 struct pf_ruleset *ruleset = NULL;
3062 r = TAILQ_FIRST(pf_main_ruleset.rules[rs_num].active.ptr);
3063 while (r && rm == NULL) {
3064 struct pf_rule_addr *src = NULL, *dst = NULL;
3065 struct pf_addr_wrap *xdst = NULL;
3066 struct pf_pooladdr *cur;
3068 if (r->action == PF_BINAT && direction == PF_IN) {
3070 cur = r->rpool.cur; /* SMP race possible */
3080 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3081 r = r->skip[PF_SKIP_IFP].ptr;
3082 else if (r->direction && r->direction != direction)
3083 r = r->skip[PF_SKIP_DIR].ptr;
3084 else if (r->af && r->af != pd->af)
3085 r = r->skip[PF_SKIP_AF].ptr;
3086 else if (r->proto && r->proto != pd->proto)
3087 r = r->skip[PF_SKIP_PROTO].ptr;
3088 else if (PF_MISMATCHAW(&src->addr, saddr, pd->af,
3090 r = r->skip[src == &r->src ? PF_SKIP_SRC_ADDR :
3091 PF_SKIP_DST_ADDR].ptr;
3092 else if (src->port_op && !pf_match_port(src->port_op,
3093 src->port[0], src->port[1], sport))
3094 r = r->skip[src == &r->src ? PF_SKIP_SRC_PORT :
3095 PF_SKIP_DST_PORT].ptr;
3096 else if (dst != NULL &&
3097 PF_MISMATCHAW(&dst->addr, daddr, pd->af, dst->neg, NULL))
3098 r = r->skip[PF_SKIP_DST_ADDR].ptr;
3099 else if (xdst != NULL && PF_MISMATCHAW(xdst, daddr, pd->af,
3101 r = TAILQ_NEXT(r, entries);
3102 else if (dst != NULL && dst->port_op &&
3103 !pf_match_port(dst->port_op, dst->port[0],
3104 dst->port[1], dport))
3105 r = r->skip[PF_SKIP_DST_PORT].ptr;
3106 else if (r->match_tag && !pf_match_tag(m, r, &tag))
3107 r = TAILQ_NEXT(r, entries);
3108 else if (r->os_fingerprint != PF_OSFP_ANY && (pd->proto !=
3109 IPPROTO_TCP || !pf_osfp_match(pf_osfp_fingerprint(pd, m,
3110 off, pd->hdr.tcp), r->os_fingerprint)))
3111 r = TAILQ_NEXT(r, entries);
3115 if (r->rtableid >= 0)
3116 rtableid = r->rtableid;
3117 if (r->anchor == NULL) {
3120 pf_step_into_anchor(&asd, &ruleset, rs_num,
3124 pf_step_out_of_anchor(&asd, &ruleset, rs_num, &r,
3127 if (pf_tag_packet(m, tag, rtableid))
3129 if (rm != NULL && (rm->action == PF_NONAT ||
3130 rm->action == PF_NORDR || rm->action == PF_NOBINAT))
3136 pf_get_translation(struct pf_pdesc *pd, struct mbuf *m, int off, int direction,
3137 struct pfi_kif *kif, struct pf_src_node **sn,
3138 struct pf_state_key **skw, struct pf_state_key **sks,
3139 struct pf_state_key **skp, struct pf_state_key **nkp,
3140 struct pf_addr *saddr, struct pf_addr *daddr,
3141 u_int16_t sport, u_int16_t dport)
3143 struct pf_rule *r = NULL;
3145 if (direction == PF_OUT) {
3146 r = pf_match_translation(pd, m, off, direction, kif, saddr,
3147 sport, daddr, dport, PF_RULESET_BINAT);
3149 r = pf_match_translation(pd, m, off, direction, kif,
3150 saddr, sport, daddr, dport, PF_RULESET_NAT);
3152 r = pf_match_translation(pd, m, off, direction, kif, saddr,
3153 sport, daddr, dport, PF_RULESET_RDR);
3155 r = pf_match_translation(pd, m, off, direction, kif,
3156 saddr, sport, daddr, dport, PF_RULESET_BINAT);
3160 struct pf_addr *naddr;
3163 if (pf_state_key_setup(pd, r, skw, sks, skp, nkp,
3164 saddr, daddr, sport, dport))
3167 /* XXX We only modify one side for now. */
3168 naddr = &(*nkp)->addr[1];
3169 nport = &(*nkp)->port[1];
3172 * NOTE: Currently all translations will clear
3173 * BRIDGE_MBUF_TAGGED, telling the bridge to
3174 * ignore the original input encapsulation.
3176 switch (r->action) {
3182 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
3183 if (pf_get_sport(pd, pd->af, pd->proto, r,
3184 saddr, daddr, sport, dport,
3185 naddr, nport, r->rpool.proxy_port[0],
3186 r->rpool.proxy_port[1], sn)) {
3187 DPFPRINTF(PF_DEBUG_MISC,
3188 ("pf: NAT proxy port allocation "
3190 r->rpool.proxy_port[0],
3191 r->rpool.proxy_port[1]));
3196 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
3197 switch (direction) {
3199 if (r->rpool.cur->addr.type == PF_ADDR_DYNIFTL){
3203 if (r->rpool.cur->addr.p.dyn->
3207 &r->rpool.cur->addr.p.dyn->
3209 &r->rpool.cur->addr.p.dyn->
3216 if (r->rpool.cur->addr.p.dyn->
3220 &r->rpool.cur->addr.p.dyn->
3222 &r->rpool.cur->addr.p.dyn->
3230 &r->rpool.cur->addr.v.a.addr,
3231 &r->rpool.cur->addr.v.a.mask,
3235 if (r->src.addr.type == PF_ADDR_DYNIFTL) {
3239 if (r->src.addr.p.dyn->
3243 &r->src.addr.p.dyn->
3245 &r->src.addr.p.dyn->
3252 if (r->src.addr.p.dyn->
3256 &r->src.addr.p.dyn->
3258 &r->src.addr.p.dyn->
3266 &r->src.addr.v.a.addr,
3267 &r->src.addr.v.a.mask, daddr,
3273 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
3274 if (pf_map_addr(pd->af, r, saddr, naddr, NULL, sn))
3276 if ((r->rpool.opts & PF_POOL_TYPEMASK) ==
3278 PF_POOLMASK(naddr, naddr,
3279 &r->rpool.cur->addr.v.a.mask, daddr,
3282 if (r->rpool.proxy_port[1]) {
3283 u_int32_t tmp_nport;
3285 tmp_nport = ((ntohs(dport) -
3286 ntohs(r->dst.port[0])) %
3287 (r->rpool.proxy_port[1] -
3288 r->rpool.proxy_port[0] + 1)) +
3289 r->rpool.proxy_port[0];
3291 /* wrap around if necessary */
3292 if (tmp_nport > 65535)
3294 *nport = htons((u_int16_t)tmp_nport);
3295 } else if (r->rpool.proxy_port[0]) {
3296 *nport = htons(r->rpool.proxy_port[0]);
3298 pd->not_cpu_localized = 1;
3309 struct netmsg_hashlookup {
3310 struct netmsg_base base;
3311 struct inpcb **nm_pinp;
3312 struct inpcbinfo *nm_pcbinfo;
3313 struct pf_addr *nm_saddr;
3314 struct pf_addr *nm_daddr;
3320 #ifdef PF_SOCKET_LOOKUP_DOMSG
3322 in_pcblookup_hash_handler(netmsg_t msg)
3324 struct netmsg_hashlookup *rmsg = (struct netmsg_hashlookup *)msg;
3326 if (rmsg->nm_af == AF_INET)
3327 *rmsg->nm_pinp = in_pcblookup_hash(rmsg->nm_pcbinfo,
3328 rmsg->nm_saddr->v4, rmsg->nm_sport, rmsg->nm_daddr->v4,
3329 rmsg->nm_dport, INPLOOKUP_WILDCARD, NULL);
3332 *rmsg->nm_pinp = in6_pcblookup_hash(rmsg->nm_pcbinfo,
3333 &rmsg->nm_saddr->v6, rmsg->nm_sport, &rmsg->nm_daddr->v6,
3334 rmsg->nm_dport, INPLOOKUP_WILDCARD, NULL);
3336 lwkt_replymsg(&rmsg->base.lmsg, 0);
3338 #endif /* PF_SOCKET_LOOKUP_DOMSG */
3341 pf_socket_lookup(int direction, struct pf_pdesc *pd)
3343 struct pf_addr *saddr, *daddr;
3344 u_int16_t sport, dport;
3345 struct inpcbinfo *pi;
3347 struct netmsg_hashlookup *msg = NULL;
3348 #ifdef PF_SOCKET_LOOKUP_DOMSG
3349 struct netmsg_hashlookup msg0;
3355 pd->lookup.uid = UID_MAX;
3356 pd->lookup.gid = GID_MAX;
3357 pd->lookup.pid = NO_PID;
3358 if (direction == PF_IN) {
3365 switch (pd->proto) {
3367 if (pd->hdr.tcp == NULL)
3369 sport = pd->hdr.tcp->th_sport;
3370 dport = pd->hdr.tcp->th_dport;
3372 pi_cpu = tcp_addrcpu(saddr->v4.s_addr, sport, daddr->v4.s_addr, dport);
3373 pi = &tcbinfo[pi_cpu];
3375 * Our netstack runs lockless on MP systems
3376 * (only for TCP connections at the moment).
3378 * As we are not allowed to read another CPU's tcbinfo,
3379 * we have to ask that CPU via remote call to search the
3382 * Prepare a msg iff data belongs to another CPU.
3384 if (pi_cpu != mycpu->gd_cpuid) {
3385 #ifdef PF_SOCKET_LOOKUP_DOMSG
3389 * Following lwkt_domsg() is dangerous and could
3390 * lockup the network system, e.g.
3393 * netisr0 domsg to netisr1 (due to lookup)
3394 * netisr1 domsg to netisr0 (due to lookup)
3396 * We simply return -1 here, since we are probably
3397 * called before NAT, so the TCP packet should
3398 * already be on the correct CPU.
3401 netmsg_init(&msg->base, NULL, &curthread->td_msgport,
3402 0, in_pcblookup_hash_handler);
3403 msg->nm_pinp = &inp;
3404 msg->nm_pcbinfo = pi;
3405 msg->nm_saddr = saddr;
3406 msg->nm_sport = sport;
3407 msg->nm_daddr = daddr;
3408 msg->nm_dport = dport;
3409 msg->nm_af = pd->af;
3410 #else /* !PF_SOCKET_LOOKUP_DOMSG */
3411 kprintf("pf_socket_lookup: tcp packet not on the "
3412 "correct cpu %d, cur cpu %d\n",
3414 print_backtrace(-1);
3416 #endif /* PF_SOCKET_LOOKUP_DOMSG */
3420 if (pd->hdr.udp == NULL)
3422 sport = pd->hdr.udp->uh_sport;
3423 dport = pd->hdr.udp->uh_dport;
3424 pi = &udbinfo[mycpuid];
3429 if (direction != PF_IN) {
3440 * Query other CPU, second part
3442 * msg only gets initialized when:
3444 * 2) the info belongs to another CPU
3446 * Use some switch/case magic to avoid code duplication.
3449 inp = in6_pcblookup_hash(pi, &saddr->v6, sport,
3450 &daddr->v6, dport, INPLOOKUP_WILDCARD, NULL);
3456 /* FALLTHROUGH if SMP and on other CPU */
3460 lwkt_domsg(netisr_cpuport(pi_cpu),
3461 &msg->base.lmsg, 0);
3464 inp = in_pcblookup_hash(pi, saddr->v4, sport, daddr->v4,
3465 dport, INPLOOKUP_WILDCARD, NULL);
3474 pd->lookup.uid = inp->inp_socket->so_cred->cr_uid;
3475 pd->lookup.gid = inp->inp_socket->so_cred->cr_groups[0];
3480 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3484 u_int8_t *opt, optlen;
3485 u_int8_t wscale = 0;
3487 hlen = th_off << 2; /* hlen <= sizeof(hdr) */
3488 if (hlen <= sizeof(struct tcphdr))
3490 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3492 opt = hdr + sizeof(struct tcphdr);
3493 hlen -= sizeof(struct tcphdr);
3503 if (wscale > TCP_MAX_WINSHIFT)
3504 wscale = TCP_MAX_WINSHIFT;
3505 wscale |= PF_WSCALE_FLAG;
3520 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3524 u_int8_t *opt, optlen;
3525 u_int16_t mss = tcp_mssdflt;
3527 hlen = th_off << 2; /* hlen <= sizeof(hdr) */
3528 if (hlen <= sizeof(struct tcphdr))
3530 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3532 opt = hdr + sizeof(struct tcphdr);
3533 hlen -= sizeof(struct tcphdr);
3534 while (hlen >= TCPOLEN_MAXSEG) {
3542 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
3557 pf_calc_mss(struct pf_addr *addr, sa_family_t af, u_int16_t offer)
3560 struct sockaddr_in *dst;
3564 struct sockaddr_in6 *dst6;
3565 struct route_in6 ro6;
3567 struct rtentry *rt = NULL;
3569 u_int16_t mss = tcp_mssdflt;
3574 hlen = sizeof(struct ip);
3575 bzero(&ro, sizeof(ro));
3576 dst = (struct sockaddr_in *)&ro.ro_dst;
3577 dst->sin_family = AF_INET;
3578 dst->sin_len = sizeof(*dst);
3579 dst->sin_addr = addr->v4;
3580 rtalloc_ign(&ro, (RTF_CLONING | RTF_PRCLONING));
3586 hlen = sizeof(struct ip6_hdr);
3587 bzero(&ro6, sizeof(ro6));
3588 dst6 = (struct sockaddr_in6 *)&ro6.ro_dst;
3589 dst6->sin6_family = AF_INET6;
3590 dst6->sin6_len = sizeof(*dst6);
3591 dst6->sin6_addr = addr->v6;
3592 rtalloc_ign((struct route *)&ro6, (RTF_CLONING | RTF_PRCLONING));
3598 if (rt && rt->rt_ifp) {
3599 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr);
3600 mss = max(tcp_mssdflt, mss);
3603 mss = min(mss, offer);
3604 mss = max(mss, 64); /* sanity - at least max opt space */
3609 pf_set_rt_ifp(struct pf_state *s, struct pf_addr *saddr)
3611 struct pf_rule *r = s->rule.ptr;
3614 if (!r->rt || r->rt == PF_FASTROUTE)
3616 switch (s->key[PF_SK_WIRE]->af) {
3619 pf_map_addr(AF_INET, r, saddr, &s->rt_addr, NULL,
3621 s->rt_kif = r->rpool.cur->kif;
3626 pf_map_addr(AF_INET6, r, saddr, &s->rt_addr, NULL,
3628 s->rt_kif = r->rpool.cur->kif;
3635 pf_tcp_iss(struct pf_pdesc *pd)
3638 u_int32_t digest[4];
3640 if (pf_tcp_secret_init == 0) {
3641 lwkt_gettoken(&pf_gtoken);
3642 if (pf_tcp_secret_init == 0) {
3643 karc4rand(pf_tcp_secret, sizeof(pf_tcp_secret));
3644 MD5Init(&pf_tcp_secret_ctx);
3645 MD5Update(&pf_tcp_secret_ctx, pf_tcp_secret,
3646 sizeof(pf_tcp_secret));
3647 pf_tcp_secret_init = 1;
3649 lwkt_reltoken(&pf_gtoken);
3651 ctx = pf_tcp_secret_ctx;
3653 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
3654 MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
3655 if (pd->af == AF_INET6) {
3656 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3657 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3659 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3660 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3662 MD5Final((u_char *)digest, &ctx);
3663 pf_tcp_iss_off += 4096;
3665 return (digest[0] + pd->hdr.tcp->th_seq + pf_tcp_iss_off);
3669 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
3670 struct pfi_kif *kif, struct mbuf *m, int off, void *h,
3671 struct pf_pdesc *pd, struct pf_rule **am, struct pf_ruleset **rsm,
3672 struct ifqueue *ifq, struct inpcb *inp)
3674 struct pf_rule *nr = NULL;
3675 struct pf_addr *saddr = pd->src, *daddr = pd->dst;
3676 sa_family_t af = pd->af;
3677 struct pf_rule *r, *a = NULL;
3678 struct pf_ruleset *ruleset = NULL;
3679 struct pf_src_node *nsn = NULL;
3680 struct tcphdr *th = pd->hdr.tcp;
3681 struct pf_state_key *skw = NULL, *sks = NULL;
3682 struct pf_state_key *sk = NULL, *nk = NULL;
3684 int rewrite = 0, hdrlen = 0;
3685 int tag = -1, rtableid = -1;
3689 u_int16_t sport = 0, dport = 0;
3690 u_int16_t bproto_sum = 0, bip_sum = 0;
3691 u_int8_t icmptype = 0, icmpcode = 0;
3694 if (direction == PF_IN && pf_check_congestion(ifq)) {
3695 REASON_SET(&reason, PFRES_CONGEST);
3700 pd->lookup.done = pf_socket_lookup(direction, pd);
3701 else if (debug_pfugidhack) {
3702 DPFPRINTF(PF_DEBUG_MISC, ("pf: unlocked lookup\n"));
3703 pd->lookup.done = pf_socket_lookup(direction, pd);
3706 switch (pd->proto) {
3708 sport = th->th_sport;
3709 dport = th->th_dport;
3710 hdrlen = sizeof(*th);
3713 sport = pd->hdr.udp->uh_sport;
3714 dport = pd->hdr.udp->uh_dport;
3715 hdrlen = sizeof(*pd->hdr.udp);
3719 if (pd->af != AF_INET)
3721 sport = dport = pd->hdr.icmp->icmp_id;
3722 hdrlen = sizeof(*pd->hdr.icmp);
3723 icmptype = pd->hdr.icmp->icmp_type;
3724 icmpcode = pd->hdr.icmp->icmp_code;
3726 if (icmptype == ICMP_UNREACH ||
3727 icmptype == ICMP_SOURCEQUENCH ||
3728 icmptype == ICMP_REDIRECT ||
3729 icmptype == ICMP_TIMXCEED ||
3730 icmptype == ICMP_PARAMPROB)
3735 case IPPROTO_ICMPV6:
3738 sport = dport = pd->hdr.icmp6->icmp6_id;
3739 hdrlen = sizeof(*pd->hdr.icmp6);
3740 icmptype = pd->hdr.icmp6->icmp6_type;
3741 icmpcode = pd->hdr.icmp6->icmp6_code;
3743 if (icmptype == ICMP6_DST_UNREACH ||
3744 icmptype == ICMP6_PACKET_TOO_BIG ||
3745 icmptype == ICMP6_TIME_EXCEEDED ||
3746 icmptype == ICMP6_PARAM_PROB)
3751 sport = dport = hdrlen = 0;
3755 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3757 /* check packet for BINAT/NAT/RDR */
3758 if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn,
3759 &skw, &sks, &sk, &nk, saddr, daddr, sport, dport)) != NULL) {
3760 if (nk == NULL || sk == NULL) {
3761 REASON_SET(&reason, PFRES_MEMORY);
3766 bip_sum = *pd->ip_sum;
3768 m->m_flags &= ~M_HASH;
3769 switch (pd->proto) {
3771 bproto_sum = th->th_sum;
3772 pd->proto_sum = &th->th_sum;
3774 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3775 nk->port[pd->sidx] != sport) {
3776 pf_change_ap(saddr, &th->th_sport, pd->ip_sum,
3777 &th->th_sum, &nk->addr[pd->sidx],
3778 nk->port[pd->sidx], 0, af);
3779 pd->sport = &th->th_sport;
3780 sport = th->th_sport;
3783 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3784 nk->port[pd->didx] != dport) {
3785 pf_change_ap(daddr, &th->th_dport, pd->ip_sum,
3786 &th->th_sum, &nk->addr[pd->didx],
3787 nk->port[pd->didx], 0, af);
3788 dport = th->th_dport;
3789 pd->dport = &th->th_dport;
3794 bproto_sum = pd->hdr.udp->uh_sum;
3795 pd->proto_sum = &pd->hdr.udp->uh_sum;
3797 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3798 nk->port[pd->sidx] != sport) {
3799 pf_change_ap(saddr, &pd->hdr.udp->uh_sport,
3800 pd->ip_sum, &pd->hdr.udp->uh_sum,
3801 &nk->addr[pd->sidx],
3802 nk->port[pd->sidx], 1, af);
3803 sport = pd->hdr.udp->uh_sport;
3804 pd->sport = &pd->hdr.udp->uh_sport;
3807 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3808 nk->port[pd->didx] != dport) {
3809 pf_change_ap(daddr, &pd->hdr.udp->uh_dport,
3810 pd->ip_sum, &pd->hdr.udp->uh_sum,
3811 &nk->addr[pd->didx],
3812 nk->port[pd->didx], 1, af);
3813 dport = pd->hdr.udp->uh_dport;
3814 pd->dport = &pd->hdr.udp->uh_dport;
3820 nk->port[0] = nk->port[1];
3821 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3822 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3823 nk->addr[pd->sidx].v4.s_addr, 0);
3825 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3826 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3827 nk->addr[pd->didx].v4.s_addr, 0);
3829 if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3830 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3831 pd->hdr.icmp->icmp_cksum, sport,
3833 pd->hdr.icmp->icmp_id = nk->port[1];
3834 pd->sport = &pd->hdr.icmp->icmp_id;
3836 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3840 case IPPROTO_ICMPV6:
3841 nk->port[0] = nk->port[1];
3842 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3843 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3844 &nk->addr[pd->sidx], 0);
3846 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3847 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3848 &nk->addr[pd->didx], 0);
3857 &nk->addr[pd->sidx], AF_INET))
3858 pf_change_a(&saddr->v4.s_addr,
3860 nk->addr[pd->sidx].v4.s_addr, 0);
3863 &nk->addr[pd->didx], AF_INET))
3864 pf_change_a(&daddr->v4.s_addr,
3866 nk->addr[pd->didx].v4.s_addr, 0);
3872 &nk->addr[pd->sidx], AF_INET6))
3873 PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3876 &nk->addr[pd->didx], AF_INET6))
3877 PF_ACPY(saddr, &nk->addr[pd->didx], af);
3890 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3891 r = r->skip[PF_SKIP_IFP].ptr;
3892 else if (r->direction && r->direction != direction)
3893 r = r->skip[PF_SKIP_DIR].ptr;
3894 else if (r->af && r->af != af)
3895 r = r->skip[PF_SKIP_AF].ptr;
3896 else if (r->proto && r->proto != pd->proto)
3897 r = r->skip[PF_SKIP_PROTO].ptr;
3898 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3900 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3901 /* tcp/udp only. port_op always 0 in other cases */
3902 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3903 r->src.port[0], r->src.port[1], sport))
3904 r = r->skip[PF_SKIP_SRC_PORT].ptr;
3905 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3907 r = r->skip[PF_SKIP_DST_ADDR].ptr;
3908 /* tcp/udp only. port_op always 0 in other cases */
3909 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3910 r->dst.port[0], r->dst.port[1], dport))
3911 r = r->skip[PF_SKIP_DST_PORT].ptr;
3912 /* icmp only. type always 0 in other cases */
3913 else if (r->type && r->type != icmptype + 1)
3914 r = TAILQ_NEXT(r, entries);
3915 /* icmp only. type always 0 in other cases */
3916 else if (r->code && r->code != icmpcode + 1)
3917 r = TAILQ_NEXT(r, entries);
3918 else if (r->tos && !(r->tos == pd->tos))
3919 r = TAILQ_NEXT(r, entries);
3920 else if (r->rule_flag & PFRULE_FRAGMENT)
3921 r = TAILQ_NEXT(r, entries);
3922 else if (pd->proto == IPPROTO_TCP &&
3923 (r->flagset & th->th_flags) != r->flags)
3924 r = TAILQ_NEXT(r, entries);
3925 /* tcp/udp only. uid.op always 0 in other cases */
3926 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3927 pf_socket_lookup(direction, pd), 1)) &&
3928 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3930 r = TAILQ_NEXT(r, entries);
3931 /* tcp/udp only. gid.op always 0 in other cases */
3932 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3933 pf_socket_lookup(direction, pd), 1)) &&
3934 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3936 r = TAILQ_NEXT(r, entries);
3938 r->prob <= karc4random())
3939 r = TAILQ_NEXT(r, entries);
3940 else if (r->match_tag && !pf_match_tag(m, r, &tag))
3941 r = TAILQ_NEXT(r, entries);
3942 else if (r->os_fingerprint != PF_OSFP_ANY &&
3943 (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3944 pf_osfp_fingerprint(pd, m, off, th),
3945 r->os_fingerprint)))
3946 r = TAILQ_NEXT(r, entries);
3950 if (r->rtableid >= 0)
3951 rtableid = r->rtableid;
3952 if (r->anchor == NULL) {
3959 r = TAILQ_NEXT(r, entries);
3961 pf_step_into_anchor(&asd, &ruleset,
3962 PF_RULESET_FILTER, &r, &a, &match);
3964 if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset,
3965 PF_RULESET_FILTER, &r, &a, &match))
3972 REASON_SET(&reason, PFRES_MATCH);
3974 if (r->log || (nr != NULL && nr->log)) {
3976 m_copyback(m, off, hdrlen, pd->hdr.any);
3977 PFLOG_PACKET(kif, h, m, af, direction, reason, r->log ? r : nr,
3981 if ((r->action == PF_DROP) &&
3982 ((r->rule_flag & PFRULE_RETURNRST) ||
3983 (r->rule_flag & PFRULE_RETURNICMP) ||
3984 (r->rule_flag & PFRULE_RETURN))) {
3985 /* undo NAT changes, if they have taken place */
3987 PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3988 PF_ACPY(daddr, &sk->addr[pd->didx], af);
3990 *pd->sport = sk->port[pd->sidx];
3992 *pd->dport = sk->port[pd->didx];
3994 *pd->proto_sum = bproto_sum;
3996 *pd->ip_sum = bip_sum;
3997 m_copyback(m, off, hdrlen, pd->hdr.any);
3999 if (pd->proto == IPPROTO_TCP &&
4000 ((r->rule_flag & PFRULE_RETURNRST) ||
4001 (r->rule_flag & PFRULE_RETURN)) &&
4002 !(th->th_flags & TH_RST)) {
4003 u_int32_t ack = ntohl(th->th_seq) + pd->p_len;
4011 h4 = mtod(m, struct ip *);
4012 len = h4->ip_len - off;
4016 h6 = mtod(m, struct ip6_hdr *);
4017 len = h6->ip6_plen - (off - sizeof(*h6));
4022 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
4023 REASON_SET(&reason, PFRES_PROTCKSUM);
4025 if (th->th_flags & TH_SYN)
4027 if (th->th_flags & TH_FIN)
4029 pf_send_tcp(r, af, pd->dst,
4030 pd->src, th->th_dport, th->th_sport,
4031 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
4032 r->return_ttl, 1, 0, pd->eh, kif->pfik_ifp);
4034 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
4036 pf_send_icmp(m, r->return_icmp >> 8,
4037 r->return_icmp & 255, af, r);
4038 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
4040 pf_send_icmp(m, r->return_icmp6 >> 8,
4041 r->return_icmp6 & 255, af, r);
4044 if (r->action == PF_DROP)
4047 if (pf_tag_packet(m, tag, rtableid)) {
4048 REASON_SET(&reason, PFRES_MEMORY);
4052 if (!state_icmp && (r->keep_state || nr != NULL ||
4053 (pd->flags & PFDESC_TCP_NORM))) {
4055 action = pf_create_state(r, nr, a, pd, nsn, skw, sks, nk, sk, m,
4056 off, sport, dport, &rewrite, kif, sm, tag, bproto_sum,
4058 if (action != PF_PASS)
4062 /* copy back packet headers if we performed NAT operations */
4064 m_copyback(m, off, hdrlen, pd->hdr.any);
4070 kfree(sk, M_PFSTATEKEYPL);
4072 kfree(nk, M_PFSTATEKEYPL);
4077 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
4078 struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *skw,
4079 struct pf_state_key *sks, struct pf_state_key *nk, struct pf_state_key *sk,
4080 struct mbuf *m, int off, u_int16_t sport, u_int16_t dport, int *rewrite,
4081 struct pfi_kif *kif, struct pf_state **sm, int tag, u_int16_t bproto_sum,
4082 u_int16_t bip_sum, int hdrlen)
4084 struct pf_state *s = NULL;
4085 struct pf_src_node *sn = NULL;
4086 struct tcphdr *th = pd->hdr.tcp;
4087 u_int16_t mss = tcp_mssdflt;
4089 int cpu = mycpu->gd_cpuid;
4091 /* check maximums */
4092 if (r->max_states && (r->states_cur >= r->max_states)) {
4093 pf_status.lcounters[LCNT_STATES]++;
4094 REASON_SET(&reason, PFRES_MAXSTATES);
4097 /* src node for filter rule */
4098 if ((r->rule_flag & PFRULE_SRCTRACK ||
4099 r->rpool.opts & PF_POOL_STICKYADDR) &&
4100 pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
4101 REASON_SET(&reason, PFRES_SRCLIMIT);
4104 /* src node for translation rule */
4105 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
4106 pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
4107 REASON_SET(&reason, PFRES_SRCLIMIT);
4110 s = kmalloc(sizeof(struct pf_state), M_PFSTATEPL, M_NOWAIT|M_ZERO);
4112 REASON_SET(&reason, PFRES_MEMORY);
4115 lockinit(&s->lk, "pfstlk", 0, 0);
4116 s->id = 0; /* XXX Do we really need that? not in OpenBSD */
4119 s->nat_rule.ptr = nr;
4121 s->state_flags = PFSTATE_CREATEINPROG;
4122 STATE_INC_COUNTERS(s);
4124 s->state_flags |= PFSTATE_ALLOWOPTS;
4125 if (r->rule_flag & PFRULE_STATESLOPPY)
4126 s->state_flags |= PFSTATE_SLOPPY;
4127 if (pd->not_cpu_localized)
4128 s->state_flags |= PFSTATE_STACK_GLOBAL;
4130 s->log = r->log & PF_LOG_ALL;
4132 s->log |= nr->log & PF_LOG_ALL;
4133 switch (pd->proto) {
4135 s->src.seqlo = ntohl(th->th_seq);
4136 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
4137 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
4138 r->keep_state == PF_STATE_MODULATE) {
4139 /* Generate sequence number modulator */
4140 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
4143 pf_change_a(&th->th_seq, &th->th_sum,
4144 htonl(s->src.seqlo + s->src.seqdiff), 0);
4148 if (th->th_flags & TH_SYN) {
4150 s->src.wscale = pf_get_wscale(m, off,
4151 th->th_off, pd->af);
4153 s->src.max_win = MAX(ntohs(th->th_win), 1);
4154 if (s->src.wscale & PF_WSCALE_MASK) {
4155 /* Remove scale factor from initial window */
4156 int win = s->src.max_win;
4157 win += 1 << (s->src.wscale & PF_WSCALE_MASK);
4158 s->src.max_win = (win - 1) >>
4159 (s->src.wscale & PF_WSCALE_MASK);
4161 if (th->th_flags & TH_FIN)
4165 s->src.state = TCPS_SYN_SENT;
4166 s->dst.state = TCPS_CLOSED;
4167 s->timeout = PFTM_TCP_FIRST_PACKET;
4170 s->src.state = PFUDPS_SINGLE;
4171 s->dst.state = PFUDPS_NO_TRAFFIC;
4172 s->timeout = PFTM_UDP_FIRST_PACKET;
4176 case IPPROTO_ICMPV6:
4178 s->timeout = PFTM_ICMP_FIRST_PACKET;
4181 s->src.state = PFOTHERS_SINGLE;
4182 s->dst.state = PFOTHERS_NO_TRAFFIC;
4183 s->timeout = PFTM_OTHER_FIRST_PACKET;
4186 s->creation = time_second;
4187 s->expire = time_second;
4191 s->src_node->states++;
4194 /* XXX We only modify one side for now. */
4195 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
4196 s->nat_src_node = nsn;
4197 s->nat_src_node->states++;
4199 if (pd->proto == IPPROTO_TCP) {
4200 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
4201 off, pd, th, &s->src, &s->dst)) {
4202 REASON_SET(&reason, PFRES_MEMORY);
4203 pf_src_tree_remove_state(s);
4204 STATE_DEC_COUNTERS(s);
4205 kfree(s, M_PFSTATEPL);
4208 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
4209 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
4210 &s->src, &s->dst, rewrite)) {
4211 /* This really shouldn't happen!!! */
4212 DPFPRINTF(PF_DEBUG_URGENT,
4213 ("pf_normalize_tcp_stateful failed on first pkt"));
4214 pf_normalize_tcp_cleanup(s);
4215 pf_src_tree_remove_state(s);
4216 STATE_DEC_COUNTERS(s);
4217 kfree(s, M_PFSTATEPL);
4221 s->direction = pd->dir;
4223 if (sk == NULL && pf_state_key_setup(pd, nr, &skw, &sks, &sk, &nk,
4224 pd->src, pd->dst, sport, dport)) {
4225 REASON_SET(&reason, PFRES_MEMORY);
4229 if (pf_state_insert(BOUND_IFACE(r, kif), skw, sks, s)) {
4230 if (pd->proto == IPPROTO_TCP)
4231 pf_normalize_tcp_cleanup(s);
4232 REASON_SET(&reason, PFRES_STATEINS);
4233 pf_src_tree_remove_state(s);
4234 STATE_DEC_COUNTERS(s);
4235 kfree(s, M_PFSTATEPL);
4240 pf_set_rt_ifp(s, pd->src); /* needs s->state_key set */
4245 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
4246 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
4247 s->src.state = PF_TCPS_PROXY_SRC;
4248 /* undo NAT changes, if they have taken place */
4250 struct pf_state_key *skt = s->key[PF_SK_WIRE];
4251 if (pd->dir == PF_OUT)
4252 skt = s->key[PF_SK_STACK];
4253 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
4254 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
4256 *pd->sport = skt->port[pd->sidx];
4258 *pd->dport = skt->port[pd->didx];
4260 *pd->proto_sum = bproto_sum;
4262 *pd->ip_sum = bip_sum;
4263 m->m_flags &= ~M_HASH;
4264 m_copyback(m, off, hdrlen, pd->hdr.any);
4266 s->src.seqhi = htonl(karc4random());
4267 /* Find mss option */
4268 mss = pf_get_mss(m, off, th->th_off, pd->af);
4269 mss = pf_calc_mss(pd->src, pd->af, mss);
4270 mss = pf_calc_mss(pd->dst, pd->af, mss);
4272 s->state_flags &= ~PFSTATE_CREATEINPROG;
4273 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport,
4274 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
4275 TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL, NULL);
4276 REASON_SET(&reason, PFRES_SYNPROXY);
4277 return (PF_SYNPROXY_DROP);
4280 s->state_flags &= ~PFSTATE_CREATEINPROG;
4285 kfree(sk, M_PFSTATEKEYPL);
4287 kfree(nk, M_PFSTATEKEYPL);
4289 if (sn != NULL && sn->states == 0 && sn->expire == 0) {
4290 RB_REMOVE(pf_src_tree, &tree_src_tracking[cpu], sn);
4291 pf_status.scounters[SCNT_SRC_NODE_REMOVALS]++;
4292 atomic_add_int(&pf_status.src_nodes, -1);
4293 kfree(sn, M_PFSRCTREEPL);
4295 if (nsn != sn && nsn != NULL && nsn->states == 0 && nsn->expire == 0) {
4296 RB_REMOVE(pf_src_tree, &tree_src_tracking[cpu], nsn);
4297 pf_status.scounters[SCNT_SRC_NODE_REMOVALS]++;
4298 atomic_add_int(&pf_status.src_nodes, -1);
4299 kfree(nsn, M_PFSRCTREEPL);
4302 pf_src_tree_remove_state(s);
4303 STATE_DEC_COUNTERS(s);
4304 kfree(s, M_PFSTATEPL);
4311 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
4312 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
4313 struct pf_ruleset **rsm)
4315 struct pf_rule *r, *a = NULL;
4316 struct pf_ruleset *ruleset = NULL;
4317 sa_family_t af = pd->af;
4323 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
4326 if (pfi_kif_match(r->kif, kif) == r->ifnot)
4327 r = r->skip[PF_SKIP_IFP].ptr;
4328 else if (r->direction && r->direction != direction)
4329 r = r->skip[PF_SKIP_DIR].ptr;
4330 else if (r->af && r->af != af)
4331 r = r->skip[PF_SKIP_AF].ptr;
4332 else if (r->proto && r->proto != pd->proto)
4333 r = r->skip[PF_SKIP_PROTO].ptr;
4334 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
4336 r = r->skip[PF_SKIP_SRC_ADDR].ptr;
4337 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
4339 r = r->skip[PF_SKIP_DST_ADDR].ptr;
4340 else if (r->tos && !(r->tos == pd->tos))
4341 r = TAILQ_NEXT(r, entries);
4342 else if (r->os_fingerprint != PF_OSFP_ANY)
4343 r = TAILQ_NEXT(r, entries);
4344 else if (pd->proto == IPPROTO_UDP &&
4345 (r->src.port_op || r->dst.port_op))
4346 r = TAILQ_NEXT(r, entries);
4347 else if (pd->proto == IPPROTO_TCP &&
4348 (r->src.port_op || r->dst.port_op || r->flagset))
4349 r = TAILQ_NEXT(r, entries);
4350 else if ((pd->proto == IPPROTO_ICMP ||
4351 pd->proto == IPPROTO_ICMPV6) &&
4352 (r->type || r->code))
4353 r = TAILQ_NEXT(r, entries);
4354 else if (r->prob && r->prob <= karc4random())
4355 r = TAILQ_NEXT(r, entries);
4356 else if (r->match_tag && !pf_match_tag(m, r, &tag))
4357 r = TAILQ_NEXT(r, entries);
4359 if (r->anchor == NULL) {
4366 r = TAILQ_NEXT(r, entries);
4368 pf_step_into_anchor(&asd, &ruleset,
4369 PF_RULESET_FILTER, &r, &a, &match);
4371 if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset,
4372 PF_RULESET_FILTER, &r, &a, &match))
4379 REASON_SET(&reason, PFRES_MATCH);
4382 PFLOG_PACKET(kif, h, m, af, direction, reason, r, a, ruleset,
4385 if (r->action != PF_PASS)
4388 if (pf_tag_packet(m, tag, -1)) {
4389 REASON_SET(&reason, PFRES_MEMORY);
4397 * Called with state locked
4400 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
4401 struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
4402 struct pf_pdesc *pd, u_short *reason, int *copyback)
4404 struct tcphdr *th = pd->hdr.tcp;
4405 u_int16_t win = ntohs(th->th_win);
4406 u_int32_t ack, end, seq, orig_seq;
4410 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
4411 sws = src->wscale & PF_WSCALE_MASK;
4412 dws = dst->wscale & PF_WSCALE_MASK;
4418 * Sequence tracking algorithm from Guido van Rooij's paper:
4419 * http://www.madison-gurkha.com/publications/tcp_filtering/
4423 orig_seq = seq = ntohl(th->th_seq);
4424 if (src->seqlo == 0) {
4425 /* First packet from this end. Set its state */
4427 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
4428 src->scrub == NULL) {
4429 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
4430 REASON_SET(reason, PFRES_MEMORY);
4435 /* Deferred generation of sequence number modulator */
4436 if (dst->seqdiff && !src->seqdiff) {
4437 /* use random iss for the TCP server */
4438 while ((src->seqdiff = karc4random() - seq) == 0)
4440 ack = ntohl(th->th_ack) - dst->seqdiff;
4441 pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
4443 pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
4446 ack = ntohl(th->th_ack);
4449 end = seq + pd->p_len;
4450 if (th->th_flags & TH_SYN) {
4452 (*state)->sync_flags |= PFSTATE_GOT_SYN2;
4453 if (dst->wscale & PF_WSCALE_FLAG) {
4454 src->wscale = pf_get_wscale(m, off, th->th_off,
4456 if (src->wscale & PF_WSCALE_FLAG) {
4457 /* Remove scale factor from initial
4459 sws = src->wscale & PF_WSCALE_MASK;
4460 win = ((u_int32_t)win + (1 << sws) - 1)
4462 dws = dst->wscale & PF_WSCALE_MASK;
4464 /* fixup other window */
4465 dst->max_win <<= dst->wscale &
4467 /* in case of a retrans SYN|ACK */
4472 if (th->th_flags & TH_FIN)
4476 if (src->state < TCPS_SYN_SENT)
4477 src->state = TCPS_SYN_SENT;
4480 * May need to slide the window (seqhi may have been set by
4481 * the crappy stack check or if we picked up the connection
4482 * after establishment)
4484 if (src->seqhi == 1 ||
4485 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
4486 src->seqhi = end + MAX(1, dst->max_win << dws);
4487 if (win > src->max_win)
4491 ack = ntohl(th->th_ack) - dst->seqdiff;
4493 /* Modulate sequence numbers */
4494 pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
4496 pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
4499 end = seq + pd->p_len;
4500 if (th->th_flags & TH_SYN)
4502 if (th->th_flags & TH_FIN)
4506 if ((th->th_flags & TH_ACK) == 0) {
4507 /* Let it pass through the ack skew check */
4509 } else if ((ack == 0 &&
4510 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
4511 /* broken tcp stacks do not set ack */
4512 (dst->state < TCPS_SYN_SENT)) {
4514 * Many stacks (ours included) will set the ACK number in an
4515 * FIN|ACK if the SYN times out -- no sequence to ACK.
4521 /* Ease sequencing restrictions on no data packets */
4526 ackskew = dst->seqlo - ack;
4530 * Need to demodulate the sequence numbers in any TCP SACK options
4531 * (Selective ACK). We could optionally validate the SACK values
4532 * against the current ACK window, either forwards or backwards, but
4533 * I'm not confident that SACK has been implemented properly
4534 * everywhere. It wouldn't surprise me if several stacks accidently
4535 * SACK too far backwards of previously ACKed data. There really aren't
4536 * any security implications of bad SACKing unless the target stack
4537 * doesn't validate the option length correctly. Someone trying to
4538 * spoof into a TCP connection won't bother blindly sending SACK
4541 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
4542 if (pf_modulate_sack(m, off, pd, th, dst))
4547 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */
4548 if (SEQ_GEQ(src->seqhi, end) &&
4549 /* Last octet inside other's window space */
4550 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
4551 /* Retrans: not more than one window back */
4552 (ackskew >= -MAXACKWINDOW) &&
4553 /* Acking not more than one reassembled fragment backwards */
4554 (ackskew <= (MAXACKWINDOW << sws)) &&
4555 /* Acking not more than one window forward */
4556 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
4557 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
4558 (pd->flags & PFDESC_IP_REAS) == 0)) {
4559 /* Require an exact/+1 sequence match on resets when possible */
4561 if (dst->scrub || src->scrub) {
4562 if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4563 *state, src, dst, copyback))
4567 /* update max window */
4568 if (src->max_win < win)
4570 /* synchronize sequencing */
4571 if (SEQ_GT(end, src->seqlo))
4573 /* slide the window of what the other end can send */
4574 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4575 dst->seqhi = ack + MAX((win << sws), 1);
4579 if (th->th_flags & TH_SYN)
4580 if (src->state < TCPS_SYN_SENT)
4581 src->state = TCPS_SYN_SENT;
4582 if (th->th_flags & TH_FIN)
4583 if (src->state < TCPS_CLOSING)
4584 src->state = TCPS_CLOSING;
4585 if (th->th_flags & TH_ACK) {
4586 if (dst->state == TCPS_SYN_SENT) {
4587 dst->state = TCPS_ESTABLISHED;
4588 if (src->state == TCPS_ESTABLISHED &&
4589 (*state)->src_node != NULL &&
4590 pf_src_connlimit(*state)) {
4591 REASON_SET(reason, PFRES_SRCLIMIT);
4594 } else if (dst->state == TCPS_CLOSING)
4595 dst->state = TCPS_FIN_WAIT_2;
4597 if (th->th_flags & TH_RST)
4598 src->state = dst->state = TCPS_TIME_WAIT;
4600 /* update expire time */
4601 (*state)->expire = time_second;
4602 if (src->state >= TCPS_FIN_WAIT_2 &&
4603 dst->state >= TCPS_FIN_WAIT_2)
4604 (*state)->timeout = PFTM_TCP_CLOSED;
4605 else if (src->state >= TCPS_CLOSING &&
4606 dst->state >= TCPS_CLOSING)
4607 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4608 else if (src->state < TCPS_ESTABLISHED ||
4609 dst->state < TCPS_ESTABLISHED)
4610 (*state)->timeout = PFTM_TCP_OPENING;
4611 else if (src->state >= TCPS_CLOSING ||
4612 dst->state >= TCPS_CLOSING)
4613 (*state)->timeout = PFTM_TCP_CLOSING;
4615 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4617 /* Fall through to PASS packet */
4619 } else if ((dst->state < TCPS_SYN_SENT ||
4620 dst->state >= TCPS_FIN_WAIT_2 ||
4621 src->state >= TCPS_FIN_WAIT_2) &&
4622 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4623 /* Within a window forward of the originating packet */
4624 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4625 /* Within a window backward of the originating packet */
4628 * This currently handles three situations:
4629 * 1) Stupid stacks will shotgun SYNs before their peer
4631 * 2) When PF catches an already established stream (the
4632 * firewall rebooted, the state table was flushed, routes
4634 * 3) Packets get funky immediately after the connection
4635 * closes (this should catch Solaris spurious ACK|FINs
4636 * that web servers like to spew after a close)
4638 * This must be a little more careful than the above code
4639 * since packet floods will also be caught here. We don't
4640 * update the TTL here to mitigate the damage of a packet
4641 * flood and so the same code can handle awkward establishment
4642 * and a loosened connection close.
4643 * In the establishment case, a correct peer response will
4644 * validate the connection, go through the normal state code
4645 * and keep updating the state TTL.
4648 if (pf_status.debug >= PF_DEBUG_MISC) {
4649 kprintf("pf: loose state match: ");
4650 pf_print_state(*state);
4651 pf_print_flags(th->th_flags);
4652 kprintf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4653 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, pd->p_len,
4654 ackskew, (unsigned long long)(*state)->packets[0],
4655 (unsigned long long)(*state)->packets[1],
4656 pd->dir == PF_IN ? "in" : "out",
4657 pd->dir == (*state)->direction ? "fwd" : "rev");
4660 if (dst->scrub || src->scrub) {
4661 if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4662 *state, src, dst, copyback))
4666 /* update max window */
4667 if (src->max_win < win)
4669 /* synchronize sequencing */
4670 if (SEQ_GT(end, src->seqlo))
4672 /* slide the window of what the other end can send */
4673 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4674 dst->seqhi = ack + MAX((win << sws), 1);
4677 * Cannot set dst->seqhi here since this could be a shotgunned
4678 * SYN and not an already established connection.
4681 if (th->th_flags & TH_FIN)
4682 if (src->state < TCPS_CLOSING)
4683 src->state = TCPS_CLOSING;
4684 if (th->th_flags & TH_RST)
4685 src->state = dst->state = TCPS_TIME_WAIT;
4687 /* Fall through to PASS packet */
4689 } else if ((*state)->pickup_mode == PF_PICKUPS_HASHONLY ||
4690 ((*state)->pickup_mode == PF_PICKUPS_ENABLED &&
4691 ((*state)->sync_flags & PFSTATE_GOT_SYN_MASK) !=
4692 PFSTATE_GOT_SYN_MASK)) {
4694 * If pickup mode is hash only, do not fail on sequence checks.
4696 * If pickup mode is enabled and we did not see the SYN in
4697 * both direction, do not fail on sequence checks because
4698 * we do not have complete information on window scale.
4700 * Adjust expiration and fall through to PASS packet.
4701 * XXX Add a FIN check to reduce timeout?
4703 (*state)->expire = time_second;
4706 * Failure processing
4708 if ((*state)->dst.state == TCPS_SYN_SENT &&
4709 (*state)->src.state == TCPS_SYN_SENT) {
4710 /* Send RST for state mismatches during handshake */
4711 if (!(th->th_flags & TH_RST))
4712 pf_send_tcp((*state)->rule.ptr, pd->af,
4713 pd->dst, pd->src, th->th_dport,
4714 th->th_sport, ntohl(th->th_ack), 0,
4716 (*state)->rule.ptr->return_ttl, 1, 0,
4717 pd->eh, kif->pfik_ifp);
4721 } else if (pf_status.debug >= PF_DEBUG_MISC) {
4722 kprintf("pf: BAD state: ");
4723 pf_print_state(*state);
4724 pf_print_flags(th->th_flags);
4725 kprintf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4726 "pkts=%llu:%llu dir=%s,%s\n",
4727 seq, orig_seq, ack, pd->p_len, ackskew,
4728 (unsigned long long)(*state)->packets[0],
4729 (unsigned long long)(*state)->packets[1],
4730 pd->dir == PF_IN ? "in" : "out",
4731 pd->dir == (*state)->direction ? "fwd" : "rev");
4732 kprintf("pf: State failure on: %c %c %c %c | %c %c\n",
4733 SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4734 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4736 (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4737 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4738 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4739 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4741 REASON_SET(reason, PFRES_BADSTATE);
4749 * Called with state locked
4752 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4753 struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4755 struct tcphdr *th = pd->hdr.tcp;
4757 if (th->th_flags & TH_SYN)
4758 if (src->state < TCPS_SYN_SENT)
4759 src->state = TCPS_SYN_SENT;
4760 if (th->th_flags & TH_FIN)
4761 if (src->state < TCPS_CLOSING)
4762 src->state = TCPS_CLOSING;
4763 if (th->th_flags & TH_ACK) {
4764 if (dst->state == TCPS_SYN_SENT) {
4765 dst->state = TCPS_ESTABLISHED;
4766 if (src->state == TCPS_ESTABLISHED &&
4767 (*state)->src_node != NULL &&
4768 pf_src_connlimit(*state)) {
4769 REASON_SET(reason, PFRES_SRCLIMIT);
4772 } else if (dst->state == TCPS_CLOSING) {
4773 dst->state = TCPS_FIN_WAIT_2;
4774 } else if (src->state == TCPS_SYN_SENT &&
4775 dst->state < TCPS_SYN_SENT) {
4777 * Handle a special sloppy case where we only see one
4778 * half of the connection. If there is a ACK after
4779 * the initial SYN without ever seeing a packet from
4780 * the destination, set the connection to established.
4782 dst->state = src->state = TCPS_ESTABLISHED;
4783 if ((*state)->src_node != NULL &&
4784 pf_src_connlimit(*state)) {
4785 REASON_SET(reason, PFRES_SRCLIMIT);
4788 } else if (src->state == TCPS_CLOSING &&
4789 dst->state == TCPS_ESTABLISHED &&
4792 * Handle the closing of half connections where we
4793 * don't see the full bidirectional FIN/ACK+ACK
4796 dst->state = TCPS_CLOSING;
4799 if (th->th_flags & TH_RST)
4800 src->state = dst->state = TCPS_TIME_WAIT;
4802 /* update expire time */
4803 (*state)->expire = time_second;
4804 if (src->state >= TCPS_FIN_WAIT_2 &&
4805 dst->state >= TCPS_FIN_WAIT_2)
4806 (*state)->timeout = PFTM_TCP_CLOSED;
4807 else if (src->state >= TCPS_CLOSING &&
4808 dst->state >= TCPS_CLOSING)
4809 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4810 else if (src->state < TCPS_ESTABLISHED ||
4811 dst->state < TCPS_ESTABLISHED)
4812 (*state)->timeout = PFTM_TCP_OPENING;
4813 else if (src->state >= TCPS_CLOSING ||
4814 dst->state >= TCPS_CLOSING)
4815 (*state)->timeout = PFTM_TCP_CLOSING;
4817 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4823 * Test TCP connection state. Caller must hold the state locked.
4826 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4827 struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4830 struct pf_state_key_cmp key;
4831 struct tcphdr *th = pd->hdr.tcp;
4834 struct pf_state_peer *src, *dst;
4835 struct pf_state_key *sk;
4837 bzero(&key, sizeof(key));
4839 key.proto = IPPROTO_TCP;
4840 if (direction == PF_IN) { /* wire side, straight */
4841 PF_ACPY(&key.addr[0], pd->src, key.af);
4842 PF_ACPY(&key.addr[1], pd->dst, key.af);
4843 key.port[0] = th->th_sport;
4844 key.port[1] = th->th_dport;
4845 if (pf_status.debug >= PF_DEBUG_MISC) {
4846 kprintf("test-tcp IN (%08x:%d) -> (%08x:%d)\n",
4847 ntohl(key.addr[0].addr32[0]),
4849 ntohl(key.addr[1].addr32[0]),
4850 ntohs(key.port[1]));
4852 } else { /* stack side, reverse */
4853 PF_ACPY(&key.addr[1], pd->src, key.af);
4854 PF_ACPY(&key.addr[0], pd->dst, key.af);
4855 key.port[1] = th->th_sport;
4856 key.port[0] = th->th_dport;
4857 if (pf_status.debug >= PF_DEBUG_MISC) {
4858 kprintf("test-tcp OUT (%08x:%d) <- (%08x:%d)\n",
4859 ntohl(key.addr[0].addr32[0]),
4861 ntohl(key.addr[1].addr32[0]),
4862 ntohs(key.port[1]));
4866 STATE_LOOKUP(kif, &key, direction, *state, m);
4867 lockmgr(&(*state)->lk, LK_EXCLUSIVE);
4869 if (direction == (*state)->direction) {
4870 src = &(*state)->src;
4871 dst = &(*state)->dst;
4873 src = &(*state)->dst;
4874 dst = &(*state)->src;
4877 sk = (*state)->key[pd->didx];
4879 if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4880 if (direction != (*state)->direction) {
4881 REASON_SET(reason, PFRES_SYNPROXY);
4882 FAIL (PF_SYNPROXY_DROP);
4884 if (th->th_flags & TH_SYN) {
4885 if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4886 REASON_SET(reason, PFRES_SYNPROXY);
4889 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
4890 pd->src, th->th_dport, th->th_sport,
4891 (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4892 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1,
4894 REASON_SET(reason, PFRES_SYNPROXY);
4895 FAIL (PF_SYNPROXY_DROP);
4896 } else if (!(th->th_flags & TH_ACK) ||
4897 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4898 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4899 REASON_SET(reason, PFRES_SYNPROXY);
4901 } else if ((*state)->src_node != NULL &&
4902 pf_src_connlimit(*state)) {
4903 REASON_SET(reason, PFRES_SRCLIMIT);
4906 (*state)->src.state = PF_TCPS_PROXY_DST;
4908 if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4909 if (direction == (*state)->direction) {
4910 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4911 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4912 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4913 REASON_SET(reason, PFRES_SYNPROXY);
4916 (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4917 if ((*state)->dst.seqhi == 1)
4918 (*state)->dst.seqhi = htonl(karc4random());
4919 pf_send_tcp((*state)->rule.ptr, pd->af,
4920 &sk->addr[pd->sidx], &sk->addr[pd->didx],
4921 sk->port[pd->sidx], sk->port[pd->didx],
4922 (*state)->dst.seqhi, 0, TH_SYN, 0,
4923 (*state)->src.mss, 0, 0, (*state)->tag, NULL, NULL);
4924 REASON_SET(reason, PFRES_SYNPROXY);
4925 FAIL (PF_SYNPROXY_DROP);
4926 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4928 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4929 REASON_SET(reason, PFRES_SYNPROXY);
4932 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4933 (*state)->dst.seqlo = ntohl(th->th_seq);
4934 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst,
4935 pd->src, th->th_dport, th->th_sport,
4936 ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4937 TH_ACK, (*state)->src.max_win, 0, 0, 0,
4938 (*state)->tag, NULL, NULL);
4939 pf_send_tcp((*state)->rule.ptr, pd->af,
4940 &sk->addr[pd->sidx], &sk->addr[pd->didx],
4941 sk->port[pd->sidx], sk->port[pd->didx],
4942 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4943 TH_ACK, (*state)->dst.max_win, 0, 0, 1,
4945 (*state)->src.seqdiff = (*state)->dst.seqhi -
4946 (*state)->src.seqlo;
4947 (*state)->dst.seqdiff = (*state)->src.seqhi -
4948 (*state)->dst.seqlo;
4949 (*state)->src.seqhi = (*state)->src.seqlo +
4950 (*state)->dst.max_win;
4951 (*state)->dst.seqhi = (*state)->dst.seqlo +
4952 (*state)->src.max_win;
4953 (*state)->src.wscale = (*state)->dst.wscale = 0;
4954 (*state)->src.state = (*state)->dst.state =
4956 REASON_SET(reason, PFRES_SYNPROXY);
4957 FAIL (PF_SYNPROXY_DROP);
4962 * Check for connection (addr+port pair) reuse. We can't actually
4963 * unlink the state if we don't own it.
4965 if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4966 dst->state >= TCPS_FIN_WAIT_2 &&
4967 src->state >= TCPS_FIN_WAIT_2) {
4968 if (pf_status.debug >= PF_DEBUG_MISC) {
4969 kprintf("pf: state reuse ");
4970 pf_print_state(*state);
4971 pf_print_flags(th->th_flags);
4974 /* XXX make sure it's the same direction ?? */
4975 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4976 if ((*state)->cpuid == mycpu->gd_cpuid) {
4977 pf_unlink_state(*state);
4980 (*state)->timeout = PFTM_PURGE;
4985 if ((*state)->state_flags & PFSTATE_SLOPPY) {
4986 if (pf_tcp_track_sloppy(src, dst, state, pd,
4987 reason) == PF_DROP) {
4991 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd,
4992 reason, ©back) == PF_DROP) {
4997 /* translate source/destination address, if necessary */
4998 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4999 struct pf_state_key *nk = (*state)->key[pd->didx];
5001 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
5002 nk->port[pd->sidx] != th->th_sport) {
5004 * The translated source address may be completely
5005 * unrelated to the saved link header, make sure
5006 * a bridge doesn't try to use it.
5008 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
5009 pf_change_ap(pd->src, &th->th_sport, pd->ip_sum,
5010 &th->th_sum, &nk->addr[pd->sidx],
5011 nk->port[pd->sidx], 0, pd->af);
5014 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
5015 nk->port[pd->didx] != th->th_dport) {
5017 * If we don't redispatch the packet will go into
5018 * the protocol stack on the wrong cpu for the
5019 * post-translated address.
5021 pf_change_ap(pd->dst, &th->th_dport, pd->ip_sum,
5022 &th->th_sum, &nk->addr[pd->didx],
5023 nk->port[pd->didx], 0, pd->af);
5028 /* Copyback sequence modulation or stateful scrub changes if needed */
5030 m->m_flags &= ~M_HASH;
5031 m_copyback(m, off, sizeof(*th), (caddr_t)th);
5034 pfsync_update_state(*state);
5038 lockmgr(&(*state)->lk, LK_RELEASE);
5043 * Test UDP connection state. Caller must hold the state locked.
5046 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
5047 struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
5049 struct pf_state_peer *src, *dst;
5050 struct pf_state_key_cmp key;
5051 struct udphdr *uh = pd->hdr.udp;
5053 bzero(&key, sizeof(key));
5055 key.proto = IPPROTO_UDP;
5056 if (direction == PF_IN) { /* wire side, straight */
5057 PF_ACPY(&key.addr[0], pd->src, key.af);
5058 PF_ACPY(&key.addr[1], pd->dst, key.af);
5059 key.port[0] = uh->uh_sport;
5060 key.port[1] = uh->uh_dport;
5061 } else { /* stack side, reverse */
5062 PF_ACPY(&key.addr[1], pd->src, key.af);
5063 PF_ACPY(&key.addr[0], pd->dst, key.af);
5064 key.port[1] = uh->uh_sport;
5065 key.port[0] = uh->uh_dport;
5068 STATE_LOOKUP(kif, &key, direction, *state, m);
5069 lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5071 if (direction == (*state)->direction) {
5072 src = &(*state)->src;
5073 dst = &(*state)->dst;
5075 src = &(*state)->dst;
5076 dst = &(*state)->src;
5080 if (src->state < PFUDPS_SINGLE)
5081 src->state = PFUDPS_SINGLE;
5082 if (dst->state == PFUDPS_SINGLE)
5083 dst->state = PFUDPS_MULTIPLE;
5085 /* update expire time */
5086 (*state)->expire = time_second;
5087 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
5088 (*state)->timeout = PFTM_UDP_MULTIPLE;
5090 (*state)->timeout = PFTM_UDP_SINGLE;
5092 /* translate source/destination address, if necessary */
5093 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5094 struct pf_state_key *nk = (*state)->key[pd->didx];
5096 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
5097 nk->port[pd->sidx] != uh->uh_sport) {
5099 * The translated source address may be completely
5100 * unrelated to the saved link header, make sure
5101 * a bridge doesn't try to use it.
5103 m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
5104 m->m_flags &= ~M_HASH;
5105 pf_change_ap(pd->src, &uh->uh_sport, pd->ip_sum,
5106 &uh->uh_sum, &nk->addr[pd->sidx],
5107 nk->port[pd->sidx], 1, pd->af);
5110 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
5111 nk->port[pd->didx] != uh->uh_dport) {
5113 * If we don't redispatch the packet will go into
5114 * the protocol stack on the wrong cpu for the
5115 * post-translated address.
5117 m->m_flags &= ~M_HASH;
5118 pf_change_ap(pd->dst, &uh->uh_dport, pd->ip_sum,
5119 &uh->uh_sum, &nk->addr[pd->didx],
5120 nk->port[pd->didx], 1, pd->af);
5122 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
5125 pfsync_update_state(*state);
5126 lockmgr(&(*state)->lk, LK_RELEASE);
5131 * Test ICMP connection state. Caller must hold the state locked.
5134 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
5135 struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
5138 struct pf_addr *saddr = pd->src, *daddr = pd->dst;
5139 u_int16_t icmpid = 0, *icmpsum = NULL;
5140 u_int8_t icmptype = 0;
5143 struct pf_state_key_cmp key;
5145 bzero(&key, sizeof(key));
5147 switch (pd->proto) {
5150 icmptype = pd->hdr.icmp->icmp_type;
5151 icmpid = pd->hdr.icmp->icmp_id;
5152 icmpsum = &pd->hdr.icmp->icmp_cksum;
5154 if (icmptype == ICMP_UNREACH ||
5155 icmptype == ICMP_SOURCEQUENCH ||
5156 icmptype == ICMP_REDIRECT ||
5157 icmptype == ICMP_TIMXCEED ||
5158 icmptype == ICMP_PARAMPROB)
5163 case IPPROTO_ICMPV6:
5164 icmptype = pd->hdr.icmp6->icmp6_type;
5165 icmpid = pd->hdr.icmp6->icmp6_id;
5166 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
5168 if (icmptype == ICMP6_DST_UNREACH ||
5169 icmptype == ICMP6_PACKET_TOO_BIG ||
5170 icmptype == ICMP6_TIME_EXCEEDED ||
5171 icmptype == ICMP6_PARAM_PROB)
5180 * ICMP query/reply message not related to a TCP/UDP packet.
5181 * Search for an ICMP state.
5184 key.proto = pd->proto;
5185 key.port[0] = key.port[1] = icmpid;
5186 if (direction == PF_IN) { /* wire side, straight */
5187 PF_ACPY(&key.addr[0], pd->src, key.af);
5188 PF_ACPY(&key.addr[1], pd->dst, key.af);
5189 } else { /* stack side, reverse */
5190 PF_ACPY(&key.addr[1], pd->src, key.af);
5191 PF_ACPY(&key.addr[0], pd->dst, key.af);
5194 STATE_LOOKUP(kif, &key, direction, *state, m);
5195 lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5197 (*state)->expire = time_second;
5198 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
5200 /* translate source/destination address, if necessary */
5201 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5202 struct pf_state_key *nk = (*state)->key[pd->didx];
5207 if (PF_ANEQ(pd->src,
5208 &nk->addr[pd->sidx], AF_INET))
5209 pf_change_a(&saddr->v4.s_addr,
5211 nk->addr[pd->sidx].v4.s_addr, 0);
5213 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
5215 pf_change_a(&daddr->v4.s_addr,
5217 nk->addr[pd->didx].v4.s_addr, 0);
5220 pd->hdr.icmp->icmp_id) {
5221 pd->hdr.icmp->icmp_cksum =
5223 pd->hdr.icmp->icmp_cksum, icmpid,
5224 nk->port[pd->sidx], 0);
5225 pd->hdr.icmp->icmp_id =
5229 m->m_flags &= ~M_HASH;
5230 m_copyback(m, off, ICMP_MINLEN,
5231 (caddr_t)pd->hdr.icmp);
5236 if (PF_ANEQ(pd->src,
5237 &nk->addr[pd->sidx], AF_INET6))
5239 &pd->hdr.icmp6->icmp6_cksum,
5240 &nk->addr[pd->sidx], 0);
5242 if (PF_ANEQ(pd->dst,
5243 &nk->addr[pd->didx], AF_INET6))
5245 &pd->hdr.icmp6->icmp6_cksum,
5246 &nk->addr[pd->didx], 0);
5248 m->m_flags &= ~M_HASH;
5250 sizeof(struct icmp6_hdr),
5251 (caddr_t)pd->hdr.icmp6);
5258 * ICMP error message in response to a TCP/UDP packet.
5259 * Extract the inner TCP/UDP header and search for that state.
5262 struct pf_pdesc pd2;
5267 struct ip6_hdr h2_6;
5273 pd2.not_cpu_localized = 1;
5275 /* Payload packet is from the opposite direction. */
5276 pd2.sidx = (direction == PF_IN) ? 1 : 0;
5277 pd2.didx = (direction == PF_IN) ? 0 : 1;
5281 /* offset of h2 in mbuf chain */
5282 ipoff2 = off + ICMP_MINLEN;
5284 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
5285 NULL, reason, pd2.af)) {
5286 DPFPRINTF(PF_DEBUG_MISC,
5287 ("pf: ICMP error message too short "
5292 * ICMP error messages don't refer to non-first
5295 if (h2.ip_off & htons(IP_OFFMASK)) {
5296 REASON_SET(reason, PFRES_FRAG);
5300 /* offset of protocol header that follows h2 */
5301 off2 = ipoff2 + (h2.ip_hl << 2);
5303 pd2.proto = h2.ip_p;
5304 pd2.src = (struct pf_addr *)&h2.ip_src;
5305 pd2.dst = (struct pf_addr *)&h2.ip_dst;
5306 pd2.ip_sum = &h2.ip_sum;
5311 ipoff2 = off + sizeof(struct icmp6_hdr);
5313 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
5314 NULL, reason, pd2.af)) {
5315 DPFPRINTF(PF_DEBUG_MISC,
5316 ("pf: ICMP error message too short "
5320 pd2.proto = h2_6.ip6_nxt;
5321 pd2.src = (struct pf_addr *)&h2_6.ip6_src;
5322 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
5324 off2 = ipoff2 + sizeof(h2_6);
5326 switch (pd2.proto) {
5327 case IPPROTO_FRAGMENT:
5329 * ICMPv6 error messages for
5330 * non-first fragments
5332 REASON_SET(reason, PFRES_FRAG);
5335 case IPPROTO_HOPOPTS:
5336 case IPPROTO_ROUTING:
5337 case IPPROTO_DSTOPTS: {
5338 /* get next header and header length */
5339 struct ip6_ext opt6;
5341 if (!pf_pull_hdr(m, off2, &opt6,
5342 sizeof(opt6), NULL, reason,
5344 DPFPRINTF(PF_DEBUG_MISC,
5345 ("pf: ICMPv6 short opt\n"));
5348 if (pd2.proto == IPPROTO_AH)
5349 off2 += (opt6.ip6e_len + 2) * 4;
5351 off2 += (opt6.ip6e_len + 1) * 8;
5352 pd2.proto = opt6.ip6e_nxt;
5353 /* goto the next header */
5360 } while (!terminal);
5364 DPFPRINTF(PF_DEBUG_MISC,
5365 ("pf: ICMP AF %d unknown (ip6)\n", pd->af));
5370 switch (pd2.proto) {
5374 struct pf_state_peer *src, *dst;
5379 * Only the first 8 bytes of the TCP header can be
5380 * expected. Don't access any TCP header fields after
5381 * th_seq, an ackskew test is not possible.
5383 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
5385 DPFPRINTF(PF_DEBUG_MISC,
5386 ("pf: ICMP error message too short "
5392 key.proto = IPPROTO_TCP;
5393 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5394 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5395 key.port[pd2.sidx] = th.th_sport;
5396 key.port[pd2.didx] = th.th_dport;
5398 STATE_LOOKUP(kif, &key, direction, *state, m);
5399 lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5401 if (direction == (*state)->direction) {
5402 src = &(*state)->dst;
5403 dst = &(*state)->src;
5405 src = &(*state)->src;
5406 dst = &(*state)->dst;
5409 if (src->wscale && dst->wscale)
5410 dws = dst->wscale & PF_WSCALE_MASK;
5414 /* Demodulate sequence number */
5415 seq = ntohl(th.th_seq) - src->seqdiff;
5417 pf_change_a(&th.th_seq, icmpsum,
5422 if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
5423 (!SEQ_GEQ(src->seqhi, seq) ||
5424 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
5425 if (pf_status.debug >= PF_DEBUG_MISC) {
5426 kprintf("pf: BAD ICMP %d:%d ",
5427 icmptype, pd->hdr.icmp->icmp_code);
5428 pf_print_host(pd->src, 0, pd->af);
5430 pf_print_host(pd->dst, 0, pd->af);
5431 kprintf(" state: ");
5432 pf_print_state(*state);
5433 kprintf(" seq=%u\n", seq);
5435 REASON_SET(reason, PFRES_BADSTATE);
5438 if (pf_status.debug >= PF_DEBUG_MISC) {
5439 kprintf("pf: OK ICMP %d:%d ",
5440 icmptype, pd->hdr.icmp->icmp_code);
5441 pf_print_host(pd->src, 0, pd->af);
5443 pf_print_host(pd->dst, 0, pd->af);
5444 kprintf(" state: ");
5445 pf_print_state(*state);
5446 kprintf(" seq=%u\n", seq);
5450 /* translate source/destination address, if necessary */
5451 if ((*state)->key[PF_SK_WIRE] !=
5452 (*state)->key[PF_SK_STACK]) {
5453 struct pf_state_key *nk =
5454 (*state)->key[pd->didx];
5456 if (PF_ANEQ(pd2.src,
5457 &nk->addr[pd2.sidx], pd2.af) ||
5458 nk->port[pd2.sidx] != th.th_sport)
5459 pf_change_icmp(pd2.src, &th.th_sport,
5460 daddr, &nk->addr[pd2.sidx],
5461 nk->port[pd2.sidx], NULL,
5462 pd2.ip_sum, icmpsum,
5463 pd->ip_sum, 0, pd2.af);
5465 if (PF_ANEQ(pd2.dst,
5466 &nk->addr[pd2.didx], pd2.af) ||
5467 nk->port[pd2.didx] != th.th_dport)
5468 pf_change_icmp(pd2.dst, &th.th_dport,
5469 NULL, /* XXX Inbound NAT? */
5470 &nk->addr[pd2.didx],
5471 nk->port[pd2.didx], NULL,
5472 pd2.ip_sum, icmpsum,
5473 pd->ip_sum, 0, pd2.af);
5481 m_copyback(m, off, ICMP_MINLEN,
5482 (caddr_t)pd->hdr.icmp);
5483 m_copyback(m, ipoff2, sizeof(h2),
5490 sizeof(struct icmp6_hdr),
5491 (caddr_t)pd->hdr.icmp6);
5492 m_copyback(m, ipoff2, sizeof(h2_6),
5497 m->m_flags &= ~M_HASH;
5498 m_copyback(m, off2, 8, (caddr_t)&th);
5505 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
5506 NULL, reason, pd2.af)) {
5507 DPFPRINTF(PF_DEBUG_MISC,
5508 ("pf: ICMP error message too short "
5514 key.proto = IPPROTO_UDP;
5515 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5516 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5517 key.port[pd2.sidx] = uh.uh_sport;
5518 key.port[pd2.didx] = uh.uh_dport;
5520 STATE_LOOKUP(kif, &key, direction, *state, m);
5521 lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5523 /* translate source/destination address, if necessary */
5524 if ((*state)->key[PF_SK_WIRE] !=
5525 (*state)->key[PF_SK_STACK]) {
5526 struct pf_state_key *nk =
5527 (*state)->key[pd->didx];
5529 if (PF_ANEQ(pd2.src,
5530 &nk->addr[pd2.sidx], pd2.af) ||
5531 nk->port[pd2.sidx] != uh.uh_sport)
5532 pf_change_icmp(pd2.src, &uh.uh_sport,
5533 daddr, &nk->addr[pd2.sidx],
5534 nk->port[pd2.sidx], &uh.uh_sum,
5535 pd2.ip_sum, icmpsum,
5536 pd->ip_sum, 1, pd2.af);
5538 if (PF_ANEQ(pd2.dst,
5539 &nk->addr[pd2.didx], pd2.af) ||
5540 nk->port[pd2.didx] != uh.uh_dport)
5541 pf_change_icmp(pd2.dst, &uh.uh_dport,
5542 NULL, /* XXX Inbound NAT? */
5543 &nk->addr[pd2.didx],
5544 nk->port[pd2.didx], &uh.uh_sum,
5545 pd2.ip_sum, icmpsum,
5546 pd->ip_sum, 1, pd2.af);
5551 m_copyback(m, off, ICMP_MINLEN,
5552 (caddr_t)pd->hdr.icmp);
5553 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5559 sizeof(struct icmp6_hdr),
5560 (caddr_t)pd->hdr.icmp6);
5561 m_copyback(m, ipoff2, sizeof(h2_6),
5566 m->m_flags &= ~M_HASH;
5567 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
5572 case IPPROTO_ICMP: {
5575 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
5576 NULL, reason, pd2.af)) {
5577 DPFPRINTF(PF_DEBUG_MISC,
5578 ("pf: ICMP error message too short i"
5584 key.proto = IPPROTO_ICMP;
5585 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5586 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5587 key.port[0] = key.port[1] = iih.icmp_id;
5589 STATE_LOOKUP(kif, &key, direction, *state, m);
5590 lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5592 /* translate source/destination address, if necessary */
5593 if ((*state)->key[PF_SK_WIRE] !=
5594 (*state)->key[PF_SK_STACK]) {
5595 struct pf_state_key *nk =
5596 (*state)->key[pd->didx];
5598 if (PF_ANEQ(pd2.src,
5599 &nk->addr[pd2.sidx], pd2.af) ||
5600 nk->port[pd2.sidx] != iih.icmp_id)
5601 pf_change_icmp(pd2.src, &iih.icmp_id,
5602 daddr, &nk->addr[pd2.sidx],
5603 nk->port[pd2.sidx], NULL,
5604 pd2.ip_sum, icmpsum,
5605 pd->ip_sum, 0, AF_INET);
5607 if (PF_ANEQ(pd2.dst,
5608 &nk->addr[pd2.didx], pd2.af) ||
5609 nk->port[pd2.didx] != iih.icmp_id)
5610 pf_change_icmp(pd2.dst, &iih.icmp_id,
5611 NULL, /* XXX Inbound NAT? */
5612 &nk->addr[pd2.didx],
5613 nk->port[pd2.didx], NULL,
5614 pd2.ip_sum, icmpsum,
5615 pd->ip_sum, 0, AF_INET);
5617 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
5618 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5619 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
5620 m->m_flags &= ~M_HASH;
5626 case IPPROTO_ICMPV6: {
5627 struct icmp6_hdr iih;
5629 if (!pf_pull_hdr(m, off2, &iih,
5630 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
5631 DPFPRINTF(PF_DEBUG_MISC,
5632 ("pf: ICMP error message too short "
5638 key.proto = IPPROTO_ICMPV6;
5639 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5640 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5641 key.port[0] = key.port[1] = iih.icmp6_id;
5643 STATE_LOOKUP(kif, &key, direction, *state, m);
5644 lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5646 /* translate source/destination address, if necessary */
5647 if ((*state)->key[PF_SK_WIRE] !=
5648 (*state)->key[PF_SK_STACK]) {
5649 struct pf_state_key *nk =
5650 (*state)->key[pd->didx];
5652 if (PF_ANEQ(pd2.src,
5653 &nk->addr[pd2.sidx], pd2.af) ||
5654 nk->port[pd2.sidx] != iih.icmp6_id)
5655 pf_change_icmp(pd2.src, &iih.icmp6_id,
5656 daddr, &nk->addr[pd2.sidx],
5657 nk->port[pd2.sidx], NULL,
5658 pd2.ip_sum, icmpsum,
5659 pd->ip_sum, 0, AF_INET6);
5661 if (PF_ANEQ(pd2.dst,
5662 &nk->addr[pd2.didx], pd2.af) ||
5663 nk->port[pd2.didx] != iih.icmp6_id)
5664 pf_change_icmp(pd2.dst, &iih.icmp6_id,
5665 NULL, /* XXX Inbound NAT? */
5666 &nk->addr[pd2.didx],
5667 nk->port[pd2.didx], NULL,
5668 pd2.ip_sum, icmpsum,
5669 pd->ip_sum, 0, AF_INET6);
5671 m_copyback(m, off, sizeof(struct icmp6_hdr),
5672 (caddr_t)pd->hdr.icmp6);
5673 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
5674 m_copyback(m, off2, sizeof(struct icmp6_hdr),
5676 m->m_flags &= ~M_HASH;
5683 key.proto = pd2.proto;
5684 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5685 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5686 key.port[0] = key.port[1] = 0;
5688 STATE_LOOKUP(kif, &key, direction, *state, m);
5689 lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5691 /* translate source/destination address, if necessary */
5692 if ((*state)->key[PF_SK_WIRE] !=
5693 (*state)->key[PF_SK_STACK]) {
5694 struct pf_state_key *nk =
5695 (*state)->key[pd->didx];
5697 if (PF_ANEQ(pd2.src,
5698 &nk->addr[pd2.sidx], pd2.af))
5699 pf_change_icmp(pd2.src, NULL, daddr,
5700 &nk->addr[pd2.sidx], 0, NULL,
5701 pd2.ip_sum, icmpsum,
5702 pd->ip_sum, 0, pd2.af);
5704 if (PF_ANEQ(pd2.dst,
5705 &nk->addr[pd2.didx], pd2.af))
5706 pf_change_icmp(pd2.src, NULL,
5707 NULL, /* XXX Inbound NAT? */
5708 &nk->addr[pd2.didx], 0, NULL,
5709 pd2.ip_sum, icmpsum,
5710 pd->ip_sum, 0, pd2.af);
5715 m_copyback(m, off, ICMP_MINLEN,
5716 (caddr_t)pd->hdr.icmp);
5717 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5718 m->m_flags &= ~M_HASH;
5724 sizeof(struct icmp6_hdr),
5725 (caddr_t)pd->hdr.icmp6);
5726 m_copyback(m, ipoff2, sizeof(h2_6),
5728 m->m_flags &= ~M_HASH;
5738 pfsync_update_state(*state);
5742 lockmgr(&(*state)->lk, LK_RELEASE);
5747 * Test other connection state. Caller must hold the state locked.
5750 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
5751 struct mbuf *m, struct pf_pdesc *pd)
5753 struct pf_state_peer *src, *dst;
5754 struct pf_state_key_cmp key;
5756 bzero(&key, sizeof(key));
5758 key.proto = pd->proto;
5759 if (direction == PF_IN) {
5760 PF_ACPY(&key.addr[0], pd->src, key.af);
5761 PF_ACPY(&key.addr[1], pd->dst, key.af);
5762 key.port[0] = key.port[1] = 0;
5764 PF_ACPY(&key.addr[1], pd->src, key.af);
5765 PF_ACPY(&key.addr[0], pd->dst, key.af);
5766 key.port[1] = key.port[0] = 0;
5769 STATE_LOOKUP(kif, &key, direction, *state, m);
5770 lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5772 if (direction == (*state)->direction) {
5773 src = &(*state)->src;
5774 dst = &(*state)->dst;
5776 src = &(*state)->dst;
5777 dst = &(*state)->src;
5781 if (src->state < PFOTHERS_SINGLE)
5782 src->state = PFOTHERS_SINGLE;
5783 if (dst->state == PFOTHERS_SINGLE)
5784 dst->state = PFOTHERS_MULTIPLE;
5786 /* update expire time */
5787 (*state)->expire = time_second;
5788 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5789 (*state)->timeout = PFTM_OTHER_MULTIPLE;
5791 (*state)->timeout = PFTM_OTHER_SINGLE;
5793 /* translate source/destination address, if necessary */
5794 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5795 struct pf_state_key *nk = (*state)->key[pd->didx];
5804 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5805 pf_change_a(&pd->src->v4.s_addr,
5807 nk->addr[pd->sidx].v4.s_addr,
5811 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5812 pf_change_a(&pd->dst->v4.s_addr,
5814 nk->addr[pd->didx].v4.s_addr,
5821 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5822 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5824 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5825 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5830 pfsync_update_state(*state);
5831 lockmgr(&(*state)->lk, LK_RELEASE);
5836 * ipoff and off are measured from the start of the mbuf chain.
5837 * h must be at "ipoff" on the mbuf chain.
5840 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5841 u_short *actionp, u_short *reasonp, sa_family_t af)
5846 struct ip *h = mtod(m, struct ip *);
5847 u_int16_t fragoff = (h->ip_off & IP_OFFMASK) << 3;
5851 ACTION_SET(actionp, PF_PASS);
5853 ACTION_SET(actionp, PF_DROP);
5854 REASON_SET(reasonp, PFRES_FRAG);
5858 if (m->m_pkthdr.len < off + len ||
5859 h->ip_len < off + len) {
5860 ACTION_SET(actionp, PF_DROP);
5861 REASON_SET(reasonp, PFRES_SHORT);
5869 struct ip6_hdr *h = mtod(m, struct ip6_hdr *);
5871 if (m->m_pkthdr.len < off + len ||
5872 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5873 (unsigned)(off + len)) {
5874 ACTION_SET(actionp, PF_DROP);
5875 REASON_SET(reasonp, PFRES_SHORT);
5882 m_copydata(m, off, len, p);
5887 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif)
5889 struct sockaddr_in *dst;
5893 struct sockaddr_in6 *dst6;
5894 struct route_in6 ro;
5898 struct radix_node *rn;
5903 bzero(&ro, sizeof(ro));
5906 dst = satosin(&ro.ro_dst);
5907 dst->sin_family = AF_INET;
5908 dst->sin_len = sizeof(*dst);
5909 dst->sin_addr = addr->v4;
5913 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5914 dst6->sin6_family = AF_INET6;
5915 dst6->sin6_len = sizeof(*dst6);
5916 dst6->sin6_addr = addr->v6;
5923 /* Skip checks for ipsec interfaces */
5924 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5927 rtalloc_ign((struct route *)&ro, 0);
5929 if (ro.ro_rt != NULL) {
5930 /* No interface given, this is a no-route check */
5934 if (kif->pfik_ifp == NULL) {
5939 /* Perform uRPF check if passed input interface */
5941 rn = (struct radix_node *)ro.ro_rt;
5943 rt = (struct rtentry *)rn;
5946 if (kif->pfik_ifp == ifp)
5949 } while (check_mpath == 1 && rn != NULL && ret == 0);
5953 if (ro.ro_rt != NULL)
5959 pf_rtlabel_match(struct pf_addr *addr, sa_family_t af, struct pf_addr_wrap *aw)
5961 struct sockaddr_in *dst;
5963 struct sockaddr_in6 *dst6;
5964 struct route_in6 ro;
5970 ASSERT_LWKT_TOKEN_HELD(&pf_token);
5972 bzero(&ro, sizeof(ro));
5975 dst = satosin(&ro.ro_dst);
5976 dst->sin_family = AF_INET;
5977 dst->sin_len = sizeof(*dst);
5978 dst->sin_addr = addr->v4;
5982 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5983 dst6->sin6_family = AF_INET6;
5984 dst6->sin6_len = sizeof(*dst6);
5985 dst6->sin6_addr = addr->v6;
5992 rtalloc_ign((struct route *)&ro, (RTF_CLONING | RTF_PRCLONING));
5994 if (ro.ro_rt != NULL) {
6003 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
6004 struct pf_state *s, struct pf_pdesc *pd)
6006 struct mbuf *m0, *m1;
6007 struct route iproute;
6008 struct route *ro = NULL;
6009 struct sockaddr_in *dst;
6011 struct ifnet *ifp = NULL;
6012 struct pf_addr naddr;
6013 struct pf_src_node *sn = NULL;
6020 ASSERT_LWKT_TOKEN_HELD(&pf_token);
6022 if (m == NULL || *m == NULL || r == NULL ||
6023 (dir != PF_IN && dir != PF_OUT) || oifp == NULL)
6024 panic("pf_route: invalid parameters");
6026 if (((*m)->m_pkthdr.fw_flags & PF_MBUF_ROUTED) == 0) {
6027 (*m)->m_pkthdr.fw_flags |= PF_MBUF_ROUTED;
6028 (*m)->m_pkthdr.pf.routed = 1;
6030 if ((*m)->m_pkthdr.pf.routed++ > 3) {
6037 if (r->rt == PF_DUPTO) {
6038 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
6042 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
6048 if (m0->m_len < sizeof(struct ip)) {
6049 DPFPRINTF(PF_DEBUG_URGENT,
6050 ("pf_route: m0->m_len < sizeof(struct ip)\n"));
6054 ip = mtod(m0, struct ip *);
6057 bzero((caddr_t)ro, sizeof(*ro));
6058 dst = satosin(&ro->ro_dst);
6059 dst->sin_family = AF_INET;
6060 dst->sin_len = sizeof(*dst);
6061 dst->sin_addr = ip->ip_dst;
6063 if (r->rt == PF_FASTROUTE) {
6065 if (ro->ro_rt == 0) {
6066 ipstat.ips_noroute++;
6070 ifp = ro->ro_rt->rt_ifp;
6071 ro->ro_rt->rt_use++;
6073 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
6074 dst = satosin(ro->ro_rt->rt_gateway);
6076 if (TAILQ_EMPTY(&r->rpool.list)) {
6077 DPFPRINTF(PF_DEBUG_URGENT,
6078 ("pf_route: TAILQ_EMPTY(&r->rpool.list)\n"));
6082 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
6084 if (!PF_AZERO(&naddr, AF_INET))
6085 dst->sin_addr.s_addr = naddr.v4.s_addr;
6086 ifp = r->rpool.cur->kif ?
6087 r->rpool.cur->kif->pfik_ifp : NULL;
6089 if (!PF_AZERO(&s->rt_addr, AF_INET))
6090 dst->sin_addr.s_addr =
6091 s->rt_addr.v4.s_addr;
6092 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
6099 if (pf_test(PF_OUT, ifp, &m0, NULL, NULL) != PF_PASS) {
6101 } else if (m0 == NULL) {
6104 if (m0->m_len < sizeof(struct ip)) {
6105 DPFPRINTF(PF_DEBUG_URGENT,
6106 ("pf_route: m0->m_len < sizeof(struct ip)\n"));
6109 ip = mtod(m0, struct ip *);
6112 /* Copied from FreeBSD 5.1-CURRENT ip_output. */
6113 m0->m_pkthdr.csum_flags |= CSUM_IP;
6114 sw_csum = m0->m_pkthdr.csum_flags & ~ifp->if_hwassist;
6115 if (sw_csum & CSUM_DELAY_DATA) {
6116 in_delayed_cksum(m0);
6117 sw_csum &= ~CSUM_DELAY_DATA;
6119 m0->m_pkthdr.csum_flags &= ifp->if_hwassist;
6120 m0->m_pkthdr.csum_iphlen = (ip->ip_hl << 2);
6123 * WARNING! We cannot fragment if the packet was modified from an
6124 * original which expected to be using TSO. In this
6125 * situation we pray that the target interface is
6126 * compatible with the originating interface.
6128 if (ip->ip_len <= ifp->if_mtu ||
6129 (m0->m_pkthdr.csum_flags & CSUM_TSO) ||
6130 ((ifp->if_hwassist & CSUM_FRAGMENT) &&
6131 (ip->ip_off & IP_DF) == 0)) {
6132 ip->ip_len = htons(ip->ip_len);
6133 ip->ip_off = htons(ip->ip_off);
6135 if (sw_csum & CSUM_DELAY_IP) {
6137 if (ip->ip_v == IPVERSION &&
6138 (ip->ip_hl << 2) == sizeof(*ip)) {
6139 ip->ip_sum = in_cksum_hdr(ip);
6141 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
6144 lwkt_reltoken(&pf_token);
6145 error = ifp->if_output(ifp, m0, sintosa(dst), ro->ro_rt);
6146 lwkt_gettoken(&pf_token);
6151 * Too large for interface; fragment if possible.
6152 * Must be able to put at least 8 bytes per fragment.
6154 if (ip->ip_off & IP_DF) {
6155 ipstat.ips_cantfrag++;
6156 if (r->rt != PF_DUPTO) {
6157 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
6165 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist, sw_csum);
6170 for (m0 = m1; m0; m0 = m1) {
6174 lwkt_reltoken(&pf_token);
6175 error = (*ifp->if_output)(ifp, m0, sintosa(dst),
6177 lwkt_gettoken(&pf_token);
6183 ipstat.ips_fragmented++;
6186 if (r->rt != PF_DUPTO)
6188 if (ro == &iproute && ro->ro_rt)
6200 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
6201 struct pf_state *s, struct pf_pdesc *pd)
6204 struct route_in6 ip6route;
6205 struct route_in6 *ro;
6206 struct sockaddr_in6 *dst;
6207 struct ip6_hdr *ip6;
6208 struct ifnet *ifp = NULL;
6209 struct pf_addr naddr;
6210 struct pf_src_node *sn = NULL;
6212 if (m == NULL || *m == NULL || r == NULL ||
6213 (dir != PF_IN && dir != PF_OUT) || oifp == NULL)
6214 panic("pf_route6: invalid parameters");
6216 if (((*m)->m_pkthdr.fw_flags & PF_MBUF_ROUTED) == 0) {
6217 (*m)->m_pkthdr.fw_flags |= PF_MBUF_ROUTED;
6218 (*m)->m_pkthdr.pf.routed = 1;
6220 if ((*m)->m_pkthdr.pf.routed++ > 3) {
6227 if (r->rt == PF_DUPTO) {
6228 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL)
6231 if ((r->rt == PF_REPLYTO) == (r->direction == dir))
6236 if (m0->m_len < sizeof(struct ip6_hdr)) {
6237 DPFPRINTF(PF_DEBUG_URGENT,
6238 ("pf_route6: m0->m_len < sizeof(struct ip6_hdr)\n"));
6241 ip6 = mtod(m0, struct ip6_hdr *);
6244 bzero((caddr_t)ro, sizeof(*ro));
6245 dst = (struct sockaddr_in6 *)&ro->ro_dst;
6246 dst->sin6_family = AF_INET6;
6247 dst->sin6_len = sizeof(*dst);
6248 dst->sin6_addr = ip6->ip6_dst;
6251 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
6252 * so make sure pf.flags is clear.
6254 * Cheat. XXX why only in the v6 case???
6256 if (r->rt == PF_FASTROUTE) {
6257 m0->m_pkthdr.fw_flags |= PF_MBUF_TAGGED;
6258 m0->m_pkthdr.pf.flags = 0;
6259 /* XXX Re-Check when Upgrading to > 4.4 */
6260 m0->m_pkthdr.pf.statekey = NULL;
6261 ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL);
6265 if (TAILQ_EMPTY(&r->rpool.list)) {
6266 DPFPRINTF(PF_DEBUG_URGENT,
6267 ("pf_route6: TAILQ_EMPTY(&r->rpool.list)\n"));
6271 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
6273 if (!PF_AZERO(&naddr, AF_INET6))
6274 PF_ACPY((struct pf_addr *)&dst->sin6_addr,
6276 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
6278 if (!PF_AZERO(&s->rt_addr, AF_INET6))
6279 PF_ACPY((struct pf_addr *)&dst->sin6_addr,
6280 &s->rt_addr, AF_INET6);
6281 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
6287 if (pf_test6(PF_OUT, ifp, &m0, NULL, NULL) != PF_PASS) {
6289 } else if (m0 == NULL) {
6292 if (m0->m_len < sizeof(struct ip6_hdr)) {
6293 DPFPRINTF(PF_DEBUG_URGENT,
6294 ("pf_route6: m0->m_len < sizeof(struct ip6_hdr)\n"));
6297 ip6 = mtod(m0, struct ip6_hdr *);
6301 * If the packet is too large for the outgoing interface,
6302 * send back an icmp6 error.
6304 if (IN6_IS_SCOPE_EMBED(&dst->sin6_addr))
6305 dst->sin6_addr.s6_addr16[1] = htons(ifp->if_index);
6306 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) {
6307 nd6_output(ifp, ifp, m0, dst, NULL);
6309 in6_ifstat_inc(ifp, ifs6_in_toobig);
6310 if (r->rt != PF_DUPTO)
6311 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
6317 if (r->rt != PF_DUPTO)
6329 * check protocol (tcp/udp/icmp/icmp6) checksum and set mbuf flag
6330 * off is the offset where the protocol header starts
6331 * len is the total length of protocol header plus payload
6332 * returns 0 when the checksum is valid, otherwise returns 1.
6336 * FreeBSD supports cksum offload for the following drivers.
6337 * em(4), gx(4), lge(4), nge(4), ti(4), xl(4)
6338 * If we can make full use of it we would outperform ipfw/ipfilter in
6339 * very heavy traffic.
6340 * I have not tested 'cause I don't have NICs that supports cksum offload.
6341 * (There might be problems. Typical phenomena would be
6342 * 1. No route message for UDP packet.
6343 * 2. No connection acceptance from external hosts regardless of rule set.)
6346 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p,
6353 if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
6355 if (m->m_pkthdr.len < off + len)
6361 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
6362 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
6363 sum = m->m_pkthdr.csum_data;
6365 ip = mtod(m, struct ip *);
6366 sum = in_pseudo(ip->ip_src.s_addr,
6367 ip->ip_dst.s_addr, htonl((u_short)len +
6368 m->m_pkthdr.csum_data + p));
6376 case IPPROTO_ICMPV6:
6386 if (p == IPPROTO_ICMP) {
6391 sum = in_cksum(m, len);
6395 if (m->m_len < sizeof(struct ip))
6397 sum = in_cksum_range(m, p, off, len);
6399 m->m_pkthdr.csum_flags |=
6402 m->m_pkthdr.csum_data = 0xffff;
6408 if (m->m_len < sizeof(struct ip6_hdr))
6410 sum = in6_cksum(m, p, off, len);
6413 * IPv6 H/W cksum off-load not supported yet!
6416 * m->m_pkthdr.csum_flags |=
6417 * (CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
6418 * m->m_pkthdr.csum_data = 0xffff;
6430 tcpstat.tcps_rcvbadsum++;
6433 udp_stat.udps_badsum++;
6436 icmpstat.icps_checksum++;
6439 case IPPROTO_ICMPV6:
6440 icmp6stat.icp6s_checksum++;
6450 pf_find_divert(struct mbuf *m)
6454 if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL)
6457 return ((struct pf_divert *)(mtag + 1));
6461 pf_get_divert(struct mbuf *m)
6465 if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL) {
6466 mtag = m_tag_get(PACKET_TAG_PF_DIVERT, sizeof(struct pf_divert),
6470 bzero(mtag + 1, sizeof(struct pf_divert));
6471 m_tag_prepend(m, mtag);
6474 return ((struct pf_divert *)(mtag + 1));
6480 * WARNING: pf_token held shared on entry, THIS IS CPU LOCALIZED CODE
6483 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0,
6484 struct ether_header *eh, struct inpcb *inp)
6486 struct pfi_kif *kif;
6487 u_short action, reason = 0, log = 0;
6488 struct mbuf *m = *m0;
6489 struct ip *h = NULL;
6490 struct pf_rule *a = NULL, *r = &pf_default_rule, *tr, *nr;
6491 struct pf_state *s = NULL;
6492 struct pf_ruleset *ruleset = NULL;
6499 if (m->m_pkthdr.fw_flags & IPFW_MBUF_CONTINUE) {
6500 /* Skip us; continue in ipfw. */
6504 if (!pf_status.running)
6507 memset(&pd, 0, sizeof(pd));
6509 if (ifp->if_type == IFT_CARP && ifp->if_carpdev)
6510 kif = (struct pfi_kif *)ifp->if_carpdev->if_pf_kif;
6513 kif = (struct pfi_kif *)ifp->if_pf_kif;
6516 DPFPRINTF(PF_DEBUG_URGENT,
6517 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
6520 if (kif->pfik_flags & PFI_IFLAG_SKIP)
6524 if ((m->m_flags & M_PKTHDR) == 0)
6525 panic("non-M_PKTHDR is passed to pf_test");
6526 #endif /* DIAGNOSTIC */
6528 if (m->m_pkthdr.len < (int)sizeof(*h)) {
6530 REASON_SET(&reason, PFRES_SHORT);
6536 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
6537 * so make sure pf.flags is clear.
6539 if (m->m_pkthdr.fw_flags & PF_MBUF_TAGGED)
6541 m->m_pkthdr.pf.flags = 0;
6542 /* Re-Check when updating to > 4.4 */
6543 m->m_pkthdr.pf.statekey = NULL;
6545 /* We do IP header normalization and packet reassembly here */
6546 if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
6550 m = *m0; /* pf_normalize messes with m0 */
6551 h = mtod(m, struct ip *);
6553 off = h->ip_hl << 2;
6554 if (off < (int)sizeof(*h)) {
6556 REASON_SET(&reason, PFRES_SHORT);
6561 pd.src = (struct pf_addr *)&h->ip_src;
6562 pd.dst = (struct pf_addr *)&h->ip_dst;
6563 pd.sport = pd.dport = NULL;
6564 pd.ip_sum = &h->ip_sum;
6565 pd.proto_sum = NULL;
6568 pd.sidx = (dir == PF_IN) ? 0 : 1;
6569 pd.didx = (dir == PF_IN) ? 1 : 0;
6572 pd.tot_len = h->ip_len;
6575 /* handle fragments that didn't get reassembled by normalization */
6576 if (h->ip_off & (IP_MF | IP_OFFMASK)) {
6577 action = pf_test_fragment(&r, dir, kif, m, h,
6588 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6589 &action, &reason, AF_INET)) {
6590 log = action != PF_PASS;
6593 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6595 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
6598 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6599 if (action == PF_DROP)
6601 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6603 if (action == PF_PASS) {
6607 } else if (s == NULL) {
6608 action = pf_test_rule(&r, &s, dir, kif,
6610 &ruleset, NULL, inp);
6619 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6620 &action, &reason, AF_INET)) {
6621 log = action != PF_PASS;
6624 if (uh.uh_dport == 0 ||
6625 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6626 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6628 REASON_SET(&reason, PFRES_SHORT);
6631 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6632 if (action == PF_PASS) {
6636 } else if (s == NULL) {
6637 action = pf_test_rule(&r, &s, dir, kif,
6639 &ruleset, NULL, inp);
6644 case IPPROTO_ICMP: {
6648 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
6649 &action, &reason, AF_INET)) {
6650 log = action != PF_PASS;
6653 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
6655 if (action == PF_PASS) {
6659 } else if (s == NULL) {
6660 action = pf_test_rule(&r, &s, dir, kif,
6662 &ruleset, NULL, inp);
6668 action = pf_test_state_other(&s, dir, kif, m, &pd);
6669 if (action == PF_PASS) {
6673 } else if (s == NULL) {
6674 action = pf_test_rule(&r, &s, dir, kif, m, off, h,
6675 &pd, &a, &ruleset, NULL, inp);
6681 if (action == PF_PASS && h->ip_hl > 5 &&
6682 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6684 REASON_SET(&reason, PFRES_IPOPTIONS);
6686 DPFPRINTF(PF_DEBUG_MISC,
6687 ("pf: dropping packet with ip options\n"));
6690 if ((s && s->tag) || r->rtableid)
6691 pf_tag_packet(m, s ? s->tag : 0, r->rtableid);
6694 if (dir == PF_IN && s && s->key[PF_SK_STACK])
6695 m->m_pkthdr.pf.statekey = s->key[PF_SK_STACK];
6700 * Generate a hash code and qid request for ALTQ. A qid of 0
6701 * is allowed and will cause altq to select the default queue.
6703 if (action == PF_PASS) {
6704 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE;
6705 if (pqid || (pd.tos & IPTOS_LOWDELAY))
6706 m->m_pkthdr.pf.qid = r->pqid;
6708 m->m_pkthdr.pf.qid = r->qid;
6709 m->m_pkthdr.pf.ecn_af = AF_INET;
6710 m->m_pkthdr.pf.hdr = h;
6711 /* add connection hash for fairq */
6714 m->m_pkthdr.pf.state_hash = s->hash;
6715 m->m_pkthdr.pf.flags |= PF_TAG_STATE_HASHED;
6721 * connections redirected to loopback should not match sockets
6722 * bound specifically to loopback due to security implications,
6723 * see tcp_input() and in_pcblookup_listen().
6725 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6726 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6727 (s->nat_rule.ptr->action == PF_RDR ||
6728 s->nat_rule.ptr->action == PF_BINAT) &&
6729 (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
6730 m->m_pkthdr.pf.flags |= PF_TAG_TRANSLATE_LOCALHOST;
6732 if (dir == PF_IN && action == PF_PASS && r->divert.port) {
6733 struct pf_divert *divert;
6735 if ((divert = pf_get_divert(m))) {
6736 m->m_pkthdr.pf.flags |= PF_TAG_DIVERTED;
6737 divert->port = r->divert.port;
6738 divert->addr.ipv4 = r->divert.addr.v4;
6745 if (s != NULL && s->nat_rule.ptr != NULL &&
6746 s->nat_rule.ptr->log & PF_LOG_ALL)
6747 lr = s->nat_rule.ptr;
6750 PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, lr, a, ruleset,
6754 kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6755 kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
6757 if (action == PF_PASS || r->action == PF_DROP) {
6758 dirndx = (dir == PF_OUT);
6759 r->packets[dirndx]++;
6760 r->bytes[dirndx] += pd.tot_len;
6762 a->packets[dirndx]++;
6763 a->bytes[dirndx] += pd.tot_len;
6766 if (s->nat_rule.ptr != NULL) {
6767 s->nat_rule.ptr->packets[dirndx]++;
6768 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6770 if (s->src_node != NULL) {
6771 s->src_node->packets[dirndx]++;
6772 s->src_node->bytes[dirndx] += pd.tot_len;
6774 if (s->nat_src_node != NULL) {
6775 s->nat_src_node->packets[dirndx]++;
6776 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6778 dirndx = (dir == s->direction) ? 0 : 1;
6779 s->packets[dirndx]++;
6780 s->bytes[dirndx] += pd.tot_len;
6783 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6784 if (nr != NULL && r == &pf_default_rule)
6786 if (tr->src.addr.type == PF_ADDR_TABLE)
6787 pfr_update_stats(tr->src.addr.p.tbl,
6788 (s == NULL) ? pd.src :
6789 &s->key[(s->direction == PF_IN)]->
6790 addr[(s->direction == PF_OUT)],
6791 pd.af, pd.tot_len, dir == PF_OUT,
6792 r->action == PF_PASS, tr->src.neg);
6793 if (tr->dst.addr.type == PF_ADDR_TABLE)
6794 pfr_update_stats(tr->dst.addr.p.tbl,
6795 (s == NULL) ? pd.dst :
6796 &s->key[(s->direction == PF_IN)]->
6797 addr[(s->direction == PF_IN)],
6798 pd.af, pd.tot_len, dir == PF_OUT,
6799 r->action == PF_PASS, tr->dst.neg);
6803 if (action == PF_SYNPROXY_DROP) {
6808 /* pf_route can free the mbuf causing *m0 to become NULL */
6809 pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
6819 * WARNING: pf_token held shared on entry, THIS IS CPU LOCALIZED CODE
6822 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0,
6823 struct ether_header *eh, struct inpcb *inp)
6825 struct pfi_kif *kif;
6826 u_short action, reason = 0, log = 0;
6827 struct mbuf *m = *m0, *n = NULL;
6828 struct ip6_hdr *h = NULL;
6829 struct pf_rule *a = NULL, *r = &pf_default_rule, *tr, *nr;
6830 struct pf_state *s = NULL;
6831 struct pf_ruleset *ruleset = NULL;
6833 int off, terminal = 0, dirndx, rh_cnt = 0;
6835 if (!pf_status.running)
6838 memset(&pd, 0, sizeof(pd));
6840 if (ifp->if_type == IFT_CARP && ifp->if_carpdev)
6841 kif = (struct pfi_kif *)ifp->if_carpdev->if_pf_kif;
6844 kif = (struct pfi_kif *)ifp->if_pf_kif;
6847 DPFPRINTF(PF_DEBUG_URGENT,
6848 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6851 if (kif->pfik_flags & PFI_IFLAG_SKIP)
6855 if ((m->m_flags & M_PKTHDR) == 0)
6856 panic("non-M_PKTHDR is passed to pf_test6");
6857 #endif /* DIAGNOSTIC */
6859 if (m->m_pkthdr.len < (int)sizeof(*h)) {
6861 REASON_SET(&reason, PFRES_SHORT);
6867 * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
6868 * so make sure pf.flags is clear.
6870 if (m->m_pkthdr.fw_flags & PF_MBUF_TAGGED)
6872 m->m_pkthdr.pf.flags = 0;
6873 /* Re-Check when updating to > 4.4 */
6874 m->m_pkthdr.pf.statekey = NULL;
6876 /* We do IP header normalization and packet reassembly here */
6877 if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6881 m = *m0; /* pf_normalize messes with m0 */
6882 h = mtod(m, struct ip6_hdr *);
6886 * we do not support jumbogram yet. if we keep going, zero ip6_plen
6887 * will do something bad, so drop the packet for now.
6889 if (htons(h->ip6_plen) == 0) {
6891 REASON_SET(&reason, PFRES_NORM); /*XXX*/
6896 pd.src = (struct pf_addr *)&h->ip6_src;
6897 pd.dst = (struct pf_addr *)&h->ip6_dst;
6898 pd.sport = pd.dport = NULL;
6900 pd.proto_sum = NULL;
6902 pd.sidx = (dir == PF_IN) ? 0 : 1;
6903 pd.didx = (dir == PF_IN) ? 1 : 0;
6906 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6909 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6910 pd.proto = h->ip6_nxt;
6913 case IPPROTO_FRAGMENT:
6914 action = pf_test_fragment(&r, dir, kif, m, h,
6916 if (action == PF_DROP)
6917 REASON_SET(&reason, PFRES_FRAG);
6919 case IPPROTO_ROUTING: {
6920 struct ip6_rthdr rthdr;
6923 DPFPRINTF(PF_DEBUG_MISC,
6924 ("pf: IPv6 more than one rthdr\n"));
6926 REASON_SET(&reason, PFRES_IPOPTIONS);
6930 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6932 DPFPRINTF(PF_DEBUG_MISC,
6933 ("pf: IPv6 short rthdr\n"));
6935 REASON_SET(&reason, PFRES_SHORT);
6939 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6940 DPFPRINTF(PF_DEBUG_MISC,
6941 ("pf: IPv6 rthdr0\n"));
6943 REASON_SET(&reason, PFRES_IPOPTIONS);
6950 case IPPROTO_HOPOPTS:
6951 case IPPROTO_DSTOPTS: {
6952 /* get next header and header length */
6953 struct ip6_ext opt6;
6955 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6956 NULL, &reason, pd.af)) {
6957 DPFPRINTF(PF_DEBUG_MISC,
6958 ("pf: IPv6 short opt\n"));
6963 if (pd.proto == IPPROTO_AH)
6964 off += (opt6.ip6e_len + 2) * 4;
6966 off += (opt6.ip6e_len + 1) * 8;
6967 pd.proto = opt6.ip6e_nxt;
6968 /* goto the next header */
6975 } while (!terminal);
6977 /* if there's no routing header, use unmodified mbuf for checksumming */
6987 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6988 &action, &reason, AF_INET6)) {
6989 log = action != PF_PASS;
6992 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6993 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6994 if (action == PF_DROP)
6996 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6998 if (action == PF_PASS) {
7002 } else if (s == NULL) {
7003 action = pf_test_rule(&r, &s, dir, kif,
7005 &ruleset, NULL, inp);
7014 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
7015 &action, &reason, AF_INET6)) {
7016 log = action != PF_PASS;
7019 if (uh.uh_dport == 0 ||
7020 ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
7021 ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
7023 REASON_SET(&reason, PFRES_SHORT);
7026 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
7027 if (action == PF_PASS) {
7031 } else if (s == NULL) {
7032 action = pf_test_rule(&r, &s, dir, kif,
7034 &ruleset, NULL, inp);
7039 case IPPROTO_ICMPV6: {
7040 struct icmp6_hdr ih;
7043 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
7044 &action, &reason, AF_INET6)) {
7045 log = action != PF_PASS;
7048 action = pf_test_state_icmp(&s, dir, kif,
7049 m, off, h, &pd, &reason);
7050 if (action == PF_PASS) {
7054 } else if (s == NULL) {
7055 action = pf_test_rule(&r, &s, dir, kif,
7057 &ruleset, NULL, inp);
7063 action = pf_test_state_other(&s, dir, kif, m, &pd);
7064 if (action == PF_PASS) {
7068 } else if (s == NULL) {
7069 action = pf_test_rule(&r, &s, dir, kif, m, off, h,
7070 &pd, &a, &ruleset, NULL, inp);
7081 /* handle dangerous IPv6 extension headers. */
7082 if (action == PF_PASS && rh_cnt &&
7083 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
7085 REASON_SET(&reason, PFRES_IPOPTIONS);
7087 DPFPRINTF(PF_DEBUG_MISC,
7088 ("pf: dropping packet with dangerous v6 headers\n"));
7091 if ((s && s->tag) || r->rtableid)
7092 pf_tag_packet(m, s ? s->tag : 0, r->rtableid);
7095 if (dir == PF_IN && s && s->key[PF_SK_STACK])
7096 m->m_pkthdr.pf.statekey = s->key[PF_SK_STACK];
7101 * Generate a hash code and qid request for ALTQ. A qid of 0
7102 * is allowed and will cause altq to select the default queue.
7104 if (action == PF_PASS) {
7105 m->m_pkthdr.fw_flags |= PF_MBUF_STRUCTURE;
7106 if (pd.tos & IPTOS_LOWDELAY)
7107 m->m_pkthdr.pf.qid = r->pqid;
7109 m->m_pkthdr.pf.qid = r->qid;
7110 m->m_pkthdr.pf.ecn_af = AF_INET6;
7111 m->m_pkthdr.pf.hdr = h;
7114 m->m_pkthdr.pf.state_hash = s->hash;
7115 m->m_pkthdr.pf.flags |= PF_TAG_STATE_HASHED;
7120 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
7121 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
7122 (s->nat_rule.ptr->action == PF_RDR ||
7123 s->nat_rule.ptr->action == PF_BINAT) &&
7124 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
7125 m->m_pkthdr.pf.flags |= PF_TAG_TRANSLATE_LOCALHOST;
7127 if (dir == PF_IN && action == PF_PASS && r->divert.port) {
7128 struct pf_divert *divert;
7130 if ((divert = pf_get_divert(m))) {
7131 m->m_pkthdr.pf.flags |= PF_TAG_DIVERTED;
7132 divert->port = r->divert.port;
7133 divert->addr.ipv6 = r->divert.addr.v6;
7140 if (s != NULL && s->nat_rule.ptr != NULL &&
7141 s->nat_rule.ptr->log & PF_LOG_ALL)
7142 lr = s->nat_rule.ptr;
7145 PFLOG_PACKET(kif, h, m, AF_INET6, dir, reason, lr, a, ruleset,
7149 kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
7150 kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
7152 if (action == PF_PASS || r->action == PF_DROP) {
7153 dirndx = (dir == PF_OUT);
7154 r->packets[dirndx]++;
7155 r->bytes[dirndx] += pd.tot_len;
7157 a->packets[dirndx]++;
7158 a->bytes[dirndx] += pd.tot_len;
7161 if (s->nat_rule.ptr != NULL) {
7162 s->nat_rule.ptr->packets[dirndx]++;
7163 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
7165 if (s->src_node != NULL) {
7166 s->src_node->packets[dirndx]++;
7167 s->src_node->bytes[dirndx] += pd.tot_len;
7169 if (s->nat_src_node != NULL) {
7170 s->nat_src_node->packets[dirndx]++;
7171 s->nat_src_node->bytes[dirndx] += pd.tot_len;
7173 dirndx = (dir == s->direction) ? 0 : 1;
7174 s->packets[dirndx]++;
7175 s->bytes[dirndx] += pd.tot_len;
7178 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
7179 if (nr != NULL && r == &pf_default_rule)
7181 if (tr->src.addr.type == PF_ADDR_TABLE)
7182 pfr_update_stats(tr->src.addr.p.tbl,
7183 (s == NULL) ? pd.src :
7184 &s->key[(s->direction == PF_IN)]->addr[0],
7185 pd.af, pd.tot_len, dir == PF_OUT,
7186 r->action == PF_PASS, tr->src.neg);
7187 if (tr->dst.addr.type == PF_ADDR_TABLE)
7188 pfr_update_stats(tr->dst.addr.p.tbl,
7189 (s == NULL) ? pd.dst :
7190 &s->key[(s->direction == PF_IN)]->addr[1],
7191 pd.af, pd.tot_len, dir == PF_OUT,
7192 r->action == PF_PASS, tr->dst.neg);
7196 if (action == PF_SYNPROXY_DROP) {
7201 /* pf_route6 can free the mbuf causing *m0 to become NULL */
7202 pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
7209 pf_check_congestion(struct ifqueue *ifq)