2 * Copyright (c) 2001 Daniel Hartmeier
3 * Copyright (c) 2002 - 2008 Henning Brauer
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
10 * - Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * - Redistributions in binary form must reproduce the above
13 * copyright notice, this list of conditions and the following
14 * disclaimer in the documentation and/or other materials provided
15 * with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
18 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
19 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
20 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
21 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
23 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
24 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
25 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
27 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28 * POSSIBILITY OF SUCH DAMAGE.
30 * Effort sponsored in part by the Defense Advanced Research Projects
31 * Agency (DARPA) and Air Force Research Laboratory, Air Force
32 * Materiel Command, USAF, under agreement number F30602-01-2-0537.
34 * $OpenBSD: pf_lb.c,v 1.2 2009/02/12 02:13:15 sthen Exp $
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
42 #include "opt_inet6.h"
44 #include <sys/param.h>
47 #include <sys/rwlock.h>
48 #include <sys/socket.h>
49 #include <sys/sysctl.h>
53 #include <net/pfvar.h>
54 #include <net/if_pflog.h>
56 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
58 static void pf_hash(struct pf_addr *, struct pf_addr *,
59 struct pf_poolhashkey *, sa_family_t);
60 static struct pf_rule *pf_match_translation(struct pf_pdesc *, struct mbuf *,
61 int, int, struct pfi_kif *,
62 struct pf_addr *, u_int16_t, struct pf_addr *,
63 uint16_t, int, struct pf_anchor_stackframe *);
64 static int pf_get_sport(sa_family_t, uint8_t, struct pf_rule *,
65 struct pf_addr *, uint16_t, struct pf_addr *, uint16_t, struct pf_addr *,
66 uint16_t *, uint16_t, uint16_t, struct pf_src_node **);
70 a -= b; a -= c; a ^= (c >> 13); \
71 b -= c; b -= a; b ^= (a << 8); \
72 c -= a; c -= b; c ^= (b >> 13); \
73 a -= b; a -= c; a ^= (c >> 12); \
74 b -= c; b -= a; b ^= (a << 16); \
75 c -= a; c -= b; c ^= (b >> 5); \
76 a -= b; a -= c; a ^= (c >> 3); \
77 b -= c; b -= a; b ^= (a << 10); \
78 c -= a; c -= b; c ^= (b >> 15); \
82 * hash function based on bridge_hash in if_bridge.c
85 pf_hash(struct pf_addr *inaddr, struct pf_addr *hash,
86 struct pf_poolhashkey *key, sa_family_t af)
88 u_int32_t a = 0x9e3779b9, b = 0x9e3779b9, c = key->key32[0];
93 a += inaddr->addr32[0];
96 hash->addr32[0] = c + key->key32[2];
101 a += inaddr->addr32[0];
102 b += inaddr->addr32[2];
105 a += inaddr->addr32[1];
106 b += inaddr->addr32[3];
110 a += inaddr->addr32[2];
111 b += inaddr->addr32[1];
115 a += inaddr->addr32[3];
116 b += inaddr->addr32[0];
125 static struct pf_rule *
126 pf_match_translation(struct pf_pdesc *pd, struct mbuf *m, int off,
127 int direction, struct pfi_kif *kif, struct pf_addr *saddr, u_int16_t sport,
128 struct pf_addr *daddr, uint16_t dport, int rs_num,
129 struct pf_anchor_stackframe *anchor_stack)
131 struct pf_rule *r, *rm = NULL;
132 struct pf_ruleset *ruleset = NULL;
137 r = TAILQ_FIRST(pf_main_ruleset.rules[rs_num].active.ptr);
138 while (r && rm == NULL) {
139 struct pf_rule_addr *src = NULL, *dst = NULL;
140 struct pf_addr_wrap *xdst = NULL;
142 if (r->action == PF_BINAT && direction == PF_IN) {
144 if (r->rpool.cur != NULL)
145 xdst = &r->rpool.cur->addr;
152 if (pfi_kif_match(r->kif, kif) == r->ifnot)
153 r = r->skip[PF_SKIP_IFP].ptr;
154 else if (r->direction && r->direction != direction)
155 r = r->skip[PF_SKIP_DIR].ptr;
156 else if (r->af && r->af != pd->af)
157 r = r->skip[PF_SKIP_AF].ptr;
158 else if (r->proto && r->proto != pd->proto)
159 r = r->skip[PF_SKIP_PROTO].ptr;
160 else if (PF_MISMATCHAW(&src->addr, saddr, pd->af,
161 src->neg, kif, M_GETFIB(m)))
162 r = r->skip[src == &r->src ? PF_SKIP_SRC_ADDR :
163 PF_SKIP_DST_ADDR].ptr;
164 else if (src->port_op && !pf_match_port(src->port_op,
165 src->port[0], src->port[1], sport))
166 r = r->skip[src == &r->src ? PF_SKIP_SRC_PORT :
167 PF_SKIP_DST_PORT].ptr;
168 else if (dst != NULL &&
169 PF_MISMATCHAW(&dst->addr, daddr, pd->af, dst->neg, NULL,
171 r = r->skip[PF_SKIP_DST_ADDR].ptr;
172 else if (xdst != NULL && PF_MISMATCHAW(xdst, daddr, pd->af,
173 0, NULL, M_GETFIB(m)))
174 r = TAILQ_NEXT(r, entries);
175 else if (dst != NULL && dst->port_op &&
176 !pf_match_port(dst->port_op, dst->port[0],
177 dst->port[1], dport))
178 r = r->skip[PF_SKIP_DST_PORT].ptr;
179 else if (r->match_tag && !pf_match_tag(m, r, &tag,
180 pd->pf_mtag ? pd->pf_mtag->tag : 0))
181 r = TAILQ_NEXT(r, entries);
182 else if (r->os_fingerprint != PF_OSFP_ANY && (pd->proto !=
183 IPPROTO_TCP || !pf_osfp_match(pf_osfp_fingerprint(pd, m,
184 off, pd->hdr.tcp), r->os_fingerprint)))
185 r = TAILQ_NEXT(r, entries);
189 if (r->rtableid >= 0)
190 rtableid = r->rtableid;
191 if (r->anchor == NULL) {
194 pf_step_into_anchor(anchor_stack, &asd,
195 &ruleset, rs_num, &r, NULL, NULL);
198 pf_step_out_of_anchor(anchor_stack, &asd, &ruleset,
199 rs_num, &r, NULL, NULL);
202 if (tag > 0 && pf_tag_packet(m, pd, tag))
205 M_SETFIB(m, rtableid);
207 if (rm != NULL && (rm->action == PF_NONAT ||
208 rm->action == PF_NORDR || rm->action == PF_NOBINAT))
214 pf_get_sport(sa_family_t af, u_int8_t proto, struct pf_rule *r,
215 struct pf_addr *saddr, uint16_t sport, struct pf_addr *daddr,
216 uint16_t dport, struct pf_addr *naddr, uint16_t *nport, uint16_t low,
217 uint16_t high, struct pf_src_node **sn)
219 struct pf_state_key_cmp key;
220 struct pf_addr init_addr;
222 bzero(&init_addr, sizeof(init_addr));
223 if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn))
226 if (proto == IPPROTO_ICMP) {
231 bzero(&key, sizeof(key));
235 PF_ACPY(&key.addr[0], daddr, key.af);
238 PF_ACPY(&key.addr[1], naddr, key.af);
241 * port search; start random, step;
242 * similar 2 portloop in in_pcbbind
244 if (!(proto == IPPROTO_TCP || proto == IPPROTO_UDP ||
245 proto == IPPROTO_ICMP) || (low == 0 && high == 0)) {
247 * XXX bug: icmp states don't use the id on both sides.
248 * (traceroute -I through nat)
251 if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
255 } else if (low == high) {
256 key.port[1] = htons(low);
257 if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
270 cut = arc4random() % (1 + high - low) + low;
271 /* low <= cut <= high */
272 for (tmp = cut; tmp <= high; ++(tmp)) {
273 key.port[1] = htons(tmp);
274 if (pf_find_state_all(&key, PF_IN, NULL) ==
280 for (tmp = cut - 1; tmp >= low; --(tmp)) {
281 key.port[1] = htons(tmp);
282 if (pf_find_state_all(&key, PF_IN, NULL) ==
290 switch (r->rpool.opts & PF_POOL_TYPEMASK) {
292 case PF_POOL_ROUNDROBIN:
293 if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn))
297 case PF_POOL_SRCHASH:
298 case PF_POOL_BITMASK:
302 } while (! PF_AEQ(&init_addr, naddr, af) );
303 return (1); /* none available */
307 pf_map_addr(sa_family_t af, struct pf_rule *r, struct pf_addr *saddr,
308 struct pf_addr *naddr, struct pf_addr *init_addr, struct pf_src_node **sn)
310 struct pf_pool *rpool = &r->rpool;
311 struct pf_addr *raddr = NULL, *rmask = NULL;
313 /* Try to find a src_node if none was given and this
314 is a sticky-address rule. */
315 if (*sn == NULL && r->rpool.opts & PF_POOL_STICKYADDR &&
316 (r->rpool.opts & PF_POOL_TYPEMASK) != PF_POOL_NONE)
317 *sn = pf_find_src_node(saddr, r, af, 0);
319 /* If a src_node was found or explicitly given and it has a non-zero
320 route address, use this address. A zeroed address is found if the
321 src node was created just a moment ago in pf_create_state and it
322 needs to be filled in with routing decision calculated here. */
323 if (*sn != NULL && !PF_AZERO(&(*sn)->raddr, af)) {
324 PF_ACPY(naddr, &(*sn)->raddr, af);
325 if (V_pf_status.debug >= PF_DEBUG_MISC) {
326 printf("pf_map_addr: src tracking maps ");
327 pf_print_host(saddr, 0, af);
329 pf_print_host(naddr, 0, af);
335 /* Find the route using chosen algorithm. Store the found route
336 in src_node if it was given or found. */
337 if (rpool->cur->addr.type == PF_ADDR_NOROUTE)
339 if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) {
343 if (rpool->cur->addr.p.dyn->pfid_acnt4 < 1 &&
344 (rpool->opts & PF_POOL_TYPEMASK) !=
347 raddr = &rpool->cur->addr.p.dyn->pfid_addr4;
348 rmask = &rpool->cur->addr.p.dyn->pfid_mask4;
353 if (rpool->cur->addr.p.dyn->pfid_acnt6 < 1 &&
354 (rpool->opts & PF_POOL_TYPEMASK) !=
357 raddr = &rpool->cur->addr.p.dyn->pfid_addr6;
358 rmask = &rpool->cur->addr.p.dyn->pfid_mask6;
362 } else if (rpool->cur->addr.type == PF_ADDR_TABLE) {
363 if ((rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_ROUNDROBIN)
364 return (1); /* unsupported */
366 raddr = &rpool->cur->addr.v.a.addr;
367 rmask = &rpool->cur->addr.v.a.mask;
370 switch (rpool->opts & PF_POOL_TYPEMASK) {
372 PF_ACPY(naddr, raddr, af);
374 case PF_POOL_BITMASK:
375 PF_POOLMASK(naddr, raddr, rmask, saddr, af);
378 if (init_addr != NULL && PF_AZERO(init_addr, af)) {
382 rpool->counter.addr32[0] = htonl(arc4random());
387 if (rmask->addr32[3] != 0xffffffff)
388 rpool->counter.addr32[3] =
392 if (rmask->addr32[2] != 0xffffffff)
393 rpool->counter.addr32[2] =
397 if (rmask->addr32[1] != 0xffffffff)
398 rpool->counter.addr32[1] =
402 if (rmask->addr32[0] != 0xffffffff)
403 rpool->counter.addr32[0] =
408 PF_POOLMASK(naddr, raddr, rmask, &rpool->counter, af);
409 PF_ACPY(init_addr, naddr, af);
412 PF_AINC(&rpool->counter, af);
413 PF_POOLMASK(naddr, raddr, rmask, &rpool->counter, af);
416 case PF_POOL_SRCHASH:
418 unsigned char hash[16];
420 pf_hash(saddr, (struct pf_addr *)&hash, &rpool->key, af);
421 PF_POOLMASK(naddr, raddr, rmask, (struct pf_addr *)&hash, af);
424 case PF_POOL_ROUNDROBIN:
426 struct pf_pooladdr *acur = rpool->cur;
429 * XXXGL: in the round-robin case we need to store
430 * the round-robin machine state in the rule, thus
431 * forwarding thread needs to modify rule.
433 * This is done w/o locking, because performance is assumed
434 * more important than round-robin precision.
436 * In the simpliest case we just update the "rpool->cur"
437 * pointer. However, if pool contains tables or dynamic
438 * addresses, then "tblidx" is also used to store machine
439 * state. Since "tblidx" is int, concurrent access to it can't
440 * lead to inconsistence, only to lost of precision.
442 * Things get worse, if table contains not hosts, but
443 * prefixes. In this case counter also stores machine state,
444 * and for IPv6 address, counter can't be updated atomically.
445 * Probably, using round-robin on a table containing IPv6
446 * prefixes (or even IPv4) would cause a panic.
449 if (rpool->cur->addr.type == PF_ADDR_TABLE) {
450 if (!pfr_pool_get(rpool->cur->addr.p.tbl,
451 &rpool->tblidx, &rpool->counter, af))
453 } else if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) {
454 if (!pfr_pool_get(rpool->cur->addr.p.dyn->pfid_kt,
455 &rpool->tblidx, &rpool->counter, af))
457 } else if (pf_match_addr(0, raddr, rmask, &rpool->counter, af))
461 if (TAILQ_NEXT(rpool->cur, entries) == NULL)
462 rpool->cur = TAILQ_FIRST(&rpool->list);
464 rpool->cur = TAILQ_NEXT(rpool->cur, entries);
465 if (rpool->cur->addr.type == PF_ADDR_TABLE) {
467 if (pfr_pool_get(rpool->cur->addr.p.tbl,
468 &rpool->tblidx, &rpool->counter, af)) {
469 /* table contains no address of type 'af' */
470 if (rpool->cur != acur)
474 } else if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) {
476 if (pfr_pool_get(rpool->cur->addr.p.dyn->pfid_kt,
477 &rpool->tblidx, &rpool->counter, af)) {
478 /* table contains no address of type 'af' */
479 if (rpool->cur != acur)
484 raddr = &rpool->cur->addr.v.a.addr;
485 rmask = &rpool->cur->addr.v.a.mask;
486 PF_ACPY(&rpool->counter, raddr, af);
490 PF_ACPY(naddr, &rpool->counter, af);
491 if (init_addr != NULL && PF_AZERO(init_addr, af))
492 PF_ACPY(init_addr, naddr, af);
493 PF_AINC(&rpool->counter, af);
498 PF_ACPY(&(*sn)->raddr, naddr, af);
500 if (V_pf_status.debug >= PF_DEBUG_MISC &&
501 (rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) {
502 printf("pf_map_addr: selected address ");
503 pf_print_host(naddr, 0, af);
511 pf_get_translation(struct pf_pdesc *pd, struct mbuf *m, int off, int direction,
512 struct pfi_kif *kif, struct pf_src_node **sn,
513 struct pf_state_key **skp, struct pf_state_key **nkp,
514 struct pf_addr *saddr, struct pf_addr *daddr,
515 uint16_t sport, uint16_t dport, struct pf_anchor_stackframe *anchor_stack)
517 struct pf_rule *r = NULL;
518 struct pf_addr *naddr;
522 KASSERT(*skp == NULL, ("*skp not NULL"));
523 KASSERT(*nkp == NULL, ("*nkp not NULL"));
525 if (direction == PF_OUT) {
526 r = pf_match_translation(pd, m, off, direction, kif, saddr,
527 sport, daddr, dport, PF_RULESET_BINAT, anchor_stack);
529 r = pf_match_translation(pd, m, off, direction, kif,
530 saddr, sport, daddr, dport, PF_RULESET_NAT,
533 r = pf_match_translation(pd, m, off, direction, kif, saddr,
534 sport, daddr, dport, PF_RULESET_RDR, anchor_stack);
536 r = pf_match_translation(pd, m, off, direction, kif,
537 saddr, sport, daddr, dport, PF_RULESET_BINAT,
551 *skp = pf_state_key_setup(pd, saddr, daddr, sport, dport);
554 *nkp = pf_state_key_clone(*skp);
556 uma_zfree(V_pf_state_key_z, skp);
561 /* XXX We only modify one side for now. */
562 naddr = &(*nkp)->addr[1];
563 nport = &(*nkp)->port[1];
567 if (pf_get_sport(pd->af, pd->proto, r, saddr, sport, daddr,
568 dport, naddr, nport, r->rpool.proxy_port[0],
569 r->rpool.proxy_port[1], sn)) {
570 DPFPRINTF(PF_DEBUG_MISC,
571 ("pf: NAT proxy port allocation (%u-%u) failed\n",
572 r->rpool.proxy_port[0], r->rpool.proxy_port[1]));
579 if (r->rpool.cur->addr.type == PF_ADDR_DYNIFTL){
583 if (r->rpool.cur->addr.p.dyn->
587 &r->rpool.cur->addr.p.dyn->
589 &r->rpool.cur->addr.p.dyn->
590 pfid_mask4, saddr, AF_INET);
595 if (r->rpool.cur->addr.p.dyn->
599 &r->rpool.cur->addr.p.dyn->
601 &r->rpool.cur->addr.p.dyn->
602 pfid_mask6, saddr, AF_INET6);
608 &r->rpool.cur->addr.v.a.addr,
609 &r->rpool.cur->addr.v.a.mask, saddr,
613 if (r->src.addr.type == PF_ADDR_DYNIFTL) {
617 if (r->src.addr.p.dyn-> pfid_acnt4 < 1)
620 &r->src.addr.p.dyn->pfid_addr4,
621 &r->src.addr.p.dyn->pfid_mask4,
627 if (r->src.addr.p.dyn->pfid_acnt6 < 1)
630 &r->src.addr.p.dyn->pfid_addr6,
631 &r->src.addr.p.dyn->pfid_mask6,
637 PF_POOLMASK(naddr, &r->src.addr.v.a.addr,
638 &r->src.addr.v.a.mask, daddr, pd->af);
643 if (pf_map_addr(pd->af, r, saddr, naddr, NULL, sn))
645 if ((r->rpool.opts & PF_POOL_TYPEMASK) == PF_POOL_BITMASK)
646 PF_POOLMASK(naddr, naddr, &r->rpool.cur->addr.v.a.mask,
649 if (r->rpool.proxy_port[1]) {
652 tmp_nport = ((ntohs(dport) - ntohs(r->dst.port[0])) %
653 (r->rpool.proxy_port[1] - r->rpool.proxy_port[0] +
654 1)) + r->rpool.proxy_port[0];
656 /* Wrap around if necessary. */
657 if (tmp_nport > 65535)
659 *nport = htons((uint16_t)tmp_nport);
660 } else if (r->rpool.proxy_port[0])
661 *nport = htons(r->rpool.proxy_port[0]);
665 panic("%s: unknown action %u", __func__, r->action);
668 /* Return success only if translation really happened. */
669 if (bcmp(*skp, *nkp, sizeof(struct pf_state_key_cmp)))
673 uma_zfree(V_pf_state_key_z, *nkp);
674 uma_zfree(V_pf_state_key_z, *skp);