1 /* $OpenBSD: pf_lb.c,v 1.2 2009/02/12 02:13:15 sthen Exp $ */
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.
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
43 #include "opt_inet6.h"
45 #include <sys/param.h>
46 #include <sys/socket.h>
47 #include <sys/sysctl.h>
50 #include <net/pfvar.h>
51 #include <net/if_pflog.h>
52 #include <net/pf_mtag.h>
54 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
56 static void pf_hash(struct pf_addr *, struct pf_addr *,
57 struct pf_poolhashkey *, sa_family_t);
58 static struct pf_rule *pf_match_translation(struct pf_pdesc *, struct mbuf *,
59 int, int, struct pfi_kif *,
60 struct pf_addr *, u_int16_t, struct pf_addr *,
62 static int pf_get_sport(sa_family_t, u_int8_t, struct pf_rule *,
63 struct pf_addr *, struct pf_addr *, u_int16_t,
64 struct pf_addr *, u_int16_t*, u_int16_t, u_int16_t,
65 struct pf_src_node **);
69 a -= b; a -= c; a ^= (c >> 13); \
70 b -= c; b -= a; b ^= (a << 8); \
71 c -= a; c -= b; c ^= (b >> 13); \
72 a -= b; a -= c; a ^= (c >> 12); \
73 b -= c; b -= a; b ^= (a << 16); \
74 c -= a; c -= b; c ^= (b >> 5); \
75 a -= b; a -= c; a ^= (c >> 3); \
76 b -= c; b -= a; b ^= (a << 10); \
77 c -= a; c -= b; c ^= (b >> 15); \
81 * hash function based on bridge_hash in if_bridge.c
84 pf_hash(struct pf_addr *inaddr, struct pf_addr *hash,
85 struct pf_poolhashkey *key, sa_family_t af)
87 u_int32_t a = 0x9e3779b9, b = 0x9e3779b9, c = key->key32[0];
92 a += inaddr->addr32[0];
95 hash->addr32[0] = c + key->key32[2];
100 a += inaddr->addr32[0];
101 b += inaddr->addr32[2];
104 a += inaddr->addr32[1];
105 b += inaddr->addr32[3];
109 a += inaddr->addr32[2];
110 b += inaddr->addr32[1];
114 a += inaddr->addr32[3];
115 b += inaddr->addr32[0];
124 static struct pf_rule *
125 pf_match_translation(struct pf_pdesc *pd, struct mbuf *m, int off,
126 int direction, struct pfi_kif *kif, struct pf_addr *saddr, u_int16_t sport,
127 struct pf_addr *daddr, u_int16_t dport, int rs_num)
129 struct pf_rule *r, *rm = NULL;
130 struct pf_ruleset *ruleset = NULL;
135 r = TAILQ_FIRST(pf_main_ruleset.rules[rs_num].active.ptr);
136 while (r && rm == NULL) {
137 struct pf_rule_addr *src = NULL, *dst = NULL;
138 struct pf_addr_wrap *xdst = NULL;
140 if (r->action == PF_BINAT && direction == PF_IN) {
142 if (r->rpool.cur != NULL)
143 xdst = &r->rpool.cur->addr;
150 if (pfi_kif_match(r->kif, kif) == r->ifnot)
151 r = r->skip[PF_SKIP_IFP].ptr;
152 else if (r->direction && r->direction != direction)
153 r = r->skip[PF_SKIP_DIR].ptr;
154 else if (r->af && r->af != pd->af)
155 r = r->skip[PF_SKIP_AF].ptr;
156 else if (r->proto && r->proto != pd->proto)
157 r = r->skip[PF_SKIP_PROTO].ptr;
158 else if (PF_MISMATCHAW(&src->addr, saddr, pd->af,
159 src->neg, kif, M_GETFIB(m)))
160 r = r->skip[src == &r->src ? PF_SKIP_SRC_ADDR :
161 PF_SKIP_DST_ADDR].ptr;
162 else if (src->port_op && !pf_match_port(src->port_op,
163 src->port[0], src->port[1], sport))
164 r = r->skip[src == &r->src ? PF_SKIP_SRC_PORT :
165 PF_SKIP_DST_PORT].ptr;
166 else if (dst != NULL &&
167 PF_MISMATCHAW(&dst->addr, daddr, pd->af, dst->neg, NULL,
169 r = r->skip[PF_SKIP_DST_ADDR].ptr;
170 else if (xdst != NULL && PF_MISMATCHAW(xdst, daddr, pd->af,
171 0, NULL, M_GETFIB(m)))
172 r = TAILQ_NEXT(r, entries);
173 else if (dst != NULL && dst->port_op &&
174 !pf_match_port(dst->port_op, dst->port[0],
175 dst->port[1], dport))
176 r = r->skip[PF_SKIP_DST_PORT].ptr;
177 else if (r->match_tag && !pf_match_tag(m, r, &tag,
178 pd->pf_mtag ? pd->pf_mtag->tag : 0))
179 r = TAILQ_NEXT(r, entries);
180 else if (r->os_fingerprint != PF_OSFP_ANY && (pd->proto !=
181 IPPROTO_TCP || !pf_osfp_match(pf_osfp_fingerprint(pd, m,
182 off, pd->hdr.tcp), r->os_fingerprint)))
183 r = TAILQ_NEXT(r, entries);
187 if (r->rtableid >= 0)
188 rtableid = r->rtableid;
189 if (r->anchor == NULL) {
192 pf_step_into_anchor(&asd, &ruleset, rs_num,
196 pf_step_out_of_anchor(&asd, &ruleset, rs_num, &r,
200 if (tag > 0 && pf_tag_packet(m, pd, tag))
203 M_SETFIB(m, rtableid);
205 if (rm != NULL && (rm->action == PF_NONAT ||
206 rm->action == PF_NORDR || rm->action == PF_NOBINAT))
212 pf_get_sport(sa_family_t af, u_int8_t proto, struct pf_rule *r,
213 struct pf_addr *saddr, struct pf_addr *daddr, u_int16_t dport,
214 struct pf_addr *naddr, u_int16_t *nport, u_int16_t low, u_int16_t high,
215 struct pf_src_node **sn)
217 struct pf_state_key_cmp key;
218 struct pf_addr init_addr;
221 bzero(&init_addr, sizeof(init_addr));
222 if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn))
225 if (proto == IPPROTO_ICMP) {
233 PF_ACPY(&key.addr[1], daddr, key.af);
234 PF_ACPY(&key.addr[0], naddr, key.af);
238 * port search; start random, step;
239 * similar 2 portloop in in_pcbbind
241 if (!(proto == IPPROTO_TCP || proto == IPPROTO_UDP ||
242 proto == IPPROTO_ICMP)) {
244 if (pf_find_state_all(&key, PF_IN, NULL) == NULL)
246 } else if (low == 0 && high == 0) {
247 key.port[0] = *nport;
248 if (pf_find_state_all(&key, PF_IN, NULL) == NULL)
250 } else if (low == high) {
251 key.port[0] = htons(low);
252 if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
265 cut = htonl(arc4random()) % (1 + high - low) + low;
266 /* low <= cut <= high */
267 for (tmp = cut; tmp <= high; ++(tmp)) {
268 key.port[0] = htons(tmp);
269 if (pf_find_state_all(&key, PF_IN, NULL) ==
275 for (tmp = cut - 1; tmp >= low; --(tmp)) {
276 key.port[0] = htons(tmp);
277 if (pf_find_state_all(&key, PF_IN, NULL) ==
285 switch (r->rpool.opts & PF_POOL_TYPEMASK) {
287 case PF_POOL_ROUNDROBIN:
288 if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn))
292 case PF_POOL_SRCHASH:
293 case PF_POOL_BITMASK:
297 } while (! PF_AEQ(&init_addr, naddr, af) );
298 return (1); /* none available */
302 pf_map_addr(sa_family_t af, struct pf_rule *r, struct pf_addr *saddr,
303 struct pf_addr *naddr, struct pf_addr *init_addr, struct pf_src_node **sn)
305 struct pf_pool *rpool = &r->rpool;
306 struct pf_addr *raddr = NULL, *rmask = NULL;
308 if (*sn == NULL && r->rpool.opts & PF_POOL_STICKYADDR &&
309 (r->rpool.opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) {
310 *sn = pf_find_src_node(saddr, r, af, 0);
311 if (*sn != NULL && !PF_AZERO(&(*sn)->raddr, af)) {
312 PF_ACPY(naddr, &(*sn)->raddr, af);
313 if (V_pf_status.debug >= PF_DEBUG_MISC) {
314 printf("pf_map_addr: src tracking maps ");
315 pf_print_host(saddr, 0, af);
317 pf_print_host(naddr, 0, af);
324 if (rpool->cur->addr.type == PF_ADDR_NOROUTE)
326 if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) {
330 if (rpool->cur->addr.p.dyn->pfid_acnt4 < 1 &&
331 (rpool->opts & PF_POOL_TYPEMASK) !=
334 raddr = &rpool->cur->addr.p.dyn->pfid_addr4;
335 rmask = &rpool->cur->addr.p.dyn->pfid_mask4;
340 if (rpool->cur->addr.p.dyn->pfid_acnt6 < 1 &&
341 (rpool->opts & PF_POOL_TYPEMASK) !=
344 raddr = &rpool->cur->addr.p.dyn->pfid_addr6;
345 rmask = &rpool->cur->addr.p.dyn->pfid_mask6;
349 } else if (rpool->cur->addr.type == PF_ADDR_TABLE) {
350 if ((rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_ROUNDROBIN)
351 return (1); /* unsupported */
353 raddr = &rpool->cur->addr.v.a.addr;
354 rmask = &rpool->cur->addr.v.a.mask;
357 switch (rpool->opts & PF_POOL_TYPEMASK) {
359 PF_ACPY(naddr, raddr, af);
361 case PF_POOL_BITMASK:
362 PF_POOLMASK(naddr, raddr, rmask, saddr, af);
365 if (init_addr != NULL && PF_AZERO(init_addr, af)) {
369 rpool->counter.addr32[0] = htonl(arc4random());
374 if (rmask->addr32[3] != 0xffffffff)
375 rpool->counter.addr32[3] =
379 if (rmask->addr32[2] != 0xffffffff)
380 rpool->counter.addr32[2] =
384 if (rmask->addr32[1] != 0xffffffff)
385 rpool->counter.addr32[1] =
389 if (rmask->addr32[0] != 0xffffffff)
390 rpool->counter.addr32[0] =
395 PF_POOLMASK(naddr, raddr, rmask, &rpool->counter, af);
396 PF_ACPY(init_addr, naddr, af);
399 PF_AINC(&rpool->counter, af);
400 PF_POOLMASK(naddr, raddr, rmask, &rpool->counter, af);
403 case PF_POOL_SRCHASH:
405 unsigned char hash[16];
407 pf_hash(saddr, (struct pf_addr *)&hash, &rpool->key, af);
408 PF_POOLMASK(naddr, raddr, rmask, (struct pf_addr *)&hash, af);
411 case PF_POOL_ROUNDROBIN:
413 struct pf_pooladdr *acur = rpool->cur;
416 * XXXGL: in the round-robin case we need to store
417 * the round-robin machine state in the rule, thus
418 * forwarding thread needs to modify rule.
420 * This is done w/o locking, because performance is assumed
421 * more important than round-robin precision.
423 * In the simpliest case we just update the "rpool->cur"
424 * pointer. However, if pool contains tables or dynamic
425 * addresses, then "tblidx" is also used to store machine
426 * state. Since "tblidx" is int, concurrent access to it can't
427 * lead to inconsistence, only to lost of precision.
429 * Things get worse, if table contains not hosts, but
430 * prefixes. In this case counter also stores machine state,
431 * and for IPv6 address, counter can't be updated atomically.
432 * Probably, using round-robin on a table containing IPv6
433 * prefixes (or even IPv4) would cause a panic.
436 if (rpool->cur->addr.type == PF_ADDR_TABLE) {
437 if (!pfr_pool_get(rpool->cur->addr.p.tbl,
438 &rpool->tblidx, &rpool->counter, af))
440 } else if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) {
441 if (!pfr_pool_get(rpool->cur->addr.p.dyn->pfid_kt,
442 &rpool->tblidx, &rpool->counter, af))
444 } else if (pf_match_addr(0, raddr, rmask, &rpool->counter, af))
448 if (TAILQ_NEXT(rpool->cur, entries) == NULL)
449 rpool->cur = TAILQ_FIRST(&rpool->list);
451 rpool->cur = TAILQ_NEXT(rpool->cur, entries);
452 if (rpool->cur->addr.type == PF_ADDR_TABLE) {
454 if (pfr_pool_get(rpool->cur->addr.p.tbl,
455 &rpool->tblidx, &rpool->counter, af)) {
456 /* table contains no address of type 'af' */
457 if (rpool->cur != acur)
461 } else if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) {
463 if (pfr_pool_get(rpool->cur->addr.p.dyn->pfid_kt,
464 &rpool->tblidx, &rpool->counter, af)) {
465 /* table contains no address of type 'af' */
466 if (rpool->cur != acur)
471 raddr = &rpool->cur->addr.v.a.addr;
472 rmask = &rpool->cur->addr.v.a.mask;
473 PF_ACPY(&rpool->counter, raddr, af);
477 PF_ACPY(naddr, &rpool->counter, af);
478 if (init_addr != NULL && PF_AZERO(init_addr, af))
479 PF_ACPY(init_addr, naddr, af);
480 PF_AINC(&rpool->counter, af);
485 PF_ACPY(&(*sn)->raddr, naddr, af);
487 if (V_pf_status.debug >= PF_DEBUG_MISC &&
488 (rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) {
489 printf("pf_map_addr: selected address ");
490 pf_print_host(naddr, 0, af);
498 pf_get_translation(struct pf_pdesc *pd, struct mbuf *m, int off, int direction,
499 struct pfi_kif *kif, struct pf_src_node **sn,
500 struct pf_state_key **skp, struct pf_state_key **nkp,
501 struct pf_addr *saddr, struct pf_addr *daddr,
502 u_int16_t sport, u_int16_t dport)
504 struct pf_rule *r = NULL;
505 struct pf_addr *naddr;
509 KASSERT(*skp == NULL, ("*skp not NULL"));
510 KASSERT(*nkp == NULL, ("*nkp not NULL"));
512 if (direction == PF_OUT) {
513 r = pf_match_translation(pd, m, off, direction, kif, saddr,
514 sport, daddr, dport, PF_RULESET_BINAT);
516 r = pf_match_translation(pd, m, off, direction, kif,
517 saddr, sport, daddr, dport, PF_RULESET_NAT);
519 r = pf_match_translation(pd, m, off, direction, kif, saddr,
520 sport, daddr, dport, PF_RULESET_RDR);
522 r = pf_match_translation(pd, m, off, direction, kif,
523 saddr, sport, daddr, dport, PF_RULESET_BINAT);
536 *skp = pf_state_key_setup(pd, saddr, daddr, sport, dport);
539 *nkp = pf_state_key_clone(*skp);
541 uma_zfree(V_pf_state_key_z, skp);
546 /* XXX We only modify one side for now. */
547 naddr = &(*nkp)->addr[1];
548 nport = &(*nkp)->port[1];
552 if (pf_get_sport(pd->af, pd->proto, r, saddr, daddr, dport,
553 naddr, nport, r->rpool.proxy_port[0],
554 r->rpool.proxy_port[1], sn)) {
555 DPFPRINTF(PF_DEBUG_MISC,
556 ("pf: NAT proxy port allocation (%u-%u) failed\n",
557 r->rpool.proxy_port[0], r->rpool.proxy_port[1]));
564 if (r->rpool.cur->addr.type == PF_ADDR_DYNIFTL){
568 if (r->rpool.cur->addr.p.dyn->
572 &r->rpool.cur->addr.p.dyn->
574 &r->rpool.cur->addr.p.dyn->
575 pfid_mask4, saddr, AF_INET);
580 if (r->rpool.cur->addr.p.dyn->
584 &r->rpool.cur->addr.p.dyn->
586 &r->rpool.cur->addr.p.dyn->
587 pfid_mask6, saddr, AF_INET6);
593 &r->rpool.cur->addr.v.a.addr,
594 &r->rpool.cur->addr.v.a.mask, saddr,
598 if (r->src.addr.type == PF_ADDR_DYNIFTL) {
602 if (r->src.addr.p.dyn-> pfid_acnt4 < 1)
605 &r->src.addr.p.dyn->pfid_addr4,
606 &r->src.addr.p.dyn->pfid_mask4,
612 if (r->src.addr.p.dyn->pfid_acnt6 < 1)
615 &r->src.addr.p.dyn->pfid_addr6,
616 &r->src.addr.p.dyn->pfid_mask6,
622 PF_POOLMASK(naddr, &r->src.addr.v.a.addr,
623 &r->src.addr.v.a.mask, daddr, pd->af);
628 if (pf_map_addr(pd->af, r, saddr, naddr, NULL, sn))
630 if ((r->rpool.opts & PF_POOL_TYPEMASK) == PF_POOL_BITMASK)
631 PF_POOLMASK(naddr, naddr, &r->rpool.cur->addr.v.a.mask,
634 if (r->rpool.proxy_port[1]) {
637 tmp_nport = ((ntohs(dport) - ntohs(r->dst.port[0])) %
638 (r->rpool.proxy_port[1] - r->rpool.proxy_port[0] +
639 1)) + r->rpool.proxy_port[0];
641 /* Wrap around if necessary. */
642 if (tmp_nport > 65535)
644 *nport = htons((uint16_t)tmp_nport);
645 } else if (r->rpool.proxy_port[0])
646 *nport = htons(r->rpool.proxy_port[0]);
650 panic("%s: unknown action %u", __func__, r->action);
653 /* Return success only if translation really happened. */
654 if (bcmp(*skp, *nkp, sizeof(struct pf_state_key_cmp)))
658 uma_zfree(V_pf_state_key_z, *nkp);
659 uma_zfree(V_pf_state_key_z, *skp);