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>
45 #include <sys/socket.h>
46 #include <sys/sysctl.h>
49 #include <net/pfvar.h>
50 #include <net/if_pflog.h>
51 #include <net/pf_mtag.h>
53 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
55 static void pf_hash(struct pf_addr *, struct pf_addr *,
56 struct pf_poolhashkey *, sa_family_t);
57 static struct pf_rule *pf_match_translation(struct pf_pdesc *, struct mbuf *,
58 int, int, struct pfi_kif *,
59 struct pf_addr *, u_int16_t, struct pf_addr *,
60 uint16_t, int, struct pf_anchor_stackframe *);
61 static int pf_get_sport(sa_family_t, uint8_t, struct pf_rule *,
62 struct pf_addr *, uint16_t, struct pf_addr *, uint16_t, struct pf_addr *,
63 uint16_t *, uint16_t, uint16_t, struct pf_src_node **);
67 a -= b; a -= c; a ^= (c >> 13); \
68 b -= c; b -= a; b ^= (a << 8); \
69 c -= a; c -= b; c ^= (b >> 13); \
70 a -= b; a -= c; a ^= (c >> 12); \
71 b -= c; b -= a; b ^= (a << 16); \
72 c -= a; c -= b; c ^= (b >> 5); \
73 a -= b; a -= c; a ^= (c >> 3); \
74 b -= c; b -= a; b ^= (a << 10); \
75 c -= a; c -= b; c ^= (b >> 15); \
79 * hash function based on bridge_hash in if_bridge.c
82 pf_hash(struct pf_addr *inaddr, struct pf_addr *hash,
83 struct pf_poolhashkey *key, sa_family_t af)
85 u_int32_t a = 0x9e3779b9, b = 0x9e3779b9, c = key->key32[0];
90 a += inaddr->addr32[0];
93 hash->addr32[0] = c + key->key32[2];
98 a += inaddr->addr32[0];
99 b += inaddr->addr32[2];
102 a += inaddr->addr32[1];
103 b += inaddr->addr32[3];
107 a += inaddr->addr32[2];
108 b += inaddr->addr32[1];
112 a += inaddr->addr32[3];
113 b += inaddr->addr32[0];
122 static struct pf_rule *
123 pf_match_translation(struct pf_pdesc *pd, struct mbuf *m, int off,
124 int direction, struct pfi_kif *kif, struct pf_addr *saddr, u_int16_t sport,
125 struct pf_addr *daddr, uint16_t dport, int rs_num,
126 struct pf_anchor_stackframe *anchor_stack)
128 struct pf_rule *r, *rm = NULL;
129 struct pf_ruleset *ruleset = NULL;
134 r = TAILQ_FIRST(pf_main_ruleset.rules[rs_num].active.ptr);
135 while (r && rm == NULL) {
136 struct pf_rule_addr *src = NULL, *dst = NULL;
137 struct pf_addr_wrap *xdst = NULL;
139 if (r->action == PF_BINAT && direction == PF_IN) {
141 if (r->rpool.cur != NULL)
142 xdst = &r->rpool.cur->addr;
149 if (pfi_kif_match(r->kif, kif) == r->ifnot)
150 r = r->skip[PF_SKIP_IFP].ptr;
151 else if (r->direction && r->direction != direction)
152 r = r->skip[PF_SKIP_DIR].ptr;
153 else if (r->af && r->af != pd->af)
154 r = r->skip[PF_SKIP_AF].ptr;
155 else if (r->proto && r->proto != pd->proto)
156 r = r->skip[PF_SKIP_PROTO].ptr;
157 else if (PF_MISMATCHAW(&src->addr, saddr, pd->af,
158 src->neg, kif, M_GETFIB(m)))
159 r = r->skip[src == &r->src ? PF_SKIP_SRC_ADDR :
160 PF_SKIP_DST_ADDR].ptr;
161 else if (src->port_op && !pf_match_port(src->port_op,
162 src->port[0], src->port[1], sport))
163 r = r->skip[src == &r->src ? PF_SKIP_SRC_PORT :
164 PF_SKIP_DST_PORT].ptr;
165 else if (dst != NULL &&
166 PF_MISMATCHAW(&dst->addr, daddr, pd->af, dst->neg, NULL,
168 r = r->skip[PF_SKIP_DST_ADDR].ptr;
169 else if (xdst != NULL && PF_MISMATCHAW(xdst, daddr, pd->af,
170 0, NULL, M_GETFIB(m)))
171 r = TAILQ_NEXT(r, entries);
172 else if (dst != NULL && dst->port_op &&
173 !pf_match_port(dst->port_op, dst->port[0],
174 dst->port[1], dport))
175 r = r->skip[PF_SKIP_DST_PORT].ptr;
176 else if (r->match_tag && !pf_match_tag(m, r, &tag,
177 pd->pf_mtag ? pd->pf_mtag->tag : 0))
178 r = TAILQ_NEXT(r, entries);
179 else if (r->os_fingerprint != PF_OSFP_ANY && (pd->proto !=
180 IPPROTO_TCP || !pf_osfp_match(pf_osfp_fingerprint(pd, m,
181 off, pd->hdr.tcp), r->os_fingerprint)))
182 r = TAILQ_NEXT(r, entries);
186 if (r->rtableid >= 0)
187 rtableid = r->rtableid;
188 if (r->anchor == NULL) {
191 pf_step_into_anchor(anchor_stack, &asd,
192 &ruleset, rs_num, &r, NULL, NULL);
195 pf_step_out_of_anchor(anchor_stack, &asd, &ruleset,
196 rs_num, &r, NULL, NULL);
199 if (tag > 0 && pf_tag_packet(m, pd, tag))
202 M_SETFIB(m, rtableid);
204 if (rm != NULL && (rm->action == PF_NONAT ||
205 rm->action == PF_NORDR || rm->action == PF_NOBINAT))
211 pf_get_sport(sa_family_t af, u_int8_t proto, struct pf_rule *r,
212 struct pf_addr *saddr, uint16_t sport, struct pf_addr *daddr,
213 uint16_t dport, struct pf_addr *naddr, uint16_t *nport, uint16_t low,
214 uint16_t high, struct pf_src_node **sn)
216 struct pf_state_key_cmp key;
217 struct pf_addr init_addr;
220 bzero(&init_addr, sizeof(init_addr));
221 if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn))
224 if (proto == IPPROTO_ICMP) {
229 bzero(&key, sizeof(key));
233 PF_ACPY(&key.addr[0], daddr, key.af);
236 PF_ACPY(&key.addr[1], naddr, key.af);
239 * port search; start random, step;
240 * similar 2 portloop in in_pcbbind
242 if (!(proto == IPPROTO_TCP || proto == IPPROTO_UDP ||
243 proto == IPPROTO_ICMP) || (low == 0 && high == 0)) {
245 * XXX bug: icmp states don't use the id on both sides.
246 * (traceroute -I through nat)
249 if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
253 } else if (low == high) {
254 key.port[1] = htons(low);
255 if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
268 cut = htonl(arc4random()) % (1 + high - low) + low;
269 /* low <= cut <= high */
270 for (tmp = cut; tmp <= high; ++(tmp)) {
271 key.port[1] = htons(tmp);
272 if (pf_find_state_all(&key, PF_IN, NULL) ==
278 for (tmp = cut - 1; tmp >= low; --(tmp)) {
279 key.port[1] = htons(tmp);
280 if (pf_find_state_all(&key, PF_IN, NULL) ==
288 switch (r->rpool.opts & PF_POOL_TYPEMASK) {
290 case PF_POOL_ROUNDROBIN:
291 if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn))
295 case PF_POOL_SRCHASH:
296 case PF_POOL_BITMASK:
300 } while (! PF_AEQ(&init_addr, naddr, af) );
301 return (1); /* none available */
305 pf_map_addr(sa_family_t af, struct pf_rule *r, struct pf_addr *saddr,
306 struct pf_addr *naddr, struct pf_addr *init_addr, struct pf_src_node **sn)
308 struct pf_pool *rpool = &r->rpool;
309 struct pf_addr *raddr = NULL, *rmask = NULL;
311 if (*sn == NULL && r->rpool.opts & PF_POOL_STICKYADDR &&
312 (r->rpool.opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) {
313 *sn = pf_find_src_node(saddr, r, af, 0);
314 if (*sn != NULL && !PF_AZERO(&(*sn)->raddr, af)) {
315 PF_ACPY(naddr, &(*sn)->raddr, af);
316 if (V_pf_status.debug >= PF_DEBUG_MISC) {
317 printf("pf_map_addr: src tracking maps ");
318 pf_print_host(saddr, 0, af);
320 pf_print_host(naddr, 0, af);
327 if (rpool->cur->addr.type == PF_ADDR_NOROUTE)
329 if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) {
333 if (rpool->cur->addr.p.dyn->pfid_acnt4 < 1 &&
334 (rpool->opts & PF_POOL_TYPEMASK) !=
337 raddr = &rpool->cur->addr.p.dyn->pfid_addr4;
338 rmask = &rpool->cur->addr.p.dyn->pfid_mask4;
343 if (rpool->cur->addr.p.dyn->pfid_acnt6 < 1 &&
344 (rpool->opts & PF_POOL_TYPEMASK) !=
347 raddr = &rpool->cur->addr.p.dyn->pfid_addr6;
348 rmask = &rpool->cur->addr.p.dyn->pfid_mask6;
352 } else if (rpool->cur->addr.type == PF_ADDR_TABLE) {
353 if ((rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_ROUNDROBIN)
354 return (1); /* unsupported */
356 raddr = &rpool->cur->addr.v.a.addr;
357 rmask = &rpool->cur->addr.v.a.mask;
360 switch (rpool->opts & PF_POOL_TYPEMASK) {
362 PF_ACPY(naddr, raddr, af);
364 case PF_POOL_BITMASK:
365 PF_POOLMASK(naddr, raddr, rmask, saddr, af);
368 if (init_addr != NULL && PF_AZERO(init_addr, af)) {
372 rpool->counter.addr32[0] = htonl(arc4random());
377 if (rmask->addr32[3] != 0xffffffff)
378 rpool->counter.addr32[3] =
382 if (rmask->addr32[2] != 0xffffffff)
383 rpool->counter.addr32[2] =
387 if (rmask->addr32[1] != 0xffffffff)
388 rpool->counter.addr32[1] =
392 if (rmask->addr32[0] != 0xffffffff)
393 rpool->counter.addr32[0] =
398 PF_POOLMASK(naddr, raddr, rmask, &rpool->counter, af);
399 PF_ACPY(init_addr, naddr, af);
402 PF_AINC(&rpool->counter, af);
403 PF_POOLMASK(naddr, raddr, rmask, &rpool->counter, af);
406 case PF_POOL_SRCHASH:
408 unsigned char hash[16];
410 pf_hash(saddr, (struct pf_addr *)&hash, &rpool->key, af);
411 PF_POOLMASK(naddr, raddr, rmask, (struct pf_addr *)&hash, af);
414 case PF_POOL_ROUNDROBIN:
416 struct pf_pooladdr *acur = rpool->cur;
419 * XXXGL: in the round-robin case we need to store
420 * the round-robin machine state in the rule, thus
421 * forwarding thread needs to modify rule.
423 * This is done w/o locking, because performance is assumed
424 * more important than round-robin precision.
426 * In the simpliest case we just update the "rpool->cur"
427 * pointer. However, if pool contains tables or dynamic
428 * addresses, then "tblidx" is also used to store machine
429 * state. Since "tblidx" is int, concurrent access to it can't
430 * lead to inconsistence, only to lost of precision.
432 * Things get worse, if table contains not hosts, but
433 * prefixes. In this case counter also stores machine state,
434 * and for IPv6 address, counter can't be updated atomically.
435 * Probably, using round-robin on a table containing IPv6
436 * prefixes (or even IPv4) would cause a panic.
439 if (rpool->cur->addr.type == PF_ADDR_TABLE) {
440 if (!pfr_pool_get(rpool->cur->addr.p.tbl,
441 &rpool->tblidx, &rpool->counter, af))
443 } else if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) {
444 if (!pfr_pool_get(rpool->cur->addr.p.dyn->pfid_kt,
445 &rpool->tblidx, &rpool->counter, af))
447 } else if (pf_match_addr(0, raddr, rmask, &rpool->counter, af))
451 if (TAILQ_NEXT(rpool->cur, entries) == NULL)
452 rpool->cur = TAILQ_FIRST(&rpool->list);
454 rpool->cur = TAILQ_NEXT(rpool->cur, entries);
455 if (rpool->cur->addr.type == PF_ADDR_TABLE) {
457 if (pfr_pool_get(rpool->cur->addr.p.tbl,
458 &rpool->tblidx, &rpool->counter, af)) {
459 /* table contains no address of type 'af' */
460 if (rpool->cur != acur)
464 } else if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) {
466 if (pfr_pool_get(rpool->cur->addr.p.dyn->pfid_kt,
467 &rpool->tblidx, &rpool->counter, af)) {
468 /* table contains no address of type 'af' */
469 if (rpool->cur != acur)
474 raddr = &rpool->cur->addr.v.a.addr;
475 rmask = &rpool->cur->addr.v.a.mask;
476 PF_ACPY(&rpool->counter, raddr, af);
480 PF_ACPY(naddr, &rpool->counter, af);
481 if (init_addr != NULL && PF_AZERO(init_addr, af))
482 PF_ACPY(init_addr, naddr, af);
483 PF_AINC(&rpool->counter, af);
488 PF_ACPY(&(*sn)->raddr, naddr, af);
490 if (V_pf_status.debug >= PF_DEBUG_MISC &&
491 (rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) {
492 printf("pf_map_addr: selected address ");
493 pf_print_host(naddr, 0, af);
501 pf_get_translation(struct pf_pdesc *pd, struct mbuf *m, int off, int direction,
502 struct pfi_kif *kif, struct pf_src_node **sn,
503 struct pf_state_key **skp, struct pf_state_key **nkp,
504 struct pf_addr *saddr, struct pf_addr *daddr,
505 uint16_t sport, uint16_t dport, struct pf_anchor_stackframe *anchor_stack)
507 struct pf_rule *r = NULL;
508 struct pf_addr *naddr;
512 KASSERT(*skp == NULL, ("*skp not NULL"));
513 KASSERT(*nkp == NULL, ("*nkp not NULL"));
515 if (direction == PF_OUT) {
516 r = pf_match_translation(pd, m, off, direction, kif, saddr,
517 sport, daddr, dport, PF_RULESET_BINAT, anchor_stack);
519 r = pf_match_translation(pd, m, off, direction, kif,
520 saddr, sport, daddr, dport, PF_RULESET_NAT,
523 r = pf_match_translation(pd, m, off, direction, kif, saddr,
524 sport, daddr, dport, PF_RULESET_RDR, anchor_stack);
526 r = pf_match_translation(pd, m, off, direction, kif,
527 saddr, sport, daddr, dport, PF_RULESET_BINAT,
541 *skp = pf_state_key_setup(pd, saddr, daddr, sport, dport);
544 *nkp = pf_state_key_clone(*skp);
546 uma_zfree(V_pf_state_key_z, skp);
551 /* XXX We only modify one side for now. */
552 naddr = &(*nkp)->addr[1];
553 nport = &(*nkp)->port[1];
557 if (pf_get_sport(pd->af, pd->proto, r, saddr, sport, daddr,
558 dport, naddr, nport, r->rpool.proxy_port[0],
559 r->rpool.proxy_port[1], sn)) {
560 DPFPRINTF(PF_DEBUG_MISC,
561 ("pf: NAT proxy port allocation (%u-%u) failed\n",
562 r->rpool.proxy_port[0], r->rpool.proxy_port[1]));
569 if (r->rpool.cur->addr.type == PF_ADDR_DYNIFTL){
573 if (r->rpool.cur->addr.p.dyn->
577 &r->rpool.cur->addr.p.dyn->
579 &r->rpool.cur->addr.p.dyn->
580 pfid_mask4, saddr, AF_INET);
585 if (r->rpool.cur->addr.p.dyn->
589 &r->rpool.cur->addr.p.dyn->
591 &r->rpool.cur->addr.p.dyn->
592 pfid_mask6, saddr, AF_INET6);
598 &r->rpool.cur->addr.v.a.addr,
599 &r->rpool.cur->addr.v.a.mask, saddr,
603 if (r->src.addr.type == PF_ADDR_DYNIFTL) {
607 if (r->src.addr.p.dyn-> pfid_acnt4 < 1)
610 &r->src.addr.p.dyn->pfid_addr4,
611 &r->src.addr.p.dyn->pfid_mask4,
617 if (r->src.addr.p.dyn->pfid_acnt6 < 1)
620 &r->src.addr.p.dyn->pfid_addr6,
621 &r->src.addr.p.dyn->pfid_mask6,
627 PF_POOLMASK(naddr, &r->src.addr.v.a.addr,
628 &r->src.addr.v.a.mask, daddr, pd->af);
633 if (pf_map_addr(pd->af, r, saddr, naddr, NULL, sn))
635 if ((r->rpool.opts & PF_POOL_TYPEMASK) == PF_POOL_BITMASK)
636 PF_POOLMASK(naddr, naddr, &r->rpool.cur->addr.v.a.mask,
639 if (r->rpool.proxy_port[1]) {
642 tmp_nport = ((ntohs(dport) - ntohs(r->dst.port[0])) %
643 (r->rpool.proxy_port[1] - r->rpool.proxy_port[0] +
644 1)) + r->rpool.proxy_port[0];
646 /* Wrap around if necessary. */
647 if (tmp_nport > 65535)
649 *nport = htons((uint16_t)tmp_nport);
650 } else if (r->rpool.proxy_port[0])
651 *nport = htons(r->rpool.proxy_port[0]);
655 panic("%s: unknown action %u", __func__, r->action);
658 /* Return success only if translation really happened. */
659 if (bcmp(*skp, *nkp, sizeof(struct pf_state_key_cmp)))
663 uma_zfree(V_pf_state_key_z, *nkp);
664 uma_zfree(V_pf_state_key_z, *skp);