]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/netpfil/pf/pf_lb.c
zfs: merge openzfs/zfs@9a865b7fb (zfs-2.1-release) into stable/13
[FreeBSD/FreeBSD.git] / sys / netpfil / pf / pf_lb.c
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2001 Daniel Hartmeier
5  * Copyright (c) 2002 - 2008 Henning Brauer
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  *
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.
18  *
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.
31  *
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.
35  *
36  *      $OpenBSD: pf_lb.c,v 1.2 2009/02/12 02:13:15 sthen Exp $
37  */
38
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41
42 #include "opt_pf.h"
43 #include "opt_inet.h"
44 #include "opt_inet6.h"
45
46 #include <sys/param.h>
47 #include <sys/lock.h>
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 #include <sys/sysctl.h>
51
52 #include <net/if.h>
53 #include <net/vnet.h>
54 #include <net/pfvar.h>
55 #include <net/if_pflog.h>
56
57 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
58
59 static void              pf_hash(struct pf_addr *, struct pf_addr *,
60                             struct pf_poolhashkey *, sa_family_t);
61 static struct pf_krule  *pf_match_translation(struct pf_pdesc *, struct mbuf *,
62                             int, int, struct pfi_kkif *,
63                             struct pf_addr *, u_int16_t, struct pf_addr *,
64                             uint16_t, int, struct pf_kanchor_stackframe *);
65 static int pf_get_sport(sa_family_t, uint8_t, struct pf_krule *,
66     struct pf_addr *, uint16_t, struct pf_addr *, uint16_t, struct pf_addr *,
67     uint16_t *, uint16_t, uint16_t, struct pf_ksrc_node **);
68
69 #define mix(a,b,c) \
70         do {                                    \
71                 a -= b; a -= c; a ^= (c >> 13); \
72                 b -= c; b -= a; b ^= (a << 8);  \
73                 c -= a; c -= b; c ^= (b >> 13); \
74                 a -= b; a -= c; a ^= (c >> 12); \
75                 b -= c; b -= a; b ^= (a << 16); \
76                 c -= a; c -= b; c ^= (b >> 5);  \
77                 a -= b; a -= c; a ^= (c >> 3);  \
78                 b -= c; b -= a; b ^= (a << 10); \
79                 c -= a; c -= b; c ^= (b >> 15); \
80         } while (0)
81
82 /*
83  * hash function based on bridge_hash in if_bridge.c
84  */
85 static void
86 pf_hash(struct pf_addr *inaddr, struct pf_addr *hash,
87     struct pf_poolhashkey *key, sa_family_t af)
88 {
89         u_int32_t       a = 0x9e3779b9, b = 0x9e3779b9, c = key->key32[0];
90
91         switch (af) {
92 #ifdef INET
93         case AF_INET:
94                 a += inaddr->addr32[0];
95                 b += key->key32[1];
96                 mix(a, b, c);
97                 hash->addr32[0] = c + key->key32[2];
98                 break;
99 #endif /* INET */
100 #ifdef INET6
101         case AF_INET6:
102                 a += inaddr->addr32[0];
103                 b += inaddr->addr32[2];
104                 mix(a, b, c);
105                 hash->addr32[0] = c;
106                 a += inaddr->addr32[1];
107                 b += inaddr->addr32[3];
108                 c += key->key32[1];
109                 mix(a, b, c);
110                 hash->addr32[1] = c;
111                 a += inaddr->addr32[2];
112                 b += inaddr->addr32[1];
113                 c += key->key32[2];
114                 mix(a, b, c);
115                 hash->addr32[2] = c;
116                 a += inaddr->addr32[3];
117                 b += inaddr->addr32[0];
118                 c += key->key32[3];
119                 mix(a, b, c);
120                 hash->addr32[3] = c;
121                 break;
122 #endif /* INET6 */
123         }
124 }
125
126 static struct pf_krule *
127 pf_match_translation(struct pf_pdesc *pd, struct mbuf *m, int off,
128     int direction, struct pfi_kkif *kif, struct pf_addr *saddr, u_int16_t sport,
129     struct pf_addr *daddr, uint16_t dport, int rs_num,
130     struct pf_kanchor_stackframe *anchor_stack)
131 {
132         struct pf_krule         *r, *rm = NULL;
133         struct pf_kruleset      *ruleset = NULL;
134         int                      tag = -1;
135         int                      rtableid = -1;
136         int                      asd = 0;
137
138         r = TAILQ_FIRST(pf_main_ruleset.rules[rs_num].active.ptr);
139         while (r && rm == NULL) {
140                 struct pf_rule_addr     *src = NULL, *dst = NULL;
141                 struct pf_addr_wrap     *xdst = NULL;
142
143                 if (r->action == PF_BINAT && direction == PF_IN) {
144                         src = &r->dst;
145                         if (r->rpool.cur != NULL)
146                                 xdst = &r->rpool.cur->addr;
147                 } else {
148                         src = &r->src;
149                         dst = &r->dst;
150                 }
151
152                 counter_u64_add(r->evaluations, 1);
153                 if (pfi_kkif_match(r->kif, kif) == r->ifnot)
154                         r = r->skip[PF_SKIP_IFP].ptr;
155                 else if (r->direction && r->direction != direction)
156                         r = r->skip[PF_SKIP_DIR].ptr;
157                 else if (r->af && r->af != pd->af)
158                         r = r->skip[PF_SKIP_AF].ptr;
159                 else if (r->proto && r->proto != pd->proto)
160                         r = r->skip[PF_SKIP_PROTO].ptr;
161                 else if (PF_MISMATCHAW(&src->addr, saddr, pd->af,
162                     src->neg, kif, M_GETFIB(m)))
163                         r = r->skip[src == &r->src ? PF_SKIP_SRC_ADDR :
164                             PF_SKIP_DST_ADDR].ptr;
165                 else if (src->port_op && !pf_match_port(src->port_op,
166                     src->port[0], src->port[1], sport))
167                         r = r->skip[src == &r->src ? PF_SKIP_SRC_PORT :
168                             PF_SKIP_DST_PORT].ptr;
169                 else if (dst != NULL &&
170                     PF_MISMATCHAW(&dst->addr, daddr, pd->af, dst->neg, NULL,
171                     M_GETFIB(m)))
172                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
173                 else if (xdst != NULL && PF_MISMATCHAW(xdst, daddr, pd->af,
174                     0, NULL, M_GETFIB(m)))
175                         r = TAILQ_NEXT(r, entries);
176                 else if (dst != NULL && dst->port_op &&
177                     !pf_match_port(dst->port_op, dst->port[0],
178                     dst->port[1], dport))
179                         r = r->skip[PF_SKIP_DST_PORT].ptr;
180                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
181                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
182                         r = TAILQ_NEXT(r, entries);
183                 else if (r->os_fingerprint != PF_OSFP_ANY && (pd->proto !=
184                     IPPROTO_TCP || !pf_osfp_match(pf_osfp_fingerprint(pd, m,
185                     off, pd->hdr.tcp), r->os_fingerprint)))
186                         r = TAILQ_NEXT(r, entries);
187                 else {
188                         if (r->tag)
189                                 tag = r->tag;
190                         if (r->rtableid >= 0)
191                                 rtableid = r->rtableid;
192                         if (r->anchor == NULL) {
193                                 rm = r;
194                         } else
195                                 pf_step_into_anchor(anchor_stack, &asd,
196                                     &ruleset, rs_num, &r, NULL, NULL);
197                 }
198                 if (r == NULL)
199                         pf_step_out_of_anchor(anchor_stack, &asd, &ruleset,
200                             rs_num, &r, NULL, NULL);
201         }
202
203         if (tag > 0 && pf_tag_packet(m, pd, tag))
204                 return (NULL);
205         if (rtableid >= 0)
206                 M_SETFIB(m, rtableid);
207
208         if (rm != NULL && (rm->action == PF_NONAT ||
209             rm->action == PF_NORDR || rm->action == PF_NOBINAT))
210                 return (NULL);
211         return (rm);
212 }
213
214 static int
215 pf_get_sport(sa_family_t af, u_int8_t proto, struct pf_krule *r,
216     struct pf_addr *saddr, uint16_t sport, struct pf_addr *daddr,
217     uint16_t dport, struct pf_addr *naddr, uint16_t *nport, uint16_t low,
218     uint16_t high, struct pf_ksrc_node **sn)
219 {
220         struct pf_state_key_cmp key;
221         struct pf_addr          init_addr;
222
223         bzero(&init_addr, sizeof(init_addr));
224         if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn))
225                 return (1);
226
227         bzero(&key, sizeof(key));
228         key.af = af;
229         key.proto = proto;
230         key.port[0] = dport;
231         PF_ACPY(&key.addr[0], daddr, key.af);
232
233         do {
234                 PF_ACPY(&key.addr[1], naddr, key.af);
235
236                 /*
237                  * port search; start random, step;
238                  * similar 2 portloop in in_pcbbind
239                  */
240                 if (!(proto == IPPROTO_TCP || proto == IPPROTO_UDP ||
241                     proto == IPPROTO_ICMP) || (low == 0 && high == 0)) {
242                         /*
243                          * XXX bug: icmp states don't use the id on both sides.
244                          * (traceroute -I through nat)
245                          */
246                         key.port[1] = sport;
247                         if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
248                                 *nport = sport;
249                                 return (0);
250                         }
251                 } else if (low == high) {
252                         key.port[1] = htons(low);
253                         if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
254                                 *nport = htons(low);
255                                 return (0);
256                         }
257                 } else {
258                         uint32_t tmp;
259                         uint16_t cut;
260
261                         if (low > high) {
262                                 tmp = low;
263                                 low = high;
264                                 high = tmp;
265                         }
266                         /* low < high */
267                         cut = arc4random() % (1 + high - low) + low;
268                         /* low <= cut <= high */
269                         for (tmp = cut; tmp <= high && tmp <= 0xffff; ++tmp) {
270                                 key.port[1] = htons(tmp);
271                                 if (pf_find_state_all(&key, PF_IN, NULL) ==
272                                     NULL) {
273                                         *nport = htons(tmp);
274                                         return (0);
275                                 }
276                         }
277                         tmp = cut;
278                         for (tmp -= 1; tmp >= low && tmp <= 0xffff; --tmp) {
279                                 key.port[1] = htons(tmp);
280                                 if (pf_find_state_all(&key, PF_IN, NULL) ==
281                                     NULL) {
282                                         *nport = htons(tmp);
283                                         return (0);
284                                 }
285                         }
286                 }
287
288                 switch (r->rpool.opts & PF_POOL_TYPEMASK) {
289                 case PF_POOL_RANDOM:
290                 case PF_POOL_ROUNDROBIN:
291                         /*
292                          * pick a different source address since we're out
293                          * of free port choices for the current one.
294                          */
295                         if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn))
296                                 return (1);
297                         break;
298                 case PF_POOL_NONE:
299                 case PF_POOL_SRCHASH:
300                 case PF_POOL_BITMASK:
301                 default:
302                         return (1);
303                 }
304         } while (! PF_AEQ(&init_addr, naddr, af) );
305         return (1);                                     /* none available */
306 }
307
308 static int
309 pf_get_mape_sport(sa_family_t af, u_int8_t proto, struct pf_krule *r,
310     struct pf_addr *saddr, uint16_t sport, struct pf_addr *daddr,
311     uint16_t dport, struct pf_addr *naddr, uint16_t *nport,
312     struct pf_ksrc_node **sn)
313 {
314         uint16_t psmask, low, highmask;
315         uint16_t i, ahigh, cut;
316         int ashift, psidshift;
317
318         ashift = 16 - r->rpool.mape.offset;
319         psidshift = ashift - r->rpool.mape.psidlen;
320         psmask = r->rpool.mape.psid & ((1U << r->rpool.mape.psidlen) - 1);
321         psmask = psmask << psidshift;
322         highmask = (1U << psidshift) - 1;
323
324         ahigh = (1U << r->rpool.mape.offset) - 1;
325         cut = arc4random() & ahigh;
326         if (cut == 0)
327                 cut = 1;
328
329         for (i = cut; i <= ahigh; i++) {
330                 low = (i << ashift) | psmask;
331                 if (!pf_get_sport(af, proto, r, saddr, sport, daddr, dport,
332                     naddr, nport, low, low | highmask, sn))
333                         return (0);
334         }
335         for (i = cut - 1; i > 0; i--) {
336                 low = (i << ashift) | psmask;
337                 if (!pf_get_sport(af, proto, r, saddr, sport, daddr, dport,
338                     naddr, nport, low, low | highmask, sn))
339                         return (0);
340         }
341         return (1);
342 }
343
344 int
345 pf_map_addr(sa_family_t af, struct pf_krule *r, struct pf_addr *saddr,
346     struct pf_addr *naddr, struct pf_addr *init_addr, struct pf_ksrc_node **sn)
347 {
348         struct pf_kpool         *rpool = &r->rpool;
349         struct pf_addr          *raddr = NULL, *rmask = NULL;
350
351         /* Try to find a src_node if none was given and this
352            is a sticky-address rule. */
353         if (*sn == NULL && r->rpool.opts & PF_POOL_STICKYADDR &&
354             (r->rpool.opts & PF_POOL_TYPEMASK) != PF_POOL_NONE)
355                 *sn = pf_find_src_node(saddr, r, af, 0);
356
357         /* If a src_node was found or explicitly given and it has a non-zero
358            route address, use this address. A zeroed address is found if the
359            src node was created just a moment ago in pf_create_state and it
360            needs to be filled in with routing decision calculated here. */
361         if (*sn != NULL && !PF_AZERO(&(*sn)->raddr, af)) {
362                 /* If the supplied address is the same as the current one we've
363                  * been asked before, so tell the caller that there's no other
364                  * address to be had. */
365                 if (PF_AEQ(naddr, &(*sn)->raddr, af))
366                         return (1);
367
368                 PF_ACPY(naddr, &(*sn)->raddr, af);
369                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
370                         printf("pf_map_addr: src tracking maps ");
371                         pf_print_host(saddr, 0, af);
372                         printf(" to ");
373                         pf_print_host(naddr, 0, af);
374                         printf("\n");
375                 }
376                 return (0);
377         }
378
379         /* Find the route using chosen algorithm. Store the found route
380            in src_node if it was given or found. */
381         if (rpool->cur->addr.type == PF_ADDR_NOROUTE)
382                 return (1);
383         if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) {
384                 switch (af) {
385 #ifdef INET
386                 case AF_INET:
387                         if (rpool->cur->addr.p.dyn->pfid_acnt4 < 1 &&
388                             (rpool->opts & PF_POOL_TYPEMASK) !=
389                             PF_POOL_ROUNDROBIN)
390                                 return (1);
391                          raddr = &rpool->cur->addr.p.dyn->pfid_addr4;
392                          rmask = &rpool->cur->addr.p.dyn->pfid_mask4;
393                         break;
394 #endif /* INET */
395 #ifdef INET6
396                 case AF_INET6:
397                         if (rpool->cur->addr.p.dyn->pfid_acnt6 < 1 &&
398                             (rpool->opts & PF_POOL_TYPEMASK) !=
399                             PF_POOL_ROUNDROBIN)
400                                 return (1);
401                         raddr = &rpool->cur->addr.p.dyn->pfid_addr6;
402                         rmask = &rpool->cur->addr.p.dyn->pfid_mask6;
403                         break;
404 #endif /* INET6 */
405                 }
406         } else if (rpool->cur->addr.type == PF_ADDR_TABLE) {
407                 if ((rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_ROUNDROBIN)
408                         return (1); /* unsupported */
409         } else {
410                 raddr = &rpool->cur->addr.v.a.addr;
411                 rmask = &rpool->cur->addr.v.a.mask;
412         }
413
414         switch (rpool->opts & PF_POOL_TYPEMASK) {
415         case PF_POOL_NONE:
416                 PF_ACPY(naddr, raddr, af);
417                 break;
418         case PF_POOL_BITMASK:
419                 PF_POOLMASK(naddr, raddr, rmask, saddr, af);
420                 break;
421         case PF_POOL_RANDOM:
422                 if (init_addr != NULL && PF_AZERO(init_addr, af)) {
423                         switch (af) {
424 #ifdef INET
425                         case AF_INET:
426                                 rpool->counter.addr32[0] = htonl(arc4random());
427                                 break;
428 #endif /* INET */
429 #ifdef INET6
430                         case AF_INET6:
431                                 if (rmask->addr32[3] != 0xffffffff)
432                                         rpool->counter.addr32[3] =
433                                             htonl(arc4random());
434                                 else
435                                         break;
436                                 if (rmask->addr32[2] != 0xffffffff)
437                                         rpool->counter.addr32[2] =
438                                             htonl(arc4random());
439                                 else
440                                         break;
441                                 if (rmask->addr32[1] != 0xffffffff)
442                                         rpool->counter.addr32[1] =
443                                             htonl(arc4random());
444                                 else
445                                         break;
446                                 if (rmask->addr32[0] != 0xffffffff)
447                                         rpool->counter.addr32[0] =
448                                             htonl(arc4random());
449                                 break;
450 #endif /* INET6 */
451                         }
452                         PF_POOLMASK(naddr, raddr, rmask, &rpool->counter, af);
453                         PF_ACPY(init_addr, naddr, af);
454
455                 } else {
456                         PF_AINC(&rpool->counter, af);
457                         PF_POOLMASK(naddr, raddr, rmask, &rpool->counter, af);
458                 }
459                 break;
460         case PF_POOL_SRCHASH:
461             {
462                 unsigned char hash[16];
463
464                 pf_hash(saddr, (struct pf_addr *)&hash, &rpool->key, af);
465                 PF_POOLMASK(naddr, raddr, rmask, (struct pf_addr *)&hash, af);
466                 break;
467             }
468         case PF_POOL_ROUNDROBIN:
469             {
470                 struct pf_kpooladdr *acur = rpool->cur;
471
472                 /*
473                  * XXXGL: in the round-robin case we need to store
474                  * the round-robin machine state in the rule, thus
475                  * forwarding thread needs to modify rule.
476                  *
477                  * This is done w/o locking, because performance is assumed
478                  * more important than round-robin precision.
479                  *
480                  * In the simpliest case we just update the "rpool->cur"
481                  * pointer. However, if pool contains tables or dynamic
482                  * addresses, then "tblidx" is also used to store machine
483                  * state. Since "tblidx" is int, concurrent access to it can't
484                  * lead to inconsistence, only to lost of precision.
485                  *
486                  * Things get worse, if table contains not hosts, but
487                  * prefixes. In this case counter also stores machine state,
488                  * and for IPv6 address, counter can't be updated atomically.
489                  * Probably, using round-robin on a table containing IPv6
490                  * prefixes (or even IPv4) would cause a panic.
491                  */
492
493                 if (rpool->cur->addr.type == PF_ADDR_TABLE) {
494                         if (!pfr_pool_get(rpool->cur->addr.p.tbl,
495                             &rpool->tblidx, &rpool->counter, af))
496                                 goto get_addr;
497                 } else if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) {
498                         if (!pfr_pool_get(rpool->cur->addr.p.dyn->pfid_kt,
499                             &rpool->tblidx, &rpool->counter, af))
500                                 goto get_addr;
501                 } else if (pf_match_addr(0, raddr, rmask, &rpool->counter, af))
502                         goto get_addr;
503
504         try_next:
505                 if (TAILQ_NEXT(rpool->cur, entries) == NULL)
506                         rpool->cur = TAILQ_FIRST(&rpool->list);
507                 else
508                         rpool->cur = TAILQ_NEXT(rpool->cur, entries);
509                 if (rpool->cur->addr.type == PF_ADDR_TABLE) {
510                         rpool->tblidx = -1;
511                         if (pfr_pool_get(rpool->cur->addr.p.tbl,
512                             &rpool->tblidx, &rpool->counter, af)) {
513                                 /* table contains no address of type 'af' */
514                                 if (rpool->cur != acur)
515                                         goto try_next;
516                                 return (1);
517                         }
518                 } else if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) {
519                         rpool->tblidx = -1;
520                         if (pfr_pool_get(rpool->cur->addr.p.dyn->pfid_kt,
521                             &rpool->tblidx, &rpool->counter, af)) {
522                                 /* table contains no address of type 'af' */
523                                 if (rpool->cur != acur)
524                                         goto try_next;
525                                 return (1);
526                         }
527                 } else {
528                         raddr = &rpool->cur->addr.v.a.addr;
529                         rmask = &rpool->cur->addr.v.a.mask;
530                         PF_ACPY(&rpool->counter, raddr, af);
531                 }
532
533         get_addr:
534                 PF_ACPY(naddr, &rpool->counter, af);
535                 if (init_addr != NULL && PF_AZERO(init_addr, af))
536                         PF_ACPY(init_addr, naddr, af);
537                 PF_AINC(&rpool->counter, af);
538                 break;
539             }
540         }
541         if (*sn != NULL)
542                 PF_ACPY(&(*sn)->raddr, naddr, af);
543
544         if (V_pf_status.debug >= PF_DEBUG_MISC &&
545             (rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) {
546                 printf("pf_map_addr: selected address ");
547                 pf_print_host(naddr, 0, af);
548                 printf("\n");
549         }
550
551         return (0);
552 }
553
554 struct pf_krule *
555 pf_get_translation(struct pf_pdesc *pd, struct mbuf *m, int off, int direction,
556     struct pfi_kkif *kif, struct pf_ksrc_node **sn,
557     struct pf_state_key **skp, struct pf_state_key **nkp,
558     struct pf_addr *saddr, struct pf_addr *daddr,
559     uint16_t sport, uint16_t dport, struct pf_kanchor_stackframe *anchor_stack)
560 {
561         struct pf_krule *r = NULL;
562         struct pf_addr  *naddr;
563         uint16_t        *nport;
564         uint16_t         low, high;
565
566         PF_RULES_RASSERT();
567         KASSERT(*skp == NULL, ("*skp not NULL"));
568         KASSERT(*nkp == NULL, ("*nkp not NULL"));
569
570         if (direction == PF_OUT) {
571                 r = pf_match_translation(pd, m, off, direction, kif, saddr,
572                     sport, daddr, dport, PF_RULESET_BINAT, anchor_stack);
573                 if (r == NULL)
574                         r = pf_match_translation(pd, m, off, direction, kif,
575                             saddr, sport, daddr, dport, PF_RULESET_NAT,
576                             anchor_stack);
577         } else {
578                 r = pf_match_translation(pd, m, off, direction, kif, saddr,
579                     sport, daddr, dport, PF_RULESET_RDR, anchor_stack);
580                 if (r == NULL)
581                         r = pf_match_translation(pd, m, off, direction, kif,
582                             saddr, sport, daddr, dport, PF_RULESET_BINAT,
583                             anchor_stack);
584         }
585
586         if (r == NULL)
587                 return (NULL);
588
589         switch (r->action) {
590         case PF_NONAT:
591         case PF_NOBINAT:
592         case PF_NORDR:
593                 return (NULL);
594         }
595
596         *skp = pf_state_key_setup(pd, saddr, daddr, sport, dport);
597         if (*skp == NULL)
598                 return (NULL);
599         *nkp = pf_state_key_clone(*skp);
600         if (*nkp == NULL) {
601                 uma_zfree(V_pf_state_key_z, *skp);
602                 *skp = NULL;
603                 return (NULL);
604         }
605
606         /* XXX We only modify one side for now. */
607         naddr = &(*nkp)->addr[1];
608         nport = &(*nkp)->port[1];
609
610         switch (r->action) {
611         case PF_NAT:
612                 if (pd->proto == IPPROTO_ICMP) {
613                         low  = 1;
614                         high = 65535;
615                 } else {
616                         low  = r->rpool.proxy_port[0];
617                         high = r->rpool.proxy_port[1];
618                 }
619                 if (r->rpool.mape.offset > 0) {
620                         if (pf_get_mape_sport(pd->af, pd->proto, r, saddr,
621                             sport, daddr, dport, naddr, nport, sn)) {
622                                 DPFPRINTF(PF_DEBUG_MISC,
623                                     ("pf: MAP-E port allocation (%u/%u/%u)"
624                                     " failed\n",
625                                     r->rpool.mape.offset,
626                                     r->rpool.mape.psidlen,
627                                     r->rpool.mape.psid));
628                                 goto notrans;
629                         }
630                 } else if (pf_get_sport(pd->af, pd->proto, r, saddr, sport,
631                     daddr, dport, naddr, nport, low, high, sn)) {
632                         DPFPRINTF(PF_DEBUG_MISC,
633                             ("pf: NAT proxy port allocation (%u-%u) failed\n",
634                             r->rpool.proxy_port[0], r->rpool.proxy_port[1]));
635                         goto notrans;
636                 }
637                 break;
638         case PF_BINAT:
639                 switch (direction) {
640                 case PF_OUT:
641                         if (r->rpool.cur->addr.type == PF_ADDR_DYNIFTL){
642                                 switch (pd->af) {
643 #ifdef INET
644                                 case AF_INET:
645                                         if (r->rpool.cur->addr.p.dyn->
646                                             pfid_acnt4 < 1)
647                                                 goto notrans;
648                                         PF_POOLMASK(naddr,
649                                             &r->rpool.cur->addr.p.dyn->
650                                             pfid_addr4,
651                                             &r->rpool.cur->addr.p.dyn->
652                                             pfid_mask4, saddr, AF_INET);
653                                         break;
654 #endif /* INET */
655 #ifdef INET6
656                                 case AF_INET6:
657                                         if (r->rpool.cur->addr.p.dyn->
658                                             pfid_acnt6 < 1)
659                                                 goto notrans;
660                                         PF_POOLMASK(naddr,
661                                             &r->rpool.cur->addr.p.dyn->
662                                             pfid_addr6,
663                                             &r->rpool.cur->addr.p.dyn->
664                                             pfid_mask6, saddr, AF_INET6);
665                                         break;
666 #endif /* INET6 */
667                                 }
668                         } else
669                                 PF_POOLMASK(naddr,
670                                     &r->rpool.cur->addr.v.a.addr,
671                                     &r->rpool.cur->addr.v.a.mask, saddr,
672                                     pd->af);
673                         break;
674                 case PF_IN:
675                         if (r->src.addr.type == PF_ADDR_DYNIFTL) {
676                                 switch (pd->af) {
677 #ifdef INET
678                                 case AF_INET:
679                                         if (r->src.addr.p.dyn-> pfid_acnt4 < 1)
680                                                 goto notrans;
681                                         PF_POOLMASK(naddr,
682                                             &r->src.addr.p.dyn->pfid_addr4,
683                                             &r->src.addr.p.dyn->pfid_mask4,
684                                             daddr, AF_INET);
685                                         break;
686 #endif /* INET */
687 #ifdef INET6
688                                 case AF_INET6:
689                                         if (r->src.addr.p.dyn->pfid_acnt6 < 1)
690                                                 goto notrans;
691                                         PF_POOLMASK(naddr,
692                                             &r->src.addr.p.dyn->pfid_addr6,
693                                             &r->src.addr.p.dyn->pfid_mask6,
694                                             daddr, AF_INET6);
695                                         break;
696 #endif /* INET6 */
697                                 }
698                         } else
699                                 PF_POOLMASK(naddr, &r->src.addr.v.a.addr,
700                                     &r->src.addr.v.a.mask, daddr, pd->af);
701                         break;
702                 }
703                 break;
704         case PF_RDR: {
705                 if (pf_map_addr(pd->af, r, saddr, naddr, NULL, sn))
706                         goto notrans;
707                 if ((r->rpool.opts & PF_POOL_TYPEMASK) == PF_POOL_BITMASK)
708                         PF_POOLMASK(naddr, naddr, &r->rpool.cur->addr.v.a.mask,
709                             daddr, pd->af);
710
711                 if (r->rpool.proxy_port[1]) {
712                         uint32_t        tmp_nport;
713
714                         tmp_nport = ((ntohs(dport) - ntohs(r->dst.port[0])) %
715                             (r->rpool.proxy_port[1] - r->rpool.proxy_port[0] +
716                             1)) + r->rpool.proxy_port[0];
717
718                         /* Wrap around if necessary. */
719                         if (tmp_nport > 65535)
720                                 tmp_nport -= 65535;
721                         *nport = htons((uint16_t)tmp_nport);
722                 } else if (r->rpool.proxy_port[0])
723                         *nport = htons(r->rpool.proxy_port[0]);
724                 break;
725         }
726         default:
727                 panic("%s: unknown action %u", __func__, r->action);
728         }
729
730         /* Return success only if translation really happened. */
731         if (bcmp(*skp, *nkp, sizeof(struct pf_state_key_cmp)))
732                 return (r);
733
734 notrans:
735         uma_zfree(V_pf_state_key_z, *nkp);
736         uma_zfree(V_pf_state_key_z, *skp);
737         *skp = *nkp = NULL;
738         *sn = NULL;
739
740         return (NULL);
741 }