sys/net/pf/pf.c

   1 /*
   2  * Copyright (c) 2004 The DragonFly Project.  All rights reserved.
   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  */
  37
  38 #include "opt_inet.h"
  39 #include "opt_inet6.h"
  40
  41 #include <sys/param.h>
  42 #include <sys/systm.h>
  43 #include <sys/malloc.h>
  44 #include <sys/mbuf.h>
  45 #include <sys/filio.h>
  46 #include <sys/socket.h>
  47 #include <sys/socketvar.h>
  48 #include <sys/kernel.h>
  49 #include <sys/time.h>
  50 #include <sys/sysctl.h>
  51 #include <sys/endian.h>
  52 #include <sys/proc.h>
  53 #include <sys/kthread.h>
  54 #include <sys/spinlock.h>
  55
  56 #include <machine/inttypes.h>
  57
  58 #include <sys/md5.h>
  59
  60 #include <net/if.h>
  61 #include <net/if_types.h>
  62 #include <net/bpf.h>
  63 #include <net/netisr2.h>
  64 #include <net/route.h>
  65
  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>
  81
  82 #include <net/pf/pfvar.h>
  83 #include <net/pf/if_pflog.h>
  84
  85 #include <net/pf/if_pfsync.h>
  86
  87 #ifdef INET6
  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>
  93 #endif /* INET6 */
  94
  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>
 102
 103 extern int ip_optcopy(struct ip *, struct ip *);
 104 extern int debug_pfugidhack;
 105
 106 /*
 107  * pf_token - shared lock for cpu-localized operations,
 108  *            exclusive lock otherwise.
 109  *
 110  * pf_gtoken- exclusive lock used for initialization.
 111  *
 112  * pf_spin  - only used to atomically fetch and increment stateid
 113  *            on 32-bit systems.
 114  */
 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");
 119 #endif
 120
 121 #define DPFPRINTF(n, x) if (pf_status.debug >= (n)) kprintf x
 122
 123 #define FAIL(code)      { error = (code); goto done; }
 124
 125 /*
 126  * Global variables
 127  */
 128
 129 /* mask radix tree */
 130 struct radix_node_head  *pf_maskhead;
 131
 132 /* state tables */
 133 struct pf_state_tree     pf_statetbl[MAXCPU+1]; /* incls one global table */
 134
 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;
 140
 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;
 145
 146 MD5_CTX                  pf_tcp_secret_ctx;
 147 u_char                   pf_tcp_secret[16];
 148 int                      pf_tcp_secret_init;
 149 int                      pf_tcp_iss_off;
 150
 151 struct pf_anchor_stackframe {
 152         struct pf_ruleset                       *rs;
 153         struct pf_rule                          *r;
 154         struct pf_anchor_node                   *parent;
 155         struct pf_anchor                        *child;
 156 } pf_anchor_stack[64];
 157
 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;
 161
 162 void                     pf_print_host(struct pf_addr *, u_int16_t, u_int8_t);
 163
 164 void                     pf_init_threshold(struct pf_threshold *, u_int32_t,
 165                             u_int32_t);
 166 void                     pf_add_threshold(struct pf_threshold *);
 167 int                      pf_check_threshold(struct pf_threshold *);
 168
 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 *);
 174 #ifdef INET6
 175 void                     pf_change_a6(struct pf_addr *, u_int16_t *,
 176                             struct pf_addr *, u_int8_t);
 177 #endif /* INET6 */
 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 *,
 192                             u_int16_t, int);
 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,
 218                             int);
 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 **,
 245                              int *);
 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 *,
 260                             struct pf_pdesc *);
 261 void                     pf_route6(struct mbuf **, struct pf_rule *, int,
 262                             struct ifnet *, struct pf_state *,
 263                             struct pf_pdesc *);
 264 u_int8_t                 pf_get_wscale(struct mbuf *, int, u_int16_t,
 265                             sa_family_t);
 266 u_int16_t                pf_get_mss(struct mbuf *, int, u_int16_t,
 267                             sa_family_t);
 268 u_int16_t                pf_calc_mss(struct pf_addr *, sa_family_t,
 269                                 u_int16_t);
 270 void                     pf_set_rt_ifp(struct pf_state *,
 271                             struct pf_addr *);
 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 *);
 283
 284 extern int pf_end_threads;
 285
 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 }
 292 };
 293
 294 /*
 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)
 299  */
 300 #define STATE_LOOKUP(i, k, d, s, m)                                     \
 301         do {                                                            \
 302                 s = pf_find_state(i, k, d, m);                          \
 303                 if (s == NULL || (s)->timeout == PFTM_PURGE)            \
 304                         return (PF_DROP);                               \
 305                 if (d == PF_OUT &&                                      \
 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 &&                              \
 311                     (s)->rt_kif != i)                                   \
 312                         return (PF_PASS);                               \
 313         } while (0)
 314
 315 #define BOUND_IFACE(r, k) \
 316         ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : pfi_all
 317
 318 #define STATE_INC_COUNTERS(s)                           \
 319         do {                                            \
 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++;    \
 325                 }                                       \
 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++;  \
 329                 }                                       \
 330         } while (0)
 331
 332 #define STATE_DEC_COUNTERS(s)                           \
 333         do {                                            \
 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);           \
 339         } while (0)
 340
 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");
 345
 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 *,
 352                                 struct pf_state *);
 353
 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];
 357
 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);
 362
 363 static __inline int
 364 pf_src_compare(struct pf_src_node *a, struct pf_src_node *b)
 365 {
 366         int     diff;
 367
 368         if (a->rule.ptr > b->rule.ptr)
 369                 return (1);
 370         if (a->rule.ptr < b->rule.ptr)
 371                 return (-1);
 372         if ((diff = a->af - b->af) != 0)
 373                 return (diff);
 374         switch (a->af) {
 375 #ifdef INET
 376         case AF_INET:
 377                 if (a->addr.addr32[0] > b->addr.addr32[0])
 378                         return (1);
 379                 if (a->addr.addr32[0] < b->addr.addr32[0])
 380                         return (-1);
 381                 break;
 382 #endif /* INET */
 383 #ifdef INET6
 384         case AF_INET6:
 385                 if (a->addr.addr32[3] > b->addr.addr32[3])
 386                         return (1);
 387                 if (a->addr.addr32[3] < b->addr.addr32[3])
 388                         return (-1);
 389                 if (a->addr.addr32[2] > b->addr.addr32[2])
 390                         return (1);
 391                 if (a->addr.addr32[2] < b->addr.addr32[2])
 392                         return (-1);
 393                 if (a->addr.addr32[1] > b->addr.addr32[1])
 394                         return (1);
 395                 if (a->addr.addr32[1] < b->addr.addr32[1])
 396                         return (-1);
 397                 if (a->addr.addr32[0] > b->addr.addr32[0])
 398                         return (1);
 399                 if (a->addr.addr32[0] < b->addr.addr32[0])
 400                         return (-1);
 401                 break;
 402 #endif /* INET6 */
 403         }
 404         return (0);
 405 }
 406
 407 u_int32_t
 408 pf_state_hash(struct pf_state_key *sk)
 409 {
 410         u_int32_t hv = (u_int32_t)(((intptr_t)sk >> 6) ^ ((intptr_t)sk >> 15));
 411         if (hv == 0)    /* disallow 0 */
 412                 hv = 1;
 413         return(hv);
 414 }
 415
 416 #ifdef INET6
 417 void
 418 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
 419 {
 420         switch (af) {
 421 #ifdef INET
 422         case AF_INET:
 423                 dst->addr32[0] = src->addr32[0];
 424                 break;
 425 #endif /* INET */
 426         case AF_INET6:
 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];
 431                 break;
 432         }
 433 }
 434 #endif /* INET6 */
 435
 436 void
 437 pf_init_threshold(struct pf_threshold *threshold,
 438     u_int32_t limit, u_int32_t seconds)
 439 {
 440         threshold->limit = limit * PF_THRESHOLD_MULT;
 441         threshold->seconds = seconds;
 442         threshold->count = 0;
 443         threshold->last = time_second;
 444 }
 445
 446 void
 447 pf_add_threshold(struct pf_threshold *threshold)
 448 {
 449         u_int32_t t = time_second, diff = t - threshold->last;
 450
 451         if (diff >= threshold->seconds)
 452                 threshold->count = 0;
 453         else
 454                 threshold->count -= threshold->count * diff /
 455                     threshold->seconds;
 456         threshold->count += PF_THRESHOLD_MULT;
 457         threshold->last = t;
 458 }
 459
 460 int
 461 pf_check_threshold(struct pf_threshold *threshold)
 462 {
 463         return (threshold->count > threshold->limit);
 464 }
 465
 466 int
 467 pf_src_connlimit(struct pf_state *state)
 468 {
 469         int bad = 0;
 470         int cpu = mycpu->gd_cpuid;
 471
 472         state->src_node->conn++;
 473         state->src.tcp_est = 1;
 474         pf_add_threshold(&state->src_node->conn_rate);
 475
 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]++;
 480                 bad++;
 481         }
 482
 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]++;
 486                 bad++;
 487         }
 488
 489         if (!bad)
 490                 return 0;
 491
 492         if (state->rule.ptr->overload_tbl) {
 493                 struct pfr_addr p;
 494                 u_int32_t       killed = 0;
 495
 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);
 501                 }
 502
 503                 bzero(&p, sizeof(p));
 504                 p.pfra_af = state->key[PF_SK_WIRE]->af;
 505                 switch (state->key[PF_SK_WIRE]->af) {
 506 #ifdef INET
 507                 case AF_INET:
 508                         p.pfra_net = 32;
 509                         p.pfra_ip4addr = state->src_node->addr.v4;
 510                         break;
 511 #endif /* INET */
 512 #ifdef INET6
 513                 case AF_INET6:
 514                         p.pfra_net = 128;
 515                         p.pfra_ip6addr = state->src_node->addr.v6;
 516                         break;
 517 #endif /* INET6 */
 518                 }
 519
 520                 pfr_insert_kentry(state->rule.ptr->overload_tbl,
 521                     &p, time_second);
 522
 523                 /* kill existing states if that's required. */
 524                 if (state->rule.ptr->flush) {
 525                         struct pf_state_key *sk;
 526                         struct pf_state *st;
 527
 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];
 531                                 /*
 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).
 535                                  */
 536                                 if (sk->af ==
 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 &
 545                                     PF_FLUSH_GLOBAL ||
 546                                     state->rule.ptr == st->rule.ptr)) {
 547                                         st->timeout = PFTM_PURGE;
 548                                         st->src.state = st->dst.state =
 549                                             TCPS_CLOSED;
 550                                         killed++;
 551                                 }
 552                         }
 553                         if (pf_status.debug >= PF_DEBUG_MISC)
 554                                 kprintf(", %u states killed", killed);
 555                 }
 556                 if (pf_status.debug >= PF_DEBUG_MISC)
 557                         kprintf("\n");
 558         }
 559
 560         /* kill this state */
 561         state->timeout = PFTM_PURGE;
 562         state->src.state = state->dst.state = TCPS_CLOSED;
 563
 564         return 1;
 565 }
 566
 567 int
 568 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
 569     struct pf_addr *src, sa_family_t af)
 570 {
 571         struct pf_src_node      k;
 572         int cpu = mycpu->gd_cpuid;
 573
 574         bzero(&k, sizeof(k));   /* avoid gcc warnings */
 575         if (*sn == NULL) {
 576                 k.af = af;
 577                 PF_ACPY(&k.addr, src, af);
 578                 if (rule->rule_flag & PFRULE_RULESRCTRACK ||
 579                     rule->rpool.opts & PF_POOL_STICKYADDR)
 580                         k.rule.ptr = rule;
 581                 else
 582                         k.rule.ptr = NULL;
 583                 pf_status.scounters[SCNT_SRC_NODE_SEARCH]++;
 584                 *sn = RB_FIND(pf_src_tree, &tree_src_tracking[cpu], &k);
 585         }
 586         if (*sn == NULL) {
 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);
 591                 else
 592                         pf_status.lcounters[LCNT_SRCNODES]++;
 593                 if ((*sn) == NULL)
 594                         return (-1);
 595
 596                 pf_init_threshold(&(*sn)->conn_rate,
 597                     rule->max_src_conn_rate.limit,
 598                     rule->max_src_conn_rate.seconds);
 599
 600                 (*sn)->af = af;
 601                 if (rule->rule_flag & PFRULE_RULESRCTRACK ||
 602                     rule->rpool.opts & PF_POOL_STICKYADDR)
 603                         (*sn)->rule.ptr = rule;
 604                 else
 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);
 612                                 kprintf("\n");
 613                         }
 614                         kfree(*sn, M_PFSRCTREEPL);
 615                         return (-1);
 616                 }
 617
 618                 /*
 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).
 622                  */
 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);
 629         } else {
 630                 if (rule->max_src_states &&
 631                     (*sn)->states >= rule->max_src_states) {
 632                         pf_status.lcounters[LCNT_SRCSTATES]++;
 633                         return (-1);
 634                 }
 635         }
 636         return (0);
 637 }
 638
 639 /*
 640  * state table (indexed by the pf_state_key structure), normal RBTREE
 641  * comparison.
 642  */
 643 static __inline int
 644 pf_state_compare_key(struct pf_state_key *a, struct pf_state_key *b)
 645 {
 646         int     diff;
 647
 648         if ((diff = a->proto - b->proto) != 0)
 649                 return (diff);
 650         if ((diff = a->af - b->af) != 0)
 651                 return (diff);
 652         switch (a->af) {
 653 #ifdef INET
 654         case AF_INET:
 655                 if (a->addr[0].addr32[0] > b->addr[0].addr32[0])
 656                         return (1);
 657                 if (a->addr[0].addr32[0] < b->addr[0].addr32[0])
 658                         return (-1);
 659                 if (a->addr[1].addr32[0] > b->addr[1].addr32[0])
 660                         return (1);
 661                 if (a->addr[1].addr32[0] < b->addr[1].addr32[0])
 662                         return (-1);
 663                 break;
 664 #endif /* INET */
 665 #ifdef INET6
 666         case AF_INET6:
 667                 if (a->addr[0].addr32[3] > b->addr[0].addr32[3])
 668                         return (1);
 669                 if (a->addr[0].addr32[3] < b->addr[0].addr32[3])
 670                         return (-1);
 671                 if (a->addr[1].addr32[3] > b->addr[1].addr32[3])
 672                         return (1);
 673                 if (a->addr[1].addr32[3] < b->addr[1].addr32[3])
 674                         return (-1);
 675                 if (a->addr[0].addr32[2] > b->addr[0].addr32[2])
 676                         return (1);
 677                 if (a->addr[0].addr32[2] < b->addr[0].addr32[2])
 678                         return (-1);
 679                 if (a->addr[1].addr32[2] > b->addr[1].addr32[2])
 680                         return (1);
 681                 if (a->addr[1].addr32[2] < b->addr[1].addr32[2])
 682                         return (-1);
 683                 if (a->addr[0].addr32[1] > b->addr[0].addr32[1])
 684                         return (1);
 685                 if (a->addr[0].addr32[1] < b->addr[0].addr32[1])
 686                         return (-1);
 687                 if (a->addr[1].addr32[1] > b->addr[1].addr32[1])
 688                         return (1);
 689                 if (a->addr[1].addr32[1] < b->addr[1].addr32[1])
 690                         return (-1);
 691                 if (a->addr[0].addr32[0] > b->addr[0].addr32[0])
 692                         return (1);
 693                 if (a->addr[0].addr32[0] < b->addr[0].addr32[0])
 694                         return (-1);
 695                 if (a->addr[1].addr32[0] > b->addr[1].addr32[0])
 696                         return (1);
 697                 if (a->addr[1].addr32[0] < b->addr[1].addr32[0])
 698                         return (-1);
 699                 break;
 700 #endif /* INET6 */
 701         }
 702
 703         if ((diff = a->port[0] - b->port[0]) != 0)
 704                 return (diff);
 705         if ((diff = a->port[1] - b->port[1]) != 0)
 706                 return (diff);
 707
 708         return (0);
 709 }
 710
 711 /*
 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).
 715  */
 716 static __inline int
 717 pf_state_compare_rkey(struct pf_state_key *a, struct pf_state_key *b)
 718 {
 719         int     diff;
 720
 721         if ((diff = a->proto - b->proto) != 0)
 722                 return (diff);
 723         if ((diff = a->af - b->af) != 0)
 724                 return (diff);
 725         switch (a->af) {
 726 #ifdef INET
 727         case AF_INET:
 728                 if (a->addr[1].addr32[0] > b->addr[0].addr32[0])
 729                         return (1);
 730                 if (a->addr[1].addr32[0] < b->addr[0].addr32[0])
 731                         return (-1);
 732                 if (a->addr[0].addr32[0] > b->addr[1].addr32[0])
 733                         return (1);
 734                 if (a->addr[0].addr32[0] < b->addr[1].addr32[0])
 735                         return (-1);
 736                 break;
 737 #endif /* INET */
 738 #ifdef INET6
 739         case AF_INET6:
 740                 if (a->addr[1].addr32[3] > b->addr[0].addr32[3])
 741                         return (1);
 742                 if (a->addr[1].addr32[3] < b->addr[0].addr32[3])
 743                         return (-1);
 744                 if (a->addr[0].addr32[3] > b->addr[1].addr32[3])
 745                         return (1);
 746                 if (a->addr[0].addr32[3] < b->addr[1].addr32[3])
 747                         return (-1);
 748                 if (a->addr[1].addr32[2] > b->addr[0].addr32[2])
 749                         return (1);
 750                 if (a->addr[1].addr32[2] < b->addr[0].addr32[2])
 751                         return (-1);
 752                 if (a->addr[0].addr32[2] > b->addr[1].addr32[2])
 753                         return (1);
 754                 if (a->addr[0].addr32[2] < b->addr[1].addr32[2])
 755                         return (-1);
 756                 if (a->addr[1].addr32[1] > b->addr[0].addr32[1])
 757                         return (1);
 758                 if (a->addr[1].addr32[1] < b->addr[0].addr32[1])
 759                         return (-1);
 760                 if (a->addr[0].addr32[1] > b->addr[1].addr32[1])
 761                         return (1);
 762                 if (a->addr[0].addr32[1] < b->addr[1].addr32[1])
 763                         return (-1);
 764                 if (a->addr[1].addr32[0] > b->addr[0].addr32[0])
 765                         return (1);
 766                 if (a->addr[1].addr32[0] < b->addr[0].addr32[0])
 767                         return (-1);
 768                 if (a->addr[0].addr32[0] > b->addr[1].addr32[0])
 769                         return (1);
 770                 if (a->addr[0].addr32[0] < b->addr[1].addr32[0])
 771                         return (-1);
 772                 break;
 773 #endif /* INET6 */
 774         }
 775
 776         if ((diff = a->port[1] - b->port[0]) != 0)
 777                 return (diff);
 778         if ((diff = a->port[0] - b->port[1]) != 0)
 779                 return (diff);
 780
 781         return (0);
 782 }
 783
 784 static __inline int
 785 pf_state_compare_id(struct pf_state *a, struct pf_state *b)
 786 {
 787         if (a->id > b->id)
 788                 return (1);
 789         if (a->id < b->id)
 790                 return (-1);
 791         if (a->creatorid > b->creatorid)
 792                 return (1);
 793         if (a->creatorid < b->creatorid)
 794                 return (-1);
 795
 796         return (0);
 797 }
 798
 799 int
 800 pf_state_key_attach(struct pf_state_key *sk, struct pf_state *s, int idx)
 801 {
 802         struct pf_state_item    *si;
 803         struct pf_state_key     *cur;
 804         int cpu;
 805         int error;
 806
 807         /*
 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.
 811          *
 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
 814          * in the state.
 815          */
 816         if (s->state_flags & PFSTATE_STACK_GLOBAL) {
 817                 cpu = MAXCPU;
 818                 lockmgr(&pf_global_statetbl_lock, LK_EXCLUSIVE);
 819         } else {
 820                 cpu = mycpu->gd_cpuid;
 821                 lockmgr(&pf_global_statetbl_lock, LK_SHARED);
 822         }
 823         KKASSERT(s->key[idx] == NULL);  /* XXX handle this? */
 824
 825         if (pf_status.debug >= PF_DEBUG_MISC) {
 826                 kprintf("state_key attach cpu %d (%08x:%d) %s (%08x:%d)\n",
 827                         cpu,
 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]));
 831         }
 832
 833         /*
 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.
 842          *
 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.
 846          *
 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.
 850          *
 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).
 856          *
 857          * pf_global_statetbl_lock will be locked shared when testing and
 858          * not entering into the global state table.
 859          */
 860         if (cpu != MAXCPU &&
 861             (cur = RB_FIND(pf_state_rtree,
 862                            (struct pf_state_rtree *)&pf_statetbl[MAXCPU],
 863                            sk)) != NULL) {
 864                 TAILQ_FOREACH(si, &cur->states, entry) {
 865                         /*
 866                          * NOTE: We must ignore direction mismatches.
 867                          */
 868                         if (si->s->kif == s->kif) {
 869                                 s->state_flags |= PFSTATE_HALF_DUPLEX |
 870                                                   PFSTATE_SLOPPY;
 871                                 if (pf_status.debug >= PF_DEBUG_MISC) {
 872                                         kprintf(
 873                                             "pf: %s key attach collision "
 874                                             "on %s: ",
 875                                             (idx == PF_SK_WIRE) ?
 876                                             "wire" : "stack",
 877                                             s->kif->pfik_name);
 878                                         pf_print_state_parts(s,
 879                                             (idx == PF_SK_WIRE) ? sk : NULL,
 880                                             (idx == PF_SK_STACK) ? sk : NULL);
 881                                         kprintf("\n");
 882                                 }
 883                                 break;
 884                         }
 885                 }
 886         }
 887
 888         /*
 889          * Enter into either the per-cpu or the global state table.
 890          *
 891          * pf_global_statetbl_lock will be locked exclusively when entering
 892          * into the global state table.
 893          */
 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) {
 900                                         kprintf(
 901                                             "pf: %s key attach failed on %s: ",
 902                                             (idx == PF_SK_WIRE) ?
 903                                             "wire" : "stack",
 904                                             s->kif->pfik_name);
 905                                         pf_print_state_parts(s,
 906                                             (idx == PF_SK_WIRE) ? sk : NULL,
 907                                             (idx == PF_SK_STACK) ? sk : NULL);
 908                                         kprintf("\n");
 909                                 }
 910                                 kfree(sk, M_PFSTATEKEYPL);
 911                                 error = -1;
 912                                 goto failed;    /* collision! */
 913                         }
 914                 }
 915                 kfree(sk, M_PFSTATEKEYPL);
 916
 917                 s->key[idx] = cur;
 918         } else {
 919                 s->key[idx] = sk;
 920         }
 921
 922         if ((si = kmalloc(sizeof(struct pf_state_item),
 923                           M_PFSTATEITEMPL, M_NOWAIT)) == NULL) {
 924                 pf_state_key_detach(s, idx);
 925                 error = -1;
 926                 goto failed;    /* collision! */
 927         }
 928         si->s = s;
 929
 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);
 933         else
 934                 TAILQ_INSERT_HEAD(&s->key[idx]->states, si, entry);
 935
 936         error = 0;
 937 failed:
 938         lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
 939         return error;
 940 }
 941
 942 /*
 943  * NOTE: Can only be called indirectly via the purge thread with pf_token
 944  *       exclusively locked.
 945  */
 946 void
 947 pf_detach_state(struct pf_state *s)
 948 {
 949         if (s->key[PF_SK_WIRE] == s->key[PF_SK_STACK])
 950                 s->key[PF_SK_WIRE] = NULL;
 951
 952         if (s->key[PF_SK_STACK] != NULL)
 953                 pf_state_key_detach(s, PF_SK_STACK);
 954
 955         if (s->key[PF_SK_WIRE] != NULL)
 956                 pf_state_key_detach(s, PF_SK_WIRE);
 957 }
 958
 959 /*
 960  * NOTE: Can only be called indirectly via the purge thread with pf_token
 961  *       exclusively locked.
 962  */
 963 void
 964 pf_state_key_detach(struct pf_state *s, int idx)
 965 {
 966         struct pf_state_item    *si;
 967         int cpu;
 968
 969         /*
 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.
 974          */
 975         if (s->state_flags & PFSTATE_STACK_GLOBAL) {
 976                 cpu = MAXCPU;
 977                 lockmgr(&pf_global_statetbl_lock, LK_EXCLUSIVE);
 978         } else {
 979                 cpu = mycpu->gd_cpuid;
 980         }
 981
 982         si = TAILQ_FIRST(&s->key[idx]->states);
 983         while (si && si->s != s)
 984                 si = TAILQ_NEXT(si, entry);
 985
 986         if (si) {
 987                 TAILQ_REMOVE(&s->key[idx]->states, si, entry);
 988                 kfree(si, M_PFSTATEITEMPL);
 989         }
 990
 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);
 998         }
 999         s->key[idx] = NULL;
1000
1001         if (s->state_flags & PFSTATE_STACK_GLOBAL)
1002                 lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
1003 }
1004
1005 struct pf_state_key *
1006 pf_alloc_state_key(int pool_flags)
1007 {
1008         struct pf_state_key     *sk;
1009
1010         sk = kmalloc(sizeof(struct pf_state_key), M_PFSTATEKEYPL, pool_flags);
1011         if (sk) {
1012                 TAILQ_INIT(&sk->states);
1013         }
1014         return (sk);
1015 }
1016
1017 int
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)
1023 {
1024         KKASSERT((*skp == NULL && *nkp == NULL));
1025
1026         if ((*skp = pf_alloc_state_key(M_NOWAIT | M_ZERO)) == NULL)
1027                 return (ENOMEM);
1028
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;
1035
1036         if (nr != NULL) {
1037                 if ((*nkp = pf_alloc_state_key(M_NOWAIT | M_ZERO)) == NULL)
1038                         return (ENOMEM); /* caller must handle cleanup */
1039
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;
1047         } else {
1048                 *nkp = *skp;
1049         }
1050
1051         if (pd->dir == PF_IN) {
1052                 *skw = *skp;
1053                 *sks = *nkp;
1054         } else {
1055                 *sks = *skp;
1056                 *skw = *nkp;
1057         }
1058         return (0);
1059 }
1060
1061 /*
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.
1065  */
1066 int
1067 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1068                 struct pf_state_key *sks, struct pf_state *s)
1069 {
1070         int cpu = mycpu->gd_cpuid;
1071
1072         s->kif = kif;
1073         s->cpuid = cpu;
1074
1075         if (skw == sks) {
1076                 if (pf_state_key_attach(skw, s, PF_SK_WIRE))
1077                         return (-1);
1078                 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1079         } else {
1080                 /*
1081                 skw->reverse = sks;
1082                 sks->reverse = skw;
1083                 */
1084                 if (pf_state_key_attach(skw, s, PF_SK_WIRE)) {
1085                         kfree(sks, M_PFSTATEKEYPL);
1086                         return (-1);
1087                 }
1088                 if (pf_state_key_attach(sks, s, PF_SK_STACK)) {
1089                         pf_state_key_detach(s, PF_SK_WIRE);
1090                         return (-1);
1091                 }
1092         }
1093
1094         if (s->id == 0 && s->creatorid == 0) {
1095                 u_int64_t sid;
1096
1097 #if __SIZEOF_LONG__ == 8
1098                 sid = atomic_fetchadd_long(&pf_status.stateid, 1);
1099 #else
1100                 spin_lock(&pf_spin);
1101                 sid = pf_status.stateid++;
1102                 spin_unlock(&pf_spin);
1103 #endif
1104                 s->id = htobe64(sid);
1105                 s->creatorid = pf_status.hostid;
1106         }
1107
1108         /*
1109          * Calculate hash code for altq
1110          */
1111         s->hash = crc32(s->key[PF_SK_WIRE], PF_STATE_KEY_HASH_LENGTH);
1112
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)");
1120                         kprintf("\n");
1121                 }
1122                 pf_detach_state(s);
1123                 return (-1);
1124         }
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);
1130         return (0);
1131 }
1132
1133 struct pf_state *
1134 pf_find_state_byid(struct pf_state_cmp *key)
1135 {
1136         int cpu = mycpu->gd_cpuid;
1137
1138         pf_status.fcounters[FCNT_STATE_SEARCH]++;
1139
1140         return (RB_FIND(pf_state_tree_id, &tree_id[cpu],
1141                         (struct pf_state *)key));
1142 }
1143
1144 /*
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
1148  */
1149 struct pf_state *
1150 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir,
1151               struct mbuf *m)
1152 {
1153         struct pf_state_key     *skey = (void *)key;
1154         struct pf_state_key     *sk;
1155         struct pf_state_item    *si;
1156         struct pf_state *s;
1157         int cpu = mycpu->gd_cpuid;
1158         int globalstl = 0;
1159
1160         pf_status.fcounters[FCNT_STATE_SEARCH]++;
1161
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;
1165         } else {
1166                 sk = RB_FIND(pf_state_tree, &pf_statetbl[cpu], skey);
1167                 if (sk == NULL) {
1168                         lockmgr(&pf_global_statetbl_lock, LK_SHARED);
1169                         sk = RB_FIND(pf_state_tree, &pf_statetbl[MAXCPU], skey);
1170                         if (sk == NULL) {
1171                                 lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
1172                                 return (NULL);
1173                         }
1174                         globalstl = 1;
1175                 }
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;
1180                 }
1181         }
1182         if (dir == PF_OUT)
1183                 m->m_pkthdr.pf.statekey = NULL;
1184
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])) {
1190                         break;
1191                 }
1192         }
1193
1194         /*
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
1200          * pf_token.
1201          */
1202         if (si) {
1203                 s = si->s;
1204                 if (s && (s->state_flags & PFSTATE_CREATEINPROG))
1205                         s = NULL;
1206         } else {
1207                 s = NULL;
1208         }
1209         if (globalstl)
1210                 lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
1211         return s;
1212 }
1213
1214 /*
1215  * WARNING! May return a state structure that was localized to another cpu,
1216  *          destruction is typically protected by the callers pf_token.
1217  */
1218 struct pf_state *
1219 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1220 {
1221         struct pf_state_key     *skey = (void *)key;
1222         struct pf_state_key     *sk;
1223         struct pf_state_item    *si, *ret = NULL;
1224         struct pf_state         *s;
1225         int cpu = mycpu->gd_cpuid;
1226         int globalstl = 0;
1227
1228         pf_status.fcounters[FCNT_STATE_SEARCH]++;
1229
1230         sk = RB_FIND(pf_state_tree, &pf_statetbl[cpu], skey);
1231         if (sk == NULL) {
1232                 lockmgr(&pf_global_statetbl_lock, LK_SHARED);
1233                 sk = RB_FIND(pf_state_tree, &pf_statetbl[MAXCPU], skey);
1234                 globalstl = 1;
1235         }
1236         if (sk != NULL) {
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]))) {
1241                                 if (more == NULL) {
1242                                         ret = si;
1243                                         break;
1244                                 }
1245                                 if (ret)
1246                                         (*more)++;
1247                                 else
1248                                         ret = si;
1249                         }
1250         }
1251
1252         /*
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
1258          * pf_token.
1259          */
1260         if (ret) {
1261                 s = ret->s;
1262                 if (s && (s->state_flags & PFSTATE_CREATEINPROG))
1263                         s = NULL;
1264         } else {
1265                 s = NULL;
1266         }
1267         if (globalstl)
1268                 lockmgr(&pf_global_statetbl_lock, LK_RELEASE);
1269         return s;
1270 }
1271
1272 /* END state table stuff */
1273
1274 void
1275 pf_purge_thread(void *v)
1276 {
1277         globaldata_t save_gd = mycpu;
1278         int nloops = 0;
1279         int locked = 0;
1280         int nn;
1281         int endingit;
1282
1283         for (;;) {
1284                 tsleep(pf_purge_thread, PWAIT, "pftm", 1 * hz);
1285
1286                 endingit = pf_end_threads;
1287
1288                 for (nn = 0; nn < ncpus; ++nn) {
1289                         lwkt_setcpu_self(globaldata_find(nn));
1290
1291                         lwkt_gettoken(&pf_token);
1292                         lockmgr(&pf_consistency_lock, LK_EXCLUSIVE);
1293                         crit_enter();
1294
1295                         /*
1296                          * process a fraction of the state table every second
1297                          */
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);
1306                         }
1307
1308                         /*
1309                          * purge other expired types every PFTM_INTERVAL
1310                          * seconds
1311                          */
1312                         if (++nloops >=
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);
1317                                 }
1318                                 nloops = 0;
1319                         }
1320
1321                         /*
1322                          * If terminating the thread, clean everything out
1323                          * (on all cpus).
1324                          */
1325                         if (endingit) {
1326                                 pf_purge_expired_states(pf_status.states, 0);
1327                                 pf_purge_expired_fragments();
1328                                 pf_purge_expired_src_nodes(1);
1329                         }
1330
1331                         crit_exit();
1332                         lockmgr(&pf_consistency_lock, LK_RELEASE);
1333                         lwkt_reltoken(&pf_token);
1334                 }
1335                 lwkt_setcpu_self(save_gd);
1336                 if (endingit)
1337                         break;
1338         }
1339
1340         /*
1341          * Thread termination
1342          */
1343         pf_end_threads++;
1344         wakeup(pf_purge_thread);
1345         kthread_exit();
1346 }
1347
1348 u_int32_t
1349 pf_state_expires(const struct pf_state *state)
1350 {
1351         u_int32_t       timeout;
1352         u_int32_t       start;
1353         u_int32_t       end;
1354         u_int32_t       states;
1355
1356         /* handle all PFTM_* > PFTM_MAX here */
1357         if (state->timeout == PFTM_PURGE)
1358                 return (time_second);
1359         if (state->timeout == PFTM_UNTIL_PACKET)
1360                 return (0);
1361         KKASSERT(state->timeout != PFTM_UNLINKED);
1362         KKASSERT(state->timeout < PFTM_MAX);
1363         timeout = state->rule.ptr->timeout[state->timeout];
1364         if (!timeout)
1365                 timeout = pf_default_rule.timeout[state->timeout];
1366         start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1367         if (start) {
1368                 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1369                 states = state->rule.ptr->states_cur;
1370         } else {
1371                 start = pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1372                 end = pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1373                 states = pf_status.states;
1374         }
1375         if (end && states > start && start < end) {
1376                 if (states < end)
1377                         return (state->expire + timeout * (end - states) /
1378                             (end - start));
1379                 else
1380                         return (time_second);
1381         }
1382         return (state->expire + timeout);
1383 }
1384
1385 /*
1386  * (called with exclusive pf_token)
1387  */
1388 int
1389 pf_purge_expired_src_nodes(int waslocked)
1390 {
1391         struct pf_src_node *cur, *next;
1392         int locked = waslocked;
1393         int cpu = mycpu->gd_cpuid;
1394
1395         for (cur = RB_MIN(pf_src_tree, &tree_src_tracking[cpu]);
1396              cur;
1397              cur = next) {
1398                 next = RB_NEXT(pf_src_tree, &tree_src_tracking[cpu], cur);
1399
1400                 if (cur->states <= 0 && cur->expire <= time_second) {
1401                          if (!locked) {
1402                                  lockmgr(&pf_consistency_lock, LK_EXCLUSIVE);
1403                                  next = RB_NEXT(pf_src_tree,
1404                                      &tree_src_tracking[cpu], cur);
1405                                  locked = 1;
1406                          }
1407                          if (cur->rule.ptr != NULL) {
1408                                 /*
1409                                  * decrements in rule should be ok, token is
1410                                  * held exclusively in this code path.
1411                                  */
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);
1416                          }
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);
1421                 }
1422         }
1423         if (locked && !waslocked)
1424                 lockmgr(&pf_consistency_lock, LK_RELEASE);
1425         return(1);
1426 }
1427
1428 void
1429 pf_src_tree_remove_state(struct pf_state *s)
1430 {
1431         u_int32_t timeout;
1432
1433         if (s->src_node != NULL) {
1434                 if (s->src.tcp_est)
1435                         --s->src_node->conn;
1436                 if (--s->src_node->states <= 0) {
1437                         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1438                         if (!timeout) {
1439                                 timeout =
1440                                     pf_default_rule.timeout[PFTM_SRC_NODE];
1441                         }
1442                         s->src_node->expire = time_second + timeout;
1443                 }
1444         }
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];
1448                         if (!timeout)
1449                                 timeout =
1450                                     pf_default_rule.timeout[PFTM_SRC_NODE];
1451                         s->nat_src_node->expire = time_second + timeout;
1452                 }
1453         }
1454         s->src_node = s->nat_src_node = NULL;
1455 }
1456
1457 /* callers should be at crit_enter() */
1458 void
1459 pf_unlink_state(struct pf_state *cur)
1460 {
1461         int cpu = mycpu->gd_cpuid;
1462
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);
1472         }
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);
1479 }
1480
1481 static struct pf_state  *purge_cur[MAXCPU];
1482
1483 /*
1484  * callers should be at crit_enter() and hold pf_consistency_lock exclusively.
1485  * pf_token must also be held exclusively.
1486  */
1487 void
1488 pf_free_state(struct pf_state *cur)
1489 {
1490         int cpu = mycpu->gd_cpuid;
1491
1492         KKASSERT(cur->cpuid == cpu);
1493
1494         if (pfsyncif != NULL &&
1495             (pfsyncif->sc_bulk_send_next == cur ||
1496             pfsyncif->sc_bulk_terminator == cur))
1497                 return;
1498         KKASSERT(cur->timeout == PFTM_UNLINKED);
1499         /*
1500          * decrements in rule should be ok, token is
1501          * held exclusively in this code path.
1502          */
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);
1510                 }
1511         }
1512         if (cur->anchor.ptr != NULL) {
1513                 if (--cur->anchor.ptr->states_cur <= 0)
1514                         pf_rm_rule(NULL, cur->anchor.ptr);
1515         }
1516         pf_normalize_tcp_cleanup(cur);
1517         pfi_kif_unref(cur->kif, PFI_KIF_REF_STATE);
1518
1519         /*
1520          * We may be freeing pf_purge_expired_states()'s saved scan entry,
1521          * adjust it if necessary.
1522          */
1523         if (purge_cur[cpu] == cur) {
1524                 kprintf("PURGE CONFLICT\n");
1525                 purge_cur[cpu] = TAILQ_NEXT(purge_cur[cpu], entry_list);
1526         }
1527         TAILQ_REMOVE(&state_list[cpu], cur, entry_list);
1528         if (cur->tag)
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);
1533 }
1534
1535 int
1536 pf_purge_expired_states(u_int32_t maxcheck, int waslocked)
1537 {
1538         struct pf_state         *cur;
1539         int locked = waslocked;
1540         int cpu = mycpu->gd_cpuid;
1541
1542         while (maxcheck--) {
1543                 /*
1544                  * Wrap to start of list when we hit the end
1545                  */
1546                 cur = purge_cur[cpu];
1547                 if (cur == NULL) {
1548                         cur = TAILQ_FIRST(&state_list[cpu]);
1549                         if (cur == NULL)
1550                                 break;  /* list empty */
1551                 }
1552
1553                 /*
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.
1557                  */
1558                 purge_cur[cpu] = TAILQ_NEXT(cur, entry_list);
1559
1560                 if (cur->timeout == PFTM_UNLINKED) {
1561                         /* free unlinked state */
1562                         if (! locked) {
1563                                 lockmgr(&pf_consistency_lock, LK_EXCLUSIVE);
1564                                 locked = 1;
1565                         }
1566                         pf_free_state(cur);
1567                 } else if (pf_state_expires(cur) <= time_second) {
1568                         /* unlink and free expired state */
1569                         pf_unlink_state(cur);
1570                         if (! locked) {
1571                                 if (!lockmgr(&pf_consistency_lock, LK_EXCLUSIVE))
1572                                         return (0);
1573                                 locked = 1;
1574                         }
1575                         pf_free_state(cur);
1576                 }
1577         }
1578
1579         if (locked)
1580                 lockmgr(&pf_consistency_lock, LK_RELEASE);
1581         return (1);
1582 }
1583
1584 int
1585 pf_tbladdr_setup(struct pf_ruleset *rs, struct pf_addr_wrap *aw)
1586 {
1587         if (aw->type != PF_ADDR_TABLE)
1588                 return (0);
1589         if ((aw->p.tbl = pfr_attach_table(rs, aw->v.tblname)) == NULL)
1590                 return (1);
1591         return (0);
1592 }
1593
1594 void
1595 pf_tbladdr_remove(struct pf_addr_wrap *aw)
1596 {
1597         if (aw->type != PF_ADDR_TABLE || aw->p.tbl == NULL)
1598                 return;
1599         pfr_detach_table(aw->p.tbl);
1600         aw->p.tbl = NULL;
1601 }
1602
1603 void
1604 pf_tbladdr_copyout(struct pf_addr_wrap *aw)
1605 {
1606         struct pfr_ktable *kt = aw->p.tbl;
1607
1608         if (aw->type != PF_ADDR_TABLE || kt == NULL)
1609                 return;
1610         if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL)
1611                 kt = kt->pfrkt_root;
1612         aw->p.tbl = NULL;
1613         aw->p.tblcnt = (kt->pfrkt_flags & PFR_TFLAG_ACTIVE) ?
1614                 kt->pfrkt_cnt : -1;
1615 }
1616
1617 void
1618 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1619 {
1620         switch (af) {
1621 #ifdef INET
1622         case AF_INET: {
1623                 u_int32_t a = ntohl(addr->addr32[0]);
1624                 kprintf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1625                     (a>>8)&255, a&255);
1626                 if (p) {
1627                         p = ntohs(p);
1628                         kprintf(":%u", p);
1629                 }
1630                 break;
1631         }
1632 #endif /* INET */
1633 #ifdef INET6
1634         case AF_INET6: {
1635                 u_int16_t b;
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)
1641                                         curstart = i;
1642                                 else
1643                                         curend = i;
1644                         } else {
1645                                 if (curstart) {
1646                                         if ((curend - curstart) >
1647                                             (maxend - maxstart)) {
1648                                                 maxstart = curstart;
1649                                                 maxend = curend;
1650                                                 curstart = 255;
1651                                         }
1652                                 }
1653                         }
1654                 }
1655                 for (i = 0; i < 8; i++) {
1656                         if (i >= maxstart && i <= maxend) {
1657                                 if (maxend != 7) {
1658                                         if (i == maxstart)
1659                                                 kprintf(":");
1660                                 } else {
1661                                         if (i == maxend)
1662                                                 kprintf(":");
1663                                 }
1664                         } else {
1665                                 b = ntohs(addr->addr16[i]);
1666                                 kprintf("%x", b);
1667                                 if (i < 7)
1668                                         kprintf(":");
1669                         }
1670                 }
1671                 if (p) {
1672                         p = ntohs(p);
1673                         kprintf("[%u]", p);
1674                 }
1675                 break;
1676         }
1677 #endif /* INET6 */
1678         }
1679 }
1680
1681 void
1682 pf_print_state(struct pf_state *s)
1683 {
1684         pf_print_state_parts(s, NULL, NULL);
1685 }
1686
1687 void
1688 pf_print_state_parts(struct pf_state *s,
1689     struct pf_state_key *skwp, struct pf_state_key *sksp)
1690 {
1691         struct pf_state_key *skw, *sks;
1692         u_int8_t proto, dir;
1693
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;
1699
1700         switch (proto) {
1701         case IPPROTO_TCP:
1702                 kprintf("TCP ");
1703                 break;
1704         case IPPROTO_UDP:
1705                 kprintf("UDP ");
1706                 break;
1707         case IPPROTO_ICMP:
1708                 kprintf("ICMP ");
1709                 break;
1710         case IPPROTO_ICMPV6:
1711                 kprintf("ICMPV6 ");
1712                 break;
1713         default:
1714                 kprintf("%u ", skw->proto);
1715                 break;
1716         }
1717         switch (dir) {
1718         case PF_IN:
1719                 kprintf(" in");
1720                 break;
1721         case PF_OUT:
1722                 kprintf(" out");
1723                 break;
1724         }
1725         if (skw) {
1726                 kprintf(" wire: ");
1727                 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1728                 kprintf(" ");
1729                 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1730         }
1731         if (sks) {
1732                 kprintf(" stack: ");
1733                 if (sks != skw) {
1734                         pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1735                         kprintf(" ");
1736                         pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1737                 } else
1738                         kprintf("-");
1739         }
1740         if (s) {
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);
1748                         kprintf("]");
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);
1755                         kprintf("]");
1756                 }
1757                 kprintf(" %u:%u", s->src.state, s->dst.state);
1758         }
1759 }
1760
1761 void
1762 pf_print_flags(u_int8_t f)
1763 {
1764         if (f)
1765                 kprintf(" ");
1766         if (f & TH_FIN)
1767                 kprintf("F");
1768         if (f & TH_SYN)
1769                 kprintf("S");
1770         if (f & TH_RST)
1771                 kprintf("R");
1772         if (f & TH_PUSH)
1773                 kprintf("P");
1774         if (f & TH_ACK)
1775                 kprintf("A");
1776         if (f & TH_URG)
1777                 kprintf("U");
1778         if (f & TH_ECE)
1779                 kprintf("E");
1780         if (f & TH_CWR)
1781                 kprintf("W");
1782 }
1783
1784 #define PF_SET_SKIP_STEPS(i)                                    \
1785         do {                                                    \
1786                 while (head[i] != cur) {                        \
1787                         head[i]->skip[i].ptr = cur;             \
1788                         head[i] = TAILQ_NEXT(head[i], entries); \
1789                 }                                               \
1790         } while (0)
1791
1792 void
1793 pf_calc_skip_steps(struct pf_rulequeue *rules)
1794 {
1795         struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1796         int i;
1797
1798         cur = TAILQ_FIRST(rules);
1799         prev = cur;
1800         for (i = 0; i < PF_SKIP_COUNT; ++i)
1801                 head[i] = cur;
1802         while (cur != NULL) {
1803
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);
1826
1827                 prev = cur;
1828                 cur = TAILQ_NEXT(cur, entries);
1829         }
1830         for (i = 0; i < PF_SKIP_COUNT; ++i)
1831                 PF_SET_SKIP_STEPS(i);
1832 }
1833
1834 int
1835 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
1836 {
1837         if (aw1->type != aw2->type)
1838                 return (1);
1839         switch (aw1->type) {
1840         case PF_ADDR_ADDRMASK:
1841         case PF_ADDR_RANGE:
1842                 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0))
1843                         return (1);
1844                 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0))
1845                         return (1);
1846                 return (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:
1851                 return (0);
1852         case PF_ADDR_TABLE:
1853                 return (aw1->p.tbl != aw2->p.tbl);
1854         case PF_ADDR_RTLABEL:
1855                 return (aw1->v.rtlabel != aw2->v.rtlabel);
1856         default:
1857                 kprintf("invalid address type: %d\n", aw1->type);
1858                 return (1);
1859         }
1860 }
1861
1862 u_int16_t
1863 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
1864 {
1865         u_int32_t       l;
1866
1867         if (udp && !cksum)
1868                 return (0x0000);
1869         l = cksum + old - new;
1870         l = (l >> 16) + (l & 65535);
1871         l = l & 65535;
1872         if (udp && !l)
1873                 return (0xFFFF);
1874         return (l);
1875 }
1876
1877 void
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)
1880 {
1881         struct pf_addr  ao;
1882         u_int16_t       po = *p;
1883
1884         PF_ACPY(&ao, a, af);
1885         PF_ACPY(a, an, af);
1886
1887         *p = pn;
1888
1889         switch (af) {
1890 #ifdef INET
1891         case AF_INET:
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);
1895                 *p = pn;
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),
1899                     po, pn, u);
1900                 break;
1901 #endif /* INET */
1902 #ifdef INET6
1903         case AF_INET6:
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),
1915                     po, pn, u);
1916                 break;
1917 #endif /* INET6 */
1918         }
1919 }
1920
1921
1922 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
1923 void
1924 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
1925 {
1926         u_int32_t       ao;
1927
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);
1932 }
1933
1934 #ifdef INET6
1935 void
1936 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
1937 {
1938         struct pf_addr  ao;
1939
1940         PF_ACPY(&ao, a, AF_INET6);
1941         PF_ACPY(a, an, AF_INET6);
1942
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);
1954 }
1955 #endif /* INET6 */
1956
1957 void
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)
1961 {
1962         struct pf_addr  oia, ooa;
1963
1964         PF_ACPY(&oia, ia, af);
1965         if (oa)
1966                 PF_ACPY(&ooa, oa, af);
1967
1968         /* Change inner protocol port, fix inner protocol checksum. */
1969         if (ip != NULL) {
1970                 u_int16_t       oip = *ip;
1971                 u_int32_t       opc = 0;
1972
1973                 if (pc != NULL)
1974                         opc = *pc;
1975                 *ip = np;
1976                 if (pc != NULL)
1977                         *pc = pf_cksum_fixup(*pc, oip, *ip, u);
1978                 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
1979                 if (pc != NULL)
1980                         *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
1981         }
1982         /* Change inner ip address, fix inner ip and icmp checksums. */
1983         PF_ACPY(ia, na, af);
1984         switch (af) {
1985 #ifdef INET
1986         case AF_INET: {
1987                 u_int32_t        oh2c = *h2c;
1988
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);
1996                 break;
1997         }
1998 #endif /* INET */
1999 #ifdef INET6
2000         case AF_INET6:
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);
2012                 break;
2013 #endif /* INET6 */
2014         }
2015         /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2016         if (oa) {
2017                 PF_ACPY(oa, na, af);
2018                 switch (af) {
2019 #ifdef INET
2020                 case AF_INET:
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);
2024                         break;
2025 #endif /* INET */
2026 #ifdef INET6
2027                 case AF_INET6:
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);
2039                         break;
2040 #endif /* INET6 */
2041                 }
2042         }
2043 }
2044
2045
2046 /*
2047  * Need to modulate the sequence numbers in the TCP SACK option
2048  * (credits to Krzysztof Pfaff for report and patch)
2049  */
2050 int
2051 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2052     struct tcphdr *th, struct pf_state_peer *dst)
2053 {
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;
2058
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))
2062                 return 0;
2063
2064         while (hlen >= TCPOLEN_SACKLEN) {
2065                 olen = opt[1];
2066                 switch (*opt) {
2067                 case TCPOPT_EOL:        /* FALLTHROUGH */
2068                 case TCPOPT_NOP:
2069                         opt++;
2070                         hlen--;
2071                         break;
2072                 case TCPOPT_SACK:
2073                         if (olen > hlen)
2074                                 olen = hlen;
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) -
2081                                             dst->seqdiff), 0);
2082                                         pf_change_a(&sack.rblk_end, &th->th_sum,
2083                                             htonl(ntohl(sack.rblk_end) -
2084                                             dst->seqdiff), 0);
2085                                         memcpy(&opt[i], &sack, sizeof(sack));
2086                                 }
2087                                 copyback = 1;
2088                         }
2089                         /* FALLTHROUGH */
2090                 default:
2091                         if (olen < 2)
2092                                 olen = 2;
2093                         hlen -= olen;
2094                         opt += olen;
2095                 }
2096         }
2097
2098         if (copyback)
2099                 m_copyback(m, off + sizeof(*th), thoptlen, opts);
2100         return (copyback);
2101 }
2102
2103 void
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)
2109 {
2110         struct mbuf     *m;
2111         int              len = 0, tlen;
2112 #ifdef INET
2113         struct ip       *h = NULL;
2114 #endif /* INET */
2115 #ifdef INET6
2116         struct ip6_hdr  *h6 = NULL;
2117 #endif /* INET6 */
2118         struct tcphdr   *th = NULL;
2119         char            *opt;
2120
2121         ASSERT_LWKT_TOKEN_HELD(&pf_token);
2122
2123         /* maximum segment size tcp option */
2124         tlen = sizeof(struct tcphdr);
2125         if (mss)
2126                 tlen += 4;
2127
2128         switch (af) {
2129 #ifdef INET
2130         case AF_INET:
2131                 len = sizeof(struct ip) + tlen;
2132                 break;
2133 #endif /* INET */
2134 #ifdef INET6
2135         case AF_INET6:
2136                 len = sizeof(struct ip6_hdr) + tlen;
2137                 break;
2138 #endif /* INET6 */
2139         }
2140
2141         /*
2142          * Create outgoing mbuf.
2143          *
2144          * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
2145          * so make sure pf.flags is clear.
2146          */
2147         m = m_gethdr(M_NOWAIT, MT_HEADER);
2148         if (m == NULL) {
2149                 return;
2150         }
2151         if (tag)
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;
2159
2160 #ifdef ALTQ
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 *);
2166         }
2167 #endif /* ALTQ */
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);
2172         switch (af) {
2173 #ifdef INET
2174         case AF_INET:
2175                 h = mtod(m, struct ip *);
2176
2177                 /* IP header fields included in the TCP checksum */
2178                 h->ip_p = IPPROTO_TCP;
2179                 h->ip_len = tlen;
2180                 h->ip_src.s_addr = saddr->v4.s_addr;
2181                 h->ip_dst.s_addr = daddr->v4.s_addr;
2182
2183                 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2184                 break;
2185 #endif /* INET */
2186 #ifdef INET6
2187         case AF_INET6:
2188                 h6 = mtod(m, struct ip6_hdr *);
2189
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));
2195
2196                 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2197                 break;
2198 #endif /* INET6 */
2199         }
2200
2201         /* TCP header */
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);
2209
2210         if (mss) {
2211                 opt = (char *)(th + 1);
2212                 opt[0] = TCPOPT_MAXSEG;
2213                 opt[1] = 4;
2214                 mss = htons(mss);
2215                 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2216         }
2217
2218         switch (af) {
2219 #ifdef INET
2220         case AF_INET:
2221                 /* TCP checksum */
2222                 th->th_sum = in_cksum(m, len);
2223
2224                 /* Finish the IP header */
2225                 h->ip_v = 4;
2226                 h->ip_hl = sizeof(*h) >> 2;
2227                 h->ip_tos = IPTOS_LOWDELAY;
2228                 h->ip_len = len;
2229                 h->ip_off = path_mtu_discovery ? IP_DF : 0;
2230                 h->ip_ttl = ttl ? ttl : ip_defttl;
2231                 h->ip_sum = 0;
2232                 if (eh == NULL) {
2233                         lwkt_reltoken(&pf_token);
2234                         ip_output(m, NULL, NULL, 0, NULL, NULL);
2235                         lwkt_gettoken(&pf_token);
2236                 } else {
2237                         struct route             ro;
2238                         struct rtentry           rt;
2239                         struct ether_header     *e = (void *)ro.ro_dst.sa_data;
2240
2241                         if (ifp == NULL) {
2242                                 m_freem(m);
2243                                 return;
2244                         }
2245                         rt.rt_ifp = ifp;
2246                         ro.ro_rt = &rt;
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);
2256                 }
2257                 break;
2258 #endif /* INET */
2259 #ifdef INET6
2260         case AF_INET6:
2261                 /* TCP checksum */
2262                 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2263                     sizeof(struct ip6_hdr), tlen);
2264
2265                 h6->ip6_vfc |= IPV6_VERSION;
2266                 h6->ip6_hlim = IPV6_DEFHLIM;
2267
2268                 lwkt_reltoken(&pf_token);
2269                 ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL);
2270                 lwkt_gettoken(&pf_token);
2271                 break;
2272 #endif /* INET6 */
2273         }
2274 }
2275
2276 void
2277 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2278     struct pf_rule *r)
2279 {
2280         struct mbuf     *m0;
2281
2282         /*
2283          * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
2284          * so make sure pf.flags is clear.
2285          */
2286         if ((m0 = m_copy(m, 0, M_COPYALL)) == NULL)
2287                 return;
2288
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;
2293
2294         if (r->rtableid >= 0)
2295                 m0->m_pkthdr.pf.rtableid = r->rtableid;
2296
2297 #ifdef ALTQ
2298         if (r->qid) {
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 *);
2303         }
2304 #endif /* ALTQ */
2305
2306         switch (af) {
2307 #ifdef INET
2308         case AF_INET:
2309                 icmp_error(m0, type, code, 0, 0);
2310                 break;
2311 #endif /* INET */
2312 #ifdef INET6
2313         case AF_INET6:
2314                 icmp6_error(m0, type, code, 0);
2315                 break;
2316 #endif /* INET6 */
2317         }
2318 }
2319
2320 /*
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
2323  * are different.
2324  */
2325 int
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)
2328 {
2329         int     match = 0;
2330
2331         switch (af) {
2332 #ifdef INET
2333         case AF_INET:
2334                 if ((a->addr32[0] & m->addr32[0]) ==
2335                     (b->addr32[0] & m->addr32[0]))
2336                         match++;
2337                 break;
2338 #endif /* INET */
2339 #ifdef INET6
2340         case AF_INET6:
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])))
2349                         match++;
2350                 break;
2351 #endif /* INET6 */
2352         }
2353         if (match) {
2354                 if (n)
2355                         return (0);
2356                 else
2357                         return (1);
2358         } else {
2359                 if (n)
2360                         return (1);
2361                 else
2362                         return (0);
2363         }
2364 }
2365
2366 /*
2367  * Return 1 if b <= a <= e, otherwise return 0.
2368  */
2369 int
2370 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2371     struct pf_addr *a, sa_family_t af)
2372 {
2373         switch (af) {
2374 #ifdef INET
2375         case AF_INET:
2376                 if ((a->addr32[0] < b->addr32[0]) ||
2377                     (a->addr32[0] > e->addr32[0]))
2378                         return (0);
2379                 break;
2380 #endif /* INET */
2381 #ifdef INET6
2382         case AF_INET6: {
2383                 int     i;
2384
2385                 /* check a >= b */
2386                 for (i = 0; i < 4; ++i)
2387                         if (a->addr32[i] > b->addr32[i])
2388                                 break;
2389                         else if (a->addr32[i] < b->addr32[i])
2390                                 return (0);
2391                 /* check a <= e */
2392                 for (i = 0; i < 4; ++i)
2393                         if (a->addr32[i] < e->addr32[i])
2394                                 break;
2395                         else if (a->addr32[i] > e->addr32[i])
2396                                 return (0);
2397                 break;
2398         }
2399 #endif /* INET6 */
2400         }
2401         return (1);
2402 }
2403
2404 int
2405 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2406 {
2407         switch (op) {
2408         case PF_OP_IRG:
2409                 return ((p > a1) && (p < a2));
2410         case PF_OP_XRG:
2411                 return ((p < a1) || (p > a2));
2412         case PF_OP_RRG:
2413                 return ((p >= a1) && (p <= a2));
2414         case PF_OP_EQ:
2415                 return (p == a1);
2416         case PF_OP_NE:
2417                 return (p != a1);
2418         case PF_OP_LT:
2419                 return (p < a1);
2420         case PF_OP_LE:
2421                 return (p <= a1);
2422         case PF_OP_GT:
2423                 return (p > a1);
2424         case PF_OP_GE:
2425                 return (p >= a1);
2426         }
2427         return (0); /* never reached */
2428 }
2429
2430 int
2431 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2432 {
2433         a1 = ntohs(a1);
2434         a2 = ntohs(a2);
2435         p = ntohs(p);
2436         return (pf_match(op, a1, a2, p));
2437 }
2438
2439 int
2440 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2441 {
2442         if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2443                 return (0);
2444         return (pf_match(op, a1, a2, u));
2445 }
2446
2447 int
2448 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2449 {
2450         if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2451                 return (0);
2452         return (pf_match(op, a1, a2, g));
2453 }
2454
2455 int
2456 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag)
2457 {
2458         if (*tag == -1)
2459                 *tag = m->m_pkthdr.pf.tag;
2460
2461         return ((!r->match_tag_not && r->match_tag == *tag) ||
2462             (r->match_tag_not && r->match_tag != *tag));
2463 }
2464
2465 int
2466 pf_tag_packet(struct mbuf *m, int tag, int rtableid)
2467 {
2468         if (tag <= 0 && rtableid < 0)
2469                 return (0);
2470
2471         if (tag > 0)
2472                 m->m_pkthdr.pf.tag = tag;
2473         if (rtableid >= 0)
2474                 m->m_pkthdr.pf.rtableid = rtableid;
2475
2476         return (0);
2477 }
2478
2479 void
2480 pf_step_into_anchor(int *depth, struct pf_ruleset **rs, int n,
2481     struct pf_rule **r, struct pf_rule **a, int *match)
2482 {
2483         struct pf_anchor_stackframe     *f;
2484
2485         (*r)->anchor->match = 0;
2486         if (match)
2487                 *match = 0;
2488         if (*depth >= NELEM(pf_anchor_stack)) {
2489                 kprintf("pf_step_into_anchor: stack overflow\n");
2490                 *r = TAILQ_NEXT(*r, entries);
2491                 return;
2492         } else if (*depth == 0 && a != NULL)
2493                 *a = *r;
2494         f = pf_anchor_stack + (*depth)++;
2495         f->rs = *rs;
2496         f->r = *r;
2497         if ((*r)->anchor_wildcard) {
2498                 f->parent = &(*r)->anchor->children;
2499                 if ((f->child = RB_MIN(pf_anchor_node, f->parent)) ==
2500                     NULL) {
2501                         *r = NULL;
2502                         return;
2503                 }
2504                 *rs = &f->child->ruleset;
2505         } else {
2506                 f->parent = NULL;
2507                 f->child = NULL;
2508                 *rs = &(*r)->anchor->ruleset;
2509         }
2510         *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2511 }
2512
2513 int
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)
2516 {
2517         struct pf_anchor_stackframe     *f;
2518         int quick = 0;
2519
2520         do {
2521                 if (*depth <= 0)
2522                         break;
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;
2528                                 *match = 0;
2529                         }
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);
2534                                 if (*r == NULL)
2535                                         continue;
2536                                 else
2537                                         break;
2538                         }
2539                 }
2540                 (*depth)--;
2541                 if (*depth == 0 && a != NULL)
2542                         *a = NULL;
2543                 *rs = f->rs;
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);
2548
2549         return (quick);
2550 }
2551
2552 #ifdef INET6
2553 void
2554 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2555     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2556 {
2557         switch (af) {
2558 #ifdef INET
2559         case AF_INET:
2560                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2561                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2562                 break;
2563 #endif /* INET */
2564         case AF_INET6:
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]);
2573                 break;
2574         }
2575 }
2576
2577 void
2578 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2579 {
2580         switch (af) {
2581 #ifdef INET
2582         case AF_INET:
2583                 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2584                 break;
2585 #endif /* INET */
2586         case AF_INET6:
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;
2593                                         addr->addr32[0] =
2594                                             htonl(ntohl(addr->addr32[0]) + 1);
2595                                 } else
2596                                         addr->addr32[1] =
2597                                             htonl(ntohl(addr->addr32[1]) + 1);
2598                         } else
2599                                 addr->addr32[2] =
2600                                     htonl(ntohl(addr->addr32[2]) + 1);
2601                 } else
2602                         addr->addr32[3] =
2603                             htonl(ntohl(addr->addr32[3]) + 1);
2604                 break;
2605         }
2606 }
2607 #endif /* INET6 */
2608
2609 #define mix(a,b,c) \
2610         do {                                    \
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); \
2620         } while (0)
2621
2622 /*
2623  * hash function based on bridge_hash in if_bridge.c
2624  */
2625 void
2626 pf_hash(struct pf_addr *inaddr, struct pf_addr *hash,
2627     struct pf_poolhashkey *key, sa_family_t af)
2628 {
2629         u_int32_t       a = 0x9e3779b9, b = 0x9e3779b9, c = key->key32[0];
2630
2631         switch (af) {
2632 #ifdef INET
2633         case AF_INET:
2634                 a += inaddr->addr32[0];
2635                 b += key->key32[1];
2636                 mix(a, b, c);
2637                 hash->addr32[0] = c + key->key32[2];
2638                 break;
2639 #endif /* INET */
2640 #ifdef INET6
2641         case AF_INET6:
2642                 a += inaddr->addr32[0];
2643                 b += inaddr->addr32[2];
2644                 mix(a, b, c);
2645                 hash->addr32[0] = c;
2646                 a += inaddr->addr32[1];
2647                 b += inaddr->addr32[3];
2648                 c += key->key32[1];
2649                 mix(a, b, c);
2650                 hash->addr32[1] = c;
2651                 a += inaddr->addr32[2];
2652                 b += inaddr->addr32[1];
2653                 c += key->key32[2];
2654                 mix(a, b, c);
2655                 hash->addr32[2] = c;
2656                 a += inaddr->addr32[3];
2657                 b += inaddr->addr32[0];
2658                 c += key->key32[3];
2659                 mix(a, b, c);
2660                 hash->addr32[3] = c;
2661                 break;
2662 #endif /* INET6 */
2663         }
2664 }
2665
2666 int
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)
2669 {
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;
2679         int tblidx;
2680
2681         bzero(hash, sizeof(hash));      /* avoid gcc warnings */
2682
2683         /*
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.
2688          */
2689         cpu_ccfence();
2690         cur = acur;
2691         raddr = &cur->addr.v.a.addr;
2692         rmask = &cur->addr.v.a.mask;
2693
2694         if (*sn == NULL && r->rpool.opts & PF_POOL_STICKYADDR &&
2695             (r->rpool.opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) {
2696                 k.af = af;
2697                 PF_ACPY(&k.addr, saddr, af);
2698                 if (r->rule_flag & PFRULE_RULESRCTRACK ||
2699                     r->rpool.opts & PF_POOL_STICKYADDR)
2700                         k.rule.ptr = r;
2701                 else
2702                         k.rule.ptr = NULL;
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);
2710                                 kprintf(" to ");
2711                                 pf_print_host(naddr, 0, af);
2712                                 kprintf("\n");
2713                         }
2714                         return (0);
2715                 }
2716         }
2717
2718         if (cur->addr.type == PF_ADDR_NOROUTE)
2719                 return (1);
2720         if (cur->addr.type == PF_ADDR_DYNIFTL) {
2721                 switch (af) {
2722 #ifdef INET
2723                 case AF_INET:
2724                         if (cur->addr.p.dyn->pfid_acnt4 < 1 &&
2725                             (rpool->opts & PF_POOL_TYPEMASK) !=
2726                             PF_POOL_ROUNDROBIN)
2727                                 return (1);
2728                         raddr = &cur->addr.p.dyn->pfid_addr4;
2729                         rmask = &cur->addr.p.dyn->pfid_mask4;
2730                         break;
2731 #endif /* INET */
2732 #ifdef INET6
2733                 case AF_INET6:
2734                         if (cur->addr.p.dyn->pfid_acnt6 < 1 &&
2735                             (rpool->opts & PF_POOL_TYPEMASK) !=
2736                             PF_POOL_ROUNDROBIN)
2737                                 return (1);
2738                         raddr = &cur->addr.p.dyn->pfid_addr6;
2739                         rmask = &cur->addr.p.dyn->pfid_mask6;
2740                         break;
2741 #endif /* INET6 */
2742                 }
2743         } else if (cur->addr.type == PF_ADDR_TABLE) {
2744                 if ((rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_ROUNDROBIN)
2745                         return (1); /* unsupported */
2746         } else {
2747                 raddr = &cur->addr.v.a.addr;
2748                 rmask = &cur->addr.v.a.mask;
2749         }
2750
2751         switch (rpool->opts & PF_POOL_TYPEMASK) {
2752         case PF_POOL_NONE:
2753                 PF_ACPY(naddr, raddr, af);
2754                 break;
2755         case PF_POOL_BITMASK:
2756                 PF_POOLMASK(naddr, raddr, rmask, saddr, af);
2757                 break;
2758         case PF_POOL_RANDOM:
2759                 if (init_addr != NULL && PF_AZERO(init_addr, af)) {
2760                         switch (af) {
2761 #ifdef INET
2762                         case AF_INET:
2763                                 counter.addr32[0] = htonl(karc4random());
2764                                 break;
2765 #endif /* INET */
2766 #ifdef INET6
2767                         case AF_INET6:
2768                                 if (rmask->addr32[3] != 0xffffffff)
2769                                         counter.addr32[3] =
2770                                                 htonl(karc4random());
2771                                 else
2772                                         break;
2773                                 if (rmask->addr32[2] != 0xffffffff)
2774                                         counter.addr32[2] =
2775                                                 htonl(karc4random());
2776                                 else
2777                                         break;
2778                                 if (rmask->addr32[1] != 0xffffffff)
2779                                         counter.addr32[1] =
2780                                                 htonl(karc4random());
2781                                 else
2782                                         break;
2783                                 if (rmask->addr32[0] != 0xffffffff)
2784                                         counter.addr32[0] =
2785                                                 htonl(karc4random());
2786                                 break;
2787 #endif /* INET6 */
2788                         }
2789                         PF_POOLMASK(naddr, raddr, rmask, &counter, af);
2790                         PF_ACPY(init_addr, naddr, af);
2791
2792                 } else {
2793                         counter = rpool->counter;
2794                         cpu_ccfence();
2795                         PF_AINC(&counter, af);
2796                         PF_POOLMASK(naddr, raddr, rmask, &counter, af);
2797                         rpool->counter = counter;
2798                 }
2799                 break;
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);
2803                 break;
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,
2809                             &tblidx, &counter,
2810                             &raddr, &rmask, af)) {
2811                                 goto get_addr;
2812                         }
2813                 } else if (cur->addr.type == PF_ADDR_DYNIFTL) {
2814                         if (!pfr_pool_get(cur->addr.p.dyn->pfid_kt,
2815                             &tblidx, &counter,
2816                             &raddr, &rmask, af)) {
2817                                 goto get_addr;
2818                         }
2819                 } else if (pf_match_addr(0, raddr, rmask,
2820                                          &counter, af)) {
2821                         goto get_addr;
2822                 }
2823
2824         try_next:
2825                 if ((cur = TAILQ_NEXT(cur, entries)) == NULL)
2826                         cur = TAILQ_FIRST(&rpool->list);
2827                 if (cur->addr.type == PF_ADDR_TABLE) {
2828                         tblidx = -1;
2829                         if (pfr_pool_get(cur->addr.p.tbl,
2830                             &tblidx, &counter,
2831                             &raddr, &rmask, af)) {
2832                                 /* table contains no address of type 'af' */
2833                                 if (cur != acur)
2834                                         goto try_next;
2835                                 return (1);
2836                         }
2837                 } else if (cur->addr.type == PF_ADDR_DYNIFTL) {
2838                         tblidx = -1;
2839                         if (pfr_pool_get(cur->addr.p.dyn->pfid_kt,
2840                             &tblidx, &counter,
2841                             &raddr, &rmask, af)) {
2842                                 /* table contains no address of type 'af' */
2843                                 if (cur != acur)
2844                                         goto try_next;
2845                                 return (1);
2846                         }
2847                 } else {
2848                         raddr = &cur->addr.v.a.addr;
2849                         rmask = &cur->addr.v.a.mask;
2850                         PF_ACPY(&counter, raddr, af);
2851                 }
2852
2853         get_addr:
2854                 rpool->cur = cur;
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;
2861                 break;
2862         }
2863         if (*sn != NULL)
2864                 PF_ACPY(&(*sn)->raddr, naddr, af);
2865
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);
2870                 kprintf("\n");
2871         }
2872
2873         return (0);
2874 }
2875
2876 int
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)
2883 {
2884         struct pf_state_key_cmp key;
2885         struct pf_addr          init_addr;
2886         u_int16_t               cut;
2887         u_int32_t               hash_base = 0;
2888         int                     do_hash = 0;
2889
2890         bzero(&init_addr, sizeof(init_addr));
2891         if (pf_map_addr(af, r, saddr, naddr, &init_addr, sn))
2892                 return (1);
2893
2894         if (proto == IPPROTO_ICMP) {
2895                 low = 1;
2896                 high = 65535;
2897         }
2898
2899         bzero(&key, sizeof(key));
2900         key.af = af;
2901         key.proto = proto;
2902         key.port[0] = dport;
2903         PF_ACPY(&key.addr[0], daddr, key.af);
2904
2905         do {
2906                 PF_ACPY(&key.addr[1], naddr, key.af);
2907
2908                 /*
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.
2912                  *
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.
2916                  */
2917                 switch(af) {
2918                 case AF_INET:
2919                         if (proto == IPPROTO_TCP) {
2920                                 do_hash = 1;
2921                                 hash_base = toeplitz_piecemeal_port(dport) ^
2922                                     toeplitz_piecemeal_addr(daddr->v4.s_addr) ^
2923                                     toeplitz_piecemeal_addr(naddr->v4.s_addr);
2924                         }
2925                         break;
2926                 case AF_INET6:
2927                         /* XXX TODO XXX */
2928                 default:
2929                         /* XXX TODO XXX */
2930                         break;
2931                 }
2932
2933                 /*
2934                  * port search; start random, step;
2935                  * similar 2 portloop in in_pcbbind
2936                  *
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.
2940                  *
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
2945                  *
2946                  * XXX fixed ports present a problem for cpu localization.
2947                  */
2948                 if (!(proto == IPPROTO_TCP ||
2949                       proto == IPPROTO_UDP ||
2950                       proto == IPPROTO_ICMP)) {
2951                         /*
2952                          * non-specific protocol, leave port intact.
2953                          */
2954                         key.port[1] = sport;
2955                         if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
2956                                 *nport = sport;
2957                                 pd->not_cpu_localized = 1;
2958                                 return (0);
2959                         }
2960                 } else if (low == 0 && high == 0) {
2961                         /*
2962                          * static-port same as originator.
2963                          */
2964                         key.port[1] = sport;
2965                         if (pf_find_state_all(&key, PF_IN, NULL) == NULL) {
2966                                 *nport = sport;
2967                                 pd->not_cpu_localized = 1;
2968                                 return (0);
2969                         }
2970                 } else if (low == high) {
2971                         /*
2972                          * specific port as specified.
2973                          */
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;
2978                                 return (0);
2979                         }
2980                 } else {
2981                         /*
2982                          * normal dynamic port
2983                          */
2984                         u_int16_t tmp;
2985
2986                         if (low > high) {
2987                                 tmp = low;
2988                                 low = high;
2989                                 high = tmp;
2990                         }
2991                         /* low < high */
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);
2996                                 if (do_hash) {
2997                                         uint32_t hash;
2998
2999                                         hash = hash_base ^
3000                                         toeplitz_piecemeal_port(key.port[1]);
3001                                         if (netisr_hashcpu(hash) != mycpuid)
3002                                                 continue;
3003                                 }
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);
3009                                         return (0);
3010                                 }
3011                         }
3012                         for (tmp = cut - 1; tmp >= low; --(tmp)) {
3013                                 key.port[1] = htons(tmp);
3014                                 if (do_hash) {
3015                                         uint32_t hash;
3016
3017                                         hash = hash_base ^
3018                                         toeplitz_piecemeal_port(key.port[1]);
3019                                         if (netisr_hashcpu(hash) != mycpuid)
3020                                                 continue;
3021                                 }
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);
3027                                         return (0);
3028                                 }
3029                         }
3030                 }
3031
3032                 /*
3033                  * Next address
3034                  */
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))
3039                                 return (1);
3040                         break;
3041                 case PF_POOL_NONE:
3042                 case PF_POOL_SRCHASH:
3043                 case PF_POOL_BITMASK:
3044                 default:
3045                         return (1);
3046                 }
3047         } while (! PF_AEQ(&init_addr, naddr, af) );
3048         return (1);                                     /* none available */
3049 }
3050
3051 struct pf_rule *
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)
3055 {
3056         struct pf_rule          *r, *rm = NULL;
3057         struct pf_ruleset       *ruleset = NULL;
3058         int                      tag = -1;
3059         int                      rtableid = -1;
3060         int                      asd = 0;
3061
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;
3067
3068                 if (r->action == PF_BINAT && direction == PF_IN) {
3069                         src = &r->dst;
3070                         cur = r->rpool.cur;     /* SMP race possible */
3071                         cpu_ccfence();
3072                         if (cur)
3073                                 xdst = &cur->addr;
3074                 } else {
3075                         src = &r->src;
3076                         dst = &r->dst;
3077                 }
3078
3079                 r->evaluations++;
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,
3089                     src->neg, kif))
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,
3100                     0, NULL))
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);
3112                 else {
3113                         if (r->tag)
3114                                 tag = r->tag;
3115                         if (r->rtableid >= 0)
3116                                 rtableid = r->rtableid;
3117                         if (r->anchor == NULL) {
3118                                 rm = r;
3119                         } else
3120                                 pf_step_into_anchor(&asd, &ruleset, rs_num,
3121                                     &r, NULL, NULL);
3122                 }
3123                 if (r == NULL)
3124                         pf_step_out_of_anchor(&asd, &ruleset, rs_num, &r,
3125                             NULL, NULL);
3126         }
3127         if (pf_tag_packet(m, tag, rtableid))
3128                 return (NULL);
3129         if (rm != NULL && (rm->action == PF_NONAT ||
3130             rm->action == PF_NORDR || rm->action == PF_NOBINAT))
3131                 return (NULL);
3132         return (rm);
3133 }
3134
3135 struct pf_rule *
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)
3142 {
3143         struct pf_rule  *r = NULL;
3144
3145         if (direction == PF_OUT) {
3146                 r = pf_match_translation(pd, m, off, direction, kif, saddr,
3147                     sport, daddr, dport, PF_RULESET_BINAT);
3148                 if (r == NULL)
3149                         r = pf_match_translation(pd, m, off, direction, kif,
3150                             saddr, sport, daddr, dport, PF_RULESET_NAT);
3151         } else {
3152                 r = pf_match_translation(pd, m, off, direction, kif, saddr,
3153                     sport, daddr, dport, PF_RULESET_RDR);
3154                 if (r == NULL)
3155                         r = pf_match_translation(pd, m, off, direction, kif,
3156                             saddr, sport, daddr, dport, PF_RULESET_BINAT);
3157         }
3158
3159         if (r != NULL) {
3160                 struct pf_addr  *naddr;
3161                 u_int16_t       *nport;
3162
3163                 if (pf_state_key_setup(pd, r, skw, sks, skp, nkp,
3164                     saddr, daddr, sport, dport))
3165                         return r;
3166
3167                 /* XXX We only modify one side for now. */
3168                 naddr = &(*nkp)->addr[1];
3169                 nport = &(*nkp)->port[1];
3170
3171                 /*
3172                  * NOTE: Currently all translations will clear
3173                  *       BRIDGE_MBUF_TAGGED, telling the bridge to
3174                  *       ignore the original input encapsulation.
3175                  */
3176                 switch (r->action) {
3177                 case PF_NONAT:
3178                 case PF_NOBINAT:
3179                 case PF_NORDR:
3180                         return (NULL);
3181                 case PF_NAT:
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 "
3189                                     "(%u-%u) failed\n",
3190                                     r->rpool.proxy_port[0],
3191                                     r->rpool.proxy_port[1]));
3192                                 return (NULL);
3193                         }
3194                         break;
3195                 case PF_BINAT:
3196                         m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
3197                         switch (direction) {
3198                         case PF_OUT:
3199                                 if (r->rpool.cur->addr.type == PF_ADDR_DYNIFTL){
3200                                         switch (pd->af) {
3201 #ifdef INET
3202                                         case AF_INET:
3203                                                 if (r->rpool.cur->addr.p.dyn->
3204                                                     pfid_acnt4 < 1)
3205                                                         return (NULL);
3206                                                 PF_POOLMASK(naddr,
3207                                                     &r->rpool.cur->addr.p.dyn->
3208                                                     pfid_addr4,
3209                                                     &r->rpool.cur->addr.p.dyn->
3210                                                     pfid_mask4,
3211                                                     saddr, AF_INET);
3212                                                 break;
3213 #endif /* INET */
3214 #ifdef INET6
3215                                         case AF_INET6:
3216                                                 if (r->rpool.cur->addr.p.dyn->
3217                                                     pfid_acnt6 < 1)
3218                                                         return (NULL);
3219                                                 PF_POOLMASK(naddr,
3220                                                     &r->rpool.cur->addr.p.dyn->
3221                                                     pfid_addr6,
3222                                                     &r->rpool.cur->addr.p.dyn->
3223                                                     pfid_mask6,
3224                                                     saddr, AF_INET6);
3225                                                 break;
3226 #endif /* INET6 */
3227                                         }
3228                                 } else
3229                                         PF_POOLMASK(naddr,
3230                                             &r->rpool.cur->addr.v.a.addr,
3231                                             &r->rpool.cur->addr.v.a.mask,
3232                                             saddr, pd->af);
3233                                 break;
3234                         case PF_IN:
3235                                 if (r->src.addr.type == PF_ADDR_DYNIFTL) {
3236                                         switch (pd->af) {
3237 #ifdef INET
3238                                         case AF_INET:
3239                                                 if (r->src.addr.p.dyn->
3240                                                     pfid_acnt4 < 1)
3241                                                         return (NULL);
3242                                                 PF_POOLMASK(naddr,
3243                                                     &r->src.addr.p.dyn->
3244                                                     pfid_addr4,
3245                                                     &r->src.addr.p.dyn->
3246                                                     pfid_mask4,
3247                                                     daddr, AF_INET);
3248                                                 break;
3249 #endif /* INET */
3250 #ifdef INET6
3251                                         case AF_INET6:
3252                                                 if (r->src.addr.p.dyn->
3253                                                     pfid_acnt6 < 1)
3254                                                         return (NULL);
3255                                                 PF_POOLMASK(naddr,
3256                                                     &r->src.addr.p.dyn->
3257                                                     pfid_addr6,
3258                                                     &r->src.addr.p.dyn->
3259                                                     pfid_mask6,
3260                                                     daddr, AF_INET6);
3261                                                 break;
3262 #endif /* INET6 */
3263                                         }
3264                                 } else
3265                                         PF_POOLMASK(naddr,
3266                                             &r->src.addr.v.a.addr,
3267                                             &r->src.addr.v.a.mask, daddr,
3268                                             pd->af);
3269                                 break;
3270                         }
3271                         break;
3272                 case PF_RDR: {
3273                         m->m_pkthdr.fw_flags &= ~BRIDGE_MBUF_TAGGED;
3274                         if (pf_map_addr(pd->af, r, saddr, naddr, NULL, sn))
3275                                 return (NULL);
3276                         if ((r->rpool.opts & PF_POOL_TYPEMASK) ==
3277                             PF_POOL_BITMASK)
3278                                 PF_POOLMASK(naddr, naddr,
3279                                     &r->rpool.cur->addr.v.a.mask, daddr,
3280                                     pd->af);
3281
3282                         if (r->rpool.proxy_port[1]) {
3283                                 u_int32_t       tmp_nport;
3284
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];
3290
3291                                 /* wrap around if necessary */
3292                                 if (tmp_nport > 65535)
3293                                         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]);
3297                         }
3298                         pd->not_cpu_localized = 1;
3299                         break;
3300                 }
3301                 default:
3302                         return (NULL);
3303                 }
3304         }
3305
3306         return (r);
3307 }
3308
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;
3315         uint16_t                nm_sport;
3316         uint16_t                nm_dport;
3317         sa_family_t             nm_af;
3318 };
3319
3320 #ifdef PF_SOCKET_LOOKUP_DOMSG
3321 static void
3322 in_pcblookup_hash_handler(netmsg_t msg)
3323 {
3324         struct netmsg_hashlookup *rmsg = (struct netmsg_hashlookup *)msg;
3325
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);
3330 #ifdef INET6
3331         else
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);
3335 #endif /* INET6 */
3336         lwkt_replymsg(&rmsg->base.lmsg, 0);
3337 }
3338 #endif  /* PF_SOCKET_LOOKUP_DOMSG */
3339
3340 int
3341 pf_socket_lookup(int direction, struct pf_pdesc *pd)
3342 {
3343         struct pf_addr          *saddr, *daddr;
3344         u_int16_t                sport, dport;
3345         struct inpcbinfo        *pi;
3346         struct inpcb            *inp;
3347         struct netmsg_hashlookup *msg = NULL;
3348 #ifdef PF_SOCKET_LOOKUP_DOMSG
3349         struct netmsg_hashlookup msg0;
3350 #endif
3351         int                      pi_cpu = 0;
3352
3353         if (pd == NULL)
3354                 return (-1);
3355         pd->lookup.uid = UID_MAX;
3356         pd->lookup.gid = GID_MAX;
3357         pd->lookup.pid = NO_PID;
3358         if (direction == PF_IN) {
3359                 saddr = pd->src;
3360                 daddr = pd->dst;
3361         } else {
3362                 saddr = pd->dst;
3363                 daddr = pd->src;
3364         }
3365         switch (pd->proto) {
3366         case IPPROTO_TCP:
3367                 if (pd->hdr.tcp == NULL)
3368                         return (-1);
3369                 sport = pd->hdr.tcp->th_sport;
3370                 dport = pd->hdr.tcp->th_dport;
3371
3372                 pi_cpu = tcp_addrcpu(saddr->v4.s_addr, sport, daddr->v4.s_addr, dport);
3373                 pi = &tcbinfo[pi_cpu];
3374                 /*
3375                  * Our netstack runs lockless on MP systems
3376                  * (only for TCP connections at the moment).
3377                  *
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
3380                  * table for us.
3381                  *
3382                  * Prepare a msg iff data belongs to another CPU.
3383                  */
3384                 if (pi_cpu != mycpu->gd_cpuid) {
3385 #ifdef PF_SOCKET_LOOKUP_DOMSG
3386                         /*
3387                          * NOTE:
3388                          *
3389                          * Following lwkt_domsg() is dangerous and could
3390                          * lockup the network system, e.g.
3391                          *
3392                          * On 2 CPU system:
3393                          * netisr0 domsg to netisr1 (due to lookup)
3394                          * netisr1 domsg to netisr0 (due to lookup)
3395                          *
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.
3399                          */
3400                         msg = &msg0;
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",
3413                                 pi_cpu, mycpuid);
3414                         print_backtrace(-1);
3415                         return -1;
3416 #endif  /* PF_SOCKET_LOOKUP_DOMSG */
3417                 }
3418                 break;
3419         case IPPROTO_UDP:
3420                 if (pd->hdr.udp == NULL)
3421                         return (-1);
3422                 sport = pd->hdr.udp->uh_sport;
3423                 dport = pd->hdr.udp->uh_dport;
3424                 pi = &udbinfo[mycpuid];
3425                 break;
3426         default:
3427                 return (-1);
3428         }
3429         if (direction != PF_IN) {
3430                 u_int16_t       p;
3431
3432                 p = sport;
3433                 sport = dport;
3434                 dport = p;
3435         }
3436         switch (pd->af) {
3437 #ifdef INET6
3438         case AF_INET6:
3439                 /*
3440                  * Query other CPU, second part
3441                  *
3442                  * msg only gets initialized when:
3443                  * 1) packet is TCP
3444                  * 2) the info belongs to another CPU
3445                  *
3446                  * Use some switch/case magic to avoid code duplication.
3447                  */
3448                 if (msg == NULL) {
3449                         inp = in6_pcblookup_hash(pi, &saddr->v6, sport,
3450                             &daddr->v6, dport, INPLOOKUP_WILDCARD, NULL);
3451
3452                         if (inp == NULL)
3453                                 return (-1);
3454                         break;
3455                 }
3456                 /* FALLTHROUGH if SMP and on other CPU */
3457 #endif /* INET6 */
3458         case AF_INET:
3459                 if (msg != NULL) {
3460                         lwkt_domsg(netisr_cpuport(pi_cpu),
3461                                      &msg->base.lmsg, 0);
3462                 } else
3463                 {
3464                         inp = in_pcblookup_hash(pi, saddr->v4, sport, daddr->v4,
3465                             dport, INPLOOKUP_WILDCARD, NULL);
3466                 }
3467                 if (inp == NULL)
3468                         return (-1);
3469                 break;
3470
3471         default:
3472                 return (-1);
3473         }
3474         pd->lookup.uid = inp->inp_socket->so_cred->cr_uid;
3475         pd->lookup.gid = inp->inp_socket->so_cred->cr_groups[0];
3476         return (1);
3477 }
3478
3479 u_int8_t
3480 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3481 {
3482         int              hlen;
3483         u_int8_t         hdr[60];
3484         u_int8_t        *opt, optlen;
3485         u_int8_t         wscale = 0;
3486
3487         hlen = th_off << 2;             /* hlen <= sizeof(hdr) */
3488         if (hlen <= sizeof(struct tcphdr))
3489                 return (0);
3490         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3491                 return (0);
3492         opt = hdr + sizeof(struct tcphdr);
3493         hlen -= sizeof(struct tcphdr);
3494         while (hlen >= 3) {
3495                 switch (*opt) {
3496                 case TCPOPT_EOL:
3497                 case TCPOPT_NOP:
3498                         ++opt;
3499                         --hlen;
3500                         break;
3501                 case TCPOPT_WINDOW:
3502                         wscale = opt[2];
3503                         if (wscale > TCP_MAX_WINSHIFT)
3504                                 wscale = TCP_MAX_WINSHIFT;
3505                         wscale |= PF_WSCALE_FLAG;
3506                         /* FALLTHROUGH */
3507                 default:
3508                         optlen = opt[1];
3509                         if (optlen < 2)
3510                                 optlen = 2;
3511                         hlen -= optlen;
3512                         opt += optlen;
3513                         break;
3514                 }
3515         }
3516         return (wscale);
3517 }
3518
3519 u_int16_t
3520 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3521 {
3522         int              hlen;
3523         u_int8_t         hdr[60];
3524         u_int8_t        *opt, optlen;
3525         u_int16_t        mss = tcp_mssdflt;
3526
3527         hlen = th_off << 2;     /* hlen <= sizeof(hdr) */
3528         if (hlen <= sizeof(struct tcphdr))
3529                 return (0);
3530         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3531                 return (0);
3532         opt = hdr + sizeof(struct tcphdr);
3533         hlen -= sizeof(struct tcphdr);
3534         while (hlen >= TCPOLEN_MAXSEG) {
3535                 switch (*opt) {
3536                 case TCPOPT_EOL:
3537                 case TCPOPT_NOP:
3538                         ++opt;
3539                         --hlen;
3540                         break;
3541                 case TCPOPT_MAXSEG:
3542                         bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
3543                         /* FALLTHROUGH */
3544                 default:
3545                         optlen = opt[1];
3546                         if (optlen < 2)
3547                                 optlen = 2;
3548                         hlen -= optlen;
3549                         opt += optlen;
3550                         break;
3551                 }
3552         }
3553         return (mss);
3554 }
3555
3556 u_int16_t
3557 pf_calc_mss(struct pf_addr *addr, sa_family_t af, u_int16_t offer)
3558 {
3559 #ifdef INET
3560         struct sockaddr_in      *dst;
3561         struct route             ro;
3562 #endif /* INET */
3563 #ifdef INET6
3564         struct sockaddr_in6     *dst6;
3565         struct route_in6         ro6;
3566 #endif /* INET6 */
3567         struct rtentry          *rt = NULL;
3568         int                      hlen = 0;
3569         u_int16_t                mss = tcp_mssdflt;
3570
3571         switch (af) {
3572 #ifdef INET
3573         case AF_INET:
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));
3581                 rt = ro.ro_rt;
3582                 break;
3583 #endif /* INET */
3584 #ifdef INET6
3585         case AF_INET6:
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));
3593                 rt = ro6.ro_rt;
3594                 break;
3595 #endif /* INET6 */
3596         }
3597
3598         if (rt && rt->rt_ifp) {
3599                 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr);
3600                 mss = max(tcp_mssdflt, mss);
3601                 RTFREE(rt);
3602         }
3603         mss = min(mss, offer);
3604         mss = max(mss, 64);             /* sanity - at least max opt space */
3605         return (mss);
3606 }
3607
3608 void
3609 pf_set_rt_ifp(struct pf_state *s, struct pf_addr *saddr)
3610 {
3611         struct pf_rule *r = s->rule.ptr;
3612
3613         s->rt_kif = NULL;
3614         if (!r->rt || r->rt == PF_FASTROUTE)
3615                 return;
3616         switch (s->key[PF_SK_WIRE]->af) {
3617 #ifdef INET
3618         case AF_INET:
3619                 pf_map_addr(AF_INET, r, saddr, &s->rt_addr, NULL,
3620                     &s->nat_src_node);
3621                 s->rt_kif = r->rpool.cur->kif;
3622                 break;
3623 #endif /* INET */
3624 #ifdef INET6
3625         case AF_INET6:
3626                 pf_map_addr(AF_INET6, r, saddr, &s->rt_addr, NULL,
3627                     &s->nat_src_node);
3628                 s->rt_kif = r->rpool.cur->kif;
3629                 break;
3630 #endif /* INET6 */
3631         }
3632 }
3633
3634 u_int32_t
3635 pf_tcp_iss(struct pf_pdesc *pd)
3636 {
3637         MD5_CTX ctx;
3638         u_int32_t digest[4];
3639
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;
3648                 }
3649                 lwkt_reltoken(&pf_gtoken);
3650         }
3651         ctx = pf_tcp_secret_ctx;
3652
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));
3658         } else {
3659                 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3660                 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3661         }
3662         MD5Final((u_char *)digest, &ctx);
3663         pf_tcp_iss_off += 4096;
3664
3665         return (digest[0] + pd->hdr.tcp->th_seq + pf_tcp_iss_off);
3666 }
3667
3668 int
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)
3673 {
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;
3683         u_short                  reason;
3684         int                      rewrite = 0, hdrlen = 0;
3685         int                      tag = -1, rtableid = -1;
3686         int                      asd = 0;
3687         int                      match = 0;
3688         int                      state_icmp = 0;
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;
3692
3693
3694         if (direction == PF_IN && pf_check_congestion(ifq)) {
3695                 REASON_SET(&reason, PFRES_CONGEST);
3696                 return (PF_DROP);
3697         }
3698
3699         if (inp != NULL)
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);
3704         }
3705
3706         switch (pd->proto) {
3707         case IPPROTO_TCP:
3708                 sport = th->th_sport;
3709                 dport = th->th_dport;
3710                 hdrlen = sizeof(*th);
3711                 break;
3712         case IPPROTO_UDP:
3713                 sport = pd->hdr.udp->uh_sport;
3714                 dport = pd->hdr.udp->uh_dport;
3715                 hdrlen = sizeof(*pd->hdr.udp);
3716                 break;
3717 #ifdef INET
3718         case IPPROTO_ICMP:
3719                 if (pd->af != AF_INET)
3720                         break;
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;
3725
3726                 if (icmptype == ICMP_UNREACH ||
3727                     icmptype == ICMP_SOURCEQUENCH ||
3728                     icmptype == ICMP_REDIRECT ||
3729                     icmptype == ICMP_TIMXCEED ||
3730                     icmptype == ICMP_PARAMPROB)
3731                         state_icmp++;
3732                 break;
3733 #endif /* INET */
3734 #ifdef INET6
3735         case IPPROTO_ICMPV6:
3736                 if (af != AF_INET6)
3737                         break;
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;
3742
3743                 if (icmptype == ICMP6_DST_UNREACH ||
3744                     icmptype == ICMP6_PACKET_TOO_BIG ||
3745                     icmptype == ICMP6_TIME_EXCEEDED ||
3746                     icmptype == ICMP6_PARAM_PROB)
3747                         state_icmp++;
3748                 break;
3749 #endif /* INET6 */
3750         default:
3751                 sport = dport = hdrlen = 0;
3752                 break;
3753         }
3754
3755         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3756
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);
3762                         goto cleanup;
3763                 }
3764
3765                 if (pd->ip_sum)
3766                         bip_sum = *pd->ip_sum;
3767
3768                 m->m_flags &= ~M_HASH;
3769                 switch (pd->proto) {
3770                 case IPPROTO_TCP:
3771                         bproto_sum = th->th_sum;
3772                         pd->proto_sum = &th->th_sum;
3773
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;
3781                         }
3782
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;
3790                         }
3791                         rewrite++;
3792                         break;
3793                 case IPPROTO_UDP:
3794                         bproto_sum = pd->hdr.udp->uh_sum;
3795                         pd->proto_sum = &pd->hdr.udp->uh_sum;
3796
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;
3805                         }
3806
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;
3815                         }
3816                         rewrite++;
3817                         break;
3818 #ifdef INET
3819                 case IPPROTO_ICMP:
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);
3824
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);
3828
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,
3832                                     nk->port[1], 0);
3833                                 pd->hdr.icmp->icmp_id = nk->port[1];
3834                                 pd->sport = &pd->hdr.icmp->icmp_id;
3835                         }
3836                         m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3837                         break;
3838 #endif /* INET */
3839 #ifdef INET6
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);
3845
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);
3849                         rewrite++;
3850                         break;
3851 #endif /* INET */
3852                 default:
3853                         switch (af) {
3854 #ifdef INET
3855                         case AF_INET:
3856                                 if (PF_ANEQ(saddr,
3857                                     &nk->addr[pd->sidx], AF_INET))
3858                                         pf_change_a(&saddr->v4.s_addr,
3859                                             pd->ip_sum,
3860                                             nk->addr[pd->sidx].v4.s_addr, 0);
3861
3862                                 if (PF_ANEQ(daddr,
3863                                     &nk->addr[pd->didx], AF_INET))
3864                                         pf_change_a(&daddr->v4.s_addr,
3865                                             pd->ip_sum,
3866                                             nk->addr[pd->didx].v4.s_addr, 0);
3867                                 break;
3868 #endif /* INET */
3869 #ifdef INET6
3870                         case AF_INET6:
3871                                 if (PF_ANEQ(saddr,
3872                                     &nk->addr[pd->sidx], AF_INET6))
3873                                         PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3874
3875                                 if (PF_ANEQ(daddr,
3876                                     &nk->addr[pd->didx], AF_INET6))
3877                                         PF_ACPY(saddr, &nk->addr[pd->didx], af);
3878                                 break;
3879 #endif /* INET */
3880                         }
3881                         break;
3882                 }
3883                 if (nr->natpass)
3884                         r = NULL;
3885                 pd->nat_rule = nr;
3886         }
3887
3888         while (r != NULL) {
3889                 r->evaluations++;
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,
3899                     r->src.neg, kif))
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,
3906                     r->dst.neg, NULL))
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],
3929                     pd->lookup.uid))
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],
3935                     pd->lookup.gid))
3936                         r = TAILQ_NEXT(r, entries);
3937                 else if (r->prob &&
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);
3947                 else {
3948                         if (r->tag)
3949                                 tag = r->tag;
3950                         if (r->rtableid >= 0)
3951                                 rtableid = r->rtableid;
3952                         if (r->anchor == NULL) {
3953                                 match = 1;
3954                                 *rm = r;
3955                                 *am = a;
3956                                 *rsm = ruleset;
3957                                 if ((*rm)->quick)
3958                                         break;
3959                                 r = TAILQ_NEXT(r, entries);
3960                         } else
3961                                 pf_step_into_anchor(&asd, &ruleset,
3962                                     PF_RULESET_FILTER, &r, &a, &match);
3963                 }
3964                 if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset,
3965                     PF_RULESET_FILTER, &r, &a, &match))
3966                         break;
3967         }
3968         r = *rm;
3969         a = *am;
3970         ruleset = *rsm;
3971
3972         REASON_SET(&reason, PFRES_MATCH);
3973
3974         if (r->log || (nr != NULL && nr->log)) {
3975                 if (rewrite)
3976                         m_copyback(m, off, hdrlen, pd->hdr.any);
3977                 PFLOG_PACKET(kif, h, m, af, direction, reason, r->log ? r : nr,
3978                     a, ruleset, pd);
3979         }
3980
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 */
3986                 if (nr != NULL) {
3987                         PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3988                         PF_ACPY(daddr, &sk->addr[pd->didx], af);
3989                         if (pd->sport)
3990                                 *pd->sport = sk->port[pd->sidx];
3991                         if (pd->dport)
3992                                 *pd->dport = sk->port[pd->didx];
3993                         if (pd->proto_sum)
3994                                 *pd->proto_sum = bproto_sum;
3995                         if (pd->ip_sum)
3996                                 *pd->ip_sum = bip_sum;
3997                         m_copyback(m, off, hdrlen, pd->hdr.any);
3998                 }
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;
4004                         int              len = 0;
4005                         struct ip       *h4;
4006 #ifdef INET6
4007                         struct ip6_hdr  *h6;
4008 #endif
4009                         switch (af) {
4010                         case AF_INET:
4011                                 h4 = mtod(m, struct ip *);
4012                                 len = h4->ip_len - off;
4013                                 break;
4014 #ifdef INET6
4015                         case AF_INET6:
4016                                 h6 = mtod(m, struct ip6_hdr *);
4017                                 len = h6->ip6_plen - (off - sizeof(*h6));
4018                                 break;
4019 #endif
4020                         }
4021
4022                         if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
4023                                 REASON_SET(&reason, PFRES_PROTCKSUM);
4024                         else {
4025                                 if (th->th_flags & TH_SYN)
4026                                         ack++;
4027                                 if (th->th_flags & TH_FIN)
4028                                         ack++;
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);
4033                         }
4034                 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
4035                     r->return_icmp)
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 &&
4039                     r->return_icmp6)
4040                         pf_send_icmp(m, r->return_icmp6 >> 8,
4041                             r->return_icmp6 & 255, af, r);
4042         }
4043
4044         if (r->action == PF_DROP)
4045                 goto cleanup;
4046
4047         if (pf_tag_packet(m, tag, rtableid)) {
4048                 REASON_SET(&reason, PFRES_MEMORY);
4049                 goto cleanup;
4050         }
4051
4052         if (!state_icmp && (r->keep_state || nr != NULL ||
4053             (pd->flags & PFDESC_TCP_NORM))) {
4054                 int action;
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,
4057                     bip_sum, hdrlen);
4058                 if (action != PF_PASS)
4059                         return (action);
4060         }
4061
4062         /* copy back packet headers if we performed NAT operations */
4063         if (rewrite)
4064                 m_copyback(m, off, hdrlen, pd->hdr.any);
4065
4066         return (PF_PASS);
4067
4068 cleanup:
4069         if (sk != NULL)
4070                 kfree(sk, M_PFSTATEKEYPL);
4071         if (nk != NULL)
4072                 kfree(nk, M_PFSTATEKEYPL);
4073         return (PF_DROP);
4074 }
4075
4076 static __inline int
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)
4083 {
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;
4088         u_short                  reason;
4089         int cpu = mycpu->gd_cpuid;
4090
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);
4095                 return (PF_DROP);
4096         }
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);
4102                 goto csfailed;
4103         }
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);
4108                 goto csfailed;
4109         }
4110         s = kmalloc(sizeof(struct pf_state), M_PFSTATEPL, M_NOWAIT|M_ZERO);
4111         if (s == NULL) {
4112                 REASON_SET(&reason, PFRES_MEMORY);
4113                 goto csfailed;
4114         }
4115         lockinit(&s->lk, "pfstlk", 0, 0);
4116         s->id = 0; /* XXX Do we really need that? not in OpenBSD */
4117         s->creatorid = 0;
4118         s->rule.ptr = r;
4119         s->nat_rule.ptr = nr;
4120         s->anchor.ptr = a;
4121         s->state_flags = PFSTATE_CREATEINPROG;
4122         STATE_INC_COUNTERS(s);
4123         if (r->allow_opts)
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;
4129
4130         s->log = r->log & PF_LOG_ALL;
4131         if (nr != NULL)
4132                 s->log |= nr->log & PF_LOG_ALL;
4133         switch (pd->proto) {
4134         case IPPROTO_TCP:
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) ==
4141                             0)
4142                                 s->src.seqdiff = 1;
4143                         pf_change_a(&th->th_seq, &th->th_sum,
4144                             htonl(s->src.seqlo + s->src.seqdiff), 0);
4145                         *rewrite = 1;
4146                 } else
4147                         s->src.seqdiff = 0;
4148                 if (th->th_flags & TH_SYN) {
4149                         s->src.seqhi++;
4150                         s->src.wscale = pf_get_wscale(m, off,
4151                             th->th_off, pd->af);
4152                 }
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);
4160                 }
4161                 if (th->th_flags & TH_FIN)
4162                         s->src.seqhi++;
4163                 s->dst.seqhi = 1;
4164                 s->dst.max_win = 1;
4165                 s->src.state = TCPS_SYN_SENT;
4166                 s->dst.state = TCPS_CLOSED;
4167                 s->timeout = PFTM_TCP_FIRST_PACKET;
4168                 break;
4169         case IPPROTO_UDP:
4170                 s->src.state = PFUDPS_SINGLE;
4171                 s->dst.state = PFUDPS_NO_TRAFFIC;
4172                 s->timeout = PFTM_UDP_FIRST_PACKET;
4173                 break;
4174         case IPPROTO_ICMP:
4175 #ifdef INET6
4176         case IPPROTO_ICMPV6:
4177 #endif
4178                 s->timeout = PFTM_ICMP_FIRST_PACKET;
4179                 break;
4180         default:
4181                 s->src.state = PFOTHERS_SINGLE;
4182                 s->dst.state = PFOTHERS_NO_TRAFFIC;
4183                 s->timeout = PFTM_OTHER_FIRST_PACKET;
4184         }
4185
4186         s->creation = time_second;
4187         s->expire = time_second;
4188
4189         if (sn != NULL) {
4190                 s->src_node = sn;
4191                 s->src_node->states++;
4192         }
4193         if (nsn != NULL) {
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++;
4198         }
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);
4206                         return (PF_DROP);
4207                 }
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);
4218                         return (PF_DROP);
4219                 }
4220         }
4221         s->direction = pd->dir;
4222
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);
4226                 goto csfailed;
4227         }
4228
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);
4236                 return (PF_DROP);
4237         } else
4238                 *sm = s;
4239
4240         pf_set_rt_ifp(s, pd->src);      /* needs s->state_key set */
4241         if (tag > 0) {
4242                 pf_tag_ref(tag);
4243                 s->tag = tag;
4244         }
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 */
4249                 if (nr != NULL) {
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);
4255                         if (pd->sport)
4256                                 *pd->sport = skt->port[pd->sidx];
4257                         if (pd->dport)
4258                                 *pd->dport = skt->port[pd->didx];
4259                         if (pd->proto_sum)
4260                                 *pd->proto_sum = bproto_sum;
4261                         if (pd->ip_sum)
4262                                 *pd->ip_sum = bip_sum;
4263                         m->m_flags &= ~M_HASH;
4264                         m_copyback(m, off, hdrlen, pd->hdr.any);
4265                 }
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);
4271                 s->src.mss = 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);
4278         }
4279
4280         s->state_flags &= ~PFSTATE_CREATEINPROG;
4281         return (PF_PASS);
4282
4283 csfailed:
4284         if (sk != NULL)
4285                 kfree(sk, M_PFSTATEKEYPL);
4286         if (nk != NULL)
4287                 kfree(nk, M_PFSTATEKEYPL);
4288
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);
4294         }
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);
4300         }
4301         if (s) {
4302                 pf_src_tree_remove_state(s);
4303                 STATE_DEC_COUNTERS(s);
4304                 kfree(s, M_PFSTATEPL);
4305         }
4306
4307         return (PF_DROP);
4308 }
4309
4310 int
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)
4314 {
4315         struct pf_rule          *r, *a = NULL;
4316         struct pf_ruleset       *ruleset = NULL;
4317         sa_family_t              af = pd->af;
4318         u_short                  reason;
4319         int                      tag = -1;
4320         int                      asd = 0;
4321         int                      match = 0;
4322
4323         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
4324         while (r != NULL) {
4325                 r->evaluations++;
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,
4335                     r->src.neg, kif))
4336                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
4337                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
4338                     r->dst.neg, NULL))
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);
4358                 else {
4359                         if (r->anchor == NULL) {
4360                                 match = 1;
4361                                 *rm = r;
4362                                 *am = a;
4363                                 *rsm = ruleset;
4364                                 if ((*rm)->quick)
4365                                         break;
4366                                 r = TAILQ_NEXT(r, entries);
4367                         } else
4368                                 pf_step_into_anchor(&asd, &ruleset,
4369                                     PF_RULESET_FILTER, &r, &a, &match);
4370                 }
4371                 if (r == NULL && pf_step_out_of_anchor(&asd, &ruleset,
4372                     PF_RULESET_FILTER, &r, &a, &match))
4373                         break;
4374         }
4375         r = *rm;
4376         a = *am;
4377         ruleset = *rsm;
4378
4379         REASON_SET(&reason, PFRES_MATCH);
4380
4381         if (r->log)
4382                 PFLOG_PACKET(kif, h, m, af, direction, reason, r, a, ruleset,
4383                     pd);
4384
4385         if (r->action != PF_PASS)
4386                 return (PF_DROP);
4387
4388         if (pf_tag_packet(m, tag, -1)) {
4389                 REASON_SET(&reason, PFRES_MEMORY);
4390                 return (PF_DROP);
4391         }
4392
4393         return (PF_PASS);
4394 }
4395
4396 /*
4397  * Called with state locked
4398  */
4399 int
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)
4403 {
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;
4407         u_int8_t                 sws, dws;
4408         int                      ackskew;
4409
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;
4413         } else {
4414                 sws = dws = 0;
4415         }
4416
4417         /*
4418          * Sequence tracking algorithm from Guido van Rooij's paper:
4419          *   http://www.madison-gurkha.com/publications/tcp_filtering/
4420          *      tcp_filtering.ps
4421          */
4422
4423         orig_seq = seq = ntohl(th->th_seq);
4424         if (src->seqlo == 0) {
4425                 /* First packet from this end. Set its state */
4426
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);
4431                                 return (PF_DROP);
4432                         }
4433                 }
4434
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)
4439                                 ;
4440                         ack = ntohl(th->th_ack) - dst->seqdiff;
4441                         pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
4442                             src->seqdiff), 0);
4443                         pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
4444                         *copyback = 1;
4445                 } else {
4446                         ack = ntohl(th->th_ack);
4447                 }
4448
4449                 end = seq + pd->p_len;
4450                 if (th->th_flags & TH_SYN) {
4451                         end++;
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,
4455                                     pd->af);
4456                                 if (src->wscale & PF_WSCALE_FLAG) {
4457                                         /* Remove scale factor from initial
4458                                          * window */
4459                                         sws = src->wscale & PF_WSCALE_MASK;
4460                                         win = ((u_int32_t)win + (1 << sws) - 1)
4461                                             >> sws;
4462                                         dws = dst->wscale & PF_WSCALE_MASK;
4463                                 } else {
4464                                         /* fixup other window */
4465                                         dst->max_win <<= dst->wscale &
4466                                             PF_WSCALE_MASK;
4467                                         /* in case of a retrans SYN|ACK */
4468                                         dst->wscale = 0;
4469                                 }
4470                         }
4471                 }
4472                 if (th->th_flags & TH_FIN)
4473                         end++;
4474
4475                 src->seqlo = seq;
4476                 if (src->state < TCPS_SYN_SENT)
4477                         src->state = TCPS_SYN_SENT;
4478
4479                 /*
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)
4483                  */
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)
4488                         src->max_win = win;
4489
4490         } else {
4491                 ack = ntohl(th->th_ack) - dst->seqdiff;
4492                 if (src->seqdiff) {
4493                         /* Modulate sequence numbers */
4494                         pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
4495                             src->seqdiff), 0);
4496                         pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
4497                         *copyback = 1;
4498                 }
4499                 end = seq + pd->p_len;
4500                 if (th->th_flags & TH_SYN)
4501                         end++;
4502                 if (th->th_flags & TH_FIN)
4503                         end++;
4504         }
4505
4506         if ((th->th_flags & TH_ACK) == 0) {
4507                 /* Let it pass through the ack skew check */
4508                 ack = dst->seqlo;
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)) {
4513                 /*
4514                  * Many stacks (ours included) will set the ACK number in an
4515                  * FIN|ACK if the SYN times out -- no sequence to ACK.
4516                  */
4517                 ack = dst->seqlo;
4518         }
4519
4520         if (seq == end) {
4521                 /* Ease sequencing restrictions on no data packets */
4522                 seq = src->seqlo;
4523                 end = seq;
4524         }
4525
4526         ackskew = dst->seqlo - ack;
4527
4528
4529         /*
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
4539          * options anyway.
4540          */
4541         if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
4542                 if (pf_modulate_sack(m, off, pd, th, dst))
4543                         *copyback = 1;
4544         }
4545
4546
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 */
4560
4561                 if (dst->scrub || src->scrub) {
4562                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4563                             *state, src, dst, copyback))
4564                                 return (PF_DROP);
4565                 }
4566
4567                 /* update max window */
4568                 if (src->max_win < win)
4569                         src->max_win = win;
4570                 /* synchronize sequencing */
4571                 if (SEQ_GT(end, src->seqlo))
4572                         src->seqlo = end;
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);
4576
4577
4578                 /* update states */
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);
4592                                         return (PF_DROP);
4593                                 }
4594                         } else if (dst->state == TCPS_CLOSING)
4595                                 dst->state = TCPS_FIN_WAIT_2;
4596                 }
4597                 if (th->th_flags & TH_RST)
4598                         src->state = dst->state = TCPS_TIME_WAIT;
4599
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;
4614                 else
4615                         (*state)->timeout = PFTM_TCP_ESTABLISHED;
4616
4617                 /* Fall through to PASS packet */
4618
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 */
4626
4627                 /*
4628                  * This currently handles three situations:
4629                  *  1) Stupid stacks will shotgun SYNs before their peer
4630                  *     replies.
4631                  *  2) When PF catches an already established stream (the
4632                  *     firewall rebooted, the state table was flushed, routes
4633                  *     changed...)
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)
4637                  *
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.
4646                  */
4647
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");
4658                 }
4659
4660                 if (dst->scrub || src->scrub) {
4661                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4662                             *state, src, dst, copyback))
4663                                 return (PF_DROP);
4664                 }
4665
4666                 /* update max window */
4667                 if (src->max_win < win)
4668                         src->max_win = win;
4669                 /* synchronize sequencing */
4670                 if (SEQ_GT(end, src->seqlo))
4671                         src->seqlo = end;
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);
4675
4676                 /*
4677                  * Cannot set dst->seqhi here since this could be a shotgunned
4678                  * SYN and not an already established connection.
4679                  */
4680
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;
4686
4687                 /* Fall through to PASS packet */
4688
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)) {
4693                 /*
4694                  * If pickup mode is hash only, do not fail on sequence checks.
4695                  *
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.
4699                  *
4700                  * Adjust expiration and fall through to PASS packet.
4701                  * XXX Add a FIN check to reduce timeout?
4702                  */
4703                 (*state)->expire = time_second;
4704         } else  {
4705                 /*
4706                  * Failure processing
4707                  */
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,
4715                                     TH_RST, 0, 0,
4716                                     (*state)->rule.ptr->return_ttl, 1, 0,
4717                                     pd->eh, kif->pfik_ifp);
4718                         src->seqlo = 0;
4719                         src->seqhi = 1;
4720                         src->max_win = 1;
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)) ?
4735                             ' ': '2',
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');
4740                 }
4741                 REASON_SET(reason, PFRES_BADSTATE);
4742                 return (PF_DROP);
4743         }
4744
4745         return (PF_PASS);
4746 }
4747
4748 /*
4749  * Called with state locked
4750  */
4751 int
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)
4754 {
4755         struct tcphdr           *th = pd->hdr.tcp;
4756
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);
4770                                 return (PF_DROP);
4771                         }
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) {
4776                         /*
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.
4781                          */
4782                         dst->state = src->state = TCPS_ESTABLISHED;
4783                         if ((*state)->src_node != NULL &&
4784                             pf_src_connlimit(*state)) {
4785                                 REASON_SET(reason, PFRES_SRCLIMIT);
4786                                 return (PF_DROP);
4787                         }
4788                 } else if (src->state == TCPS_CLOSING &&
4789                     dst->state == TCPS_ESTABLISHED &&
4790                     dst->seqlo == 0) {
4791                         /*
4792                          * Handle the closing of half connections where we
4793                          * don't see the full bidirectional FIN/ACK+ACK
4794                          * handshake.
4795                          */
4796                         dst->state = TCPS_CLOSING;
4797                 }
4798         }
4799         if (th->th_flags & TH_RST)
4800                 src->state = dst->state = TCPS_TIME_WAIT;
4801
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;
4816         else
4817                 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4818
4819         return (PF_PASS);
4820 }
4821
4822 /*
4823  * Test TCP connection state.  Caller must hold the state locked.
4824  */
4825 int
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,
4828                   u_short *reason)
4829 {
4830         struct pf_state_key_cmp  key;
4831         struct tcphdr           *th = pd->hdr.tcp;
4832         int                      copyback = 0;
4833         int                      error;
4834         struct pf_state_peer    *src, *dst;
4835         struct pf_state_key     *sk;
4836
4837         bzero(&key, sizeof(key));
4838         key.af = pd->af;
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]),
4848                                 ntohs(key.port[0]),
4849                                 ntohl(key.addr[1].addr32[0]),
4850                                 ntohs(key.port[1]));
4851                 }
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]),
4860                                 ntohs(key.port[0]),
4861                                 ntohl(key.addr[1].addr32[0]),
4862                                 ntohs(key.port[1]));
4863                 }
4864         }
4865
4866         STATE_LOOKUP(kif, &key, direction, *state, m);
4867         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
4868
4869         if (direction == (*state)->direction) {
4870                 src = &(*state)->src;
4871                 dst = &(*state)->dst;
4872         } else {
4873                 src = &(*state)->dst;
4874                 dst = &(*state)->src;
4875         }
4876
4877         sk = (*state)->key[pd->didx];
4878
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);
4883                 }
4884                 if (th->th_flags & TH_SYN) {
4885                         if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4886                                 REASON_SET(reason, PFRES_SYNPROXY);
4887                                 FAIL (PF_DROP);
4888                         }
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,
4893                             0, NULL, NULL);
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);
4900                         FAIL (PF_DROP);
4901                 } else if ((*state)->src_node != NULL &&
4902                     pf_src_connlimit(*state)) {
4903                         REASON_SET(reason, PFRES_SRCLIMIT);
4904                         FAIL (PF_DROP);
4905                 } else
4906                         (*state)->src.state = PF_TCPS_PROXY_DST;
4907         }
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);
4914                                 FAIL (PF_DROP);
4915                         }
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)) !=
4927                     (TH_SYN|TH_ACK)) ||
4928                     (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4929                         REASON_SET(reason, PFRES_SYNPROXY);
4930                         FAIL (PF_DROP);
4931                 } else {
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,
4944                             0, NULL, NULL);
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 =
4955                             TCPS_ESTABLISHED;
4956                         REASON_SET(reason, PFRES_SYNPROXY);
4957                         FAIL (PF_SYNPROXY_DROP);
4958                 }
4959         }
4960
4961         /*
4962          * Check for connection (addr+port pair) reuse.  We can't actually
4963          * unlink the state if we don't own it.
4964          */
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);
4972                         kprintf("\n");
4973                 }
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);
4978                         *state = NULL;
4979                 } else {
4980                         (*state)->timeout = PFTM_PURGE;
4981                 }
4982                 FAIL (PF_DROP);
4983         }
4984
4985         if ((*state)->state_flags & PFSTATE_SLOPPY) {
4986                 if (pf_tcp_track_sloppy(src, dst, state, pd,
4987                                         reason) == PF_DROP) {
4988                         FAIL (PF_DROP);
4989                 }
4990         } else {
4991                 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd,
4992                                       reason, &copyback) == PF_DROP) {
4993                         FAIL (PF_DROP);
4994                 }
4995         }
4996
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];
5000
5001                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
5002                     nk->port[pd->sidx] != th->th_sport)  {
5003                         /*
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.
5007                          */
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);
5012                 }
5013
5014                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
5015                     nk->port[pd->didx] != th->th_dport) {
5016                         /*
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.
5020                          */
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);
5024                 }
5025                 copyback = 1;
5026         }
5027
5028         /* Copyback sequence modulation or stateful scrub changes if needed */
5029         if (copyback) {
5030                 m->m_flags &= ~M_HASH;
5031                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
5032         }
5033
5034         pfsync_update_state(*state);
5035         error = PF_PASS;
5036 done:
5037         if (*state)
5038                 lockmgr(&(*state)->lk, LK_RELEASE);
5039         return (error);
5040 }
5041
5042 /*
5043  * Test UDP connection state.  Caller must hold the state locked.
5044  */
5045 int
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)
5048 {
5049         struct pf_state_peer    *src, *dst;
5050         struct pf_state_key_cmp  key;
5051         struct udphdr           *uh = pd->hdr.udp;
5052
5053         bzero(&key, sizeof(key));
5054         key.af = pd->af;
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;
5066         }
5067
5068         STATE_LOOKUP(kif, &key, direction, *state, m);
5069         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5070
5071         if (direction == (*state)->direction) {
5072                 src = &(*state)->src;
5073                 dst = &(*state)->dst;
5074         } else {
5075                 src = &(*state)->dst;
5076                 dst = &(*state)->src;
5077         }
5078
5079         /* update states */
5080         if (src->state < PFUDPS_SINGLE)
5081                 src->state = PFUDPS_SINGLE;
5082         if (dst->state == PFUDPS_SINGLE)
5083                 dst->state = PFUDPS_MULTIPLE;
5084
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;
5089         else
5090                 (*state)->timeout = PFTM_UDP_SINGLE;
5091
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];
5095
5096                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
5097                     nk->port[pd->sidx] != uh->uh_sport) {
5098                         /*
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.
5102                          */
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);
5108                 }
5109
5110                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
5111                     nk->port[pd->didx] != uh->uh_dport) {
5112                         /*
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.
5116                          */
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);
5121                 }
5122                 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
5123         }
5124
5125         pfsync_update_state(*state);
5126         lockmgr(&(*state)->lk, LK_RELEASE);
5127         return (PF_PASS);
5128 }
5129
5130 /*
5131  * Test ICMP connection state.  Caller must hold the state locked.
5132  */
5133 int
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,
5136                    u_short *reason)
5137 {
5138         struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
5139         u_int16_t        icmpid = 0, *icmpsum = NULL;
5140         u_int8_t         icmptype = 0;
5141         int              state_icmp = 0;
5142         int              error;
5143         struct pf_state_key_cmp key;
5144
5145         bzero(&key, sizeof(key));
5146
5147         switch (pd->proto) {
5148 #ifdef INET
5149         case IPPROTO_ICMP:
5150                 icmptype = pd->hdr.icmp->icmp_type;
5151                 icmpid = pd->hdr.icmp->icmp_id;
5152                 icmpsum = &pd->hdr.icmp->icmp_cksum;
5153
5154                 if (icmptype == ICMP_UNREACH ||
5155                     icmptype == ICMP_SOURCEQUENCH ||
5156                     icmptype == ICMP_REDIRECT ||
5157                     icmptype == ICMP_TIMXCEED ||
5158                     icmptype == ICMP_PARAMPROB)
5159                         state_icmp++;
5160                 break;
5161 #endif /* INET */
5162 #ifdef INET6
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;
5167
5168                 if (icmptype == ICMP6_DST_UNREACH ||
5169                     icmptype == ICMP6_PACKET_TOO_BIG ||
5170                     icmptype == ICMP6_TIME_EXCEEDED ||
5171                     icmptype == ICMP6_PARAM_PROB)
5172                         state_icmp++;
5173                 break;
5174 #endif /* INET6 */
5175         }
5176
5177         if (!state_icmp) {
5178
5179                 /*
5180                  * ICMP query/reply message not related to a TCP/UDP packet.
5181                  * Search for an ICMP state.
5182                  */
5183                 key.af = pd->af;
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);
5192                 }
5193
5194                 STATE_LOOKUP(kif, &key, direction, *state, m);
5195                 lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5196
5197                 (*state)->expire = time_second;
5198                 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
5199
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];
5203
5204                         switch (pd->af) {
5205 #ifdef INET
5206                         case AF_INET:
5207                                 if (PF_ANEQ(pd->src,
5208                                     &nk->addr[pd->sidx], AF_INET))
5209                                         pf_change_a(&saddr->v4.s_addr,
5210                                             pd->ip_sum,
5211                                             nk->addr[pd->sidx].v4.s_addr, 0);
5212
5213                                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
5214                                     AF_INET))
5215                                         pf_change_a(&daddr->v4.s_addr,
5216                                             pd->ip_sum,
5217                                             nk->addr[pd->didx].v4.s_addr, 0);
5218
5219                                 if (nk->port[0] !=
5220                                     pd->hdr.icmp->icmp_id) {
5221                                         pd->hdr.icmp->icmp_cksum =
5222                                             pf_cksum_fixup(
5223                                             pd->hdr.icmp->icmp_cksum, icmpid,
5224                                             nk->port[pd->sidx], 0);
5225                                         pd->hdr.icmp->icmp_id =
5226                                             nk->port[pd->sidx];
5227                                 }
5228
5229                                 m->m_flags &= ~M_HASH;
5230                                 m_copyback(m, off, ICMP_MINLEN,
5231                                     (caddr_t)pd->hdr.icmp);
5232                                 break;
5233 #endif /* INET */
5234 #ifdef INET6
5235                         case AF_INET6:
5236                                 if (PF_ANEQ(pd->src,
5237                                     &nk->addr[pd->sidx], AF_INET6))
5238                                         pf_change_a6(saddr,
5239                                             &pd->hdr.icmp6->icmp6_cksum,
5240                                             &nk->addr[pd->sidx], 0);
5241
5242                                 if (PF_ANEQ(pd->dst,
5243                                     &nk->addr[pd->didx], AF_INET6))
5244                                         pf_change_a6(daddr,
5245                                             &pd->hdr.icmp6->icmp6_cksum,
5246                                             &nk->addr[pd->didx], 0);
5247
5248                                 m->m_flags &= ~M_HASH;
5249                                 m_copyback(m, off,
5250                                         sizeof(struct icmp6_hdr),
5251                                         (caddr_t)pd->hdr.icmp6);
5252                                 break;
5253 #endif /* INET6 */
5254                         }
5255                 }
5256         } else {
5257                 /*
5258                  * ICMP error message in response to a TCP/UDP packet.
5259                  * Extract the inner TCP/UDP header and search for that state.
5260                  */
5261
5262                 struct pf_pdesc pd2;
5263 #ifdef INET
5264                 struct ip       h2;
5265 #endif /* INET */
5266 #ifdef INET6
5267                 struct ip6_hdr  h2_6;
5268                 int             terminal = 0;
5269 #endif /* INET6 */
5270                 int             ipoff2;
5271                 int             off2;
5272
5273                 pd2.not_cpu_localized = 1;
5274                 pd2.af = pd->af;
5275                 /* Payload packet is from the opposite direction. */
5276                 pd2.sidx = (direction == PF_IN) ? 1 : 0;
5277                 pd2.didx = (direction == PF_IN) ? 0 : 1;
5278                 switch (pd->af) {
5279 #ifdef INET
5280                 case AF_INET:
5281                         /* offset of h2 in mbuf chain */
5282                         ipoff2 = off + ICMP_MINLEN;
5283
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 "
5288                                     "(ip)\n"));
5289                                 FAIL (PF_DROP);
5290                         }
5291                         /*
5292                          * ICMP error messages don't refer to non-first
5293                          * fragments
5294                          */
5295                         if (h2.ip_off & htons(IP_OFFMASK)) {
5296                                 REASON_SET(reason, PFRES_FRAG);
5297                                 FAIL (PF_DROP);
5298                         }
5299
5300                         /* offset of protocol header that follows h2 */
5301                         off2 = ipoff2 + (h2.ip_hl << 2);
5302
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;
5307                         break;
5308 #endif /* INET */
5309 #ifdef INET6
5310                 case AF_INET6:
5311                         ipoff2 = off + sizeof(struct icmp6_hdr);
5312
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 "
5317                                     "(ip6)\n"));
5318                                 FAIL (PF_DROP);
5319                         }
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;
5323                         pd2.ip_sum = NULL;
5324                         off2 = ipoff2 + sizeof(h2_6);
5325                         do {
5326                                 switch (pd2.proto) {
5327                                 case IPPROTO_FRAGMENT:
5328                                         /*
5329                                          * ICMPv6 error messages for
5330                                          * non-first fragments
5331                                          */
5332                                         REASON_SET(reason, PFRES_FRAG);
5333                                         FAIL (PF_DROP);
5334                                 case IPPROTO_AH:
5335                                 case IPPROTO_HOPOPTS:
5336                                 case IPPROTO_ROUTING:
5337                                 case IPPROTO_DSTOPTS: {
5338                                         /* get next header and header length */
5339                                         struct ip6_ext opt6;
5340
5341                                         if (!pf_pull_hdr(m, off2, &opt6,
5342                                             sizeof(opt6), NULL, reason,
5343                                             pd2.af)) {
5344                                                 DPFPRINTF(PF_DEBUG_MISC,
5345                                                     ("pf: ICMPv6 short opt\n"));
5346                                                 FAIL (PF_DROP);
5347                                         }
5348                                         if (pd2.proto == IPPROTO_AH)
5349                                                 off2 += (opt6.ip6e_len + 2) * 4;
5350                                         else
5351                                                 off2 += (opt6.ip6e_len + 1) * 8;
5352                                         pd2.proto = opt6.ip6e_nxt;
5353                                         /* goto the next header */
5354                                         break;
5355                                 }
5356                                 default:
5357                                         terminal++;
5358                                         break;
5359                                 }
5360                         } while (!terminal);
5361                         break;
5362 #endif /* INET6 */
5363                 default:
5364                         DPFPRINTF(PF_DEBUG_MISC,
5365                             ("pf: ICMP AF %d unknown (ip6)\n", pd->af));
5366                         FAIL (PF_DROP);
5367                         break;
5368                 }
5369
5370                 switch (pd2.proto) {
5371                 case IPPROTO_TCP: {
5372                         struct tcphdr            th;
5373                         u_int32_t                seq;
5374                         struct pf_state_peer    *src, *dst;
5375                         u_int8_t                 dws;
5376                         int                      copyback = 0;
5377
5378                         /*
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.
5382                          */
5383                         if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
5384                             pd2.af)) {
5385                                 DPFPRINTF(PF_DEBUG_MISC,
5386                                     ("pf: ICMP error message too short "
5387                                     "(tcp)\n"));
5388                                 FAIL (PF_DROP);
5389                         }
5390
5391                         key.af = pd2.af;
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;
5397
5398                         STATE_LOOKUP(kif, &key, direction, *state, m);
5399                         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5400
5401                         if (direction == (*state)->direction) {
5402                                 src = &(*state)->dst;
5403                                 dst = &(*state)->src;
5404                         } else {
5405                                 src = &(*state)->src;
5406                                 dst = &(*state)->dst;
5407                         }
5408
5409                         if (src->wscale && dst->wscale)
5410                                 dws = dst->wscale & PF_WSCALE_MASK;
5411                         else
5412                                 dws = 0;
5413
5414                         /* Demodulate sequence number */
5415                         seq = ntohl(th.th_seq) - src->seqdiff;
5416                         if (src->seqdiff) {
5417                                 pf_change_a(&th.th_seq, icmpsum,
5418                                     htonl(seq), 0);
5419                                 copyback = 1;
5420                         }
5421
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);
5429                                         kprintf(" -> ");
5430                                         pf_print_host(pd->dst, 0, pd->af);
5431                                         kprintf(" state: ");
5432                                         pf_print_state(*state);
5433                                         kprintf(" seq=%u\n", seq);
5434                                 }
5435                                 REASON_SET(reason, PFRES_BADSTATE);
5436                                 FAIL (PF_DROP);
5437                         } else {
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);
5442                                         kprintf(" -> ");
5443                                         pf_print_host(pd->dst, 0, pd->af);
5444                                         kprintf(" state: ");
5445                                         pf_print_state(*state);
5446                                         kprintf(" seq=%u\n", seq);
5447                                 }
5448                         }
5449
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];
5455
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);
5464
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);
5474                                 copyback = 1;
5475                         }
5476
5477                         if (copyback) {
5478                                 switch (pd2.af) {
5479 #ifdef INET
5480                                 case AF_INET:
5481                                         m_copyback(m, off, ICMP_MINLEN,
5482                                             (caddr_t)pd->hdr.icmp);
5483                                         m_copyback(m, ipoff2, sizeof(h2),
5484                                             (caddr_t)&h2);
5485                                         break;
5486 #endif /* INET */
5487 #ifdef INET6
5488                                 case AF_INET6:
5489                                         m_copyback(m, off,
5490                                             sizeof(struct icmp6_hdr),
5491                                             (caddr_t)pd->hdr.icmp6);
5492                                         m_copyback(m, ipoff2, sizeof(h2_6),
5493                                             (caddr_t)&h2_6);
5494                                         break;
5495 #endif /* INET6 */
5496                                 }
5497                                 m->m_flags &= ~M_HASH;
5498                                 m_copyback(m, off2, 8, (caddr_t)&th);
5499                         }
5500                         break;
5501                 }
5502                 case IPPROTO_UDP: {
5503                         struct udphdr           uh;
5504
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 "
5509                                     "(udp)\n"));
5510                                 return (PF_DROP);
5511                         }
5512
5513                         key.af = pd2.af;
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;
5519
5520                         STATE_LOOKUP(kif, &key, direction, *state, m);
5521                         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5522
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];
5528
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);
5537
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);
5547
5548                                 switch (pd2.af) {
5549 #ifdef INET
5550                                 case AF_INET:
5551                                         m_copyback(m, off, ICMP_MINLEN,
5552                                             (caddr_t)pd->hdr.icmp);
5553                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5554                                         break;
5555 #endif /* INET */
5556 #ifdef INET6
5557                                 case AF_INET6:
5558                                         m_copyback(m, off,
5559                                             sizeof(struct icmp6_hdr),
5560                                             (caddr_t)pd->hdr.icmp6);
5561                                         m_copyback(m, ipoff2, sizeof(h2_6),
5562                                             (caddr_t)&h2_6);
5563                                         break;
5564 #endif /* INET6 */
5565                                 }
5566                                 m->m_flags &= ~M_HASH;
5567                                 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
5568                         }
5569                         break;
5570                 }
5571 #ifdef INET
5572                 case IPPROTO_ICMP: {
5573                         struct icmp             iih;
5574
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"
5579                                     "(icmp)\n"));
5580                                 return (PF_DROP);
5581                         }
5582
5583                         key.af = pd2.af;
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;
5588
5589                         STATE_LOOKUP(kif, &key, direction, *state, m);
5590                         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5591
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];
5597
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);
5606
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);
5616
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;
5621                         }
5622                         break;
5623                 }
5624 #endif /* INET */
5625 #ifdef INET6
5626                 case IPPROTO_ICMPV6: {
5627                         struct icmp6_hdr        iih;
5628
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 "
5633                                     "(icmp6)\n"));
5634                                 FAIL (PF_DROP);
5635                         }
5636
5637                         key.af = pd2.af;
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;
5642
5643                         STATE_LOOKUP(kif, &key, direction, *state, m);
5644                         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5645
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];
5651
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);
5660
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);
5670
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),
5675                                     (caddr_t)&iih);
5676                                 m->m_flags &= ~M_HASH;
5677                         }
5678                         break;
5679                 }
5680 #endif /* INET6 */
5681                 default: {
5682                         key.af = pd2.af;
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;
5687
5688                         STATE_LOOKUP(kif, &key, direction, *state, m);
5689                         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5690
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];
5696
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);
5703
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);
5711
5712                                 switch (pd2.af) {
5713 #ifdef INET
5714                                 case AF_INET:
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;
5719                                         break;
5720 #endif /* INET */
5721 #ifdef INET6
5722                                 case AF_INET6:
5723                                         m_copyback(m, off,
5724                                             sizeof(struct icmp6_hdr),
5725                                             (caddr_t)pd->hdr.icmp6);
5726                                         m_copyback(m, ipoff2, sizeof(h2_6),
5727                                             (caddr_t)&h2_6);
5728                                         m->m_flags &= ~M_HASH;
5729                                         break;
5730 #endif /* INET6 */
5731                                 }
5732                         }
5733                         break;
5734                 }
5735                 }
5736         }
5737
5738         pfsync_update_state(*state);
5739         error = PF_PASS;
5740 done:
5741         if (*state)
5742                 lockmgr(&(*state)->lk, LK_RELEASE);
5743         return (error);
5744 }
5745
5746 /*
5747  * Test other connection state.  Caller must hold the state locked.
5748  */
5749 int
5750 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
5751                     struct mbuf *m, struct pf_pdesc *pd)
5752 {
5753         struct pf_state_peer    *src, *dst;
5754         struct pf_state_key_cmp  key;
5755
5756         bzero(&key, sizeof(key));
5757         key.af = pd->af;
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;
5763         } else {
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;
5767         }
5768
5769         STATE_LOOKUP(kif, &key, direction, *state, m);
5770         lockmgr(&(*state)->lk, LK_EXCLUSIVE);
5771
5772         if (direction == (*state)->direction) {
5773                 src = &(*state)->src;
5774                 dst = &(*state)->dst;
5775         } else {
5776                 src = &(*state)->dst;
5777                 dst = &(*state)->src;
5778         }
5779
5780         /* update states */
5781         if (src->state < PFOTHERS_SINGLE)
5782                 src->state = PFOTHERS_SINGLE;
5783         if (dst->state == PFOTHERS_SINGLE)
5784                 dst->state = PFOTHERS_MULTIPLE;
5785
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;
5790         else
5791                 (*state)->timeout = PFTM_OTHER_SINGLE;
5792
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];
5796
5797                 KKASSERT(nk);
5798                 KKASSERT(pd);
5799                 KKASSERT(pd->src);
5800                 KKASSERT(pd->dst);
5801                 switch (pd->af) {
5802 #ifdef INET
5803                 case AF_INET:
5804                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5805                                 pf_change_a(&pd->src->v4.s_addr,
5806                                     pd->ip_sum,
5807                                     nk->addr[pd->sidx].v4.s_addr,
5808                                     0);
5809
5810
5811                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5812                                 pf_change_a(&pd->dst->v4.s_addr,
5813                                     pd->ip_sum,
5814                                     nk->addr[pd->didx].v4.s_addr,
5815                                     0);
5816
5817                         break;
5818 #endif /* INET */
5819 #ifdef INET6
5820                 case AF_INET6:
5821                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5822                                 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5823
5824                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5825                                 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5826 #endif /* INET6 */
5827                 }
5828         }
5829
5830         pfsync_update_state(*state);
5831         lockmgr(&(*state)->lk, LK_RELEASE);
5832         return (PF_PASS);
5833 }
5834
5835 /*
5836  * ipoff and off are measured from the start of the mbuf chain.
5837  * h must be at "ipoff" on the mbuf chain.
5838  */
5839 void *
5840 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5841     u_short *actionp, u_short *reasonp, sa_family_t af)
5842 {
5843         switch (af) {
5844 #ifdef INET
5845         case AF_INET: {
5846                 struct ip       *h = mtod(m, struct ip *);
5847                 u_int16_t        fragoff = (h->ip_off & IP_OFFMASK) << 3;
5848
5849                 if (fragoff) {
5850                         if (fragoff >= len)
5851                                 ACTION_SET(actionp, PF_PASS);
5852                         else {
5853                                 ACTION_SET(actionp, PF_DROP);
5854                                 REASON_SET(reasonp, PFRES_FRAG);
5855                         }
5856                         return (NULL);
5857                 }
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);
5862                         return (NULL);
5863                 }
5864                 break;
5865         }
5866 #endif /* INET */
5867 #ifdef INET6
5868         case AF_INET6: {
5869                 struct ip6_hdr  *h = mtod(m, struct ip6_hdr *);
5870
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);
5876                         return (NULL);
5877                 }
5878                 break;
5879         }
5880 #endif /* INET6 */
5881         }
5882         m_copydata(m, off, len, p);
5883         return (p);
5884 }
5885
5886 int
5887 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif)
5888 {
5889         struct sockaddr_in      *dst;
5890         int                      ret = 1;
5891         int                      check_mpath;
5892 #ifdef INET6
5893         struct sockaddr_in6     *dst6;
5894         struct route_in6         ro;
5895 #else
5896         struct route             ro;
5897 #endif
5898         struct radix_node       *rn;
5899         struct rtentry          *rt;
5900         struct ifnet            *ifp;
5901
5902         check_mpath = 0;
5903         bzero(&ro, sizeof(ro));
5904         switch (af) {
5905         case AF_INET:
5906                 dst = satosin(&ro.ro_dst);
5907                 dst->sin_family = AF_INET;
5908                 dst->sin_len = sizeof(*dst);
5909                 dst->sin_addr = addr->v4;
5910                 break;
5911 #ifdef INET6
5912         case AF_INET6:
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;
5917                 break;
5918 #endif /* INET6 */
5919         default:
5920                 return (0);
5921         }
5922
5923         /* Skip checks for ipsec interfaces */
5924         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5925                 goto out;
5926
5927         rtalloc_ign((struct route *)&ro, 0);
5928
5929         if (ro.ro_rt != NULL) {
5930                 /* No interface given, this is a no-route check */
5931                 if (kif == NULL)
5932                         goto out;
5933
5934                 if (kif->pfik_ifp == NULL) {
5935                         ret = 0;
5936                         goto out;
5937                 }
5938
5939                 /* Perform uRPF check if passed input interface */
5940                 ret = 0;
5941                 rn = (struct radix_node *)ro.ro_rt;
5942                 do {
5943                         rt = (struct rtentry *)rn;
5944                         ifp = rt->rt_ifp;
5945
5946                         if (kif->pfik_ifp == ifp)
5947                                 ret = 1;
5948                         rn = NULL;
5949                 } while (check_mpath == 1 && rn != NULL && ret == 0);
5950         } else
5951                 ret = 0;
5952 out:
5953         if (ro.ro_rt != NULL)
5954                 RTFREE(ro.ro_rt);
5955         return (ret);
5956 }
5957
5958 int
5959 pf_rtlabel_match(struct pf_addr *addr, sa_family_t af, struct pf_addr_wrap *aw)
5960 {
5961         struct sockaddr_in      *dst;
5962 #ifdef INET6
5963         struct sockaddr_in6     *dst6;
5964         struct route_in6         ro;
5965 #else
5966         struct route             ro;
5967 #endif
5968         int                      ret = 0;
5969
5970         ASSERT_LWKT_TOKEN_HELD(&pf_token);
5971
5972         bzero(&ro, sizeof(ro));
5973         switch (af) {
5974         case AF_INET:
5975                 dst = satosin(&ro.ro_dst);
5976                 dst->sin_family = AF_INET;
5977                 dst->sin_len = sizeof(*dst);
5978                 dst->sin_addr = addr->v4;
5979                 break;
5980 #ifdef INET6
5981         case AF_INET6:
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;
5986                 break;
5987 #endif /* INET6 */
5988         default:
5989                 return (0);
5990         }
5991
5992 rtalloc_ign((struct route *)&ro, (RTF_CLONING | RTF_PRCLONING));
5993
5994         if (ro.ro_rt != NULL) {
5995                 RTFREE(ro.ro_rt);
5996         }
5997
5998         return (ret);
5999 }
6000
6001 #ifdef INET
6002 void
6003 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
6004     struct pf_state *s, struct pf_pdesc *pd)
6005 {
6006         struct mbuf             *m0, *m1;
6007         struct route             iproute;
6008         struct route            *ro = NULL;
6009         struct sockaddr_in      *dst;
6010         struct ip               *ip;
6011         struct ifnet            *ifp = NULL;
6012         struct pf_addr           naddr;
6013         struct pf_src_node      *sn = NULL;
6014         int                      error = 0;
6015         int sw_csum;
6016 #ifdef IPSEC
6017         struct m_tag            *mtag;
6018 #endif /* IPSEC */
6019
6020         ASSERT_LWKT_TOKEN_HELD(&pf_token);
6021
6022         if (m == NULL || *m == NULL || r == NULL ||
6023             (dir != PF_IN && dir != PF_OUT) || oifp == NULL)
6024                 panic("pf_route: invalid parameters");
6025
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;
6029         } else {
6030                 if ((*m)->m_pkthdr.pf.routed++ > 3) {
6031                         m0 = *m;
6032                         *m = NULL;
6033                         goto bad;
6034                 }
6035         }
6036
6037         if (r->rt == PF_DUPTO) {
6038                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
6039                         return;
6040                 }
6041         } else {
6042                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
6043                         return;
6044                 }
6045                 m0 = *m;
6046         }
6047
6048         if (m0->m_len < sizeof(struct ip)) {
6049                 DPFPRINTF(PF_DEBUG_URGENT,
6050                     ("pf_route: m0->m_len < sizeof(struct ip)\n"));
6051                 goto bad;
6052         }
6053
6054         ip = mtod(m0, struct ip *);
6055
6056         ro = &iproute;
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;
6062
6063         if (r->rt == PF_FASTROUTE) {
6064                 rtalloc(ro);
6065                 if (ro->ro_rt == 0) {
6066                         ipstat.ips_noroute++;
6067                         goto bad;
6068                 }
6069
6070                 ifp = ro->ro_rt->rt_ifp;
6071                 ro->ro_rt->rt_use++;
6072
6073                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
6074                         dst = satosin(ro->ro_rt->rt_gateway);
6075         } else {
6076                 if (TAILQ_EMPTY(&r->rpool.list)) {
6077                         DPFPRINTF(PF_DEBUG_URGENT,
6078                             ("pf_route: TAILQ_EMPTY(&r->rpool.list)\n"));
6079                         goto bad;
6080                 }
6081                 if (s == NULL) {
6082                         pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
6083                             &naddr, NULL, &sn);
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;
6088                 } else {
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;
6093                 }
6094         }
6095         if (ifp == NULL)
6096                 goto bad;
6097
6098         if (oifp != ifp) {
6099                 if (pf_test(PF_OUT, ifp, &m0, NULL, NULL) != PF_PASS) {
6100                         goto bad;
6101                 } else if (m0 == NULL) {
6102                         goto done;
6103                 }
6104                 if (m0->m_len < sizeof(struct ip)) {
6105                         DPFPRINTF(PF_DEBUG_URGENT,
6106                             ("pf_route: m0->m_len < sizeof(struct ip)\n"));
6107                         goto bad;
6108                 }
6109                 ip = mtod(m0, struct ip *);
6110         }
6111
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;
6118         }
6119         m0->m_pkthdr.csum_flags &= ifp->if_hwassist;
6120         m0->m_pkthdr.csum_iphlen = (ip->ip_hl << 2);
6121
6122         /*
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.
6127          */
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);
6134                 ip->ip_sum = 0;
6135                 if (sw_csum & CSUM_DELAY_IP) {
6136                         /* From KAME */
6137                         if (ip->ip_v == IPVERSION &&
6138                             (ip->ip_hl << 2) == sizeof(*ip)) {
6139                                 ip->ip_sum = in_cksum_hdr(ip);
6140                         } else {
6141                                 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
6142                         }
6143                 }
6144                 lwkt_reltoken(&pf_token);
6145                 error = ifp->if_output(ifp, m0, sintosa(dst), ro->ro_rt);
6146                 lwkt_gettoken(&pf_token);
6147                 goto done;
6148         }
6149
6150         /*
6151          * Too large for interface; fragment if possible.
6152          * Must be able to put at least 8 bytes per fragment.
6153          */
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,
6158                                    ifp->if_mtu);
6159                         goto done;
6160                 } else
6161                         goto bad;
6162         }
6163
6164         m1 = m0;
6165         error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist, sw_csum);
6166         if (error) {
6167                 goto bad;
6168         }
6169
6170         for (m0 = m1; m0; m0 = m1) {
6171                 m1 = m0->m_nextpkt;
6172                 m0->m_nextpkt = 0;
6173                 if (error == 0) {
6174                         lwkt_reltoken(&pf_token);
6175                         error = (*ifp->if_output)(ifp, m0, sintosa(dst),
6176                                                   NULL);
6177                         lwkt_gettoken(&pf_token);
6178                 } else
6179                         m_freem(m0);
6180         }
6181
6182         if (error == 0)
6183                 ipstat.ips_fragmented++;
6184
6185 done:
6186         if (r->rt != PF_DUPTO)
6187                 *m = NULL;
6188         if (ro == &iproute && ro->ro_rt)
6189                 RTFREE(ro->ro_rt);
6190         return;
6191
6192 bad:
6193         m_freem(m0);
6194         goto done;
6195 }
6196 #endif /* INET */
6197
6198 #ifdef INET6
6199 void
6200 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
6201     struct pf_state *s, struct pf_pdesc *pd)
6202 {
6203         struct mbuf             *m0;
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;
6211
6212         if (m == NULL || *m == NULL || r == NULL ||
6213             (dir != PF_IN && dir != PF_OUT) || oifp == NULL)
6214                 panic("pf_route6: invalid parameters");
6215
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;
6219         } else {
6220                 if ((*m)->m_pkthdr.pf.routed++ > 3) {
6221                         m0 = *m;
6222                         *m = NULL;
6223                         goto bad;
6224                 }
6225         }
6226
6227         if (r->rt == PF_DUPTO) {
6228                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL)
6229                         return;
6230         } else {
6231                 if ((r->rt == PF_REPLYTO) == (r->direction == dir))
6232                         return;
6233                 m0 = *m;
6234         }
6235
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"));
6239                 goto bad;
6240         }
6241         ip6 = mtod(m0, struct ip6_hdr *);
6242
6243         ro = &ip6route;
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;
6249
6250         /*
6251          * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
6252          * so make sure pf.flags is clear.
6253          *
6254          * Cheat. XXX why only in the v6 case???
6255          */
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);
6262                 return;
6263         }
6264
6265         if (TAILQ_EMPTY(&r->rpool.list)) {
6266                 DPFPRINTF(PF_DEBUG_URGENT,
6267                     ("pf_route6: TAILQ_EMPTY(&r->rpool.list)\n"));
6268                 goto bad;
6269         }
6270         if (s == NULL) {
6271                 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
6272                     &naddr, NULL, &sn);
6273                 if (!PF_AZERO(&naddr, AF_INET6))
6274                         PF_ACPY((struct pf_addr *)&dst->sin6_addr,
6275                             &naddr, AF_INET6);
6276                 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
6277         } else {
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;
6282         }
6283         if (ifp == NULL)
6284                 goto bad;
6285
6286         if (oifp != ifp) {
6287                 if (pf_test6(PF_OUT, ifp, &m0, NULL, NULL) != PF_PASS) {
6288                         goto bad;
6289                 } else if (m0 == NULL) {
6290                         goto done;
6291                 }
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"));
6295                         goto bad;
6296                 }
6297                 ip6 = mtod(m0, struct ip6_hdr *);
6298         }
6299
6300         /*
6301          * If the packet is too large for the outgoing interface,
6302          * send back an icmp6 error.
6303          */
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);
6308         } else {
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);
6312                 else
6313                         goto bad;
6314         }
6315
6316 done:
6317         if (r->rt != PF_DUPTO)
6318                 *m = NULL;
6319         return;
6320
6321 bad:
6322         m_freem(m0);
6323         goto done;
6324 }
6325 #endif /* INET6 */
6326
6327
6328 /*
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.
6333  */
6334 /*
6335  * XXX
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.)
6344  */
6345 int
6346 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p,
6347     sa_family_t af)
6348 {
6349         u_int16_t sum = 0;
6350         int hw_assist = 0;
6351         struct ip *ip;
6352
6353         if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
6354                 return (1);
6355         if (m->m_pkthdr.len < off + len)
6356                 return (1);
6357
6358         switch (p) {
6359         case IPPROTO_TCP:
6360         case IPPROTO_UDP:
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;
6364                         } else {
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));
6369                         }
6370                         sum ^= 0xffff;
6371                         ++hw_assist;
6372                 }
6373                 break;
6374         case IPPROTO_ICMP:
6375 #ifdef INET6
6376         case IPPROTO_ICMPV6:
6377 #endif /* INET6 */
6378                 break;
6379         default:
6380                 return (1);
6381         }
6382
6383         if (!hw_assist) {
6384                 switch (af) {
6385                 case AF_INET:
6386                         if (p == IPPROTO_ICMP) {
6387                                 if (m->m_len < off)
6388                                         return (1);
6389                                 m->m_data += off;
6390                                 m->m_len -= off;
6391                                 sum = in_cksum(m, len);
6392                                 m->m_data -= off;
6393                                 m->m_len += off;
6394                         } else {
6395                                 if (m->m_len < sizeof(struct ip))
6396                                         return (1);
6397                                 sum = in_cksum_range(m, p, off, len);
6398                                 if (sum == 0) {
6399                                         m->m_pkthdr.csum_flags |=
6400                                             (CSUM_DATA_VALID |
6401                                              CSUM_PSEUDO_HDR);
6402                                         m->m_pkthdr.csum_data = 0xffff;
6403                                 }
6404                         }
6405                         break;
6406 #ifdef INET6
6407                 case AF_INET6:
6408                         if (m->m_len < sizeof(struct ip6_hdr))
6409                                 return (1);
6410                         sum = in6_cksum(m, p, off, len);
6411                         /*
6412                          * XXX
6413                          * IPv6 H/W cksum off-load not supported yet!
6414                          *
6415                          * if (sum == 0) {
6416                          *      m->m_pkthdr.csum_flags |=
6417                          *          (CSUM_DATA_VALID|CSUM_PSEUDO_HDR);
6418                          *      m->m_pkthdr.csum_data = 0xffff;
6419                          *}
6420                          */
6421                         break;
6422 #endif /* INET6 */
6423                 default:
6424                         return (1);
6425                 }
6426         }
6427         if (sum) {
6428                 switch (p) {
6429                 case IPPROTO_TCP:
6430                         tcpstat.tcps_rcvbadsum++;
6431                         break;
6432                 case IPPROTO_UDP:
6433                         udp_stat.udps_badsum++;
6434                         break;
6435                 case IPPROTO_ICMP:
6436                         icmpstat.icps_checksum++;
6437                         break;
6438 #ifdef INET6
6439                 case IPPROTO_ICMPV6:
6440                         icmp6stat.icp6s_checksum++;
6441                         break;
6442 #endif /* INET6 */
6443                 }
6444                 return (1);
6445         }
6446         return (0);
6447 }
6448
6449 struct pf_divert *
6450 pf_find_divert(struct mbuf *m)
6451 {
6452         struct m_tag    *mtag;
6453
6454         if ((mtag = m_tag_find(m, PACKET_TAG_PF_DIVERT, NULL)) == NULL)
6455                 return (NULL);
6456
6457         return ((struct pf_divert *)(mtag + 1));
6458 }
6459
6460 struct pf_divert *
6461 pf_get_divert(struct mbuf *m)
6462 {
6463         struct m_tag    *mtag;
6464
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),
6467                     M_NOWAIT);
6468                 if (mtag == NULL)
6469                         return (NULL);
6470                 bzero(mtag + 1, sizeof(struct pf_divert));
6471                 m_tag_prepend(m, mtag);
6472         }
6473
6474         return ((struct pf_divert *)(mtag + 1));
6475 }
6476
6477 #ifdef INET
6478
6479 /*
6480  * WARNING: pf_token held shared on entry, THIS IS CPU LOCALIZED CODE
6481  */
6482 int
6483 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0,
6484     struct ether_header *eh, struct inpcb *inp)
6485 {
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;
6493         struct pf_pdesc          pd;
6494         int                      off, dirndx;
6495 #ifdef ALTQ
6496         int                      pqid = 0;
6497 #endif
6498
6499         if (m->m_pkthdr.fw_flags & IPFW_MBUF_CONTINUE) {
6500                 /* Skip us; continue in ipfw. */
6501                 return (PF_PASS);
6502         }
6503
6504         if (!pf_status.running)
6505                 return (PF_PASS);
6506
6507         memset(&pd, 0, sizeof(pd));
6508 #ifdef foo
6509         if (ifp->if_type == IFT_CARP && ifp->if_carpdev)
6510                 kif = (struct pfi_kif *)ifp->if_carpdev->if_pf_kif;
6511         else
6512 #endif
6513                 kif = (struct pfi_kif *)ifp->if_pf_kif;
6514
6515         if (kif == NULL) {
6516                 DPFPRINTF(PF_DEBUG_URGENT,
6517                     ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
6518                 return (PF_DROP);
6519         }
6520         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6521                 return (PF_PASS);
6522
6523 #ifdef DIAGNOSTIC
6524         if ((m->m_flags & M_PKTHDR) == 0)
6525                 panic("non-M_PKTHDR is passed to pf_test");
6526 #endif /* DIAGNOSTIC */
6527
6528         if (m->m_pkthdr.len < (int)sizeof(*h)) {
6529                 action = PF_DROP;
6530                 REASON_SET(&reason, PFRES_SHORT);
6531                 log = 1;
6532                 goto done;
6533         }
6534
6535         /*
6536          * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
6537          * so make sure pf.flags is clear.
6538          */
6539         if (m->m_pkthdr.fw_flags & PF_MBUF_TAGGED)
6540                 return (PF_PASS);
6541         m->m_pkthdr.pf.flags = 0;
6542         /* Re-Check when updating to > 4.4 */
6543         m->m_pkthdr.pf.statekey = NULL;
6544
6545         /* We do IP header normalization and packet reassembly here */
6546         if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
6547                 action = PF_DROP;
6548                 goto done;
6549         }
6550         m = *m0;        /* pf_normalize messes with m0 */
6551         h = mtod(m, struct ip *);
6552
6553         off = h->ip_hl << 2;
6554         if (off < (int)sizeof(*h)) {
6555                 action = PF_DROP;
6556                 REASON_SET(&reason, PFRES_SHORT);
6557                 log = 1;
6558                 goto done;
6559         }
6560
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;
6566         pd.proto = h->ip_p;
6567         pd.dir = dir;
6568         pd.sidx = (dir == PF_IN) ? 0 : 1;
6569         pd.didx = (dir == PF_IN) ? 1 : 0;
6570         pd.af = AF_INET;
6571         pd.tos = h->ip_tos;
6572         pd.tot_len = h->ip_len;
6573         pd.eh = eh;
6574
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,
6578                     &pd, &a, &ruleset);
6579                 goto done;
6580         }
6581
6582         switch (h->ip_p) {
6583
6584         case IPPROTO_TCP: {
6585                 struct tcphdr   th;
6586
6587                 pd.hdr.tcp = &th;
6588                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6589                     &action, &reason, AF_INET)) {
6590                         log = action != PF_PASS;
6591                         goto done;
6592                 }
6593                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6594 #ifdef ALTQ
6595                 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
6596                         pqid = 1;
6597 #endif
6598                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6599                 if (action == PF_DROP)
6600                         goto done;
6601                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6602                                            &reason);
6603                 if (action == PF_PASS) {
6604                         r = s->rule.ptr;
6605                         a = s->anchor.ptr;
6606                         log = s->log;
6607                 } else if (s == NULL) {
6608                         action = pf_test_rule(&r, &s, dir, kif,
6609                                               m, off, h, &pd, &a,
6610                                               &ruleset, NULL, inp);
6611                 }
6612                 break;
6613         }
6614
6615         case IPPROTO_UDP: {
6616                 struct udphdr   uh;
6617
6618                 pd.hdr.udp = &uh;
6619                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6620                     &action, &reason, AF_INET)) {
6621                         log = action != PF_PASS;
6622                         goto done;
6623                 }
6624                 if (uh.uh_dport == 0 ||
6625                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6626                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6627                         action = PF_DROP;
6628                         REASON_SET(&reason, PFRES_SHORT);
6629                         goto done;
6630                 }
6631                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6632                 if (action == PF_PASS) {
6633                         r = s->rule.ptr;
6634                         a = s->anchor.ptr;
6635                         log = s->log;
6636                 } else if (s == NULL) {
6637                         action = pf_test_rule(&r, &s, dir, kif,
6638                                               m, off, h, &pd, &a,
6639                                               &ruleset, NULL, inp);
6640                 }
6641                 break;
6642         }
6643
6644         case IPPROTO_ICMP: {
6645                 struct icmp     ih;
6646
6647                 pd.hdr.icmp = &ih;
6648                 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
6649                     &action, &reason, AF_INET)) {
6650                         log = action != PF_PASS;
6651                         goto done;
6652                 }
6653                 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
6654                                             &reason);
6655                 if (action == PF_PASS) {
6656                         r = s->rule.ptr;
6657                         a = s->anchor.ptr;
6658                         log = s->log;
6659                 } else if (s == NULL) {
6660                         action = pf_test_rule(&r, &s, dir, kif,
6661                                               m, off, h, &pd, &a,
6662                                               &ruleset, NULL, inp);
6663                 }
6664                 break;
6665         }
6666
6667         default:
6668                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6669                 if (action == PF_PASS) {
6670                         r = s->rule.ptr;
6671                         a = s->anchor.ptr;
6672                         log = s->log;
6673                 } else if (s == NULL) {
6674                         action = pf_test_rule(&r, &s, dir, kif, m, off, h,
6675                                               &pd, &a, &ruleset, NULL, inp);
6676                 }
6677                 break;
6678         }
6679
6680 done:
6681         if (action == PF_PASS && h->ip_hl > 5 &&
6682             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6683                 action = PF_DROP;
6684                 REASON_SET(&reason, PFRES_IPOPTIONS);
6685                 log = 1;
6686                 DPFPRINTF(PF_DEBUG_MISC,
6687                     ("pf: dropping packet with ip options\n"));
6688         }
6689
6690         if ((s && s->tag) || r->rtableid)
6691                 pf_tag_packet(m, s ? s->tag : 0, r->rtableid);
6692
6693 #if 0
6694         if (dir == PF_IN && s && s->key[PF_SK_STACK])
6695                 m->m_pkthdr.pf.statekey = s->key[PF_SK_STACK];
6696 #endif
6697
6698 #ifdef ALTQ
6699         /*
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.
6702          */
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;
6707                 else
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 */
6712                 if (s) {
6713                         /* for fairq */
6714                         m->m_pkthdr.pf.state_hash = s->hash;
6715                         m->m_pkthdr.pf.flags |= PF_TAG_STATE_HASHED;
6716                 }
6717         }
6718 #endif /* ALTQ */
6719
6720         /*
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().
6724          */
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;
6731
6732         if (dir == PF_IN && action == PF_PASS && r->divert.port) {
6733                 struct pf_divert *divert;
6734
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;
6739                 }
6740         }
6741
6742         if (log) {
6743                 struct pf_rule *lr;
6744
6745                 if (s != NULL && s->nat_rule.ptr != NULL &&
6746                     s->nat_rule.ptr->log & PF_LOG_ALL)
6747                         lr = s->nat_rule.ptr;
6748                 else
6749                         lr = r;
6750                 PFLOG_PACKET(kif, h, m, AF_INET, dir, reason, lr, a, ruleset,
6751                     &pd);
6752         }
6753
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]++;
6756
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;
6761                 if (a != NULL) {
6762                         a->packets[dirndx]++;
6763                         a->bytes[dirndx] += pd.tot_len;
6764                 }
6765                 if (s != NULL) {
6766                         if (s->nat_rule.ptr != NULL) {
6767                                 s->nat_rule.ptr->packets[dirndx]++;
6768                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6769                         }
6770                         if (s->src_node != NULL) {
6771                                 s->src_node->packets[dirndx]++;
6772                                 s->src_node->bytes[dirndx] += pd.tot_len;
6773                         }
6774                         if (s->nat_src_node != NULL) {
6775                                 s->nat_src_node->packets[dirndx]++;
6776                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6777                         }
6778                         dirndx = (dir == s->direction) ? 0 : 1;
6779                         s->packets[dirndx]++;
6780                         s->bytes[dirndx] += pd.tot_len;
6781                 }
6782                 tr = r;
6783                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6784                 if (nr != NULL && r == &pf_default_rule)
6785                         tr = nr;
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);
6800         }
6801
6802
6803         if (action == PF_SYNPROXY_DROP) {
6804                 m_freem(*m0);
6805                 *m0 = NULL;
6806                 action = PF_PASS;
6807         } else if (r->rt) {
6808                 /* pf_route can free the mbuf causing *m0 to become NULL */
6809                 pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
6810         }
6811
6812         return (action);
6813 }
6814 #endif /* INET */
6815
6816 #ifdef INET6
6817
6818 /*
6819  * WARNING: pf_token held shared on entry, THIS IS CPU LOCALIZED CODE
6820  */
6821 int
6822 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0,
6823     struct ether_header *eh, struct inpcb *inp)
6824 {
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;
6832         struct pf_pdesc          pd;
6833         int                      off, terminal = 0, dirndx, rh_cnt = 0;
6834
6835         if (!pf_status.running)
6836                 return (PF_PASS);
6837
6838         memset(&pd, 0, sizeof(pd));
6839 #ifdef foo
6840         if (ifp->if_type == IFT_CARP && ifp->if_carpdev)
6841                 kif = (struct pfi_kif *)ifp->if_carpdev->if_pf_kif;
6842         else
6843 #endif
6844                 kif = (struct pfi_kif *)ifp->if_pf_kif;
6845
6846         if (kif == NULL) {
6847                 DPFPRINTF(PF_DEBUG_URGENT,
6848                     ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6849                 return (PF_DROP);
6850         }
6851         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6852                 return (PF_PASS);
6853
6854 #ifdef DIAGNOSTIC
6855         if ((m->m_flags & M_PKTHDR) == 0)
6856                 panic("non-M_PKTHDR is passed to pf_test6");
6857 #endif /* DIAGNOSTIC */
6858
6859         if (m->m_pkthdr.len < (int)sizeof(*h)) {
6860                 action = PF_DROP;
6861                 REASON_SET(&reason, PFRES_SHORT);
6862                 log = 1;
6863                 goto done;
6864         }
6865
6866         /*
6867          * DragonFly doesn't zero the auxillary pkghdr fields, only fw_flags,
6868          * so make sure pf.flags is clear.
6869          */
6870         if (m->m_pkthdr.fw_flags & PF_MBUF_TAGGED)
6871                 return (PF_PASS);
6872         m->m_pkthdr.pf.flags = 0;
6873         /* Re-Check when updating to > 4.4 */
6874         m->m_pkthdr.pf.statekey = NULL;
6875
6876         /* We do IP header normalization and packet reassembly here */
6877         if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6878                 action = PF_DROP;
6879                 goto done;
6880         }
6881         m = *m0;        /* pf_normalize messes with m0 */
6882         h = mtod(m, struct ip6_hdr *);
6883
6884 #if 1
6885         /*
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.
6888          */
6889         if (htons(h->ip6_plen) == 0) {
6890                 action = PF_DROP;
6891                 REASON_SET(&reason, PFRES_NORM);        /*XXX*/
6892                 goto done;
6893         }
6894 #endif
6895
6896         pd.src = (struct pf_addr *)&h->ip6_src;
6897         pd.dst = (struct pf_addr *)&h->ip6_dst;
6898         pd.sport = pd.dport = NULL;
6899         pd.ip_sum = NULL;
6900         pd.proto_sum = NULL;
6901         pd.dir = dir;
6902         pd.sidx = (dir == PF_IN) ? 0 : 1;
6903         pd.didx = (dir == PF_IN) ? 1 : 0;
6904         pd.af = AF_INET6;
6905         pd.tos = 0;
6906         pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6907         pd.eh = eh;
6908
6909         off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6910         pd.proto = h->ip6_nxt;
6911         do {
6912                 switch (pd.proto) {
6913                 case IPPROTO_FRAGMENT:
6914                         action = pf_test_fragment(&r, dir, kif, m, h,
6915                             &pd, &a, &ruleset);
6916                         if (action == PF_DROP)
6917                                 REASON_SET(&reason, PFRES_FRAG);
6918                         goto done;
6919                 case IPPROTO_ROUTING: {
6920                         struct ip6_rthdr rthdr;
6921
6922                         if (rh_cnt++) {
6923                                 DPFPRINTF(PF_DEBUG_MISC,
6924                                     ("pf: IPv6 more than one rthdr\n"));
6925                                 action = PF_DROP;
6926                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6927                                 log = 1;
6928                                 goto done;
6929                         }
6930                         if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6931                             &reason, pd.af)) {
6932                                 DPFPRINTF(PF_DEBUG_MISC,
6933                                     ("pf: IPv6 short rthdr\n"));
6934                                 action = PF_DROP;
6935                                 REASON_SET(&reason, PFRES_SHORT);
6936                                 log = 1;
6937                                 goto done;
6938                         }
6939                         if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6940                                 DPFPRINTF(PF_DEBUG_MISC,
6941                                     ("pf: IPv6 rthdr0\n"));
6942                                 action = PF_DROP;
6943                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6944                                 log = 1;
6945                                 goto done;
6946                         }
6947                         /* FALLTHROUGH */
6948                 }
6949                 case IPPROTO_AH:
6950                 case IPPROTO_HOPOPTS:
6951                 case IPPROTO_DSTOPTS: {
6952                         /* get next header and header length */
6953                         struct ip6_ext  opt6;
6954
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"));
6959                                 action = PF_DROP;
6960                                 log = 1;
6961                                 goto done;
6962                         }
6963                         if (pd.proto == IPPROTO_AH)
6964                                 off += (opt6.ip6e_len + 2) * 4;
6965                         else
6966                                 off += (opt6.ip6e_len + 1) * 8;
6967                         pd.proto = opt6.ip6e_nxt;
6968                         /* goto the next header */
6969                         break;
6970                 }
6971                 default:
6972                         terminal++;
6973                         break;
6974                 }
6975         } while (!terminal);
6976
6977         /* if there's no routing header, use unmodified mbuf for checksumming */
6978         if (!n)
6979                 n = m;
6980
6981         switch (pd.proto) {
6982
6983         case IPPROTO_TCP: {
6984                 struct tcphdr   th;
6985
6986                 pd.hdr.tcp = &th;
6987                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6988                     &action, &reason, AF_INET6)) {
6989                         log = action != PF_PASS;
6990                         goto done;
6991                 }
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)
6995                         goto done;
6996                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6997                                            &reason);
6998                 if (action == PF_PASS) {
6999                         r = s->rule.ptr;
7000                         a = s->anchor.ptr;
7001                         log = s->log;
7002                 } else if (s == NULL) {
7003                         action = pf_test_rule(&r, &s, dir, kif,
7004                                               m, off, h, &pd, &a,
7005                                               &ruleset, NULL, inp);
7006                 }
7007                 break;
7008         }
7009
7010         case IPPROTO_UDP: {
7011                 struct udphdr   uh;
7012
7013                 pd.hdr.udp = &uh;
7014                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
7015                     &action, &reason, AF_INET6)) {
7016                         log = action != PF_PASS;
7017                         goto done;
7018                 }
7019                 if (uh.uh_dport == 0 ||
7020                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
7021                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
7022                         action = PF_DROP;
7023                         REASON_SET(&reason, PFRES_SHORT);
7024                         goto done;
7025                 }
7026                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
7027                 if (action == PF_PASS) {
7028                         r = s->rule.ptr;
7029                         a = s->anchor.ptr;
7030                         log = s->log;
7031                 } else if (s == NULL) {
7032                         action = pf_test_rule(&r, &s, dir, kif,
7033                                               m, off, h, &pd, &a,
7034                                               &ruleset, NULL, inp);
7035                 }
7036                 break;
7037         }
7038
7039         case IPPROTO_ICMPV6: {
7040                 struct icmp6_hdr        ih;
7041
7042                 pd.hdr.icmp6 = &ih;
7043                 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
7044                     &action, &reason, AF_INET6)) {
7045                         log = action != PF_PASS;
7046                         goto done;
7047                 }
7048                 action = pf_test_state_icmp(&s, dir, kif,
7049                                             m, off, h, &pd, &reason);
7050                 if (action == PF_PASS) {
7051                         r = s->rule.ptr;
7052                         a = s->anchor.ptr;
7053                         log = s->log;
7054                 } else if (s == NULL) {
7055                         action = pf_test_rule(&r, &s, dir, kif,
7056                                               m, off, h, &pd, &a,
7057                                               &ruleset, NULL, inp);
7058                 }
7059                 break;
7060         }
7061
7062         default:
7063                 action = pf_test_state_other(&s, dir, kif, m, &pd);
7064                 if (action == PF_PASS) {
7065                         r = s->rule.ptr;
7066                         a = s->anchor.ptr;
7067                         log = s->log;
7068                 } else if (s == NULL) {
7069                         action = pf_test_rule(&r, &s, dir, kif, m, off, h,
7070                                               &pd, &a, &ruleset, NULL, inp);
7071                 }
7072                 break;
7073         }
7074
7075 done:
7076         if (n != m) {
7077                 m_freem(n);
7078                 n = NULL;
7079         }
7080
7081         /* handle dangerous IPv6 extension headers. */
7082         if (action == PF_PASS && rh_cnt &&
7083             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
7084                 action = PF_DROP;
7085                 REASON_SET(&reason, PFRES_IPOPTIONS);
7086                 log = 1;
7087                 DPFPRINTF(PF_DEBUG_MISC,
7088                     ("pf: dropping packet with dangerous v6 headers\n"));
7089         }
7090
7091         if ((s && s->tag) || r->rtableid)
7092                 pf_tag_packet(m, s ? s->tag : 0, r->rtableid);
7093
7094 #if 0
7095         if (dir == PF_IN && s && s->key[PF_SK_STACK])
7096                 m->m_pkthdr.pf.statekey = s->key[PF_SK_STACK];
7097 #endif
7098
7099 #ifdef ALTQ
7100         /*
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.
7103          */
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;
7108                 else
7109                         m->m_pkthdr.pf.qid = r->qid;
7110                 m->m_pkthdr.pf.ecn_af = AF_INET6;
7111                 m->m_pkthdr.pf.hdr = h;
7112                 if (s) {
7113                         /* for fairq */
7114                         m->m_pkthdr.pf.state_hash = s->hash;
7115                         m->m_pkthdr.pf.flags |= PF_TAG_STATE_HASHED;
7116                 }
7117         }
7118 #endif /* ALTQ */
7119
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;
7126
7127         if (dir == PF_IN && action == PF_PASS && r->divert.port) {
7128                 struct pf_divert *divert;
7129
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;
7134                 }
7135         }
7136
7137         if (log) {
7138                 struct pf_rule *lr;
7139
7140                 if (s != NULL && s->nat_rule.ptr != NULL &&
7141                     s->nat_rule.ptr->log & PF_LOG_ALL)
7142                         lr = s->nat_rule.ptr;
7143                 else
7144                         lr = r;
7145                 PFLOG_PACKET(kif, h, m, AF_INET6, dir, reason, lr, a, ruleset,
7146                     &pd);
7147         }
7148
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]++;
7151
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;
7156                 if (a != NULL) {
7157                         a->packets[dirndx]++;
7158                         a->bytes[dirndx] += pd.tot_len;
7159                 }
7160                 if (s != NULL) {
7161                         if (s->nat_rule.ptr != NULL) {
7162                                 s->nat_rule.ptr->packets[dirndx]++;
7163                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
7164                         }
7165                         if (s->src_node != NULL) {
7166                                 s->src_node->packets[dirndx]++;
7167                                 s->src_node->bytes[dirndx] += pd.tot_len;
7168                         }
7169                         if (s->nat_src_node != NULL) {
7170                                 s->nat_src_node->packets[dirndx]++;
7171                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
7172                         }
7173                         dirndx = (dir == s->direction) ? 0 : 1;
7174                         s->packets[dirndx]++;
7175                         s->bytes[dirndx] += pd.tot_len;
7176                 }
7177                 tr = r;
7178                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
7179                 if (nr != NULL && r == &pf_default_rule)
7180                         tr = nr;
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);
7193         }
7194
7195
7196         if (action == PF_SYNPROXY_DROP) {
7197                 m_freem(*m0);
7198                 *m0 = NULL;
7199                 action = PF_PASS;
7200         } else if (r->rt)
7201                 /* pf_route6 can free the mbuf causing *m0 to become NULL */
7202                 pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
7203
7204         return (action);
7205 }
7206 #endif /* INET6 */
7207
7208 int
7209 pf_check_congestion(struct ifqueue *ifq)
7210 {
7211                 return (0);
7212 }