sys/netpfil/pf/pf.c

   1 /*-
   2  * Copyright (c) 2001 Daniel Hartmeier
   3  * Copyright (c) 2002 - 2008 Henning Brauer
   4  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  *    - Redistributions of source code must retain the above copyright
  12  *      notice, this list of conditions and the following disclaimer.
  13  *    - Redistributions in binary form must reproduce the above
  14  *      copyright notice, this list of conditions and the following
  15  *      disclaimer in the documentation and/or other materials provided
  16  *      with the distribution.
  17  *
  18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  19  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  20  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  21  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
  22  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
  24  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  25  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
  26  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
  28  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  29  * POSSIBILITY OF SUCH DAMAGE.
  30  *
  31  * Effort sponsored in part by the Defense Advanced Research Projects
  32  * Agency (DARPA) and Air Force Research Laboratory, Air Force
  33  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
  34  *
  35  *      $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
  36  */
  37
  38 #include <sys/cdefs.h>
  39 __FBSDID("$FreeBSD$");
  40
  41 #include "opt_inet.h"
  42 #include "opt_inet6.h"
  43 #include "opt_bpf.h"
  44 #include "opt_pf.h"
  45
  46 #include <sys/param.h>
  47 #include <sys/bus.h>
  48 #include <sys/endian.h>
  49 #include <sys/hash.h>
  50 #include <sys/interrupt.h>
  51 #include <sys/kernel.h>
  52 #include <sys/kthread.h>
  53 #include <sys/limits.h>
  54 #include <sys/mbuf.h>
  55 #include <sys/md5.h>
  56 #include <sys/random.h>
  57 #include <sys/refcount.h>
  58 #include <sys/socket.h>
  59 #include <sys/sysctl.h>
  60 #include <sys/taskqueue.h>
  61 #include <sys/ucred.h>
  62
  63 #include <net/if.h>
  64 #include <net/if_types.h>
  65 #include <net/route.h>
  66 #include <net/radix_mpath.h>
  67 #include <net/vnet.h>
  68
  69 #include <net/pfvar.h>
  70 #include <net/if_pflog.h>
  71 #include <net/if_pfsync.h>
  72
  73 #include <netinet/in_pcb.h>
  74 #include <netinet/in_var.h>
  75 #include <netinet/ip.h>
  76 #include <netinet/ip_fw.h>
  77 #include <netinet/ip_icmp.h>
  78 #include <netinet/icmp_var.h>
  79 #include <netinet/ip_var.h>
  80 #include <netinet/tcp.h>
  81 #include <netinet/tcp_fsm.h>
  82 #include <netinet/tcp_seq.h>
  83 #include <netinet/tcp_timer.h>
  84 #include <netinet/tcp_var.h>
  85 #include <netinet/udp.h>
  86 #include <netinet/udp_var.h>
  87
  88 #include <netpfil/ipfw/ip_fw_private.h> /* XXX: only for DIR_IN/DIR_OUT */
  89
  90 #ifdef INET6
  91 #include <netinet/ip6.h>
  92 #include <netinet/icmp6.h>
  93 #include <netinet6/nd6.h>
  94 #include <netinet6/ip6_var.h>
  95 #include <netinet6/in6_pcb.h>
  96 #endif /* INET6 */
  97
  98 #include <machine/in_cksum.h>
  99 #include <security/mac/mac_framework.h>
 100
 101 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
 102
 103 /*
 104  * Global variables
 105  */
 106
 107 /* state tables */
 108 VNET_DEFINE(struct pf_altqqueue,         pf_altqs[2]);
 109 VNET_DEFINE(struct pf_palist,            pf_pabuf);
 110 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_active);
 111 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_inactive);
 112 VNET_DEFINE(struct pf_status,            pf_status);
 113
 114 VNET_DEFINE(u_int32_t,                   ticket_altqs_active);
 115 VNET_DEFINE(u_int32_t,                   ticket_altqs_inactive);
 116 VNET_DEFINE(int,                         altqs_inactive_open);
 117 VNET_DEFINE(u_int32_t,                   ticket_pabuf);
 118
 119 VNET_DEFINE(MD5_CTX,                     pf_tcp_secret_ctx);
 120 #define V_pf_tcp_secret_ctx              VNET(pf_tcp_secret_ctx)
 121 VNET_DEFINE(u_char,                      pf_tcp_secret[16]);
 122 #define V_pf_tcp_secret                  VNET(pf_tcp_secret)
 123 VNET_DEFINE(int,                         pf_tcp_secret_init);
 124 #define V_pf_tcp_secret_init             VNET(pf_tcp_secret_init)
 125 VNET_DEFINE(int,                         pf_tcp_iss_off);
 126 #define V_pf_tcp_iss_off                 VNET(pf_tcp_iss_off)
 127
 128 /*
 129  * Queue for pf_intr() sends.
 130  */
 131 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
 132 struct pf_send_entry {
 133         STAILQ_ENTRY(pf_send_entry)     pfse_next;
 134         struct mbuf                     *pfse_m;
 135         enum {
 136                 PFSE_IP,
 137                 PFSE_IP6,
 138                 PFSE_ICMP,
 139                 PFSE_ICMP6,
 140         }                               pfse_type;
 141         union {
 142                 struct route            ro;
 143                 struct {
 144                         int             type;
 145                         int             code;
 146                         int             mtu;
 147                 } icmpopts;
 148         } u;
 149 #define pfse_ro         u.ro
 150 #define pfse_icmp_type  u.icmpopts.type
 151 #define pfse_icmp_code  u.icmpopts.code
 152 #define pfse_icmp_mtu   u.icmpopts.mtu
 153 };
 154
 155 STAILQ_HEAD(pf_send_head, pf_send_entry);
 156 static VNET_DEFINE(struct pf_send_head, pf_sendqueue);
 157 #define V_pf_sendqueue  VNET(pf_sendqueue)
 158
 159 static struct mtx pf_sendqueue_mtx;
 160 #define PF_SENDQ_LOCK()         mtx_lock(&pf_sendqueue_mtx)
 161 #define PF_SENDQ_UNLOCK()       mtx_unlock(&pf_sendqueue_mtx)
 162
 163 /*
 164  * Queue for pf_overload_task() tasks.
 165  */
 166 struct pf_overload_entry {
 167         SLIST_ENTRY(pf_overload_entry)  next;
 168         struct pf_addr                  addr;
 169         sa_family_t                     af;
 170         uint8_t                         dir;
 171         struct pf_rule                  *rule;
 172 };
 173
 174 SLIST_HEAD(pf_overload_head, pf_overload_entry);
 175 static VNET_DEFINE(struct pf_overload_head, pf_overloadqueue);
 176 #define V_pf_overloadqueue      VNET(pf_overloadqueue)
 177 static VNET_DEFINE(struct task, pf_overloadtask);
 178 #define V_pf_overloadtask       VNET(pf_overloadtask)
 179
 180 static struct mtx pf_overloadqueue_mtx;
 181 #define PF_OVERLOADQ_LOCK()     mtx_lock(&pf_overloadqueue_mtx)
 182 #define PF_OVERLOADQ_UNLOCK()   mtx_unlock(&pf_overloadqueue_mtx)
 183
 184 VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules);
 185 struct mtx pf_unlnkdrules_mtx;
 186
 187 static VNET_DEFINE(uma_zone_t,  pf_sources_z);
 188 #define V_pf_sources_z  VNET(pf_sources_z)
 189 static VNET_DEFINE(uma_zone_t,  pf_mtag_z);
 190 #define V_pf_mtag_z     VNET(pf_mtag_z)
 191 VNET_DEFINE(uma_zone_t,  pf_state_z);
 192 VNET_DEFINE(uma_zone_t,  pf_state_key_z);
 193
 194 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
 195 #define PFID_CPUBITS    8
 196 #define PFID_CPUSHIFT   (sizeof(uint64_t) * NBBY - PFID_CPUBITS)
 197 #define PFID_CPUMASK    ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT)
 198 #define PFID_MAXID      (~PFID_CPUMASK)
 199 CTASSERT((1 << PFID_CPUBITS) > MAXCPU);
 200
 201 static void              pf_src_tree_remove_state(struct pf_state *);
 202 static void              pf_init_threshold(struct pf_threshold *, u_int32_t,
 203                             u_int32_t);
 204 static void              pf_add_threshold(struct pf_threshold *);
 205 static int               pf_check_threshold(struct pf_threshold *);
 206
 207 static void              pf_change_ap(struct pf_addr *, u_int16_t *,
 208                             u_int16_t *, u_int16_t *, struct pf_addr *,
 209                             u_int16_t, u_int8_t, sa_family_t);
 210 static int               pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
 211                             struct tcphdr *, struct pf_state_peer *);
 212 static void              pf_change_icmp(struct pf_addr *, u_int16_t *,
 213                             struct pf_addr *, struct pf_addr *, u_int16_t,
 214                             u_int16_t *, u_int16_t *, u_int16_t *,
 215                             u_int16_t *, u_int8_t, sa_family_t);
 216 static void              pf_send_tcp(struct mbuf *,
 217                             const struct pf_rule *, sa_family_t,
 218                             const struct pf_addr *, const struct pf_addr *,
 219                             u_int16_t, u_int16_t, u_int32_t, u_int32_t,
 220                             u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
 221                             u_int16_t, struct ifnet *);
 222 static void              pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
 223                             sa_family_t, struct pf_rule *);
 224 static void              pf_detach_state(struct pf_state *);
 225 static int               pf_state_key_attach(struct pf_state_key *,
 226                             struct pf_state_key *, struct pf_state *);
 227 static void              pf_state_key_detach(struct pf_state *, int);
 228 static int               pf_state_key_ctor(void *, int, void *, int);
 229 static u_int32_t         pf_tcp_iss(struct pf_pdesc *);
 230 static int               pf_test_rule(struct pf_rule **, struct pf_state **,
 231                             int, struct pfi_kif *, struct mbuf *, int,
 232                             struct pf_pdesc *, struct pf_rule **,
 233                             struct pf_ruleset **, struct inpcb *);
 234 static int               pf_create_state(struct pf_rule *, struct pf_rule *,
 235                             struct pf_rule *, struct pf_pdesc *,
 236                             struct pf_src_node *, struct pf_state_key *,
 237                             struct pf_state_key *, struct mbuf *, int,
 238                             u_int16_t, u_int16_t, int *, struct pfi_kif *,
 239                             struct pf_state **, int, u_int16_t, u_int16_t,
 240                             int);
 241 static int               pf_test_fragment(struct pf_rule **, int,
 242                             struct pfi_kif *, struct mbuf *, void *,
 243                             struct pf_pdesc *, struct pf_rule **,
 244                             struct pf_ruleset **);
 245 static int               pf_tcp_track_full(struct pf_state_peer *,
 246                             struct pf_state_peer *, struct pf_state **,
 247                             struct pfi_kif *, struct mbuf *, int,
 248                             struct pf_pdesc *, u_short *, int *);
 249 static int               pf_tcp_track_sloppy(struct pf_state_peer *,
 250                             struct pf_state_peer *, struct pf_state **,
 251                             struct pf_pdesc *, u_short *);
 252 static int               pf_test_state_tcp(struct pf_state **, int,
 253                             struct pfi_kif *, struct mbuf *, int,
 254                             void *, struct pf_pdesc *, u_short *);
 255 static int               pf_test_state_udp(struct pf_state **, int,
 256                             struct pfi_kif *, struct mbuf *, int,
 257                             void *, struct pf_pdesc *);
 258 static int               pf_test_state_icmp(struct pf_state **, int,
 259                             struct pfi_kif *, struct mbuf *, int,
 260                             void *, struct pf_pdesc *, u_short *);
 261 static int               pf_test_state_other(struct pf_state **, int,
 262                             struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
 263 static u_int8_t          pf_get_wscale(struct mbuf *, int, u_int16_t,
 264                             sa_family_t);
 265 static u_int16_t         pf_get_mss(struct mbuf *, int, u_int16_t,
 266                             sa_family_t);
 267 static u_int16_t         pf_calc_mss(struct pf_addr *, sa_family_t,
 268                                 int, u_int16_t);
 269 static void              pf_set_rt_ifp(struct pf_state *,
 270                             struct pf_addr *);
 271 static int               pf_check_proto_cksum(struct mbuf *, int, int,
 272                             u_int8_t, sa_family_t);
 273 static void              pf_print_state_parts(struct pf_state *,
 274                             struct pf_state_key *, struct pf_state_key *);
 275 static int               pf_addr_wrap_neq(struct pf_addr_wrap *,
 276                             struct pf_addr_wrap *);
 277 static struct pf_state  *pf_find_state(struct pfi_kif *,
 278                             struct pf_state_key_cmp *, u_int);
 279 static int               pf_src_connlimit(struct pf_state **);
 280 static void              pf_overload_task(void *c, int pending);
 281 static int               pf_insert_src_node(struct pf_src_node **,
 282                             struct pf_rule *, struct pf_addr *, sa_family_t);
 283 static u_int             pf_purge_expired_states(u_int, int);
 284 static void              pf_purge_unlinked_rules(void);
 285 static int               pf_mtag_init(void *, int, int);
 286 static void              pf_mtag_free(struct m_tag *);
 287 #ifdef INET
 288 static void              pf_route(struct mbuf **, struct pf_rule *, int,
 289                             struct ifnet *, struct pf_state *,
 290                             struct pf_pdesc *);
 291 #endif /* INET */
 292 #ifdef INET6
 293 static void              pf_change_a6(struct pf_addr *, u_int16_t *,
 294                             struct pf_addr *, u_int8_t);
 295 static void              pf_route6(struct mbuf **, struct pf_rule *, int,
 296                             struct ifnet *, struct pf_state *,
 297                             struct pf_pdesc *);
 298 #endif /* INET6 */
 299
 300 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
 301
 302 VNET_DECLARE(int, pf_end_threads);
 303
 304 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
 305
 306 #define PACKET_LOOPED(pd)       ((pd)->pf_mtag &&                       \
 307                                  (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
 308
 309 #define STATE_LOOKUP(i, k, d, s, pd)                                    \
 310         do {                                                            \
 311                 (s) = pf_find_state((i), (k), (d));                     \
 312                 if ((s) == NULL)                                        \
 313                         return (PF_DROP);                               \
 314                 if (PACKET_LOOPED(pd))                                  \
 315                         return (PF_PASS);                               \
 316                 if ((d) == PF_OUT &&                                    \
 317                     (((s)->rule.ptr->rt == PF_ROUTETO &&                \
 318                     (s)->rule.ptr->direction == PF_OUT) ||              \
 319                     ((s)->rule.ptr->rt == PF_REPLYTO &&                 \
 320                     (s)->rule.ptr->direction == PF_IN)) &&              \
 321                     (s)->rt_kif != NULL &&                              \
 322                     (s)->rt_kif != (i))                                 \
 323                         return (PF_PASS);                               \
 324         } while (0)
 325
 326 #define BOUND_IFACE(r, k) \
 327         ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
 328
 329 #define STATE_INC_COUNTERS(s)                                           \
 330         do {                                                            \
 331                 counter_u64_add(s->rule.ptr->states_cur, 1);            \
 332                 counter_u64_add(s->rule.ptr->states_tot, 1);            \
 333                 if (s->anchor.ptr != NULL) {                            \
 334                         counter_u64_add(s->anchor.ptr->states_cur, 1);  \
 335                         counter_u64_add(s->anchor.ptr->states_tot, 1);  \
 336                 }                                                       \
 337                 if (s->nat_rule.ptr != NULL) {                          \
 338                         counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
 339                         counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
 340                 }                                                       \
 341         } while (0)
 342
 343 #define STATE_DEC_COUNTERS(s)                                           \
 344         do {                                                            \
 345                 if (s->nat_rule.ptr != NULL)                            \
 346                         counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
 347                 if (s->anchor.ptr != NULL)                              \
 348                         counter_u64_add(s->anchor.ptr->states_cur, -1); \
 349                 counter_u64_add(s->rule.ptr->states_cur, -1);           \
 350         } while (0)
 351
 352 static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
 353 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
 354 VNET_DEFINE(struct pf_idhash *, pf_idhash);
 355 VNET_DEFINE(u_long, pf_hashmask);
 356 VNET_DEFINE(struct pf_srchash *, pf_srchash);
 357 VNET_DEFINE(u_long, pf_srchashmask);
 358
 359 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)");
 360
 361 VNET_DEFINE(u_long, pf_hashsize);
 362 #define V_pf_hashsize   VNET(pf_hashsize)
 363 SYSCTL_VNET_UINT(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
 364     &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable");
 365
 366 VNET_DEFINE(u_long, pf_srchashsize);
 367 #define V_pf_srchashsize        VNET(pf_srchashsize)
 368 SYSCTL_VNET_UINT(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
 369     &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable");
 370
 371 VNET_DEFINE(void *, pf_swi_cookie);
 372
 373 VNET_DEFINE(uint32_t, pf_hashseed);
 374 #define V_pf_hashseed   VNET(pf_hashseed)
 375
 376 static __inline uint32_t
 377 pf_hashkey(struct pf_state_key *sk)
 378 {
 379         uint32_t h;
 380
 381         h = jenkins_hash32((uint32_t *)sk,
 382             sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
 383             V_pf_hashseed);
 384
 385         return (h & V_pf_hashmask);
 386 }
 387
 388 static __inline uint32_t
 389 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
 390 {
 391         uint32_t h;
 392
 393         switch (af) {
 394         case AF_INET:
 395                 h = jenkins_hash32((uint32_t *)&addr->v4,
 396                     sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
 397                 break;
 398         case AF_INET6:
 399                 h = jenkins_hash32((uint32_t *)&addr->v6,
 400                     sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
 401                 break;
 402         default:
 403                 panic("%s: unknown address family %u", __func__, af);
 404         }
 405
 406         return (h & V_pf_srchashmask);
 407 }
 408
 409 #ifdef INET6
 410 void
 411 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
 412 {
 413         switch (af) {
 414 #ifdef INET
 415         case AF_INET:
 416                 dst->addr32[0] = src->addr32[0];
 417                 break;
 418 #endif /* INET */
 419         case AF_INET6:
 420                 dst->addr32[0] = src->addr32[0];
 421                 dst->addr32[1] = src->addr32[1];
 422                 dst->addr32[2] = src->addr32[2];
 423                 dst->addr32[3] = src->addr32[3];
 424                 break;
 425         }
 426 }
 427 #endif /* INET6 */
 428
 429 static void
 430 pf_init_threshold(struct pf_threshold *threshold,
 431     u_int32_t limit, u_int32_t seconds)
 432 {
 433         threshold->limit = limit * PF_THRESHOLD_MULT;
 434         threshold->seconds = seconds;
 435         threshold->count = 0;
 436         threshold->last = time_uptime;
 437 }
 438
 439 static void
 440 pf_add_threshold(struct pf_threshold *threshold)
 441 {
 442         u_int32_t t = time_uptime, diff = t - threshold->last;
 443
 444         if (diff >= threshold->seconds)
 445                 threshold->count = 0;
 446         else
 447                 threshold->count -= threshold->count * diff /
 448                     threshold->seconds;
 449         threshold->count += PF_THRESHOLD_MULT;
 450         threshold->last = t;
 451 }
 452
 453 static int
 454 pf_check_threshold(struct pf_threshold *threshold)
 455 {
 456         return (threshold->count > threshold->limit);
 457 }
 458
 459 static int
 460 pf_src_connlimit(struct pf_state **state)
 461 {
 462         struct pf_overload_entry *pfoe;
 463         int bad = 0;
 464
 465         PF_STATE_LOCK_ASSERT(*state);
 466
 467         (*state)->src_node->conn++;
 468         (*state)->src.tcp_est = 1;
 469         pf_add_threshold(&(*state)->src_node->conn_rate);
 470
 471         if ((*state)->rule.ptr->max_src_conn &&
 472             (*state)->rule.ptr->max_src_conn <
 473             (*state)->src_node->conn) {
 474                 V_pf_status.lcounters[LCNT_SRCCONN]++;
 475                 bad++;
 476         }
 477
 478         if ((*state)->rule.ptr->max_src_conn_rate.limit &&
 479             pf_check_threshold(&(*state)->src_node->conn_rate)) {
 480                 V_pf_status.lcounters[LCNT_SRCCONNRATE]++;
 481                 bad++;
 482         }
 483
 484         if (!bad)
 485                 return (0);
 486
 487         /* Kill this state. */
 488         (*state)->timeout = PFTM_PURGE;
 489         (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
 490
 491         if ((*state)->rule.ptr->overload_tbl == NULL)
 492                 return (1);
 493
 494         /* Schedule overloading and flushing task. */
 495         pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
 496         if (pfoe == NULL)
 497                 return (1);     /* too bad :( */
 498
 499         bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
 500         pfoe->af = (*state)->key[PF_SK_WIRE]->af;
 501         pfoe->rule = (*state)->rule.ptr;
 502         pfoe->dir = (*state)->direction;
 503         PF_OVERLOADQ_LOCK();
 504         SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
 505         PF_OVERLOADQ_UNLOCK();
 506         taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
 507
 508         return (1);
 509 }
 510
 511 static void
 512 pf_overload_task(void *c, int pending)
 513 {
 514         struct pf_overload_head queue;
 515         struct pfr_addr p;
 516         struct pf_overload_entry *pfoe, *pfoe1;
 517         uint32_t killed = 0;
 518
 519         PF_OVERLOADQ_LOCK();
 520         queue = *(struct pf_overload_head *)c;
 521         SLIST_INIT((struct pf_overload_head *)c);
 522         PF_OVERLOADQ_UNLOCK();
 523
 524         bzero(&p, sizeof(p));
 525         SLIST_FOREACH(pfoe, &queue, next) {
 526                 V_pf_status.lcounters[LCNT_OVERLOAD_TABLE]++;
 527                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
 528                         printf("%s: blocking address ", __func__);
 529                         pf_print_host(&pfoe->addr, 0, pfoe->af);
 530                         printf("\n");
 531                 }
 532
 533                 p.pfra_af = pfoe->af;
 534                 switch (pfoe->af) {
 535 #ifdef INET
 536                 case AF_INET:
 537                         p.pfra_net = 32;
 538                         p.pfra_ip4addr = pfoe->addr.v4;
 539                         break;
 540 #endif
 541 #ifdef INET6
 542                 case AF_INET6:
 543                         p.pfra_net = 128;
 544                         p.pfra_ip6addr = pfoe->addr.v6;
 545                         break;
 546 #endif
 547                 }
 548
 549                 PF_RULES_WLOCK();
 550                 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
 551                 PF_RULES_WUNLOCK();
 552         }
 553
 554         /*
 555          * Remove those entries, that don't need flushing.
 556          */
 557         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 558                 if (pfoe->rule->flush == 0) {
 559                         SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
 560                         free(pfoe, M_PFTEMP);
 561                 } else
 562                         V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH]++;
 563
 564         /* If nothing to flush, return. */
 565         if (SLIST_EMPTY(&queue))
 566                 return;
 567
 568         for (int i = 0; i <= V_pf_hashmask; i++) {
 569                 struct pf_idhash *ih = &V_pf_idhash[i];
 570                 struct pf_state_key *sk;
 571                 struct pf_state *s;
 572
 573                 PF_HASHROW_LOCK(ih);
 574                 LIST_FOREACH(s, &ih->states, entry) {
 575                     sk = s->key[PF_SK_WIRE];
 576                     SLIST_FOREACH(pfoe, &queue, next)
 577                         if (sk->af == pfoe->af &&
 578                             ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
 579                             pfoe->rule == s->rule.ptr) &&
 580                             ((pfoe->dir == PF_OUT &&
 581                             PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
 582                             (pfoe->dir == PF_IN &&
 583                             PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
 584                                 s->timeout = PFTM_PURGE;
 585                                 s->src.state = s->dst.state = TCPS_CLOSED;
 586                                 killed++;
 587                         }
 588                 }
 589                 PF_HASHROW_UNLOCK(ih);
 590         }
 591         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 592                 free(pfoe, M_PFTEMP);
 593         if (V_pf_status.debug >= PF_DEBUG_MISC)
 594                 printf("%s: %u states killed", __func__, killed);
 595 }
 596
 597 /*
 598  * Can return locked on failure, so that we can consistently
 599  * allocate and insert a new one.
 600  */
 601 struct pf_src_node *
 602 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
 603         int returnlocked)
 604 {
 605         struct pf_srchash *sh;
 606         struct pf_src_node *n;
 607
 608         V_pf_status.scounters[SCNT_SRC_NODE_SEARCH]++;
 609
 610         sh = &V_pf_srchash[pf_hashsrc(src, af)];
 611         PF_HASHROW_LOCK(sh);
 612         LIST_FOREACH(n, &sh->nodes, entry)
 613                 if (n->rule.ptr == rule && n->af == af &&
 614                     ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
 615                     (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
 616                         break;
 617         if (n != NULL || returnlocked == 0)
 618                 PF_HASHROW_UNLOCK(sh);
 619
 620         return (n);
 621 }
 622
 623 static int
 624 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
 625     struct pf_addr *src, sa_family_t af)
 626 {
 627
 628         KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
 629             rule->rpool.opts & PF_POOL_STICKYADDR),
 630             ("%s for non-tracking rule %p", __func__, rule));
 631
 632         if (*sn == NULL)
 633                 *sn = pf_find_src_node(src, rule, af, 1);
 634
 635         if (*sn == NULL) {
 636                 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
 637
 638                 PF_HASHROW_ASSERT(sh);
 639
 640                 if (!rule->max_src_nodes ||
 641                     counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
 642                         (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
 643                 else
 644                         V_pf_status.lcounters[LCNT_SRCNODES]++;
 645                 if ((*sn) == NULL) {
 646                         PF_HASHROW_UNLOCK(sh);
 647                         return (-1);
 648                 }
 649
 650                 pf_init_threshold(&(*sn)->conn_rate,
 651                     rule->max_src_conn_rate.limit,
 652                     rule->max_src_conn_rate.seconds);
 653
 654                 (*sn)->af = af;
 655                 (*sn)->rule.ptr = rule;
 656                 PF_ACPY(&(*sn)->addr, src, af);
 657                 LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
 658                 (*sn)->creation = time_uptime;
 659                 (*sn)->ruletype = rule->action;
 660                 if ((*sn)->rule.ptr != NULL)
 661                         counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
 662                 PF_HASHROW_UNLOCK(sh);
 663                 V_pf_status.scounters[SCNT_SRC_NODE_INSERT]++;
 664                 V_pf_status.src_nodes++;
 665         } else {
 666                 if (rule->max_src_states &&
 667                     (*sn)->states >= rule->max_src_states) {
 668                         V_pf_status.lcounters[LCNT_SRCSTATES]++;
 669                         return (-1);
 670                 }
 671         }
 672         return (0);
 673 }
 674
 675 void
 676 pf_unlink_src_node_locked(struct pf_src_node *src)
 677 {
 678 #ifdef INVARIANTS
 679         struct pf_srchash *sh;
 680
 681         sh = &V_pf_srchash[pf_hashsrc(&src->addr, src->af)];
 682         PF_HASHROW_ASSERT(sh);
 683 #endif
 684         LIST_REMOVE(src, entry);
 685         if (src->rule.ptr)
 686                 counter_u64_add(src->rule.ptr->src_nodes, -1);
 687         V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS]++;
 688         V_pf_status.src_nodes--;
 689 }
 690
 691 void
 692 pf_unlink_src_node(struct pf_src_node *src)
 693 {
 694         struct pf_srchash *sh;
 695
 696         sh = &V_pf_srchash[pf_hashsrc(&src->addr, src->af)];
 697         PF_HASHROW_LOCK(sh);
 698         pf_unlink_src_node_locked(src);
 699         PF_HASHROW_UNLOCK(sh);
 700 }
 701
 702 static void
 703 pf_free_src_node(struct pf_src_node *sn)
 704 {
 705
 706         KASSERT(sn->states == 0, ("%s: %p has refs", __func__, sn));
 707         uma_zfree(V_pf_sources_z, sn);
 708 }
 709
 710 u_int
 711 pf_free_src_nodes(struct pf_src_node_list *head)
 712 {
 713         struct pf_src_node *sn, *tmp;
 714         u_int count = 0;
 715
 716         LIST_FOREACH_SAFE(sn, head, entry, tmp) {
 717                 pf_free_src_node(sn);
 718                 count++;
 719         }
 720
 721         return (count);
 722 }
 723
 724 /* Data storage structures initialization. */
 725 void
 726 pf_initialize()
 727 {
 728         struct pf_keyhash       *kh;
 729         struct pf_idhash        *ih;
 730         struct pf_srchash       *sh;
 731         u_int i;
 732
 733         TUNABLE_ULONG_FETCH("net.pf.states_hashsize", &V_pf_hashsize);
 734         if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize))
 735                 V_pf_hashsize = PF_HASHSIZ;
 736         TUNABLE_ULONG_FETCH("net.pf.source_nodes_hashsize", &V_pf_srchashsize);
 737         if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize))
 738                 V_pf_srchashsize = PF_HASHSIZ / 4;
 739
 740         V_pf_hashseed = arc4random();
 741
 742         /* States and state keys storage. */
 743         V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
 744             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 745         V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
 746         uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
 747         uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
 748
 749         V_pf_state_key_z = uma_zcreate("pf state keys",
 750             sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
 751             UMA_ALIGN_PTR, 0);
 752         V_pf_keyhash = malloc(V_pf_hashsize * sizeof(struct pf_keyhash),
 753             M_PFHASH, M_WAITOK | M_ZERO);
 754         V_pf_idhash = malloc(V_pf_hashsize * sizeof(struct pf_idhash),
 755             M_PFHASH, M_WAITOK | M_ZERO);
 756         V_pf_hashmask = V_pf_hashsize - 1;
 757         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
 758             i++, kh++, ih++) {
 759                 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
 760                 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
 761         }
 762
 763         /* Source nodes. */
 764         V_pf_sources_z = uma_zcreate("pf source nodes",
 765             sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
 766             0);
 767         V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
 768         uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
 769         uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
 770         V_pf_srchash = malloc(V_pf_srchashsize * sizeof(struct pf_srchash),
 771           M_PFHASH, M_WAITOK|M_ZERO);
 772         V_pf_srchashmask = V_pf_srchashsize - 1;
 773         for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++)
 774                 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
 775
 776         /* ALTQ */
 777         TAILQ_INIT(&V_pf_altqs[0]);
 778         TAILQ_INIT(&V_pf_altqs[1]);
 779         TAILQ_INIT(&V_pf_pabuf);
 780         V_pf_altqs_active = &V_pf_altqs[0];
 781         V_pf_altqs_inactive = &V_pf_altqs[1];
 782
 783         /* Mbuf tags */
 784         V_pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
 785             sizeof(struct pf_mtag), NULL, NULL, pf_mtag_init, NULL,
 786             UMA_ALIGN_PTR, 0);
 787
 788         /* Send & overload+flush queues. */
 789         STAILQ_INIT(&V_pf_sendqueue);
 790         SLIST_INIT(&V_pf_overloadqueue);
 791         TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, &V_pf_overloadqueue);
 792         mtx_init(&pf_sendqueue_mtx, "pf send queue", NULL, MTX_DEF);
 793         mtx_init(&pf_overloadqueue_mtx, "pf overload/flush queue", NULL,
 794             MTX_DEF);
 795
 796         /* Unlinked, but may be referenced rules. */
 797         TAILQ_INIT(&V_pf_unlinked_rules);
 798         mtx_init(&pf_unlnkdrules_mtx, "pf unlinked rules", NULL, MTX_DEF);
 799 }
 800
 801 void
 802 pf_cleanup()
 803 {
 804         struct pf_keyhash       *kh;
 805         struct pf_idhash        *ih;
 806         struct pf_srchash       *sh;
 807         struct pf_send_entry    *pfse, *next;
 808         u_int i;
 809
 810         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
 811             i++, kh++, ih++) {
 812                 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
 813                     __func__));
 814                 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
 815                     __func__));
 816                 mtx_destroy(&kh->lock);
 817                 mtx_destroy(&ih->lock);
 818         }
 819         free(V_pf_keyhash, M_PFHASH);
 820         free(V_pf_idhash, M_PFHASH);
 821
 822         for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
 823                 KASSERT(LIST_EMPTY(&sh->nodes),
 824                     ("%s: source node hash not empty", __func__));
 825                 mtx_destroy(&sh->lock);
 826         }
 827         free(V_pf_srchash, M_PFHASH);
 828
 829         STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
 830                 m_freem(pfse->pfse_m);
 831                 free(pfse, M_PFTEMP);
 832         }
 833
 834         mtx_destroy(&pf_sendqueue_mtx);
 835         mtx_destroy(&pf_overloadqueue_mtx);
 836         mtx_destroy(&pf_unlnkdrules_mtx);
 837
 838         uma_zdestroy(V_pf_mtag_z);
 839         uma_zdestroy(V_pf_sources_z);
 840         uma_zdestroy(V_pf_state_z);
 841         uma_zdestroy(V_pf_state_key_z);
 842 }
 843
 844 static int
 845 pf_mtag_init(void *mem, int size, int how)
 846 {
 847         struct m_tag *t;
 848
 849         t = (struct m_tag *)mem;
 850         t->m_tag_cookie = MTAG_ABI_COMPAT;
 851         t->m_tag_id = PACKET_TAG_PF;
 852         t->m_tag_len = sizeof(struct pf_mtag);
 853         t->m_tag_free = pf_mtag_free;
 854
 855         return (0);
 856 }
 857
 858 static void
 859 pf_mtag_free(struct m_tag *t)
 860 {
 861
 862         uma_zfree(V_pf_mtag_z, t);
 863 }
 864
 865 struct pf_mtag *
 866 pf_get_mtag(struct mbuf *m)
 867 {
 868         struct m_tag *mtag;
 869
 870         if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
 871                 return ((struct pf_mtag *)(mtag + 1));
 872
 873         mtag = uma_zalloc(V_pf_mtag_z, M_NOWAIT);
 874         if (mtag == NULL)
 875                 return (NULL);
 876         bzero(mtag + 1, sizeof(struct pf_mtag));
 877         m_tag_prepend(m, mtag);
 878
 879         return ((struct pf_mtag *)(mtag + 1));
 880 }
 881
 882 static int
 883 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
 884     struct pf_state *s)
 885 {
 886         struct pf_keyhash       *khs, *khw, *kh;
 887         struct pf_state_key     *sk, *cur;
 888         struct pf_state         *si, *olds = NULL;
 889         int idx;
 890
 891         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
 892         KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
 893         KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
 894
 895         /*
 896          * We need to lock hash slots of both keys. To avoid deadlock
 897          * we always lock the slot with lower address first. Unlock order
 898          * isn't important.
 899          *
 900          * We also need to lock ID hash slot before dropping key
 901          * locks. On success we return with ID hash slot locked.
 902          */
 903
 904         if (skw == sks) {
 905                 khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
 906                 PF_HASHROW_LOCK(khs);
 907         } else {
 908                 khs = &V_pf_keyhash[pf_hashkey(sks)];
 909                 khw = &V_pf_keyhash[pf_hashkey(skw)];
 910                 if (khs == khw) {
 911                         PF_HASHROW_LOCK(khs);
 912                 } else if (khs < khw) {
 913                         PF_HASHROW_LOCK(khs);
 914                         PF_HASHROW_LOCK(khw);
 915                 } else {
 916                         PF_HASHROW_LOCK(khw);
 917                         PF_HASHROW_LOCK(khs);
 918                 }
 919         }
 920
 921 #define KEYS_UNLOCK()   do {                    \
 922         if (khs != khw) {                       \
 923                 PF_HASHROW_UNLOCK(khs);         \
 924                 PF_HASHROW_UNLOCK(khw);         \
 925         } else                                  \
 926                 PF_HASHROW_UNLOCK(khs);         \
 927 } while (0)
 928
 929         /*
 930          * First run: start with wire key.
 931          */
 932         sk = skw;
 933         kh = khw;
 934         idx = PF_SK_WIRE;
 935
 936 keyattach:
 937         LIST_FOREACH(cur, &kh->keys, entry)
 938                 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
 939                         break;
 940
 941         if (cur != NULL) {
 942                 /* Key exists. Check for same kif, if none, add to key. */
 943                 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
 944                         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
 945
 946                         PF_HASHROW_LOCK(ih);
 947                         if (si->kif == s->kif &&
 948                             si->direction == s->direction) {
 949                                 if (sk->proto == IPPROTO_TCP &&
 950                                     si->src.state >= TCPS_FIN_WAIT_2 &&
 951                                     si->dst.state >= TCPS_FIN_WAIT_2) {
 952                                         /*
 953                                          * New state matches an old >FIN_WAIT_2
 954                                          * state. We can't drop key hash locks,
 955                                          * thus we can't unlink it properly.
 956                                          *
 957                                          * As a workaround we drop it into
 958                                          * TCPS_CLOSED state, schedule purge
 959                                          * ASAP and push it into the very end
 960                                          * of the slot TAILQ, so that it won't
 961                                          * conflict with our new state.
 962                                          */
 963                                         si->src.state = si->dst.state =
 964                                             TCPS_CLOSED;
 965                                         si->timeout = PFTM_PURGE;
 966                                         olds = si;
 967                                 } else {
 968                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
 969                                                 printf("pf: %s key attach "
 970                                                     "failed on %s: ",
 971                                                     (idx == PF_SK_WIRE) ?
 972                                                     "wire" : "stack",
 973                                                     s->kif->pfik_name);
 974                                                 pf_print_state_parts(s,
 975                                                     (idx == PF_SK_WIRE) ?
 976                                                     sk : NULL,
 977                                                     (idx == PF_SK_STACK) ?
 978                                                     sk : NULL);
 979                                                 printf(", existing: ");
 980                                                 pf_print_state_parts(si,
 981                                                     (idx == PF_SK_WIRE) ?
 982                                                     sk : NULL,
 983                                                     (idx == PF_SK_STACK) ?
 984                                                     sk : NULL);
 985                                                 printf("\n");
 986                                         }
 987                                         PF_HASHROW_UNLOCK(ih);
 988                                         KEYS_UNLOCK();
 989                                         uma_zfree(V_pf_state_key_z, sk);
 990                                         if (idx == PF_SK_STACK)
 991                                                 pf_detach_state(s);
 992                                         return (EEXIST); /* collision! */
 993                                 }
 994                         }
 995                         PF_HASHROW_UNLOCK(ih);
 996                 }
 997                 uma_zfree(V_pf_state_key_z, sk);
 998                 s->key[idx] = cur;
 999         } else {
1000                 LIST_INSERT_HEAD(&kh->keys, sk, entry);
1001                 s->key[idx] = sk;
1002         }
1003
1004 stateattach:
1005         /* List is sorted, if-bound states before floating. */
1006         if (s->kif == V_pfi_all)
1007                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1008         else
1009                 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1010
1011         if (olds) {
1012                 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1013                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1014                     key_list[idx]);
1015                 olds = NULL;
1016         }
1017
1018         /*
1019          * Attach done. See how should we (or should not?)
1020          * attach a second key.
1021          */
1022         if (sks == skw) {
1023                 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1024                 idx = PF_SK_STACK;
1025                 sks = NULL;
1026                 goto stateattach;
1027         } else if (sks != NULL) {
1028                 /*
1029                  * Continue attaching with stack key.
1030                  */
1031                 sk = sks;
1032                 kh = khs;
1033                 idx = PF_SK_STACK;
1034                 sks = NULL;
1035                 goto keyattach;
1036         }
1037
1038         PF_STATE_LOCK(s);
1039         KEYS_UNLOCK();
1040
1041         KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1042             ("%s failure", __func__));
1043
1044         return (0);
1045 #undef  KEYS_UNLOCK
1046 }
1047
1048 static void
1049 pf_detach_state(struct pf_state *s)
1050 {
1051         struct pf_state_key *sks = s->key[PF_SK_STACK];
1052         struct pf_keyhash *kh;
1053
1054         if (sks != NULL) {
1055                 kh = &V_pf_keyhash[pf_hashkey(sks)];
1056                 PF_HASHROW_LOCK(kh);
1057                 if (s->key[PF_SK_STACK] != NULL)
1058                         pf_state_key_detach(s, PF_SK_STACK);
1059                 /*
1060                  * If both point to same key, then we are done.
1061                  */
1062                 if (sks == s->key[PF_SK_WIRE]) {
1063                         pf_state_key_detach(s, PF_SK_WIRE);
1064                         PF_HASHROW_UNLOCK(kh);
1065                         return;
1066                 }
1067                 PF_HASHROW_UNLOCK(kh);
1068         }
1069
1070         if (s->key[PF_SK_WIRE] != NULL) {
1071                 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1072                 PF_HASHROW_LOCK(kh);
1073                 if (s->key[PF_SK_WIRE] != NULL)
1074                         pf_state_key_detach(s, PF_SK_WIRE);
1075                 PF_HASHROW_UNLOCK(kh);
1076         }
1077 }
1078
1079 static void
1080 pf_state_key_detach(struct pf_state *s, int idx)
1081 {
1082         struct pf_state_key *sk = s->key[idx];
1083 #ifdef INVARIANTS
1084         struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1085
1086         PF_HASHROW_ASSERT(kh);
1087 #endif
1088         TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1089         s->key[idx] = NULL;
1090
1091         if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1092                 LIST_REMOVE(sk, entry);
1093                 uma_zfree(V_pf_state_key_z, sk);
1094         }
1095 }
1096
1097 static int
1098 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1099 {
1100         struct pf_state_key *sk = mem;
1101
1102         bzero(sk, sizeof(struct pf_state_key_cmp));
1103         TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1104         TAILQ_INIT(&sk->states[PF_SK_STACK]);
1105
1106         return (0);
1107 }
1108
1109 struct pf_state_key *
1110 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1111         struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1112 {
1113         struct pf_state_key *sk;
1114
1115         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1116         if (sk == NULL)
1117                 return (NULL);
1118
1119         PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1120         PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1121         sk->port[pd->sidx] = sport;
1122         sk->port[pd->didx] = dport;
1123         sk->proto = pd->proto;
1124         sk->af = pd->af;
1125
1126         return (sk);
1127 }
1128
1129 struct pf_state_key *
1130 pf_state_key_clone(struct pf_state_key *orig)
1131 {
1132         struct pf_state_key *sk;
1133
1134         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1135         if (sk == NULL)
1136                 return (NULL);
1137
1138         bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1139
1140         return (sk);
1141 }
1142
1143 int
1144 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1145     struct pf_state_key *sks, struct pf_state *s)
1146 {
1147         struct pf_idhash *ih;
1148         struct pf_state *cur;
1149         int error;
1150
1151         KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1152             ("%s: sks not pristine", __func__));
1153         KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1154             ("%s: skw not pristine", __func__));
1155         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1156
1157         s->kif = kif;
1158
1159         if (s->id == 0 && s->creatorid == 0) {
1160                 /* XXX: should be atomic, but probability of collision low */
1161                 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1162                         V_pf_stateid[curcpu] = 1;
1163                 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1164                 s->id = htobe64(s->id);
1165                 s->creatorid = V_pf_status.hostid;
1166         }
1167
1168         /* Returns with ID locked on success. */
1169         if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1170                 return (error);
1171
1172         ih = &V_pf_idhash[PF_IDHASH(s)];
1173         PF_HASHROW_ASSERT(ih);
1174         LIST_FOREACH(cur, &ih->states, entry)
1175                 if (cur->id == s->id && cur->creatorid == s->creatorid)
1176                         break;
1177
1178         if (cur != NULL) {
1179                 PF_HASHROW_UNLOCK(ih);
1180                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1181                         printf("pf: state ID collision: "
1182                             "id: %016llx creatorid: %08x\n",
1183                             (unsigned long long)be64toh(s->id),
1184                             ntohl(s->creatorid));
1185                 }
1186                 pf_detach_state(s);
1187                 return (EEXIST);
1188         }
1189         LIST_INSERT_HEAD(&ih->states, s, entry);
1190         /* One for keys, one for ID hash. */
1191         refcount_init(&s->refs, 2);
1192
1193         V_pf_status.fcounters[FCNT_STATE_INSERT]++;
1194         if (pfsync_insert_state_ptr != NULL)
1195                 pfsync_insert_state_ptr(s);
1196
1197         /* Returns locked. */
1198         return (0);
1199 }
1200
1201 /*
1202  * Find state by ID: returns with locked row on success.
1203  */
1204 struct pf_state *
1205 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1206 {
1207         struct pf_idhash *ih;
1208         struct pf_state *s;
1209
1210         V_pf_status.fcounters[FCNT_STATE_SEARCH]++;
1211
1212         ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))];
1213
1214         PF_HASHROW_LOCK(ih);
1215         LIST_FOREACH(s, &ih->states, entry)
1216                 if (s->id == id && s->creatorid == creatorid)
1217                         break;
1218
1219         if (s == NULL)
1220                 PF_HASHROW_UNLOCK(ih);
1221
1222         return (s);
1223 }
1224
1225 /*
1226  * Find state by key.
1227  * Returns with ID hash slot locked on success.
1228  */
1229 static struct pf_state *
1230 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1231 {
1232         struct pf_keyhash       *kh;
1233         struct pf_state_key     *sk;
1234         struct pf_state         *s;
1235         int idx;
1236
1237         V_pf_status.fcounters[FCNT_STATE_SEARCH]++;
1238
1239         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1240
1241         PF_HASHROW_LOCK(kh);
1242         LIST_FOREACH(sk, &kh->keys, entry)
1243                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1244                         break;
1245         if (sk == NULL) {
1246                 PF_HASHROW_UNLOCK(kh);
1247                 return (NULL);
1248         }
1249
1250         idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1251
1252         /* List is sorted, if-bound states before floating ones. */
1253         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1254                 if (s->kif == V_pfi_all || s->kif == kif) {
1255                         PF_STATE_LOCK(s);
1256                         PF_HASHROW_UNLOCK(kh);
1257                         if (s->timeout >= PFTM_MAX) {
1258                                 /*
1259                                  * State is either being processed by
1260                                  * pf_unlink_state() in an other thread, or
1261                                  * is scheduled for immediate expiry.
1262                                  */
1263                                 PF_STATE_UNLOCK(s);
1264                                 return (NULL);
1265                         }
1266                         return (s);
1267                 }
1268         PF_HASHROW_UNLOCK(kh);
1269
1270         return (NULL);
1271 }
1272
1273 struct pf_state *
1274 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1275 {
1276         struct pf_keyhash       *kh;
1277         struct pf_state_key     *sk;
1278         struct pf_state         *s, *ret = NULL;
1279         int                      idx, inout = 0;
1280
1281         V_pf_status.fcounters[FCNT_STATE_SEARCH]++;
1282
1283         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1284
1285         PF_HASHROW_LOCK(kh);
1286         LIST_FOREACH(sk, &kh->keys, entry)
1287                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1288                         break;
1289         if (sk == NULL) {
1290                 PF_HASHROW_UNLOCK(kh);
1291                 return (NULL);
1292         }
1293         switch (dir) {
1294         case PF_IN:
1295                 idx = PF_SK_WIRE;
1296                 break;
1297         case PF_OUT:
1298                 idx = PF_SK_STACK;
1299                 break;
1300         case PF_INOUT:
1301                 idx = PF_SK_WIRE;
1302                 inout = 1;
1303                 break;
1304         default:
1305                 panic("%s: dir %u", __func__, dir);
1306         }
1307 second_run:
1308         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1309                 if (more == NULL) {
1310                         PF_HASHROW_UNLOCK(kh);
1311                         return (s);
1312                 }
1313
1314                 if (ret)
1315                         (*more)++;
1316                 else
1317                         ret = s;
1318         }
1319         if (inout == 1) {
1320                 inout = 0;
1321                 idx = PF_SK_STACK;
1322                 goto second_run;
1323         }
1324         PF_HASHROW_UNLOCK(kh);
1325
1326         return (ret);
1327 }
1328
1329 /* END state table stuff */
1330
1331 static void
1332 pf_send(struct pf_send_entry *pfse)
1333 {
1334
1335         PF_SENDQ_LOCK();
1336         STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1337         PF_SENDQ_UNLOCK();
1338         swi_sched(V_pf_swi_cookie, 0);
1339 }
1340
1341 void
1342 pf_intr(void *v)
1343 {
1344         struct pf_send_head queue;
1345         struct pf_send_entry *pfse, *next;
1346
1347         CURVNET_SET((struct vnet *)v);
1348
1349         PF_SENDQ_LOCK();
1350         queue = V_pf_sendqueue;
1351         STAILQ_INIT(&V_pf_sendqueue);
1352         PF_SENDQ_UNLOCK();
1353
1354         STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1355                 switch (pfse->pfse_type) {
1356 #ifdef INET
1357                 case PFSE_IP:
1358                         ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1359                         break;
1360                 case PFSE_ICMP:
1361                         icmp_error(pfse->pfse_m, pfse->pfse_icmp_type,
1362                             pfse->pfse_icmp_code, 0, pfse->pfse_icmp_mtu);
1363                         break;
1364 #endif /* INET */
1365 #ifdef INET6
1366                 case PFSE_IP6:
1367                         ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1368                             NULL);
1369                         break;
1370                 case PFSE_ICMP6:
1371                         icmp6_error(pfse->pfse_m, pfse->pfse_icmp_type,
1372                             pfse->pfse_icmp_code, pfse->pfse_icmp_mtu);
1373                         break;
1374 #endif /* INET6 */
1375                 default:
1376                         panic("%s: unknown type", __func__);
1377                 }
1378                 free(pfse, M_PFTEMP);
1379         }
1380         CURVNET_RESTORE();
1381 }
1382
1383 void
1384 pf_purge_thread(void *v)
1385 {
1386         u_int idx = 0;
1387
1388         CURVNET_SET((struct vnet *)v);
1389
1390         for (;;) {
1391                 PF_RULES_RLOCK();
1392                 rw_sleep(pf_purge_thread, &pf_rules_lock, 0, "pftm", hz / 10);
1393
1394                 if (V_pf_end_threads) {
1395                         /*
1396                          * To cleanse up all kifs and rules we need
1397                          * two runs: first one clears reference flags,
1398                          * then pf_purge_expired_states() doesn't
1399                          * raise them, and then second run frees.
1400                          */
1401                         PF_RULES_RUNLOCK();
1402                         pf_purge_unlinked_rules();
1403                         pfi_kif_purge();
1404
1405                         /*
1406                          * Now purge everything.
1407                          */
1408                         pf_purge_expired_states(0, V_pf_hashmask);
1409                         pf_purge_expired_fragments();
1410                         pf_purge_expired_src_nodes();
1411
1412                         /*
1413                          * Now all kifs & rules should be unreferenced,
1414                          * thus should be successfully freed.
1415                          */
1416                         pf_purge_unlinked_rules();
1417                         pfi_kif_purge();
1418
1419                         /*
1420                          * Announce success and exit.
1421                          */
1422                         PF_RULES_RLOCK();
1423                         V_pf_end_threads++;
1424                         PF_RULES_RUNLOCK();
1425                         wakeup(pf_purge_thread);
1426                         kproc_exit(0);
1427                 }
1428                 PF_RULES_RUNLOCK();
1429
1430                 /* Process 1/interval fraction of the state table every run. */
1431                 idx = pf_purge_expired_states(idx, V_pf_hashmask /
1432                             (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1433
1434                 /* Purge other expired types every PFTM_INTERVAL seconds. */
1435                 if (idx == 0) {
1436                         /*
1437                          * Order is important:
1438                          * - states and src nodes reference rules
1439                          * - states and rules reference kifs
1440                          */
1441                         pf_purge_expired_fragments();
1442                         pf_purge_expired_src_nodes();
1443                         pf_purge_unlinked_rules();
1444                         pfi_kif_purge();
1445                 }
1446         }
1447         /* not reached */
1448         CURVNET_RESTORE();
1449 }
1450
1451 u_int32_t
1452 pf_state_expires(const struct pf_state *state)
1453 {
1454         u_int32_t       timeout;
1455         u_int32_t       start;
1456         u_int32_t       end;
1457         u_int32_t       states;
1458
1459         /* handle all PFTM_* > PFTM_MAX here */
1460         if (state->timeout == PFTM_PURGE)
1461                 return (time_uptime);
1462         KASSERT(state->timeout != PFTM_UNLINKED,
1463             ("pf_state_expires: timeout == PFTM_UNLINKED"));
1464         KASSERT((state->timeout < PFTM_MAX),
1465             ("pf_state_expires: timeout > PFTM_MAX"));
1466         timeout = state->rule.ptr->timeout[state->timeout];
1467         if (!timeout)
1468                 timeout = V_pf_default_rule.timeout[state->timeout];
1469         start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1470         if (start) {
1471                 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1472                 states = counter_u64_fetch(state->rule.ptr->states_cur);
1473         } else {
1474                 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1475                 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1476                 states = V_pf_status.states;
1477         }
1478         if (end && states > start && start < end) {
1479                 if (states < end)
1480                         return (state->expire + timeout * (end - states) /
1481                             (end - start));
1482                 else
1483                         return (time_uptime);
1484         }
1485         return (state->expire + timeout);
1486 }
1487
1488 void
1489 pf_purge_expired_src_nodes()
1490 {
1491         struct pf_src_node_list  freelist;
1492         struct pf_srchash       *sh;
1493         struct pf_src_node      *cur, *next;
1494         int i;
1495
1496         LIST_INIT(&freelist);
1497         for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
1498             PF_HASHROW_LOCK(sh);
1499             LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1500                 if (cur->states == 0 && cur->expire <= time_uptime) {
1501                         pf_unlink_src_node_locked(cur);
1502                         LIST_INSERT_HEAD(&freelist, cur, entry);
1503                 } else if (cur->rule.ptr != NULL)
1504                         cur->rule.ptr->rule_flag |= PFRULE_REFS;
1505             PF_HASHROW_UNLOCK(sh);
1506         }
1507
1508         pf_free_src_nodes(&freelist);
1509 }
1510
1511 static void
1512 pf_src_tree_remove_state(struct pf_state *s)
1513 {
1514         u_int32_t timeout;
1515
1516         if (s->src_node != NULL) {
1517                 if (s->src.tcp_est)
1518                         --s->src_node->conn;
1519                 if (--s->src_node->states == 0) {
1520                         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1521                         if (!timeout)
1522                                 timeout =
1523                                     V_pf_default_rule.timeout[PFTM_SRC_NODE];
1524                         s->src_node->expire = time_uptime + timeout;
1525                 }
1526         }
1527         if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1528                 if (--s->nat_src_node->states == 0) {
1529                         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1530                         if (!timeout)
1531                                 timeout =
1532                                     V_pf_default_rule.timeout[PFTM_SRC_NODE];
1533                         s->nat_src_node->expire = time_uptime + timeout;
1534                 }
1535         }
1536         s->src_node = s->nat_src_node = NULL;
1537 }
1538
1539 /*
1540  * Unlink and potentilly free a state. Function may be
1541  * called with ID hash row locked, but always returns
1542  * unlocked, since it needs to go through key hash locking.
1543  */
1544 int
1545 pf_unlink_state(struct pf_state *s, u_int flags)
1546 {
1547         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1548
1549         if ((flags & PF_ENTER_LOCKED) == 0)
1550                 PF_HASHROW_LOCK(ih);
1551         else
1552                 PF_HASHROW_ASSERT(ih);
1553
1554         if (s->timeout == PFTM_UNLINKED) {
1555                 /*
1556                  * State is being processed
1557                  * by pf_unlink_state() in
1558                  * an other thread.
1559                  */
1560                 PF_HASHROW_UNLOCK(ih);
1561                 return (0);     /* XXXGL: undefined actually */
1562         }
1563
1564         if (s->src.state == PF_TCPS_PROXY_DST) {
1565                 /* XXX wire key the right one? */
1566                 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1567                     &s->key[PF_SK_WIRE]->addr[1],
1568                     &s->key[PF_SK_WIRE]->addr[0],
1569                     s->key[PF_SK_WIRE]->port[1],
1570                     s->key[PF_SK_WIRE]->port[0],
1571                     s->src.seqhi, s->src.seqlo + 1,
1572                     TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1573         }
1574
1575         LIST_REMOVE(s, entry);
1576         pf_src_tree_remove_state(s);
1577
1578         if (pfsync_delete_state_ptr != NULL)
1579                 pfsync_delete_state_ptr(s);
1580
1581         STATE_DEC_COUNTERS(s);
1582
1583         s->timeout = PFTM_UNLINKED;
1584
1585         PF_HASHROW_UNLOCK(ih);
1586
1587         pf_detach_state(s);
1588         refcount_release(&s->refs);
1589
1590         return (pf_release_state(s));
1591 }
1592
1593 void
1594 pf_free_state(struct pf_state *cur)
1595 {
1596
1597         KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1598         KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1599             cur->timeout));
1600
1601         pf_normalize_tcp_cleanup(cur);
1602         uma_zfree(V_pf_state_z, cur);
1603         V_pf_status.fcounters[FCNT_STATE_REMOVALS]++;
1604 }
1605
1606 /*
1607  * Called only from pf_purge_thread(), thus serialized.
1608  */
1609 static u_int
1610 pf_purge_expired_states(u_int i, int maxcheck)
1611 {
1612         struct pf_idhash *ih;
1613         struct pf_state *s;
1614
1615         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1616
1617         /*
1618          * Go through hash and unlink states that expire now.
1619          */
1620         while (maxcheck > 0) {
1621
1622                 ih = &V_pf_idhash[i];
1623 relock:
1624                 PF_HASHROW_LOCK(ih);
1625                 LIST_FOREACH(s, &ih->states, entry) {
1626                         if (pf_state_expires(s) <= time_uptime) {
1627                                 V_pf_status.states -=
1628                                     pf_unlink_state(s, PF_ENTER_LOCKED);
1629                                 goto relock;
1630                         }
1631                         s->rule.ptr->rule_flag |= PFRULE_REFS;
1632                         if (s->nat_rule.ptr != NULL)
1633                                 s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1634                         if (s->anchor.ptr != NULL)
1635                                 s->anchor.ptr->rule_flag |= PFRULE_REFS;
1636                         s->kif->pfik_flags |= PFI_IFLAG_REFS;
1637                         if (s->rt_kif)
1638                                 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1639                 }
1640                 PF_HASHROW_UNLOCK(ih);
1641
1642                 /* Return when we hit end of hash. */
1643                 if (++i > V_pf_hashmask) {
1644                         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1645                         return (0);
1646                 }
1647
1648                 maxcheck--;
1649         }
1650
1651         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1652
1653         return (i);
1654 }
1655
1656 static void
1657 pf_purge_unlinked_rules()
1658 {
1659         struct pf_rulequeue tmpq;
1660         struct pf_rule *r, *r1;
1661
1662         /*
1663          * If we have overloading task pending, then we'd
1664          * better skip purging this time. There is a tiny
1665          * probability that overloading task references
1666          * an already unlinked rule.
1667          */
1668         PF_OVERLOADQ_LOCK();
1669         if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1670                 PF_OVERLOADQ_UNLOCK();
1671                 return;
1672         }
1673         PF_OVERLOADQ_UNLOCK();
1674
1675         /*
1676          * Do naive mark-and-sweep garbage collecting of old rules.
1677          * Reference flag is raised by pf_purge_expired_states()
1678          * and pf_purge_expired_src_nodes().
1679          *
1680          * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1681          * use a temporary queue.
1682          */
1683         TAILQ_INIT(&tmpq);
1684         PF_UNLNKDRULES_LOCK();
1685         TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1686                 if (!(r->rule_flag & PFRULE_REFS)) {
1687                         TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1688                         TAILQ_INSERT_TAIL(&tmpq, r, entries);
1689                 } else
1690                         r->rule_flag &= ~PFRULE_REFS;
1691         }
1692         PF_UNLNKDRULES_UNLOCK();
1693
1694         if (!TAILQ_EMPTY(&tmpq)) {
1695                 PF_RULES_WLOCK();
1696                 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1697                         TAILQ_REMOVE(&tmpq, r, entries);
1698                         pf_free_rule(r);
1699                 }
1700                 PF_RULES_WUNLOCK();
1701         }
1702 }
1703
1704 void
1705 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1706 {
1707         switch (af) {
1708 #ifdef INET
1709         case AF_INET: {
1710                 u_int32_t a = ntohl(addr->addr32[0]);
1711                 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1712                     (a>>8)&255, a&255);
1713                 if (p) {
1714                         p = ntohs(p);
1715                         printf(":%u", p);
1716                 }
1717                 break;
1718         }
1719 #endif /* INET */
1720 #ifdef INET6
1721         case AF_INET6: {
1722                 u_int16_t b;
1723                 u_int8_t i, curstart, curend, maxstart, maxend;
1724                 curstart = curend = maxstart = maxend = 255;
1725                 for (i = 0; i < 8; i++) {
1726                         if (!addr->addr16[i]) {
1727                                 if (curstart == 255)
1728                                         curstart = i;
1729                                 curend = i;
1730                         } else {
1731                                 if ((curend - curstart) >
1732                                     (maxend - maxstart)) {
1733                                         maxstart = curstart;
1734                                         maxend = curend;
1735                                 }
1736                                 curstart = curend = 255;
1737                         }
1738                 }
1739                 if ((curend - curstart) >
1740                     (maxend - maxstart)) {
1741                         maxstart = curstart;
1742                         maxend = curend;
1743                 }
1744                 for (i = 0; i < 8; i++) {
1745                         if (i >= maxstart && i <= maxend) {
1746                                 if (i == 0)
1747                                         printf(":");
1748                                 if (i == maxend)
1749                                         printf(":");
1750                         } else {
1751                                 b = ntohs(addr->addr16[i]);
1752                                 printf("%x", b);
1753                                 if (i < 7)
1754                                         printf(":");
1755                         }
1756                 }
1757                 if (p) {
1758                         p = ntohs(p);
1759                         printf("[%u]", p);
1760                 }
1761                 break;
1762         }
1763 #endif /* INET6 */
1764         }
1765 }
1766
1767 void
1768 pf_print_state(struct pf_state *s)
1769 {
1770         pf_print_state_parts(s, NULL, NULL);
1771 }
1772
1773 static void
1774 pf_print_state_parts(struct pf_state *s,
1775     struct pf_state_key *skwp, struct pf_state_key *sksp)
1776 {
1777         struct pf_state_key *skw, *sks;
1778         u_int8_t proto, dir;
1779
1780         /* Do our best to fill these, but they're skipped if NULL */
1781         skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1782         sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1783         proto = skw ? skw->proto : (sks ? sks->proto : 0);
1784         dir = s ? s->direction : 0;
1785
1786         switch (proto) {
1787         case IPPROTO_IPV4:
1788                 printf("IPv4");
1789                 break;
1790         case IPPROTO_IPV6:
1791                 printf("IPv6");
1792                 break;
1793         case IPPROTO_TCP:
1794                 printf("TCP");
1795                 break;
1796         case IPPROTO_UDP:
1797                 printf("UDP");
1798                 break;
1799         case IPPROTO_ICMP:
1800                 printf("ICMP");
1801                 break;
1802         case IPPROTO_ICMPV6:
1803                 printf("ICMPv6");
1804                 break;
1805         default:
1806                 printf("%u", skw->proto);
1807                 break;
1808         }
1809         switch (dir) {
1810         case PF_IN:
1811                 printf(" in");
1812                 break;
1813         case PF_OUT:
1814                 printf(" out");
1815                 break;
1816         }
1817         if (skw) {
1818                 printf(" wire: ");
1819                 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1820                 printf(" ");
1821                 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1822         }
1823         if (sks) {
1824                 printf(" stack: ");
1825                 if (sks != skw) {
1826                         pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1827                         printf(" ");
1828                         pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1829                 } else
1830                         printf("-");
1831         }
1832         if (s) {
1833                 if (proto == IPPROTO_TCP) {
1834                         printf(" [lo=%u high=%u win=%u modulator=%u",
1835                             s->src.seqlo, s->src.seqhi,
1836                             s->src.max_win, s->src.seqdiff);
1837                         if (s->src.wscale && s->dst.wscale)
1838                                 printf(" wscale=%u",
1839                                     s->src.wscale & PF_WSCALE_MASK);
1840                         printf("]");
1841                         printf(" [lo=%u high=%u win=%u modulator=%u",
1842                             s->dst.seqlo, s->dst.seqhi,
1843                             s->dst.max_win, s->dst.seqdiff);
1844                         if (s->src.wscale && s->dst.wscale)
1845                                 printf(" wscale=%u",
1846                                 s->dst.wscale & PF_WSCALE_MASK);
1847                         printf("]");
1848                 }
1849                 printf(" %u:%u", s->src.state, s->dst.state);
1850         }
1851 }
1852
1853 void
1854 pf_print_flags(u_int8_t f)
1855 {
1856         if (f)
1857                 printf(" ");
1858         if (f & TH_FIN)
1859                 printf("F");
1860         if (f & TH_SYN)
1861                 printf("S");
1862         if (f & TH_RST)
1863                 printf("R");
1864         if (f & TH_PUSH)
1865                 printf("P");
1866         if (f & TH_ACK)
1867                 printf("A");
1868         if (f & TH_URG)
1869                 printf("U");
1870         if (f & TH_ECE)
1871                 printf("E");
1872         if (f & TH_CWR)
1873                 printf("W");
1874 }
1875
1876 #define PF_SET_SKIP_STEPS(i)                                    \
1877         do {                                                    \
1878                 while (head[i] != cur) {                        \
1879                         head[i]->skip[i].ptr = cur;             \
1880                         head[i] = TAILQ_NEXT(head[i], entries); \
1881                 }                                               \
1882         } while (0)
1883
1884 void
1885 pf_calc_skip_steps(struct pf_rulequeue *rules)
1886 {
1887         struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1888         int i;
1889
1890         cur = TAILQ_FIRST(rules);
1891         prev = cur;
1892         for (i = 0; i < PF_SKIP_COUNT; ++i)
1893                 head[i] = cur;
1894         while (cur != NULL) {
1895
1896                 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1897                         PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1898                 if (cur->direction != prev->direction)
1899                         PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1900                 if (cur->af != prev->af)
1901                         PF_SET_SKIP_STEPS(PF_SKIP_AF);
1902                 if (cur->proto != prev->proto)
1903                         PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
1904                 if (cur->src.neg != prev->src.neg ||
1905                     pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
1906                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
1907                 if (cur->src.port[0] != prev->src.port[0] ||
1908                     cur->src.port[1] != prev->src.port[1] ||
1909                     cur->src.port_op != prev->src.port_op)
1910                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
1911                 if (cur->dst.neg != prev->dst.neg ||
1912                     pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
1913                         PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
1914                 if (cur->dst.port[0] != prev->dst.port[0] ||
1915                     cur->dst.port[1] != prev->dst.port[1] ||
1916                     cur->dst.port_op != prev->dst.port_op)
1917                         PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
1918
1919                 prev = cur;
1920                 cur = TAILQ_NEXT(cur, entries);
1921         }
1922         for (i = 0; i < PF_SKIP_COUNT; ++i)
1923                 PF_SET_SKIP_STEPS(i);
1924 }
1925
1926 static int
1927 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
1928 {
1929         if (aw1->type != aw2->type)
1930                 return (1);
1931         switch (aw1->type) {
1932         case PF_ADDR_ADDRMASK:
1933         case PF_ADDR_RANGE:
1934                 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0))
1935                         return (1);
1936                 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0))
1937                         return (1);
1938                 return (0);
1939         case PF_ADDR_DYNIFTL:
1940                 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
1941         case PF_ADDR_NOROUTE:
1942         case PF_ADDR_URPFFAILED:
1943                 return (0);
1944         case PF_ADDR_TABLE:
1945                 return (aw1->p.tbl != aw2->p.tbl);
1946         default:
1947                 printf("invalid address type: %d\n", aw1->type);
1948                 return (1);
1949         }
1950 }
1951
1952 u_int16_t
1953 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
1954 {
1955         u_int32_t       l;
1956
1957         if (udp && !cksum)
1958                 return (0x0000);
1959         l = cksum + old - new;
1960         l = (l >> 16) + (l & 65535);
1961         l = l & 65535;
1962         if (udp && !l)
1963                 return (0xFFFF);
1964         return (l);
1965 }
1966
1967 static void
1968 pf_change_ap(struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc,
1969     struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af)
1970 {
1971         struct pf_addr  ao;
1972         u_int16_t       po = *p;
1973
1974         PF_ACPY(&ao, a, af);
1975         PF_ACPY(a, an, af);
1976
1977         *p = pn;
1978
1979         switch (af) {
1980 #ifdef INET
1981         case AF_INET:
1982                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
1983                     ao.addr16[0], an->addr16[0], 0),
1984                     ao.addr16[1], an->addr16[1], 0);
1985                 *p = pn;
1986                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
1987                     ao.addr16[0], an->addr16[0], u),
1988                     ao.addr16[1], an->addr16[1], u),
1989                     po, pn, u);
1990                 break;
1991 #endif /* INET */
1992 #ifdef INET6
1993         case AF_INET6:
1994                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1995                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
1996                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
1997                     ao.addr16[0], an->addr16[0], u),
1998                     ao.addr16[1], an->addr16[1], u),
1999                     ao.addr16[2], an->addr16[2], u),
2000                     ao.addr16[3], an->addr16[3], u),
2001                     ao.addr16[4], an->addr16[4], u),
2002                     ao.addr16[5], an->addr16[5], u),
2003                     ao.addr16[6], an->addr16[6], u),
2004                     ao.addr16[7], an->addr16[7], u),
2005                     po, pn, u);
2006                 break;
2007 #endif /* INET6 */
2008         }
2009 }
2010
2011
2012 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2013 void
2014 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2015 {
2016         u_int32_t       ao;
2017
2018         memcpy(&ao, a, sizeof(ao));
2019         memcpy(a, &an, sizeof(u_int32_t));
2020         *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2021             ao % 65536, an % 65536, u);
2022 }
2023
2024 #ifdef INET6
2025 static void
2026 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2027 {
2028         struct pf_addr  ao;
2029
2030         PF_ACPY(&ao, a, AF_INET6);
2031         PF_ACPY(a, an, AF_INET6);
2032
2033         *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2034             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2035             pf_cksum_fixup(pf_cksum_fixup(*c,
2036             ao.addr16[0], an->addr16[0], u),
2037             ao.addr16[1], an->addr16[1], u),
2038             ao.addr16[2], an->addr16[2], u),
2039             ao.addr16[3], an->addr16[3], u),
2040             ao.addr16[4], an->addr16[4], u),
2041             ao.addr16[5], an->addr16[5], u),
2042             ao.addr16[6], an->addr16[6], u),
2043             ao.addr16[7], an->addr16[7], u);
2044 }
2045 #endif /* INET6 */
2046
2047 static void
2048 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2049     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2050     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2051 {
2052         struct pf_addr  oia, ooa;
2053
2054         PF_ACPY(&oia, ia, af);
2055         if (oa)
2056                 PF_ACPY(&ooa, oa, af);
2057
2058         /* Change inner protocol port, fix inner protocol checksum. */
2059         if (ip != NULL) {
2060                 u_int16_t       oip = *ip;
2061                 u_int32_t       opc;
2062
2063                 if (pc != NULL)
2064                         opc = *pc;
2065                 *ip = np;
2066                 if (pc != NULL)
2067                         *pc = pf_cksum_fixup(*pc, oip, *ip, u);
2068                 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2069                 if (pc != NULL)
2070                         *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2071         }
2072         /* Change inner ip address, fix inner ip and icmp checksums. */
2073         PF_ACPY(ia, na, af);
2074         switch (af) {
2075 #ifdef INET
2076         case AF_INET: {
2077                 u_int32_t        oh2c = *h2c;
2078
2079                 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2080                     oia.addr16[0], ia->addr16[0], 0),
2081                     oia.addr16[1], ia->addr16[1], 0);
2082                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2083                     oia.addr16[0], ia->addr16[0], 0),
2084                     oia.addr16[1], ia->addr16[1], 0);
2085                 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2086                 break;
2087         }
2088 #endif /* INET */
2089 #ifdef INET6
2090         case AF_INET6:
2091                 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2092                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2093                     pf_cksum_fixup(pf_cksum_fixup(*ic,
2094                     oia.addr16[0], ia->addr16[0], u),
2095                     oia.addr16[1], ia->addr16[1], u),
2096                     oia.addr16[2], ia->addr16[2], u),
2097                     oia.addr16[3], ia->addr16[3], u),
2098                     oia.addr16[4], ia->addr16[4], u),
2099                     oia.addr16[5], ia->addr16[5], u),
2100                     oia.addr16[6], ia->addr16[6], u),
2101                     oia.addr16[7], ia->addr16[7], u);
2102                 break;
2103 #endif /* INET6 */
2104         }
2105         /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2106         if (oa) {
2107                 PF_ACPY(oa, na, af);
2108                 switch (af) {
2109 #ifdef INET
2110                 case AF_INET:
2111                         *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2112                             ooa.addr16[0], oa->addr16[0], 0),
2113                             ooa.addr16[1], oa->addr16[1], 0);
2114                         break;
2115 #endif /* INET */
2116 #ifdef INET6
2117                 case AF_INET6:
2118                         *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2119                             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2120                             pf_cksum_fixup(pf_cksum_fixup(*ic,
2121                             ooa.addr16[0], oa->addr16[0], u),
2122                             ooa.addr16[1], oa->addr16[1], u),
2123                             ooa.addr16[2], oa->addr16[2], u),
2124                             ooa.addr16[3], oa->addr16[3], u),
2125                             ooa.addr16[4], oa->addr16[4], u),
2126                             ooa.addr16[5], oa->addr16[5], u),
2127                             ooa.addr16[6], oa->addr16[6], u),
2128                             ooa.addr16[7], oa->addr16[7], u);
2129                         break;
2130 #endif /* INET6 */
2131                 }
2132         }
2133 }
2134
2135
2136 /*
2137  * Need to modulate the sequence numbers in the TCP SACK option
2138  * (credits to Krzysztof Pfaff for report and patch)
2139  */
2140 static int
2141 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2142     struct tcphdr *th, struct pf_state_peer *dst)
2143 {
2144         int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2145         u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2146         int copyback = 0, i, olen;
2147         struct sackblk sack;
2148
2149 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2150         if (hlen < TCPOLEN_SACKLEN ||
2151             !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2152                 return 0;
2153
2154         while (hlen >= TCPOLEN_SACKLEN) {
2155                 olen = opt[1];
2156                 switch (*opt) {
2157                 case TCPOPT_EOL:        /* FALLTHROUGH */
2158                 case TCPOPT_NOP:
2159                         opt++;
2160                         hlen--;
2161                         break;
2162                 case TCPOPT_SACK:
2163                         if (olen > hlen)
2164                                 olen = hlen;
2165                         if (olen >= TCPOLEN_SACKLEN) {
2166                                 for (i = 2; i + TCPOLEN_SACK <= olen;
2167                                     i += TCPOLEN_SACK) {
2168                                         memcpy(&sack, &opt[i], sizeof(sack));
2169                                         pf_change_a(&sack.start, &th->th_sum,
2170                                             htonl(ntohl(sack.start) -
2171                                             dst->seqdiff), 0);
2172                                         pf_change_a(&sack.end, &th->th_sum,
2173                                             htonl(ntohl(sack.end) -
2174                                             dst->seqdiff), 0);
2175                                         memcpy(&opt[i], &sack, sizeof(sack));
2176                                 }
2177                                 copyback = 1;
2178                         }
2179                         /* FALLTHROUGH */
2180                 default:
2181                         if (olen < 2)
2182                                 olen = 2;
2183                         hlen -= olen;
2184                         opt += olen;
2185                 }
2186         }
2187
2188         if (copyback)
2189                 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2190         return (copyback);
2191 }
2192
2193 static void
2194 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2195     const struct pf_addr *saddr, const struct pf_addr *daddr,
2196     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2197     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2198     u_int16_t rtag, struct ifnet *ifp)
2199 {
2200         struct pf_send_entry *pfse;
2201         struct mbuf     *m;
2202         int              len, tlen;
2203 #ifdef INET
2204         struct ip       *h = NULL;
2205 #endif /* INET */
2206 #ifdef INET6
2207         struct ip6_hdr  *h6 = NULL;
2208 #endif /* INET6 */
2209         struct tcphdr   *th;
2210         char            *opt;
2211         struct pf_mtag  *pf_mtag;
2212
2213         len = 0;
2214         th = NULL;
2215
2216         /* maximum segment size tcp option */
2217         tlen = sizeof(struct tcphdr);
2218         if (mss)
2219                 tlen += 4;
2220
2221         switch (af) {
2222 #ifdef INET
2223         case AF_INET:
2224                 len = sizeof(struct ip) + tlen;
2225                 break;
2226 #endif /* INET */
2227 #ifdef INET6
2228         case AF_INET6:
2229                 len = sizeof(struct ip6_hdr) + tlen;
2230                 break;
2231 #endif /* INET6 */
2232         default:
2233                 panic("%s: unsupported af %d", __func__, af);
2234         }
2235
2236         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2237         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2238         if (pfse == NULL)
2239                 return;
2240         m = m_gethdr(M_NOWAIT, MT_DATA);
2241         if (m == NULL) {
2242                 free(pfse, M_PFTEMP);
2243                 return;
2244         }
2245 #ifdef MAC
2246         mac_netinet_firewall_send(m);
2247 #endif
2248         if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2249                 free(pfse, M_PFTEMP);
2250                 m_freem(m);
2251                 return;
2252         }
2253         if (tag)
2254                 m->m_flags |= M_SKIP_FIREWALL;
2255         pf_mtag->tag = rtag;
2256
2257         if (r != NULL && r->rtableid >= 0)
2258                 M_SETFIB(m, r->rtableid);
2259
2260 #ifdef ALTQ
2261         if (r != NULL && r->qid) {
2262                 pf_mtag->qid = r->qid;
2263
2264                 /* add hints for ecn */
2265                 pf_mtag->hdr = mtod(m, struct ip *);
2266         }
2267 #endif /* ALTQ */
2268         m->m_data += max_linkhdr;
2269         m->m_pkthdr.len = m->m_len = len;
2270         m->m_pkthdr.rcvif = NULL;
2271         bzero(m->m_data, len);
2272         switch (af) {
2273 #ifdef INET
2274         case AF_INET:
2275                 h = mtod(m, struct ip *);
2276
2277                 /* IP header fields included in the TCP checksum */
2278                 h->ip_p = IPPROTO_TCP;
2279                 h->ip_len = htons(tlen);
2280                 h->ip_src.s_addr = saddr->v4.s_addr;
2281                 h->ip_dst.s_addr = daddr->v4.s_addr;
2282
2283                 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2284                 break;
2285 #endif /* INET */
2286 #ifdef INET6
2287         case AF_INET6:
2288                 h6 = mtod(m, struct ip6_hdr *);
2289
2290                 /* IP header fields included in the TCP checksum */
2291                 h6->ip6_nxt = IPPROTO_TCP;
2292                 h6->ip6_plen = htons(tlen);
2293                 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2294                 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2295
2296                 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2297                 break;
2298 #endif /* INET6 */
2299         }
2300
2301         /* TCP header */
2302         th->th_sport = sport;
2303         th->th_dport = dport;
2304         th->th_seq = htonl(seq);
2305         th->th_ack = htonl(ack);
2306         th->th_off = tlen >> 2;
2307         th->th_flags = flags;
2308         th->th_win = htons(win);
2309
2310         if (mss) {
2311                 opt = (char *)(th + 1);
2312                 opt[0] = TCPOPT_MAXSEG;
2313                 opt[1] = 4;
2314                 HTONS(mss);
2315                 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2316         }
2317
2318         switch (af) {
2319 #ifdef INET
2320         case AF_INET:
2321                 /* TCP checksum */
2322                 th->th_sum = in_cksum(m, len);
2323
2324                 /* Finish the IP header */
2325                 h->ip_v = 4;
2326                 h->ip_hl = sizeof(*h) >> 2;
2327                 h->ip_tos = IPTOS_LOWDELAY;
2328                 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2329                 h->ip_len = htons(len);
2330                 h->ip_ttl = ttl ? ttl : V_ip_defttl;
2331                 h->ip_sum = 0;
2332
2333                 pfse->pfse_type = PFSE_IP;
2334                 break;
2335 #endif /* INET */
2336 #ifdef INET6
2337         case AF_INET6:
2338                 /* TCP checksum */
2339                 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2340                     sizeof(struct ip6_hdr), tlen);
2341
2342                 h6->ip6_vfc |= IPV6_VERSION;
2343                 h6->ip6_hlim = IPV6_DEFHLIM;
2344
2345                 pfse->pfse_type = PFSE_IP6;
2346                 break;
2347 #endif /* INET6 */
2348         }
2349         pfse->pfse_m = m;
2350         pf_send(pfse);
2351 }
2352
2353 static void
2354 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2355     struct pf_rule *r)
2356 {
2357         struct pf_send_entry *pfse;
2358         struct mbuf *m0;
2359         struct pf_mtag *pf_mtag;
2360
2361         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2362         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2363         if (pfse == NULL)
2364                 return;
2365
2366         if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2367                 free(pfse, M_PFTEMP);
2368                 return;
2369         }
2370
2371         if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2372                 free(pfse, M_PFTEMP);
2373                 return;
2374         }
2375         /* XXX: revisit */
2376         m0->m_flags |= M_SKIP_FIREWALL;
2377
2378         if (r->rtableid >= 0)
2379                 M_SETFIB(m0, r->rtableid);
2380
2381 #ifdef ALTQ
2382         if (r->qid) {
2383                 pf_mtag->qid = r->qid;
2384                 /* add hints for ecn */
2385                 pf_mtag->hdr = mtod(m0, struct ip *);
2386         }
2387 #endif /* ALTQ */
2388
2389         switch (af) {
2390 #ifdef INET
2391         case AF_INET:
2392                 pfse->pfse_type = PFSE_ICMP;
2393                 break;
2394 #endif /* INET */
2395 #ifdef INET6
2396         case AF_INET6:
2397                 pfse->pfse_type = PFSE_ICMP6;
2398                 break;
2399 #endif /* INET6 */
2400         }
2401         pfse->pfse_m = m0;
2402         pfse->pfse_icmp_type = type;
2403         pfse->pfse_icmp_code = code;
2404         pf_send(pfse);
2405 }
2406
2407 /*
2408  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2409  * If n is 0, they match if they are equal. If n is != 0, they match if they
2410  * are different.
2411  */
2412 int
2413 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2414     struct pf_addr *b, sa_family_t af)
2415 {
2416         int     match = 0;
2417
2418         switch (af) {
2419 #ifdef INET
2420         case AF_INET:
2421                 if ((a->addr32[0] & m->addr32[0]) ==
2422                     (b->addr32[0] & m->addr32[0]))
2423                         match++;
2424                 break;
2425 #endif /* INET */
2426 #ifdef INET6
2427         case AF_INET6:
2428                 if (((a->addr32[0] & m->addr32[0]) ==
2429                      (b->addr32[0] & m->addr32[0])) &&
2430                     ((a->addr32[1] & m->addr32[1]) ==
2431                      (b->addr32[1] & m->addr32[1])) &&
2432                     ((a->addr32[2] & m->addr32[2]) ==
2433                      (b->addr32[2] & m->addr32[2])) &&
2434                     ((a->addr32[3] & m->addr32[3]) ==
2435                      (b->addr32[3] & m->addr32[3])))
2436                         match++;
2437                 break;
2438 #endif /* INET6 */
2439         }
2440         if (match) {
2441                 if (n)
2442                         return (0);
2443                 else
2444                         return (1);
2445         } else {
2446                 if (n)
2447                         return (1);
2448                 else
2449                         return (0);
2450         }
2451 }
2452
2453 /*
2454  * Return 1 if b <= a <= e, otherwise return 0.
2455  */
2456 int
2457 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2458     struct pf_addr *a, sa_family_t af)
2459 {
2460         switch (af) {
2461 #ifdef INET
2462         case AF_INET:
2463                 if ((a->addr32[0] < b->addr32[0]) ||
2464                     (a->addr32[0] > e->addr32[0]))
2465                         return (0);
2466                 break;
2467 #endif /* INET */
2468 #ifdef INET6
2469         case AF_INET6: {
2470                 int     i;
2471
2472                 /* check a >= b */
2473                 for (i = 0; i < 4; ++i)
2474                         if (a->addr32[i] > b->addr32[i])
2475                                 break;
2476                         else if (a->addr32[i] < b->addr32[i])
2477                                 return (0);
2478                 /* check a <= e */
2479                 for (i = 0; i < 4; ++i)
2480                         if (a->addr32[i] < e->addr32[i])
2481                                 break;
2482                         else if (a->addr32[i] > e->addr32[i])
2483                                 return (0);
2484                 break;
2485         }
2486 #endif /* INET6 */
2487         }
2488         return (1);
2489 }
2490
2491 static int
2492 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2493 {
2494         switch (op) {
2495         case PF_OP_IRG:
2496                 return ((p > a1) && (p < a2));
2497         case PF_OP_XRG:
2498                 return ((p < a1) || (p > a2));
2499         case PF_OP_RRG:
2500                 return ((p >= a1) && (p <= a2));
2501         case PF_OP_EQ:
2502                 return (p == a1);
2503         case PF_OP_NE:
2504                 return (p != a1);
2505         case PF_OP_LT:
2506                 return (p < a1);
2507         case PF_OP_LE:
2508                 return (p <= a1);
2509         case PF_OP_GT:
2510                 return (p > a1);
2511         case PF_OP_GE:
2512                 return (p >= a1);
2513         }
2514         return (0); /* never reached */
2515 }
2516
2517 int
2518 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2519 {
2520         NTOHS(a1);
2521         NTOHS(a2);
2522         NTOHS(p);
2523         return (pf_match(op, a1, a2, p));
2524 }
2525
2526 static int
2527 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2528 {
2529         if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2530                 return (0);
2531         return (pf_match(op, a1, a2, u));
2532 }
2533
2534 static int
2535 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2536 {
2537         if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2538                 return (0);
2539         return (pf_match(op, a1, a2, g));
2540 }
2541
2542 int
2543 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2544 {
2545         if (*tag == -1)
2546                 *tag = mtag;
2547
2548         return ((!r->match_tag_not && r->match_tag == *tag) ||
2549             (r->match_tag_not && r->match_tag != *tag));
2550 }
2551
2552 int
2553 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2554 {
2555
2556         KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2557
2558         if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2559                 return (ENOMEM);
2560
2561         pd->pf_mtag->tag = tag;
2562
2563         return (0);
2564 }
2565
2566 #define PF_ANCHOR_STACKSIZE     32
2567 struct pf_anchor_stackframe {
2568         struct pf_ruleset       *rs;
2569         struct pf_rule          *r;     /* XXX: + match bit */
2570         struct pf_anchor        *child;
2571 };
2572
2573 /*
2574  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2575  */
2576 #define PF_ANCHORSTACK_MATCH    0x00000001
2577 #define PF_ANCHORSTACK_MASK     (PF_ANCHORSTACK_MATCH)
2578
2579 #define PF_ANCHOR_MATCH(f)      ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2580 #define PF_ANCHOR_RULE(f)       (struct pf_rule *)                      \
2581                                 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2582 #define PF_ANCHOR_SET_MATCH(f)  do { (f)->r = (void *)                  \
2583                                 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2584 } while (0)
2585
2586 void
2587 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2588     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2589     int *match)
2590 {
2591         struct pf_anchor_stackframe     *f;
2592
2593         PF_RULES_RASSERT();
2594
2595         if (match)
2596                 *match = 0;
2597         if (*depth >= PF_ANCHOR_STACKSIZE) {
2598                 printf("%s: anchor stack overflow on %s\n",
2599                     __func__, (*r)->anchor->name);
2600                 *r = TAILQ_NEXT(*r, entries);
2601                 return;
2602         } else if (*depth == 0 && a != NULL)
2603                 *a = *r;
2604         f = stack + (*depth)++;
2605         f->rs = *rs;
2606         f->r = *r;
2607         if ((*r)->anchor_wildcard) {
2608                 struct pf_anchor_node *parent = &(*r)->anchor->children;
2609
2610                 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2611                         *r = NULL;
2612                         return;
2613                 }
2614                 *rs = &f->child->ruleset;
2615         } else {
2616                 f->child = NULL;
2617                 *rs = &(*r)->anchor->ruleset;
2618         }
2619         *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2620 }
2621
2622 int
2623 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2624     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2625     int *match)
2626 {
2627         struct pf_anchor_stackframe     *f;
2628         struct pf_rule *fr;
2629         int quick = 0;
2630
2631         PF_RULES_RASSERT();
2632
2633         do {
2634                 if (*depth <= 0)
2635                         break;
2636                 f = stack + *depth - 1;
2637                 fr = PF_ANCHOR_RULE(f);
2638                 if (f->child != NULL) {
2639                         struct pf_anchor_node *parent;
2640
2641                         /*
2642                          * This block traverses through
2643                          * a wildcard anchor.
2644                          */
2645                         parent = &fr->anchor->children;
2646                         if (match != NULL && *match) {
2647                                 /*
2648                                  * If any of "*" matched, then
2649                                  * "foo/ *" matched, mark frame
2650                                  * appropriately.
2651                                  */
2652                                 PF_ANCHOR_SET_MATCH(f);
2653                                 *match = 0;
2654                         }
2655                         f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2656                         if (f->child != NULL) {
2657                                 *rs = &f->child->ruleset;
2658                                 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2659                                 if (*r == NULL)
2660                                         continue;
2661                                 else
2662                                         break;
2663                         }
2664                 }
2665                 (*depth)--;
2666                 if (*depth == 0 && a != NULL)
2667                         *a = NULL;
2668                 *rs = f->rs;
2669                 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2670                         quick = fr->quick;
2671                 *r = TAILQ_NEXT(fr, entries);
2672         } while (*r == NULL);
2673
2674         return (quick);
2675 }
2676
2677 #ifdef INET6
2678 void
2679 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2680     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2681 {
2682         switch (af) {
2683 #ifdef INET
2684         case AF_INET:
2685                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2686                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2687                 break;
2688 #endif /* INET */
2689         case AF_INET6:
2690                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2691                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2692                 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2693                 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2694                 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2695                 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2696                 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2697                 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2698                 break;
2699         }
2700 }
2701
2702 void
2703 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2704 {
2705         switch (af) {
2706 #ifdef INET
2707         case AF_INET:
2708                 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2709                 break;
2710 #endif /* INET */
2711         case AF_INET6:
2712                 if (addr->addr32[3] == 0xffffffff) {
2713                         addr->addr32[3] = 0;
2714                         if (addr->addr32[2] == 0xffffffff) {
2715                                 addr->addr32[2] = 0;
2716                                 if (addr->addr32[1] == 0xffffffff) {
2717                                         addr->addr32[1] = 0;
2718                                         addr->addr32[0] =
2719                                             htonl(ntohl(addr->addr32[0]) + 1);
2720                                 } else
2721                                         addr->addr32[1] =
2722                                             htonl(ntohl(addr->addr32[1]) + 1);
2723                         } else
2724                                 addr->addr32[2] =
2725                                     htonl(ntohl(addr->addr32[2]) + 1);
2726                 } else
2727                         addr->addr32[3] =
2728                             htonl(ntohl(addr->addr32[3]) + 1);
2729                 break;
2730         }
2731 }
2732 #endif /* INET6 */
2733
2734 int
2735 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
2736 {
2737         struct pf_addr          *saddr, *daddr;
2738         u_int16_t                sport, dport;
2739         struct inpcbinfo        *pi;
2740         struct inpcb            *inp;
2741
2742         pd->lookup.uid = UID_MAX;
2743         pd->lookup.gid = GID_MAX;
2744
2745         switch (pd->proto) {
2746         case IPPROTO_TCP:
2747                 if (pd->hdr.tcp == NULL)
2748                         return (-1);
2749                 sport = pd->hdr.tcp->th_sport;
2750                 dport = pd->hdr.tcp->th_dport;
2751                 pi = &V_tcbinfo;
2752                 break;
2753         case IPPROTO_UDP:
2754                 if (pd->hdr.udp == NULL)
2755                         return (-1);
2756                 sport = pd->hdr.udp->uh_sport;
2757                 dport = pd->hdr.udp->uh_dport;
2758                 pi = &V_udbinfo;
2759                 break;
2760         default:
2761                 return (-1);
2762         }
2763         if (direction == PF_IN) {
2764                 saddr = pd->src;
2765                 daddr = pd->dst;
2766         } else {
2767                 u_int16_t       p;
2768
2769                 p = sport;
2770                 sport = dport;
2771                 dport = p;
2772                 saddr = pd->dst;
2773                 daddr = pd->src;
2774         }
2775         switch (pd->af) {
2776 #ifdef INET
2777         case AF_INET:
2778                 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
2779                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2780                 if (inp == NULL) {
2781                         inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
2782                            daddr->v4, dport, INPLOOKUP_WILDCARD |
2783                            INPLOOKUP_RLOCKPCB, NULL, m);
2784                         if (inp == NULL)
2785                                 return (-1);
2786                 }
2787                 break;
2788 #endif /* INET */
2789 #ifdef INET6
2790         case AF_INET6:
2791                 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
2792                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2793                 if (inp == NULL) {
2794                         inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
2795                             &daddr->v6, dport, INPLOOKUP_WILDCARD |
2796                             INPLOOKUP_RLOCKPCB, NULL, m);
2797                         if (inp == NULL)
2798                                 return (-1);
2799                 }
2800                 break;
2801 #endif /* INET6 */
2802
2803         default:
2804                 return (-1);
2805         }
2806         INP_RLOCK_ASSERT(inp);
2807         pd->lookup.uid = inp->inp_cred->cr_uid;
2808         pd->lookup.gid = inp->inp_cred->cr_groups[0];
2809         INP_RUNLOCK(inp);
2810
2811         return (1);
2812 }
2813
2814 static u_int8_t
2815 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2816 {
2817         int              hlen;
2818         u_int8_t         hdr[60];
2819         u_int8_t        *opt, optlen;
2820         u_int8_t         wscale = 0;
2821
2822         hlen = th_off << 2;             /* hlen <= sizeof(hdr) */
2823         if (hlen <= sizeof(struct tcphdr))
2824                 return (0);
2825         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2826                 return (0);
2827         opt = hdr + sizeof(struct tcphdr);
2828         hlen -= sizeof(struct tcphdr);
2829         while (hlen >= 3) {
2830                 switch (*opt) {
2831                 case TCPOPT_EOL:
2832                 case TCPOPT_NOP:
2833                         ++opt;
2834                         --hlen;
2835                         break;
2836                 case TCPOPT_WINDOW:
2837                         wscale = opt[2];
2838                         if (wscale > TCP_MAX_WINSHIFT)
2839                                 wscale = TCP_MAX_WINSHIFT;
2840                         wscale |= PF_WSCALE_FLAG;
2841                         /* FALLTHROUGH */
2842                 default:
2843                         optlen = opt[1];
2844                         if (optlen < 2)
2845                                 optlen = 2;
2846                         hlen -= optlen;
2847                         opt += optlen;
2848                         break;
2849                 }
2850         }
2851         return (wscale);
2852 }
2853
2854 static u_int16_t
2855 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2856 {
2857         int              hlen;
2858         u_int8_t         hdr[60];
2859         u_int8_t        *opt, optlen;
2860         u_int16_t        mss = V_tcp_mssdflt;
2861
2862         hlen = th_off << 2;     /* hlen <= sizeof(hdr) */
2863         if (hlen <= sizeof(struct tcphdr))
2864                 return (0);
2865         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2866                 return (0);
2867         opt = hdr + sizeof(struct tcphdr);
2868         hlen -= sizeof(struct tcphdr);
2869         while (hlen >= TCPOLEN_MAXSEG) {
2870                 switch (*opt) {
2871                 case TCPOPT_EOL:
2872                 case TCPOPT_NOP:
2873                         ++opt;
2874                         --hlen;
2875                         break;
2876                 case TCPOPT_MAXSEG:
2877                         bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
2878                         NTOHS(mss);
2879                         /* FALLTHROUGH */
2880                 default:
2881                         optlen = opt[1];
2882                         if (optlen < 2)
2883                                 optlen = 2;
2884                         hlen -= optlen;
2885                         opt += optlen;
2886                         break;
2887                 }
2888         }
2889         return (mss);
2890 }
2891
2892 static u_int16_t
2893 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
2894 {
2895 #ifdef INET
2896         struct sockaddr_in      *dst;
2897         struct route             ro;
2898 #endif /* INET */
2899 #ifdef INET6
2900         struct sockaddr_in6     *dst6;
2901         struct route_in6         ro6;
2902 #endif /* INET6 */
2903         struct rtentry          *rt = NULL;
2904         int                      hlen = 0;
2905         u_int16_t                mss = V_tcp_mssdflt;
2906
2907         switch (af) {
2908 #ifdef INET
2909         case AF_INET:
2910                 hlen = sizeof(struct ip);
2911                 bzero(&ro, sizeof(ro));
2912                 dst = (struct sockaddr_in *)&ro.ro_dst;
2913                 dst->sin_family = AF_INET;
2914                 dst->sin_len = sizeof(*dst);
2915                 dst->sin_addr = addr->v4;
2916                 in_rtalloc_ign(&ro, 0, rtableid);
2917                 rt = ro.ro_rt;
2918                 break;
2919 #endif /* INET */
2920 #ifdef INET6
2921         case AF_INET6:
2922                 hlen = sizeof(struct ip6_hdr);
2923                 bzero(&ro6, sizeof(ro6));
2924                 dst6 = (struct sockaddr_in6 *)&ro6.ro_dst;
2925                 dst6->sin6_family = AF_INET6;
2926                 dst6->sin6_len = sizeof(*dst6);
2927                 dst6->sin6_addr = addr->v6;
2928                 in6_rtalloc_ign(&ro6, 0, rtableid);
2929                 rt = ro6.ro_rt;
2930                 break;
2931 #endif /* INET6 */
2932         }
2933
2934         if (rt && rt->rt_ifp) {
2935                 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr);
2936                 mss = max(V_tcp_mssdflt, mss);
2937                 RTFREE(rt);
2938         }
2939         mss = min(mss, offer);
2940         mss = max(mss, 64);             /* sanity - at least max opt space */
2941         return (mss);
2942 }
2943
2944 static void
2945 pf_set_rt_ifp(struct pf_state *s, struct pf_addr *saddr)
2946 {
2947         struct pf_rule *r = s->rule.ptr;
2948         struct pf_src_node *sn = NULL;
2949
2950         s->rt_kif = NULL;
2951         if (!r->rt || r->rt == PF_FASTROUTE)
2952                 return;
2953         switch (s->key[PF_SK_WIRE]->af) {
2954 #ifdef INET
2955         case AF_INET:
2956                 pf_map_addr(AF_INET, r, saddr, &s->rt_addr, NULL, &sn);
2957                 s->rt_kif = r->rpool.cur->kif;
2958                 break;
2959 #endif /* INET */
2960 #ifdef INET6
2961         case AF_INET6:
2962                 pf_map_addr(AF_INET6, r, saddr, &s->rt_addr, NULL, &sn);
2963                 s->rt_kif = r->rpool.cur->kif;
2964                 break;
2965 #endif /* INET6 */
2966         }
2967 }
2968
2969 static u_int32_t
2970 pf_tcp_iss(struct pf_pdesc *pd)
2971 {
2972         MD5_CTX ctx;
2973         u_int32_t digest[4];
2974
2975         if (V_pf_tcp_secret_init == 0) {
2976                 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
2977                 MD5Init(&V_pf_tcp_secret_ctx);
2978                 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
2979                     sizeof(V_pf_tcp_secret));
2980                 V_pf_tcp_secret_init = 1;
2981         }
2982
2983         ctx = V_pf_tcp_secret_ctx;
2984
2985         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
2986         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
2987         if (pd->af == AF_INET6) {
2988                 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
2989                 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
2990         } else {
2991                 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
2992                 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
2993         }
2994         MD5Final((u_char *)digest, &ctx);
2995         V_pf_tcp_iss_off += 4096;
2996 #define ISN_RANDOM_INCREMENT (4096 - 1)
2997         return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
2998             V_pf_tcp_iss_off);
2999 #undef  ISN_RANDOM_INCREMENT
3000 }
3001
3002 static int
3003 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
3004     struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
3005     struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
3006 {
3007         struct pf_rule          *nr = NULL;
3008         struct pf_addr          * const saddr = pd->src;
3009         struct pf_addr          * const daddr = pd->dst;
3010         sa_family_t              af = pd->af;
3011         struct pf_rule          *r, *a = NULL;
3012         struct pf_ruleset       *ruleset = NULL;
3013         struct pf_src_node      *nsn = NULL;
3014         struct tcphdr           *th = pd->hdr.tcp;
3015         struct pf_state_key     *sk = NULL, *nk = NULL;
3016         u_short                  reason;
3017         int                      rewrite = 0, hdrlen = 0;
3018         int                      tag = -1, rtableid = -1;
3019         int                      asd = 0;
3020         int                      match = 0;
3021         int                      state_icmp = 0;
3022         u_int16_t                sport = 0, dport = 0;
3023         u_int16_t                bproto_sum = 0, bip_sum = 0;
3024         u_int8_t                 icmptype = 0, icmpcode = 0;
3025         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3026
3027         PF_RULES_RASSERT();
3028
3029         if (inp != NULL) {
3030                 INP_LOCK_ASSERT(inp);
3031                 pd->lookup.uid = inp->inp_cred->cr_uid;
3032                 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3033                 pd->lookup.done = 1;
3034         }
3035
3036         switch (pd->proto) {
3037         case IPPROTO_TCP:
3038                 sport = th->th_sport;
3039                 dport = th->th_dport;
3040                 hdrlen = sizeof(*th);
3041                 break;
3042         case IPPROTO_UDP:
3043                 sport = pd->hdr.udp->uh_sport;
3044                 dport = pd->hdr.udp->uh_dport;
3045                 hdrlen = sizeof(*pd->hdr.udp);
3046                 break;
3047 #ifdef INET
3048         case IPPROTO_ICMP:
3049                 if (pd->af != AF_INET)
3050                         break;
3051                 sport = dport = pd->hdr.icmp->icmp_id;
3052                 hdrlen = sizeof(*pd->hdr.icmp);
3053                 icmptype = pd->hdr.icmp->icmp_type;
3054                 icmpcode = pd->hdr.icmp->icmp_code;
3055
3056                 if (icmptype == ICMP_UNREACH ||
3057                     icmptype == ICMP_SOURCEQUENCH ||
3058                     icmptype == ICMP_REDIRECT ||
3059                     icmptype == ICMP_TIMXCEED ||
3060                     icmptype == ICMP_PARAMPROB)
3061                         state_icmp++;
3062                 break;
3063 #endif /* INET */
3064 #ifdef INET6
3065         case IPPROTO_ICMPV6:
3066                 if (af != AF_INET6)
3067                         break;
3068                 sport = dport = pd->hdr.icmp6->icmp6_id;
3069                 hdrlen = sizeof(*pd->hdr.icmp6);
3070                 icmptype = pd->hdr.icmp6->icmp6_type;
3071                 icmpcode = pd->hdr.icmp6->icmp6_code;
3072
3073                 if (icmptype == ICMP6_DST_UNREACH ||
3074                     icmptype == ICMP6_PACKET_TOO_BIG ||
3075                     icmptype == ICMP6_TIME_EXCEEDED ||
3076                     icmptype == ICMP6_PARAM_PROB)
3077                         state_icmp++;
3078                 break;
3079 #endif /* INET6 */
3080         default:
3081                 sport = dport = hdrlen = 0;
3082                 break;
3083         }
3084
3085         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3086
3087         /* check packet for BINAT/NAT/RDR */
3088         if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3089             &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3090                 KASSERT(sk != NULL, ("%s: null sk", __func__));
3091                 KASSERT(nk != NULL, ("%s: null nk", __func__));
3092
3093                 if (pd->ip_sum)
3094                         bip_sum = *pd->ip_sum;
3095
3096                 switch (pd->proto) {
3097                 case IPPROTO_TCP:
3098                         bproto_sum = th->th_sum;
3099                         pd->proto_sum = &th->th_sum;
3100
3101                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3102                             nk->port[pd->sidx] != sport) {
3103                                 pf_change_ap(saddr, &th->th_sport, pd->ip_sum,
3104                                     &th->th_sum, &nk->addr[pd->sidx],
3105                                     nk->port[pd->sidx], 0, af);
3106                                 pd->sport = &th->th_sport;
3107                                 sport = th->th_sport;
3108                         }
3109
3110                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3111                             nk->port[pd->didx] != dport) {
3112                                 pf_change_ap(daddr, &th->th_dport, pd->ip_sum,
3113                                     &th->th_sum, &nk->addr[pd->didx],
3114                                     nk->port[pd->didx], 0, af);
3115                                 dport = th->th_dport;
3116                                 pd->dport = &th->th_dport;
3117                         }
3118                         rewrite++;
3119                         break;
3120                 case IPPROTO_UDP:
3121                         bproto_sum = pd->hdr.udp->uh_sum;
3122                         pd->proto_sum = &pd->hdr.udp->uh_sum;
3123
3124                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3125                             nk->port[pd->sidx] != sport) {
3126                                 pf_change_ap(saddr, &pd->hdr.udp->uh_sport,
3127                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3128                                     &nk->addr[pd->sidx],
3129                                     nk->port[pd->sidx], 1, af);
3130                                 sport = pd->hdr.udp->uh_sport;
3131                                 pd->sport = &pd->hdr.udp->uh_sport;
3132                         }
3133
3134                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3135                             nk->port[pd->didx] != dport) {
3136                                 pf_change_ap(daddr, &pd->hdr.udp->uh_dport,
3137                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3138                                     &nk->addr[pd->didx],
3139                                     nk->port[pd->didx], 1, af);
3140                                 dport = pd->hdr.udp->uh_dport;
3141                                 pd->dport = &pd->hdr.udp->uh_dport;
3142                         }
3143                         rewrite++;
3144                         break;
3145 #ifdef INET
3146                 case IPPROTO_ICMP:
3147                         nk->port[0] = nk->port[1];
3148                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3149                                 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3150                                     nk->addr[pd->sidx].v4.s_addr, 0);
3151
3152                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3153                                 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3154                                     nk->addr[pd->didx].v4.s_addr, 0);
3155
3156                         if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3157                                 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3158                                     pd->hdr.icmp->icmp_cksum, sport,
3159                                     nk->port[1], 0);
3160                                 pd->hdr.icmp->icmp_id = nk->port[1];
3161                                 pd->sport = &pd->hdr.icmp->icmp_id;
3162                         }
3163                         m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3164                         break;
3165 #endif /* INET */
3166 #ifdef INET6
3167                 case IPPROTO_ICMPV6:
3168                         nk->port[0] = nk->port[1];
3169                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3170                                 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3171                                     &nk->addr[pd->sidx], 0);
3172
3173                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3174                                 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3175                                     &nk->addr[pd->didx], 0);
3176                         rewrite++;
3177                         break;
3178 #endif /* INET */
3179                 default:
3180                         switch (af) {
3181 #ifdef INET
3182                         case AF_INET:
3183                                 if (PF_ANEQ(saddr,
3184                                     &nk->addr[pd->sidx], AF_INET))
3185                                         pf_change_a(&saddr->v4.s_addr,
3186                                             pd->ip_sum,
3187                                             nk->addr[pd->sidx].v4.s_addr, 0);
3188
3189                                 if (PF_ANEQ(daddr,
3190                                     &nk->addr[pd->didx], AF_INET))
3191                                         pf_change_a(&daddr->v4.s_addr,
3192                                             pd->ip_sum,
3193                                             nk->addr[pd->didx].v4.s_addr, 0);
3194                                 break;
3195 #endif /* INET */
3196 #ifdef INET6
3197                         case AF_INET6:
3198                                 if (PF_ANEQ(saddr,
3199                                     &nk->addr[pd->sidx], AF_INET6))
3200                                         PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3201
3202                                 if (PF_ANEQ(daddr,
3203                                     &nk->addr[pd->didx], AF_INET6))
3204                                         PF_ACPY(saddr, &nk->addr[pd->didx], af);
3205                                 break;
3206 #endif /* INET */
3207                         }
3208                         break;
3209                 }
3210                 if (nr->natpass)
3211                         r = NULL;
3212                 pd->nat_rule = nr;
3213         }
3214
3215         while (r != NULL) {
3216                 r->evaluations++;
3217                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3218                         r = r->skip[PF_SKIP_IFP].ptr;
3219                 else if (r->direction && r->direction != direction)
3220                         r = r->skip[PF_SKIP_DIR].ptr;
3221                 else if (r->af && r->af != af)
3222                         r = r->skip[PF_SKIP_AF].ptr;
3223                 else if (r->proto && r->proto != pd->proto)
3224                         r = r->skip[PF_SKIP_PROTO].ptr;
3225                 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3226                     r->src.neg, kif, M_GETFIB(m)))
3227                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3228                 /* tcp/udp only. port_op always 0 in other cases */
3229                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3230                     r->src.port[0], r->src.port[1], sport))
3231                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
3232                 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3233                     r->dst.neg, NULL, M_GETFIB(m)))
3234                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3235                 /* tcp/udp only. port_op always 0 in other cases */
3236                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3237                     r->dst.port[0], r->dst.port[1], dport))
3238                         r = r->skip[PF_SKIP_DST_PORT].ptr;
3239                 /* icmp only. type always 0 in other cases */
3240                 else if (r->type && r->type != icmptype + 1)
3241                         r = TAILQ_NEXT(r, entries);
3242                 /* icmp only. type always 0 in other cases */
3243                 else if (r->code && r->code != icmpcode + 1)
3244                         r = TAILQ_NEXT(r, entries);
3245                 else if (r->tos && !(r->tos == pd->tos))
3246                         r = TAILQ_NEXT(r, entries);
3247                 else if (r->rule_flag & PFRULE_FRAGMENT)
3248                         r = TAILQ_NEXT(r, entries);
3249                 else if (pd->proto == IPPROTO_TCP &&
3250                     (r->flagset & th->th_flags) != r->flags)
3251                         r = TAILQ_NEXT(r, entries);
3252                 /* tcp/udp only. uid.op always 0 in other cases */
3253                 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3254                     pf_socket_lookup(direction, pd, m), 1)) &&
3255                     !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3256                     pd->lookup.uid))
3257                         r = TAILQ_NEXT(r, entries);
3258                 /* tcp/udp only. gid.op always 0 in other cases */
3259                 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3260                     pf_socket_lookup(direction, pd, m), 1)) &&
3261                     !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3262                     pd->lookup.gid))
3263                         r = TAILQ_NEXT(r, entries);
3264                 else if (r->prob &&
3265                     r->prob <= arc4random())
3266                         r = TAILQ_NEXT(r, entries);
3267                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3268                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3269                         r = TAILQ_NEXT(r, entries);
3270                 else if (r->os_fingerprint != PF_OSFP_ANY &&
3271                     (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3272                     pf_osfp_fingerprint(pd, m, off, th),
3273                     r->os_fingerprint)))
3274                         r = TAILQ_NEXT(r, entries);
3275                 else {
3276                         if (r->tag)
3277                                 tag = r->tag;
3278                         if (r->rtableid >= 0)
3279                                 rtableid = r->rtableid;
3280                         if (r->anchor == NULL) {
3281                                 match = 1;
3282                                 *rm = r;
3283                                 *am = a;
3284                                 *rsm = ruleset;
3285                                 if ((*rm)->quick)
3286                                         break;
3287                                 r = TAILQ_NEXT(r, entries);
3288                         } else
3289                                 pf_step_into_anchor(anchor_stack, &asd,
3290                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3291                                     &match);
3292                 }
3293                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3294                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3295                         break;
3296         }
3297         r = *rm;
3298         a = *am;
3299         ruleset = *rsm;
3300
3301         REASON_SET(&reason, PFRES_MATCH);
3302
3303         if (r->log || (nr != NULL && nr->log)) {
3304                 if (rewrite)
3305                         m_copyback(m, off, hdrlen, pd->hdr.any);
3306                 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3307                     ruleset, pd, 1);
3308         }
3309
3310         if ((r->action == PF_DROP) &&
3311             ((r->rule_flag & PFRULE_RETURNRST) ||
3312             (r->rule_flag & PFRULE_RETURNICMP) ||
3313             (r->rule_flag & PFRULE_RETURN))) {
3314                 /* undo NAT changes, if they have taken place */
3315                 if (nr != NULL) {
3316                         PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3317                         PF_ACPY(daddr, &sk->addr[pd->didx], af);
3318                         if (pd->sport)
3319                                 *pd->sport = sk->port[pd->sidx];
3320                         if (pd->dport)
3321                                 *pd->dport = sk->port[pd->didx];
3322                         if (pd->proto_sum)
3323                                 *pd->proto_sum = bproto_sum;
3324                         if (pd->ip_sum)
3325                                 *pd->ip_sum = bip_sum;
3326                         m_copyback(m, off, hdrlen, pd->hdr.any);
3327                 }
3328                 if (pd->proto == IPPROTO_TCP &&
3329                     ((r->rule_flag & PFRULE_RETURNRST) ||
3330                     (r->rule_flag & PFRULE_RETURN)) &&
3331                     !(th->th_flags & TH_RST)) {
3332                         u_int32_t        ack = ntohl(th->th_seq) + pd->p_len;
3333                         int              len = 0;
3334 #ifdef INET
3335                         struct ip       *h4;
3336 #endif
3337 #ifdef INET6
3338                         struct ip6_hdr  *h6;
3339 #endif
3340
3341                         switch (af) {
3342 #ifdef INET
3343                         case AF_INET:
3344                                 h4 = mtod(m, struct ip *);
3345                                 len = ntohs(h4->ip_len) - off;
3346                                 break;
3347 #endif
3348 #ifdef INET6
3349                         case AF_INET6:
3350                                 h6 = mtod(m, struct ip6_hdr *);
3351                                 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3352                                 break;
3353 #endif
3354                         }
3355
3356                         if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3357                                 REASON_SET(&reason, PFRES_PROTCKSUM);
3358                         else {
3359                                 if (th->th_flags & TH_SYN)
3360                                         ack++;
3361                                 if (th->th_flags & TH_FIN)
3362                                         ack++;
3363                                 pf_send_tcp(m, r, af, pd->dst,
3364                                     pd->src, th->th_dport, th->th_sport,
3365                                     ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3366                                     r->return_ttl, 1, 0, kif->pfik_ifp);
3367                         }
3368                 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3369                     r->return_icmp)
3370                         pf_send_icmp(m, r->return_icmp >> 8,
3371                             r->return_icmp & 255, af, r);
3372                 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3373                     r->return_icmp6)
3374                         pf_send_icmp(m, r->return_icmp6 >> 8,
3375                             r->return_icmp6 & 255, af, r);
3376         }
3377
3378         if (r->action == PF_DROP)
3379                 goto cleanup;
3380
3381         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3382                 REASON_SET(&reason, PFRES_MEMORY);
3383                 goto cleanup;
3384         }
3385         if (rtableid >= 0)
3386                 M_SETFIB(m, rtableid);
3387
3388         if (!state_icmp && (r->keep_state || nr != NULL ||
3389             (pd->flags & PFDESC_TCP_NORM))) {
3390                 int action;
3391                 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3392                     sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3393                     hdrlen);
3394                 if (action != PF_PASS)
3395                         return (action);
3396         } else {
3397                 if (sk != NULL)
3398                         uma_zfree(V_pf_state_key_z, sk);
3399                 if (nk != NULL)
3400                         uma_zfree(V_pf_state_key_z, nk);
3401         }
3402
3403         /* copy back packet headers if we performed NAT operations */
3404         if (rewrite)
3405                 m_copyback(m, off, hdrlen, pd->hdr.any);
3406
3407         if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3408             direction == PF_OUT &&
3409             pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m))
3410                 /*
3411                  * We want the state created, but we dont
3412                  * want to send this in case a partner
3413                  * firewall has to know about it to allow
3414                  * replies through it.
3415                  */
3416                 return (PF_DEFER);
3417
3418         return (PF_PASS);
3419
3420 cleanup:
3421         if (sk != NULL)
3422                 uma_zfree(V_pf_state_key_z, sk);
3423         if (nk != NULL)
3424                 uma_zfree(V_pf_state_key_z, nk);
3425         return (PF_DROP);
3426 }
3427
3428 static int
3429 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3430     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3431     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3432     u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3433     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3434 {
3435         struct pf_state         *s = NULL;
3436         struct pf_src_node      *sn = NULL;
3437         struct tcphdr           *th = pd->hdr.tcp;
3438         u_int16_t                mss = V_tcp_mssdflt;
3439         u_short                  reason;
3440
3441         /* check maximums */
3442         if (r->max_states &&
3443             (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3444                 V_pf_status.lcounters[LCNT_STATES]++;
3445                 REASON_SET(&reason, PFRES_MAXSTATES);
3446                 return (PF_DROP);
3447         }
3448         /* src node for filter rule */
3449         if ((r->rule_flag & PFRULE_SRCTRACK ||
3450             r->rpool.opts & PF_POOL_STICKYADDR) &&
3451             pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3452                 REASON_SET(&reason, PFRES_SRCLIMIT);
3453                 goto csfailed;
3454         }
3455         /* src node for translation rule */
3456         if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3457             pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3458                 REASON_SET(&reason, PFRES_SRCLIMIT);
3459                 goto csfailed;
3460         }
3461         s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3462         if (s == NULL) {
3463                 REASON_SET(&reason, PFRES_MEMORY);
3464                 goto csfailed;
3465         }
3466         s->rule.ptr = r;
3467         s->nat_rule.ptr = nr;
3468         s->anchor.ptr = a;
3469         STATE_INC_COUNTERS(s);
3470         if (r->allow_opts)
3471                 s->state_flags |= PFSTATE_ALLOWOPTS;
3472         if (r->rule_flag & PFRULE_STATESLOPPY)
3473                 s->state_flags |= PFSTATE_SLOPPY;
3474         s->log = r->log & PF_LOG_ALL;
3475         s->sync_state = PFSYNC_S_NONE;
3476         if (nr != NULL)
3477                 s->log |= nr->log & PF_LOG_ALL;
3478         switch (pd->proto) {
3479         case IPPROTO_TCP:
3480                 s->src.seqlo = ntohl(th->th_seq);
3481                 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3482                 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3483                     r->keep_state == PF_STATE_MODULATE) {
3484                         /* Generate sequence number modulator */
3485                         if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3486                             0)
3487                                 s->src.seqdiff = 1;
3488                         pf_change_a(&th->th_seq, &th->th_sum,
3489                             htonl(s->src.seqlo + s->src.seqdiff), 0);
3490                         *rewrite = 1;
3491                 } else
3492                         s->src.seqdiff = 0;
3493                 if (th->th_flags & TH_SYN) {
3494                         s->src.seqhi++;
3495                         s->src.wscale = pf_get_wscale(m, off,
3496                             th->th_off, pd->af);
3497                 }
3498                 s->src.max_win = MAX(ntohs(th->th_win), 1);
3499                 if (s->src.wscale & PF_WSCALE_MASK) {
3500                         /* Remove scale factor from initial window */
3501                         int win = s->src.max_win;
3502                         win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3503                         s->src.max_win = (win - 1) >>
3504                             (s->src.wscale & PF_WSCALE_MASK);
3505                 }
3506                 if (th->th_flags & TH_FIN)
3507                         s->src.seqhi++;
3508                 s->dst.seqhi = 1;
3509                 s->dst.max_win = 1;
3510                 s->src.state = TCPS_SYN_SENT;
3511                 s->dst.state = TCPS_CLOSED;
3512                 s->timeout = PFTM_TCP_FIRST_PACKET;
3513                 break;
3514         case IPPROTO_UDP:
3515                 s->src.state = PFUDPS_SINGLE;
3516                 s->dst.state = PFUDPS_NO_TRAFFIC;
3517                 s->timeout = PFTM_UDP_FIRST_PACKET;
3518                 break;
3519         case IPPROTO_ICMP:
3520 #ifdef INET6
3521         case IPPROTO_ICMPV6:
3522 #endif
3523                 s->timeout = PFTM_ICMP_FIRST_PACKET;
3524                 break;
3525         default:
3526                 s->src.state = PFOTHERS_SINGLE;
3527                 s->dst.state = PFOTHERS_NO_TRAFFIC;
3528                 s->timeout = PFTM_OTHER_FIRST_PACKET;
3529         }
3530
3531         s->creation = time_uptime;
3532         s->expire = time_uptime;
3533
3534         if (sn != NULL) {
3535                 s->src_node = sn;
3536                 s->src_node->states++;
3537         }
3538         if (nsn != NULL) {
3539                 /* XXX We only modify one side for now. */
3540                 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3541                 s->nat_src_node = nsn;
3542                 s->nat_src_node->states++;
3543         }
3544         if (pd->proto == IPPROTO_TCP) {
3545                 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3546                     off, pd, th, &s->src, &s->dst)) {
3547                         REASON_SET(&reason, PFRES_MEMORY);
3548                         pf_src_tree_remove_state(s);
3549                         STATE_DEC_COUNTERS(s);
3550                         uma_zfree(V_pf_state_z, s);
3551                         return (PF_DROP);
3552                 }
3553                 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3554                     pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3555                     &s->src, &s->dst, rewrite)) {
3556                         /* This really shouldn't happen!!! */
3557                         DPFPRINTF(PF_DEBUG_URGENT,
3558                             ("pf_normalize_tcp_stateful failed on first pkt"));
3559                         pf_normalize_tcp_cleanup(s);
3560                         pf_src_tree_remove_state(s);
3561                         STATE_DEC_COUNTERS(s);
3562                         uma_zfree(V_pf_state_z, s);
3563                         return (PF_DROP);
3564                 }
3565         }
3566         s->direction = pd->dir;
3567
3568         /*
3569          * sk/nk could already been setup by pf_get_translation().
3570          */
3571         if (nr == NULL) {
3572                 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3573                     __func__, nr, sk, nk));
3574                 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3575                 if (sk == NULL)
3576                         goto csfailed;
3577                 nk = sk;
3578         } else
3579                 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3580                     __func__, nr, sk, nk));
3581
3582         /* Swap sk/nk for PF_OUT. */
3583         if (pf_state_insert(BOUND_IFACE(r, kif),
3584             (pd->dir == PF_IN) ? sk : nk,
3585             (pd->dir == PF_IN) ? nk : sk, s)) {
3586                 if (pd->proto == IPPROTO_TCP)
3587                         pf_normalize_tcp_cleanup(s);
3588                 REASON_SET(&reason, PFRES_STATEINS);
3589                 pf_src_tree_remove_state(s);
3590                 STATE_DEC_COUNTERS(s);
3591                 uma_zfree(V_pf_state_z, s);
3592                 return (PF_DROP);
3593         } else
3594                 *sm = s;
3595
3596         pf_set_rt_ifp(s, pd->src);      /* needs s->state_key set */
3597         if (tag > 0)
3598                 s->tag = tag;
3599         if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3600             TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3601                 s->src.state = PF_TCPS_PROXY_SRC;
3602                 /* undo NAT changes, if they have taken place */
3603                 if (nr != NULL) {
3604                         struct pf_state_key *skt = s->key[PF_SK_WIRE];
3605                         if (pd->dir == PF_OUT)
3606                                 skt = s->key[PF_SK_STACK];
3607                         PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3608                         PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3609                         if (pd->sport)
3610                                 *pd->sport = skt->port[pd->sidx];
3611                         if (pd->dport)
3612                                 *pd->dport = skt->port[pd->didx];
3613                         if (pd->proto_sum)
3614                                 *pd->proto_sum = bproto_sum;
3615                         if (pd->ip_sum)
3616                                 *pd->ip_sum = bip_sum;
3617                         m_copyback(m, off, hdrlen, pd->hdr.any);
3618                 }
3619                 s->src.seqhi = htonl(arc4random());
3620                 /* Find mss option */
3621                 int rtid = M_GETFIB(m);
3622                 mss = pf_get_mss(m, off, th->th_off, pd->af);
3623                 mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3624                 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3625                 s->src.mss = mss;
3626                 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3627                     th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3628                     TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3629                 REASON_SET(&reason, PFRES_SYNPROXY);
3630                 return (PF_SYNPROXY_DROP);
3631         }
3632
3633         return (PF_PASS);
3634
3635 csfailed:
3636         if (sk != NULL)
3637                 uma_zfree(V_pf_state_key_z, sk);
3638         if (nk != NULL)
3639                 uma_zfree(V_pf_state_key_z, nk);
3640
3641         if (sn != NULL && sn->states == 0 && sn->expire == 0) {
3642                 pf_unlink_src_node(sn);
3643                 pf_free_src_node(sn);
3644         }
3645
3646         if (nsn != sn && nsn != NULL && nsn->states == 0 && nsn->expire == 0) {
3647                 pf_unlink_src_node(nsn);
3648                 pf_free_src_node(nsn);
3649         }
3650
3651         return (PF_DROP);
3652 }
3653
3654 static int
3655 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
3656     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
3657     struct pf_ruleset **rsm)
3658 {
3659         struct pf_rule          *r, *a = NULL;
3660         struct pf_ruleset       *ruleset = NULL;
3661         sa_family_t              af = pd->af;
3662         u_short                  reason;
3663         int                      tag = -1;
3664         int                      asd = 0;
3665         int                      match = 0;
3666         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3667
3668         PF_RULES_RASSERT();
3669
3670         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3671         while (r != NULL) {
3672                 r->evaluations++;
3673                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3674                         r = r->skip[PF_SKIP_IFP].ptr;
3675                 else if (r->direction && r->direction != direction)
3676                         r = r->skip[PF_SKIP_DIR].ptr;
3677                 else if (r->af && r->af != af)
3678                         r = r->skip[PF_SKIP_AF].ptr;
3679                 else if (r->proto && r->proto != pd->proto)
3680                         r = r->skip[PF_SKIP_PROTO].ptr;
3681                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3682                     r->src.neg, kif, M_GETFIB(m)))
3683                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3684                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3685                     r->dst.neg, NULL, M_GETFIB(m)))
3686                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3687                 else if (r->tos && !(r->tos == pd->tos))
3688                         r = TAILQ_NEXT(r, entries);
3689                 else if (r->os_fingerprint != PF_OSFP_ANY)
3690                         r = TAILQ_NEXT(r, entries);
3691                 else if (pd->proto == IPPROTO_UDP &&
3692                     (r->src.port_op || r->dst.port_op))
3693                         r = TAILQ_NEXT(r, entries);
3694                 else if (pd->proto == IPPROTO_TCP &&
3695                     (r->src.port_op || r->dst.port_op || r->flagset))
3696                         r = TAILQ_NEXT(r, entries);
3697                 else if ((pd->proto == IPPROTO_ICMP ||
3698                     pd->proto == IPPROTO_ICMPV6) &&
3699                     (r->type || r->code))
3700                         r = TAILQ_NEXT(r, entries);
3701                 else if (r->prob && r->prob <=
3702                     (arc4random() % (UINT_MAX - 1) + 1))
3703                         r = TAILQ_NEXT(r, entries);
3704                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3705                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3706                         r = TAILQ_NEXT(r, entries);
3707                 else {
3708                         if (r->anchor == NULL) {
3709                                 match = 1;
3710                                 *rm = r;
3711                                 *am = a;
3712                                 *rsm = ruleset;
3713                                 if ((*rm)->quick)
3714                                         break;
3715                                 r = TAILQ_NEXT(r, entries);
3716                         } else
3717                                 pf_step_into_anchor(anchor_stack, &asd,
3718                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3719                                     &match);
3720                 }
3721                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3722                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3723                         break;
3724         }
3725         r = *rm;
3726         a = *am;
3727         ruleset = *rsm;
3728
3729         REASON_SET(&reason, PFRES_MATCH);
3730
3731         if (r->log)
3732                 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
3733                     1);
3734
3735         if (r->action != PF_PASS)
3736                 return (PF_DROP);
3737
3738         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3739                 REASON_SET(&reason, PFRES_MEMORY);
3740                 return (PF_DROP);
3741         }
3742
3743         return (PF_PASS);
3744 }
3745
3746 static int
3747 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
3748         struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
3749         struct pf_pdesc *pd, u_short *reason, int *copyback)
3750 {
3751         struct tcphdr           *th = pd->hdr.tcp;
3752         u_int16_t                win = ntohs(th->th_win);
3753         u_int32_t                ack, end, seq, orig_seq;
3754         u_int8_t                 sws, dws;
3755         int                      ackskew;
3756
3757         if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
3758                 sws = src->wscale & PF_WSCALE_MASK;
3759                 dws = dst->wscale & PF_WSCALE_MASK;
3760         } else
3761                 sws = dws = 0;
3762
3763         /*
3764          * Sequence tracking algorithm from Guido van Rooij's paper:
3765          *   http://www.madison-gurkha.com/publications/tcp_filtering/
3766          *      tcp_filtering.ps
3767          */
3768
3769         orig_seq = seq = ntohl(th->th_seq);
3770         if (src->seqlo == 0) {
3771                 /* First packet from this end. Set its state */
3772
3773                 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
3774                     src->scrub == NULL) {
3775                         if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
3776                                 REASON_SET(reason, PFRES_MEMORY);
3777                                 return (PF_DROP);
3778                         }
3779                 }
3780
3781                 /* Deferred generation of sequence number modulator */
3782                 if (dst->seqdiff && !src->seqdiff) {
3783                         /* use random iss for the TCP server */
3784                         while ((src->seqdiff = arc4random() - seq) == 0)
3785                                 ;
3786                         ack = ntohl(th->th_ack) - dst->seqdiff;
3787                         pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
3788                             src->seqdiff), 0);
3789                         pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
3790                         *copyback = 1;
3791                 } else {
3792                         ack = ntohl(th->th_ack);
3793                 }
3794
3795                 end = seq + pd->p_len;
3796                 if (th->th_flags & TH_SYN) {
3797                         end++;
3798                         if (dst->wscale & PF_WSCALE_FLAG) {
3799                                 src->wscale = pf_get_wscale(m, off, th->th_off,
3800                                     pd->af);
3801                                 if (src->wscale & PF_WSCALE_FLAG) {
3802                                         /* Remove scale factor from initial
3803                                          * window */
3804                                         sws = src->wscale & PF_WSCALE_MASK;
3805                                         win = ((u_int32_t)win + (1 << sws) - 1)
3806                                             >> sws;
3807                                         dws = dst->wscale & PF_WSCALE_MASK;
3808                                 } else {
3809                                         /* fixup other window */
3810                                         dst->max_win <<= dst->wscale &
3811                                             PF_WSCALE_MASK;
3812                                         /* in case of a retrans SYN|ACK */
3813                                         dst->wscale = 0;
3814                                 }
3815                         }
3816                 }
3817                 if (th->th_flags & TH_FIN)
3818                         end++;
3819
3820                 src->seqlo = seq;
3821                 if (src->state < TCPS_SYN_SENT)
3822                         src->state = TCPS_SYN_SENT;
3823
3824                 /*
3825                  * May need to slide the window (seqhi may have been set by
3826                  * the crappy stack check or if we picked up the connection
3827                  * after establishment)
3828                  */
3829                 if (src->seqhi == 1 ||
3830                     SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
3831                         src->seqhi = end + MAX(1, dst->max_win << dws);
3832                 if (win > src->max_win)
3833                         src->max_win = win;
3834
3835         } else {
3836                 ack = ntohl(th->th_ack) - dst->seqdiff;
3837                 if (src->seqdiff) {
3838                         /* Modulate sequence numbers */
3839                         pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
3840                             src->seqdiff), 0);
3841                         pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
3842                         *copyback = 1;
3843                 }
3844                 end = seq + pd->p_len;
3845                 if (th->th_flags & TH_SYN)
3846                         end++;
3847                 if (th->th_flags & TH_FIN)
3848                         end++;
3849         }
3850
3851         if ((th->th_flags & TH_ACK) == 0) {
3852                 /* Let it pass through the ack skew check */
3853                 ack = dst->seqlo;
3854         } else if ((ack == 0 &&
3855             (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
3856             /* broken tcp stacks do not set ack */
3857             (dst->state < TCPS_SYN_SENT)) {
3858                 /*
3859                  * Many stacks (ours included) will set the ACK number in an
3860                  * FIN|ACK if the SYN times out -- no sequence to ACK.
3861                  */
3862                 ack = dst->seqlo;
3863         }
3864
3865         if (seq == end) {
3866                 /* Ease sequencing restrictions on no data packets */
3867                 seq = src->seqlo;
3868                 end = seq;
3869         }
3870
3871         ackskew = dst->seqlo - ack;
3872
3873
3874         /*
3875          * Need to demodulate the sequence numbers in any TCP SACK options
3876          * (Selective ACK). We could optionally validate the SACK values
3877          * against the current ACK window, either forwards or backwards, but
3878          * I'm not confident that SACK has been implemented properly
3879          * everywhere. It wouldn't surprise me if several stacks accidently
3880          * SACK too far backwards of previously ACKed data. There really aren't
3881          * any security implications of bad SACKing unless the target stack
3882          * doesn't validate the option length correctly. Someone trying to
3883          * spoof into a TCP connection won't bother blindly sending SACK
3884          * options anyway.
3885          */
3886         if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
3887                 if (pf_modulate_sack(m, off, pd, th, dst))
3888                         *copyback = 1;
3889         }
3890
3891
3892 #define MAXACKWINDOW (0xffff + 1500)    /* 1500 is an arbitrary fudge factor */
3893         if (SEQ_GEQ(src->seqhi, end) &&
3894             /* Last octet inside other's window space */
3895             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
3896             /* Retrans: not more than one window back */
3897             (ackskew >= -MAXACKWINDOW) &&
3898             /* Acking not more than one reassembled fragment backwards */
3899             (ackskew <= (MAXACKWINDOW << sws)) &&
3900             /* Acking not more than one window forward */
3901             ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
3902             (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
3903             (pd->flags & PFDESC_IP_REAS) == 0)) {
3904             /* Require an exact/+1 sequence match on resets when possible */
3905
3906                 if (dst->scrub || src->scrub) {
3907                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
3908                             *state, src, dst, copyback))
3909                                 return (PF_DROP);
3910                 }
3911
3912                 /* update max window */
3913                 if (src->max_win < win)
3914                         src->max_win = win;
3915                 /* synchronize sequencing */
3916                 if (SEQ_GT(end, src->seqlo))
3917                         src->seqlo = end;
3918                 /* slide the window of what the other end can send */
3919                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
3920                         dst->seqhi = ack + MAX((win << sws), 1);
3921
3922
3923                 /* update states */
3924                 if (th->th_flags & TH_SYN)
3925                         if (src->state < TCPS_SYN_SENT)
3926                                 src->state = TCPS_SYN_SENT;
3927                 if (th->th_flags & TH_FIN)
3928                         if (src->state < TCPS_CLOSING)
3929                                 src->state = TCPS_CLOSING;
3930                 if (th->th_flags & TH_ACK) {
3931                         if (dst->state == TCPS_SYN_SENT) {
3932                                 dst->state = TCPS_ESTABLISHED;
3933                                 if (src->state == TCPS_ESTABLISHED &&
3934                                     (*state)->src_node != NULL &&
3935                                     pf_src_connlimit(state)) {
3936                                         REASON_SET(reason, PFRES_SRCLIMIT);
3937                                         return (PF_DROP);
3938                                 }
3939                         } else if (dst->state == TCPS_CLOSING)
3940                                 dst->state = TCPS_FIN_WAIT_2;
3941                 }
3942                 if (th->th_flags & TH_RST)
3943                         src->state = dst->state = TCPS_TIME_WAIT;
3944
3945                 /* update expire time */
3946                 (*state)->expire = time_uptime;
3947                 if (src->state >= TCPS_FIN_WAIT_2 &&
3948                     dst->state >= TCPS_FIN_WAIT_2)
3949                         (*state)->timeout = PFTM_TCP_CLOSED;
3950                 else if (src->state >= TCPS_CLOSING &&
3951                     dst->state >= TCPS_CLOSING)
3952                         (*state)->timeout = PFTM_TCP_FIN_WAIT;
3953                 else if (src->state < TCPS_ESTABLISHED ||
3954                     dst->state < TCPS_ESTABLISHED)
3955                         (*state)->timeout = PFTM_TCP_OPENING;
3956                 else if (src->state >= TCPS_CLOSING ||
3957                     dst->state >= TCPS_CLOSING)
3958                         (*state)->timeout = PFTM_TCP_CLOSING;
3959                 else
3960                         (*state)->timeout = PFTM_TCP_ESTABLISHED;
3961
3962                 /* Fall through to PASS packet */
3963
3964         } else if ((dst->state < TCPS_SYN_SENT ||
3965                 dst->state >= TCPS_FIN_WAIT_2 ||
3966                 src->state >= TCPS_FIN_WAIT_2) &&
3967             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
3968             /* Within a window forward of the originating packet */
3969             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
3970             /* Within a window backward of the originating packet */
3971
3972                 /*
3973                  * This currently handles three situations:
3974                  *  1) Stupid stacks will shotgun SYNs before their peer
3975                  *     replies.
3976                  *  2) When PF catches an already established stream (the
3977                  *     firewall rebooted, the state table was flushed, routes
3978                  *     changed...)
3979                  *  3) Packets get funky immediately after the connection
3980                  *     closes (this should catch Solaris spurious ACK|FINs
3981                  *     that web servers like to spew after a close)
3982                  *
3983                  * This must be a little more careful than the above code
3984                  * since packet floods will also be caught here. We don't
3985                  * update the TTL here to mitigate the damage of a packet
3986                  * flood and so the same code can handle awkward establishment
3987                  * and a loosened connection close.
3988                  * In the establishment case, a correct peer response will
3989                  * validate the connection, go through the normal state code
3990                  * and keep updating the state TTL.
3991                  */
3992
3993                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
3994                         printf("pf: loose state match: ");
3995                         pf_print_state(*state);
3996                         pf_print_flags(th->th_flags);
3997                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
3998                             "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
3999                             pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
4000                             (unsigned long long)(*state)->packets[1],
4001                             pd->dir == PF_IN ? "in" : "out",
4002                             pd->dir == (*state)->direction ? "fwd" : "rev");
4003                 }
4004
4005                 if (dst->scrub || src->scrub) {
4006                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4007                             *state, src, dst, copyback))
4008                                 return (PF_DROP);
4009                 }
4010
4011                 /* update max window */
4012                 if (src->max_win < win)
4013                         src->max_win = win;
4014                 /* synchronize sequencing */
4015                 if (SEQ_GT(end, src->seqlo))
4016                         src->seqlo = end;
4017                 /* slide the window of what the other end can send */
4018                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4019                         dst->seqhi = ack + MAX((win << sws), 1);
4020
4021                 /*
4022                  * Cannot set dst->seqhi here since this could be a shotgunned
4023                  * SYN and not an already established connection.
4024                  */
4025
4026                 if (th->th_flags & TH_FIN)
4027                         if (src->state < TCPS_CLOSING)
4028                                 src->state = TCPS_CLOSING;
4029                 if (th->th_flags & TH_RST)
4030                         src->state = dst->state = TCPS_TIME_WAIT;
4031
4032                 /* Fall through to PASS packet */
4033
4034         } else {
4035                 if ((*state)->dst.state == TCPS_SYN_SENT &&
4036                     (*state)->src.state == TCPS_SYN_SENT) {
4037                         /* Send RST for state mismatches during handshake */
4038                         if (!(th->th_flags & TH_RST))
4039                                 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4040                                     pd->dst, pd->src, th->th_dport,
4041                                     th->th_sport, ntohl(th->th_ack), 0,
4042                                     TH_RST, 0, 0,
4043                                     (*state)->rule.ptr->return_ttl, 1, 0,
4044                                     kif->pfik_ifp);
4045                         src->seqlo = 0;
4046                         src->seqhi = 1;
4047                         src->max_win = 1;
4048                 } else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4049                         printf("pf: BAD state: ");
4050                         pf_print_state(*state);
4051                         pf_print_flags(th->th_flags);
4052                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4053                             "pkts=%llu:%llu dir=%s,%s\n",
4054                             seq, orig_seq, ack, pd->p_len, ackskew,
4055                             (unsigned long long)(*state)->packets[0],
4056                             (unsigned long long)(*state)->packets[1],
4057                             pd->dir == PF_IN ? "in" : "out",
4058                             pd->dir == (*state)->direction ? "fwd" : "rev");
4059                         printf("pf: State failure on: %c %c %c %c | %c %c\n",
4060                             SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4061                             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4062                             ' ': '2',
4063                             (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4064                             (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4065                             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4066                             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4067                 }
4068                 REASON_SET(reason, PFRES_BADSTATE);
4069                 return (PF_DROP);
4070         }
4071
4072         return (PF_PASS);
4073 }
4074
4075 static int
4076 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4077         struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4078 {
4079         struct tcphdr           *th = pd->hdr.tcp;
4080
4081         if (th->th_flags & TH_SYN)
4082                 if (src->state < TCPS_SYN_SENT)
4083                         src->state = TCPS_SYN_SENT;
4084         if (th->th_flags & TH_FIN)
4085                 if (src->state < TCPS_CLOSING)
4086                         src->state = TCPS_CLOSING;
4087         if (th->th_flags & TH_ACK) {
4088                 if (dst->state == TCPS_SYN_SENT) {
4089                         dst->state = TCPS_ESTABLISHED;
4090                         if (src->state == TCPS_ESTABLISHED &&
4091                             (*state)->src_node != NULL &&
4092                             pf_src_connlimit(state)) {
4093                                 REASON_SET(reason, PFRES_SRCLIMIT);
4094                                 return (PF_DROP);
4095                         }
4096                 } else if (dst->state == TCPS_CLOSING) {
4097                         dst->state = TCPS_FIN_WAIT_2;
4098                 } else if (src->state == TCPS_SYN_SENT &&
4099                     dst->state < TCPS_SYN_SENT) {
4100                         /*
4101                          * Handle a special sloppy case where we only see one
4102                          * half of the connection. If there is a ACK after
4103                          * the initial SYN without ever seeing a packet from
4104                          * the destination, set the connection to established.
4105                          */
4106                         dst->state = src->state = TCPS_ESTABLISHED;
4107                         if ((*state)->src_node != NULL &&
4108                             pf_src_connlimit(state)) {
4109                                 REASON_SET(reason, PFRES_SRCLIMIT);
4110                                 return (PF_DROP);
4111                         }
4112                 } else if (src->state == TCPS_CLOSING &&
4113                     dst->state == TCPS_ESTABLISHED &&
4114                     dst->seqlo == 0) {
4115                         /*
4116                          * Handle the closing of half connections where we
4117                          * don't see the full bidirectional FIN/ACK+ACK
4118                          * handshake.
4119                          */
4120                         dst->state = TCPS_CLOSING;
4121                 }
4122         }
4123         if (th->th_flags & TH_RST)
4124                 src->state = dst->state = TCPS_TIME_WAIT;
4125
4126         /* update expire time */
4127         (*state)->expire = time_uptime;
4128         if (src->state >= TCPS_FIN_WAIT_2 &&
4129             dst->state >= TCPS_FIN_WAIT_2)
4130                 (*state)->timeout = PFTM_TCP_CLOSED;
4131         else if (src->state >= TCPS_CLOSING &&
4132             dst->state >= TCPS_CLOSING)
4133                 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4134         else if (src->state < TCPS_ESTABLISHED ||
4135             dst->state < TCPS_ESTABLISHED)
4136                 (*state)->timeout = PFTM_TCP_OPENING;
4137         else if (src->state >= TCPS_CLOSING ||
4138             dst->state >= TCPS_CLOSING)
4139                 (*state)->timeout = PFTM_TCP_CLOSING;
4140         else
4141                 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4142
4143         return (PF_PASS);
4144 }
4145
4146 static int
4147 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4148     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4149     u_short *reason)
4150 {
4151         struct pf_state_key_cmp  key;
4152         struct tcphdr           *th = pd->hdr.tcp;
4153         int                      copyback = 0;
4154         struct pf_state_peer    *src, *dst;
4155         struct pf_state_key     *sk;
4156
4157         bzero(&key, sizeof(key));
4158         key.af = pd->af;
4159         key.proto = IPPROTO_TCP;
4160         if (direction == PF_IN) {       /* wire side, straight */
4161                 PF_ACPY(&key.addr[0], pd->src, key.af);
4162                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4163                 key.port[0] = th->th_sport;
4164                 key.port[1] = th->th_dport;
4165         } else {                        /* stack side, reverse */
4166                 PF_ACPY(&key.addr[1], pd->src, key.af);
4167                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4168                 key.port[1] = th->th_sport;
4169                 key.port[0] = th->th_dport;
4170         }
4171
4172         STATE_LOOKUP(kif, &key, direction, *state, pd);
4173
4174         if (direction == (*state)->direction) {
4175                 src = &(*state)->src;
4176                 dst = &(*state)->dst;
4177         } else {
4178                 src = &(*state)->dst;
4179                 dst = &(*state)->src;
4180         }
4181
4182         sk = (*state)->key[pd->didx];
4183
4184         if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4185                 if (direction != (*state)->direction) {
4186                         REASON_SET(reason, PFRES_SYNPROXY);
4187                         return (PF_SYNPROXY_DROP);
4188                 }
4189                 if (th->th_flags & TH_SYN) {
4190                         if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4191                                 REASON_SET(reason, PFRES_SYNPROXY);
4192                                 return (PF_DROP);
4193                         }
4194                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4195                             pd->src, th->th_dport, th->th_sport,
4196                             (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4197                             TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4198                         REASON_SET(reason, PFRES_SYNPROXY);
4199                         return (PF_SYNPROXY_DROP);
4200                 } else if (!(th->th_flags & TH_ACK) ||
4201                     (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4202                     (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4203                         REASON_SET(reason, PFRES_SYNPROXY);
4204                         return (PF_DROP);
4205                 } else if ((*state)->src_node != NULL &&
4206                     pf_src_connlimit(state)) {
4207                         REASON_SET(reason, PFRES_SRCLIMIT);
4208                         return (PF_DROP);
4209                 } else
4210                         (*state)->src.state = PF_TCPS_PROXY_DST;
4211         }
4212         if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4213                 if (direction == (*state)->direction) {
4214                         if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4215                             (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4216                             (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4217                                 REASON_SET(reason, PFRES_SYNPROXY);
4218                                 return (PF_DROP);
4219                         }
4220                         (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4221                         if ((*state)->dst.seqhi == 1)
4222                                 (*state)->dst.seqhi = htonl(arc4random());
4223                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4224                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4225                             sk->port[pd->sidx], sk->port[pd->didx],
4226                             (*state)->dst.seqhi, 0, TH_SYN, 0,
4227                             (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4228                         REASON_SET(reason, PFRES_SYNPROXY);
4229                         return (PF_SYNPROXY_DROP);
4230                 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4231                     (TH_SYN|TH_ACK)) ||
4232                     (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4233                         REASON_SET(reason, PFRES_SYNPROXY);
4234                         return (PF_DROP);
4235                 } else {
4236                         (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4237                         (*state)->dst.seqlo = ntohl(th->th_seq);
4238                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4239                             pd->src, th->th_dport, th->th_sport,
4240                             ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4241                             TH_ACK, (*state)->src.max_win, 0, 0, 0,
4242                             (*state)->tag, NULL);
4243                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4244                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4245                             sk->port[pd->sidx], sk->port[pd->didx],
4246                             (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4247                             TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4248                         (*state)->src.seqdiff = (*state)->dst.seqhi -
4249                             (*state)->src.seqlo;
4250                         (*state)->dst.seqdiff = (*state)->src.seqhi -
4251                             (*state)->dst.seqlo;
4252                         (*state)->src.seqhi = (*state)->src.seqlo +
4253                             (*state)->dst.max_win;
4254                         (*state)->dst.seqhi = (*state)->dst.seqlo +
4255                             (*state)->src.max_win;
4256                         (*state)->src.wscale = (*state)->dst.wscale = 0;
4257                         (*state)->src.state = (*state)->dst.state =
4258                             TCPS_ESTABLISHED;
4259                         REASON_SET(reason, PFRES_SYNPROXY);
4260                         return (PF_SYNPROXY_DROP);
4261                 }
4262         }
4263
4264         if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4265             dst->state >= TCPS_FIN_WAIT_2 &&
4266             src->state >= TCPS_FIN_WAIT_2) {
4267                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4268                         printf("pf: state reuse ");
4269                         pf_print_state(*state);
4270                         pf_print_flags(th->th_flags);
4271                         printf("\n");
4272                 }
4273                 /* XXX make sure it's the same direction ?? */
4274                 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4275                 pf_unlink_state(*state, PF_ENTER_LOCKED);
4276                 *state = NULL;
4277                 return (PF_DROP);
4278         }
4279
4280         if ((*state)->state_flags & PFSTATE_SLOPPY) {
4281                 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4282                         return (PF_DROP);
4283         } else {
4284                 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4285                     &copyback) == PF_DROP)
4286                         return (PF_DROP);
4287         }
4288
4289         /* translate source/destination address, if necessary */
4290         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4291                 struct pf_state_key *nk = (*state)->key[pd->didx];
4292
4293                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4294                     nk->port[pd->sidx] != th->th_sport)
4295                         pf_change_ap(pd->src, &th->th_sport, pd->ip_sum,
4296                             &th->th_sum, &nk->addr[pd->sidx],
4297                             nk->port[pd->sidx], 0, pd->af);
4298
4299                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4300                     nk->port[pd->didx] != th->th_dport)
4301                         pf_change_ap(pd->dst, &th->th_dport, pd->ip_sum,
4302                             &th->th_sum, &nk->addr[pd->didx],
4303                             nk->port[pd->didx], 0, pd->af);
4304                 copyback = 1;
4305         }
4306
4307         /* Copyback sequence modulation or stateful scrub changes if needed */
4308         if (copyback)
4309                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
4310
4311         return (PF_PASS);
4312 }
4313
4314 static int
4315 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
4316     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4317 {
4318         struct pf_state_peer    *src, *dst;
4319         struct pf_state_key_cmp  key;
4320         struct udphdr           *uh = pd->hdr.udp;
4321
4322         bzero(&key, sizeof(key));
4323         key.af = pd->af;
4324         key.proto = IPPROTO_UDP;
4325         if (direction == PF_IN) {       /* wire side, straight */
4326                 PF_ACPY(&key.addr[0], pd->src, key.af);
4327                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4328                 key.port[0] = uh->uh_sport;
4329                 key.port[1] = uh->uh_dport;
4330         } else {                        /* stack side, reverse */
4331                 PF_ACPY(&key.addr[1], pd->src, key.af);
4332                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4333                 key.port[1] = uh->uh_sport;
4334                 key.port[0] = uh->uh_dport;
4335         }
4336
4337         STATE_LOOKUP(kif, &key, direction, *state, pd);
4338
4339         if (direction == (*state)->direction) {
4340                 src = &(*state)->src;
4341                 dst = &(*state)->dst;
4342         } else {
4343                 src = &(*state)->dst;
4344                 dst = &(*state)->src;
4345         }
4346
4347         /* update states */
4348         if (src->state < PFUDPS_SINGLE)
4349                 src->state = PFUDPS_SINGLE;
4350         if (dst->state == PFUDPS_SINGLE)
4351                 dst->state = PFUDPS_MULTIPLE;
4352
4353         /* update expire time */
4354         (*state)->expire = time_uptime;
4355         if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4356                 (*state)->timeout = PFTM_UDP_MULTIPLE;
4357         else
4358                 (*state)->timeout = PFTM_UDP_SINGLE;
4359
4360         /* translate source/destination address, if necessary */
4361         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4362                 struct pf_state_key *nk = (*state)->key[pd->didx];
4363
4364                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4365                     nk->port[pd->sidx] != uh->uh_sport)
4366                         pf_change_ap(pd->src, &uh->uh_sport, pd->ip_sum,
4367                             &uh->uh_sum, &nk->addr[pd->sidx],
4368                             nk->port[pd->sidx], 1, pd->af);
4369
4370                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4371                     nk->port[pd->didx] != uh->uh_dport)
4372                         pf_change_ap(pd->dst, &uh->uh_dport, pd->ip_sum,
4373                             &uh->uh_sum, &nk->addr[pd->didx],
4374                             nk->port[pd->didx], 1, pd->af);
4375                 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4376         }
4377
4378         return (PF_PASS);
4379 }
4380
4381 static int
4382 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
4383     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4384 {
4385         struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
4386         u_int16_t        icmpid = 0, *icmpsum;
4387         u_int8_t         icmptype;
4388         int              state_icmp = 0;
4389         struct pf_state_key_cmp key;
4390
4391         bzero(&key, sizeof(key));
4392         switch (pd->proto) {
4393 #ifdef INET
4394         case IPPROTO_ICMP:
4395                 icmptype = pd->hdr.icmp->icmp_type;
4396                 icmpid = pd->hdr.icmp->icmp_id;
4397                 icmpsum = &pd->hdr.icmp->icmp_cksum;
4398
4399                 if (icmptype == ICMP_UNREACH ||
4400                     icmptype == ICMP_SOURCEQUENCH ||
4401                     icmptype == ICMP_REDIRECT ||
4402                     icmptype == ICMP_TIMXCEED ||
4403                     icmptype == ICMP_PARAMPROB)
4404                         state_icmp++;
4405                 break;
4406 #endif /* INET */
4407 #ifdef INET6
4408         case IPPROTO_ICMPV6:
4409                 icmptype = pd->hdr.icmp6->icmp6_type;
4410                 icmpid = pd->hdr.icmp6->icmp6_id;
4411                 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4412
4413                 if (icmptype == ICMP6_DST_UNREACH ||
4414                     icmptype == ICMP6_PACKET_TOO_BIG ||
4415                     icmptype == ICMP6_TIME_EXCEEDED ||
4416                     icmptype == ICMP6_PARAM_PROB)
4417                         state_icmp++;
4418                 break;
4419 #endif /* INET6 */
4420         }
4421
4422         if (!state_icmp) {
4423
4424                 /*
4425                  * ICMP query/reply message not related to a TCP/UDP packet.
4426                  * Search for an ICMP state.
4427                  */
4428                 key.af = pd->af;
4429                 key.proto = pd->proto;
4430                 key.port[0] = key.port[1] = icmpid;
4431                 if (direction == PF_IN) {       /* wire side, straight */
4432                         PF_ACPY(&key.addr[0], pd->src, key.af);
4433                         PF_ACPY(&key.addr[1], pd->dst, key.af);
4434                 } else {                        /* stack side, reverse */
4435                         PF_ACPY(&key.addr[1], pd->src, key.af);
4436                         PF_ACPY(&key.addr[0], pd->dst, key.af);
4437                 }
4438
4439                 STATE_LOOKUP(kif, &key, direction, *state, pd);
4440
4441                 (*state)->expire = time_uptime;
4442                 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4443
4444                 /* translate source/destination address, if necessary */
4445                 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4446                         struct pf_state_key *nk = (*state)->key[pd->didx];
4447
4448                         switch (pd->af) {
4449 #ifdef INET
4450                         case AF_INET:
4451                                 if (PF_ANEQ(pd->src,
4452                                     &nk->addr[pd->sidx], AF_INET))
4453                                         pf_change_a(&saddr->v4.s_addr,
4454                                             pd->ip_sum,
4455                                             nk->addr[pd->sidx].v4.s_addr, 0);
4456
4457                                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4458                                     AF_INET))
4459                                         pf_change_a(&daddr->v4.s_addr,
4460                                             pd->ip_sum,
4461                                             nk->addr[pd->didx].v4.s_addr, 0);
4462
4463                                 if (nk->port[0] !=
4464                                     pd->hdr.icmp->icmp_id) {
4465                                         pd->hdr.icmp->icmp_cksum =
4466                                             pf_cksum_fixup(
4467                                             pd->hdr.icmp->icmp_cksum, icmpid,
4468                                             nk->port[pd->sidx], 0);
4469                                         pd->hdr.icmp->icmp_id =
4470                                             nk->port[pd->sidx];
4471                                 }
4472
4473                                 m_copyback(m, off, ICMP_MINLEN,
4474                                     (caddr_t )pd->hdr.icmp);
4475                                 break;
4476 #endif /* INET */
4477 #ifdef INET6
4478                         case AF_INET6:
4479                                 if (PF_ANEQ(pd->src,
4480                                     &nk->addr[pd->sidx], AF_INET6))
4481                                         pf_change_a6(saddr,
4482                                             &pd->hdr.icmp6->icmp6_cksum,
4483                                             &nk->addr[pd->sidx], 0);
4484
4485                                 if (PF_ANEQ(pd->dst,
4486                                     &nk->addr[pd->didx], AF_INET6))
4487                                         pf_change_a6(daddr,
4488                                             &pd->hdr.icmp6->icmp6_cksum,
4489                                             &nk->addr[pd->didx], 0);
4490
4491                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4492                                     (caddr_t )pd->hdr.icmp6);
4493                                 break;
4494 #endif /* INET6 */
4495                         }
4496                 }
4497                 return (PF_PASS);
4498
4499         } else {
4500                 /*
4501                  * ICMP error message in response to a TCP/UDP packet.
4502                  * Extract the inner TCP/UDP header and search for that state.
4503                  */
4504
4505                 struct pf_pdesc pd2;
4506                 bzero(&pd2, sizeof pd2);
4507 #ifdef INET
4508                 struct ip       h2;
4509 #endif /* INET */
4510 #ifdef INET6
4511                 struct ip6_hdr  h2_6;
4512                 int             terminal = 0;
4513 #endif /* INET6 */
4514                 int             ipoff2 = 0;
4515                 int             off2 = 0;
4516
4517                 pd2.af = pd->af;
4518                 /* Payload packet is from the opposite direction. */
4519                 pd2.sidx = (direction == PF_IN) ? 1 : 0;
4520                 pd2.didx = (direction == PF_IN) ? 0 : 1;
4521                 switch (pd->af) {
4522 #ifdef INET
4523                 case AF_INET:
4524                         /* offset of h2 in mbuf chain */
4525                         ipoff2 = off + ICMP_MINLEN;
4526
4527                         if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4528                             NULL, reason, pd2.af)) {
4529                                 DPFPRINTF(PF_DEBUG_MISC,
4530                                     ("pf: ICMP error message too short "
4531                                     "(ip)\n"));
4532                                 return (PF_DROP);
4533                         }
4534                         /*
4535                          * ICMP error messages don't refer to non-first
4536                          * fragments
4537                          */
4538                         if (h2.ip_off & htons(IP_OFFMASK)) {
4539                                 REASON_SET(reason, PFRES_FRAG);
4540                                 return (PF_DROP);
4541                         }
4542
4543                         /* offset of protocol header that follows h2 */
4544                         off2 = ipoff2 + (h2.ip_hl << 2);
4545
4546                         pd2.proto = h2.ip_p;
4547                         pd2.src = (struct pf_addr *)&h2.ip_src;
4548                         pd2.dst = (struct pf_addr *)&h2.ip_dst;
4549                         pd2.ip_sum = &h2.ip_sum;
4550                         break;
4551 #endif /* INET */
4552 #ifdef INET6
4553                 case AF_INET6:
4554                         ipoff2 = off + sizeof(struct icmp6_hdr);
4555
4556                         if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4557                             NULL, reason, pd2.af)) {
4558                                 DPFPRINTF(PF_DEBUG_MISC,
4559                                     ("pf: ICMP error message too short "
4560                                     "(ip6)\n"));
4561                                 return (PF_DROP);
4562                         }
4563                         pd2.proto = h2_6.ip6_nxt;
4564                         pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4565                         pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4566                         pd2.ip_sum = NULL;
4567                         off2 = ipoff2 + sizeof(h2_6);
4568                         do {
4569                                 switch (pd2.proto) {
4570                                 case IPPROTO_FRAGMENT:
4571                                         /*
4572                                          * ICMPv6 error messages for
4573                                          * non-first fragments
4574                                          */
4575                                         REASON_SET(reason, PFRES_FRAG);
4576                                         return (PF_DROP);
4577                                 case IPPROTO_AH:
4578                                 case IPPROTO_HOPOPTS:
4579                                 case IPPROTO_ROUTING:
4580                                 case IPPROTO_DSTOPTS: {
4581                                         /* get next header and header length */
4582                                         struct ip6_ext opt6;
4583
4584                                         if (!pf_pull_hdr(m, off2, &opt6,
4585                                             sizeof(opt6), NULL, reason,
4586                                             pd2.af)) {
4587                                                 DPFPRINTF(PF_DEBUG_MISC,
4588                                                     ("pf: ICMPv6 short opt\n"));
4589                                                 return (PF_DROP);
4590                                         }
4591                                         if (pd2.proto == IPPROTO_AH)
4592                                                 off2 += (opt6.ip6e_len + 2) * 4;
4593                                         else
4594                                                 off2 += (opt6.ip6e_len + 1) * 8;
4595                                         pd2.proto = opt6.ip6e_nxt;
4596                                         /* goto the next header */
4597                                         break;
4598                                 }
4599                                 default:
4600                                         terminal++;
4601                                         break;
4602                                 }
4603                         } while (!terminal);
4604                         break;
4605 #endif /* INET6 */
4606                 }
4607
4608                 switch (pd2.proto) {
4609                 case IPPROTO_TCP: {
4610                         struct tcphdr            th;
4611                         u_int32_t                seq;
4612                         struct pf_state_peer    *src, *dst;
4613                         u_int8_t                 dws;
4614                         int                      copyback = 0;
4615
4616                         /*
4617                          * Only the first 8 bytes of the TCP header can be
4618                          * expected. Don't access any TCP header fields after
4619                          * th_seq, an ackskew test is not possible.
4620                          */
4621                         if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4622                             pd2.af)) {
4623                                 DPFPRINTF(PF_DEBUG_MISC,
4624                                     ("pf: ICMP error message too short "
4625                                     "(tcp)\n"));
4626                                 return (PF_DROP);
4627                         }
4628
4629                         key.af = pd2.af;
4630                         key.proto = IPPROTO_TCP;
4631                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4632                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4633                         key.port[pd2.sidx] = th.th_sport;
4634                         key.port[pd2.didx] = th.th_dport;
4635
4636                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4637
4638                         if (direction == (*state)->direction) {
4639                                 src = &(*state)->dst;
4640                                 dst = &(*state)->src;
4641                         } else {
4642                                 src = &(*state)->src;
4643                                 dst = &(*state)->dst;
4644                         }
4645
4646                         if (src->wscale && dst->wscale)
4647                                 dws = dst->wscale & PF_WSCALE_MASK;
4648                         else
4649                                 dws = 0;
4650
4651                         /* Demodulate sequence number */
4652                         seq = ntohl(th.th_seq) - src->seqdiff;
4653                         if (src->seqdiff) {
4654                                 pf_change_a(&th.th_seq, icmpsum,
4655                                     htonl(seq), 0);
4656                                 copyback = 1;
4657                         }
4658
4659                         if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4660                             (!SEQ_GEQ(src->seqhi, seq) ||
4661                             !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4662                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4663                                         printf("pf: BAD ICMP %d:%d ",
4664                                             icmptype, pd->hdr.icmp->icmp_code);
4665                                         pf_print_host(pd->src, 0, pd->af);
4666                                         printf(" -> ");
4667                                         pf_print_host(pd->dst, 0, pd->af);
4668                                         printf(" state: ");
4669                                         pf_print_state(*state);
4670                                         printf(" seq=%u\n", seq);
4671                                 }
4672                                 REASON_SET(reason, PFRES_BADSTATE);
4673                                 return (PF_DROP);
4674                         } else {
4675                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4676                                         printf("pf: OK ICMP %d:%d ",
4677                                             icmptype, pd->hdr.icmp->icmp_code);
4678                                         pf_print_host(pd->src, 0, pd->af);
4679                                         printf(" -> ");
4680                                         pf_print_host(pd->dst, 0, pd->af);
4681                                         printf(" state: ");
4682                                         pf_print_state(*state);
4683                                         printf(" seq=%u\n", seq);
4684                                 }
4685                         }
4686
4687                         /* translate source/destination address, if necessary */
4688                         if ((*state)->key[PF_SK_WIRE] !=
4689                             (*state)->key[PF_SK_STACK]) {
4690                                 struct pf_state_key *nk =
4691                                     (*state)->key[pd->didx];
4692
4693                                 if (PF_ANEQ(pd2.src,
4694                                     &nk->addr[pd2.sidx], pd2.af) ||
4695                                     nk->port[pd2.sidx] != th.th_sport)
4696                                         pf_change_icmp(pd2.src, &th.th_sport,
4697                                             daddr, &nk->addr[pd2.sidx],
4698                                             nk->port[pd2.sidx], NULL,
4699                                             pd2.ip_sum, icmpsum,
4700                                             pd->ip_sum, 0, pd2.af);
4701
4702                                 if (PF_ANEQ(pd2.dst,
4703                                     &nk->addr[pd2.didx], pd2.af) ||
4704                                     nk->port[pd2.didx] != th.th_dport)
4705                                         pf_change_icmp(pd2.dst, &th.th_dport,
4706                                             NULL, /* XXX Inbound NAT? */
4707                                             &nk->addr[pd2.didx],
4708                                             nk->port[pd2.didx], NULL,
4709                                             pd2.ip_sum, icmpsum,
4710                                             pd->ip_sum, 0, pd2.af);
4711                                 copyback = 1;
4712                         }
4713
4714                         if (copyback) {
4715                                 switch (pd2.af) {
4716 #ifdef INET
4717                                 case AF_INET:
4718                                         m_copyback(m, off, ICMP_MINLEN,
4719                                             (caddr_t )pd->hdr.icmp);
4720                                         m_copyback(m, ipoff2, sizeof(h2),
4721                                             (caddr_t )&h2);
4722                                         break;
4723 #endif /* INET */
4724 #ifdef INET6
4725                                 case AF_INET6:
4726                                         m_copyback(m, off,
4727                                             sizeof(struct icmp6_hdr),
4728                                             (caddr_t )pd->hdr.icmp6);
4729                                         m_copyback(m, ipoff2, sizeof(h2_6),
4730                                             (caddr_t )&h2_6);
4731                                         break;
4732 #endif /* INET6 */
4733                                 }
4734                                 m_copyback(m, off2, 8, (caddr_t)&th);
4735                         }
4736
4737                         return (PF_PASS);
4738                         break;
4739                 }
4740                 case IPPROTO_UDP: {
4741                         struct udphdr           uh;
4742
4743                         if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
4744                             NULL, reason, pd2.af)) {
4745                                 DPFPRINTF(PF_DEBUG_MISC,
4746                                     ("pf: ICMP error message too short "
4747                                     "(udp)\n"));
4748                                 return (PF_DROP);
4749                         }
4750
4751                         key.af = pd2.af;
4752                         key.proto = IPPROTO_UDP;
4753                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4754                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4755                         key.port[pd2.sidx] = uh.uh_sport;
4756                         key.port[pd2.didx] = uh.uh_dport;
4757
4758                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4759
4760                         /* translate source/destination address, if necessary */
4761                         if ((*state)->key[PF_SK_WIRE] !=
4762                             (*state)->key[PF_SK_STACK]) {
4763                                 struct pf_state_key *nk =
4764                                     (*state)->key[pd->didx];
4765
4766                                 if (PF_ANEQ(pd2.src,
4767                                     &nk->addr[pd2.sidx], pd2.af) ||
4768                                     nk->port[pd2.sidx] != uh.uh_sport)
4769                                         pf_change_icmp(pd2.src, &uh.uh_sport,
4770                                             daddr, &nk->addr[pd2.sidx],
4771                                             nk->port[pd2.sidx], &uh.uh_sum,
4772                                             pd2.ip_sum, icmpsum,
4773                                             pd->ip_sum, 1, pd2.af);
4774
4775                                 if (PF_ANEQ(pd2.dst,
4776                                     &nk->addr[pd2.didx], pd2.af) ||
4777                                     nk->port[pd2.didx] != uh.uh_dport)
4778                                         pf_change_icmp(pd2.dst, &uh.uh_dport,
4779                                             NULL, /* XXX Inbound NAT? */
4780                                             &nk->addr[pd2.didx],
4781                                             nk->port[pd2.didx], &uh.uh_sum,
4782                                             pd2.ip_sum, icmpsum,
4783                                             pd->ip_sum, 1, pd2.af);
4784
4785                                 switch (pd2.af) {
4786 #ifdef INET
4787                                 case AF_INET:
4788                                         m_copyback(m, off, ICMP_MINLEN,
4789                                             (caddr_t )pd->hdr.icmp);
4790                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4791                                         break;
4792 #endif /* INET */
4793 #ifdef INET6
4794                                 case AF_INET6:
4795                                         m_copyback(m, off,
4796                                             sizeof(struct icmp6_hdr),
4797                                             (caddr_t )pd->hdr.icmp6);
4798                                         m_copyback(m, ipoff2, sizeof(h2_6),
4799                                             (caddr_t )&h2_6);
4800                                         break;
4801 #endif /* INET6 */
4802                                 }
4803                                 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
4804                         }
4805                         return (PF_PASS);
4806                         break;
4807                 }
4808 #ifdef INET
4809                 case IPPROTO_ICMP: {
4810                         struct icmp             iih;
4811
4812                         if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
4813                             NULL, reason, pd2.af)) {
4814                                 DPFPRINTF(PF_DEBUG_MISC,
4815                                     ("pf: ICMP error message too short i"
4816                                     "(icmp)\n"));
4817                                 return (PF_DROP);
4818                         }
4819
4820                         key.af = pd2.af;
4821                         key.proto = IPPROTO_ICMP;
4822                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4823                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4824                         key.port[0] = key.port[1] = iih.icmp_id;
4825
4826                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4827
4828                         /* translate source/destination address, if necessary */
4829                         if ((*state)->key[PF_SK_WIRE] !=
4830                             (*state)->key[PF_SK_STACK]) {
4831                                 struct pf_state_key *nk =
4832                                     (*state)->key[pd->didx];
4833
4834                                 if (PF_ANEQ(pd2.src,
4835                                     &nk->addr[pd2.sidx], pd2.af) ||
4836                                     nk->port[pd2.sidx] != iih.icmp_id)
4837                                         pf_change_icmp(pd2.src, &iih.icmp_id,
4838                                             daddr, &nk->addr[pd2.sidx],
4839                                             nk->port[pd2.sidx], NULL,
4840                                             pd2.ip_sum, icmpsum,
4841                                             pd->ip_sum, 0, AF_INET);
4842
4843                                 if (PF_ANEQ(pd2.dst,
4844                                     &nk->addr[pd2.didx], pd2.af) ||
4845                                     nk->port[pd2.didx] != iih.icmp_id)
4846                                         pf_change_icmp(pd2.dst, &iih.icmp_id,
4847                                             NULL, /* XXX Inbound NAT? */
4848                                             &nk->addr[pd2.didx],
4849                                             nk->port[pd2.didx], NULL,
4850                                             pd2.ip_sum, icmpsum,
4851                                             pd->ip_sum, 0, AF_INET);
4852
4853                                 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
4854                                 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4855                                 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
4856                         }
4857                         return (PF_PASS);
4858                         break;
4859                 }
4860 #endif /* INET */
4861 #ifdef INET6
4862                 case IPPROTO_ICMPV6: {
4863                         struct icmp6_hdr        iih;
4864
4865                         if (!pf_pull_hdr(m, off2, &iih,
4866                             sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
4867                                 DPFPRINTF(PF_DEBUG_MISC,
4868                                     ("pf: ICMP error message too short "
4869                                     "(icmp6)\n"));
4870                                 return (PF_DROP);
4871                         }
4872
4873                         key.af = pd2.af;
4874                         key.proto = IPPROTO_ICMPV6;
4875                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4876                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4877                         key.port[0] = key.port[1] = iih.icmp6_id;
4878
4879                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4880
4881                         /* translate source/destination address, if necessary */
4882                         if ((*state)->key[PF_SK_WIRE] !=
4883                             (*state)->key[PF_SK_STACK]) {
4884                                 struct pf_state_key *nk =
4885                                     (*state)->key[pd->didx];
4886
4887                                 if (PF_ANEQ(pd2.src,
4888                                     &nk->addr[pd2.sidx], pd2.af) ||
4889                                     nk->port[pd2.sidx] != iih.icmp6_id)
4890                                         pf_change_icmp(pd2.src, &iih.icmp6_id,
4891                                             daddr, &nk->addr[pd2.sidx],
4892                                             nk->port[pd2.sidx], NULL,
4893                                             pd2.ip_sum, icmpsum,
4894                                             pd->ip_sum, 0, AF_INET6);
4895
4896                                 if (PF_ANEQ(pd2.dst,
4897                                     &nk->addr[pd2.didx], pd2.af) ||
4898                                     nk->port[pd2.didx] != iih.icmp6_id)
4899                                         pf_change_icmp(pd2.dst, &iih.icmp6_id,
4900                                             NULL, /* XXX Inbound NAT? */
4901                                             &nk->addr[pd2.didx],
4902                                             nk->port[pd2.didx], NULL,
4903                                             pd2.ip_sum, icmpsum,
4904                                             pd->ip_sum, 0, AF_INET6);
4905
4906                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4907                                     (caddr_t)pd->hdr.icmp6);
4908                                 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
4909                                 m_copyback(m, off2, sizeof(struct icmp6_hdr),
4910                                     (caddr_t)&iih);
4911                         }
4912                         return (PF_PASS);
4913                         break;
4914                 }
4915 #endif /* INET6 */
4916                 default: {
4917                         key.af = pd2.af;
4918                         key.proto = pd2.proto;
4919                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4920                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4921                         key.port[0] = key.port[1] = 0;
4922
4923                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4924
4925                         /* translate source/destination address, if necessary */
4926                         if ((*state)->key[PF_SK_WIRE] !=
4927                             (*state)->key[PF_SK_STACK]) {
4928                                 struct pf_state_key *nk =
4929                                     (*state)->key[pd->didx];
4930
4931                                 if (PF_ANEQ(pd2.src,
4932                                     &nk->addr[pd2.sidx], pd2.af))
4933                                         pf_change_icmp(pd2.src, NULL, daddr,
4934                                             &nk->addr[pd2.sidx], 0, NULL,
4935                                             pd2.ip_sum, icmpsum,
4936                                             pd->ip_sum, 0, pd2.af);
4937
4938                                 if (PF_ANEQ(pd2.dst,
4939                                     &nk->addr[pd2.didx], pd2.af))
4940                                         pf_change_icmp(pd2.src, NULL,
4941                                             NULL, /* XXX Inbound NAT? */
4942                                             &nk->addr[pd2.didx], 0, NULL,
4943                                             pd2.ip_sum, icmpsum,
4944                                             pd->ip_sum, 0, pd2.af);
4945
4946                                 switch (pd2.af) {
4947 #ifdef INET
4948                                 case AF_INET:
4949                                         m_copyback(m, off, ICMP_MINLEN,
4950                                             (caddr_t)pd->hdr.icmp);
4951                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4952                                         break;
4953 #endif /* INET */
4954 #ifdef INET6
4955                                 case AF_INET6:
4956                                         m_copyback(m, off,
4957                                             sizeof(struct icmp6_hdr),
4958                                             (caddr_t )pd->hdr.icmp6);
4959                                         m_copyback(m, ipoff2, sizeof(h2_6),
4960                                             (caddr_t )&h2_6);
4961                                         break;
4962 #endif /* INET6 */
4963                                 }
4964                         }
4965                         return (PF_PASS);
4966                         break;
4967                 }
4968                 }
4969         }
4970 }
4971
4972 static int
4973 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
4974     struct mbuf *m, struct pf_pdesc *pd)
4975 {
4976         struct pf_state_peer    *src, *dst;
4977         struct pf_state_key_cmp  key;
4978
4979         bzero(&key, sizeof(key));
4980         key.af = pd->af;
4981         key.proto = pd->proto;
4982         if (direction == PF_IN) {
4983                 PF_ACPY(&key.addr[0], pd->src, key.af);
4984                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4985                 key.port[0] = key.port[1] = 0;
4986         } else {
4987                 PF_ACPY(&key.addr[1], pd->src, key.af);
4988                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4989                 key.port[1] = key.port[0] = 0;
4990         }
4991
4992         STATE_LOOKUP(kif, &key, direction, *state, pd);
4993
4994         if (direction == (*state)->direction) {
4995                 src = &(*state)->src;
4996                 dst = &(*state)->dst;
4997         } else {
4998                 src = &(*state)->dst;
4999                 dst = &(*state)->src;
5000         }
5001
5002         /* update states */
5003         if (src->state < PFOTHERS_SINGLE)
5004                 src->state = PFOTHERS_SINGLE;
5005         if (dst->state == PFOTHERS_SINGLE)
5006                 dst->state = PFOTHERS_MULTIPLE;
5007
5008         /* update expire time */
5009         (*state)->expire = time_uptime;
5010         if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5011                 (*state)->timeout = PFTM_OTHER_MULTIPLE;
5012         else
5013                 (*state)->timeout = PFTM_OTHER_SINGLE;
5014
5015         /* translate source/destination address, if necessary */
5016         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5017                 struct pf_state_key *nk = (*state)->key[pd->didx];
5018
5019                 KASSERT(nk, ("%s: nk is null", __func__));
5020                 KASSERT(pd, ("%s: pd is null", __func__));
5021                 KASSERT(pd->src, ("%s: pd->src is null", __func__));
5022                 KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5023                 switch (pd->af) {
5024 #ifdef INET
5025                 case AF_INET:
5026                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5027                                 pf_change_a(&pd->src->v4.s_addr,
5028                                     pd->ip_sum,
5029                                     nk->addr[pd->sidx].v4.s_addr,
5030                                     0);
5031
5032
5033                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5034                                 pf_change_a(&pd->dst->v4.s_addr,
5035                                     pd->ip_sum,
5036                                     nk->addr[pd->didx].v4.s_addr,
5037                                     0);
5038
5039                                 break;
5040 #endif /* INET */
5041 #ifdef INET6
5042                 case AF_INET6:
5043                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5044                                 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5045
5046                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5047                                 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5048 #endif /* INET6 */
5049                 }
5050         }
5051         return (PF_PASS);
5052 }
5053
5054 /*
5055  * ipoff and off are measured from the start of the mbuf chain.
5056  * h must be at "ipoff" on the mbuf chain.
5057  */
5058 void *
5059 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5060     u_short *actionp, u_short *reasonp, sa_family_t af)
5061 {
5062         switch (af) {
5063 #ifdef INET
5064         case AF_INET: {
5065                 struct ip       *h = mtod(m, struct ip *);
5066                 u_int16_t        fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5067
5068                 if (fragoff) {
5069                         if (fragoff >= len)
5070                                 ACTION_SET(actionp, PF_PASS);
5071                         else {
5072                                 ACTION_SET(actionp, PF_DROP);
5073                                 REASON_SET(reasonp, PFRES_FRAG);
5074                         }
5075                         return (NULL);
5076                 }
5077                 if (m->m_pkthdr.len < off + len ||
5078                     ntohs(h->ip_len) < off + len) {
5079                         ACTION_SET(actionp, PF_DROP);
5080                         REASON_SET(reasonp, PFRES_SHORT);
5081                         return (NULL);
5082                 }
5083                 break;
5084         }
5085 #endif /* INET */
5086 #ifdef INET6
5087         case AF_INET6: {
5088                 struct ip6_hdr  *h = mtod(m, struct ip6_hdr *);
5089
5090                 if (m->m_pkthdr.len < off + len ||
5091                     (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5092                     (unsigned)(off + len)) {
5093                         ACTION_SET(actionp, PF_DROP);
5094                         REASON_SET(reasonp, PFRES_SHORT);
5095                         return (NULL);
5096                 }
5097                 break;
5098         }
5099 #endif /* INET6 */
5100         }
5101         m_copydata(m, off, len, p);
5102         return (p);
5103 }
5104
5105 int
5106 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5107     int rtableid)
5108 {
5109 #ifdef RADIX_MPATH
5110         struct radix_node_head  *rnh;
5111 #endif
5112         struct sockaddr_in      *dst;
5113         int                      ret = 1;
5114         int                      check_mpath;
5115 #ifdef INET6
5116         struct sockaddr_in6     *dst6;
5117         struct route_in6         ro;
5118 #else
5119         struct route             ro;
5120 #endif
5121         struct radix_node       *rn;
5122         struct rtentry          *rt;
5123         struct ifnet            *ifp;
5124
5125         check_mpath = 0;
5126 #ifdef RADIX_MPATH
5127         /* XXX: stick to table 0 for now */
5128         rnh = rt_tables_get_rnh(0, af);
5129         if (rnh != NULL && rn_mpath_capable(rnh))
5130                 check_mpath = 1;
5131 #endif
5132         bzero(&ro, sizeof(ro));
5133         switch (af) {
5134         case AF_INET:
5135                 dst = satosin(&ro.ro_dst);
5136                 dst->sin_family = AF_INET;
5137                 dst->sin_len = sizeof(*dst);
5138                 dst->sin_addr = addr->v4;
5139                 break;
5140 #ifdef INET6
5141         case AF_INET6:
5142                 /*
5143                  * Skip check for addresses with embedded interface scope,
5144                  * as they would always match anyway.
5145                  */
5146                 if (IN6_IS_SCOPE_EMBED(&addr->v6))
5147                         goto out;
5148                 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5149                 dst6->sin6_family = AF_INET6;
5150                 dst6->sin6_len = sizeof(*dst6);
5151                 dst6->sin6_addr = addr->v6;
5152                 break;
5153 #endif /* INET6 */
5154         default:
5155                 return (0);
5156         }
5157
5158         /* Skip checks for ipsec interfaces */
5159         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5160                 goto out;
5161
5162         switch (af) {
5163 #ifdef INET6
5164         case AF_INET6:
5165                 in6_rtalloc_ign(&ro, 0, rtableid);
5166                 break;
5167 #endif
5168 #ifdef INET
5169         case AF_INET:
5170                 in_rtalloc_ign((struct route *)&ro, 0, rtableid);
5171                 break;
5172 #endif
5173         default:
5174                 rtalloc_ign((struct route *)&ro, 0);    /* No/default FIB. */
5175                 break;
5176         }
5177
5178         if (ro.ro_rt != NULL) {
5179                 /* No interface given, this is a no-route check */
5180                 if (kif == NULL)
5181                         goto out;
5182
5183                 if (kif->pfik_ifp == NULL) {
5184                         ret = 0;
5185                         goto out;
5186                 }
5187
5188                 /* Perform uRPF check if passed input interface */
5189                 ret = 0;
5190                 rn = (struct radix_node *)ro.ro_rt;
5191                 do {
5192                         rt = (struct rtentry *)rn;
5193                         ifp = rt->rt_ifp;
5194
5195                         if (kif->pfik_ifp == ifp)
5196                                 ret = 1;
5197 #ifdef RADIX_MPATH
5198                         rn = rn_mpath_next(rn);
5199 #endif
5200                 } while (check_mpath == 1 && rn != NULL && ret == 0);
5201         } else
5202                 ret = 0;
5203 out:
5204         if (ro.ro_rt != NULL)
5205                 RTFREE(ro.ro_rt);
5206         return (ret);
5207 }
5208
5209 #ifdef INET
5210 static void
5211 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5212     struct pf_state *s, struct pf_pdesc *pd)
5213 {
5214         struct mbuf             *m0, *m1;
5215         struct sockaddr_in      dst;
5216         struct ip               *ip;
5217         struct ifnet            *ifp = NULL;
5218         struct pf_addr           naddr;
5219         struct pf_src_node      *sn = NULL;
5220         int                      error = 0;
5221         uint16_t                 ip_len, ip_off;
5222
5223         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5224         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5225             __func__));
5226
5227         if ((pd->pf_mtag == NULL &&
5228             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5229             pd->pf_mtag->routed++ > 3) {
5230                 m0 = *m;
5231                 *m = NULL;
5232                 goto bad_locked;
5233         }
5234
5235         if (r->rt == PF_DUPTO) {
5236                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5237                         if (s)
5238                                 PF_STATE_UNLOCK(s);
5239                         return;
5240                 }
5241         } else {
5242                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5243                         if (s)
5244                                 PF_STATE_UNLOCK(s);
5245                         return;
5246                 }
5247                 m0 = *m;
5248         }
5249
5250         ip = mtod(m0, struct ip *);
5251
5252         bzero(&dst, sizeof(dst));
5253         dst.sin_family = AF_INET;
5254         dst.sin_len = sizeof(dst);
5255         dst.sin_addr = ip->ip_dst;
5256
5257         if (r->rt == PF_FASTROUTE) {
5258                 struct rtentry *rt;
5259
5260                 if (s)
5261                         PF_STATE_UNLOCK(s);
5262                 rt = rtalloc1_fib(sintosa(&dst), 0, 0, M_GETFIB(m0));
5263                 if (rt == NULL) {
5264                         KMOD_IPSTAT_INC(ips_noroute);
5265                         error = EHOSTUNREACH;
5266                         goto bad;
5267                 }
5268
5269                 ifp = rt->rt_ifp;
5270                 rt->rt_rmx.rmx_pksent++;
5271
5272                 if (rt->rt_flags & RTF_GATEWAY)
5273                         bcopy(satosin(rt->rt_gateway), &dst, sizeof(dst));
5274                 RTFREE_LOCKED(rt);
5275         } else {
5276                 if (TAILQ_EMPTY(&r->rpool.list)) {
5277                         DPFPRINTF(PF_DEBUG_URGENT,
5278                             ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5279                         goto bad_locked;
5280                 }
5281                 if (s == NULL) {
5282                         pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5283                             &naddr, NULL, &sn);
5284                         if (!PF_AZERO(&naddr, AF_INET))
5285                                 dst.sin_addr.s_addr = naddr.v4.s_addr;
5286                         ifp = r->rpool.cur->kif ?
5287                             r->rpool.cur->kif->pfik_ifp : NULL;
5288                 } else {
5289                         if (!PF_AZERO(&s->rt_addr, AF_INET))
5290                                 dst.sin_addr.s_addr =
5291                                     s->rt_addr.v4.s_addr;
5292                         ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5293                         PF_STATE_UNLOCK(s);
5294                 }
5295         }
5296         if (ifp == NULL)
5297                 goto bad;
5298
5299         if (oifp != ifp) {
5300                 if (pf_test(PF_OUT, ifp, &m0, NULL) != PF_PASS)
5301                         goto bad;
5302                 else if (m0 == NULL)
5303                         goto done;
5304                 if (m0->m_len < sizeof(struct ip)) {
5305                         DPFPRINTF(PF_DEBUG_URGENT,
5306                             ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5307                         goto bad;
5308                 }
5309                 ip = mtod(m0, struct ip *);
5310         }
5311
5312         if (ifp->if_flags & IFF_LOOPBACK)
5313                 m0->m_flags |= M_SKIP_FIREWALL;
5314
5315         ip_len = ntohs(ip->ip_len);
5316         ip_off = ntohs(ip->ip_off);
5317
5318         /* Copied from FreeBSD 10.0-CURRENT ip_output. */
5319         m0->m_pkthdr.csum_flags |= CSUM_IP;
5320         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5321                 in_delayed_cksum(m0);
5322                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5323         }
5324 #ifdef SCTP
5325         if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5326                 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
5327                 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5328         }
5329 #endif
5330
5331         /*
5332          * If small enough for interface, or the interface will take
5333          * care of the fragmentation for us, we can just send directly.
5334          */
5335         if (ip_len <= ifp->if_mtu ||
5336             (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 ||
5337             ((ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) {
5338                 ip->ip_sum = 0;
5339                 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5340                         ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5341                         m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5342                 }
5343                 m_clrprotoflags(m0);    /* Avoid confusing lower layers. */
5344                 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5345                 goto done;
5346         }
5347
5348         /* Balk when DF bit is set or the interface didn't support TSO. */
5349         if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5350                 error = EMSGSIZE;
5351                 KMOD_IPSTAT_INC(ips_cantfrag);
5352                 if (r->rt != PF_DUPTO) {
5353                         icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5354                             ifp->if_mtu);
5355                         goto done;
5356                 } else
5357                         goto bad;
5358         }
5359
5360         error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5361         if (error)
5362                 goto bad;
5363
5364         for (; m0; m0 = m1) {
5365                 m1 = m0->m_nextpkt;
5366                 m0->m_nextpkt = NULL;
5367                 if (error == 0) {
5368                         m_clrprotoflags(m0);
5369                         error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5370                 } else
5371                         m_freem(m0);
5372         }
5373
5374         if (error == 0)
5375                 KMOD_IPSTAT_INC(ips_fragmented);
5376
5377 done:
5378         if (r->rt != PF_DUPTO)
5379                 *m = NULL;
5380         return;
5381
5382 bad_locked:
5383         if (s)
5384                 PF_STATE_UNLOCK(s);
5385 bad:
5386         m_freem(m0);
5387         goto done;
5388 }
5389 #endif /* INET */
5390
5391 #ifdef INET6
5392 static void
5393 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5394     struct pf_state *s, struct pf_pdesc *pd)
5395 {
5396         struct mbuf             *m0;
5397         struct sockaddr_in6     dst;
5398         struct ip6_hdr          *ip6;
5399         struct ifnet            *ifp = NULL;
5400         struct pf_addr           naddr;
5401         struct pf_src_node      *sn = NULL;
5402
5403         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5404         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5405             __func__));
5406
5407         if ((pd->pf_mtag == NULL &&
5408             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5409             pd->pf_mtag->routed++ > 3) {
5410                 m0 = *m;
5411                 *m = NULL;
5412                 goto bad_locked;
5413         }
5414
5415         if (r->rt == PF_DUPTO) {
5416                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5417                         if (s)
5418                                 PF_STATE_UNLOCK(s);
5419                         return;
5420                 }
5421         } else {
5422                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5423                         if (s)
5424                                 PF_STATE_UNLOCK(s);
5425                         return;
5426                 }
5427                 m0 = *m;
5428         }
5429
5430         ip6 = mtod(m0, struct ip6_hdr *);
5431
5432         bzero(&dst, sizeof(dst));
5433         dst.sin6_family = AF_INET6;
5434         dst.sin6_len = sizeof(dst);
5435         dst.sin6_addr = ip6->ip6_dst;
5436
5437         /* Cheat. XXX why only in the v6 case??? */
5438         if (r->rt == PF_FASTROUTE) {
5439                 if (s)
5440                         PF_STATE_UNLOCK(s);
5441                 m0->m_flags |= M_SKIP_FIREWALL;
5442                 ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL);
5443                 return;
5444         }
5445
5446         if (TAILQ_EMPTY(&r->rpool.list)) {
5447                 DPFPRINTF(PF_DEBUG_URGENT,
5448                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5449                 goto bad_locked;
5450         }
5451         if (s == NULL) {
5452                 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5453                     &naddr, NULL, &sn);
5454                 if (!PF_AZERO(&naddr, AF_INET6))
5455                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5456                             &naddr, AF_INET6);
5457                 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5458         } else {
5459                 if (!PF_AZERO(&s->rt_addr, AF_INET6))
5460                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5461                             &s->rt_addr, AF_INET6);
5462                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5463         }
5464
5465         if (s)
5466                 PF_STATE_UNLOCK(s);
5467
5468         if (ifp == NULL)
5469                 goto bad;
5470
5471         if (oifp != ifp) {
5472                 if (pf_test6(PF_OUT, ifp, &m0, NULL) != PF_PASS)
5473                         goto bad;
5474                 else if (m0 == NULL)
5475                         goto done;
5476                 if (m0->m_len < sizeof(struct ip6_hdr)) {
5477                         DPFPRINTF(PF_DEBUG_URGENT,
5478                             ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5479                             __func__));
5480                         goto bad;
5481                 }
5482                 ip6 = mtod(m0, struct ip6_hdr *);
5483         }
5484
5485         if (ifp->if_flags & IFF_LOOPBACK)
5486                 m0->m_flags |= M_SKIP_FIREWALL;
5487
5488         /*
5489          * If the packet is too large for the outgoing interface,
5490          * send back an icmp6 error.
5491          */
5492         if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5493                 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5494         if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5495                 nd6_output(ifp, ifp, m0, &dst, NULL);
5496         else {
5497                 in6_ifstat_inc(ifp, ifs6_in_toobig);
5498                 if (r->rt != PF_DUPTO)
5499                         icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5500                 else
5501                         goto bad;
5502         }
5503
5504 done:
5505         if (r->rt != PF_DUPTO)
5506                 *m = NULL;
5507         return;
5508
5509 bad_locked:
5510         if (s)
5511                 PF_STATE_UNLOCK(s);
5512 bad:
5513         m_freem(m0);
5514         goto done;
5515 }
5516 #endif /* INET6 */
5517
5518 /*
5519  * FreeBSD supports cksum offloads for the following drivers.
5520  *  em(4), fxp(4), ixgb(4), lge(4), ndis(4), nge(4), re(4),
5521  *   ti(4), txp(4), xl(4)
5522  *
5523  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5524  *  network driver performed cksum including pseudo header, need to verify
5525  *   csum_data
5526  * CSUM_DATA_VALID :
5527  *  network driver performed cksum, needs to additional pseudo header
5528  *  cksum computation with partial csum_data(i.e. lack of H/W support for
5529  *  pseudo header, for instance hme(4), sk(4) and possibly gem(4))
5530  *
5531  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5532  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5533  * TCP/UDP layer.
5534  * Also, set csum_data to 0xffff to force cksum validation.
5535  */
5536 static int
5537 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5538 {
5539         u_int16_t sum = 0;
5540         int hw_assist = 0;
5541         struct ip *ip;
5542
5543         if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5544                 return (1);
5545         if (m->m_pkthdr.len < off + len)
5546                 return (1);
5547
5548         switch (p) {
5549         case IPPROTO_TCP:
5550                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5551                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5552                                 sum = m->m_pkthdr.csum_data;
5553                         } else {
5554                                 ip = mtod(m, struct ip *);
5555                                 sum = in_pseudo(ip->ip_src.s_addr,
5556                                 ip->ip_dst.s_addr, htonl((u_short)len +
5557                                 m->m_pkthdr.csum_data + IPPROTO_TCP));
5558                         }
5559                         sum ^= 0xffff;
5560                         ++hw_assist;
5561                 }
5562                 break;
5563         case IPPROTO_UDP:
5564                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5565                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5566                                 sum = m->m_pkthdr.csum_data;
5567                         } else {
5568                                 ip = mtod(m, struct ip *);
5569                                 sum = in_pseudo(ip->ip_src.s_addr,
5570                                 ip->ip_dst.s_addr, htonl((u_short)len +
5571                                 m->m_pkthdr.csum_data + IPPROTO_UDP));
5572                         }
5573                         sum ^= 0xffff;
5574                         ++hw_assist;
5575                 }
5576                 break;
5577         case IPPROTO_ICMP:
5578 #ifdef INET6
5579         case IPPROTO_ICMPV6:
5580 #endif /* INET6 */
5581                 break;
5582         default:
5583                 return (1);
5584         }
5585
5586         if (!hw_assist) {
5587                 switch (af) {
5588                 case AF_INET:
5589                         if (p == IPPROTO_ICMP) {
5590                                 if (m->m_len < off)
5591                                         return (1);
5592                                 m->m_data += off;
5593                                 m->m_len -= off;
5594                                 sum = in_cksum(m, len);
5595                                 m->m_data -= off;
5596                                 m->m_len += off;
5597                         } else {
5598                                 if (m->m_len < sizeof(struct ip))
5599                                         return (1);
5600                                 sum = in4_cksum(m, p, off, len);
5601                         }
5602                         break;
5603 #ifdef INET6
5604                 case AF_INET6:
5605                         if (m->m_len < sizeof(struct ip6_hdr))
5606                                 return (1);
5607                         sum = in6_cksum(m, p, off, len);
5608                         break;
5609 #endif /* INET6 */
5610                 default:
5611                         return (1);
5612                 }
5613         }
5614         if (sum) {
5615                 switch (p) {
5616                 case IPPROTO_TCP:
5617                     {
5618                         KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5619                         break;
5620                     }
5621                 case IPPROTO_UDP:
5622                     {
5623                         KMOD_UDPSTAT_INC(udps_badsum);
5624                         break;
5625                     }
5626 #ifdef INET
5627                 case IPPROTO_ICMP:
5628                     {
5629                         KMOD_ICMPSTAT_INC(icps_checksum);
5630                         break;
5631                     }
5632 #endif
5633 #ifdef INET6
5634                 case IPPROTO_ICMPV6:
5635                     {
5636                         KMOD_ICMP6STAT_INC(icp6s_checksum);
5637                         break;
5638                     }
5639 #endif /* INET6 */
5640                 }
5641                 return (1);
5642         } else {
5643                 if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5644                         m->m_pkthdr.csum_flags |=
5645                             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5646                         m->m_pkthdr.csum_data = 0xffff;
5647                 }
5648         }
5649         return (0);
5650 }
5651
5652
5653 #ifdef INET
5654 int
5655 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5656 {
5657         struct pfi_kif          *kif;
5658         u_short                  action, reason = 0, log = 0;
5659         struct mbuf             *m = *m0;
5660         struct ip               *h = NULL;
5661         struct m_tag            *ipfwtag;
5662         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5663         struct pf_state         *s = NULL;
5664         struct pf_ruleset       *ruleset = NULL;
5665         struct pf_pdesc          pd;
5666         int                      off, dirndx, pqid = 0;
5667
5668         M_ASSERTPKTHDR(m);
5669
5670         if (!V_pf_status.running)
5671                 return (PF_PASS);
5672
5673         memset(&pd, 0, sizeof(pd));
5674
5675         kif = (struct pfi_kif *)ifp->if_pf_kif;
5676
5677         if (kif == NULL) {
5678                 DPFPRINTF(PF_DEBUG_URGENT,
5679                     ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5680                 return (PF_DROP);
5681         }
5682         if (kif->pfik_flags & PFI_IFLAG_SKIP)
5683                 return (PF_PASS);
5684
5685         if (m->m_flags & M_SKIP_FIREWALL)
5686                 return (PF_PASS);
5687
5688         pd.pf_mtag = pf_find_mtag(m);
5689
5690         PF_RULES_RLOCK();
5691
5692         if (ip_divert_ptr != NULL &&
5693             ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5694                 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5695                 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5696                         if (pd.pf_mtag == NULL &&
5697                             ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5698                                 action = PF_DROP;
5699                                 goto done;
5700                         }
5701                         pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5702                         m_tag_delete(m, ipfwtag);
5703                 }
5704                 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5705                         m->m_flags |= M_FASTFWD_OURS;
5706                         pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5707                 }
5708         } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5709                 /* We do IP header normalization and packet reassembly here */
5710                 action = PF_DROP;
5711                 goto done;
5712         }
5713         m = *m0;        /* pf_normalize messes with m0 */
5714         h = mtod(m, struct ip *);
5715
5716         off = h->ip_hl << 2;
5717         if (off < (int)sizeof(struct ip)) {
5718                 action = PF_DROP;
5719                 REASON_SET(&reason, PFRES_SHORT);
5720                 log = 1;
5721                 goto done;
5722         }
5723
5724         pd.src = (struct pf_addr *)&h->ip_src;
5725         pd.dst = (struct pf_addr *)&h->ip_dst;
5726         pd.sport = pd.dport = NULL;
5727         pd.ip_sum = &h->ip_sum;
5728         pd.proto_sum = NULL;
5729         pd.proto = h->ip_p;
5730         pd.dir = dir;
5731         pd.sidx = (dir == PF_IN) ? 0 : 1;
5732         pd.didx = (dir == PF_IN) ? 1 : 0;
5733         pd.af = AF_INET;
5734         pd.tos = h->ip_tos;
5735         pd.tot_len = ntohs(h->ip_len);
5736
5737         /* handle fragments that didn't get reassembled by normalization */
5738         if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
5739                 action = pf_test_fragment(&r, dir, kif, m, h,
5740                     &pd, &a, &ruleset);
5741                 goto done;
5742         }
5743
5744         switch (h->ip_p) {
5745
5746         case IPPROTO_TCP: {
5747                 struct tcphdr   th;
5748
5749                 pd.hdr.tcp = &th;
5750                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
5751                     &action, &reason, AF_INET)) {
5752                         log = action != PF_PASS;
5753                         goto done;
5754                 }
5755                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
5756                 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
5757                         pqid = 1;
5758                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
5759                 if (action == PF_DROP)
5760                         goto done;
5761                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
5762                     &reason);
5763                 if (action == PF_PASS) {
5764                         if (pfsync_update_state_ptr != NULL)
5765                                 pfsync_update_state_ptr(s);
5766                         r = s->rule.ptr;
5767                         a = s->anchor.ptr;
5768                         log = s->log;
5769                 } else if (s == NULL)
5770                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5771                             &a, &ruleset, inp);
5772                 break;
5773         }
5774
5775         case IPPROTO_UDP: {
5776                 struct udphdr   uh;
5777
5778                 pd.hdr.udp = &uh;
5779                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
5780                     &action, &reason, AF_INET)) {
5781                         log = action != PF_PASS;
5782                         goto done;
5783                 }
5784                 if (uh.uh_dport == 0 ||
5785                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
5786                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
5787                         action = PF_DROP;
5788                         REASON_SET(&reason, PFRES_SHORT);
5789                         goto done;
5790                 }
5791                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
5792                 if (action == PF_PASS) {
5793                         if (pfsync_update_state_ptr != NULL)
5794                                 pfsync_update_state_ptr(s);
5795                         r = s->rule.ptr;
5796                         a = s->anchor.ptr;
5797                         log = s->log;
5798                 } else if (s == NULL)
5799                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5800                             &a, &ruleset, inp);
5801                 break;
5802         }
5803
5804         case IPPROTO_ICMP: {
5805                 struct icmp     ih;
5806
5807                 pd.hdr.icmp = &ih;
5808                 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
5809                     &action, &reason, AF_INET)) {
5810                         log = action != PF_PASS;
5811                         goto done;
5812                 }
5813                 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
5814                     &reason);
5815                 if (action == PF_PASS) {
5816                         if (pfsync_update_state_ptr != NULL)
5817                                 pfsync_update_state_ptr(s);
5818                         r = s->rule.ptr;
5819                         a = s->anchor.ptr;
5820                         log = s->log;
5821                 } else if (s == NULL)
5822                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5823                             &a, &ruleset, inp);
5824                 break;
5825         }
5826
5827 #ifdef INET6
5828         case IPPROTO_ICMPV6: {
5829                 action = PF_DROP;
5830                 DPFPRINTF(PF_DEBUG_MISC,
5831                     ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
5832                 goto done;
5833         }
5834 #endif
5835
5836         default:
5837                 action = pf_test_state_other(&s, dir, kif, m, &pd);
5838                 if (action == PF_PASS) {
5839                         if (pfsync_update_state_ptr != NULL)
5840                                 pfsync_update_state_ptr(s);
5841                         r = s->rule.ptr;
5842                         a = s->anchor.ptr;
5843                         log = s->log;
5844                 } else if (s == NULL)
5845                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5846                             &a, &ruleset, inp);
5847                 break;
5848         }
5849
5850 done:
5851         PF_RULES_RUNLOCK();
5852         if (action == PF_PASS && h->ip_hl > 5 &&
5853             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
5854                 action = PF_DROP;
5855                 REASON_SET(&reason, PFRES_IPOPTIONS);
5856                 log = 1;
5857                 DPFPRINTF(PF_DEBUG_MISC,
5858                     ("pf: dropping packet with ip options\n"));
5859         }
5860
5861         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
5862                 action = PF_DROP;
5863                 REASON_SET(&reason, PFRES_MEMORY);
5864         }
5865         if (r->rtableid >= 0)
5866                 M_SETFIB(m, r->rtableid);
5867
5868 #ifdef ALTQ
5869         if (action == PF_PASS && r->qid) {
5870                 if (pd.pf_mtag == NULL &&
5871                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5872                         action = PF_DROP;
5873                         REASON_SET(&reason, PFRES_MEMORY);
5874                 }
5875                 if (pqid || (pd.tos & IPTOS_LOWDELAY))
5876                         pd.pf_mtag->qid = r->pqid;
5877                 else
5878                         pd.pf_mtag->qid = r->qid;
5879                 /* add hints for ecn */
5880                 pd.pf_mtag->hdr = h;
5881
5882         }
5883 #endif /* ALTQ */
5884
5885         /*
5886          * connections redirected to loopback should not match sockets
5887          * bound specifically to loopback due to security implications,
5888          * see tcp_input() and in_pcblookup_listen().
5889          */
5890         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
5891             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
5892             (s->nat_rule.ptr->action == PF_RDR ||
5893             s->nat_rule.ptr->action == PF_BINAT) &&
5894             (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
5895                 m->m_flags |= M_SKIP_FIREWALL;
5896
5897         if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
5898             !PACKET_LOOPED(&pd)) {
5899
5900                 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
5901                     sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
5902                 if (ipfwtag != NULL) {
5903                         ((struct ipfw_rule_ref *)(ipfwtag+1))->info =
5904                             ntohs(r->divert.port);
5905                         ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
5906
5907                         if (s)
5908                                 PF_STATE_UNLOCK(s);
5909
5910                         m_tag_prepend(m, ipfwtag);
5911                         if (m->m_flags & M_FASTFWD_OURS) {
5912                                 if (pd.pf_mtag == NULL &&
5913                                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5914                                         action = PF_DROP;
5915                                         REASON_SET(&reason, PFRES_MEMORY);
5916                                         log = 1;
5917                                         DPFPRINTF(PF_DEBUG_MISC,
5918                                             ("pf: failed to allocate tag\n"));
5919                                 }
5920                                 pd.pf_mtag->flags |= PF_FASTFWD_OURS_PRESENT;
5921                                 m->m_flags &= ~M_FASTFWD_OURS;
5922                         }
5923                         ip_divert_ptr(*m0, dir ==  PF_IN ? DIR_IN : DIR_OUT);
5924                         *m0 = NULL;
5925
5926                         return (action);
5927                 } else {
5928                         /* XXX: ipfw has the same behaviour! */
5929                         action = PF_DROP;
5930                         REASON_SET(&reason, PFRES_MEMORY);
5931                         log = 1;
5932                         DPFPRINTF(PF_DEBUG_MISC,
5933                             ("pf: failed to allocate divert tag\n"));
5934                 }
5935         }
5936
5937         if (log) {
5938                 struct pf_rule *lr;
5939
5940                 if (s != NULL && s->nat_rule.ptr != NULL &&
5941                     s->nat_rule.ptr->log & PF_LOG_ALL)
5942                         lr = s->nat_rule.ptr;
5943                 else
5944                         lr = r;
5945                 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
5946                     (s == NULL));
5947         }
5948
5949         kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
5950         kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
5951
5952         if (action == PF_PASS || r->action == PF_DROP) {
5953                 dirndx = (dir == PF_OUT);
5954                 r->packets[dirndx]++;
5955                 r->bytes[dirndx] += pd.tot_len;
5956                 if (a != NULL) {
5957                         a->packets[dirndx]++;
5958                         a->bytes[dirndx] += pd.tot_len;
5959                 }
5960                 if (s != NULL) {
5961                         if (s->nat_rule.ptr != NULL) {
5962                                 s->nat_rule.ptr->packets[dirndx]++;
5963                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
5964                         }
5965                         if (s->src_node != NULL) {
5966                                 s->src_node->packets[dirndx]++;
5967                                 s->src_node->bytes[dirndx] += pd.tot_len;
5968                         }
5969                         if (s->nat_src_node != NULL) {
5970                                 s->nat_src_node->packets[dirndx]++;
5971                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
5972                         }
5973                         dirndx = (dir == s->direction) ? 0 : 1;
5974                         s->packets[dirndx]++;
5975                         s->bytes[dirndx] += pd.tot_len;
5976                 }
5977                 tr = r;
5978                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
5979                 if (nr != NULL && r == &V_pf_default_rule)
5980                         tr = nr;
5981                 if (tr->src.addr.type == PF_ADDR_TABLE)
5982                         pfr_update_stats(tr->src.addr.p.tbl,
5983                             (s == NULL) ? pd.src :
5984                             &s->key[(s->direction == PF_IN)]->
5985                                 addr[(s->direction == PF_OUT)],
5986                             pd.af, pd.tot_len, dir == PF_OUT,
5987                             r->action == PF_PASS, tr->src.neg);
5988                 if (tr->dst.addr.type == PF_ADDR_TABLE)
5989                         pfr_update_stats(tr->dst.addr.p.tbl,
5990                             (s == NULL) ? pd.dst :
5991                             &s->key[(s->direction == PF_IN)]->
5992                                 addr[(s->direction == PF_IN)],
5993                             pd.af, pd.tot_len, dir == PF_OUT,
5994                             r->action == PF_PASS, tr->dst.neg);
5995         }
5996
5997         switch (action) {
5998         case PF_SYNPROXY_DROP:
5999                 m_freem(*m0);
6000         case PF_DEFER:
6001                 *m0 = NULL;
6002                 action = PF_PASS;
6003                 break;
6004         default:
6005                 /* pf_route() returns unlocked. */
6006                 if (r->rt) {
6007                         pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
6008                         return (action);
6009                 }
6010                 break;
6011         }
6012         if (s)
6013                 PF_STATE_UNLOCK(s);
6014
6015         return (action);
6016 }
6017 #endif /* INET */
6018
6019 #ifdef INET6
6020 int
6021 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6022 {
6023         struct pfi_kif          *kif;
6024         u_short                  action, reason = 0, log = 0;
6025         struct mbuf             *m = *m0, *n = NULL;
6026         struct ip6_hdr          *h = NULL;
6027         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6028         struct pf_state         *s = NULL;
6029         struct pf_ruleset       *ruleset = NULL;
6030         struct pf_pdesc          pd;
6031         int                      off, terminal = 0, dirndx, rh_cnt = 0;
6032
6033         M_ASSERTPKTHDR(m);
6034
6035         if (!V_pf_status.running)
6036                 return (PF_PASS);
6037
6038         memset(&pd, 0, sizeof(pd));
6039         pd.pf_mtag = pf_find_mtag(m);
6040
6041         if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6042                 return (PF_PASS);
6043
6044         kif = (struct pfi_kif *)ifp->if_pf_kif;
6045         if (kif == NULL) {
6046                 DPFPRINTF(PF_DEBUG_URGENT,
6047                     ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6048                 return (PF_DROP);
6049         }
6050         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6051                 return (PF_PASS);
6052
6053         PF_RULES_RLOCK();
6054
6055         /* We do IP header normalization and packet reassembly here */
6056         if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6057                 action = PF_DROP;
6058                 goto done;
6059         }
6060         m = *m0;        /* pf_normalize messes with m0 */
6061         h = mtod(m, struct ip6_hdr *);
6062
6063 #if 1
6064         /*
6065          * we do not support jumbogram yet.  if we keep going, zero ip6_plen
6066          * will do something bad, so drop the packet for now.
6067          */
6068         if (htons(h->ip6_plen) == 0) {
6069                 action = PF_DROP;
6070                 REASON_SET(&reason, PFRES_NORM);        /*XXX*/
6071                 goto done;
6072         }
6073 #endif
6074
6075         pd.src = (struct pf_addr *)&h->ip6_src;
6076         pd.dst = (struct pf_addr *)&h->ip6_dst;
6077         pd.sport = pd.dport = NULL;
6078         pd.ip_sum = NULL;
6079         pd.proto_sum = NULL;
6080         pd.dir = dir;
6081         pd.sidx = (dir == PF_IN) ? 0 : 1;
6082         pd.didx = (dir == PF_IN) ? 1 : 0;
6083         pd.af = AF_INET6;
6084         pd.tos = 0;
6085         pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6086
6087         off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6088         pd.proto = h->ip6_nxt;
6089         do {
6090                 switch (pd.proto) {
6091                 case IPPROTO_FRAGMENT:
6092                         action = pf_test_fragment(&r, dir, kif, m, h,
6093                             &pd, &a, &ruleset);
6094                         if (action == PF_DROP)
6095                                 REASON_SET(&reason, PFRES_FRAG);
6096                         goto done;
6097                 case IPPROTO_ROUTING: {
6098                         struct ip6_rthdr rthdr;
6099
6100                         if (rh_cnt++) {
6101                                 DPFPRINTF(PF_DEBUG_MISC,
6102                                     ("pf: IPv6 more than one rthdr\n"));
6103                                 action = PF_DROP;
6104                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6105                                 log = 1;
6106                                 goto done;
6107                         }
6108                         if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6109                             &reason, pd.af)) {
6110                                 DPFPRINTF(PF_DEBUG_MISC,
6111                                     ("pf: IPv6 short rthdr\n"));
6112                                 action = PF_DROP;
6113                                 REASON_SET(&reason, PFRES_SHORT);
6114                                 log = 1;
6115                                 goto done;
6116                         }
6117                         if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6118                                 DPFPRINTF(PF_DEBUG_MISC,
6119                                     ("pf: IPv6 rthdr0\n"));
6120                                 action = PF_DROP;
6121                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6122                                 log = 1;
6123                                 goto done;
6124                         }
6125                         /* FALLTHROUGH */
6126                 }
6127                 case IPPROTO_AH:
6128                 case IPPROTO_HOPOPTS:
6129                 case IPPROTO_DSTOPTS: {
6130                         /* get next header and header length */
6131                         struct ip6_ext  opt6;
6132
6133                         if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6134                             NULL, &reason, pd.af)) {
6135                                 DPFPRINTF(PF_DEBUG_MISC,
6136                                     ("pf: IPv6 short opt\n"));
6137                                 action = PF_DROP;
6138                                 log = 1;
6139                                 goto done;
6140                         }
6141                         if (pd.proto == IPPROTO_AH)
6142                                 off += (opt6.ip6e_len + 2) * 4;
6143                         else
6144                                 off += (opt6.ip6e_len + 1) * 8;
6145                         pd.proto = opt6.ip6e_nxt;
6146                         /* goto the next header */
6147                         break;
6148                 }
6149                 default:
6150                         terminal++;
6151                         break;
6152                 }
6153         } while (!terminal);
6154
6155         /* if there's no routing header, use unmodified mbuf for checksumming */
6156         if (!n)
6157                 n = m;
6158
6159         switch (pd.proto) {
6160
6161         case IPPROTO_TCP: {
6162                 struct tcphdr   th;
6163
6164                 pd.hdr.tcp = &th;
6165                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6166                     &action, &reason, AF_INET6)) {
6167                         log = action != PF_PASS;
6168                         goto done;
6169                 }
6170                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6171                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6172                 if (action == PF_DROP)
6173                         goto done;
6174                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6175                     &reason);
6176                 if (action == PF_PASS) {
6177                         if (pfsync_update_state_ptr != NULL)
6178                                 pfsync_update_state_ptr(s);
6179                         r = s->rule.ptr;
6180                         a = s->anchor.ptr;
6181                         log = s->log;
6182                 } else if (s == NULL)
6183                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6184                             &a, &ruleset, inp);
6185                 break;
6186         }
6187
6188         case IPPROTO_UDP: {
6189                 struct udphdr   uh;
6190
6191                 pd.hdr.udp = &uh;
6192                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6193                     &action, &reason, AF_INET6)) {
6194                         log = action != PF_PASS;
6195                         goto done;
6196                 }
6197                 if (uh.uh_dport == 0 ||
6198                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6199                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6200                         action = PF_DROP;
6201                         REASON_SET(&reason, PFRES_SHORT);
6202                         goto done;
6203                 }
6204                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6205                 if (action == PF_PASS) {
6206                         if (pfsync_update_state_ptr != NULL)
6207                                 pfsync_update_state_ptr(s);
6208                         r = s->rule.ptr;
6209                         a = s->anchor.ptr;
6210                         log = s->log;
6211                 } else if (s == NULL)
6212                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6213                             &a, &ruleset, inp);
6214                 break;
6215         }
6216
6217         case IPPROTO_ICMP: {
6218                 action = PF_DROP;
6219                 DPFPRINTF(PF_DEBUG_MISC,
6220                     ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6221                 goto done;
6222         }
6223
6224         case IPPROTO_ICMPV6: {
6225                 struct icmp6_hdr        ih;
6226
6227                 pd.hdr.icmp6 = &ih;
6228                 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6229                     &action, &reason, AF_INET6)) {
6230                         log = action != PF_PASS;
6231                         goto done;
6232                 }
6233                 action = pf_test_state_icmp(&s, dir, kif,
6234                     m, off, h, &pd, &reason);
6235                 if (action == PF_PASS) {
6236                         if (pfsync_update_state_ptr != NULL)
6237                                 pfsync_update_state_ptr(s);
6238                         r = s->rule.ptr;
6239                         a = s->anchor.ptr;
6240                         log = s->log;
6241                 } else if (s == NULL)
6242                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6243                             &a, &ruleset, inp);
6244                 break;
6245         }
6246
6247         default:
6248                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6249                 if (action == PF_PASS) {
6250                         if (pfsync_update_state_ptr != NULL)
6251                                 pfsync_update_state_ptr(s);
6252                         r = s->rule.ptr;
6253                         a = s->anchor.ptr;
6254                         log = s->log;
6255                 } else if (s == NULL)
6256                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6257                             &a, &ruleset, inp);
6258                 break;
6259         }
6260
6261 done:
6262         PF_RULES_RUNLOCK();
6263         if (n != m) {
6264                 m_freem(n);
6265                 n = NULL;
6266         }
6267
6268         /* handle dangerous IPv6 extension headers. */
6269         if (action == PF_PASS && rh_cnt &&
6270             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6271                 action = PF_DROP;
6272                 REASON_SET(&reason, PFRES_IPOPTIONS);
6273                 log = 1;
6274                 DPFPRINTF(PF_DEBUG_MISC,
6275                     ("pf: dropping packet with dangerous v6 headers\n"));
6276         }
6277
6278         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6279                 action = PF_DROP;
6280                 REASON_SET(&reason, PFRES_MEMORY);
6281         }
6282         if (r->rtableid >= 0)
6283                 M_SETFIB(m, r->rtableid);
6284
6285 #ifdef ALTQ
6286         if (action == PF_PASS && r->qid) {
6287                 if (pd.pf_mtag == NULL &&
6288                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6289                         action = PF_DROP;
6290                         REASON_SET(&reason, PFRES_MEMORY);
6291                 }
6292                 if (pd.tos & IPTOS_LOWDELAY)
6293                         pd.pf_mtag->qid = r->pqid;
6294                 else
6295                         pd.pf_mtag->qid = r->qid;
6296                 /* add hints for ecn */
6297                 pd.pf_mtag->hdr = h;
6298         }
6299 #endif /* ALTQ */
6300
6301         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6302             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6303             (s->nat_rule.ptr->action == PF_RDR ||
6304             s->nat_rule.ptr->action == PF_BINAT) &&
6305             IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6306                 m->m_flags |= M_SKIP_FIREWALL;
6307
6308         /* XXX: Anybody working on it?! */
6309         if (r->divert.port)
6310                 printf("pf: divert(9) is not supported for IPv6\n");
6311
6312         if (log) {
6313                 struct pf_rule *lr;
6314
6315                 if (s != NULL && s->nat_rule.ptr != NULL &&
6316                     s->nat_rule.ptr->log & PF_LOG_ALL)
6317                         lr = s->nat_rule.ptr;
6318                 else
6319                         lr = r;
6320                 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6321                     &pd, (s == NULL));
6322         }
6323
6324         kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6325         kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
6326
6327         if (action == PF_PASS || r->action == PF_DROP) {
6328                 dirndx = (dir == PF_OUT);
6329                 r->packets[dirndx]++;
6330                 r->bytes[dirndx] += pd.tot_len;
6331                 if (a != NULL) {
6332                         a->packets[dirndx]++;
6333                         a->bytes[dirndx] += pd.tot_len;
6334                 }
6335                 if (s != NULL) {
6336                         if (s->nat_rule.ptr != NULL) {
6337                                 s->nat_rule.ptr->packets[dirndx]++;
6338                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6339                         }
6340                         if (s->src_node != NULL) {
6341                                 s->src_node->packets[dirndx]++;
6342                                 s->src_node->bytes[dirndx] += pd.tot_len;
6343                         }
6344                         if (s->nat_src_node != NULL) {
6345                                 s->nat_src_node->packets[dirndx]++;
6346                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6347                         }
6348                         dirndx = (dir == s->direction) ? 0 : 1;
6349                         s->packets[dirndx]++;
6350                         s->bytes[dirndx] += pd.tot_len;
6351                 }
6352                 tr = r;
6353                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6354                 if (nr != NULL && r == &V_pf_default_rule)
6355                         tr = nr;
6356                 if (tr->src.addr.type == PF_ADDR_TABLE)
6357                         pfr_update_stats(tr->src.addr.p.tbl,
6358                             (s == NULL) ? pd.src :
6359                             &s->key[(s->direction == PF_IN)]->addr[0],
6360                             pd.af, pd.tot_len, dir == PF_OUT,
6361                             r->action == PF_PASS, tr->src.neg);
6362                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6363                         pfr_update_stats(tr->dst.addr.p.tbl,
6364                             (s == NULL) ? pd.dst :
6365                             &s->key[(s->direction == PF_IN)]->addr[1],
6366                             pd.af, pd.tot_len, dir == PF_OUT,
6367                             r->action == PF_PASS, tr->dst.neg);
6368         }
6369
6370         switch (action) {
6371         case PF_SYNPROXY_DROP:
6372                 m_freem(*m0);
6373         case PF_DEFER:
6374                 *m0 = NULL;
6375                 action = PF_PASS;
6376                 break;
6377         default:
6378                 /* pf_route6() returns unlocked. */
6379                 if (r->rt) {
6380                         pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
6381                         return (action);
6382                 }
6383                 break;
6384         }
6385
6386         if (s)
6387                 PF_STATE_UNLOCK(s);
6388
6389         return (action);
6390 }
6391 #endif /* INET6 */