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_kstatus,           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 uma_zone_t              pf_mtag_z;
 190 VNET_DEFINE(uma_zone_t,  pf_state_z);
 191 VNET_DEFINE(uma_zone_t,  pf_state_key_z);
 192
 193 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
 194 #define PFID_CPUBITS    8
 195 #define PFID_CPUSHIFT   (sizeof(uint64_t) * NBBY - PFID_CPUBITS)
 196 #define PFID_CPUMASK    ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT)
 197 #define PFID_MAXID      (~PFID_CPUMASK)
 198 CTASSERT((1 << PFID_CPUBITS) > MAXCPU);
 199
 200 static void              pf_src_tree_remove_state(struct pf_state *);
 201 static void              pf_init_threshold(struct pf_threshold *, u_int32_t,
 202                             u_int32_t);
 203 static void              pf_add_threshold(struct pf_threshold *);
 204 static int               pf_check_threshold(struct pf_threshold *);
 205
 206 static void              pf_change_ap(struct pf_addr *, u_int16_t *,
 207                             u_int16_t *, u_int16_t *, struct pf_addr *,
 208                             u_int16_t, u_int8_t, sa_family_t);
 209 static int               pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
 210                             struct tcphdr *, struct pf_state_peer *);
 211 static void              pf_change_icmp(struct pf_addr *, u_int16_t *,
 212                             struct pf_addr *, struct pf_addr *, u_int16_t,
 213                             u_int16_t *, u_int16_t *, u_int16_t *,
 214                             u_int16_t *, u_int8_t, sa_family_t);
 215 static void              pf_send_tcp(struct mbuf *,
 216                             const struct pf_rule *, sa_family_t,
 217                             const struct pf_addr *, const struct pf_addr *,
 218                             u_int16_t, u_int16_t, u_int32_t, u_int32_t,
 219                             u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
 220                             u_int16_t, struct ifnet *);
 221 static void              pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
 222                             sa_family_t, struct pf_rule *);
 223 static void              pf_detach_state(struct pf_state *);
 224 static int               pf_state_key_attach(struct pf_state_key *,
 225                             struct pf_state_key *, struct pf_state *);
 226 static void              pf_state_key_detach(struct pf_state *, int);
 227 static int               pf_state_key_ctor(void *, int, void *, int);
 228 static u_int32_t         pf_tcp_iss(struct pf_pdesc *);
 229 static int               pf_test_rule(struct pf_rule **, struct pf_state **,
 230                             int, struct pfi_kif *, struct mbuf *, int,
 231                             struct pf_pdesc *, struct pf_rule **,
 232                             struct pf_ruleset **, struct inpcb *);
 233 static int               pf_create_state(struct pf_rule *, struct pf_rule *,
 234                             struct pf_rule *, struct pf_pdesc *,
 235                             struct pf_src_node *, struct pf_state_key *,
 236                             struct pf_state_key *, struct mbuf *, int,
 237                             u_int16_t, u_int16_t, int *, struct pfi_kif *,
 238                             struct pf_state **, int, u_int16_t, u_int16_t,
 239                             int);
 240 static int               pf_test_fragment(struct pf_rule **, int,
 241                             struct pfi_kif *, struct mbuf *, void *,
 242                             struct pf_pdesc *, struct pf_rule **,
 243                             struct pf_ruleset **);
 244 static int               pf_tcp_track_full(struct pf_state_peer *,
 245                             struct pf_state_peer *, struct pf_state **,
 246                             struct pfi_kif *, struct mbuf *, int,
 247                             struct pf_pdesc *, u_short *, int *);
 248 static int               pf_tcp_track_sloppy(struct pf_state_peer *,
 249                             struct pf_state_peer *, struct pf_state **,
 250                             struct pf_pdesc *, u_short *);
 251 static int               pf_test_state_tcp(struct pf_state **, int,
 252                             struct pfi_kif *, struct mbuf *, int,
 253                             void *, struct pf_pdesc *, u_short *);
 254 static int               pf_test_state_udp(struct pf_state **, int,
 255                             struct pfi_kif *, struct mbuf *, int,
 256                             void *, struct pf_pdesc *);
 257 static int               pf_test_state_icmp(struct pf_state **, int,
 258                             struct pfi_kif *, struct mbuf *, int,
 259                             void *, struct pf_pdesc *, u_short *);
 260 static int               pf_test_state_other(struct pf_state **, int,
 261                             struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
 262 static u_int8_t          pf_get_wscale(struct mbuf *, int, u_int16_t,
 263                             sa_family_t);
 264 static u_int16_t         pf_get_mss(struct mbuf *, int, u_int16_t,
 265                             sa_family_t);
 266 static u_int16_t         pf_calc_mss(struct pf_addr *, sa_family_t,
 267                                 int, u_int16_t);
 268 static int               pf_check_proto_cksum(struct mbuf *, int, int,
 269                             u_int8_t, sa_family_t);
 270 static void              pf_print_state_parts(struct pf_state *,
 271                             struct pf_state_key *, struct pf_state_key *);
 272 static int               pf_addr_wrap_neq(struct pf_addr_wrap *,
 273                             struct pf_addr_wrap *);
 274 static struct pf_state  *pf_find_state(struct pfi_kif *,
 275                             struct pf_state_key_cmp *, u_int);
 276 static int               pf_src_connlimit(struct pf_state **);
 277 static void              pf_overload_task(void *v, int pending);
 278 static int               pf_insert_src_node(struct pf_src_node **,
 279                             struct pf_rule *, struct pf_addr *, sa_family_t);
 280 static u_int             pf_purge_expired_states(u_int, int);
 281 static void              pf_purge_unlinked_rules(void);
 282 static int               pf_mtag_uminit(void *, int, int);
 283 static void              pf_mtag_free(struct m_tag *);
 284 #ifdef INET
 285 static void              pf_route(struct mbuf **, struct pf_rule *, int,
 286                             struct ifnet *, struct pf_state *,
 287                             struct pf_pdesc *);
 288 #endif /* INET */
 289 #ifdef INET6
 290 static void              pf_change_a6(struct pf_addr *, u_int16_t *,
 291                             struct pf_addr *, u_int8_t);
 292 static void              pf_route6(struct mbuf **, struct pf_rule *, int,
 293                             struct ifnet *, struct pf_state *,
 294                             struct pf_pdesc *);
 295 #endif /* INET6 */
 296
 297 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
 298
 299 VNET_DECLARE(int, pf_end_threads);
 300
 301 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
 302
 303 #define PACKET_LOOPED(pd)       ((pd)->pf_mtag &&                       \
 304                                  (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
 305
 306 #define STATE_LOOKUP(i, k, d, s, pd)                                    \
 307         do {                                                            \
 308                 (s) = pf_find_state((i), (k), (d));                     \
 309                 if ((s) == NULL)                                        \
 310                         return (PF_DROP);                               \
 311                 if (PACKET_LOOPED(pd))                                  \
 312                         return (PF_PASS);                               \
 313                 if ((d) == PF_OUT &&                                    \
 314                     (((s)->rule.ptr->rt == PF_ROUTETO &&                \
 315                     (s)->rule.ptr->direction == PF_OUT) ||              \
 316                     ((s)->rule.ptr->rt == PF_REPLYTO &&                 \
 317                     (s)->rule.ptr->direction == PF_IN)) &&              \
 318                     (s)->rt_kif != NULL &&                              \
 319                     (s)->rt_kif != (i))                                 \
 320                         return (PF_PASS);                               \
 321         } while (0)
 322
 323 #define BOUND_IFACE(r, k) \
 324         ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
 325
 326 #define STATE_INC_COUNTERS(s)                                           \
 327         do {                                                            \
 328                 counter_u64_add(s->rule.ptr->states_cur, 1);            \
 329                 counter_u64_add(s->rule.ptr->states_tot, 1);            \
 330                 if (s->anchor.ptr != NULL) {                            \
 331                         counter_u64_add(s->anchor.ptr->states_cur, 1);  \
 332                         counter_u64_add(s->anchor.ptr->states_tot, 1);  \
 333                 }                                                       \
 334                 if (s->nat_rule.ptr != NULL) {                          \
 335                         counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
 336                         counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
 337                 }                                                       \
 338         } while (0)
 339
 340 #define STATE_DEC_COUNTERS(s)                                           \
 341         do {                                                            \
 342                 if (s->nat_rule.ptr != NULL)                            \
 343                         counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
 344                 if (s->anchor.ptr != NULL)                              \
 345                         counter_u64_add(s->anchor.ptr->states_cur, -1); \
 346                 counter_u64_add(s->rule.ptr->states_cur, -1);           \
 347         } while (0)
 348
 349 static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
 350 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
 351 VNET_DEFINE(struct pf_idhash *, pf_idhash);
 352 VNET_DEFINE(struct pf_srchash *, pf_srchash);
 353
 354 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)");
 355
 356 u_long  pf_hashmask;
 357 u_long  pf_srchashmask;
 358 static u_long   pf_hashsize;
 359 static u_long   pf_srchashsize;
 360
 361 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
 362     &pf_hashsize, 0, "Size of pf(4) states hashtable");
 363 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
 364     &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
 365
 366 VNET_DEFINE(void *, pf_swi_cookie);
 367
 368 VNET_DEFINE(uint32_t, pf_hashseed);
 369 #define V_pf_hashseed   VNET(pf_hashseed)
 370
 371 static __inline uint32_t
 372 pf_hashkey(struct pf_state_key *sk)
 373 {
 374         uint32_t h;
 375
 376         h = jenkins_hash32((uint32_t *)sk,
 377             sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
 378             V_pf_hashseed);
 379
 380         return (h & pf_hashmask);
 381 }
 382
 383 static __inline uint32_t
 384 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
 385 {
 386         uint32_t h;
 387
 388         switch (af) {
 389         case AF_INET:
 390                 h = jenkins_hash32((uint32_t *)&addr->v4,
 391                     sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
 392                 break;
 393         case AF_INET6:
 394                 h = jenkins_hash32((uint32_t *)&addr->v6,
 395                     sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
 396                 break;
 397         default:
 398                 panic("%s: unknown address family %u", __func__, af);
 399         }
 400
 401         return (h & pf_srchashmask);
 402 }
 403
 404 #ifdef INET6
 405 void
 406 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
 407 {
 408         switch (af) {
 409 #ifdef INET
 410         case AF_INET:
 411                 dst->addr32[0] = src->addr32[0];
 412                 break;
 413 #endif /* INET */
 414         case AF_INET6:
 415                 dst->addr32[0] = src->addr32[0];
 416                 dst->addr32[1] = src->addr32[1];
 417                 dst->addr32[2] = src->addr32[2];
 418                 dst->addr32[3] = src->addr32[3];
 419                 break;
 420         }
 421 }
 422 #endif /* INET6 */
 423
 424 static void
 425 pf_init_threshold(struct pf_threshold *threshold,
 426     u_int32_t limit, u_int32_t seconds)
 427 {
 428         threshold->limit = limit * PF_THRESHOLD_MULT;
 429         threshold->seconds = seconds;
 430         threshold->count = 0;
 431         threshold->last = time_uptime;
 432 }
 433
 434 static void
 435 pf_add_threshold(struct pf_threshold *threshold)
 436 {
 437         u_int32_t t = time_uptime, diff = t - threshold->last;
 438
 439         if (diff >= threshold->seconds)
 440                 threshold->count = 0;
 441         else
 442                 threshold->count -= threshold->count * diff /
 443                     threshold->seconds;
 444         threshold->count += PF_THRESHOLD_MULT;
 445         threshold->last = t;
 446 }
 447
 448 static int
 449 pf_check_threshold(struct pf_threshold *threshold)
 450 {
 451         return (threshold->count > threshold->limit);
 452 }
 453
 454 static int
 455 pf_src_connlimit(struct pf_state **state)
 456 {
 457         struct pf_overload_entry *pfoe;
 458         int bad = 0;
 459
 460         PF_STATE_LOCK_ASSERT(*state);
 461
 462         (*state)->src_node->conn++;
 463         (*state)->src.tcp_est = 1;
 464         pf_add_threshold(&(*state)->src_node->conn_rate);
 465
 466         if ((*state)->rule.ptr->max_src_conn &&
 467             (*state)->rule.ptr->max_src_conn <
 468             (*state)->src_node->conn) {
 469                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
 470                 bad++;
 471         }
 472
 473         if ((*state)->rule.ptr->max_src_conn_rate.limit &&
 474             pf_check_threshold(&(*state)->src_node->conn_rate)) {
 475                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
 476                 bad++;
 477         }
 478
 479         if (!bad)
 480                 return (0);
 481
 482         /* Kill this state. */
 483         (*state)->timeout = PFTM_PURGE;
 484         (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
 485
 486         if ((*state)->rule.ptr->overload_tbl == NULL)
 487                 return (1);
 488
 489         /* Schedule overloading and flushing task. */
 490         pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
 491         if (pfoe == NULL)
 492                 return (1);     /* too bad :( */
 493
 494         bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
 495         pfoe->af = (*state)->key[PF_SK_WIRE]->af;
 496         pfoe->rule = (*state)->rule.ptr;
 497         pfoe->dir = (*state)->direction;
 498         PF_OVERLOADQ_LOCK();
 499         SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
 500         PF_OVERLOADQ_UNLOCK();
 501         taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
 502
 503         return (1);
 504 }
 505
 506 static void
 507 pf_overload_task(void *v, int pending)
 508 {
 509         struct pf_overload_head queue;
 510         struct pfr_addr p;
 511         struct pf_overload_entry *pfoe, *pfoe1;
 512         uint32_t killed = 0;
 513
 514         CURVNET_SET((struct vnet *)v);
 515
 516         PF_OVERLOADQ_LOCK();
 517         queue = V_pf_overloadqueue;
 518         SLIST_INIT(&V_pf_overloadqueue);
 519         PF_OVERLOADQ_UNLOCK();
 520
 521         bzero(&p, sizeof(p));
 522         SLIST_FOREACH(pfoe, &queue, next) {
 523                 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
 524                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
 525                         printf("%s: blocking address ", __func__);
 526                         pf_print_host(&pfoe->addr, 0, pfoe->af);
 527                         printf("\n");
 528                 }
 529
 530                 p.pfra_af = pfoe->af;
 531                 switch (pfoe->af) {
 532 #ifdef INET
 533                 case AF_INET:
 534                         p.pfra_net = 32;
 535                         p.pfra_ip4addr = pfoe->addr.v4;
 536                         break;
 537 #endif
 538 #ifdef INET6
 539                 case AF_INET6:
 540                         p.pfra_net = 128;
 541                         p.pfra_ip6addr = pfoe->addr.v6;
 542                         break;
 543 #endif
 544                 }
 545
 546                 PF_RULES_WLOCK();
 547                 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
 548                 PF_RULES_WUNLOCK();
 549         }
 550
 551         /*
 552          * Remove those entries, that don't need flushing.
 553          */
 554         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 555                 if (pfoe->rule->flush == 0) {
 556                         SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
 557                         free(pfoe, M_PFTEMP);
 558                 } else
 559                         counter_u64_add(
 560                             V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
 561
 562         /* If nothing to flush, return. */
 563         if (SLIST_EMPTY(&queue)) {
 564                 CURVNET_RESTORE();
 565                 return;
 566         }
 567
 568         for (int i = 0; i <= 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         CURVNET_RESTORE();
 597 }
 598
 599 /*
 600  * Can return locked on failure, so that we can consistently
 601  * allocate and insert a new one.
 602  */
 603 struct pf_src_node *
 604 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
 605         int returnlocked)
 606 {
 607         struct pf_srchash *sh;
 608         struct pf_src_node *n;
 609
 610         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
 611
 612         sh = &V_pf_srchash[pf_hashsrc(src, af)];
 613         PF_HASHROW_LOCK(sh);
 614         LIST_FOREACH(n, &sh->nodes, entry)
 615                 if (n->rule.ptr == rule && n->af == af &&
 616                     ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
 617                     (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
 618                         break;
 619         if (n != NULL || returnlocked == 0)
 620                 PF_HASHROW_UNLOCK(sh);
 621
 622         return (n);
 623 }
 624
 625 static int
 626 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
 627     struct pf_addr *src, sa_family_t af)
 628 {
 629
 630         KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
 631             rule->rpool.opts & PF_POOL_STICKYADDR),
 632             ("%s for non-tracking rule %p", __func__, rule));
 633
 634         if (*sn == NULL)
 635                 *sn = pf_find_src_node(src, rule, af, 1);
 636
 637         if (*sn == NULL) {
 638                 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
 639
 640                 PF_HASHROW_ASSERT(sh);
 641
 642                 if (!rule->max_src_nodes ||
 643                     counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
 644                         (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
 645                 else
 646                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
 647                             1);
 648                 if ((*sn) == NULL) {
 649                         PF_HASHROW_UNLOCK(sh);
 650                         return (-1);
 651                 }
 652
 653                 pf_init_threshold(&(*sn)->conn_rate,
 654                     rule->max_src_conn_rate.limit,
 655                     rule->max_src_conn_rate.seconds);
 656
 657                 (*sn)->af = af;
 658                 (*sn)->rule.ptr = rule;
 659                 PF_ACPY(&(*sn)->addr, src, af);
 660                 LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
 661                 (*sn)->creation = time_uptime;
 662                 (*sn)->ruletype = rule->action;
 663                 if ((*sn)->rule.ptr != NULL)
 664                         counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
 665                 PF_HASHROW_UNLOCK(sh);
 666                 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
 667         } else {
 668                 if (rule->max_src_states &&
 669                     (*sn)->states >= rule->max_src_states) {
 670                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
 671                             1);
 672                         return (-1);
 673                 }
 674         }
 675         return (0);
 676 }
 677
 678 void
 679 pf_unlink_src_node_locked(struct pf_src_node *src)
 680 {
 681 #ifdef INVARIANTS
 682         struct pf_srchash *sh;
 683
 684         sh = &V_pf_srchash[pf_hashsrc(&src->addr, src->af)];
 685         PF_HASHROW_ASSERT(sh);
 686 #endif
 687         LIST_REMOVE(src, entry);
 688         if (src->rule.ptr)
 689                 counter_u64_add(src->rule.ptr->src_nodes, -1);
 690         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
 691 }
 692
 693 void
 694 pf_unlink_src_node(struct pf_src_node *src)
 695 {
 696         struct pf_srchash *sh;
 697
 698         sh = &V_pf_srchash[pf_hashsrc(&src->addr, src->af)];
 699         PF_HASHROW_LOCK(sh);
 700         pf_unlink_src_node_locked(src);
 701         PF_HASHROW_UNLOCK(sh);
 702 }
 703
 704 static void
 705 pf_free_src_node(struct pf_src_node *sn)
 706 {
 707
 708         KASSERT(sn->states == 0, ("%s: %p has refs", __func__, sn));
 709         uma_zfree(V_pf_sources_z, sn);
 710 }
 711
 712 u_int
 713 pf_free_src_nodes(struct pf_src_node_list *head)
 714 {
 715         struct pf_src_node *sn, *tmp;
 716         u_int count = 0;
 717
 718         LIST_FOREACH_SAFE(sn, head, entry, tmp) {
 719                 pf_free_src_node(sn);
 720                 count++;
 721         }
 722
 723         return (count);
 724 }
 725
 726 void
 727 pf_mtag_initialize()
 728 {
 729
 730         pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
 731             sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
 732             UMA_ALIGN_PTR, 0);
 733 }
 734
 735 /* Per-vnet data storage structures initialization. */
 736 void
 737 pf_initialize()
 738 {
 739         struct pf_keyhash       *kh;
 740         struct pf_idhash        *ih;
 741         struct pf_srchash       *sh;
 742         u_int i;
 743
 744         TUNABLE_ULONG_FETCH("net.pf.states_hashsize", &pf_hashsize);
 745         if (pf_hashsize == 0 || !powerof2(pf_hashsize))
 746                 pf_hashsize = PF_HASHSIZ;
 747         TUNABLE_ULONG_FETCH("net.pf.source_nodes_hashsize", &pf_srchashsize);
 748         if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
 749                 pf_srchashsize = PF_HASHSIZ / 4;
 750
 751         V_pf_hashseed = arc4random();
 752
 753         /* States and state keys storage. */
 754         V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
 755             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 756         V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
 757         uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
 758         uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
 759
 760         V_pf_state_key_z = uma_zcreate("pf state keys",
 761             sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
 762             UMA_ALIGN_PTR, 0);
 763         V_pf_keyhash = malloc(pf_hashsize * sizeof(struct pf_keyhash),
 764             M_PFHASH, M_WAITOK | M_ZERO);
 765         V_pf_idhash = malloc(pf_hashsize * sizeof(struct pf_idhash),
 766             M_PFHASH, M_WAITOK | M_ZERO);
 767         pf_hashmask = pf_hashsize - 1;
 768         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 769             i++, kh++, ih++) {
 770                 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
 771                 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
 772         }
 773
 774         /* Source nodes. */
 775         V_pf_sources_z = uma_zcreate("pf source nodes",
 776             sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
 777             0);
 778         V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
 779         uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
 780         uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
 781         V_pf_srchash = malloc(pf_srchashsize * sizeof(struct pf_srchash),
 782           M_PFHASH, M_WAITOK|M_ZERO);
 783         pf_srchashmask = pf_srchashsize - 1;
 784         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
 785                 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
 786
 787         /* ALTQ */
 788         TAILQ_INIT(&V_pf_altqs[0]);
 789         TAILQ_INIT(&V_pf_altqs[1]);
 790         TAILQ_INIT(&V_pf_pabuf);
 791         V_pf_altqs_active = &V_pf_altqs[0];
 792         V_pf_altqs_inactive = &V_pf_altqs[1];
 793
 794
 795         /* Send & overload+flush queues. */
 796         STAILQ_INIT(&V_pf_sendqueue);
 797         SLIST_INIT(&V_pf_overloadqueue);
 798         TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
 799         mtx_init(&pf_sendqueue_mtx, "pf send queue", NULL, MTX_DEF);
 800         mtx_init(&pf_overloadqueue_mtx, "pf overload/flush queue", NULL,
 801             MTX_DEF);
 802
 803         /* Unlinked, but may be referenced rules. */
 804         TAILQ_INIT(&V_pf_unlinked_rules);
 805         mtx_init(&pf_unlnkdrules_mtx, "pf unlinked rules", NULL, MTX_DEF);
 806 }
 807
 808 void
 809 pf_mtag_cleanup()
 810 {
 811
 812         uma_zdestroy(pf_mtag_z);
 813 }
 814
 815 void
 816 pf_cleanup()
 817 {
 818         struct pf_keyhash       *kh;
 819         struct pf_idhash        *ih;
 820         struct pf_srchash       *sh;
 821         struct pf_send_entry    *pfse, *next;
 822         u_int i;
 823
 824         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 825             i++, kh++, ih++) {
 826                 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
 827                     __func__));
 828                 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
 829                     __func__));
 830                 mtx_destroy(&kh->lock);
 831                 mtx_destroy(&ih->lock);
 832         }
 833         free(V_pf_keyhash, M_PFHASH);
 834         free(V_pf_idhash, M_PFHASH);
 835
 836         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
 837                 KASSERT(LIST_EMPTY(&sh->nodes),
 838                     ("%s: source node hash not empty", __func__));
 839                 mtx_destroy(&sh->lock);
 840         }
 841         free(V_pf_srchash, M_PFHASH);
 842
 843         STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
 844                 m_freem(pfse->pfse_m);
 845                 free(pfse, M_PFTEMP);
 846         }
 847
 848         mtx_destroy(&pf_sendqueue_mtx);
 849         mtx_destroy(&pf_overloadqueue_mtx);
 850         mtx_destroy(&pf_unlnkdrules_mtx);
 851
 852         uma_zdestroy(V_pf_sources_z);
 853         uma_zdestroy(V_pf_state_z);
 854         uma_zdestroy(V_pf_state_key_z);
 855 }
 856
 857 static int
 858 pf_mtag_uminit(void *mem, int size, int how)
 859 {
 860         struct m_tag *t;
 861
 862         t = (struct m_tag *)mem;
 863         t->m_tag_cookie = MTAG_ABI_COMPAT;
 864         t->m_tag_id = PACKET_TAG_PF;
 865         t->m_tag_len = sizeof(struct pf_mtag);
 866         t->m_tag_free = pf_mtag_free;
 867
 868         return (0);
 869 }
 870
 871 static void
 872 pf_mtag_free(struct m_tag *t)
 873 {
 874
 875         uma_zfree(pf_mtag_z, t);
 876 }
 877
 878 struct pf_mtag *
 879 pf_get_mtag(struct mbuf *m)
 880 {
 881         struct m_tag *mtag;
 882
 883         if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
 884                 return ((struct pf_mtag *)(mtag + 1));
 885
 886         mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
 887         if (mtag == NULL)
 888                 return (NULL);
 889         bzero(mtag + 1, sizeof(struct pf_mtag));
 890         m_tag_prepend(m, mtag);
 891
 892         return ((struct pf_mtag *)(mtag + 1));
 893 }
 894
 895 static int
 896 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
 897     struct pf_state *s)
 898 {
 899         struct pf_keyhash       *khs, *khw, *kh;
 900         struct pf_state_key     *sk, *cur;
 901         struct pf_state         *si, *olds = NULL;
 902         int idx;
 903
 904         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
 905         KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
 906         KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
 907
 908         /*
 909          * We need to lock hash slots of both keys. To avoid deadlock
 910          * we always lock the slot with lower address first. Unlock order
 911          * isn't important.
 912          *
 913          * We also need to lock ID hash slot before dropping key
 914          * locks. On success we return with ID hash slot locked.
 915          */
 916
 917         if (skw == sks) {
 918                 khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
 919                 PF_HASHROW_LOCK(khs);
 920         } else {
 921                 khs = &V_pf_keyhash[pf_hashkey(sks)];
 922                 khw = &V_pf_keyhash[pf_hashkey(skw)];
 923                 if (khs == khw) {
 924                         PF_HASHROW_LOCK(khs);
 925                 } else if (khs < khw) {
 926                         PF_HASHROW_LOCK(khs);
 927                         PF_HASHROW_LOCK(khw);
 928                 } else {
 929                         PF_HASHROW_LOCK(khw);
 930                         PF_HASHROW_LOCK(khs);
 931                 }
 932         }
 933
 934 #define KEYS_UNLOCK()   do {                    \
 935         if (khs != khw) {                       \
 936                 PF_HASHROW_UNLOCK(khs);         \
 937                 PF_HASHROW_UNLOCK(khw);         \
 938         } else                                  \
 939                 PF_HASHROW_UNLOCK(khs);         \
 940 } while (0)
 941
 942         /*
 943          * First run: start with wire key.
 944          */
 945         sk = skw;
 946         kh = khw;
 947         idx = PF_SK_WIRE;
 948
 949 keyattach:
 950         LIST_FOREACH(cur, &kh->keys, entry)
 951                 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
 952                         break;
 953
 954         if (cur != NULL) {
 955                 /* Key exists. Check for same kif, if none, add to key. */
 956                 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
 957                         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
 958
 959                         PF_HASHROW_LOCK(ih);
 960                         if (si->kif == s->kif &&
 961                             si->direction == s->direction) {
 962                                 if (sk->proto == IPPROTO_TCP &&
 963                                     si->src.state >= TCPS_FIN_WAIT_2 &&
 964                                     si->dst.state >= TCPS_FIN_WAIT_2) {
 965                                         /*
 966                                          * New state matches an old >FIN_WAIT_2
 967                                          * state. We can't drop key hash locks,
 968                                          * thus we can't unlink it properly.
 969                                          *
 970                                          * As a workaround we drop it into
 971                                          * TCPS_CLOSED state, schedule purge
 972                                          * ASAP and push it into the very end
 973                                          * of the slot TAILQ, so that it won't
 974                                          * conflict with our new state.
 975                                          */
 976                                         si->src.state = si->dst.state =
 977                                             TCPS_CLOSED;
 978                                         si->timeout = PFTM_PURGE;
 979                                         olds = si;
 980                                 } else {
 981                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
 982                                                 printf("pf: %s key attach "
 983                                                     "failed on %s: ",
 984                                                     (idx == PF_SK_WIRE) ?
 985                                                     "wire" : "stack",
 986                                                     s->kif->pfik_name);
 987                                                 pf_print_state_parts(s,
 988                                                     (idx == PF_SK_WIRE) ?
 989                                                     sk : NULL,
 990                                                     (idx == PF_SK_STACK) ?
 991                                                     sk : NULL);
 992                                                 printf(", existing: ");
 993                                                 pf_print_state_parts(si,
 994                                                     (idx == PF_SK_WIRE) ?
 995                                                     sk : NULL,
 996                                                     (idx == PF_SK_STACK) ?
 997                                                     sk : NULL);
 998                                                 printf("\n");
 999                                         }
1000                                         PF_HASHROW_UNLOCK(ih);
1001                                         KEYS_UNLOCK();
1002                                         uma_zfree(V_pf_state_key_z, sk);
1003                                         if (idx == PF_SK_STACK)
1004                                                 pf_detach_state(s);
1005                                         return (EEXIST); /* collision! */
1006                                 }
1007                         }
1008                         PF_HASHROW_UNLOCK(ih);
1009                 }
1010                 uma_zfree(V_pf_state_key_z, sk);
1011                 s->key[idx] = cur;
1012         } else {
1013                 LIST_INSERT_HEAD(&kh->keys, sk, entry);
1014                 s->key[idx] = sk;
1015         }
1016
1017 stateattach:
1018         /* List is sorted, if-bound states before floating. */
1019         if (s->kif == V_pfi_all)
1020                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1021         else
1022                 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1023
1024         if (olds) {
1025                 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1026                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1027                     key_list[idx]);
1028                 olds = NULL;
1029         }
1030
1031         /*
1032          * Attach done. See how should we (or should not?)
1033          * attach a second key.
1034          */
1035         if (sks == skw) {
1036                 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1037                 idx = PF_SK_STACK;
1038                 sks = NULL;
1039                 goto stateattach;
1040         } else if (sks != NULL) {
1041                 /*
1042                  * Continue attaching with stack key.
1043                  */
1044                 sk = sks;
1045                 kh = khs;
1046                 idx = PF_SK_STACK;
1047                 sks = NULL;
1048                 goto keyattach;
1049         }
1050
1051         PF_STATE_LOCK(s);
1052         KEYS_UNLOCK();
1053
1054         KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1055             ("%s failure", __func__));
1056
1057         return (0);
1058 #undef  KEYS_UNLOCK
1059 }
1060
1061 static void
1062 pf_detach_state(struct pf_state *s)
1063 {
1064         struct pf_state_key *sks = s->key[PF_SK_STACK];
1065         struct pf_keyhash *kh;
1066
1067         if (sks != NULL) {
1068                 kh = &V_pf_keyhash[pf_hashkey(sks)];
1069                 PF_HASHROW_LOCK(kh);
1070                 if (s->key[PF_SK_STACK] != NULL)
1071                         pf_state_key_detach(s, PF_SK_STACK);
1072                 /*
1073                  * If both point to same key, then we are done.
1074                  */
1075                 if (sks == s->key[PF_SK_WIRE]) {
1076                         pf_state_key_detach(s, PF_SK_WIRE);
1077                         PF_HASHROW_UNLOCK(kh);
1078                         return;
1079                 }
1080                 PF_HASHROW_UNLOCK(kh);
1081         }
1082
1083         if (s->key[PF_SK_WIRE] != NULL) {
1084                 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1085                 PF_HASHROW_LOCK(kh);
1086                 if (s->key[PF_SK_WIRE] != NULL)
1087                         pf_state_key_detach(s, PF_SK_WIRE);
1088                 PF_HASHROW_UNLOCK(kh);
1089         }
1090 }
1091
1092 static void
1093 pf_state_key_detach(struct pf_state *s, int idx)
1094 {
1095         struct pf_state_key *sk = s->key[idx];
1096 #ifdef INVARIANTS
1097         struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1098
1099         PF_HASHROW_ASSERT(kh);
1100 #endif
1101         TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1102         s->key[idx] = NULL;
1103
1104         if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1105                 LIST_REMOVE(sk, entry);
1106                 uma_zfree(V_pf_state_key_z, sk);
1107         }
1108 }
1109
1110 static int
1111 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1112 {
1113         struct pf_state_key *sk = mem;
1114
1115         bzero(sk, sizeof(struct pf_state_key_cmp));
1116         TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1117         TAILQ_INIT(&sk->states[PF_SK_STACK]);
1118
1119         return (0);
1120 }
1121
1122 struct pf_state_key *
1123 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1124         struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1125 {
1126         struct pf_state_key *sk;
1127
1128         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1129         if (sk == NULL)
1130                 return (NULL);
1131
1132         PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1133         PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1134         sk->port[pd->sidx] = sport;
1135         sk->port[pd->didx] = dport;
1136         sk->proto = pd->proto;
1137         sk->af = pd->af;
1138
1139         return (sk);
1140 }
1141
1142 struct pf_state_key *
1143 pf_state_key_clone(struct pf_state_key *orig)
1144 {
1145         struct pf_state_key *sk;
1146
1147         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1148         if (sk == NULL)
1149                 return (NULL);
1150
1151         bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1152
1153         return (sk);
1154 }
1155
1156 int
1157 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1158     struct pf_state_key *sks, struct pf_state *s)
1159 {
1160         struct pf_idhash *ih;
1161         struct pf_state *cur;
1162         int error;
1163
1164         KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1165             ("%s: sks not pristine", __func__));
1166         KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1167             ("%s: skw not pristine", __func__));
1168         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1169
1170         s->kif = kif;
1171
1172         if (s->id == 0 && s->creatorid == 0) {
1173                 /* XXX: should be atomic, but probability of collision low */
1174                 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1175                         V_pf_stateid[curcpu] = 1;
1176                 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1177                 s->id = htobe64(s->id);
1178                 s->creatorid = V_pf_status.hostid;
1179         }
1180
1181         /* Returns with ID locked on success. */
1182         if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1183                 return (error);
1184
1185         ih = &V_pf_idhash[PF_IDHASH(s)];
1186         PF_HASHROW_ASSERT(ih);
1187         LIST_FOREACH(cur, &ih->states, entry)
1188                 if (cur->id == s->id && cur->creatorid == s->creatorid)
1189                         break;
1190
1191         if (cur != NULL) {
1192                 PF_HASHROW_UNLOCK(ih);
1193                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1194                         printf("pf: state ID collision: "
1195                             "id: %016llx creatorid: %08x\n",
1196                             (unsigned long long)be64toh(s->id),
1197                             ntohl(s->creatorid));
1198                 }
1199                 pf_detach_state(s);
1200                 return (EEXIST);
1201         }
1202         LIST_INSERT_HEAD(&ih->states, s, entry);
1203         /* One for keys, one for ID hash. */
1204         refcount_init(&s->refs, 2);
1205
1206         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1207         if (pfsync_insert_state_ptr != NULL)
1208                 pfsync_insert_state_ptr(s);
1209
1210         /* Returns locked. */
1211         return (0);
1212 }
1213
1214 /*
1215  * Find state by ID: returns with locked row on success.
1216  */
1217 struct pf_state *
1218 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1219 {
1220         struct pf_idhash *ih;
1221         struct pf_state *s;
1222
1223         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1224
1225         ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1226
1227         PF_HASHROW_LOCK(ih);
1228         LIST_FOREACH(s, &ih->states, entry)
1229                 if (s->id == id && s->creatorid == creatorid)
1230                         break;
1231
1232         if (s == NULL)
1233                 PF_HASHROW_UNLOCK(ih);
1234
1235         return (s);
1236 }
1237
1238 /*
1239  * Find state by key.
1240  * Returns with ID hash slot locked on success.
1241  */
1242 static struct pf_state *
1243 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1244 {
1245         struct pf_keyhash       *kh;
1246         struct pf_state_key     *sk;
1247         struct pf_state         *s;
1248         int idx;
1249
1250         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1251
1252         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1253
1254         PF_HASHROW_LOCK(kh);
1255         LIST_FOREACH(sk, &kh->keys, entry)
1256                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1257                         break;
1258         if (sk == NULL) {
1259                 PF_HASHROW_UNLOCK(kh);
1260                 return (NULL);
1261         }
1262
1263         idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1264
1265         /* List is sorted, if-bound states before floating ones. */
1266         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1267                 if (s->kif == V_pfi_all || s->kif == kif) {
1268                         PF_STATE_LOCK(s);
1269                         PF_HASHROW_UNLOCK(kh);
1270                         if (s->timeout >= PFTM_MAX) {
1271                                 /*
1272                                  * State is either being processed by
1273                                  * pf_unlink_state() in an other thread, or
1274                                  * is scheduled for immediate expiry.
1275                                  */
1276                                 PF_STATE_UNLOCK(s);
1277                                 return (NULL);
1278                         }
1279                         return (s);
1280                 }
1281         PF_HASHROW_UNLOCK(kh);
1282
1283         return (NULL);
1284 }
1285
1286 struct pf_state *
1287 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1288 {
1289         struct pf_keyhash       *kh;
1290         struct pf_state_key     *sk;
1291         struct pf_state         *s, *ret = NULL;
1292         int                      idx, inout = 0;
1293
1294         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1295
1296         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1297
1298         PF_HASHROW_LOCK(kh);
1299         LIST_FOREACH(sk, &kh->keys, entry)
1300                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1301                         break;
1302         if (sk == NULL) {
1303                 PF_HASHROW_UNLOCK(kh);
1304                 return (NULL);
1305         }
1306         switch (dir) {
1307         case PF_IN:
1308                 idx = PF_SK_WIRE;
1309                 break;
1310         case PF_OUT:
1311                 idx = PF_SK_STACK;
1312                 break;
1313         case PF_INOUT:
1314                 idx = PF_SK_WIRE;
1315                 inout = 1;
1316                 break;
1317         default:
1318                 panic("%s: dir %u", __func__, dir);
1319         }
1320 second_run:
1321         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1322                 if (more == NULL) {
1323                         PF_HASHROW_UNLOCK(kh);
1324                         return (s);
1325                 }
1326
1327                 if (ret)
1328                         (*more)++;
1329                 else
1330                         ret = s;
1331         }
1332         if (inout == 1) {
1333                 inout = 0;
1334                 idx = PF_SK_STACK;
1335                 goto second_run;
1336         }
1337         PF_HASHROW_UNLOCK(kh);
1338
1339         return (ret);
1340 }
1341
1342 /* END state table stuff */
1343
1344 static void
1345 pf_send(struct pf_send_entry *pfse)
1346 {
1347
1348         PF_SENDQ_LOCK();
1349         STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1350         PF_SENDQ_UNLOCK();
1351         swi_sched(V_pf_swi_cookie, 0);
1352 }
1353
1354 void
1355 pf_intr(void *v)
1356 {
1357         struct pf_send_head queue;
1358         struct pf_send_entry *pfse, *next;
1359
1360         CURVNET_SET((struct vnet *)v);
1361
1362         PF_SENDQ_LOCK();
1363         queue = V_pf_sendqueue;
1364         STAILQ_INIT(&V_pf_sendqueue);
1365         PF_SENDQ_UNLOCK();
1366
1367         STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1368                 switch (pfse->pfse_type) {
1369 #ifdef INET
1370                 case PFSE_IP:
1371                         ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1372                         break;
1373                 case PFSE_ICMP:
1374                         icmp_error(pfse->pfse_m, pfse->pfse_icmp_type,
1375                             pfse->pfse_icmp_code, 0, pfse->pfse_icmp_mtu);
1376                         break;
1377 #endif /* INET */
1378 #ifdef INET6
1379                 case PFSE_IP6:
1380                         ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1381                             NULL);
1382                         break;
1383                 case PFSE_ICMP6:
1384                         icmp6_error(pfse->pfse_m, pfse->pfse_icmp_type,
1385                             pfse->pfse_icmp_code, pfse->pfse_icmp_mtu);
1386                         break;
1387 #endif /* INET6 */
1388                 default:
1389                         panic("%s: unknown type", __func__);
1390                 }
1391                 free(pfse, M_PFTEMP);
1392         }
1393         CURVNET_RESTORE();
1394 }
1395
1396 void
1397 pf_purge_thread(void *v)
1398 {
1399         u_int idx = 0;
1400
1401         CURVNET_SET((struct vnet *)v);
1402
1403         for (;;) {
1404                 PF_RULES_RLOCK();
1405                 rw_sleep(pf_purge_thread, &pf_rules_lock, 0, "pftm", hz / 10);
1406
1407                 if (V_pf_end_threads) {
1408                         /*
1409                          * To cleanse up all kifs and rules we need
1410                          * two runs: first one clears reference flags,
1411                          * then pf_purge_expired_states() doesn't
1412                          * raise them, and then second run frees.
1413                          */
1414                         PF_RULES_RUNLOCK();
1415                         pf_purge_unlinked_rules();
1416                         pfi_kif_purge();
1417
1418                         /*
1419                          * Now purge everything.
1420                          */
1421                         pf_purge_expired_states(0, pf_hashmask);
1422                         pf_purge_expired_fragments();
1423                         pf_purge_expired_src_nodes();
1424
1425                         /*
1426                          * Now all kifs & rules should be unreferenced,
1427                          * thus should be successfully freed.
1428                          */
1429                         pf_purge_unlinked_rules();
1430                         pfi_kif_purge();
1431
1432                         /*
1433                          * Announce success and exit.
1434                          */
1435                         PF_RULES_RLOCK();
1436                         V_pf_end_threads++;
1437                         PF_RULES_RUNLOCK();
1438                         wakeup(pf_purge_thread);
1439                         kproc_exit(0);
1440                 }
1441                 PF_RULES_RUNLOCK();
1442
1443                 /* Process 1/interval fraction of the state table every run. */
1444                 idx = pf_purge_expired_states(idx, pf_hashmask /
1445                             (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1446
1447                 /* Purge other expired types every PFTM_INTERVAL seconds. */
1448                 if (idx == 0) {
1449                         /*
1450                          * Order is important:
1451                          * - states and src nodes reference rules
1452                          * - states and rules reference kifs
1453                          */
1454                         pf_purge_expired_fragments();
1455                         pf_purge_expired_src_nodes();
1456                         pf_purge_unlinked_rules();
1457                         pfi_kif_purge();
1458                 }
1459         }
1460         /* not reached */
1461         CURVNET_RESTORE();
1462 }
1463
1464 u_int32_t
1465 pf_state_expires(const struct pf_state *state)
1466 {
1467         u_int32_t       timeout;
1468         u_int32_t       start;
1469         u_int32_t       end;
1470         u_int32_t       states;
1471
1472         /* handle all PFTM_* > PFTM_MAX here */
1473         if (state->timeout == PFTM_PURGE)
1474                 return (time_uptime);
1475         KASSERT(state->timeout != PFTM_UNLINKED,
1476             ("pf_state_expires: timeout == PFTM_UNLINKED"));
1477         KASSERT((state->timeout < PFTM_MAX),
1478             ("pf_state_expires: timeout > PFTM_MAX"));
1479         timeout = state->rule.ptr->timeout[state->timeout];
1480         if (!timeout)
1481                 timeout = V_pf_default_rule.timeout[state->timeout];
1482         start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1483         if (start) {
1484                 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1485                 states = counter_u64_fetch(state->rule.ptr->states_cur);
1486         } else {
1487                 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1488                 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1489                 states = V_pf_status.states;
1490         }
1491         if (end && states > start && start < end) {
1492                 if (states < end)
1493                         return (state->expire + timeout * (end - states) /
1494                             (end - start));
1495                 else
1496                         return (time_uptime);
1497         }
1498         return (state->expire + timeout);
1499 }
1500
1501 void
1502 pf_purge_expired_src_nodes()
1503 {
1504         struct pf_src_node_list  freelist;
1505         struct pf_srchash       *sh;
1506         struct pf_src_node      *cur, *next;
1507         int i;
1508
1509         LIST_INIT(&freelist);
1510         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1511             PF_HASHROW_LOCK(sh);
1512             LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1513                 if (cur->states == 0 && cur->expire <= time_uptime) {
1514                         pf_unlink_src_node_locked(cur);
1515                         LIST_INSERT_HEAD(&freelist, cur, entry);
1516                 } else if (cur->rule.ptr != NULL)
1517                         cur->rule.ptr->rule_flag |= PFRULE_REFS;
1518             PF_HASHROW_UNLOCK(sh);
1519         }
1520
1521         pf_free_src_nodes(&freelist);
1522
1523         V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1524 }
1525
1526 static void
1527 pf_src_tree_remove_state(struct pf_state *s)
1528 {
1529         u_int32_t timeout;
1530
1531         if (s->src_node != NULL) {
1532                 if (s->src.tcp_est)
1533                         --s->src_node->conn;
1534                 if (--s->src_node->states == 0) {
1535                         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1536                         if (!timeout)
1537                                 timeout =
1538                                     V_pf_default_rule.timeout[PFTM_SRC_NODE];
1539                         s->src_node->expire = time_uptime + timeout;
1540                 }
1541         }
1542         if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1543                 if (--s->nat_src_node->states == 0) {
1544                         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1545                         if (!timeout)
1546                                 timeout =
1547                                     V_pf_default_rule.timeout[PFTM_SRC_NODE];
1548                         s->nat_src_node->expire = time_uptime + timeout;
1549                 }
1550         }
1551         s->src_node = s->nat_src_node = NULL;
1552 }
1553
1554 /*
1555  * Unlink and potentilly free a state. Function may be
1556  * called with ID hash row locked, but always returns
1557  * unlocked, since it needs to go through key hash locking.
1558  */
1559 int
1560 pf_unlink_state(struct pf_state *s, u_int flags)
1561 {
1562         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1563
1564         if ((flags & PF_ENTER_LOCKED) == 0)
1565                 PF_HASHROW_LOCK(ih);
1566         else
1567                 PF_HASHROW_ASSERT(ih);
1568
1569         if (s->timeout == PFTM_UNLINKED) {
1570                 /*
1571                  * State is being processed
1572                  * by pf_unlink_state() in
1573                  * an other thread.
1574                  */
1575                 PF_HASHROW_UNLOCK(ih);
1576                 return (0);     /* XXXGL: undefined actually */
1577         }
1578
1579         if (s->src.state == PF_TCPS_PROXY_DST) {
1580                 /* XXX wire key the right one? */
1581                 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1582                     &s->key[PF_SK_WIRE]->addr[1],
1583                     &s->key[PF_SK_WIRE]->addr[0],
1584                     s->key[PF_SK_WIRE]->port[1],
1585                     s->key[PF_SK_WIRE]->port[0],
1586                     s->src.seqhi, s->src.seqlo + 1,
1587                     TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1588         }
1589
1590         LIST_REMOVE(s, entry);
1591         pf_src_tree_remove_state(s);
1592
1593         if (pfsync_delete_state_ptr != NULL)
1594                 pfsync_delete_state_ptr(s);
1595
1596         STATE_DEC_COUNTERS(s);
1597
1598         s->timeout = PFTM_UNLINKED;
1599
1600         PF_HASHROW_UNLOCK(ih);
1601
1602         pf_detach_state(s);
1603         refcount_release(&s->refs);
1604
1605         return (pf_release_state(s));
1606 }
1607
1608 void
1609 pf_free_state(struct pf_state *cur)
1610 {
1611
1612         KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1613         KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1614             cur->timeout));
1615
1616         pf_normalize_tcp_cleanup(cur);
1617         uma_zfree(V_pf_state_z, cur);
1618         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
1619 }
1620
1621 /*
1622  * Called only from pf_purge_thread(), thus serialized.
1623  */
1624 static u_int
1625 pf_purge_expired_states(u_int i, int maxcheck)
1626 {
1627         struct pf_idhash *ih;
1628         struct pf_state *s;
1629
1630         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1631
1632         /*
1633          * Go through hash and unlink states that expire now.
1634          */
1635         while (maxcheck > 0) {
1636
1637                 ih = &V_pf_idhash[i];
1638 relock:
1639                 PF_HASHROW_LOCK(ih);
1640                 LIST_FOREACH(s, &ih->states, entry) {
1641                         if (pf_state_expires(s) <= time_uptime) {
1642                                 V_pf_status.states -=
1643                                     pf_unlink_state(s, PF_ENTER_LOCKED);
1644                                 goto relock;
1645                         }
1646                         s->rule.ptr->rule_flag |= PFRULE_REFS;
1647                         if (s->nat_rule.ptr != NULL)
1648                                 s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1649                         if (s->anchor.ptr != NULL)
1650                                 s->anchor.ptr->rule_flag |= PFRULE_REFS;
1651                         s->kif->pfik_flags |= PFI_IFLAG_REFS;
1652                         if (s->rt_kif)
1653                                 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1654                 }
1655                 PF_HASHROW_UNLOCK(ih);
1656
1657                 /* Return when we hit end of hash. */
1658                 if (++i > pf_hashmask) {
1659                         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1660                         return (0);
1661                 }
1662
1663                 maxcheck--;
1664         }
1665
1666         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1667
1668         return (i);
1669 }
1670
1671 static void
1672 pf_purge_unlinked_rules()
1673 {
1674         struct pf_rulequeue tmpq;
1675         struct pf_rule *r, *r1;
1676
1677         /*
1678          * If we have overloading task pending, then we'd
1679          * better skip purging this time. There is a tiny
1680          * probability that overloading task references
1681          * an already unlinked rule.
1682          */
1683         PF_OVERLOADQ_LOCK();
1684         if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1685                 PF_OVERLOADQ_UNLOCK();
1686                 return;
1687         }
1688         PF_OVERLOADQ_UNLOCK();
1689
1690         /*
1691          * Do naive mark-and-sweep garbage collecting of old rules.
1692          * Reference flag is raised by pf_purge_expired_states()
1693          * and pf_purge_expired_src_nodes().
1694          *
1695          * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1696          * use a temporary queue.
1697          */
1698         TAILQ_INIT(&tmpq);
1699         PF_UNLNKDRULES_LOCK();
1700         TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1701                 if (!(r->rule_flag & PFRULE_REFS)) {
1702                         TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1703                         TAILQ_INSERT_TAIL(&tmpq, r, entries);
1704                 } else
1705                         r->rule_flag &= ~PFRULE_REFS;
1706         }
1707         PF_UNLNKDRULES_UNLOCK();
1708
1709         if (!TAILQ_EMPTY(&tmpq)) {
1710                 PF_RULES_WLOCK();
1711                 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1712                         TAILQ_REMOVE(&tmpq, r, entries);
1713                         pf_free_rule(r);
1714                 }
1715                 PF_RULES_WUNLOCK();
1716         }
1717 }
1718
1719 void
1720 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1721 {
1722         switch (af) {
1723 #ifdef INET
1724         case AF_INET: {
1725                 u_int32_t a = ntohl(addr->addr32[0]);
1726                 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1727                     (a>>8)&255, a&255);
1728                 if (p) {
1729                         p = ntohs(p);
1730                         printf(":%u", p);
1731                 }
1732                 break;
1733         }
1734 #endif /* INET */
1735 #ifdef INET6
1736         case AF_INET6: {
1737                 u_int16_t b;
1738                 u_int8_t i, curstart, curend, maxstart, maxend;
1739                 curstart = curend = maxstart = maxend = 255;
1740                 for (i = 0; i < 8; i++) {
1741                         if (!addr->addr16[i]) {
1742                                 if (curstart == 255)
1743                                         curstart = i;
1744                                 curend = i;
1745                         } else {
1746                                 if ((curend - curstart) >
1747                                     (maxend - maxstart)) {
1748                                         maxstart = curstart;
1749                                         maxend = curend;
1750                                 }
1751                                 curstart = curend = 255;
1752                         }
1753                 }
1754                 if ((curend - curstart) >
1755                     (maxend - maxstart)) {
1756                         maxstart = curstart;
1757                         maxend = curend;
1758                 }
1759                 for (i = 0; i < 8; i++) {
1760                         if (i >= maxstart && i <= maxend) {
1761                                 if (i == 0)
1762                                         printf(":");
1763                                 if (i == maxend)
1764                                         printf(":");
1765                         } else {
1766                                 b = ntohs(addr->addr16[i]);
1767                                 printf("%x", b);
1768                                 if (i < 7)
1769                                         printf(":");
1770                         }
1771                 }
1772                 if (p) {
1773                         p = ntohs(p);
1774                         printf("[%u]", p);
1775                 }
1776                 break;
1777         }
1778 #endif /* INET6 */
1779         }
1780 }
1781
1782 void
1783 pf_print_state(struct pf_state *s)
1784 {
1785         pf_print_state_parts(s, NULL, NULL);
1786 }
1787
1788 static void
1789 pf_print_state_parts(struct pf_state *s,
1790     struct pf_state_key *skwp, struct pf_state_key *sksp)
1791 {
1792         struct pf_state_key *skw, *sks;
1793         u_int8_t proto, dir;
1794
1795         /* Do our best to fill these, but they're skipped if NULL */
1796         skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1797         sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1798         proto = skw ? skw->proto : (sks ? sks->proto : 0);
1799         dir = s ? s->direction : 0;
1800
1801         switch (proto) {
1802         case IPPROTO_IPV4:
1803                 printf("IPv4");
1804                 break;
1805         case IPPROTO_IPV6:
1806                 printf("IPv6");
1807                 break;
1808         case IPPROTO_TCP:
1809                 printf("TCP");
1810                 break;
1811         case IPPROTO_UDP:
1812                 printf("UDP");
1813                 break;
1814         case IPPROTO_ICMP:
1815                 printf("ICMP");
1816                 break;
1817         case IPPROTO_ICMPV6:
1818                 printf("ICMPv6");
1819                 break;
1820         default:
1821                 printf("%u", skw->proto);
1822                 break;
1823         }
1824         switch (dir) {
1825         case PF_IN:
1826                 printf(" in");
1827                 break;
1828         case PF_OUT:
1829                 printf(" out");
1830                 break;
1831         }
1832         if (skw) {
1833                 printf(" wire: ");
1834                 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1835                 printf(" ");
1836                 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1837         }
1838         if (sks) {
1839                 printf(" stack: ");
1840                 if (sks != skw) {
1841                         pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1842                         printf(" ");
1843                         pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1844                 } else
1845                         printf("-");
1846         }
1847         if (s) {
1848                 if (proto == IPPROTO_TCP) {
1849                         printf(" [lo=%u high=%u win=%u modulator=%u",
1850                             s->src.seqlo, s->src.seqhi,
1851                             s->src.max_win, s->src.seqdiff);
1852                         if (s->src.wscale && s->dst.wscale)
1853                                 printf(" wscale=%u",
1854                                     s->src.wscale & PF_WSCALE_MASK);
1855                         printf("]");
1856                         printf(" [lo=%u high=%u win=%u modulator=%u",
1857                             s->dst.seqlo, s->dst.seqhi,
1858                             s->dst.max_win, s->dst.seqdiff);
1859                         if (s->src.wscale && s->dst.wscale)
1860                                 printf(" wscale=%u",
1861                                 s->dst.wscale & PF_WSCALE_MASK);
1862                         printf("]");
1863                 }
1864                 printf(" %u:%u", s->src.state, s->dst.state);
1865         }
1866 }
1867
1868 void
1869 pf_print_flags(u_int8_t f)
1870 {
1871         if (f)
1872                 printf(" ");
1873         if (f & TH_FIN)
1874                 printf("F");
1875         if (f & TH_SYN)
1876                 printf("S");
1877         if (f & TH_RST)
1878                 printf("R");
1879         if (f & TH_PUSH)
1880                 printf("P");
1881         if (f & TH_ACK)
1882                 printf("A");
1883         if (f & TH_URG)
1884                 printf("U");
1885         if (f & TH_ECE)
1886                 printf("E");
1887         if (f & TH_CWR)
1888                 printf("W");
1889 }
1890
1891 #define PF_SET_SKIP_STEPS(i)                                    \
1892         do {                                                    \
1893                 while (head[i] != cur) {                        \
1894                         head[i]->skip[i].ptr = cur;             \
1895                         head[i] = TAILQ_NEXT(head[i], entries); \
1896                 }                                               \
1897         } while (0)
1898
1899 void
1900 pf_calc_skip_steps(struct pf_rulequeue *rules)
1901 {
1902         struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1903         int i;
1904
1905         cur = TAILQ_FIRST(rules);
1906         prev = cur;
1907         for (i = 0; i < PF_SKIP_COUNT; ++i)
1908                 head[i] = cur;
1909         while (cur != NULL) {
1910
1911                 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1912                         PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1913                 if (cur->direction != prev->direction)
1914                         PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1915                 if (cur->af != prev->af)
1916                         PF_SET_SKIP_STEPS(PF_SKIP_AF);
1917                 if (cur->proto != prev->proto)
1918                         PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
1919                 if (cur->src.neg != prev->src.neg ||
1920                     pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
1921                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
1922                 if (cur->src.port[0] != prev->src.port[0] ||
1923                     cur->src.port[1] != prev->src.port[1] ||
1924                     cur->src.port_op != prev->src.port_op)
1925                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
1926                 if (cur->dst.neg != prev->dst.neg ||
1927                     pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
1928                         PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
1929                 if (cur->dst.port[0] != prev->dst.port[0] ||
1930                     cur->dst.port[1] != prev->dst.port[1] ||
1931                     cur->dst.port_op != prev->dst.port_op)
1932                         PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
1933
1934                 prev = cur;
1935                 cur = TAILQ_NEXT(cur, entries);
1936         }
1937         for (i = 0; i < PF_SKIP_COUNT; ++i)
1938                 PF_SET_SKIP_STEPS(i);
1939 }
1940
1941 static int
1942 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
1943 {
1944         if (aw1->type != aw2->type)
1945                 return (1);
1946         switch (aw1->type) {
1947         case PF_ADDR_ADDRMASK:
1948         case PF_ADDR_RANGE:
1949                 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0))
1950                         return (1);
1951                 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0))
1952                         return (1);
1953                 return (0);
1954         case PF_ADDR_DYNIFTL:
1955                 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
1956         case PF_ADDR_NOROUTE:
1957         case PF_ADDR_URPFFAILED:
1958                 return (0);
1959         case PF_ADDR_TABLE:
1960                 return (aw1->p.tbl != aw2->p.tbl);
1961         default:
1962                 printf("invalid address type: %d\n", aw1->type);
1963                 return (1);
1964         }
1965 }
1966
1967 u_int16_t
1968 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
1969 {
1970         u_int32_t       l;
1971
1972         if (udp && !cksum)
1973                 return (0x0000);
1974         l = cksum + old - new;
1975         l = (l >> 16) + (l & 65535);
1976         l = l & 65535;
1977         if (udp && !l)
1978                 return (0xFFFF);
1979         return (l);
1980 }
1981
1982 static void
1983 pf_change_ap(struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc,
1984     struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af)
1985 {
1986         struct pf_addr  ao;
1987         u_int16_t       po = *p;
1988
1989         PF_ACPY(&ao, a, af);
1990         PF_ACPY(a, an, af);
1991
1992         *p = pn;
1993
1994         switch (af) {
1995 #ifdef INET
1996         case AF_INET:
1997                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
1998                     ao.addr16[0], an->addr16[0], 0),
1999                     ao.addr16[1], an->addr16[1], 0);
2000                 *p = pn;
2001                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
2002                     ao.addr16[0], an->addr16[0], u),
2003                     ao.addr16[1], an->addr16[1], u),
2004                     po, pn, u);
2005                 break;
2006 #endif /* INET */
2007 #ifdef INET6
2008         case AF_INET6:
2009                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2010                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2011                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
2012                     ao.addr16[0], an->addr16[0], u),
2013                     ao.addr16[1], an->addr16[1], u),
2014                     ao.addr16[2], an->addr16[2], u),
2015                     ao.addr16[3], an->addr16[3], u),
2016                     ao.addr16[4], an->addr16[4], u),
2017                     ao.addr16[5], an->addr16[5], u),
2018                     ao.addr16[6], an->addr16[6], u),
2019                     ao.addr16[7], an->addr16[7], u),
2020                     po, pn, u);
2021                 break;
2022 #endif /* INET6 */
2023         }
2024 }
2025
2026
2027 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2028 void
2029 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2030 {
2031         u_int32_t       ao;
2032
2033         memcpy(&ao, a, sizeof(ao));
2034         memcpy(a, &an, sizeof(u_int32_t));
2035         *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2036             ao % 65536, an % 65536, u);
2037 }
2038
2039 #ifdef INET6
2040 static void
2041 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2042 {
2043         struct pf_addr  ao;
2044
2045         PF_ACPY(&ao, a, AF_INET6);
2046         PF_ACPY(a, an, AF_INET6);
2047
2048         *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2049             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2050             pf_cksum_fixup(pf_cksum_fixup(*c,
2051             ao.addr16[0], an->addr16[0], u),
2052             ao.addr16[1], an->addr16[1], u),
2053             ao.addr16[2], an->addr16[2], u),
2054             ao.addr16[3], an->addr16[3], u),
2055             ao.addr16[4], an->addr16[4], u),
2056             ao.addr16[5], an->addr16[5], u),
2057             ao.addr16[6], an->addr16[6], u),
2058             ao.addr16[7], an->addr16[7], u);
2059 }
2060 #endif /* INET6 */
2061
2062 static void
2063 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2064     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2065     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2066 {
2067         struct pf_addr  oia, ooa;
2068
2069         PF_ACPY(&oia, ia, af);
2070         if (oa)
2071                 PF_ACPY(&ooa, oa, af);
2072
2073         /* Change inner protocol port, fix inner protocol checksum. */
2074         if (ip != NULL) {
2075                 u_int16_t       oip = *ip;
2076                 u_int32_t       opc;
2077
2078                 if (pc != NULL)
2079                         opc = *pc;
2080                 *ip = np;
2081                 if (pc != NULL)
2082                         *pc = pf_cksum_fixup(*pc, oip, *ip, u);
2083                 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2084                 if (pc != NULL)
2085                         *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2086         }
2087         /* Change inner ip address, fix inner ip and icmp checksums. */
2088         PF_ACPY(ia, na, af);
2089         switch (af) {
2090 #ifdef INET
2091         case AF_INET: {
2092                 u_int32_t        oh2c = *h2c;
2093
2094                 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2095                     oia.addr16[0], ia->addr16[0], 0),
2096                     oia.addr16[1], ia->addr16[1], 0);
2097                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2098                     oia.addr16[0], ia->addr16[0], 0),
2099                     oia.addr16[1], ia->addr16[1], 0);
2100                 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2101                 break;
2102         }
2103 #endif /* INET */
2104 #ifdef INET6
2105         case AF_INET6:
2106                 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2107                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2108                     pf_cksum_fixup(pf_cksum_fixup(*ic,
2109                     oia.addr16[0], ia->addr16[0], u),
2110                     oia.addr16[1], ia->addr16[1], u),
2111                     oia.addr16[2], ia->addr16[2], u),
2112                     oia.addr16[3], ia->addr16[3], u),
2113                     oia.addr16[4], ia->addr16[4], u),
2114                     oia.addr16[5], ia->addr16[5], u),
2115                     oia.addr16[6], ia->addr16[6], u),
2116                     oia.addr16[7], ia->addr16[7], u);
2117                 break;
2118 #endif /* INET6 */
2119         }
2120         /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2121         if (oa) {
2122                 PF_ACPY(oa, na, af);
2123                 switch (af) {
2124 #ifdef INET
2125                 case AF_INET:
2126                         *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2127                             ooa.addr16[0], oa->addr16[0], 0),
2128                             ooa.addr16[1], oa->addr16[1], 0);
2129                         break;
2130 #endif /* INET */
2131 #ifdef INET6
2132                 case AF_INET6:
2133                         *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2134                             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2135                             pf_cksum_fixup(pf_cksum_fixup(*ic,
2136                             ooa.addr16[0], oa->addr16[0], u),
2137                             ooa.addr16[1], oa->addr16[1], u),
2138                             ooa.addr16[2], oa->addr16[2], u),
2139                             ooa.addr16[3], oa->addr16[3], u),
2140                             ooa.addr16[4], oa->addr16[4], u),
2141                             ooa.addr16[5], oa->addr16[5], u),
2142                             ooa.addr16[6], oa->addr16[6], u),
2143                             ooa.addr16[7], oa->addr16[7], u);
2144                         break;
2145 #endif /* INET6 */
2146                 }
2147         }
2148 }
2149
2150
2151 /*
2152  * Need to modulate the sequence numbers in the TCP SACK option
2153  * (credits to Krzysztof Pfaff for report and patch)
2154  */
2155 static int
2156 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2157     struct tcphdr *th, struct pf_state_peer *dst)
2158 {
2159         int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2160         u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2161         int copyback = 0, i, olen;
2162         struct sackblk sack;
2163
2164 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2165         if (hlen < TCPOLEN_SACKLEN ||
2166             !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2167                 return 0;
2168
2169         while (hlen >= TCPOLEN_SACKLEN) {
2170                 olen = opt[1];
2171                 switch (*opt) {
2172                 case TCPOPT_EOL:        /* FALLTHROUGH */
2173                 case TCPOPT_NOP:
2174                         opt++;
2175                         hlen--;
2176                         break;
2177                 case TCPOPT_SACK:
2178                         if (olen > hlen)
2179                                 olen = hlen;
2180                         if (olen >= TCPOLEN_SACKLEN) {
2181                                 for (i = 2; i + TCPOLEN_SACK <= olen;
2182                                     i += TCPOLEN_SACK) {
2183                                         memcpy(&sack, &opt[i], sizeof(sack));
2184                                         pf_change_a(&sack.start, &th->th_sum,
2185                                             htonl(ntohl(sack.start) -
2186                                             dst->seqdiff), 0);
2187                                         pf_change_a(&sack.end, &th->th_sum,
2188                                             htonl(ntohl(sack.end) -
2189                                             dst->seqdiff), 0);
2190                                         memcpy(&opt[i], &sack, sizeof(sack));
2191                                 }
2192                                 copyback = 1;
2193                         }
2194                         /* FALLTHROUGH */
2195                 default:
2196                         if (olen < 2)
2197                                 olen = 2;
2198                         hlen -= olen;
2199                         opt += olen;
2200                 }
2201         }
2202
2203         if (copyback)
2204                 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2205         return (copyback);
2206 }
2207
2208 static void
2209 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2210     const struct pf_addr *saddr, const struct pf_addr *daddr,
2211     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2212     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2213     u_int16_t rtag, struct ifnet *ifp)
2214 {
2215         struct pf_send_entry *pfse;
2216         struct mbuf     *m;
2217         int              len, tlen;
2218 #ifdef INET
2219         struct ip       *h = NULL;
2220 #endif /* INET */
2221 #ifdef INET6
2222         struct ip6_hdr  *h6 = NULL;
2223 #endif /* INET6 */
2224         struct tcphdr   *th;
2225         char            *opt;
2226         struct pf_mtag  *pf_mtag;
2227
2228         len = 0;
2229         th = NULL;
2230
2231         /* maximum segment size tcp option */
2232         tlen = sizeof(struct tcphdr);
2233         if (mss)
2234                 tlen += 4;
2235
2236         switch (af) {
2237 #ifdef INET
2238         case AF_INET:
2239                 len = sizeof(struct ip) + tlen;
2240                 break;
2241 #endif /* INET */
2242 #ifdef INET6
2243         case AF_INET6:
2244                 len = sizeof(struct ip6_hdr) + tlen;
2245                 break;
2246 #endif /* INET6 */
2247         default:
2248                 panic("%s: unsupported af %d", __func__, af);
2249         }
2250
2251         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2252         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2253         if (pfse == NULL)
2254                 return;
2255         m = m_gethdr(M_NOWAIT, MT_DATA);
2256         if (m == NULL) {
2257                 free(pfse, M_PFTEMP);
2258                 return;
2259         }
2260 #ifdef MAC
2261         mac_netinet_firewall_send(m);
2262 #endif
2263         if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2264                 free(pfse, M_PFTEMP);
2265                 m_freem(m);
2266                 return;
2267         }
2268         if (tag)
2269                 m->m_flags |= M_SKIP_FIREWALL;
2270         pf_mtag->tag = rtag;
2271
2272         if (r != NULL && r->rtableid >= 0)
2273                 M_SETFIB(m, r->rtableid);
2274
2275 #ifdef ALTQ
2276         if (r != NULL && r->qid) {
2277                 pf_mtag->qid = r->qid;
2278
2279                 /* add hints for ecn */
2280                 pf_mtag->hdr = mtod(m, struct ip *);
2281         }
2282 #endif /* ALTQ */
2283         m->m_data += max_linkhdr;
2284         m->m_pkthdr.len = m->m_len = len;
2285         m->m_pkthdr.rcvif = NULL;
2286         bzero(m->m_data, len);
2287         switch (af) {
2288 #ifdef INET
2289         case AF_INET:
2290                 h = mtod(m, struct ip *);
2291
2292                 /* IP header fields included in the TCP checksum */
2293                 h->ip_p = IPPROTO_TCP;
2294                 h->ip_len = htons(tlen);
2295                 h->ip_src.s_addr = saddr->v4.s_addr;
2296                 h->ip_dst.s_addr = daddr->v4.s_addr;
2297
2298                 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2299                 break;
2300 #endif /* INET */
2301 #ifdef INET6
2302         case AF_INET6:
2303                 h6 = mtod(m, struct ip6_hdr *);
2304
2305                 /* IP header fields included in the TCP checksum */
2306                 h6->ip6_nxt = IPPROTO_TCP;
2307                 h6->ip6_plen = htons(tlen);
2308                 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2309                 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2310
2311                 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2312                 break;
2313 #endif /* INET6 */
2314         }
2315
2316         /* TCP header */
2317         th->th_sport = sport;
2318         th->th_dport = dport;
2319         th->th_seq = htonl(seq);
2320         th->th_ack = htonl(ack);
2321         th->th_off = tlen >> 2;
2322         th->th_flags = flags;
2323         th->th_win = htons(win);
2324
2325         if (mss) {
2326                 opt = (char *)(th + 1);
2327                 opt[0] = TCPOPT_MAXSEG;
2328                 opt[1] = 4;
2329                 HTONS(mss);
2330                 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2331         }
2332
2333         switch (af) {
2334 #ifdef INET
2335         case AF_INET:
2336                 /* TCP checksum */
2337                 th->th_sum = in_cksum(m, len);
2338
2339                 /* Finish the IP header */
2340                 h->ip_v = 4;
2341                 h->ip_hl = sizeof(*h) >> 2;
2342                 h->ip_tos = IPTOS_LOWDELAY;
2343                 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2344                 h->ip_len = htons(len);
2345                 h->ip_ttl = ttl ? ttl : V_ip_defttl;
2346                 h->ip_sum = 0;
2347
2348                 pfse->pfse_type = PFSE_IP;
2349                 break;
2350 #endif /* INET */
2351 #ifdef INET6
2352         case AF_INET6:
2353                 /* TCP checksum */
2354                 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2355                     sizeof(struct ip6_hdr), tlen);
2356
2357                 h6->ip6_vfc |= IPV6_VERSION;
2358                 h6->ip6_hlim = IPV6_DEFHLIM;
2359
2360                 pfse->pfse_type = PFSE_IP6;
2361                 break;
2362 #endif /* INET6 */
2363         }
2364         pfse->pfse_m = m;
2365         pf_send(pfse);
2366 }
2367
2368 static void
2369 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2370     struct pf_rule *r)
2371 {
2372         struct pf_send_entry *pfse;
2373         struct mbuf *m0;
2374         struct pf_mtag *pf_mtag;
2375
2376         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2377         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2378         if (pfse == NULL)
2379                 return;
2380
2381         if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2382                 free(pfse, M_PFTEMP);
2383                 return;
2384         }
2385
2386         if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2387                 free(pfse, M_PFTEMP);
2388                 return;
2389         }
2390         /* XXX: revisit */
2391         m0->m_flags |= M_SKIP_FIREWALL;
2392
2393         if (r->rtableid >= 0)
2394                 M_SETFIB(m0, r->rtableid);
2395
2396 #ifdef ALTQ
2397         if (r->qid) {
2398                 pf_mtag->qid = r->qid;
2399                 /* add hints for ecn */
2400                 pf_mtag->hdr = mtod(m0, struct ip *);
2401         }
2402 #endif /* ALTQ */
2403
2404         switch (af) {
2405 #ifdef INET
2406         case AF_INET:
2407                 pfse->pfse_type = PFSE_ICMP;
2408                 break;
2409 #endif /* INET */
2410 #ifdef INET6
2411         case AF_INET6:
2412                 pfse->pfse_type = PFSE_ICMP6;
2413                 break;
2414 #endif /* INET6 */
2415         }
2416         pfse->pfse_m = m0;
2417         pfse->pfse_icmp_type = type;
2418         pfse->pfse_icmp_code = code;
2419         pf_send(pfse);
2420 }
2421
2422 /*
2423  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2424  * If n is 0, they match if they are equal. If n is != 0, they match if they
2425  * are different.
2426  */
2427 int
2428 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2429     struct pf_addr *b, sa_family_t af)
2430 {
2431         int     match = 0;
2432
2433         switch (af) {
2434 #ifdef INET
2435         case AF_INET:
2436                 if ((a->addr32[0] & m->addr32[0]) ==
2437                     (b->addr32[0] & m->addr32[0]))
2438                         match++;
2439                 break;
2440 #endif /* INET */
2441 #ifdef INET6
2442         case AF_INET6:
2443                 if (((a->addr32[0] & m->addr32[0]) ==
2444                      (b->addr32[0] & m->addr32[0])) &&
2445                     ((a->addr32[1] & m->addr32[1]) ==
2446                      (b->addr32[1] & m->addr32[1])) &&
2447                     ((a->addr32[2] & m->addr32[2]) ==
2448                      (b->addr32[2] & m->addr32[2])) &&
2449                     ((a->addr32[3] & m->addr32[3]) ==
2450                      (b->addr32[3] & m->addr32[3])))
2451                         match++;
2452                 break;
2453 #endif /* INET6 */
2454         }
2455         if (match) {
2456                 if (n)
2457                         return (0);
2458                 else
2459                         return (1);
2460         } else {
2461                 if (n)
2462                         return (1);
2463                 else
2464                         return (0);
2465         }
2466 }
2467
2468 /*
2469  * Return 1 if b <= a <= e, otherwise return 0.
2470  */
2471 int
2472 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2473     struct pf_addr *a, sa_family_t af)
2474 {
2475         switch (af) {
2476 #ifdef INET
2477         case AF_INET:
2478                 if ((a->addr32[0] < b->addr32[0]) ||
2479                     (a->addr32[0] > e->addr32[0]))
2480                         return (0);
2481                 break;
2482 #endif /* INET */
2483 #ifdef INET6
2484         case AF_INET6: {
2485                 int     i;
2486
2487                 /* check a >= b */
2488                 for (i = 0; i < 4; ++i)
2489                         if (a->addr32[i] > b->addr32[i])
2490                                 break;
2491                         else if (a->addr32[i] < b->addr32[i])
2492                                 return (0);
2493                 /* check a <= e */
2494                 for (i = 0; i < 4; ++i)
2495                         if (a->addr32[i] < e->addr32[i])
2496                                 break;
2497                         else if (a->addr32[i] > e->addr32[i])
2498                                 return (0);
2499                 break;
2500         }
2501 #endif /* INET6 */
2502         }
2503         return (1);
2504 }
2505
2506 static int
2507 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2508 {
2509         switch (op) {
2510         case PF_OP_IRG:
2511                 return ((p > a1) && (p < a2));
2512         case PF_OP_XRG:
2513                 return ((p < a1) || (p > a2));
2514         case PF_OP_RRG:
2515                 return ((p >= a1) && (p <= a2));
2516         case PF_OP_EQ:
2517                 return (p == a1);
2518         case PF_OP_NE:
2519                 return (p != a1);
2520         case PF_OP_LT:
2521                 return (p < a1);
2522         case PF_OP_LE:
2523                 return (p <= a1);
2524         case PF_OP_GT:
2525                 return (p > a1);
2526         case PF_OP_GE:
2527                 return (p >= a1);
2528         }
2529         return (0); /* never reached */
2530 }
2531
2532 int
2533 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2534 {
2535         NTOHS(a1);
2536         NTOHS(a2);
2537         NTOHS(p);
2538         return (pf_match(op, a1, a2, p));
2539 }
2540
2541 static int
2542 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2543 {
2544         if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2545                 return (0);
2546         return (pf_match(op, a1, a2, u));
2547 }
2548
2549 static int
2550 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2551 {
2552         if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2553                 return (0);
2554         return (pf_match(op, a1, a2, g));
2555 }
2556
2557 int
2558 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2559 {
2560         if (*tag == -1)
2561                 *tag = mtag;
2562
2563         return ((!r->match_tag_not && r->match_tag == *tag) ||
2564             (r->match_tag_not && r->match_tag != *tag));
2565 }
2566
2567 int
2568 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2569 {
2570
2571         KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2572
2573         if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2574                 return (ENOMEM);
2575
2576         pd->pf_mtag->tag = tag;
2577
2578         return (0);
2579 }
2580
2581 #define PF_ANCHOR_STACKSIZE     32
2582 struct pf_anchor_stackframe {
2583         struct pf_ruleset       *rs;
2584         struct pf_rule          *r;     /* XXX: + match bit */
2585         struct pf_anchor        *child;
2586 };
2587
2588 /*
2589  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2590  */
2591 #define PF_ANCHORSTACK_MATCH    0x00000001
2592 #define PF_ANCHORSTACK_MASK     (PF_ANCHORSTACK_MATCH)
2593
2594 #define PF_ANCHOR_MATCH(f)      ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2595 #define PF_ANCHOR_RULE(f)       (struct pf_rule *)                      \
2596                                 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2597 #define PF_ANCHOR_SET_MATCH(f)  do { (f)->r = (void *)                  \
2598                                 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2599 } while (0)
2600
2601 void
2602 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2603     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2604     int *match)
2605 {
2606         struct pf_anchor_stackframe     *f;
2607
2608         PF_RULES_RASSERT();
2609
2610         if (match)
2611                 *match = 0;
2612         if (*depth >= PF_ANCHOR_STACKSIZE) {
2613                 printf("%s: anchor stack overflow on %s\n",
2614                     __func__, (*r)->anchor->name);
2615                 *r = TAILQ_NEXT(*r, entries);
2616                 return;
2617         } else if (*depth == 0 && a != NULL)
2618                 *a = *r;
2619         f = stack + (*depth)++;
2620         f->rs = *rs;
2621         f->r = *r;
2622         if ((*r)->anchor_wildcard) {
2623                 struct pf_anchor_node *parent = &(*r)->anchor->children;
2624
2625                 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2626                         *r = NULL;
2627                         return;
2628                 }
2629                 *rs = &f->child->ruleset;
2630         } else {
2631                 f->child = NULL;
2632                 *rs = &(*r)->anchor->ruleset;
2633         }
2634         *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2635 }
2636
2637 int
2638 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2639     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2640     int *match)
2641 {
2642         struct pf_anchor_stackframe     *f;
2643         struct pf_rule *fr;
2644         int quick = 0;
2645
2646         PF_RULES_RASSERT();
2647
2648         do {
2649                 if (*depth <= 0)
2650                         break;
2651                 f = stack + *depth - 1;
2652                 fr = PF_ANCHOR_RULE(f);
2653                 if (f->child != NULL) {
2654                         struct pf_anchor_node *parent;
2655
2656                         /*
2657                          * This block traverses through
2658                          * a wildcard anchor.
2659                          */
2660                         parent = &fr->anchor->children;
2661                         if (match != NULL && *match) {
2662                                 /*
2663                                  * If any of "*" matched, then
2664                                  * "foo/ *" matched, mark frame
2665                                  * appropriately.
2666                                  */
2667                                 PF_ANCHOR_SET_MATCH(f);
2668                                 *match = 0;
2669                         }
2670                         f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2671                         if (f->child != NULL) {
2672                                 *rs = &f->child->ruleset;
2673                                 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2674                                 if (*r == NULL)
2675                                         continue;
2676                                 else
2677                                         break;
2678                         }
2679                 }
2680                 (*depth)--;
2681                 if (*depth == 0 && a != NULL)
2682                         *a = NULL;
2683                 *rs = f->rs;
2684                 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2685                         quick = fr->quick;
2686                 *r = TAILQ_NEXT(fr, entries);
2687         } while (*r == NULL);
2688
2689         return (quick);
2690 }
2691
2692 #ifdef INET6
2693 void
2694 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2695     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2696 {
2697         switch (af) {
2698 #ifdef INET
2699         case AF_INET:
2700                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2701                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2702                 break;
2703 #endif /* INET */
2704         case AF_INET6:
2705                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2706                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2707                 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2708                 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2709                 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2710                 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2711                 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2712                 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2713                 break;
2714         }
2715 }
2716
2717 void
2718 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2719 {
2720         switch (af) {
2721 #ifdef INET
2722         case AF_INET:
2723                 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2724                 break;
2725 #endif /* INET */
2726         case AF_INET6:
2727                 if (addr->addr32[3] == 0xffffffff) {
2728                         addr->addr32[3] = 0;
2729                         if (addr->addr32[2] == 0xffffffff) {
2730                                 addr->addr32[2] = 0;
2731                                 if (addr->addr32[1] == 0xffffffff) {
2732                                         addr->addr32[1] = 0;
2733                                         addr->addr32[0] =
2734                                             htonl(ntohl(addr->addr32[0]) + 1);
2735                                 } else
2736                                         addr->addr32[1] =
2737                                             htonl(ntohl(addr->addr32[1]) + 1);
2738                         } else
2739                                 addr->addr32[2] =
2740                                     htonl(ntohl(addr->addr32[2]) + 1);
2741                 } else
2742                         addr->addr32[3] =
2743                             htonl(ntohl(addr->addr32[3]) + 1);
2744                 break;
2745         }
2746 }
2747 #endif /* INET6 */
2748
2749 int
2750 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
2751 {
2752         struct pf_addr          *saddr, *daddr;
2753         u_int16_t                sport, dport;
2754         struct inpcbinfo        *pi;
2755         struct inpcb            *inp;
2756
2757         pd->lookup.uid = UID_MAX;
2758         pd->lookup.gid = GID_MAX;
2759
2760         switch (pd->proto) {
2761         case IPPROTO_TCP:
2762                 if (pd->hdr.tcp == NULL)
2763                         return (-1);
2764                 sport = pd->hdr.tcp->th_sport;
2765                 dport = pd->hdr.tcp->th_dport;
2766                 pi = &V_tcbinfo;
2767                 break;
2768         case IPPROTO_UDP:
2769                 if (pd->hdr.udp == NULL)
2770                         return (-1);
2771                 sport = pd->hdr.udp->uh_sport;
2772                 dport = pd->hdr.udp->uh_dport;
2773                 pi = &V_udbinfo;
2774                 break;
2775         default:
2776                 return (-1);
2777         }
2778         if (direction == PF_IN) {
2779                 saddr = pd->src;
2780                 daddr = pd->dst;
2781         } else {
2782                 u_int16_t       p;
2783
2784                 p = sport;
2785                 sport = dport;
2786                 dport = p;
2787                 saddr = pd->dst;
2788                 daddr = pd->src;
2789         }
2790         switch (pd->af) {
2791 #ifdef INET
2792         case AF_INET:
2793                 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
2794                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2795                 if (inp == NULL) {
2796                         inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
2797                            daddr->v4, dport, INPLOOKUP_WILDCARD |
2798                            INPLOOKUP_RLOCKPCB, NULL, m);
2799                         if (inp == NULL)
2800                                 return (-1);
2801                 }
2802                 break;
2803 #endif /* INET */
2804 #ifdef INET6
2805         case AF_INET6:
2806                 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
2807                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2808                 if (inp == NULL) {
2809                         inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
2810                             &daddr->v6, dport, INPLOOKUP_WILDCARD |
2811                             INPLOOKUP_RLOCKPCB, NULL, m);
2812                         if (inp == NULL)
2813                                 return (-1);
2814                 }
2815                 break;
2816 #endif /* INET6 */
2817
2818         default:
2819                 return (-1);
2820         }
2821         INP_RLOCK_ASSERT(inp);
2822         pd->lookup.uid = inp->inp_cred->cr_uid;
2823         pd->lookup.gid = inp->inp_cred->cr_groups[0];
2824         INP_RUNLOCK(inp);
2825
2826         return (1);
2827 }
2828
2829 static u_int8_t
2830 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2831 {
2832         int              hlen;
2833         u_int8_t         hdr[60];
2834         u_int8_t        *opt, optlen;
2835         u_int8_t         wscale = 0;
2836
2837         hlen = th_off << 2;             /* hlen <= sizeof(hdr) */
2838         if (hlen <= sizeof(struct tcphdr))
2839                 return (0);
2840         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2841                 return (0);
2842         opt = hdr + sizeof(struct tcphdr);
2843         hlen -= sizeof(struct tcphdr);
2844         while (hlen >= 3) {
2845                 switch (*opt) {
2846                 case TCPOPT_EOL:
2847                 case TCPOPT_NOP:
2848                         ++opt;
2849                         --hlen;
2850                         break;
2851                 case TCPOPT_WINDOW:
2852                         wscale = opt[2];
2853                         if (wscale > TCP_MAX_WINSHIFT)
2854                                 wscale = TCP_MAX_WINSHIFT;
2855                         wscale |= PF_WSCALE_FLAG;
2856                         /* FALLTHROUGH */
2857                 default:
2858                         optlen = opt[1];
2859                         if (optlen < 2)
2860                                 optlen = 2;
2861                         hlen -= optlen;
2862                         opt += optlen;
2863                         break;
2864                 }
2865         }
2866         return (wscale);
2867 }
2868
2869 static u_int16_t
2870 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2871 {
2872         int              hlen;
2873         u_int8_t         hdr[60];
2874         u_int8_t        *opt, optlen;
2875         u_int16_t        mss = V_tcp_mssdflt;
2876
2877         hlen = th_off << 2;     /* hlen <= sizeof(hdr) */
2878         if (hlen <= sizeof(struct tcphdr))
2879                 return (0);
2880         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2881                 return (0);
2882         opt = hdr + sizeof(struct tcphdr);
2883         hlen -= sizeof(struct tcphdr);
2884         while (hlen >= TCPOLEN_MAXSEG) {
2885                 switch (*opt) {
2886                 case TCPOPT_EOL:
2887                 case TCPOPT_NOP:
2888                         ++opt;
2889                         --hlen;
2890                         break;
2891                 case TCPOPT_MAXSEG:
2892                         bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
2893                         NTOHS(mss);
2894                         /* FALLTHROUGH */
2895                 default:
2896                         optlen = opt[1];
2897                         if (optlen < 2)
2898                                 optlen = 2;
2899                         hlen -= optlen;
2900                         opt += optlen;
2901                         break;
2902                 }
2903         }
2904         return (mss);
2905 }
2906
2907 static u_int16_t
2908 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
2909 {
2910 #ifdef INET
2911         struct sockaddr_in      *dst;
2912         struct route             ro;
2913 #endif /* INET */
2914 #ifdef INET6
2915         struct sockaddr_in6     *dst6;
2916         struct route_in6         ro6;
2917 #endif /* INET6 */
2918         struct rtentry          *rt = NULL;
2919         int                      hlen = 0;
2920         u_int16_t                mss = V_tcp_mssdflt;
2921
2922         switch (af) {
2923 #ifdef INET
2924         case AF_INET:
2925                 hlen = sizeof(struct ip);
2926                 bzero(&ro, sizeof(ro));
2927                 dst = (struct sockaddr_in *)&ro.ro_dst;
2928                 dst->sin_family = AF_INET;
2929                 dst->sin_len = sizeof(*dst);
2930                 dst->sin_addr = addr->v4;
2931                 in_rtalloc_ign(&ro, 0, rtableid);
2932                 rt = ro.ro_rt;
2933                 break;
2934 #endif /* INET */
2935 #ifdef INET6
2936         case AF_INET6:
2937                 hlen = sizeof(struct ip6_hdr);
2938                 bzero(&ro6, sizeof(ro6));
2939                 dst6 = (struct sockaddr_in6 *)&ro6.ro_dst;
2940                 dst6->sin6_family = AF_INET6;
2941                 dst6->sin6_len = sizeof(*dst6);
2942                 dst6->sin6_addr = addr->v6;
2943                 in6_rtalloc_ign(&ro6, 0, rtableid);
2944                 rt = ro6.ro_rt;
2945                 break;
2946 #endif /* INET6 */
2947         }
2948
2949         if (rt && rt->rt_ifp) {
2950                 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr);
2951                 mss = max(V_tcp_mssdflt, mss);
2952                 RTFREE(rt);
2953         }
2954         mss = min(mss, offer);
2955         mss = max(mss, 64);             /* sanity - at least max opt space */
2956         return (mss);
2957 }
2958
2959 static u_int32_t
2960 pf_tcp_iss(struct pf_pdesc *pd)
2961 {
2962         MD5_CTX ctx;
2963         u_int32_t digest[4];
2964
2965         if (V_pf_tcp_secret_init == 0) {
2966                 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
2967                 MD5Init(&V_pf_tcp_secret_ctx);
2968                 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
2969                     sizeof(V_pf_tcp_secret));
2970                 V_pf_tcp_secret_init = 1;
2971         }
2972
2973         ctx = V_pf_tcp_secret_ctx;
2974
2975         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
2976         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
2977         if (pd->af == AF_INET6) {
2978                 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
2979                 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
2980         } else {
2981                 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
2982                 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
2983         }
2984         MD5Final((u_char *)digest, &ctx);
2985         V_pf_tcp_iss_off += 4096;
2986 #define ISN_RANDOM_INCREMENT (4096 - 1)
2987         return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
2988             V_pf_tcp_iss_off);
2989 #undef  ISN_RANDOM_INCREMENT
2990 }
2991
2992 static int
2993 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
2994     struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
2995     struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
2996 {
2997         struct pf_rule          *nr = NULL;
2998         struct pf_addr          * const saddr = pd->src;
2999         struct pf_addr          * const daddr = pd->dst;
3000         sa_family_t              af = pd->af;
3001         struct pf_rule          *r, *a = NULL;
3002         struct pf_ruleset       *ruleset = NULL;
3003         struct pf_src_node      *nsn = NULL;
3004         struct tcphdr           *th = pd->hdr.tcp;
3005         struct pf_state_key     *sk = NULL, *nk = NULL;
3006         u_short                  reason;
3007         int                      rewrite = 0, hdrlen = 0;
3008         int                      tag = -1, rtableid = -1;
3009         int                      asd = 0;
3010         int                      match = 0;
3011         int                      state_icmp = 0;
3012         u_int16_t                sport = 0, dport = 0;
3013         u_int16_t                bproto_sum = 0, bip_sum = 0;
3014         u_int8_t                 icmptype = 0, icmpcode = 0;
3015         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3016
3017         PF_RULES_RASSERT();
3018
3019         if (inp != NULL) {
3020                 INP_LOCK_ASSERT(inp);
3021                 pd->lookup.uid = inp->inp_cred->cr_uid;
3022                 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3023                 pd->lookup.done = 1;
3024         }
3025
3026         switch (pd->proto) {
3027         case IPPROTO_TCP:
3028                 sport = th->th_sport;
3029                 dport = th->th_dport;
3030                 hdrlen = sizeof(*th);
3031                 break;
3032         case IPPROTO_UDP:
3033                 sport = pd->hdr.udp->uh_sport;
3034                 dport = pd->hdr.udp->uh_dport;
3035                 hdrlen = sizeof(*pd->hdr.udp);
3036                 break;
3037 #ifdef INET
3038         case IPPROTO_ICMP:
3039                 if (pd->af != AF_INET)
3040                         break;
3041                 sport = dport = pd->hdr.icmp->icmp_id;
3042                 hdrlen = sizeof(*pd->hdr.icmp);
3043                 icmptype = pd->hdr.icmp->icmp_type;
3044                 icmpcode = pd->hdr.icmp->icmp_code;
3045
3046                 if (icmptype == ICMP_UNREACH ||
3047                     icmptype == ICMP_SOURCEQUENCH ||
3048                     icmptype == ICMP_REDIRECT ||
3049                     icmptype == ICMP_TIMXCEED ||
3050                     icmptype == ICMP_PARAMPROB)
3051                         state_icmp++;
3052                 break;
3053 #endif /* INET */
3054 #ifdef INET6
3055         case IPPROTO_ICMPV6:
3056                 if (af != AF_INET6)
3057                         break;
3058                 sport = dport = pd->hdr.icmp6->icmp6_id;
3059                 hdrlen = sizeof(*pd->hdr.icmp6);
3060                 icmptype = pd->hdr.icmp6->icmp6_type;
3061                 icmpcode = pd->hdr.icmp6->icmp6_code;
3062
3063                 if (icmptype == ICMP6_DST_UNREACH ||
3064                     icmptype == ICMP6_PACKET_TOO_BIG ||
3065                     icmptype == ICMP6_TIME_EXCEEDED ||
3066                     icmptype == ICMP6_PARAM_PROB)
3067                         state_icmp++;
3068                 break;
3069 #endif /* INET6 */
3070         default:
3071                 sport = dport = hdrlen = 0;
3072                 break;
3073         }
3074
3075         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3076
3077         /* check packet for BINAT/NAT/RDR */
3078         if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3079             &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3080                 KASSERT(sk != NULL, ("%s: null sk", __func__));
3081                 KASSERT(nk != NULL, ("%s: null nk", __func__));
3082
3083                 if (pd->ip_sum)
3084                         bip_sum = *pd->ip_sum;
3085
3086                 switch (pd->proto) {
3087                 case IPPROTO_TCP:
3088                         bproto_sum = th->th_sum;
3089                         pd->proto_sum = &th->th_sum;
3090
3091                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3092                             nk->port[pd->sidx] != sport) {
3093                                 pf_change_ap(saddr, &th->th_sport, pd->ip_sum,
3094                                     &th->th_sum, &nk->addr[pd->sidx],
3095                                     nk->port[pd->sidx], 0, af);
3096                                 pd->sport = &th->th_sport;
3097                                 sport = th->th_sport;
3098                         }
3099
3100                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3101                             nk->port[pd->didx] != dport) {
3102                                 pf_change_ap(daddr, &th->th_dport, pd->ip_sum,
3103                                     &th->th_sum, &nk->addr[pd->didx],
3104                                     nk->port[pd->didx], 0, af);
3105                                 dport = th->th_dport;
3106                                 pd->dport = &th->th_dport;
3107                         }
3108                         rewrite++;
3109                         break;
3110                 case IPPROTO_UDP:
3111                         bproto_sum = pd->hdr.udp->uh_sum;
3112                         pd->proto_sum = &pd->hdr.udp->uh_sum;
3113
3114                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3115                             nk->port[pd->sidx] != sport) {
3116                                 pf_change_ap(saddr, &pd->hdr.udp->uh_sport,
3117                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3118                                     &nk->addr[pd->sidx],
3119                                     nk->port[pd->sidx], 1, af);
3120                                 sport = pd->hdr.udp->uh_sport;
3121                                 pd->sport = &pd->hdr.udp->uh_sport;
3122                         }
3123
3124                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3125                             nk->port[pd->didx] != dport) {
3126                                 pf_change_ap(daddr, &pd->hdr.udp->uh_dport,
3127                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3128                                     &nk->addr[pd->didx],
3129                                     nk->port[pd->didx], 1, af);
3130                                 dport = pd->hdr.udp->uh_dport;
3131                                 pd->dport = &pd->hdr.udp->uh_dport;
3132                         }
3133                         rewrite++;
3134                         break;
3135 #ifdef INET
3136                 case IPPROTO_ICMP:
3137                         nk->port[0] = nk->port[1];
3138                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3139                                 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3140                                     nk->addr[pd->sidx].v4.s_addr, 0);
3141
3142                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3143                                 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3144                                     nk->addr[pd->didx].v4.s_addr, 0);
3145
3146                         if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3147                                 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3148                                     pd->hdr.icmp->icmp_cksum, sport,
3149                                     nk->port[1], 0);
3150                                 pd->hdr.icmp->icmp_id = nk->port[1];
3151                                 pd->sport = &pd->hdr.icmp->icmp_id;
3152                         }
3153                         m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3154                         break;
3155 #endif /* INET */
3156 #ifdef INET6
3157                 case IPPROTO_ICMPV6:
3158                         nk->port[0] = nk->port[1];
3159                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3160                                 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3161                                     &nk->addr[pd->sidx], 0);
3162
3163                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3164                                 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3165                                     &nk->addr[pd->didx], 0);
3166                         rewrite++;
3167                         break;
3168 #endif /* INET */
3169                 default:
3170                         switch (af) {
3171 #ifdef INET
3172                         case AF_INET:
3173                                 if (PF_ANEQ(saddr,
3174                                     &nk->addr[pd->sidx], AF_INET))
3175                                         pf_change_a(&saddr->v4.s_addr,
3176                                             pd->ip_sum,
3177                                             nk->addr[pd->sidx].v4.s_addr, 0);
3178
3179                                 if (PF_ANEQ(daddr,
3180                                     &nk->addr[pd->didx], AF_INET))
3181                                         pf_change_a(&daddr->v4.s_addr,
3182                                             pd->ip_sum,
3183                                             nk->addr[pd->didx].v4.s_addr, 0);
3184                                 break;
3185 #endif /* INET */
3186 #ifdef INET6
3187                         case AF_INET6:
3188                                 if (PF_ANEQ(saddr,
3189                                     &nk->addr[pd->sidx], AF_INET6))
3190                                         PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3191
3192                                 if (PF_ANEQ(daddr,
3193                                     &nk->addr[pd->didx], AF_INET6))
3194                                         PF_ACPY(saddr, &nk->addr[pd->didx], af);
3195                                 break;
3196 #endif /* INET */
3197                         }
3198                         break;
3199                 }
3200                 if (nr->natpass)
3201                         r = NULL;
3202                 pd->nat_rule = nr;
3203         }
3204
3205         while (r != NULL) {
3206                 r->evaluations++;
3207                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3208                         r = r->skip[PF_SKIP_IFP].ptr;
3209                 else if (r->direction && r->direction != direction)
3210                         r = r->skip[PF_SKIP_DIR].ptr;
3211                 else if (r->af && r->af != af)
3212                         r = r->skip[PF_SKIP_AF].ptr;
3213                 else if (r->proto && r->proto != pd->proto)
3214                         r = r->skip[PF_SKIP_PROTO].ptr;
3215                 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3216                     r->src.neg, kif, M_GETFIB(m)))
3217                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3218                 /* tcp/udp only. port_op always 0 in other cases */
3219                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3220                     r->src.port[0], r->src.port[1], sport))
3221                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
3222                 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3223                     r->dst.neg, NULL, M_GETFIB(m)))
3224                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3225                 /* tcp/udp only. port_op always 0 in other cases */
3226                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3227                     r->dst.port[0], r->dst.port[1], dport))
3228                         r = r->skip[PF_SKIP_DST_PORT].ptr;
3229                 /* icmp only. type always 0 in other cases */
3230                 else if (r->type && r->type != icmptype + 1)
3231                         r = TAILQ_NEXT(r, entries);
3232                 /* icmp only. type always 0 in other cases */
3233                 else if (r->code && r->code != icmpcode + 1)
3234                         r = TAILQ_NEXT(r, entries);
3235                 else if (r->tos && !(r->tos == pd->tos))
3236                         r = TAILQ_NEXT(r, entries);
3237                 else if (r->rule_flag & PFRULE_FRAGMENT)
3238                         r = TAILQ_NEXT(r, entries);
3239                 else if (pd->proto == IPPROTO_TCP &&
3240                     (r->flagset & th->th_flags) != r->flags)
3241                         r = TAILQ_NEXT(r, entries);
3242                 /* tcp/udp only. uid.op always 0 in other cases */
3243                 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3244                     pf_socket_lookup(direction, pd, m), 1)) &&
3245                     !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3246                     pd->lookup.uid))
3247                         r = TAILQ_NEXT(r, entries);
3248                 /* tcp/udp only. gid.op always 0 in other cases */
3249                 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3250                     pf_socket_lookup(direction, pd, m), 1)) &&
3251                     !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3252                     pd->lookup.gid))
3253                         r = TAILQ_NEXT(r, entries);
3254                 else if (r->prob &&
3255                     r->prob <= arc4random())
3256                         r = TAILQ_NEXT(r, entries);
3257                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3258                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3259                         r = TAILQ_NEXT(r, entries);
3260                 else if (r->os_fingerprint != PF_OSFP_ANY &&
3261                     (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3262                     pf_osfp_fingerprint(pd, m, off, th),
3263                     r->os_fingerprint)))
3264                         r = TAILQ_NEXT(r, entries);
3265                 else {
3266                         if (r->tag)
3267                                 tag = r->tag;
3268                         if (r->rtableid >= 0)
3269                                 rtableid = r->rtableid;
3270                         if (r->anchor == NULL) {
3271                                 match = 1;
3272                                 *rm = r;
3273                                 *am = a;
3274                                 *rsm = ruleset;
3275                                 if ((*rm)->quick)
3276                                         break;
3277                                 r = TAILQ_NEXT(r, entries);
3278                         } else
3279                                 pf_step_into_anchor(anchor_stack, &asd,
3280                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3281                                     &match);
3282                 }
3283                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3284                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3285                         break;
3286         }
3287         r = *rm;
3288         a = *am;
3289         ruleset = *rsm;
3290
3291         REASON_SET(&reason, PFRES_MATCH);
3292
3293         if (r->log || (nr != NULL && nr->log)) {
3294                 if (rewrite)
3295                         m_copyback(m, off, hdrlen, pd->hdr.any);
3296                 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3297                     ruleset, pd, 1);
3298         }
3299
3300         if ((r->action == PF_DROP) &&
3301             ((r->rule_flag & PFRULE_RETURNRST) ||
3302             (r->rule_flag & PFRULE_RETURNICMP) ||
3303             (r->rule_flag & PFRULE_RETURN))) {
3304                 /* undo NAT changes, if they have taken place */
3305                 if (nr != NULL) {
3306                         PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3307                         PF_ACPY(daddr, &sk->addr[pd->didx], af);
3308                         if (pd->sport)
3309                                 *pd->sport = sk->port[pd->sidx];
3310                         if (pd->dport)
3311                                 *pd->dport = sk->port[pd->didx];
3312                         if (pd->proto_sum)
3313                                 *pd->proto_sum = bproto_sum;
3314                         if (pd->ip_sum)
3315                                 *pd->ip_sum = bip_sum;
3316                         m_copyback(m, off, hdrlen, pd->hdr.any);
3317                 }
3318                 if (pd->proto == IPPROTO_TCP &&
3319                     ((r->rule_flag & PFRULE_RETURNRST) ||
3320                     (r->rule_flag & PFRULE_RETURN)) &&
3321                     !(th->th_flags & TH_RST)) {
3322                         u_int32_t        ack = ntohl(th->th_seq) + pd->p_len;
3323                         int              len = 0;
3324 #ifdef INET
3325                         struct ip       *h4;
3326 #endif
3327 #ifdef INET6
3328                         struct ip6_hdr  *h6;
3329 #endif
3330
3331                         switch (af) {
3332 #ifdef INET
3333                         case AF_INET:
3334                                 h4 = mtod(m, struct ip *);
3335                                 len = ntohs(h4->ip_len) - off;
3336                                 break;
3337 #endif
3338 #ifdef INET6
3339                         case AF_INET6:
3340                                 h6 = mtod(m, struct ip6_hdr *);
3341                                 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3342                                 break;
3343 #endif
3344                         }
3345
3346                         if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3347                                 REASON_SET(&reason, PFRES_PROTCKSUM);
3348                         else {
3349                                 if (th->th_flags & TH_SYN)
3350                                         ack++;
3351                                 if (th->th_flags & TH_FIN)
3352                                         ack++;
3353                                 pf_send_tcp(m, r, af, pd->dst,
3354                                     pd->src, th->th_dport, th->th_sport,
3355                                     ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3356                                     r->return_ttl, 1, 0, kif->pfik_ifp);
3357                         }
3358                 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3359                     r->return_icmp)
3360                         pf_send_icmp(m, r->return_icmp >> 8,
3361                             r->return_icmp & 255, af, r);
3362                 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3363                     r->return_icmp6)
3364                         pf_send_icmp(m, r->return_icmp6 >> 8,
3365                             r->return_icmp6 & 255, af, r);
3366         }
3367
3368         if (r->action == PF_DROP)
3369                 goto cleanup;
3370
3371         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3372                 REASON_SET(&reason, PFRES_MEMORY);
3373                 goto cleanup;
3374         }
3375         if (rtableid >= 0)
3376                 M_SETFIB(m, rtableid);
3377
3378         if (!state_icmp && (r->keep_state || nr != NULL ||
3379             (pd->flags & PFDESC_TCP_NORM))) {
3380                 int action;
3381                 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3382                     sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3383                     hdrlen);
3384                 if (action != PF_PASS)
3385                         return (action);
3386         } else {
3387                 if (sk != NULL)
3388                         uma_zfree(V_pf_state_key_z, sk);
3389                 if (nk != NULL)
3390                         uma_zfree(V_pf_state_key_z, nk);
3391         }
3392
3393         /* copy back packet headers if we performed NAT operations */
3394         if (rewrite)
3395                 m_copyback(m, off, hdrlen, pd->hdr.any);
3396
3397         if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3398             direction == PF_OUT &&
3399             pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m))
3400                 /*
3401                  * We want the state created, but we dont
3402                  * want to send this in case a partner
3403                  * firewall has to know about it to allow
3404                  * replies through it.
3405                  */
3406                 return (PF_DEFER);
3407
3408         return (PF_PASS);
3409
3410 cleanup:
3411         if (sk != NULL)
3412                 uma_zfree(V_pf_state_key_z, sk);
3413         if (nk != NULL)
3414                 uma_zfree(V_pf_state_key_z, nk);
3415         return (PF_DROP);
3416 }
3417
3418 static int
3419 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3420     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3421     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3422     u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3423     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3424 {
3425         struct pf_state         *s = NULL;
3426         struct pf_src_node      *sn = NULL;
3427         struct tcphdr           *th = pd->hdr.tcp;
3428         u_int16_t                mss = V_tcp_mssdflt;
3429         u_short                  reason;
3430
3431         /* check maximums */
3432         if (r->max_states &&
3433             (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3434                 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
3435                 REASON_SET(&reason, PFRES_MAXSTATES);
3436                 return (PF_DROP);
3437         }
3438         /* src node for filter rule */
3439         if ((r->rule_flag & PFRULE_SRCTRACK ||
3440             r->rpool.opts & PF_POOL_STICKYADDR) &&
3441             pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3442                 REASON_SET(&reason, PFRES_SRCLIMIT);
3443                 goto csfailed;
3444         }
3445         /* src node for translation rule */
3446         if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3447             pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3448                 REASON_SET(&reason, PFRES_SRCLIMIT);
3449                 goto csfailed;
3450         }
3451         s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3452         if (s == NULL) {
3453                 REASON_SET(&reason, PFRES_MEMORY);
3454                 goto csfailed;
3455         }
3456         s->rule.ptr = r;
3457         s->nat_rule.ptr = nr;
3458         s->anchor.ptr = a;
3459         STATE_INC_COUNTERS(s);
3460         if (r->allow_opts)
3461                 s->state_flags |= PFSTATE_ALLOWOPTS;
3462         if (r->rule_flag & PFRULE_STATESLOPPY)
3463                 s->state_flags |= PFSTATE_SLOPPY;
3464         s->log = r->log & PF_LOG_ALL;
3465         s->sync_state = PFSYNC_S_NONE;
3466         if (nr != NULL)
3467                 s->log |= nr->log & PF_LOG_ALL;
3468         switch (pd->proto) {
3469         case IPPROTO_TCP:
3470                 s->src.seqlo = ntohl(th->th_seq);
3471                 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3472                 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3473                     r->keep_state == PF_STATE_MODULATE) {
3474                         /* Generate sequence number modulator */
3475                         if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3476                             0)
3477                                 s->src.seqdiff = 1;
3478                         pf_change_a(&th->th_seq, &th->th_sum,
3479                             htonl(s->src.seqlo + s->src.seqdiff), 0);
3480                         *rewrite = 1;
3481                 } else
3482                         s->src.seqdiff = 0;
3483                 if (th->th_flags & TH_SYN) {
3484                         s->src.seqhi++;
3485                         s->src.wscale = pf_get_wscale(m, off,
3486                             th->th_off, pd->af);
3487                 }
3488                 s->src.max_win = MAX(ntohs(th->th_win), 1);
3489                 if (s->src.wscale & PF_WSCALE_MASK) {
3490                         /* Remove scale factor from initial window */
3491                         int win = s->src.max_win;
3492                         win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3493                         s->src.max_win = (win - 1) >>
3494                             (s->src.wscale & PF_WSCALE_MASK);
3495                 }
3496                 if (th->th_flags & TH_FIN)
3497                         s->src.seqhi++;
3498                 s->dst.seqhi = 1;
3499                 s->dst.max_win = 1;
3500                 s->src.state = TCPS_SYN_SENT;
3501                 s->dst.state = TCPS_CLOSED;
3502                 s->timeout = PFTM_TCP_FIRST_PACKET;
3503                 break;
3504         case IPPROTO_UDP:
3505                 s->src.state = PFUDPS_SINGLE;
3506                 s->dst.state = PFUDPS_NO_TRAFFIC;
3507                 s->timeout = PFTM_UDP_FIRST_PACKET;
3508                 break;
3509         case IPPROTO_ICMP:
3510 #ifdef INET6
3511         case IPPROTO_ICMPV6:
3512 #endif
3513                 s->timeout = PFTM_ICMP_FIRST_PACKET;
3514                 break;
3515         default:
3516                 s->src.state = PFOTHERS_SINGLE;
3517                 s->dst.state = PFOTHERS_NO_TRAFFIC;
3518                 s->timeout = PFTM_OTHER_FIRST_PACKET;
3519         }
3520
3521         if (r->rt && r->rt != PF_FASTROUTE) {
3522                 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) {
3523                         REASON_SET(&reason, PFRES_BADSTATE);
3524                         pf_src_tree_remove_state(s);
3525                         STATE_DEC_COUNTERS(s);
3526                         uma_zfree(V_pf_state_z, s);
3527                         goto csfailed;
3528                 }
3529                 s->rt_kif = r->rpool.cur->kif;
3530         }
3531
3532         s->creation = time_uptime;
3533         s->expire = time_uptime;
3534
3535         if (sn != NULL) {
3536                 s->src_node = sn;
3537                 s->src_node->states++;
3538         }
3539         if (nsn != NULL) {
3540                 /* XXX We only modify one side for now. */
3541                 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3542                 s->nat_src_node = nsn;
3543                 s->nat_src_node->states++;
3544         }
3545         if (pd->proto == IPPROTO_TCP) {
3546                 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3547                     off, pd, th, &s->src, &s->dst)) {
3548                         REASON_SET(&reason, PFRES_MEMORY);
3549                         pf_src_tree_remove_state(s);
3550                         STATE_DEC_COUNTERS(s);
3551                         uma_zfree(V_pf_state_z, s);
3552                         return (PF_DROP);
3553                 }
3554                 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3555                     pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3556                     &s->src, &s->dst, rewrite)) {
3557                         /* This really shouldn't happen!!! */
3558                         DPFPRINTF(PF_DEBUG_URGENT,
3559                             ("pf_normalize_tcp_stateful failed on first pkt"));
3560                         pf_normalize_tcp_cleanup(s);
3561                         pf_src_tree_remove_state(s);
3562                         STATE_DEC_COUNTERS(s);
3563                         uma_zfree(V_pf_state_z, s);
3564                         return (PF_DROP);
3565                 }
3566         }
3567         s->direction = pd->dir;
3568
3569         /*
3570          * sk/nk could already been setup by pf_get_translation().
3571          */
3572         if (nr == NULL) {
3573                 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3574                     __func__, nr, sk, nk));
3575                 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3576                 if (sk == NULL)
3577                         goto csfailed;
3578                 nk = sk;
3579         } else
3580                 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3581                     __func__, nr, sk, nk));
3582
3583         /* Swap sk/nk for PF_OUT. */
3584         if (pf_state_insert(BOUND_IFACE(r, kif),
3585             (pd->dir == PF_IN) ? sk : nk,
3586             (pd->dir == PF_IN) ? nk : sk, s)) {
3587                 if (pd->proto == IPPROTO_TCP)
3588                         pf_normalize_tcp_cleanup(s);
3589                 REASON_SET(&reason, PFRES_STATEINS);
3590                 pf_src_tree_remove_state(s);
3591                 STATE_DEC_COUNTERS(s);
3592                 uma_zfree(V_pf_state_z, s);
3593                 return (PF_DROP);
3594         } else
3595                 *sm = s;
3596
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                 counter_u64_add(rt->rt_pksent, 1);
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         case PF_DROP:
6005                 m_freem(*m0);
6006                 *m0 = NULL;
6007                 break;
6008         default:
6009                 /* pf_route() returns unlocked. */
6010                 if (r->rt) {
6011                         pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
6012                         return (action);
6013                 }
6014                 break;
6015         }
6016         if (s)
6017                 PF_STATE_UNLOCK(s);
6018
6019         return (action);
6020 }
6021 #endif /* INET */
6022
6023 #ifdef INET6
6024 int
6025 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6026 {
6027         struct pfi_kif          *kif;
6028         u_short                  action, reason = 0, log = 0;
6029         struct mbuf             *m = *m0, *n = NULL;
6030         struct ip6_hdr          *h = NULL;
6031         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6032         struct pf_state         *s = NULL;
6033         struct pf_ruleset       *ruleset = NULL;
6034         struct pf_pdesc          pd;
6035         int                      off, terminal = 0, dirndx, rh_cnt = 0;
6036
6037         M_ASSERTPKTHDR(m);
6038
6039         if (!V_pf_status.running)
6040                 return (PF_PASS);
6041
6042         memset(&pd, 0, sizeof(pd));
6043         pd.pf_mtag = pf_find_mtag(m);
6044
6045         if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6046                 return (PF_PASS);
6047
6048         kif = (struct pfi_kif *)ifp->if_pf_kif;
6049         if (kif == NULL) {
6050                 DPFPRINTF(PF_DEBUG_URGENT,
6051                     ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6052                 return (PF_DROP);
6053         }
6054         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6055                 return (PF_PASS);
6056
6057         if (m->m_flags & M_SKIP_FIREWALL)
6058                 return (PF_PASS);
6059
6060         PF_RULES_RLOCK();
6061
6062         /* We do IP header normalization and packet reassembly here */
6063         if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6064                 action = PF_DROP;
6065                 goto done;
6066         }
6067         m = *m0;        /* pf_normalize messes with m0 */
6068         h = mtod(m, struct ip6_hdr *);
6069
6070 #if 1
6071         /*
6072          * we do not support jumbogram yet.  if we keep going, zero ip6_plen
6073          * will do something bad, so drop the packet for now.
6074          */
6075         if (htons(h->ip6_plen) == 0) {
6076                 action = PF_DROP;
6077                 REASON_SET(&reason, PFRES_NORM);        /*XXX*/
6078                 goto done;
6079         }
6080 #endif
6081
6082         pd.src = (struct pf_addr *)&h->ip6_src;
6083         pd.dst = (struct pf_addr *)&h->ip6_dst;
6084         pd.sport = pd.dport = NULL;
6085         pd.ip_sum = NULL;
6086         pd.proto_sum = NULL;
6087         pd.dir = dir;
6088         pd.sidx = (dir == PF_IN) ? 0 : 1;
6089         pd.didx = (dir == PF_IN) ? 1 : 0;
6090         pd.af = AF_INET6;
6091         pd.tos = 0;
6092         pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6093
6094         off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6095         pd.proto = h->ip6_nxt;
6096         do {
6097                 switch (pd.proto) {
6098                 case IPPROTO_FRAGMENT:
6099                         action = pf_test_fragment(&r, dir, kif, m, h,
6100                             &pd, &a, &ruleset);
6101                         if (action == PF_DROP)
6102                                 REASON_SET(&reason, PFRES_FRAG);
6103                         goto done;
6104                 case IPPROTO_ROUTING: {
6105                         struct ip6_rthdr rthdr;
6106
6107                         if (rh_cnt++) {
6108                                 DPFPRINTF(PF_DEBUG_MISC,
6109                                     ("pf: IPv6 more than one rthdr\n"));
6110                                 action = PF_DROP;
6111                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6112                                 log = 1;
6113                                 goto done;
6114                         }
6115                         if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6116                             &reason, pd.af)) {
6117                                 DPFPRINTF(PF_DEBUG_MISC,
6118                                     ("pf: IPv6 short rthdr\n"));
6119                                 action = PF_DROP;
6120                                 REASON_SET(&reason, PFRES_SHORT);
6121                                 log = 1;
6122                                 goto done;
6123                         }
6124                         if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6125                                 DPFPRINTF(PF_DEBUG_MISC,
6126                                     ("pf: IPv6 rthdr0\n"));
6127                                 action = PF_DROP;
6128                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6129                                 log = 1;
6130                                 goto done;
6131                         }
6132                         /* FALLTHROUGH */
6133                 }
6134                 case IPPROTO_AH:
6135                 case IPPROTO_HOPOPTS:
6136                 case IPPROTO_DSTOPTS: {
6137                         /* get next header and header length */
6138                         struct ip6_ext  opt6;
6139
6140                         if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6141                             NULL, &reason, pd.af)) {
6142                                 DPFPRINTF(PF_DEBUG_MISC,
6143                                     ("pf: IPv6 short opt\n"));
6144                                 action = PF_DROP;
6145                                 log = 1;
6146                                 goto done;
6147                         }
6148                         if (pd.proto == IPPROTO_AH)
6149                                 off += (opt6.ip6e_len + 2) * 4;
6150                         else
6151                                 off += (opt6.ip6e_len + 1) * 8;
6152                         pd.proto = opt6.ip6e_nxt;
6153                         /* goto the next header */
6154                         break;
6155                 }
6156                 default:
6157                         terminal++;
6158                         break;
6159                 }
6160         } while (!terminal);
6161
6162         /* if there's no routing header, use unmodified mbuf for checksumming */
6163         if (!n)
6164                 n = m;
6165
6166         switch (pd.proto) {
6167
6168         case IPPROTO_TCP: {
6169                 struct tcphdr   th;
6170
6171                 pd.hdr.tcp = &th;
6172                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6173                     &action, &reason, AF_INET6)) {
6174                         log = action != PF_PASS;
6175                         goto done;
6176                 }
6177                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6178                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6179                 if (action == PF_DROP)
6180                         goto done;
6181                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6182                     &reason);
6183                 if (action == PF_PASS) {
6184                         if (pfsync_update_state_ptr != NULL)
6185                                 pfsync_update_state_ptr(s);
6186                         r = s->rule.ptr;
6187                         a = s->anchor.ptr;
6188                         log = s->log;
6189                 } else if (s == NULL)
6190                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6191                             &a, &ruleset, inp);
6192                 break;
6193         }
6194
6195         case IPPROTO_UDP: {
6196                 struct udphdr   uh;
6197
6198                 pd.hdr.udp = &uh;
6199                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6200                     &action, &reason, AF_INET6)) {
6201                         log = action != PF_PASS;
6202                         goto done;
6203                 }
6204                 if (uh.uh_dport == 0 ||
6205                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6206                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6207                         action = PF_DROP;
6208                         REASON_SET(&reason, PFRES_SHORT);
6209                         goto done;
6210                 }
6211                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6212                 if (action == PF_PASS) {
6213                         if (pfsync_update_state_ptr != NULL)
6214                                 pfsync_update_state_ptr(s);
6215                         r = s->rule.ptr;
6216                         a = s->anchor.ptr;
6217                         log = s->log;
6218                 } else if (s == NULL)
6219                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6220                             &a, &ruleset, inp);
6221                 break;
6222         }
6223
6224         case IPPROTO_ICMP: {
6225                 action = PF_DROP;
6226                 DPFPRINTF(PF_DEBUG_MISC,
6227                     ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6228                 goto done;
6229         }
6230
6231         case IPPROTO_ICMPV6: {
6232                 struct icmp6_hdr        ih;
6233
6234                 pd.hdr.icmp6 = &ih;
6235                 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6236                     &action, &reason, AF_INET6)) {
6237                         log = action != PF_PASS;
6238                         goto done;
6239                 }
6240                 action = pf_test_state_icmp(&s, dir, kif,
6241                     m, off, h, &pd, &reason);
6242                 if (action == PF_PASS) {
6243                         if (pfsync_update_state_ptr != NULL)
6244                                 pfsync_update_state_ptr(s);
6245                         r = s->rule.ptr;
6246                         a = s->anchor.ptr;
6247                         log = s->log;
6248                 } else if (s == NULL)
6249                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6250                             &a, &ruleset, inp);
6251                 break;
6252         }
6253
6254         default:
6255                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6256                 if (action == PF_PASS) {
6257                         if (pfsync_update_state_ptr != NULL)
6258                                 pfsync_update_state_ptr(s);
6259                         r = s->rule.ptr;
6260                         a = s->anchor.ptr;
6261                         log = s->log;
6262                 } else if (s == NULL)
6263                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6264                             &a, &ruleset, inp);
6265                 break;
6266         }
6267
6268 done:
6269         PF_RULES_RUNLOCK();
6270         if (n != m) {
6271                 m_freem(n);
6272                 n = NULL;
6273         }
6274
6275         /* handle dangerous IPv6 extension headers. */
6276         if (action == PF_PASS && rh_cnt &&
6277             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6278                 action = PF_DROP;
6279                 REASON_SET(&reason, PFRES_IPOPTIONS);
6280                 log = 1;
6281                 DPFPRINTF(PF_DEBUG_MISC,
6282                     ("pf: dropping packet with dangerous v6 headers\n"));
6283         }
6284
6285         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6286                 action = PF_DROP;
6287                 REASON_SET(&reason, PFRES_MEMORY);
6288         }
6289         if (r->rtableid >= 0)
6290                 M_SETFIB(m, r->rtableid);
6291
6292 #ifdef ALTQ
6293         if (action == PF_PASS && r->qid) {
6294                 if (pd.pf_mtag == NULL &&
6295                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6296                         action = PF_DROP;
6297                         REASON_SET(&reason, PFRES_MEMORY);
6298                 }
6299                 if (pd.tos & IPTOS_LOWDELAY)
6300                         pd.pf_mtag->qid = r->pqid;
6301                 else
6302                         pd.pf_mtag->qid = r->qid;
6303                 /* add hints for ecn */
6304                 pd.pf_mtag->hdr = h;
6305         }
6306 #endif /* ALTQ */
6307
6308         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6309             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6310             (s->nat_rule.ptr->action == PF_RDR ||
6311             s->nat_rule.ptr->action == PF_BINAT) &&
6312             IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6313                 m->m_flags |= M_SKIP_FIREWALL;
6314
6315         /* XXX: Anybody working on it?! */
6316         if (r->divert.port)
6317                 printf("pf: divert(9) is not supported for IPv6\n");
6318
6319         if (log) {
6320                 struct pf_rule *lr;
6321
6322                 if (s != NULL && s->nat_rule.ptr != NULL &&
6323                     s->nat_rule.ptr->log & PF_LOG_ALL)
6324                         lr = s->nat_rule.ptr;
6325                 else
6326                         lr = r;
6327                 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6328                     &pd, (s == NULL));
6329         }
6330
6331         kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6332         kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
6333
6334         if (action == PF_PASS || r->action == PF_DROP) {
6335                 dirndx = (dir == PF_OUT);
6336                 r->packets[dirndx]++;
6337                 r->bytes[dirndx] += pd.tot_len;
6338                 if (a != NULL) {
6339                         a->packets[dirndx]++;
6340                         a->bytes[dirndx] += pd.tot_len;
6341                 }
6342                 if (s != NULL) {
6343                         if (s->nat_rule.ptr != NULL) {
6344                                 s->nat_rule.ptr->packets[dirndx]++;
6345                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6346                         }
6347                         if (s->src_node != NULL) {
6348                                 s->src_node->packets[dirndx]++;
6349                                 s->src_node->bytes[dirndx] += pd.tot_len;
6350                         }
6351                         if (s->nat_src_node != NULL) {
6352                                 s->nat_src_node->packets[dirndx]++;
6353                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6354                         }
6355                         dirndx = (dir == s->direction) ? 0 : 1;
6356                         s->packets[dirndx]++;
6357                         s->bytes[dirndx] += pd.tot_len;
6358                 }
6359                 tr = r;
6360                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6361                 if (nr != NULL && r == &V_pf_default_rule)
6362                         tr = nr;
6363                 if (tr->src.addr.type == PF_ADDR_TABLE)
6364                         pfr_update_stats(tr->src.addr.p.tbl,
6365                             (s == NULL) ? pd.src :
6366                             &s->key[(s->direction == PF_IN)]->addr[0],
6367                             pd.af, pd.tot_len, dir == PF_OUT,
6368                             r->action == PF_PASS, tr->src.neg);
6369                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6370                         pfr_update_stats(tr->dst.addr.p.tbl,
6371                             (s == NULL) ? pd.dst :
6372                             &s->key[(s->direction == PF_IN)]->addr[1],
6373                             pd.af, pd.tot_len, dir == PF_OUT,
6374                             r->action == PF_PASS, tr->dst.neg);
6375         }
6376
6377         switch (action) {
6378         case PF_SYNPROXY_DROP:
6379                 m_freem(*m0);
6380         case PF_DEFER:
6381                 *m0 = NULL;
6382                 action = PF_PASS;
6383                 break;
6384         case PF_DROP:
6385                 m_freem(*m0);
6386                 *m0 = NULL;
6387                 break;
6388         default:
6389                 /* pf_route6() returns unlocked. */
6390                 if (r->rt) {
6391                         pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
6392                         return (action);
6393                 }
6394                 break;
6395         }
6396
6397         if (s)
6398                 PF_STATE_UNLOCK(s);
6399
6400         return (action);
6401 }
6402 #endif /* INET6 */