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(u_long, pf_hashmask);
 353 VNET_DEFINE(struct pf_srchash *, pf_srchash);
 354 VNET_DEFINE(u_long, pf_srchashmask);
 355
 356 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)");
 357
 358 VNET_DEFINE(u_long, pf_hashsize);
 359 #define V_pf_hashsize   VNET(pf_hashsize)
 360 SYSCTL_VNET_UINT(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
 361     &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable");
 362
 363 VNET_DEFINE(u_long, pf_srchashsize);
 364 #define V_pf_srchashsize        VNET(pf_srchashsize)
 365 SYSCTL_VNET_UINT(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
 366     &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable");
 367
 368 VNET_DEFINE(void *, pf_swi_cookie);
 369
 370 VNET_DEFINE(uint32_t, pf_hashseed);
 371 #define V_pf_hashseed   VNET(pf_hashseed)
 372
 373 static __inline uint32_t
 374 pf_hashkey(struct pf_state_key *sk)
 375 {
 376         uint32_t h;
 377
 378         h = jenkins_hash32((uint32_t *)sk,
 379             sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
 380             V_pf_hashseed);
 381
 382         return (h & V_pf_hashmask);
 383 }
 384
 385 static __inline uint32_t
 386 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
 387 {
 388         uint32_t h;
 389
 390         switch (af) {
 391         case AF_INET:
 392                 h = jenkins_hash32((uint32_t *)&addr->v4,
 393                     sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
 394                 break;
 395         case AF_INET6:
 396                 h = jenkins_hash32((uint32_t *)&addr->v6,
 397                     sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
 398                 break;
 399         default:
 400                 panic("%s: unknown address family %u", __func__, af);
 401         }
 402
 403         return (h & V_pf_srchashmask);
 404 }
 405
 406 #ifdef INET6
 407 void
 408 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
 409 {
 410         switch (af) {
 411 #ifdef INET
 412         case AF_INET:
 413                 dst->addr32[0] = src->addr32[0];
 414                 break;
 415 #endif /* INET */
 416         case AF_INET6:
 417                 dst->addr32[0] = src->addr32[0];
 418                 dst->addr32[1] = src->addr32[1];
 419                 dst->addr32[2] = src->addr32[2];
 420                 dst->addr32[3] = src->addr32[3];
 421                 break;
 422         }
 423 }
 424 #endif /* INET6 */
 425
 426 static void
 427 pf_init_threshold(struct pf_threshold *threshold,
 428     u_int32_t limit, u_int32_t seconds)
 429 {
 430         threshold->limit = limit * PF_THRESHOLD_MULT;
 431         threshold->seconds = seconds;
 432         threshold->count = 0;
 433         threshold->last = time_uptime;
 434 }
 435
 436 static void
 437 pf_add_threshold(struct pf_threshold *threshold)
 438 {
 439         u_int32_t t = time_uptime, diff = t - threshold->last;
 440
 441         if (diff >= threshold->seconds)
 442                 threshold->count = 0;
 443         else
 444                 threshold->count -= threshold->count * diff /
 445                     threshold->seconds;
 446         threshold->count += PF_THRESHOLD_MULT;
 447         threshold->last = t;
 448 }
 449
 450 static int
 451 pf_check_threshold(struct pf_threshold *threshold)
 452 {
 453         return (threshold->count > threshold->limit);
 454 }
 455
 456 static int
 457 pf_src_connlimit(struct pf_state **state)
 458 {
 459         struct pf_overload_entry *pfoe;
 460         int bad = 0;
 461
 462         PF_STATE_LOCK_ASSERT(*state);
 463
 464         (*state)->src_node->conn++;
 465         (*state)->src.tcp_est = 1;
 466         pf_add_threshold(&(*state)->src_node->conn_rate);
 467
 468         if ((*state)->rule.ptr->max_src_conn &&
 469             (*state)->rule.ptr->max_src_conn <
 470             (*state)->src_node->conn) {
 471                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
 472                 bad++;
 473         }
 474
 475         if ((*state)->rule.ptr->max_src_conn_rate.limit &&
 476             pf_check_threshold(&(*state)->src_node->conn_rate)) {
 477                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
 478                 bad++;
 479         }
 480
 481         if (!bad)
 482                 return (0);
 483
 484         /* Kill this state. */
 485         (*state)->timeout = PFTM_PURGE;
 486         (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
 487
 488         if ((*state)->rule.ptr->overload_tbl == NULL)
 489                 return (1);
 490
 491         /* Schedule overloading and flushing task. */
 492         pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
 493         if (pfoe == NULL)
 494                 return (1);     /* too bad :( */
 495
 496         bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
 497         pfoe->af = (*state)->key[PF_SK_WIRE]->af;
 498         pfoe->rule = (*state)->rule.ptr;
 499         pfoe->dir = (*state)->direction;
 500         PF_OVERLOADQ_LOCK();
 501         SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
 502         PF_OVERLOADQ_UNLOCK();
 503         taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
 504
 505         return (1);
 506 }
 507
 508 static void
 509 pf_overload_task(void *v, int pending)
 510 {
 511         struct pf_overload_head queue;
 512         struct pfr_addr p;
 513         struct pf_overload_entry *pfoe, *pfoe1;
 514         uint32_t killed = 0;
 515
 516         CURVNET_SET((struct vnet *)v);
 517
 518         PF_OVERLOADQ_LOCK();
 519         queue = V_pf_overloadqueue;
 520         SLIST_INIT(&V_pf_overloadqueue);
 521         PF_OVERLOADQ_UNLOCK();
 522
 523         bzero(&p, sizeof(p));
 524         SLIST_FOREACH(pfoe, &queue, next) {
 525                 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
 526                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
 527                         printf("%s: blocking address ", __func__);
 528                         pf_print_host(&pfoe->addr, 0, pfoe->af);
 529                         printf("\n");
 530                 }
 531
 532                 p.pfra_af = pfoe->af;
 533                 switch (pfoe->af) {
 534 #ifdef INET
 535                 case AF_INET:
 536                         p.pfra_net = 32;
 537                         p.pfra_ip4addr = pfoe->addr.v4;
 538                         break;
 539 #endif
 540 #ifdef INET6
 541                 case AF_INET6:
 542                         p.pfra_net = 128;
 543                         p.pfra_ip6addr = pfoe->addr.v6;
 544                         break;
 545 #endif
 546                 }
 547
 548                 PF_RULES_WLOCK();
 549                 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
 550                 PF_RULES_WUNLOCK();
 551         }
 552
 553         /*
 554          * Remove those entries, that don't need flushing.
 555          */
 556         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 557                 if (pfoe->rule->flush == 0) {
 558                         SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
 559                         free(pfoe, M_PFTEMP);
 560                 } else
 561                         counter_u64_add(
 562                             V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
 563
 564         /* If nothing to flush, return. */
 565         if (SLIST_EMPTY(&queue)) {
 566                 CURVNET_RESTORE();
 567                 return;
 568         }
 569
 570         for (int i = 0; i <= V_pf_hashmask; i++) {
 571                 struct pf_idhash *ih = &V_pf_idhash[i];
 572                 struct pf_state_key *sk;
 573                 struct pf_state *s;
 574
 575                 PF_HASHROW_LOCK(ih);
 576                 LIST_FOREACH(s, &ih->states, entry) {
 577                     sk = s->key[PF_SK_WIRE];
 578                     SLIST_FOREACH(pfoe, &queue, next)
 579                         if (sk->af == pfoe->af &&
 580                             ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
 581                             pfoe->rule == s->rule.ptr) &&
 582                             ((pfoe->dir == PF_OUT &&
 583                             PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
 584                             (pfoe->dir == PF_IN &&
 585                             PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
 586                                 s->timeout = PFTM_PURGE;
 587                                 s->src.state = s->dst.state = TCPS_CLOSED;
 588                                 killed++;
 589                         }
 590                 }
 591                 PF_HASHROW_UNLOCK(ih);
 592         }
 593         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 594                 free(pfoe, M_PFTEMP);
 595         if (V_pf_status.debug >= PF_DEBUG_MISC)
 596                 printf("%s: %u states killed", __func__, killed);
 597
 598         CURVNET_RESTORE();
 599 }
 600
 601 /*
 602  * Can return locked on failure, so that we can consistently
 603  * allocate and insert a new one.
 604  */
 605 struct pf_src_node *
 606 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
 607         int returnlocked)
 608 {
 609         struct pf_srchash *sh;
 610         struct pf_src_node *n;
 611
 612         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
 613
 614         sh = &V_pf_srchash[pf_hashsrc(src, af)];
 615         PF_HASHROW_LOCK(sh);
 616         LIST_FOREACH(n, &sh->nodes, entry)
 617                 if (n->rule.ptr == rule && n->af == af &&
 618                     ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
 619                     (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
 620                         break;
 621         if (n != NULL || returnlocked == 0)
 622                 PF_HASHROW_UNLOCK(sh);
 623
 624         return (n);
 625 }
 626
 627 static int
 628 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
 629     struct pf_addr *src, sa_family_t af)
 630 {
 631
 632         KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
 633             rule->rpool.opts & PF_POOL_STICKYADDR),
 634             ("%s for non-tracking rule %p", __func__, rule));
 635
 636         if (*sn == NULL)
 637                 *sn = pf_find_src_node(src, rule, af, 1);
 638
 639         if (*sn == NULL) {
 640                 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
 641
 642                 PF_HASHROW_ASSERT(sh);
 643
 644                 if (!rule->max_src_nodes ||
 645                     counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
 646                         (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
 647                 else
 648                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
 649                             1);
 650                 if ((*sn) == NULL) {
 651                         PF_HASHROW_UNLOCK(sh);
 652                         return (-1);
 653                 }
 654
 655                 pf_init_threshold(&(*sn)->conn_rate,
 656                     rule->max_src_conn_rate.limit,
 657                     rule->max_src_conn_rate.seconds);
 658
 659                 (*sn)->af = af;
 660                 (*sn)->rule.ptr = rule;
 661                 PF_ACPY(&(*sn)->addr, src, af);
 662                 LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
 663                 (*sn)->creation = time_uptime;
 664                 (*sn)->ruletype = rule->action;
 665                 if ((*sn)->rule.ptr != NULL)
 666                         counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
 667                 PF_HASHROW_UNLOCK(sh);
 668                 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
 669         } else {
 670                 if (rule->max_src_states &&
 671                     (*sn)->states >= rule->max_src_states) {
 672                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
 673                             1);
 674                         return (-1);
 675                 }
 676         }
 677         return (0);
 678 }
 679
 680 void
 681 pf_unlink_src_node_locked(struct pf_src_node *src)
 682 {
 683 #ifdef INVARIANTS
 684         struct pf_srchash *sh;
 685
 686         sh = &V_pf_srchash[pf_hashsrc(&src->addr, src->af)];
 687         PF_HASHROW_ASSERT(sh);
 688 #endif
 689         LIST_REMOVE(src, entry);
 690         if (src->rule.ptr)
 691                 counter_u64_add(src->rule.ptr->src_nodes, -1);
 692         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
 693 }
 694
 695 void
 696 pf_unlink_src_node(struct pf_src_node *src)
 697 {
 698         struct pf_srchash *sh;
 699
 700         sh = &V_pf_srchash[pf_hashsrc(&src->addr, src->af)];
 701         PF_HASHROW_LOCK(sh);
 702         pf_unlink_src_node_locked(src);
 703         PF_HASHROW_UNLOCK(sh);
 704 }
 705
 706 static void
 707 pf_free_src_node(struct pf_src_node *sn)
 708 {
 709
 710         KASSERT(sn->states == 0, ("%s: %p has refs", __func__, sn));
 711         uma_zfree(V_pf_sources_z, sn);
 712 }
 713
 714 u_int
 715 pf_free_src_nodes(struct pf_src_node_list *head)
 716 {
 717         struct pf_src_node *sn, *tmp;
 718         u_int count = 0;
 719
 720         LIST_FOREACH_SAFE(sn, head, entry, tmp) {
 721                 pf_free_src_node(sn);
 722                 count++;
 723         }
 724
 725         return (count);
 726 }
 727
 728 void
 729 pf_mtag_initialize()
 730 {
 731
 732         pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
 733             sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
 734             UMA_ALIGN_PTR, 0);
 735 }
 736
 737 /* Per-vnet data storage structures initialization. */
 738 void
 739 pf_initialize()
 740 {
 741         struct pf_keyhash       *kh;
 742         struct pf_idhash        *ih;
 743         struct pf_srchash       *sh;
 744         u_int i;
 745
 746         TUNABLE_ULONG_FETCH("net.pf.states_hashsize", &V_pf_hashsize);
 747         if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize))
 748                 V_pf_hashsize = PF_HASHSIZ;
 749         TUNABLE_ULONG_FETCH("net.pf.source_nodes_hashsize", &V_pf_srchashsize);
 750         if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize))
 751                 V_pf_srchashsize = PF_HASHSIZ / 4;
 752
 753         V_pf_hashseed = arc4random();
 754
 755         /* States and state keys storage. */
 756         V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
 757             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 758         V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
 759         uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
 760         uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
 761
 762         V_pf_state_key_z = uma_zcreate("pf state keys",
 763             sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
 764             UMA_ALIGN_PTR, 0);
 765         V_pf_keyhash = malloc(V_pf_hashsize * sizeof(struct pf_keyhash),
 766             M_PFHASH, M_WAITOK | M_ZERO);
 767         V_pf_idhash = malloc(V_pf_hashsize * sizeof(struct pf_idhash),
 768             M_PFHASH, M_WAITOK | M_ZERO);
 769         V_pf_hashmask = V_pf_hashsize - 1;
 770         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
 771             i++, kh++, ih++) {
 772                 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
 773                 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
 774         }
 775
 776         /* Source nodes. */
 777         V_pf_sources_z = uma_zcreate("pf source nodes",
 778             sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
 779             0);
 780         V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
 781         uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
 782         uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
 783         V_pf_srchash = malloc(V_pf_srchashsize * sizeof(struct pf_srchash),
 784           M_PFHASH, M_WAITOK|M_ZERO);
 785         V_pf_srchashmask = V_pf_srchashsize - 1;
 786         for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++)
 787                 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
 788
 789         /* ALTQ */
 790         TAILQ_INIT(&V_pf_altqs[0]);
 791         TAILQ_INIT(&V_pf_altqs[1]);
 792         TAILQ_INIT(&V_pf_pabuf);
 793         V_pf_altqs_active = &V_pf_altqs[0];
 794         V_pf_altqs_inactive = &V_pf_altqs[1];
 795
 796
 797         /* Send & overload+flush queues. */
 798         STAILQ_INIT(&V_pf_sendqueue);
 799         SLIST_INIT(&V_pf_overloadqueue);
 800         TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
 801         mtx_init(&pf_sendqueue_mtx, "pf send queue", NULL, MTX_DEF);
 802         mtx_init(&pf_overloadqueue_mtx, "pf overload/flush queue", NULL,
 803             MTX_DEF);
 804
 805         /* Unlinked, but may be referenced rules. */
 806         TAILQ_INIT(&V_pf_unlinked_rules);
 807         mtx_init(&pf_unlnkdrules_mtx, "pf unlinked rules", NULL, MTX_DEF);
 808 }
 809
 810 void
 811 pf_mtag_cleanup()
 812 {
 813
 814         uma_zdestroy(pf_mtag_z);
 815 }
 816
 817 void
 818 pf_cleanup()
 819 {
 820         struct pf_keyhash       *kh;
 821         struct pf_idhash        *ih;
 822         struct pf_srchash       *sh;
 823         struct pf_send_entry    *pfse, *next;
 824         u_int i;
 825
 826         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
 827             i++, kh++, ih++) {
 828                 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
 829                     __func__));
 830                 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
 831                     __func__));
 832                 mtx_destroy(&kh->lock);
 833                 mtx_destroy(&ih->lock);
 834         }
 835         free(V_pf_keyhash, M_PFHASH);
 836         free(V_pf_idhash, M_PFHASH);
 837
 838         for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
 839                 KASSERT(LIST_EMPTY(&sh->nodes),
 840                     ("%s: source node hash not empty", __func__));
 841                 mtx_destroy(&sh->lock);
 842         }
 843         free(V_pf_srchash, M_PFHASH);
 844
 845         STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
 846                 m_freem(pfse->pfse_m);
 847                 free(pfse, M_PFTEMP);
 848         }
 849
 850         mtx_destroy(&pf_sendqueue_mtx);
 851         mtx_destroy(&pf_overloadqueue_mtx);
 852         mtx_destroy(&pf_unlnkdrules_mtx);
 853
 854         uma_zdestroy(V_pf_sources_z);
 855         uma_zdestroy(V_pf_state_z);
 856         uma_zdestroy(V_pf_state_key_z);
 857 }
 858
 859 static int
 860 pf_mtag_uminit(void *mem, int size, int how)
 861 {
 862         struct m_tag *t;
 863
 864         t = (struct m_tag *)mem;
 865         t->m_tag_cookie = MTAG_ABI_COMPAT;
 866         t->m_tag_id = PACKET_TAG_PF;
 867         t->m_tag_len = sizeof(struct pf_mtag);
 868         t->m_tag_free = pf_mtag_free;
 869
 870         return (0);
 871 }
 872
 873 static void
 874 pf_mtag_free(struct m_tag *t)
 875 {
 876
 877         uma_zfree(pf_mtag_z, t);
 878 }
 879
 880 struct pf_mtag *
 881 pf_get_mtag(struct mbuf *m)
 882 {
 883         struct m_tag *mtag;
 884
 885         if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
 886                 return ((struct pf_mtag *)(mtag + 1));
 887
 888         mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
 889         if (mtag == NULL)
 890                 return (NULL);
 891         bzero(mtag + 1, sizeof(struct pf_mtag));
 892         m_tag_prepend(m, mtag);
 893
 894         return ((struct pf_mtag *)(mtag + 1));
 895 }
 896
 897 static int
 898 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
 899     struct pf_state *s)
 900 {
 901         struct pf_keyhash       *khs, *khw, *kh;
 902         struct pf_state_key     *sk, *cur;
 903         struct pf_state         *si, *olds = NULL;
 904         int idx;
 905
 906         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
 907         KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
 908         KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
 909
 910         /*
 911          * We need to lock hash slots of both keys. To avoid deadlock
 912          * we always lock the slot with lower address first. Unlock order
 913          * isn't important.
 914          *
 915          * We also need to lock ID hash slot before dropping key
 916          * locks. On success we return with ID hash slot locked.
 917          */
 918
 919         if (skw == sks) {
 920                 khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
 921                 PF_HASHROW_LOCK(khs);
 922         } else {
 923                 khs = &V_pf_keyhash[pf_hashkey(sks)];
 924                 khw = &V_pf_keyhash[pf_hashkey(skw)];
 925                 if (khs == khw) {
 926                         PF_HASHROW_LOCK(khs);
 927                 } else if (khs < khw) {
 928                         PF_HASHROW_LOCK(khs);
 929                         PF_HASHROW_LOCK(khw);
 930                 } else {
 931                         PF_HASHROW_LOCK(khw);
 932                         PF_HASHROW_LOCK(khs);
 933                 }
 934         }
 935
 936 #define KEYS_UNLOCK()   do {                    \
 937         if (khs != khw) {                       \
 938                 PF_HASHROW_UNLOCK(khs);         \
 939                 PF_HASHROW_UNLOCK(khw);         \
 940         } else                                  \
 941                 PF_HASHROW_UNLOCK(khs);         \
 942 } while (0)
 943
 944         /*
 945          * First run: start with wire key.
 946          */
 947         sk = skw;
 948         kh = khw;
 949         idx = PF_SK_WIRE;
 950
 951 keyattach:
 952         LIST_FOREACH(cur, &kh->keys, entry)
 953                 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
 954                         break;
 955
 956         if (cur != NULL) {
 957                 /* Key exists. Check for same kif, if none, add to key. */
 958                 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
 959                         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
 960
 961                         PF_HASHROW_LOCK(ih);
 962                         if (si->kif == s->kif &&
 963                             si->direction == s->direction) {
 964                                 if (sk->proto == IPPROTO_TCP &&
 965                                     si->src.state >= TCPS_FIN_WAIT_2 &&
 966                                     si->dst.state >= TCPS_FIN_WAIT_2) {
 967                                         /*
 968                                          * New state matches an old >FIN_WAIT_2
 969                                          * state. We can't drop key hash locks,
 970                                          * thus we can't unlink it properly.
 971                                          *
 972                                          * As a workaround we drop it into
 973                                          * TCPS_CLOSED state, schedule purge
 974                                          * ASAP and push it into the very end
 975                                          * of the slot TAILQ, so that it won't
 976                                          * conflict with our new state.
 977                                          */
 978                                         si->src.state = si->dst.state =
 979                                             TCPS_CLOSED;
 980                                         si->timeout = PFTM_PURGE;
 981                                         olds = si;
 982                                 } else {
 983                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
 984                                                 printf("pf: %s key attach "
 985                                                     "failed on %s: ",
 986                                                     (idx == PF_SK_WIRE) ?
 987                                                     "wire" : "stack",
 988                                                     s->kif->pfik_name);
 989                                                 pf_print_state_parts(s,
 990                                                     (idx == PF_SK_WIRE) ?
 991                                                     sk : NULL,
 992                                                     (idx == PF_SK_STACK) ?
 993                                                     sk : NULL);
 994                                                 printf(", existing: ");
 995                                                 pf_print_state_parts(si,
 996                                                     (idx == PF_SK_WIRE) ?
 997                                                     sk : NULL,
 998                                                     (idx == PF_SK_STACK) ?
 999                                                     sk : NULL);
1000                                                 printf("\n");
1001                                         }
1002                                         PF_HASHROW_UNLOCK(ih);
1003                                         KEYS_UNLOCK();
1004                                         uma_zfree(V_pf_state_key_z, sk);
1005                                         if (idx == PF_SK_STACK)
1006                                                 pf_detach_state(s);
1007                                         return (EEXIST); /* collision! */
1008                                 }
1009                         }
1010                         PF_HASHROW_UNLOCK(ih);
1011                 }
1012                 uma_zfree(V_pf_state_key_z, sk);
1013                 s->key[idx] = cur;
1014         } else {
1015                 LIST_INSERT_HEAD(&kh->keys, sk, entry);
1016                 s->key[idx] = sk;
1017         }
1018
1019 stateattach:
1020         /* List is sorted, if-bound states before floating. */
1021         if (s->kif == V_pfi_all)
1022                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1023         else
1024                 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1025
1026         if (olds) {
1027                 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1028                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1029                     key_list[idx]);
1030                 olds = NULL;
1031         }
1032
1033         /*
1034          * Attach done. See how should we (or should not?)
1035          * attach a second key.
1036          */
1037         if (sks == skw) {
1038                 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1039                 idx = PF_SK_STACK;
1040                 sks = NULL;
1041                 goto stateattach;
1042         } else if (sks != NULL) {
1043                 /*
1044                  * Continue attaching with stack key.
1045                  */
1046                 sk = sks;
1047                 kh = khs;
1048                 idx = PF_SK_STACK;
1049                 sks = NULL;
1050                 goto keyattach;
1051         }
1052
1053         PF_STATE_LOCK(s);
1054         KEYS_UNLOCK();
1055
1056         KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1057             ("%s failure", __func__));
1058
1059         return (0);
1060 #undef  KEYS_UNLOCK
1061 }
1062
1063 static void
1064 pf_detach_state(struct pf_state *s)
1065 {
1066         struct pf_state_key *sks = s->key[PF_SK_STACK];
1067         struct pf_keyhash *kh;
1068
1069         if (sks != NULL) {
1070                 kh = &V_pf_keyhash[pf_hashkey(sks)];
1071                 PF_HASHROW_LOCK(kh);
1072                 if (s->key[PF_SK_STACK] != NULL)
1073                         pf_state_key_detach(s, PF_SK_STACK);
1074                 /*
1075                  * If both point to same key, then we are done.
1076                  */
1077                 if (sks == s->key[PF_SK_WIRE]) {
1078                         pf_state_key_detach(s, PF_SK_WIRE);
1079                         PF_HASHROW_UNLOCK(kh);
1080                         return;
1081                 }
1082                 PF_HASHROW_UNLOCK(kh);
1083         }
1084
1085         if (s->key[PF_SK_WIRE] != NULL) {
1086                 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1087                 PF_HASHROW_LOCK(kh);
1088                 if (s->key[PF_SK_WIRE] != NULL)
1089                         pf_state_key_detach(s, PF_SK_WIRE);
1090                 PF_HASHROW_UNLOCK(kh);
1091         }
1092 }
1093
1094 static void
1095 pf_state_key_detach(struct pf_state *s, int idx)
1096 {
1097         struct pf_state_key *sk = s->key[idx];
1098 #ifdef INVARIANTS
1099         struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1100
1101         PF_HASHROW_ASSERT(kh);
1102 #endif
1103         TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1104         s->key[idx] = NULL;
1105
1106         if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1107                 LIST_REMOVE(sk, entry);
1108                 uma_zfree(V_pf_state_key_z, sk);
1109         }
1110 }
1111
1112 static int
1113 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1114 {
1115         struct pf_state_key *sk = mem;
1116
1117         bzero(sk, sizeof(struct pf_state_key_cmp));
1118         TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1119         TAILQ_INIT(&sk->states[PF_SK_STACK]);
1120
1121         return (0);
1122 }
1123
1124 struct pf_state_key *
1125 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1126         struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1127 {
1128         struct pf_state_key *sk;
1129
1130         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1131         if (sk == NULL)
1132                 return (NULL);
1133
1134         PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1135         PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1136         sk->port[pd->sidx] = sport;
1137         sk->port[pd->didx] = dport;
1138         sk->proto = pd->proto;
1139         sk->af = pd->af;
1140
1141         return (sk);
1142 }
1143
1144 struct pf_state_key *
1145 pf_state_key_clone(struct pf_state_key *orig)
1146 {
1147         struct pf_state_key *sk;
1148
1149         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1150         if (sk == NULL)
1151                 return (NULL);
1152
1153         bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1154
1155         return (sk);
1156 }
1157
1158 int
1159 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1160     struct pf_state_key *sks, struct pf_state *s)
1161 {
1162         struct pf_idhash *ih;
1163         struct pf_state *cur;
1164         int error;
1165
1166         KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1167             ("%s: sks not pristine", __func__));
1168         KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1169             ("%s: skw not pristine", __func__));
1170         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1171
1172         s->kif = kif;
1173
1174         if (s->id == 0 && s->creatorid == 0) {
1175                 /* XXX: should be atomic, but probability of collision low */
1176                 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1177                         V_pf_stateid[curcpu] = 1;
1178                 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1179                 s->id = htobe64(s->id);
1180                 s->creatorid = V_pf_status.hostid;
1181         }
1182
1183         /* Returns with ID locked on success. */
1184         if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1185                 return (error);
1186
1187         ih = &V_pf_idhash[PF_IDHASH(s)];
1188         PF_HASHROW_ASSERT(ih);
1189         LIST_FOREACH(cur, &ih->states, entry)
1190                 if (cur->id == s->id && cur->creatorid == s->creatorid)
1191                         break;
1192
1193         if (cur != NULL) {
1194                 PF_HASHROW_UNLOCK(ih);
1195                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1196                         printf("pf: state ID collision: "
1197                             "id: %016llx creatorid: %08x\n",
1198                             (unsigned long long)be64toh(s->id),
1199                             ntohl(s->creatorid));
1200                 }
1201                 pf_detach_state(s);
1202                 return (EEXIST);
1203         }
1204         LIST_INSERT_HEAD(&ih->states, s, entry);
1205         /* One for keys, one for ID hash. */
1206         refcount_init(&s->refs, 2);
1207
1208         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1209         if (pfsync_insert_state_ptr != NULL)
1210                 pfsync_insert_state_ptr(s);
1211
1212         /* Returns locked. */
1213         return (0);
1214 }
1215
1216 /*
1217  * Find state by ID: returns with locked row on success.
1218  */
1219 struct pf_state *
1220 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1221 {
1222         struct pf_idhash *ih;
1223         struct pf_state *s;
1224
1225         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1226
1227         ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))];
1228
1229         PF_HASHROW_LOCK(ih);
1230         LIST_FOREACH(s, &ih->states, entry)
1231                 if (s->id == id && s->creatorid == creatorid)
1232                         break;
1233
1234         if (s == NULL)
1235                 PF_HASHROW_UNLOCK(ih);
1236
1237         return (s);
1238 }
1239
1240 /*
1241  * Find state by key.
1242  * Returns with ID hash slot locked on success.
1243  */
1244 static struct pf_state *
1245 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1246 {
1247         struct pf_keyhash       *kh;
1248         struct pf_state_key     *sk;
1249         struct pf_state         *s;
1250         int idx;
1251
1252         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1253
1254         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1255
1256         PF_HASHROW_LOCK(kh);
1257         LIST_FOREACH(sk, &kh->keys, entry)
1258                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1259                         break;
1260         if (sk == NULL) {
1261                 PF_HASHROW_UNLOCK(kh);
1262                 return (NULL);
1263         }
1264
1265         idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1266
1267         /* List is sorted, if-bound states before floating ones. */
1268         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1269                 if (s->kif == V_pfi_all || s->kif == kif) {
1270                         PF_STATE_LOCK(s);
1271                         PF_HASHROW_UNLOCK(kh);
1272                         if (s->timeout >= PFTM_MAX) {
1273                                 /*
1274                                  * State is either being processed by
1275                                  * pf_unlink_state() in an other thread, or
1276                                  * is scheduled for immediate expiry.
1277                                  */
1278                                 PF_STATE_UNLOCK(s);
1279                                 return (NULL);
1280                         }
1281                         return (s);
1282                 }
1283         PF_HASHROW_UNLOCK(kh);
1284
1285         return (NULL);
1286 }
1287
1288 struct pf_state *
1289 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1290 {
1291         struct pf_keyhash       *kh;
1292         struct pf_state_key     *sk;
1293         struct pf_state         *s, *ret = NULL;
1294         int                      idx, inout = 0;
1295
1296         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1297
1298         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1299
1300         PF_HASHROW_LOCK(kh);
1301         LIST_FOREACH(sk, &kh->keys, entry)
1302                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1303                         break;
1304         if (sk == NULL) {
1305                 PF_HASHROW_UNLOCK(kh);
1306                 return (NULL);
1307         }
1308         switch (dir) {
1309         case PF_IN:
1310                 idx = PF_SK_WIRE;
1311                 break;
1312         case PF_OUT:
1313                 idx = PF_SK_STACK;
1314                 break;
1315         case PF_INOUT:
1316                 idx = PF_SK_WIRE;
1317                 inout = 1;
1318                 break;
1319         default:
1320                 panic("%s: dir %u", __func__, dir);
1321         }
1322 second_run:
1323         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1324                 if (more == NULL) {
1325                         PF_HASHROW_UNLOCK(kh);
1326                         return (s);
1327                 }
1328
1329                 if (ret)
1330                         (*more)++;
1331                 else
1332                         ret = s;
1333         }
1334         if (inout == 1) {
1335                 inout = 0;
1336                 idx = PF_SK_STACK;
1337                 goto second_run;
1338         }
1339         PF_HASHROW_UNLOCK(kh);
1340
1341         return (ret);
1342 }
1343
1344 /* END state table stuff */
1345
1346 static void
1347 pf_send(struct pf_send_entry *pfse)
1348 {
1349
1350         PF_SENDQ_LOCK();
1351         STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1352         PF_SENDQ_UNLOCK();
1353         swi_sched(V_pf_swi_cookie, 0);
1354 }
1355
1356 void
1357 pf_intr(void *v)
1358 {
1359         struct pf_send_head queue;
1360         struct pf_send_entry *pfse, *next;
1361
1362         CURVNET_SET((struct vnet *)v);
1363
1364         PF_SENDQ_LOCK();
1365         queue = V_pf_sendqueue;
1366         STAILQ_INIT(&V_pf_sendqueue);
1367         PF_SENDQ_UNLOCK();
1368
1369         STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1370                 switch (pfse->pfse_type) {
1371 #ifdef INET
1372                 case PFSE_IP:
1373                         ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1374                         break;
1375                 case PFSE_ICMP:
1376                         icmp_error(pfse->pfse_m, pfse->pfse_icmp_type,
1377                             pfse->pfse_icmp_code, 0, pfse->pfse_icmp_mtu);
1378                         break;
1379 #endif /* INET */
1380 #ifdef INET6
1381                 case PFSE_IP6:
1382                         ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1383                             NULL);
1384                         break;
1385                 case PFSE_ICMP6:
1386                         icmp6_error(pfse->pfse_m, pfse->pfse_icmp_type,
1387                             pfse->pfse_icmp_code, pfse->pfse_icmp_mtu);
1388                         break;
1389 #endif /* INET6 */
1390                 default:
1391                         panic("%s: unknown type", __func__);
1392                 }
1393                 free(pfse, M_PFTEMP);
1394         }
1395         CURVNET_RESTORE();
1396 }
1397
1398 void
1399 pf_purge_thread(void *v)
1400 {
1401         u_int idx = 0;
1402
1403         CURVNET_SET((struct vnet *)v);
1404
1405         for (;;) {
1406                 PF_RULES_RLOCK();
1407                 rw_sleep(pf_purge_thread, &pf_rules_lock, 0, "pftm", hz / 10);
1408
1409                 if (V_pf_end_threads) {
1410                         /*
1411                          * To cleanse up all kifs and rules we need
1412                          * two runs: first one clears reference flags,
1413                          * then pf_purge_expired_states() doesn't
1414                          * raise them, and then second run frees.
1415                          */
1416                         PF_RULES_RUNLOCK();
1417                         pf_purge_unlinked_rules();
1418                         pfi_kif_purge();
1419
1420                         /*
1421                          * Now purge everything.
1422                          */
1423                         pf_purge_expired_states(0, V_pf_hashmask);
1424                         pf_purge_expired_fragments();
1425                         pf_purge_expired_src_nodes();
1426
1427                         /*
1428                          * Now all kifs & rules should be unreferenced,
1429                          * thus should be successfully freed.
1430                          */
1431                         pf_purge_unlinked_rules();
1432                         pfi_kif_purge();
1433
1434                         /*
1435                          * Announce success and exit.
1436                          */
1437                         PF_RULES_RLOCK();
1438                         V_pf_end_threads++;
1439                         PF_RULES_RUNLOCK();
1440                         wakeup(pf_purge_thread);
1441                         kproc_exit(0);
1442                 }
1443                 PF_RULES_RUNLOCK();
1444
1445                 /* Process 1/interval fraction of the state table every run. */
1446                 idx = pf_purge_expired_states(idx, V_pf_hashmask /
1447                             (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1448
1449                 /* Purge other expired types every PFTM_INTERVAL seconds. */
1450                 if (idx == 0) {
1451                         /*
1452                          * Order is important:
1453                          * - states and src nodes reference rules
1454                          * - states and rules reference kifs
1455                          */
1456                         pf_purge_expired_fragments();
1457                         pf_purge_expired_src_nodes();
1458                         pf_purge_unlinked_rules();
1459                         pfi_kif_purge();
1460                 }
1461         }
1462         /* not reached */
1463         CURVNET_RESTORE();
1464 }
1465
1466 u_int32_t
1467 pf_state_expires(const struct pf_state *state)
1468 {
1469         u_int32_t       timeout;
1470         u_int32_t       start;
1471         u_int32_t       end;
1472         u_int32_t       states;
1473
1474         /* handle all PFTM_* > PFTM_MAX here */
1475         if (state->timeout == PFTM_PURGE)
1476                 return (time_uptime);
1477         KASSERT(state->timeout != PFTM_UNLINKED,
1478             ("pf_state_expires: timeout == PFTM_UNLINKED"));
1479         KASSERT((state->timeout < PFTM_MAX),
1480             ("pf_state_expires: timeout > PFTM_MAX"));
1481         timeout = state->rule.ptr->timeout[state->timeout];
1482         if (!timeout)
1483                 timeout = V_pf_default_rule.timeout[state->timeout];
1484         start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1485         if (start) {
1486                 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1487                 states = counter_u64_fetch(state->rule.ptr->states_cur);
1488         } else {
1489                 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1490                 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1491                 states = V_pf_status.states;
1492         }
1493         if (end && states > start && start < end) {
1494                 if (states < end)
1495                         return (state->expire + timeout * (end - states) /
1496                             (end - start));
1497                 else
1498                         return (time_uptime);
1499         }
1500         return (state->expire + timeout);
1501 }
1502
1503 void
1504 pf_purge_expired_src_nodes()
1505 {
1506         struct pf_src_node_list  freelist;
1507         struct pf_srchash       *sh;
1508         struct pf_src_node      *cur, *next;
1509         int i;
1510
1511         LIST_INIT(&freelist);
1512         for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
1513             PF_HASHROW_LOCK(sh);
1514             LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1515                 if (cur->states == 0 && cur->expire <= time_uptime) {
1516                         pf_unlink_src_node_locked(cur);
1517                         LIST_INSERT_HEAD(&freelist, cur, entry);
1518                 } else if (cur->rule.ptr != NULL)
1519                         cur->rule.ptr->rule_flag |= PFRULE_REFS;
1520             PF_HASHROW_UNLOCK(sh);
1521         }
1522
1523         pf_free_src_nodes(&freelist);
1524
1525         V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1526 }
1527
1528 static void
1529 pf_src_tree_remove_state(struct pf_state *s)
1530 {
1531         u_int32_t timeout;
1532
1533         if (s->src_node != NULL) {
1534                 if (s->src.tcp_est)
1535                         --s->src_node->conn;
1536                 if (--s->src_node->states == 0) {
1537                         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1538                         if (!timeout)
1539                                 timeout =
1540                                     V_pf_default_rule.timeout[PFTM_SRC_NODE];
1541                         s->src_node->expire = time_uptime + timeout;
1542                 }
1543         }
1544         if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1545                 if (--s->nat_src_node->states == 0) {
1546                         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1547                         if (!timeout)
1548                                 timeout =
1549                                     V_pf_default_rule.timeout[PFTM_SRC_NODE];
1550                         s->nat_src_node->expire = time_uptime + timeout;
1551                 }
1552         }
1553         s->src_node = s->nat_src_node = NULL;
1554 }
1555
1556 /*
1557  * Unlink and potentilly free a state. Function may be
1558  * called with ID hash row locked, but always returns
1559  * unlocked, since it needs to go through key hash locking.
1560  */
1561 int
1562 pf_unlink_state(struct pf_state *s, u_int flags)
1563 {
1564         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1565
1566         if ((flags & PF_ENTER_LOCKED) == 0)
1567                 PF_HASHROW_LOCK(ih);
1568         else
1569                 PF_HASHROW_ASSERT(ih);
1570
1571         if (s->timeout == PFTM_UNLINKED) {
1572                 /*
1573                  * State is being processed
1574                  * by pf_unlink_state() in
1575                  * an other thread.
1576                  */
1577                 PF_HASHROW_UNLOCK(ih);
1578                 return (0);     /* XXXGL: undefined actually */
1579         }
1580
1581         if (s->src.state == PF_TCPS_PROXY_DST) {
1582                 /* XXX wire key the right one? */
1583                 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1584                     &s->key[PF_SK_WIRE]->addr[1],
1585                     &s->key[PF_SK_WIRE]->addr[0],
1586                     s->key[PF_SK_WIRE]->port[1],
1587                     s->key[PF_SK_WIRE]->port[0],
1588                     s->src.seqhi, s->src.seqlo + 1,
1589                     TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1590         }
1591
1592         LIST_REMOVE(s, entry);
1593         pf_src_tree_remove_state(s);
1594
1595         if (pfsync_delete_state_ptr != NULL)
1596                 pfsync_delete_state_ptr(s);
1597
1598         STATE_DEC_COUNTERS(s);
1599
1600         s->timeout = PFTM_UNLINKED;
1601
1602         PF_HASHROW_UNLOCK(ih);
1603
1604         pf_detach_state(s);
1605         refcount_release(&s->refs);
1606
1607         return (pf_release_state(s));
1608 }
1609
1610 void
1611 pf_free_state(struct pf_state *cur)
1612 {
1613
1614         KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1615         KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1616             cur->timeout));
1617
1618         pf_normalize_tcp_cleanup(cur);
1619         uma_zfree(V_pf_state_z, cur);
1620         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
1621 }
1622
1623 /*
1624  * Called only from pf_purge_thread(), thus serialized.
1625  */
1626 static u_int
1627 pf_purge_expired_states(u_int i, int maxcheck)
1628 {
1629         struct pf_idhash *ih;
1630         struct pf_state *s;
1631
1632         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1633
1634         /*
1635          * Go through hash and unlink states that expire now.
1636          */
1637         while (maxcheck > 0) {
1638
1639                 ih = &V_pf_idhash[i];
1640 relock:
1641                 PF_HASHROW_LOCK(ih);
1642                 LIST_FOREACH(s, &ih->states, entry) {
1643                         if (pf_state_expires(s) <= time_uptime) {
1644                                 V_pf_status.states -=
1645                                     pf_unlink_state(s, PF_ENTER_LOCKED);
1646                                 goto relock;
1647                         }
1648                         s->rule.ptr->rule_flag |= PFRULE_REFS;
1649                         if (s->nat_rule.ptr != NULL)
1650                                 s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1651                         if (s->anchor.ptr != NULL)
1652                                 s->anchor.ptr->rule_flag |= PFRULE_REFS;
1653                         s->kif->pfik_flags |= PFI_IFLAG_REFS;
1654                         if (s->rt_kif)
1655                                 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1656                 }
1657                 PF_HASHROW_UNLOCK(ih);
1658
1659                 /* Return when we hit end of hash. */
1660                 if (++i > V_pf_hashmask) {
1661                         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1662                         return (0);
1663                 }
1664
1665                 maxcheck--;
1666         }
1667
1668         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1669
1670         return (i);
1671 }
1672
1673 static void
1674 pf_purge_unlinked_rules()
1675 {
1676         struct pf_rulequeue tmpq;
1677         struct pf_rule *r, *r1;
1678
1679         /*
1680          * If we have overloading task pending, then we'd
1681          * better skip purging this time. There is a tiny
1682          * probability that overloading task references
1683          * an already unlinked rule.
1684          */
1685         PF_OVERLOADQ_LOCK();
1686         if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1687                 PF_OVERLOADQ_UNLOCK();
1688                 return;
1689         }
1690         PF_OVERLOADQ_UNLOCK();
1691
1692         /*
1693          * Do naive mark-and-sweep garbage collecting of old rules.
1694          * Reference flag is raised by pf_purge_expired_states()
1695          * and pf_purge_expired_src_nodes().
1696          *
1697          * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1698          * use a temporary queue.
1699          */
1700         TAILQ_INIT(&tmpq);
1701         PF_UNLNKDRULES_LOCK();
1702         TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1703                 if (!(r->rule_flag & PFRULE_REFS)) {
1704                         TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1705                         TAILQ_INSERT_TAIL(&tmpq, r, entries);
1706                 } else
1707                         r->rule_flag &= ~PFRULE_REFS;
1708         }
1709         PF_UNLNKDRULES_UNLOCK();
1710
1711         if (!TAILQ_EMPTY(&tmpq)) {
1712                 PF_RULES_WLOCK();
1713                 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1714                         TAILQ_REMOVE(&tmpq, r, entries);
1715                         pf_free_rule(r);
1716                 }
1717                 PF_RULES_WUNLOCK();
1718         }
1719 }
1720
1721 void
1722 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1723 {
1724         switch (af) {
1725 #ifdef INET
1726         case AF_INET: {
1727                 u_int32_t a = ntohl(addr->addr32[0]);
1728                 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1729                     (a>>8)&255, a&255);
1730                 if (p) {
1731                         p = ntohs(p);
1732                         printf(":%u", p);
1733                 }
1734                 break;
1735         }
1736 #endif /* INET */
1737 #ifdef INET6
1738         case AF_INET6: {
1739                 u_int16_t b;
1740                 u_int8_t i, curstart, curend, maxstart, maxend;
1741                 curstart = curend = maxstart = maxend = 255;
1742                 for (i = 0; i < 8; i++) {
1743                         if (!addr->addr16[i]) {
1744                                 if (curstart == 255)
1745                                         curstart = i;
1746                                 curend = i;
1747                         } else {
1748                                 if ((curend - curstart) >
1749                                     (maxend - maxstart)) {
1750                                         maxstart = curstart;
1751                                         maxend = curend;
1752                                 }
1753                                 curstart = curend = 255;
1754                         }
1755                 }
1756                 if ((curend - curstart) >
1757                     (maxend - maxstart)) {
1758                         maxstart = curstart;
1759                         maxend = curend;
1760                 }
1761                 for (i = 0; i < 8; i++) {
1762                         if (i >= maxstart && i <= maxend) {
1763                                 if (i == 0)
1764                                         printf(":");
1765                                 if (i == maxend)
1766                                         printf(":");
1767                         } else {
1768                                 b = ntohs(addr->addr16[i]);
1769                                 printf("%x", b);
1770                                 if (i < 7)
1771                                         printf(":");
1772                         }
1773                 }
1774                 if (p) {
1775                         p = ntohs(p);
1776                         printf("[%u]", p);
1777                 }
1778                 break;
1779         }
1780 #endif /* INET6 */
1781         }
1782 }
1783
1784 void
1785 pf_print_state(struct pf_state *s)
1786 {
1787         pf_print_state_parts(s, NULL, NULL);
1788 }
1789
1790 static void
1791 pf_print_state_parts(struct pf_state *s,
1792     struct pf_state_key *skwp, struct pf_state_key *sksp)
1793 {
1794         struct pf_state_key *skw, *sks;
1795         u_int8_t proto, dir;
1796
1797         /* Do our best to fill these, but they're skipped if NULL */
1798         skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1799         sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1800         proto = skw ? skw->proto : (sks ? sks->proto : 0);
1801         dir = s ? s->direction : 0;
1802
1803         switch (proto) {
1804         case IPPROTO_IPV4:
1805                 printf("IPv4");
1806                 break;
1807         case IPPROTO_IPV6:
1808                 printf("IPv6");
1809                 break;
1810         case IPPROTO_TCP:
1811                 printf("TCP");
1812                 break;
1813         case IPPROTO_UDP:
1814                 printf("UDP");
1815                 break;
1816         case IPPROTO_ICMP:
1817                 printf("ICMP");
1818                 break;
1819         case IPPROTO_ICMPV6:
1820                 printf("ICMPv6");
1821                 break;
1822         default:
1823                 printf("%u", skw->proto);
1824                 break;
1825         }
1826         switch (dir) {
1827         case PF_IN:
1828                 printf(" in");
1829                 break;
1830         case PF_OUT:
1831                 printf(" out");
1832                 break;
1833         }
1834         if (skw) {
1835                 printf(" wire: ");
1836                 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1837                 printf(" ");
1838                 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1839         }
1840         if (sks) {
1841                 printf(" stack: ");
1842                 if (sks != skw) {
1843                         pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1844                         printf(" ");
1845                         pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1846                 } else
1847                         printf("-");
1848         }
1849         if (s) {
1850                 if (proto == IPPROTO_TCP) {
1851                         printf(" [lo=%u high=%u win=%u modulator=%u",
1852                             s->src.seqlo, s->src.seqhi,
1853                             s->src.max_win, s->src.seqdiff);
1854                         if (s->src.wscale && s->dst.wscale)
1855                                 printf(" wscale=%u",
1856                                     s->src.wscale & PF_WSCALE_MASK);
1857                         printf("]");
1858                         printf(" [lo=%u high=%u win=%u modulator=%u",
1859                             s->dst.seqlo, s->dst.seqhi,
1860                             s->dst.max_win, s->dst.seqdiff);
1861                         if (s->src.wscale && s->dst.wscale)
1862                                 printf(" wscale=%u",
1863                                 s->dst.wscale & PF_WSCALE_MASK);
1864                         printf("]");
1865                 }
1866                 printf(" %u:%u", s->src.state, s->dst.state);
1867         }
1868 }
1869
1870 void
1871 pf_print_flags(u_int8_t f)
1872 {
1873         if (f)
1874                 printf(" ");
1875         if (f & TH_FIN)
1876                 printf("F");
1877         if (f & TH_SYN)
1878                 printf("S");
1879         if (f & TH_RST)
1880                 printf("R");
1881         if (f & TH_PUSH)
1882                 printf("P");
1883         if (f & TH_ACK)
1884                 printf("A");
1885         if (f & TH_URG)
1886                 printf("U");
1887         if (f & TH_ECE)
1888                 printf("E");
1889         if (f & TH_CWR)
1890                 printf("W");
1891 }
1892
1893 #define PF_SET_SKIP_STEPS(i)                                    \
1894         do {                                                    \
1895                 while (head[i] != cur) {                        \
1896                         head[i]->skip[i].ptr = cur;             \
1897                         head[i] = TAILQ_NEXT(head[i], entries); \
1898                 }                                               \
1899         } while (0)
1900
1901 void
1902 pf_calc_skip_steps(struct pf_rulequeue *rules)
1903 {
1904         struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1905         int i;
1906
1907         cur = TAILQ_FIRST(rules);
1908         prev = cur;
1909         for (i = 0; i < PF_SKIP_COUNT; ++i)
1910                 head[i] = cur;
1911         while (cur != NULL) {
1912
1913                 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1914                         PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1915                 if (cur->direction != prev->direction)
1916                         PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1917                 if (cur->af != prev->af)
1918                         PF_SET_SKIP_STEPS(PF_SKIP_AF);
1919                 if (cur->proto != prev->proto)
1920                         PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
1921                 if (cur->src.neg != prev->src.neg ||
1922                     pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
1923                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
1924                 if (cur->src.port[0] != prev->src.port[0] ||
1925                     cur->src.port[1] != prev->src.port[1] ||
1926                     cur->src.port_op != prev->src.port_op)
1927                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
1928                 if (cur->dst.neg != prev->dst.neg ||
1929                     pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
1930                         PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
1931                 if (cur->dst.port[0] != prev->dst.port[0] ||
1932                     cur->dst.port[1] != prev->dst.port[1] ||
1933                     cur->dst.port_op != prev->dst.port_op)
1934                         PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
1935
1936                 prev = cur;
1937                 cur = TAILQ_NEXT(cur, entries);
1938         }
1939         for (i = 0; i < PF_SKIP_COUNT; ++i)
1940                 PF_SET_SKIP_STEPS(i);
1941 }
1942
1943 static int
1944 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
1945 {
1946         if (aw1->type != aw2->type)
1947                 return (1);
1948         switch (aw1->type) {
1949         case PF_ADDR_ADDRMASK:
1950         case PF_ADDR_RANGE:
1951                 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0))
1952                         return (1);
1953                 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0))
1954                         return (1);
1955                 return (0);
1956         case PF_ADDR_DYNIFTL:
1957                 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
1958         case PF_ADDR_NOROUTE:
1959         case PF_ADDR_URPFFAILED:
1960                 return (0);
1961         case PF_ADDR_TABLE:
1962                 return (aw1->p.tbl != aw2->p.tbl);
1963         default:
1964                 printf("invalid address type: %d\n", aw1->type);
1965                 return (1);
1966         }
1967 }
1968
1969 u_int16_t
1970 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
1971 {
1972         u_int32_t       l;
1973
1974         if (udp && !cksum)
1975                 return (0x0000);
1976         l = cksum + old - new;
1977         l = (l >> 16) + (l & 65535);
1978         l = l & 65535;
1979         if (udp && !l)
1980                 return (0xFFFF);
1981         return (l);
1982 }
1983
1984 static void
1985 pf_change_ap(struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc,
1986     struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af)
1987 {
1988         struct pf_addr  ao;
1989         u_int16_t       po = *p;
1990
1991         PF_ACPY(&ao, a, af);
1992         PF_ACPY(a, an, af);
1993
1994         *p = pn;
1995
1996         switch (af) {
1997 #ifdef INET
1998         case AF_INET:
1999                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2000                     ao.addr16[0], an->addr16[0], 0),
2001                     ao.addr16[1], an->addr16[1], 0);
2002                 *p = pn;
2003                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
2004                     ao.addr16[0], an->addr16[0], u),
2005                     ao.addr16[1], an->addr16[1], u),
2006                     po, pn, u);
2007                 break;
2008 #endif /* INET */
2009 #ifdef INET6
2010         case AF_INET6:
2011                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2012                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2013                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
2014                     ao.addr16[0], an->addr16[0], u),
2015                     ao.addr16[1], an->addr16[1], u),
2016                     ao.addr16[2], an->addr16[2], u),
2017                     ao.addr16[3], an->addr16[3], u),
2018                     ao.addr16[4], an->addr16[4], u),
2019                     ao.addr16[5], an->addr16[5], u),
2020                     ao.addr16[6], an->addr16[6], u),
2021                     ao.addr16[7], an->addr16[7], u),
2022                     po, pn, u);
2023                 break;
2024 #endif /* INET6 */
2025         }
2026 }
2027
2028
2029 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2030 void
2031 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2032 {
2033         u_int32_t       ao;
2034
2035         memcpy(&ao, a, sizeof(ao));
2036         memcpy(a, &an, sizeof(u_int32_t));
2037         *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2038             ao % 65536, an % 65536, u);
2039 }
2040
2041 #ifdef INET6
2042 static void
2043 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2044 {
2045         struct pf_addr  ao;
2046
2047         PF_ACPY(&ao, a, AF_INET6);
2048         PF_ACPY(a, an, AF_INET6);
2049
2050         *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2051             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2052             pf_cksum_fixup(pf_cksum_fixup(*c,
2053             ao.addr16[0], an->addr16[0], u),
2054             ao.addr16[1], an->addr16[1], u),
2055             ao.addr16[2], an->addr16[2], u),
2056             ao.addr16[3], an->addr16[3], u),
2057             ao.addr16[4], an->addr16[4], u),
2058             ao.addr16[5], an->addr16[5], u),
2059             ao.addr16[6], an->addr16[6], u),
2060             ao.addr16[7], an->addr16[7], u);
2061 }
2062 #endif /* INET6 */
2063
2064 static void
2065 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2066     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2067     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2068 {
2069         struct pf_addr  oia, ooa;
2070
2071         PF_ACPY(&oia, ia, af);
2072         if (oa)
2073                 PF_ACPY(&ooa, oa, af);
2074
2075         /* Change inner protocol port, fix inner protocol checksum. */
2076         if (ip != NULL) {
2077                 u_int16_t       oip = *ip;
2078                 u_int32_t       opc;
2079
2080                 if (pc != NULL)
2081                         opc = *pc;
2082                 *ip = np;
2083                 if (pc != NULL)
2084                         *pc = pf_cksum_fixup(*pc, oip, *ip, u);
2085                 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2086                 if (pc != NULL)
2087                         *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2088         }
2089         /* Change inner ip address, fix inner ip and icmp checksums. */
2090         PF_ACPY(ia, na, af);
2091         switch (af) {
2092 #ifdef INET
2093         case AF_INET: {
2094                 u_int32_t        oh2c = *h2c;
2095
2096                 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2097                     oia.addr16[0], ia->addr16[0], 0),
2098                     oia.addr16[1], ia->addr16[1], 0);
2099                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2100                     oia.addr16[0], ia->addr16[0], 0),
2101                     oia.addr16[1], ia->addr16[1], 0);
2102                 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2103                 break;
2104         }
2105 #endif /* INET */
2106 #ifdef INET6
2107         case AF_INET6:
2108                 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2109                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2110                     pf_cksum_fixup(pf_cksum_fixup(*ic,
2111                     oia.addr16[0], ia->addr16[0], u),
2112                     oia.addr16[1], ia->addr16[1], u),
2113                     oia.addr16[2], ia->addr16[2], u),
2114                     oia.addr16[3], ia->addr16[3], u),
2115                     oia.addr16[4], ia->addr16[4], u),
2116                     oia.addr16[5], ia->addr16[5], u),
2117                     oia.addr16[6], ia->addr16[6], u),
2118                     oia.addr16[7], ia->addr16[7], u);
2119                 break;
2120 #endif /* INET6 */
2121         }
2122         /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2123         if (oa) {
2124                 PF_ACPY(oa, na, af);
2125                 switch (af) {
2126 #ifdef INET
2127                 case AF_INET:
2128                         *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2129                             ooa.addr16[0], oa->addr16[0], 0),
2130                             ooa.addr16[1], oa->addr16[1], 0);
2131                         break;
2132 #endif /* INET */
2133 #ifdef INET6
2134                 case AF_INET6:
2135                         *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2136                             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2137                             pf_cksum_fixup(pf_cksum_fixup(*ic,
2138                             ooa.addr16[0], oa->addr16[0], u),
2139                             ooa.addr16[1], oa->addr16[1], u),
2140                             ooa.addr16[2], oa->addr16[2], u),
2141                             ooa.addr16[3], oa->addr16[3], u),
2142                             ooa.addr16[4], oa->addr16[4], u),
2143                             ooa.addr16[5], oa->addr16[5], u),
2144                             ooa.addr16[6], oa->addr16[6], u),
2145                             ooa.addr16[7], oa->addr16[7], u);
2146                         break;
2147 #endif /* INET6 */
2148                 }
2149         }
2150 }
2151
2152
2153 /*
2154  * Need to modulate the sequence numbers in the TCP SACK option
2155  * (credits to Krzysztof Pfaff for report and patch)
2156  */
2157 static int
2158 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2159     struct tcphdr *th, struct pf_state_peer *dst)
2160 {
2161         int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2162         u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2163         int copyback = 0, i, olen;
2164         struct sackblk sack;
2165
2166 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2167         if (hlen < TCPOLEN_SACKLEN ||
2168             !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2169                 return 0;
2170
2171         while (hlen >= TCPOLEN_SACKLEN) {
2172                 olen = opt[1];
2173                 switch (*opt) {
2174                 case TCPOPT_EOL:        /* FALLTHROUGH */
2175                 case TCPOPT_NOP:
2176                         opt++;
2177                         hlen--;
2178                         break;
2179                 case TCPOPT_SACK:
2180                         if (olen > hlen)
2181                                 olen = hlen;
2182                         if (olen >= TCPOLEN_SACKLEN) {
2183                                 for (i = 2; i + TCPOLEN_SACK <= olen;
2184                                     i += TCPOLEN_SACK) {
2185                                         memcpy(&sack, &opt[i], sizeof(sack));
2186                                         pf_change_a(&sack.start, &th->th_sum,
2187                                             htonl(ntohl(sack.start) -
2188                                             dst->seqdiff), 0);
2189                                         pf_change_a(&sack.end, &th->th_sum,
2190                                             htonl(ntohl(sack.end) -
2191                                             dst->seqdiff), 0);
2192                                         memcpy(&opt[i], &sack, sizeof(sack));
2193                                 }
2194                                 copyback = 1;
2195                         }
2196                         /* FALLTHROUGH */
2197                 default:
2198                         if (olen < 2)
2199                                 olen = 2;
2200                         hlen -= olen;
2201                         opt += olen;
2202                 }
2203         }
2204
2205         if (copyback)
2206                 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2207         return (copyback);
2208 }
2209
2210 static void
2211 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2212     const struct pf_addr *saddr, const struct pf_addr *daddr,
2213     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2214     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2215     u_int16_t rtag, struct ifnet *ifp)
2216 {
2217         struct pf_send_entry *pfse;
2218         struct mbuf     *m;
2219         int              len, tlen;
2220 #ifdef INET
2221         struct ip       *h = NULL;
2222 #endif /* INET */
2223 #ifdef INET6
2224         struct ip6_hdr  *h6 = NULL;
2225 #endif /* INET6 */
2226         struct tcphdr   *th;
2227         char            *opt;
2228         struct pf_mtag  *pf_mtag;
2229
2230         len = 0;
2231         th = NULL;
2232
2233         /* maximum segment size tcp option */
2234         tlen = sizeof(struct tcphdr);
2235         if (mss)
2236                 tlen += 4;
2237
2238         switch (af) {
2239 #ifdef INET
2240         case AF_INET:
2241                 len = sizeof(struct ip) + tlen;
2242                 break;
2243 #endif /* INET */
2244 #ifdef INET6
2245         case AF_INET6:
2246                 len = sizeof(struct ip6_hdr) + tlen;
2247                 break;
2248 #endif /* INET6 */
2249         default:
2250                 panic("%s: unsupported af %d", __func__, af);
2251         }
2252
2253         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2254         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2255         if (pfse == NULL)
2256                 return;
2257         m = m_gethdr(M_NOWAIT, MT_DATA);
2258         if (m == NULL) {
2259                 free(pfse, M_PFTEMP);
2260                 return;
2261         }
2262 #ifdef MAC
2263         mac_netinet_firewall_send(m);
2264 #endif
2265         if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2266                 free(pfse, M_PFTEMP);
2267                 m_freem(m);
2268                 return;
2269         }
2270         if (tag)
2271                 m->m_flags |= M_SKIP_FIREWALL;
2272         pf_mtag->tag = rtag;
2273
2274         if (r != NULL && r->rtableid >= 0)
2275                 M_SETFIB(m, r->rtableid);
2276
2277 #ifdef ALTQ
2278         if (r != NULL && r->qid) {
2279                 pf_mtag->qid = r->qid;
2280
2281                 /* add hints for ecn */
2282                 pf_mtag->hdr = mtod(m, struct ip *);
2283         }
2284 #endif /* ALTQ */
2285         m->m_data += max_linkhdr;
2286         m->m_pkthdr.len = m->m_len = len;
2287         m->m_pkthdr.rcvif = NULL;
2288         bzero(m->m_data, len);
2289         switch (af) {
2290 #ifdef INET
2291         case AF_INET:
2292                 h = mtod(m, struct ip *);
2293
2294                 /* IP header fields included in the TCP checksum */
2295                 h->ip_p = IPPROTO_TCP;
2296                 h->ip_len = htons(tlen);
2297                 h->ip_src.s_addr = saddr->v4.s_addr;
2298                 h->ip_dst.s_addr = daddr->v4.s_addr;
2299
2300                 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2301                 break;
2302 #endif /* INET */
2303 #ifdef INET6
2304         case AF_INET6:
2305                 h6 = mtod(m, struct ip6_hdr *);
2306
2307                 /* IP header fields included in the TCP checksum */
2308                 h6->ip6_nxt = IPPROTO_TCP;
2309                 h6->ip6_plen = htons(tlen);
2310                 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2311                 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2312
2313                 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2314                 break;
2315 #endif /* INET6 */
2316         }
2317
2318         /* TCP header */
2319         th->th_sport = sport;
2320         th->th_dport = dport;
2321         th->th_seq = htonl(seq);
2322         th->th_ack = htonl(ack);
2323         th->th_off = tlen >> 2;
2324         th->th_flags = flags;
2325         th->th_win = htons(win);
2326
2327         if (mss) {
2328                 opt = (char *)(th + 1);
2329                 opt[0] = TCPOPT_MAXSEG;
2330                 opt[1] = 4;
2331                 HTONS(mss);
2332                 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2333         }
2334
2335         switch (af) {
2336 #ifdef INET
2337         case AF_INET:
2338                 /* TCP checksum */
2339                 th->th_sum = in_cksum(m, len);
2340
2341                 /* Finish the IP header */
2342                 h->ip_v = 4;
2343                 h->ip_hl = sizeof(*h) >> 2;
2344                 h->ip_tos = IPTOS_LOWDELAY;
2345                 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2346                 h->ip_len = htons(len);
2347                 h->ip_ttl = ttl ? ttl : V_ip_defttl;
2348                 h->ip_sum = 0;
2349
2350                 pfse->pfse_type = PFSE_IP;
2351                 break;
2352 #endif /* INET */
2353 #ifdef INET6
2354         case AF_INET6:
2355                 /* TCP checksum */
2356                 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2357                     sizeof(struct ip6_hdr), tlen);
2358
2359                 h6->ip6_vfc |= IPV6_VERSION;
2360                 h6->ip6_hlim = IPV6_DEFHLIM;
2361
2362                 pfse->pfse_type = PFSE_IP6;
2363                 break;
2364 #endif /* INET6 */
2365         }
2366         pfse->pfse_m = m;
2367         pf_send(pfse);
2368 }
2369
2370 static void
2371 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2372     struct pf_rule *r)
2373 {
2374         struct pf_send_entry *pfse;
2375         struct mbuf *m0;
2376         struct pf_mtag *pf_mtag;
2377
2378         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2379         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2380         if (pfse == NULL)
2381                 return;
2382
2383         if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2384                 free(pfse, M_PFTEMP);
2385                 return;
2386         }
2387
2388         if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2389                 free(pfse, M_PFTEMP);
2390                 return;
2391         }
2392         /* XXX: revisit */
2393         m0->m_flags |= M_SKIP_FIREWALL;
2394
2395         if (r->rtableid >= 0)
2396                 M_SETFIB(m0, r->rtableid);
2397
2398 #ifdef ALTQ
2399         if (r->qid) {
2400                 pf_mtag->qid = r->qid;
2401                 /* add hints for ecn */
2402                 pf_mtag->hdr = mtod(m0, struct ip *);
2403         }
2404 #endif /* ALTQ */
2405
2406         switch (af) {
2407 #ifdef INET
2408         case AF_INET:
2409                 pfse->pfse_type = PFSE_ICMP;
2410                 break;
2411 #endif /* INET */
2412 #ifdef INET6
2413         case AF_INET6:
2414                 pfse->pfse_type = PFSE_ICMP6;
2415                 break;
2416 #endif /* INET6 */
2417         }
2418         pfse->pfse_m = m0;
2419         pfse->pfse_icmp_type = type;
2420         pfse->pfse_icmp_code = code;
2421         pf_send(pfse);
2422 }
2423
2424 /*
2425  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2426  * If n is 0, they match if they are equal. If n is != 0, they match if they
2427  * are different.
2428  */
2429 int
2430 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2431     struct pf_addr *b, sa_family_t af)
2432 {
2433         int     match = 0;
2434
2435         switch (af) {
2436 #ifdef INET
2437         case AF_INET:
2438                 if ((a->addr32[0] & m->addr32[0]) ==
2439                     (b->addr32[0] & m->addr32[0]))
2440                         match++;
2441                 break;
2442 #endif /* INET */
2443 #ifdef INET6
2444         case AF_INET6:
2445                 if (((a->addr32[0] & m->addr32[0]) ==
2446                      (b->addr32[0] & m->addr32[0])) &&
2447                     ((a->addr32[1] & m->addr32[1]) ==
2448                      (b->addr32[1] & m->addr32[1])) &&
2449                     ((a->addr32[2] & m->addr32[2]) ==
2450                      (b->addr32[2] & m->addr32[2])) &&
2451                     ((a->addr32[3] & m->addr32[3]) ==
2452                      (b->addr32[3] & m->addr32[3])))
2453                         match++;
2454                 break;
2455 #endif /* INET6 */
2456         }
2457         if (match) {
2458                 if (n)
2459                         return (0);
2460                 else
2461                         return (1);
2462         } else {
2463                 if (n)
2464                         return (1);
2465                 else
2466                         return (0);
2467         }
2468 }
2469
2470 /*
2471  * Return 1 if b <= a <= e, otherwise return 0.
2472  */
2473 int
2474 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2475     struct pf_addr *a, sa_family_t af)
2476 {
2477         switch (af) {
2478 #ifdef INET
2479         case AF_INET:
2480                 if ((a->addr32[0] < b->addr32[0]) ||
2481                     (a->addr32[0] > e->addr32[0]))
2482                         return (0);
2483                 break;
2484 #endif /* INET */
2485 #ifdef INET6
2486         case AF_INET6: {
2487                 int     i;
2488
2489                 /* check a >= b */
2490                 for (i = 0; i < 4; ++i)
2491                         if (a->addr32[i] > b->addr32[i])
2492                                 break;
2493                         else if (a->addr32[i] < b->addr32[i])
2494                                 return (0);
2495                 /* check a <= e */
2496                 for (i = 0; i < 4; ++i)
2497                         if (a->addr32[i] < e->addr32[i])
2498                                 break;
2499                         else if (a->addr32[i] > e->addr32[i])
2500                                 return (0);
2501                 break;
2502         }
2503 #endif /* INET6 */
2504         }
2505         return (1);
2506 }
2507
2508 static int
2509 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2510 {
2511         switch (op) {
2512         case PF_OP_IRG:
2513                 return ((p > a1) && (p < a2));
2514         case PF_OP_XRG:
2515                 return ((p < a1) || (p > a2));
2516         case PF_OP_RRG:
2517                 return ((p >= a1) && (p <= a2));
2518         case PF_OP_EQ:
2519                 return (p == a1);
2520         case PF_OP_NE:
2521                 return (p != a1);
2522         case PF_OP_LT:
2523                 return (p < a1);
2524         case PF_OP_LE:
2525                 return (p <= a1);
2526         case PF_OP_GT:
2527                 return (p > a1);
2528         case PF_OP_GE:
2529                 return (p >= a1);
2530         }
2531         return (0); /* never reached */
2532 }
2533
2534 int
2535 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2536 {
2537         NTOHS(a1);
2538         NTOHS(a2);
2539         NTOHS(p);
2540         return (pf_match(op, a1, a2, p));
2541 }
2542
2543 static int
2544 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2545 {
2546         if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2547                 return (0);
2548         return (pf_match(op, a1, a2, u));
2549 }
2550
2551 static int
2552 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2553 {
2554         if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2555                 return (0);
2556         return (pf_match(op, a1, a2, g));
2557 }
2558
2559 int
2560 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2561 {
2562         if (*tag == -1)
2563                 *tag = mtag;
2564
2565         return ((!r->match_tag_not && r->match_tag == *tag) ||
2566             (r->match_tag_not && r->match_tag != *tag));
2567 }
2568
2569 int
2570 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2571 {
2572
2573         KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2574
2575         if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2576                 return (ENOMEM);
2577
2578         pd->pf_mtag->tag = tag;
2579
2580         return (0);
2581 }
2582
2583 #define PF_ANCHOR_STACKSIZE     32
2584 struct pf_anchor_stackframe {
2585         struct pf_ruleset       *rs;
2586         struct pf_rule          *r;     /* XXX: + match bit */
2587         struct pf_anchor        *child;
2588 };
2589
2590 /*
2591  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2592  */
2593 #define PF_ANCHORSTACK_MATCH    0x00000001
2594 #define PF_ANCHORSTACK_MASK     (PF_ANCHORSTACK_MATCH)
2595
2596 #define PF_ANCHOR_MATCH(f)      ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2597 #define PF_ANCHOR_RULE(f)       (struct pf_rule *)                      \
2598                                 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2599 #define PF_ANCHOR_SET_MATCH(f)  do { (f)->r = (void *)                  \
2600                                 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2601 } while (0)
2602
2603 void
2604 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2605     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2606     int *match)
2607 {
2608         struct pf_anchor_stackframe     *f;
2609
2610         PF_RULES_RASSERT();
2611
2612         if (match)
2613                 *match = 0;
2614         if (*depth >= PF_ANCHOR_STACKSIZE) {
2615                 printf("%s: anchor stack overflow on %s\n",
2616                     __func__, (*r)->anchor->name);
2617                 *r = TAILQ_NEXT(*r, entries);
2618                 return;
2619         } else if (*depth == 0 && a != NULL)
2620                 *a = *r;
2621         f = stack + (*depth)++;
2622         f->rs = *rs;
2623         f->r = *r;
2624         if ((*r)->anchor_wildcard) {
2625                 struct pf_anchor_node *parent = &(*r)->anchor->children;
2626
2627                 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2628                         *r = NULL;
2629                         return;
2630                 }
2631                 *rs = &f->child->ruleset;
2632         } else {
2633                 f->child = NULL;
2634                 *rs = &(*r)->anchor->ruleset;
2635         }
2636         *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2637 }
2638
2639 int
2640 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2641     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2642     int *match)
2643 {
2644         struct pf_anchor_stackframe     *f;
2645         struct pf_rule *fr;
2646         int quick = 0;
2647
2648         PF_RULES_RASSERT();
2649
2650         do {
2651                 if (*depth <= 0)
2652                         break;
2653                 f = stack + *depth - 1;
2654                 fr = PF_ANCHOR_RULE(f);
2655                 if (f->child != NULL) {
2656                         struct pf_anchor_node *parent;
2657
2658                         /*
2659                          * This block traverses through
2660                          * a wildcard anchor.
2661                          */
2662                         parent = &fr->anchor->children;
2663                         if (match != NULL && *match) {
2664                                 /*
2665                                  * If any of "*" matched, then
2666                                  * "foo/ *" matched, mark frame
2667                                  * appropriately.
2668                                  */
2669                                 PF_ANCHOR_SET_MATCH(f);
2670                                 *match = 0;
2671                         }
2672                         f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2673                         if (f->child != NULL) {
2674                                 *rs = &f->child->ruleset;
2675                                 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2676                                 if (*r == NULL)
2677                                         continue;
2678                                 else
2679                                         break;
2680                         }
2681                 }
2682                 (*depth)--;
2683                 if (*depth == 0 && a != NULL)
2684                         *a = NULL;
2685                 *rs = f->rs;
2686                 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2687                         quick = fr->quick;
2688                 *r = TAILQ_NEXT(fr, entries);
2689         } while (*r == NULL);
2690
2691         return (quick);
2692 }
2693
2694 #ifdef INET6
2695 void
2696 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2697     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2698 {
2699         switch (af) {
2700 #ifdef INET
2701         case AF_INET:
2702                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2703                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2704                 break;
2705 #endif /* INET */
2706         case AF_INET6:
2707                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2708                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2709                 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2710                 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2711                 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2712                 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2713                 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2714                 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2715                 break;
2716         }
2717 }
2718
2719 void
2720 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2721 {
2722         switch (af) {
2723 #ifdef INET
2724         case AF_INET:
2725                 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2726                 break;
2727 #endif /* INET */
2728         case AF_INET6:
2729                 if (addr->addr32[3] == 0xffffffff) {
2730                         addr->addr32[3] = 0;
2731                         if (addr->addr32[2] == 0xffffffff) {
2732                                 addr->addr32[2] = 0;
2733                                 if (addr->addr32[1] == 0xffffffff) {
2734                                         addr->addr32[1] = 0;
2735                                         addr->addr32[0] =
2736                                             htonl(ntohl(addr->addr32[0]) + 1);
2737                                 } else
2738                                         addr->addr32[1] =
2739                                             htonl(ntohl(addr->addr32[1]) + 1);
2740                         } else
2741                                 addr->addr32[2] =
2742                                     htonl(ntohl(addr->addr32[2]) + 1);
2743                 } else
2744                         addr->addr32[3] =
2745                             htonl(ntohl(addr->addr32[3]) + 1);
2746                 break;
2747         }
2748 }
2749 #endif /* INET6 */
2750
2751 int
2752 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
2753 {
2754         struct pf_addr          *saddr, *daddr;
2755         u_int16_t                sport, dport;
2756         struct inpcbinfo        *pi;
2757         struct inpcb            *inp;
2758
2759         pd->lookup.uid = UID_MAX;
2760         pd->lookup.gid = GID_MAX;
2761
2762         switch (pd->proto) {
2763         case IPPROTO_TCP:
2764                 if (pd->hdr.tcp == NULL)
2765                         return (-1);
2766                 sport = pd->hdr.tcp->th_sport;
2767                 dport = pd->hdr.tcp->th_dport;
2768                 pi = &V_tcbinfo;
2769                 break;
2770         case IPPROTO_UDP:
2771                 if (pd->hdr.udp == NULL)
2772                         return (-1);
2773                 sport = pd->hdr.udp->uh_sport;
2774                 dport = pd->hdr.udp->uh_dport;
2775                 pi = &V_udbinfo;
2776                 break;
2777         default:
2778                 return (-1);
2779         }
2780         if (direction == PF_IN) {
2781                 saddr = pd->src;
2782                 daddr = pd->dst;
2783         } else {
2784                 u_int16_t       p;
2785
2786                 p = sport;
2787                 sport = dport;
2788                 dport = p;
2789                 saddr = pd->dst;
2790                 daddr = pd->src;
2791         }
2792         switch (pd->af) {
2793 #ifdef INET
2794         case AF_INET:
2795                 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
2796                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2797                 if (inp == NULL) {
2798                         inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
2799                            daddr->v4, dport, INPLOOKUP_WILDCARD |
2800                            INPLOOKUP_RLOCKPCB, NULL, m);
2801                         if (inp == NULL)
2802                                 return (-1);
2803                 }
2804                 break;
2805 #endif /* INET */
2806 #ifdef INET6
2807         case AF_INET6:
2808                 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
2809                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2810                 if (inp == NULL) {
2811                         inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
2812                             &daddr->v6, dport, INPLOOKUP_WILDCARD |
2813                             INPLOOKUP_RLOCKPCB, NULL, m);
2814                         if (inp == NULL)
2815                                 return (-1);
2816                 }
2817                 break;
2818 #endif /* INET6 */
2819
2820         default:
2821                 return (-1);
2822         }
2823         INP_RLOCK_ASSERT(inp);
2824         pd->lookup.uid = inp->inp_cred->cr_uid;
2825         pd->lookup.gid = inp->inp_cred->cr_groups[0];
2826         INP_RUNLOCK(inp);
2827
2828         return (1);
2829 }
2830
2831 static u_int8_t
2832 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2833 {
2834         int              hlen;
2835         u_int8_t         hdr[60];
2836         u_int8_t        *opt, optlen;
2837         u_int8_t         wscale = 0;
2838
2839         hlen = th_off << 2;             /* hlen <= sizeof(hdr) */
2840         if (hlen <= sizeof(struct tcphdr))
2841                 return (0);
2842         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2843                 return (0);
2844         opt = hdr + sizeof(struct tcphdr);
2845         hlen -= sizeof(struct tcphdr);
2846         while (hlen >= 3) {
2847                 switch (*opt) {
2848                 case TCPOPT_EOL:
2849                 case TCPOPT_NOP:
2850                         ++opt;
2851                         --hlen;
2852                         break;
2853                 case TCPOPT_WINDOW:
2854                         wscale = opt[2];
2855                         if (wscale > TCP_MAX_WINSHIFT)
2856                                 wscale = TCP_MAX_WINSHIFT;
2857                         wscale |= PF_WSCALE_FLAG;
2858                         /* FALLTHROUGH */
2859                 default:
2860                         optlen = opt[1];
2861                         if (optlen < 2)
2862                                 optlen = 2;
2863                         hlen -= optlen;
2864                         opt += optlen;
2865                         break;
2866                 }
2867         }
2868         return (wscale);
2869 }
2870
2871 static u_int16_t
2872 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2873 {
2874         int              hlen;
2875         u_int8_t         hdr[60];
2876         u_int8_t        *opt, optlen;
2877         u_int16_t        mss = V_tcp_mssdflt;
2878
2879         hlen = th_off << 2;     /* hlen <= sizeof(hdr) */
2880         if (hlen <= sizeof(struct tcphdr))
2881                 return (0);
2882         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2883                 return (0);
2884         opt = hdr + sizeof(struct tcphdr);
2885         hlen -= sizeof(struct tcphdr);
2886         while (hlen >= TCPOLEN_MAXSEG) {
2887                 switch (*opt) {
2888                 case TCPOPT_EOL:
2889                 case TCPOPT_NOP:
2890                         ++opt;
2891                         --hlen;
2892                         break;
2893                 case TCPOPT_MAXSEG:
2894                         bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
2895                         NTOHS(mss);
2896                         /* FALLTHROUGH */
2897                 default:
2898                         optlen = opt[1];
2899                         if (optlen < 2)
2900                                 optlen = 2;
2901                         hlen -= optlen;
2902                         opt += optlen;
2903                         break;
2904                 }
2905         }
2906         return (mss);
2907 }
2908
2909 static u_int16_t
2910 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
2911 {
2912 #ifdef INET
2913         struct sockaddr_in      *dst;
2914         struct route             ro;
2915 #endif /* INET */
2916 #ifdef INET6
2917         struct sockaddr_in6     *dst6;
2918         struct route_in6         ro6;
2919 #endif /* INET6 */
2920         struct rtentry          *rt = NULL;
2921         int                      hlen = 0;
2922         u_int16_t                mss = V_tcp_mssdflt;
2923
2924         switch (af) {
2925 #ifdef INET
2926         case AF_INET:
2927                 hlen = sizeof(struct ip);
2928                 bzero(&ro, sizeof(ro));
2929                 dst = (struct sockaddr_in *)&ro.ro_dst;
2930                 dst->sin_family = AF_INET;
2931                 dst->sin_len = sizeof(*dst);
2932                 dst->sin_addr = addr->v4;
2933                 in_rtalloc_ign(&ro, 0, rtableid);
2934                 rt = ro.ro_rt;
2935                 break;
2936 #endif /* INET */
2937 #ifdef INET6
2938         case AF_INET6:
2939                 hlen = sizeof(struct ip6_hdr);
2940                 bzero(&ro6, sizeof(ro6));
2941                 dst6 = (struct sockaddr_in6 *)&ro6.ro_dst;
2942                 dst6->sin6_family = AF_INET6;
2943                 dst6->sin6_len = sizeof(*dst6);
2944                 dst6->sin6_addr = addr->v6;
2945                 in6_rtalloc_ign(&ro6, 0, rtableid);
2946                 rt = ro6.ro_rt;
2947                 break;
2948 #endif /* INET6 */
2949         }
2950
2951         if (rt && rt->rt_ifp) {
2952                 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr);
2953                 mss = max(V_tcp_mssdflt, mss);
2954                 RTFREE(rt);
2955         }
2956         mss = min(mss, offer);
2957         mss = max(mss, 64);             /* sanity - at least max opt space */
2958         return (mss);
2959 }
2960
2961 static u_int32_t
2962 pf_tcp_iss(struct pf_pdesc *pd)
2963 {
2964         MD5_CTX ctx;
2965         u_int32_t digest[4];
2966
2967         if (V_pf_tcp_secret_init == 0) {
2968                 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
2969                 MD5Init(&V_pf_tcp_secret_ctx);
2970                 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
2971                     sizeof(V_pf_tcp_secret));
2972                 V_pf_tcp_secret_init = 1;
2973         }
2974
2975         ctx = V_pf_tcp_secret_ctx;
2976
2977         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
2978         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
2979         if (pd->af == AF_INET6) {
2980                 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
2981                 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
2982         } else {
2983                 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
2984                 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
2985         }
2986         MD5Final((u_char *)digest, &ctx);
2987         V_pf_tcp_iss_off += 4096;
2988 #define ISN_RANDOM_INCREMENT (4096 - 1)
2989         return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
2990             V_pf_tcp_iss_off);
2991 #undef  ISN_RANDOM_INCREMENT
2992 }
2993
2994 static int
2995 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
2996     struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
2997     struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
2998 {
2999         struct pf_rule          *nr = NULL;
3000         struct pf_addr          * const saddr = pd->src;
3001         struct pf_addr          * const daddr = pd->dst;
3002         sa_family_t              af = pd->af;
3003         struct pf_rule          *r, *a = NULL;
3004         struct pf_ruleset       *ruleset = NULL;
3005         struct pf_src_node      *nsn = NULL;
3006         struct tcphdr           *th = pd->hdr.tcp;
3007         struct pf_state_key     *sk = NULL, *nk = NULL;
3008         u_short                  reason;
3009         int                      rewrite = 0, hdrlen = 0;
3010         int                      tag = -1, rtableid = -1;
3011         int                      asd = 0;
3012         int                      match = 0;
3013         int                      state_icmp = 0;
3014         u_int16_t                sport = 0, dport = 0;
3015         u_int16_t                bproto_sum = 0, bip_sum = 0;
3016         u_int8_t                 icmptype = 0, icmpcode = 0;
3017         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3018
3019         PF_RULES_RASSERT();
3020
3021         if (inp != NULL) {
3022                 INP_LOCK_ASSERT(inp);
3023                 pd->lookup.uid = inp->inp_cred->cr_uid;
3024                 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3025                 pd->lookup.done = 1;
3026         }
3027
3028         switch (pd->proto) {
3029         case IPPROTO_TCP:
3030                 sport = th->th_sport;
3031                 dport = th->th_dport;
3032                 hdrlen = sizeof(*th);
3033                 break;
3034         case IPPROTO_UDP:
3035                 sport = pd->hdr.udp->uh_sport;
3036                 dport = pd->hdr.udp->uh_dport;
3037                 hdrlen = sizeof(*pd->hdr.udp);
3038                 break;
3039 #ifdef INET
3040         case IPPROTO_ICMP:
3041                 if (pd->af != AF_INET)
3042                         break;
3043                 sport = dport = pd->hdr.icmp->icmp_id;
3044                 hdrlen = sizeof(*pd->hdr.icmp);
3045                 icmptype = pd->hdr.icmp->icmp_type;
3046                 icmpcode = pd->hdr.icmp->icmp_code;
3047
3048                 if (icmptype == ICMP_UNREACH ||
3049                     icmptype == ICMP_SOURCEQUENCH ||
3050                     icmptype == ICMP_REDIRECT ||
3051                     icmptype == ICMP_TIMXCEED ||
3052                     icmptype == ICMP_PARAMPROB)
3053                         state_icmp++;
3054                 break;
3055 #endif /* INET */
3056 #ifdef INET6
3057         case IPPROTO_ICMPV6:
3058                 if (af != AF_INET6)
3059                         break;
3060                 sport = dport = pd->hdr.icmp6->icmp6_id;
3061                 hdrlen = sizeof(*pd->hdr.icmp6);
3062                 icmptype = pd->hdr.icmp6->icmp6_type;
3063                 icmpcode = pd->hdr.icmp6->icmp6_code;
3064
3065                 if (icmptype == ICMP6_DST_UNREACH ||
3066                     icmptype == ICMP6_PACKET_TOO_BIG ||
3067                     icmptype == ICMP6_TIME_EXCEEDED ||
3068                     icmptype == ICMP6_PARAM_PROB)
3069                         state_icmp++;
3070                 break;
3071 #endif /* INET6 */
3072         default:
3073                 sport = dport = hdrlen = 0;
3074                 break;
3075         }
3076
3077         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3078
3079         /* check packet for BINAT/NAT/RDR */
3080         if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3081             &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3082                 KASSERT(sk != NULL, ("%s: null sk", __func__));
3083                 KASSERT(nk != NULL, ("%s: null nk", __func__));
3084
3085                 if (pd->ip_sum)
3086                         bip_sum = *pd->ip_sum;
3087
3088                 switch (pd->proto) {
3089                 case IPPROTO_TCP:
3090                         bproto_sum = th->th_sum;
3091                         pd->proto_sum = &th->th_sum;
3092
3093                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3094                             nk->port[pd->sidx] != sport) {
3095                                 pf_change_ap(saddr, &th->th_sport, pd->ip_sum,
3096                                     &th->th_sum, &nk->addr[pd->sidx],
3097                                     nk->port[pd->sidx], 0, af);
3098                                 pd->sport = &th->th_sport;
3099                                 sport = th->th_sport;
3100                         }
3101
3102                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3103                             nk->port[pd->didx] != dport) {
3104                                 pf_change_ap(daddr, &th->th_dport, pd->ip_sum,
3105                                     &th->th_sum, &nk->addr[pd->didx],
3106                                     nk->port[pd->didx], 0, af);
3107                                 dport = th->th_dport;
3108                                 pd->dport = &th->th_dport;
3109                         }
3110                         rewrite++;
3111                         break;
3112                 case IPPROTO_UDP:
3113                         bproto_sum = pd->hdr.udp->uh_sum;
3114                         pd->proto_sum = &pd->hdr.udp->uh_sum;
3115
3116                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3117                             nk->port[pd->sidx] != sport) {
3118                                 pf_change_ap(saddr, &pd->hdr.udp->uh_sport,
3119                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3120                                     &nk->addr[pd->sidx],
3121                                     nk->port[pd->sidx], 1, af);
3122                                 sport = pd->hdr.udp->uh_sport;
3123                                 pd->sport = &pd->hdr.udp->uh_sport;
3124                         }
3125
3126                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3127                             nk->port[pd->didx] != dport) {
3128                                 pf_change_ap(daddr, &pd->hdr.udp->uh_dport,
3129                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3130                                     &nk->addr[pd->didx],
3131                                     nk->port[pd->didx], 1, af);
3132                                 dport = pd->hdr.udp->uh_dport;
3133                                 pd->dport = &pd->hdr.udp->uh_dport;
3134                         }
3135                         rewrite++;
3136                         break;
3137 #ifdef INET
3138                 case IPPROTO_ICMP:
3139                         nk->port[0] = nk->port[1];
3140                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3141                                 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3142                                     nk->addr[pd->sidx].v4.s_addr, 0);
3143
3144                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3145                                 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3146                                     nk->addr[pd->didx].v4.s_addr, 0);
3147
3148                         if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3149                                 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3150                                     pd->hdr.icmp->icmp_cksum, sport,
3151                                     nk->port[1], 0);
3152                                 pd->hdr.icmp->icmp_id = nk->port[1];
3153                                 pd->sport = &pd->hdr.icmp->icmp_id;
3154                         }
3155                         m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3156                         break;
3157 #endif /* INET */
3158 #ifdef INET6
3159                 case IPPROTO_ICMPV6:
3160                         nk->port[0] = nk->port[1];
3161                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3162                                 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3163                                     &nk->addr[pd->sidx], 0);
3164
3165                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3166                                 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3167                                     &nk->addr[pd->didx], 0);
3168                         rewrite++;
3169                         break;
3170 #endif /* INET */
3171                 default:
3172                         switch (af) {
3173 #ifdef INET
3174                         case AF_INET:
3175                                 if (PF_ANEQ(saddr,
3176                                     &nk->addr[pd->sidx], AF_INET))
3177                                         pf_change_a(&saddr->v4.s_addr,
3178                                             pd->ip_sum,
3179                                             nk->addr[pd->sidx].v4.s_addr, 0);
3180
3181                                 if (PF_ANEQ(daddr,
3182                                     &nk->addr[pd->didx], AF_INET))
3183                                         pf_change_a(&daddr->v4.s_addr,
3184                                             pd->ip_sum,
3185                                             nk->addr[pd->didx].v4.s_addr, 0);
3186                                 break;
3187 #endif /* INET */
3188 #ifdef INET6
3189                         case AF_INET6:
3190                                 if (PF_ANEQ(saddr,
3191                                     &nk->addr[pd->sidx], AF_INET6))
3192                                         PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3193
3194                                 if (PF_ANEQ(daddr,
3195                                     &nk->addr[pd->didx], AF_INET6))
3196                                         PF_ACPY(saddr, &nk->addr[pd->didx], af);
3197                                 break;
3198 #endif /* INET */
3199                         }
3200                         break;
3201                 }
3202                 if (nr->natpass)
3203                         r = NULL;
3204                 pd->nat_rule = nr;
3205         }
3206
3207         while (r != NULL) {
3208                 r->evaluations++;
3209                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3210                         r = r->skip[PF_SKIP_IFP].ptr;
3211                 else if (r->direction && r->direction != direction)
3212                         r = r->skip[PF_SKIP_DIR].ptr;
3213                 else if (r->af && r->af != af)
3214                         r = r->skip[PF_SKIP_AF].ptr;
3215                 else if (r->proto && r->proto != pd->proto)
3216                         r = r->skip[PF_SKIP_PROTO].ptr;
3217                 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3218                     r->src.neg, kif, M_GETFIB(m)))
3219                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3220                 /* tcp/udp only. port_op always 0 in other cases */
3221                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3222                     r->src.port[0], r->src.port[1], sport))
3223                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
3224                 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3225                     r->dst.neg, NULL, M_GETFIB(m)))
3226                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3227                 /* tcp/udp only. port_op always 0 in other cases */
3228                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3229                     r->dst.port[0], r->dst.port[1], dport))
3230                         r = r->skip[PF_SKIP_DST_PORT].ptr;
3231                 /* icmp only. type always 0 in other cases */
3232                 else if (r->type && r->type != icmptype + 1)
3233                         r = TAILQ_NEXT(r, entries);
3234                 /* icmp only. type always 0 in other cases */
3235                 else if (r->code && r->code != icmpcode + 1)
3236                         r = TAILQ_NEXT(r, entries);
3237                 else if (r->tos && !(r->tos == pd->tos))
3238                         r = TAILQ_NEXT(r, entries);
3239                 else if (r->rule_flag & PFRULE_FRAGMENT)
3240                         r = TAILQ_NEXT(r, entries);
3241                 else if (pd->proto == IPPROTO_TCP &&
3242                     (r->flagset & th->th_flags) != r->flags)
3243                         r = TAILQ_NEXT(r, entries);
3244                 /* tcp/udp only. uid.op always 0 in other cases */
3245                 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3246                     pf_socket_lookup(direction, pd, m), 1)) &&
3247                     !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3248                     pd->lookup.uid))
3249                         r = TAILQ_NEXT(r, entries);
3250                 /* tcp/udp only. gid.op always 0 in other cases */
3251                 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3252                     pf_socket_lookup(direction, pd, m), 1)) &&
3253                     !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3254                     pd->lookup.gid))
3255                         r = TAILQ_NEXT(r, entries);
3256                 else if (r->prob &&
3257                     r->prob <= arc4random())
3258                         r = TAILQ_NEXT(r, entries);
3259                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3260                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3261                         r = TAILQ_NEXT(r, entries);
3262                 else if (r->os_fingerprint != PF_OSFP_ANY &&
3263                     (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3264                     pf_osfp_fingerprint(pd, m, off, th),
3265                     r->os_fingerprint)))
3266                         r = TAILQ_NEXT(r, entries);
3267                 else {
3268                         if (r->tag)
3269                                 tag = r->tag;
3270                         if (r->rtableid >= 0)
3271                                 rtableid = r->rtableid;
3272                         if (r->anchor == NULL) {
3273                                 match = 1;
3274                                 *rm = r;
3275                                 *am = a;
3276                                 *rsm = ruleset;
3277                                 if ((*rm)->quick)
3278                                         break;
3279                                 r = TAILQ_NEXT(r, entries);
3280                         } else
3281                                 pf_step_into_anchor(anchor_stack, &asd,
3282                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3283                                     &match);
3284                 }
3285                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3286                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3287                         break;
3288         }
3289         r = *rm;
3290         a = *am;
3291         ruleset = *rsm;
3292
3293         REASON_SET(&reason, PFRES_MATCH);
3294
3295         if (r->log || (nr != NULL && nr->log)) {
3296                 if (rewrite)
3297                         m_copyback(m, off, hdrlen, pd->hdr.any);
3298                 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3299                     ruleset, pd, 1);
3300         }
3301
3302         if ((r->action == PF_DROP) &&
3303             ((r->rule_flag & PFRULE_RETURNRST) ||
3304             (r->rule_flag & PFRULE_RETURNICMP) ||
3305             (r->rule_flag & PFRULE_RETURN))) {
3306                 /* undo NAT changes, if they have taken place */
3307                 if (nr != NULL) {
3308                         PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3309                         PF_ACPY(daddr, &sk->addr[pd->didx], af);
3310                         if (pd->sport)
3311                                 *pd->sport = sk->port[pd->sidx];
3312                         if (pd->dport)
3313                                 *pd->dport = sk->port[pd->didx];
3314                         if (pd->proto_sum)
3315                                 *pd->proto_sum = bproto_sum;
3316                         if (pd->ip_sum)
3317                                 *pd->ip_sum = bip_sum;
3318                         m_copyback(m, off, hdrlen, pd->hdr.any);
3319                 }
3320                 if (pd->proto == IPPROTO_TCP &&
3321                     ((r->rule_flag & PFRULE_RETURNRST) ||
3322                     (r->rule_flag & PFRULE_RETURN)) &&
3323                     !(th->th_flags & TH_RST)) {
3324                         u_int32_t        ack = ntohl(th->th_seq) + pd->p_len;
3325                         int              len = 0;
3326 #ifdef INET
3327                         struct ip       *h4;
3328 #endif
3329 #ifdef INET6
3330                         struct ip6_hdr  *h6;
3331 #endif
3332
3333                         switch (af) {
3334 #ifdef INET
3335                         case AF_INET:
3336                                 h4 = mtod(m, struct ip *);
3337                                 len = ntohs(h4->ip_len) - off;
3338                                 break;
3339 #endif
3340 #ifdef INET6
3341                         case AF_INET6:
3342                                 h6 = mtod(m, struct ip6_hdr *);
3343                                 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3344                                 break;
3345 #endif
3346                         }
3347
3348                         if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3349                                 REASON_SET(&reason, PFRES_PROTCKSUM);
3350                         else {
3351                                 if (th->th_flags & TH_SYN)
3352                                         ack++;
3353                                 if (th->th_flags & TH_FIN)
3354                                         ack++;
3355                                 pf_send_tcp(m, r, af, pd->dst,
3356                                     pd->src, th->th_dport, th->th_sport,
3357                                     ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3358                                     r->return_ttl, 1, 0, kif->pfik_ifp);
3359                         }
3360                 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3361                     r->return_icmp)
3362                         pf_send_icmp(m, r->return_icmp >> 8,
3363                             r->return_icmp & 255, af, r);
3364                 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3365                     r->return_icmp6)
3366                         pf_send_icmp(m, r->return_icmp6 >> 8,
3367                             r->return_icmp6 & 255, af, r);
3368         }
3369
3370         if (r->action == PF_DROP)
3371                 goto cleanup;
3372
3373         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3374                 REASON_SET(&reason, PFRES_MEMORY);
3375                 goto cleanup;
3376         }
3377         if (rtableid >= 0)
3378                 M_SETFIB(m, rtableid);
3379
3380         if (!state_icmp && (r->keep_state || nr != NULL ||
3381             (pd->flags & PFDESC_TCP_NORM))) {
3382                 int action;
3383                 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3384                     sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3385                     hdrlen);
3386                 if (action != PF_PASS)
3387                         return (action);
3388         } else {
3389                 if (sk != NULL)
3390                         uma_zfree(V_pf_state_key_z, sk);
3391                 if (nk != NULL)
3392                         uma_zfree(V_pf_state_key_z, nk);
3393         }
3394
3395         /* copy back packet headers if we performed NAT operations */
3396         if (rewrite)
3397                 m_copyback(m, off, hdrlen, pd->hdr.any);
3398
3399         if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3400             direction == PF_OUT &&
3401             pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m))
3402                 /*
3403                  * We want the state created, but we dont
3404                  * want to send this in case a partner
3405                  * firewall has to know about it to allow
3406                  * replies through it.
3407                  */
3408                 return (PF_DEFER);
3409
3410         return (PF_PASS);
3411
3412 cleanup:
3413         if (sk != NULL)
3414                 uma_zfree(V_pf_state_key_z, sk);
3415         if (nk != NULL)
3416                 uma_zfree(V_pf_state_key_z, nk);
3417         return (PF_DROP);
3418 }
3419
3420 static int
3421 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3422     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3423     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3424     u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3425     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3426 {
3427         struct pf_state         *s = NULL;
3428         struct pf_src_node      *sn = NULL;
3429         struct tcphdr           *th = pd->hdr.tcp;
3430         u_int16_t                mss = V_tcp_mssdflt;
3431         u_short                  reason;
3432
3433         /* check maximums */
3434         if (r->max_states &&
3435             (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3436                 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
3437                 REASON_SET(&reason, PFRES_MAXSTATES);
3438                 return (PF_DROP);
3439         }
3440         /* src node for filter rule */
3441         if ((r->rule_flag & PFRULE_SRCTRACK ||
3442             r->rpool.opts & PF_POOL_STICKYADDR) &&
3443             pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3444                 REASON_SET(&reason, PFRES_SRCLIMIT);
3445                 goto csfailed;
3446         }
3447         /* src node for translation rule */
3448         if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3449             pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3450                 REASON_SET(&reason, PFRES_SRCLIMIT);
3451                 goto csfailed;
3452         }
3453         s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3454         if (s == NULL) {
3455                 REASON_SET(&reason, PFRES_MEMORY);
3456                 goto csfailed;
3457         }
3458         s->rule.ptr = r;
3459         s->nat_rule.ptr = nr;
3460         s->anchor.ptr = a;
3461         STATE_INC_COUNTERS(s);
3462         if (r->allow_opts)
3463                 s->state_flags |= PFSTATE_ALLOWOPTS;
3464         if (r->rule_flag & PFRULE_STATESLOPPY)
3465                 s->state_flags |= PFSTATE_SLOPPY;
3466         s->log = r->log & PF_LOG_ALL;
3467         s->sync_state = PFSYNC_S_NONE;
3468         if (nr != NULL)
3469                 s->log |= nr->log & PF_LOG_ALL;
3470         switch (pd->proto) {
3471         case IPPROTO_TCP:
3472                 s->src.seqlo = ntohl(th->th_seq);
3473                 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3474                 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3475                     r->keep_state == PF_STATE_MODULATE) {
3476                         /* Generate sequence number modulator */
3477                         if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3478                             0)
3479                                 s->src.seqdiff = 1;
3480                         pf_change_a(&th->th_seq, &th->th_sum,
3481                             htonl(s->src.seqlo + s->src.seqdiff), 0);
3482                         *rewrite = 1;
3483                 } else
3484                         s->src.seqdiff = 0;
3485                 if (th->th_flags & TH_SYN) {
3486                         s->src.seqhi++;
3487                         s->src.wscale = pf_get_wscale(m, off,
3488                             th->th_off, pd->af);
3489                 }
3490                 s->src.max_win = MAX(ntohs(th->th_win), 1);
3491                 if (s->src.wscale & PF_WSCALE_MASK) {
3492                         /* Remove scale factor from initial window */
3493                         int win = s->src.max_win;
3494                         win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3495                         s->src.max_win = (win - 1) >>
3496                             (s->src.wscale & PF_WSCALE_MASK);
3497                 }
3498                 if (th->th_flags & TH_FIN)
3499                         s->src.seqhi++;
3500                 s->dst.seqhi = 1;
3501                 s->dst.max_win = 1;
3502                 s->src.state = TCPS_SYN_SENT;
3503                 s->dst.state = TCPS_CLOSED;
3504                 s->timeout = PFTM_TCP_FIRST_PACKET;
3505                 break;
3506         case IPPROTO_UDP:
3507                 s->src.state = PFUDPS_SINGLE;
3508                 s->dst.state = PFUDPS_NO_TRAFFIC;
3509                 s->timeout = PFTM_UDP_FIRST_PACKET;
3510                 break;
3511         case IPPROTO_ICMP:
3512 #ifdef INET6
3513         case IPPROTO_ICMPV6:
3514 #endif
3515                 s->timeout = PFTM_ICMP_FIRST_PACKET;
3516                 break;
3517         default:
3518                 s->src.state = PFOTHERS_SINGLE;
3519                 s->dst.state = PFOTHERS_NO_TRAFFIC;
3520                 s->timeout = PFTM_OTHER_FIRST_PACKET;
3521         }
3522
3523         if (r->rt && r->rt != PF_FASTROUTE) {
3524                 struct pf_src_node *sn = NULL;
3525
3526                 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) {
3527                         REASON_SET(&reason, PFRES_MAPFAILED);
3528                         pf_src_tree_remove_state(s);
3529                         STATE_DEC_COUNTERS(s);
3530                         uma_zfree(V_pf_state_z, s);
3531                         goto csfailed;
3532                 }
3533                 s->rt_kif = r->rpool.cur->kif;
3534         }
3535
3536         s->creation = time_uptime;
3537         s->expire = time_uptime;
3538
3539         if (sn != NULL) {
3540                 s->src_node = sn;
3541                 s->src_node->states++;
3542         }
3543         if (nsn != NULL) {
3544                 /* XXX We only modify one side for now. */
3545                 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3546                 s->nat_src_node = nsn;
3547                 s->nat_src_node->states++;
3548         }
3549         if (pd->proto == IPPROTO_TCP) {
3550                 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3551                     off, pd, th, &s->src, &s->dst)) {
3552                         REASON_SET(&reason, PFRES_MEMORY);
3553                         pf_src_tree_remove_state(s);
3554                         STATE_DEC_COUNTERS(s);
3555                         uma_zfree(V_pf_state_z, s);
3556                         return (PF_DROP);
3557                 }
3558                 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3559                     pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3560                     &s->src, &s->dst, rewrite)) {
3561                         /* This really shouldn't happen!!! */
3562                         DPFPRINTF(PF_DEBUG_URGENT,
3563                             ("pf_normalize_tcp_stateful failed on first pkt"));
3564                         pf_normalize_tcp_cleanup(s);
3565                         pf_src_tree_remove_state(s);
3566                         STATE_DEC_COUNTERS(s);
3567                         uma_zfree(V_pf_state_z, s);
3568                         return (PF_DROP);
3569                 }
3570         }
3571         s->direction = pd->dir;
3572
3573         /*
3574          * sk/nk could already been setup by pf_get_translation().
3575          */
3576         if (nr == NULL) {
3577                 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3578                     __func__, nr, sk, nk));
3579                 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3580                 if (sk == NULL)
3581                         goto csfailed;
3582                 nk = sk;
3583         } else
3584                 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3585                     __func__, nr, sk, nk));
3586
3587         /* Swap sk/nk for PF_OUT. */
3588         if (pf_state_insert(BOUND_IFACE(r, kif),
3589             (pd->dir == PF_IN) ? sk : nk,
3590             (pd->dir == PF_IN) ? nk : sk, s)) {
3591                 if (pd->proto == IPPROTO_TCP)
3592                         pf_normalize_tcp_cleanup(s);
3593                 REASON_SET(&reason, PFRES_STATEINS);
3594                 pf_src_tree_remove_state(s);
3595                 STATE_DEC_COUNTERS(s);
3596                 uma_zfree(V_pf_state_z, s);
3597                 return (PF_DROP);
3598         } else
3599                 *sm = s;
3600
3601         if (tag > 0)
3602                 s->tag = tag;
3603         if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3604             TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3605                 s->src.state = PF_TCPS_PROXY_SRC;
3606                 /* undo NAT changes, if they have taken place */
3607                 if (nr != NULL) {
3608                         struct pf_state_key *skt = s->key[PF_SK_WIRE];
3609                         if (pd->dir == PF_OUT)
3610                                 skt = s->key[PF_SK_STACK];
3611                         PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3612                         PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3613                         if (pd->sport)
3614                                 *pd->sport = skt->port[pd->sidx];
3615                         if (pd->dport)
3616                                 *pd->dport = skt->port[pd->didx];
3617                         if (pd->proto_sum)
3618                                 *pd->proto_sum = bproto_sum;
3619                         if (pd->ip_sum)
3620                                 *pd->ip_sum = bip_sum;
3621                         m_copyback(m, off, hdrlen, pd->hdr.any);
3622                 }
3623                 s->src.seqhi = htonl(arc4random());
3624                 /* Find mss option */
3625                 int rtid = M_GETFIB(m);
3626                 mss = pf_get_mss(m, off, th->th_off, pd->af);
3627                 mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3628                 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3629                 s->src.mss = mss;
3630                 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3631                     th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3632                     TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3633                 REASON_SET(&reason, PFRES_SYNPROXY);
3634                 return (PF_SYNPROXY_DROP);
3635         }
3636
3637         return (PF_PASS);
3638
3639 csfailed:
3640         if (sk != NULL)
3641                 uma_zfree(V_pf_state_key_z, sk);
3642         if (nk != NULL)
3643                 uma_zfree(V_pf_state_key_z, nk);
3644
3645         if (sn != NULL && sn->states == 0 && sn->expire == 0) {
3646                 pf_unlink_src_node(sn);
3647                 pf_free_src_node(sn);
3648         }
3649
3650         if (nsn != sn && nsn != NULL && nsn->states == 0 && nsn->expire == 0) {
3651                 pf_unlink_src_node(nsn);
3652                 pf_free_src_node(nsn);
3653         }
3654
3655         return (PF_DROP);
3656 }
3657
3658 static int
3659 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
3660     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
3661     struct pf_ruleset **rsm)
3662 {
3663         struct pf_rule          *r, *a = NULL;
3664         struct pf_ruleset       *ruleset = NULL;
3665         sa_family_t              af = pd->af;
3666         u_short                  reason;
3667         int                      tag = -1;
3668         int                      asd = 0;
3669         int                      match = 0;
3670         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3671
3672         PF_RULES_RASSERT();
3673
3674         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3675         while (r != NULL) {
3676                 r->evaluations++;
3677                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3678                         r = r->skip[PF_SKIP_IFP].ptr;
3679                 else if (r->direction && r->direction != direction)
3680                         r = r->skip[PF_SKIP_DIR].ptr;
3681                 else if (r->af && r->af != af)
3682                         r = r->skip[PF_SKIP_AF].ptr;
3683                 else if (r->proto && r->proto != pd->proto)
3684                         r = r->skip[PF_SKIP_PROTO].ptr;
3685                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3686                     r->src.neg, kif, M_GETFIB(m)))
3687                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3688                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3689                     r->dst.neg, NULL, M_GETFIB(m)))
3690                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3691                 else if (r->tos && !(r->tos == pd->tos))
3692                         r = TAILQ_NEXT(r, entries);
3693                 else if (r->os_fingerprint != PF_OSFP_ANY)
3694                         r = TAILQ_NEXT(r, entries);
3695                 else if (pd->proto == IPPROTO_UDP &&
3696                     (r->src.port_op || r->dst.port_op))
3697                         r = TAILQ_NEXT(r, entries);
3698                 else if (pd->proto == IPPROTO_TCP &&
3699                     (r->src.port_op || r->dst.port_op || r->flagset))
3700                         r = TAILQ_NEXT(r, entries);
3701                 else if ((pd->proto == IPPROTO_ICMP ||
3702                     pd->proto == IPPROTO_ICMPV6) &&
3703                     (r->type || r->code))
3704                         r = TAILQ_NEXT(r, entries);
3705                 else if (r->prob && r->prob <=
3706                     (arc4random() % (UINT_MAX - 1) + 1))
3707                         r = TAILQ_NEXT(r, entries);
3708                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3709                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3710                         r = TAILQ_NEXT(r, entries);
3711                 else {
3712                         if (r->anchor == NULL) {
3713                                 match = 1;
3714                                 *rm = r;
3715                                 *am = a;
3716                                 *rsm = ruleset;
3717                                 if ((*rm)->quick)
3718                                         break;
3719                                 r = TAILQ_NEXT(r, entries);
3720                         } else
3721                                 pf_step_into_anchor(anchor_stack, &asd,
3722                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3723                                     &match);
3724                 }
3725                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3726                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3727                         break;
3728         }
3729         r = *rm;
3730         a = *am;
3731         ruleset = *rsm;
3732
3733         REASON_SET(&reason, PFRES_MATCH);
3734
3735         if (r->log)
3736                 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
3737                     1);
3738
3739         if (r->action != PF_PASS)
3740                 return (PF_DROP);
3741
3742         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3743                 REASON_SET(&reason, PFRES_MEMORY);
3744                 return (PF_DROP);
3745         }
3746
3747         return (PF_PASS);
3748 }
3749
3750 static int
3751 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
3752         struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
3753         struct pf_pdesc *pd, u_short *reason, int *copyback)
3754 {
3755         struct tcphdr           *th = pd->hdr.tcp;
3756         u_int16_t                win = ntohs(th->th_win);
3757         u_int32_t                ack, end, seq, orig_seq;
3758         u_int8_t                 sws, dws;
3759         int                      ackskew;
3760
3761         if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
3762                 sws = src->wscale & PF_WSCALE_MASK;
3763                 dws = dst->wscale & PF_WSCALE_MASK;
3764         } else
3765                 sws = dws = 0;
3766
3767         /*
3768          * Sequence tracking algorithm from Guido van Rooij's paper:
3769          *   http://www.madison-gurkha.com/publications/tcp_filtering/
3770          *      tcp_filtering.ps
3771          */
3772
3773         orig_seq = seq = ntohl(th->th_seq);
3774         if (src->seqlo == 0) {
3775                 /* First packet from this end. Set its state */
3776
3777                 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
3778                     src->scrub == NULL) {
3779                         if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
3780                                 REASON_SET(reason, PFRES_MEMORY);
3781                                 return (PF_DROP);
3782                         }
3783                 }
3784
3785                 /* Deferred generation of sequence number modulator */
3786                 if (dst->seqdiff && !src->seqdiff) {
3787                         /* use random iss for the TCP server */
3788                         while ((src->seqdiff = arc4random() - seq) == 0)
3789                                 ;
3790                         ack = ntohl(th->th_ack) - dst->seqdiff;
3791                         pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
3792                             src->seqdiff), 0);
3793                         pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
3794                         *copyback = 1;
3795                 } else {
3796                         ack = ntohl(th->th_ack);
3797                 }
3798
3799                 end = seq + pd->p_len;
3800                 if (th->th_flags & TH_SYN) {
3801                         end++;
3802                         if (dst->wscale & PF_WSCALE_FLAG) {
3803                                 src->wscale = pf_get_wscale(m, off, th->th_off,
3804                                     pd->af);
3805                                 if (src->wscale & PF_WSCALE_FLAG) {
3806                                         /* Remove scale factor from initial
3807                                          * window */
3808                                         sws = src->wscale & PF_WSCALE_MASK;
3809                                         win = ((u_int32_t)win + (1 << sws) - 1)
3810                                             >> sws;
3811                                         dws = dst->wscale & PF_WSCALE_MASK;
3812                                 } else {
3813                                         /* fixup other window */
3814                                         dst->max_win <<= dst->wscale &
3815                                             PF_WSCALE_MASK;
3816                                         /* in case of a retrans SYN|ACK */
3817                                         dst->wscale = 0;
3818                                 }
3819                         }
3820                 }
3821                 if (th->th_flags & TH_FIN)
3822                         end++;
3823
3824                 src->seqlo = seq;
3825                 if (src->state < TCPS_SYN_SENT)
3826                         src->state = TCPS_SYN_SENT;
3827
3828                 /*
3829                  * May need to slide the window (seqhi may have been set by
3830                  * the crappy stack check or if we picked up the connection
3831                  * after establishment)
3832                  */
3833                 if (src->seqhi == 1 ||
3834                     SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
3835                         src->seqhi = end + MAX(1, dst->max_win << dws);
3836                 if (win > src->max_win)
3837                         src->max_win = win;
3838
3839         } else {
3840                 ack = ntohl(th->th_ack) - dst->seqdiff;
3841                 if (src->seqdiff) {
3842                         /* Modulate sequence numbers */
3843                         pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
3844                             src->seqdiff), 0);
3845                         pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
3846                         *copyback = 1;
3847                 }
3848                 end = seq + pd->p_len;
3849                 if (th->th_flags & TH_SYN)
3850                         end++;
3851                 if (th->th_flags & TH_FIN)
3852                         end++;
3853         }
3854
3855         if ((th->th_flags & TH_ACK) == 0) {
3856                 /* Let it pass through the ack skew check */
3857                 ack = dst->seqlo;
3858         } else if ((ack == 0 &&
3859             (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
3860             /* broken tcp stacks do not set ack */
3861             (dst->state < TCPS_SYN_SENT)) {
3862                 /*
3863                  * Many stacks (ours included) will set the ACK number in an
3864                  * FIN|ACK if the SYN times out -- no sequence to ACK.
3865                  */
3866                 ack = dst->seqlo;
3867         }
3868
3869         if (seq == end) {
3870                 /* Ease sequencing restrictions on no data packets */
3871                 seq = src->seqlo;
3872                 end = seq;
3873         }
3874
3875         ackskew = dst->seqlo - ack;
3876
3877
3878         /*
3879          * Need to demodulate the sequence numbers in any TCP SACK options
3880          * (Selective ACK). We could optionally validate the SACK values
3881          * against the current ACK window, either forwards or backwards, but
3882          * I'm not confident that SACK has been implemented properly
3883          * everywhere. It wouldn't surprise me if several stacks accidently
3884          * SACK too far backwards of previously ACKed data. There really aren't
3885          * any security implications of bad SACKing unless the target stack
3886          * doesn't validate the option length correctly. Someone trying to
3887          * spoof into a TCP connection won't bother blindly sending SACK
3888          * options anyway.
3889          */
3890         if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
3891                 if (pf_modulate_sack(m, off, pd, th, dst))
3892                         *copyback = 1;
3893         }
3894
3895
3896 #define MAXACKWINDOW (0xffff + 1500)    /* 1500 is an arbitrary fudge factor */
3897         if (SEQ_GEQ(src->seqhi, end) &&
3898             /* Last octet inside other's window space */
3899             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
3900             /* Retrans: not more than one window back */
3901             (ackskew >= -MAXACKWINDOW) &&
3902             /* Acking not more than one reassembled fragment backwards */
3903             (ackskew <= (MAXACKWINDOW << sws)) &&
3904             /* Acking not more than one window forward */
3905             ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
3906             (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
3907             (pd->flags & PFDESC_IP_REAS) == 0)) {
3908             /* Require an exact/+1 sequence match on resets when possible */
3909
3910                 if (dst->scrub || src->scrub) {
3911                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
3912                             *state, src, dst, copyback))
3913                                 return (PF_DROP);
3914                 }
3915
3916                 /* update max window */
3917                 if (src->max_win < win)
3918                         src->max_win = win;
3919                 /* synchronize sequencing */
3920                 if (SEQ_GT(end, src->seqlo))
3921                         src->seqlo = end;
3922                 /* slide the window of what the other end can send */
3923                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
3924                         dst->seqhi = ack + MAX((win << sws), 1);
3925
3926
3927                 /* update states */
3928                 if (th->th_flags & TH_SYN)
3929                         if (src->state < TCPS_SYN_SENT)
3930                                 src->state = TCPS_SYN_SENT;
3931                 if (th->th_flags & TH_FIN)
3932                         if (src->state < TCPS_CLOSING)
3933                                 src->state = TCPS_CLOSING;
3934                 if (th->th_flags & TH_ACK) {
3935                         if (dst->state == TCPS_SYN_SENT) {
3936                                 dst->state = TCPS_ESTABLISHED;
3937                                 if (src->state == TCPS_ESTABLISHED &&
3938                                     (*state)->src_node != NULL &&
3939                                     pf_src_connlimit(state)) {
3940                                         REASON_SET(reason, PFRES_SRCLIMIT);
3941                                         return (PF_DROP);
3942                                 }
3943                         } else if (dst->state == TCPS_CLOSING)
3944                                 dst->state = TCPS_FIN_WAIT_2;
3945                 }
3946                 if (th->th_flags & TH_RST)
3947                         src->state = dst->state = TCPS_TIME_WAIT;
3948
3949                 /* update expire time */
3950                 (*state)->expire = time_uptime;
3951                 if (src->state >= TCPS_FIN_WAIT_2 &&
3952                     dst->state >= TCPS_FIN_WAIT_2)
3953                         (*state)->timeout = PFTM_TCP_CLOSED;
3954                 else if (src->state >= TCPS_CLOSING &&
3955                     dst->state >= TCPS_CLOSING)
3956                         (*state)->timeout = PFTM_TCP_FIN_WAIT;
3957                 else if (src->state < TCPS_ESTABLISHED ||
3958                     dst->state < TCPS_ESTABLISHED)
3959                         (*state)->timeout = PFTM_TCP_OPENING;
3960                 else if (src->state >= TCPS_CLOSING ||
3961                     dst->state >= TCPS_CLOSING)
3962                         (*state)->timeout = PFTM_TCP_CLOSING;
3963                 else
3964                         (*state)->timeout = PFTM_TCP_ESTABLISHED;
3965
3966                 /* Fall through to PASS packet */
3967
3968         } else if ((dst->state < TCPS_SYN_SENT ||
3969                 dst->state >= TCPS_FIN_WAIT_2 ||
3970                 src->state >= TCPS_FIN_WAIT_2) &&
3971             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
3972             /* Within a window forward of the originating packet */
3973             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
3974             /* Within a window backward of the originating packet */
3975
3976                 /*
3977                  * This currently handles three situations:
3978                  *  1) Stupid stacks will shotgun SYNs before their peer
3979                  *     replies.
3980                  *  2) When PF catches an already established stream (the
3981                  *     firewall rebooted, the state table was flushed, routes
3982                  *     changed...)
3983                  *  3) Packets get funky immediately after the connection
3984                  *     closes (this should catch Solaris spurious ACK|FINs
3985                  *     that web servers like to spew after a close)
3986                  *
3987                  * This must be a little more careful than the above code
3988                  * since packet floods will also be caught here. We don't
3989                  * update the TTL here to mitigate the damage of a packet
3990                  * flood and so the same code can handle awkward establishment
3991                  * and a loosened connection close.
3992                  * In the establishment case, a correct peer response will
3993                  * validate the connection, go through the normal state code
3994                  * and keep updating the state TTL.
3995                  */
3996
3997                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
3998                         printf("pf: loose state match: ");
3999                         pf_print_state(*state);
4000                         pf_print_flags(th->th_flags);
4001                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4002                             "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
4003                             pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
4004                             (unsigned long long)(*state)->packets[1],
4005                             pd->dir == PF_IN ? "in" : "out",
4006                             pd->dir == (*state)->direction ? "fwd" : "rev");
4007                 }
4008
4009                 if (dst->scrub || src->scrub) {
4010                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4011                             *state, src, dst, copyback))
4012                                 return (PF_DROP);
4013                 }
4014
4015                 /* update max window */
4016                 if (src->max_win < win)
4017                         src->max_win = win;
4018                 /* synchronize sequencing */
4019                 if (SEQ_GT(end, src->seqlo))
4020                         src->seqlo = end;
4021                 /* slide the window of what the other end can send */
4022                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4023                         dst->seqhi = ack + MAX((win << sws), 1);
4024
4025                 /*
4026                  * Cannot set dst->seqhi here since this could be a shotgunned
4027                  * SYN and not an already established connection.
4028                  */
4029
4030                 if (th->th_flags & TH_FIN)
4031                         if (src->state < TCPS_CLOSING)
4032                                 src->state = TCPS_CLOSING;
4033                 if (th->th_flags & TH_RST)
4034                         src->state = dst->state = TCPS_TIME_WAIT;
4035
4036                 /* Fall through to PASS packet */
4037
4038         } else {
4039                 if ((*state)->dst.state == TCPS_SYN_SENT &&
4040                     (*state)->src.state == TCPS_SYN_SENT) {
4041                         /* Send RST for state mismatches during handshake */
4042                         if (!(th->th_flags & TH_RST))
4043                                 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4044                                     pd->dst, pd->src, th->th_dport,
4045                                     th->th_sport, ntohl(th->th_ack), 0,
4046                                     TH_RST, 0, 0,
4047                                     (*state)->rule.ptr->return_ttl, 1, 0,
4048                                     kif->pfik_ifp);
4049                         src->seqlo = 0;
4050                         src->seqhi = 1;
4051                         src->max_win = 1;
4052                 } else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4053                         printf("pf: BAD state: ");
4054                         pf_print_state(*state);
4055                         pf_print_flags(th->th_flags);
4056                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4057                             "pkts=%llu:%llu dir=%s,%s\n",
4058                             seq, orig_seq, ack, pd->p_len, ackskew,
4059                             (unsigned long long)(*state)->packets[0],
4060                             (unsigned long long)(*state)->packets[1],
4061                             pd->dir == PF_IN ? "in" : "out",
4062                             pd->dir == (*state)->direction ? "fwd" : "rev");
4063                         printf("pf: State failure on: %c %c %c %c | %c %c\n",
4064                             SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4065                             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4066                             ' ': '2',
4067                             (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4068                             (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4069                             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4070                             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4071                 }
4072                 REASON_SET(reason, PFRES_BADSTATE);
4073                 return (PF_DROP);
4074         }
4075
4076         return (PF_PASS);
4077 }
4078
4079 static int
4080 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4081         struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4082 {
4083         struct tcphdr           *th = pd->hdr.tcp;
4084
4085         if (th->th_flags & TH_SYN)
4086                 if (src->state < TCPS_SYN_SENT)
4087                         src->state = TCPS_SYN_SENT;
4088         if (th->th_flags & TH_FIN)
4089                 if (src->state < TCPS_CLOSING)
4090                         src->state = TCPS_CLOSING;
4091         if (th->th_flags & TH_ACK) {
4092                 if (dst->state == TCPS_SYN_SENT) {
4093                         dst->state = TCPS_ESTABLISHED;
4094                         if (src->state == TCPS_ESTABLISHED &&
4095                             (*state)->src_node != NULL &&
4096                             pf_src_connlimit(state)) {
4097                                 REASON_SET(reason, PFRES_SRCLIMIT);
4098                                 return (PF_DROP);
4099                         }
4100                 } else if (dst->state == TCPS_CLOSING) {
4101                         dst->state = TCPS_FIN_WAIT_2;
4102                 } else if (src->state == TCPS_SYN_SENT &&
4103                     dst->state < TCPS_SYN_SENT) {
4104                         /*
4105                          * Handle a special sloppy case where we only see one
4106                          * half of the connection. If there is a ACK after
4107                          * the initial SYN without ever seeing a packet from
4108                          * the destination, set the connection to established.
4109                          */
4110                         dst->state = src->state = TCPS_ESTABLISHED;
4111                         if ((*state)->src_node != NULL &&
4112                             pf_src_connlimit(state)) {
4113                                 REASON_SET(reason, PFRES_SRCLIMIT);
4114                                 return (PF_DROP);
4115                         }
4116                 } else if (src->state == TCPS_CLOSING &&
4117                     dst->state == TCPS_ESTABLISHED &&
4118                     dst->seqlo == 0) {
4119                         /*
4120                          * Handle the closing of half connections where we
4121                          * don't see the full bidirectional FIN/ACK+ACK
4122                          * handshake.
4123                          */
4124                         dst->state = TCPS_CLOSING;
4125                 }
4126         }
4127         if (th->th_flags & TH_RST)
4128                 src->state = dst->state = TCPS_TIME_WAIT;
4129
4130         /* update expire time */
4131         (*state)->expire = time_uptime;
4132         if (src->state >= TCPS_FIN_WAIT_2 &&
4133             dst->state >= TCPS_FIN_WAIT_2)
4134                 (*state)->timeout = PFTM_TCP_CLOSED;
4135         else if (src->state >= TCPS_CLOSING &&
4136             dst->state >= TCPS_CLOSING)
4137                 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4138         else if (src->state < TCPS_ESTABLISHED ||
4139             dst->state < TCPS_ESTABLISHED)
4140                 (*state)->timeout = PFTM_TCP_OPENING;
4141         else if (src->state >= TCPS_CLOSING ||
4142             dst->state >= TCPS_CLOSING)
4143                 (*state)->timeout = PFTM_TCP_CLOSING;
4144         else
4145                 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4146
4147         return (PF_PASS);
4148 }
4149
4150 static int
4151 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4152     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4153     u_short *reason)
4154 {
4155         struct pf_state_key_cmp  key;
4156         struct tcphdr           *th = pd->hdr.tcp;
4157         int                      copyback = 0;
4158         struct pf_state_peer    *src, *dst;
4159         struct pf_state_key     *sk;
4160
4161         bzero(&key, sizeof(key));
4162         key.af = pd->af;
4163         key.proto = IPPROTO_TCP;
4164         if (direction == PF_IN) {       /* wire side, straight */
4165                 PF_ACPY(&key.addr[0], pd->src, key.af);
4166                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4167                 key.port[0] = th->th_sport;
4168                 key.port[1] = th->th_dport;
4169         } else {                        /* stack side, reverse */
4170                 PF_ACPY(&key.addr[1], pd->src, key.af);
4171                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4172                 key.port[1] = th->th_sport;
4173                 key.port[0] = th->th_dport;
4174         }
4175
4176         STATE_LOOKUP(kif, &key, direction, *state, pd);
4177
4178         if (direction == (*state)->direction) {
4179                 src = &(*state)->src;
4180                 dst = &(*state)->dst;
4181         } else {
4182                 src = &(*state)->dst;
4183                 dst = &(*state)->src;
4184         }
4185
4186         sk = (*state)->key[pd->didx];
4187
4188         if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4189                 if (direction != (*state)->direction) {
4190                         REASON_SET(reason, PFRES_SYNPROXY);
4191                         return (PF_SYNPROXY_DROP);
4192                 }
4193                 if (th->th_flags & TH_SYN) {
4194                         if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4195                                 REASON_SET(reason, PFRES_SYNPROXY);
4196                                 return (PF_DROP);
4197                         }
4198                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4199                             pd->src, th->th_dport, th->th_sport,
4200                             (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4201                             TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4202                         REASON_SET(reason, PFRES_SYNPROXY);
4203                         return (PF_SYNPROXY_DROP);
4204                 } else if (!(th->th_flags & TH_ACK) ||
4205                     (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4206                     (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4207                         REASON_SET(reason, PFRES_SYNPROXY);
4208                         return (PF_DROP);
4209                 } else if ((*state)->src_node != NULL &&
4210                     pf_src_connlimit(state)) {
4211                         REASON_SET(reason, PFRES_SRCLIMIT);
4212                         return (PF_DROP);
4213                 } else
4214                         (*state)->src.state = PF_TCPS_PROXY_DST;
4215         }
4216         if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4217                 if (direction == (*state)->direction) {
4218                         if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4219                             (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4220                             (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4221                                 REASON_SET(reason, PFRES_SYNPROXY);
4222                                 return (PF_DROP);
4223                         }
4224                         (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4225                         if ((*state)->dst.seqhi == 1)
4226                                 (*state)->dst.seqhi = htonl(arc4random());
4227                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4228                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4229                             sk->port[pd->sidx], sk->port[pd->didx],
4230                             (*state)->dst.seqhi, 0, TH_SYN, 0,
4231                             (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4232                         REASON_SET(reason, PFRES_SYNPROXY);
4233                         return (PF_SYNPROXY_DROP);
4234                 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4235                     (TH_SYN|TH_ACK)) ||
4236                     (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4237                         REASON_SET(reason, PFRES_SYNPROXY);
4238                         return (PF_DROP);
4239                 } else {
4240                         (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4241                         (*state)->dst.seqlo = ntohl(th->th_seq);
4242                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4243                             pd->src, th->th_dport, th->th_sport,
4244                             ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4245                             TH_ACK, (*state)->src.max_win, 0, 0, 0,
4246                             (*state)->tag, NULL);
4247                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4248                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4249                             sk->port[pd->sidx], sk->port[pd->didx],
4250                             (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4251                             TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4252                         (*state)->src.seqdiff = (*state)->dst.seqhi -
4253                             (*state)->src.seqlo;
4254                         (*state)->dst.seqdiff = (*state)->src.seqhi -
4255                             (*state)->dst.seqlo;
4256                         (*state)->src.seqhi = (*state)->src.seqlo +
4257                             (*state)->dst.max_win;
4258                         (*state)->dst.seqhi = (*state)->dst.seqlo +
4259                             (*state)->src.max_win;
4260                         (*state)->src.wscale = (*state)->dst.wscale = 0;
4261                         (*state)->src.state = (*state)->dst.state =
4262                             TCPS_ESTABLISHED;
4263                         REASON_SET(reason, PFRES_SYNPROXY);
4264                         return (PF_SYNPROXY_DROP);
4265                 }
4266         }
4267
4268         if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4269             dst->state >= TCPS_FIN_WAIT_2 &&
4270             src->state >= TCPS_FIN_WAIT_2) {
4271                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4272                         printf("pf: state reuse ");
4273                         pf_print_state(*state);
4274                         pf_print_flags(th->th_flags);
4275                         printf("\n");
4276                 }
4277                 /* XXX make sure it's the same direction ?? */
4278                 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4279                 pf_unlink_state(*state, PF_ENTER_LOCKED);
4280                 *state = NULL;
4281                 return (PF_DROP);
4282         }
4283
4284         if ((*state)->state_flags & PFSTATE_SLOPPY) {
4285                 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4286                         return (PF_DROP);
4287         } else {
4288                 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4289                     &copyback) == PF_DROP)
4290                         return (PF_DROP);
4291         }
4292
4293         /* translate source/destination address, if necessary */
4294         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4295                 struct pf_state_key *nk = (*state)->key[pd->didx];
4296
4297                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4298                     nk->port[pd->sidx] != th->th_sport)
4299                         pf_change_ap(pd->src, &th->th_sport, pd->ip_sum,
4300                             &th->th_sum, &nk->addr[pd->sidx],
4301                             nk->port[pd->sidx], 0, pd->af);
4302
4303                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4304                     nk->port[pd->didx] != th->th_dport)
4305                         pf_change_ap(pd->dst, &th->th_dport, pd->ip_sum,
4306                             &th->th_sum, &nk->addr[pd->didx],
4307                             nk->port[pd->didx], 0, pd->af);
4308                 copyback = 1;
4309         }
4310
4311         /* Copyback sequence modulation or stateful scrub changes if needed */
4312         if (copyback)
4313                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
4314
4315         return (PF_PASS);
4316 }
4317
4318 static int
4319 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
4320     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4321 {
4322         struct pf_state_peer    *src, *dst;
4323         struct pf_state_key_cmp  key;
4324         struct udphdr           *uh = pd->hdr.udp;
4325
4326         bzero(&key, sizeof(key));
4327         key.af = pd->af;
4328         key.proto = IPPROTO_UDP;
4329         if (direction == PF_IN) {       /* wire side, straight */
4330                 PF_ACPY(&key.addr[0], pd->src, key.af);
4331                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4332                 key.port[0] = uh->uh_sport;
4333                 key.port[1] = uh->uh_dport;
4334         } else {                        /* stack side, reverse */
4335                 PF_ACPY(&key.addr[1], pd->src, key.af);
4336                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4337                 key.port[1] = uh->uh_sport;
4338                 key.port[0] = uh->uh_dport;
4339         }
4340
4341         STATE_LOOKUP(kif, &key, direction, *state, pd);
4342
4343         if (direction == (*state)->direction) {
4344                 src = &(*state)->src;
4345                 dst = &(*state)->dst;
4346         } else {
4347                 src = &(*state)->dst;
4348                 dst = &(*state)->src;
4349         }
4350
4351         /* update states */
4352         if (src->state < PFUDPS_SINGLE)
4353                 src->state = PFUDPS_SINGLE;
4354         if (dst->state == PFUDPS_SINGLE)
4355                 dst->state = PFUDPS_MULTIPLE;
4356
4357         /* update expire time */
4358         (*state)->expire = time_uptime;
4359         if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4360                 (*state)->timeout = PFTM_UDP_MULTIPLE;
4361         else
4362                 (*state)->timeout = PFTM_UDP_SINGLE;
4363
4364         /* translate source/destination address, if necessary */
4365         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4366                 struct pf_state_key *nk = (*state)->key[pd->didx];
4367
4368                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4369                     nk->port[pd->sidx] != uh->uh_sport)
4370                         pf_change_ap(pd->src, &uh->uh_sport, pd->ip_sum,
4371                             &uh->uh_sum, &nk->addr[pd->sidx],
4372                             nk->port[pd->sidx], 1, pd->af);
4373
4374                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4375                     nk->port[pd->didx] != uh->uh_dport)
4376                         pf_change_ap(pd->dst, &uh->uh_dport, pd->ip_sum,
4377                             &uh->uh_sum, &nk->addr[pd->didx],
4378                             nk->port[pd->didx], 1, pd->af);
4379                 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4380         }
4381
4382         return (PF_PASS);
4383 }
4384
4385 static int
4386 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
4387     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4388 {
4389         struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
4390         u_int16_t        icmpid = 0, *icmpsum;
4391         u_int8_t         icmptype;
4392         int              state_icmp = 0;
4393         struct pf_state_key_cmp key;
4394
4395         bzero(&key, sizeof(key));
4396         switch (pd->proto) {
4397 #ifdef INET
4398         case IPPROTO_ICMP:
4399                 icmptype = pd->hdr.icmp->icmp_type;
4400                 icmpid = pd->hdr.icmp->icmp_id;
4401                 icmpsum = &pd->hdr.icmp->icmp_cksum;
4402
4403                 if (icmptype == ICMP_UNREACH ||
4404                     icmptype == ICMP_SOURCEQUENCH ||
4405                     icmptype == ICMP_REDIRECT ||
4406                     icmptype == ICMP_TIMXCEED ||
4407                     icmptype == ICMP_PARAMPROB)
4408                         state_icmp++;
4409                 break;
4410 #endif /* INET */
4411 #ifdef INET6
4412         case IPPROTO_ICMPV6:
4413                 icmptype = pd->hdr.icmp6->icmp6_type;
4414                 icmpid = pd->hdr.icmp6->icmp6_id;
4415                 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4416
4417                 if (icmptype == ICMP6_DST_UNREACH ||
4418                     icmptype == ICMP6_PACKET_TOO_BIG ||
4419                     icmptype == ICMP6_TIME_EXCEEDED ||
4420                     icmptype == ICMP6_PARAM_PROB)
4421                         state_icmp++;
4422                 break;
4423 #endif /* INET6 */
4424         }
4425
4426         if (!state_icmp) {
4427
4428                 /*
4429                  * ICMP query/reply message not related to a TCP/UDP packet.
4430                  * Search for an ICMP state.
4431                  */
4432                 key.af = pd->af;
4433                 key.proto = pd->proto;
4434                 key.port[0] = key.port[1] = icmpid;
4435                 if (direction == PF_IN) {       /* wire side, straight */
4436                         PF_ACPY(&key.addr[0], pd->src, key.af);
4437                         PF_ACPY(&key.addr[1], pd->dst, key.af);
4438                 } else {                        /* stack side, reverse */
4439                         PF_ACPY(&key.addr[1], pd->src, key.af);
4440                         PF_ACPY(&key.addr[0], pd->dst, key.af);
4441                 }
4442
4443                 STATE_LOOKUP(kif, &key, direction, *state, pd);
4444
4445                 (*state)->expire = time_uptime;
4446                 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4447
4448                 /* translate source/destination address, if necessary */
4449                 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4450                         struct pf_state_key *nk = (*state)->key[pd->didx];
4451
4452                         switch (pd->af) {
4453 #ifdef INET
4454                         case AF_INET:
4455                                 if (PF_ANEQ(pd->src,
4456                                     &nk->addr[pd->sidx], AF_INET))
4457                                         pf_change_a(&saddr->v4.s_addr,
4458                                             pd->ip_sum,
4459                                             nk->addr[pd->sidx].v4.s_addr, 0);
4460
4461                                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4462                                     AF_INET))
4463                                         pf_change_a(&daddr->v4.s_addr,
4464                                             pd->ip_sum,
4465                                             nk->addr[pd->didx].v4.s_addr, 0);
4466
4467                                 if (nk->port[0] !=
4468                                     pd->hdr.icmp->icmp_id) {
4469                                         pd->hdr.icmp->icmp_cksum =
4470                                             pf_cksum_fixup(
4471                                             pd->hdr.icmp->icmp_cksum, icmpid,
4472                                             nk->port[pd->sidx], 0);
4473                                         pd->hdr.icmp->icmp_id =
4474                                             nk->port[pd->sidx];
4475                                 }
4476
4477                                 m_copyback(m, off, ICMP_MINLEN,
4478                                     (caddr_t )pd->hdr.icmp);
4479                                 break;
4480 #endif /* INET */
4481 #ifdef INET6
4482                         case AF_INET6:
4483                                 if (PF_ANEQ(pd->src,
4484                                     &nk->addr[pd->sidx], AF_INET6))
4485                                         pf_change_a6(saddr,
4486                                             &pd->hdr.icmp6->icmp6_cksum,
4487                                             &nk->addr[pd->sidx], 0);
4488
4489                                 if (PF_ANEQ(pd->dst,
4490                                     &nk->addr[pd->didx], AF_INET6))
4491                                         pf_change_a6(daddr,
4492                                             &pd->hdr.icmp6->icmp6_cksum,
4493                                             &nk->addr[pd->didx], 0);
4494
4495                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4496                                     (caddr_t )pd->hdr.icmp6);
4497                                 break;
4498 #endif /* INET6 */
4499                         }
4500                 }
4501                 return (PF_PASS);
4502
4503         } else {
4504                 /*
4505                  * ICMP error message in response to a TCP/UDP packet.
4506                  * Extract the inner TCP/UDP header and search for that state.
4507                  */
4508
4509                 struct pf_pdesc pd2;
4510                 bzero(&pd2, sizeof pd2);
4511 #ifdef INET
4512                 struct ip       h2;
4513 #endif /* INET */
4514 #ifdef INET6
4515                 struct ip6_hdr  h2_6;
4516                 int             terminal = 0;
4517 #endif /* INET6 */
4518                 int             ipoff2 = 0;
4519                 int             off2 = 0;
4520
4521                 pd2.af = pd->af;
4522                 /* Payload packet is from the opposite direction. */
4523                 pd2.sidx = (direction == PF_IN) ? 1 : 0;
4524                 pd2.didx = (direction == PF_IN) ? 0 : 1;
4525                 switch (pd->af) {
4526 #ifdef INET
4527                 case AF_INET:
4528                         /* offset of h2 in mbuf chain */
4529                         ipoff2 = off + ICMP_MINLEN;
4530
4531                         if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4532                             NULL, reason, pd2.af)) {
4533                                 DPFPRINTF(PF_DEBUG_MISC,
4534                                     ("pf: ICMP error message too short "
4535                                     "(ip)\n"));
4536                                 return (PF_DROP);
4537                         }
4538                         /*
4539                          * ICMP error messages don't refer to non-first
4540                          * fragments
4541                          */
4542                         if (h2.ip_off & htons(IP_OFFMASK)) {
4543                                 REASON_SET(reason, PFRES_FRAG);
4544                                 return (PF_DROP);
4545                         }
4546
4547                         /* offset of protocol header that follows h2 */
4548                         off2 = ipoff2 + (h2.ip_hl << 2);
4549
4550                         pd2.proto = h2.ip_p;
4551                         pd2.src = (struct pf_addr *)&h2.ip_src;
4552                         pd2.dst = (struct pf_addr *)&h2.ip_dst;
4553                         pd2.ip_sum = &h2.ip_sum;
4554                         break;
4555 #endif /* INET */
4556 #ifdef INET6
4557                 case AF_INET6:
4558                         ipoff2 = off + sizeof(struct icmp6_hdr);
4559
4560                         if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4561                             NULL, reason, pd2.af)) {
4562                                 DPFPRINTF(PF_DEBUG_MISC,
4563                                     ("pf: ICMP error message too short "
4564                                     "(ip6)\n"));
4565                                 return (PF_DROP);
4566                         }
4567                         pd2.proto = h2_6.ip6_nxt;
4568                         pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4569                         pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4570                         pd2.ip_sum = NULL;
4571                         off2 = ipoff2 + sizeof(h2_6);
4572                         do {
4573                                 switch (pd2.proto) {
4574                                 case IPPROTO_FRAGMENT:
4575                                         /*
4576                                          * ICMPv6 error messages for
4577                                          * non-first fragments
4578                                          */
4579                                         REASON_SET(reason, PFRES_FRAG);
4580                                         return (PF_DROP);
4581                                 case IPPROTO_AH:
4582                                 case IPPROTO_HOPOPTS:
4583                                 case IPPROTO_ROUTING:
4584                                 case IPPROTO_DSTOPTS: {
4585                                         /* get next header and header length */
4586                                         struct ip6_ext opt6;
4587
4588                                         if (!pf_pull_hdr(m, off2, &opt6,
4589                                             sizeof(opt6), NULL, reason,
4590                                             pd2.af)) {
4591                                                 DPFPRINTF(PF_DEBUG_MISC,
4592                                                     ("pf: ICMPv6 short opt\n"));
4593                                                 return (PF_DROP);
4594                                         }
4595                                         if (pd2.proto == IPPROTO_AH)
4596                                                 off2 += (opt6.ip6e_len + 2) * 4;
4597                                         else
4598                                                 off2 += (opt6.ip6e_len + 1) * 8;
4599                                         pd2.proto = opt6.ip6e_nxt;
4600                                         /* goto the next header */
4601                                         break;
4602                                 }
4603                                 default:
4604                                         terminal++;
4605                                         break;
4606                                 }
4607                         } while (!terminal);
4608                         break;
4609 #endif /* INET6 */
4610                 }
4611
4612                 switch (pd2.proto) {
4613                 case IPPROTO_TCP: {
4614                         struct tcphdr            th;
4615                         u_int32_t                seq;
4616                         struct pf_state_peer    *src, *dst;
4617                         u_int8_t                 dws;
4618                         int                      copyback = 0;
4619
4620                         /*
4621                          * Only the first 8 bytes of the TCP header can be
4622                          * expected. Don't access any TCP header fields after
4623                          * th_seq, an ackskew test is not possible.
4624                          */
4625                         if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4626                             pd2.af)) {
4627                                 DPFPRINTF(PF_DEBUG_MISC,
4628                                     ("pf: ICMP error message too short "
4629                                     "(tcp)\n"));
4630                                 return (PF_DROP);
4631                         }
4632
4633                         key.af = pd2.af;
4634                         key.proto = IPPROTO_TCP;
4635                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4636                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4637                         key.port[pd2.sidx] = th.th_sport;
4638                         key.port[pd2.didx] = th.th_dport;
4639
4640                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4641
4642                         if (direction == (*state)->direction) {
4643                                 src = &(*state)->dst;
4644                                 dst = &(*state)->src;
4645                         } else {
4646                                 src = &(*state)->src;
4647                                 dst = &(*state)->dst;
4648                         }
4649
4650                         if (src->wscale && dst->wscale)
4651                                 dws = dst->wscale & PF_WSCALE_MASK;
4652                         else
4653                                 dws = 0;
4654
4655                         /* Demodulate sequence number */
4656                         seq = ntohl(th.th_seq) - src->seqdiff;
4657                         if (src->seqdiff) {
4658                                 pf_change_a(&th.th_seq, icmpsum,
4659                                     htonl(seq), 0);
4660                                 copyback = 1;
4661                         }
4662
4663                         if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4664                             (!SEQ_GEQ(src->seqhi, seq) ||
4665                             !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4666                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4667                                         printf("pf: BAD ICMP %d:%d ",
4668                                             icmptype, pd->hdr.icmp->icmp_code);
4669                                         pf_print_host(pd->src, 0, pd->af);
4670                                         printf(" -> ");
4671                                         pf_print_host(pd->dst, 0, pd->af);
4672                                         printf(" state: ");
4673                                         pf_print_state(*state);
4674                                         printf(" seq=%u\n", seq);
4675                                 }
4676                                 REASON_SET(reason, PFRES_BADSTATE);
4677                                 return (PF_DROP);
4678                         } else {
4679                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4680                                         printf("pf: OK ICMP %d:%d ",
4681                                             icmptype, pd->hdr.icmp->icmp_code);
4682                                         pf_print_host(pd->src, 0, pd->af);
4683                                         printf(" -> ");
4684                                         pf_print_host(pd->dst, 0, pd->af);
4685                                         printf(" state: ");
4686                                         pf_print_state(*state);
4687                                         printf(" seq=%u\n", seq);
4688                                 }
4689                         }
4690
4691                         /* translate source/destination address, if necessary */
4692                         if ((*state)->key[PF_SK_WIRE] !=
4693                             (*state)->key[PF_SK_STACK]) {
4694                                 struct pf_state_key *nk =
4695                                     (*state)->key[pd->didx];
4696
4697                                 if (PF_ANEQ(pd2.src,
4698                                     &nk->addr[pd2.sidx], pd2.af) ||
4699                                     nk->port[pd2.sidx] != th.th_sport)
4700                                         pf_change_icmp(pd2.src, &th.th_sport,
4701                                             daddr, &nk->addr[pd2.sidx],
4702                                             nk->port[pd2.sidx], NULL,
4703                                             pd2.ip_sum, icmpsum,
4704                                             pd->ip_sum, 0, pd2.af);
4705
4706                                 if (PF_ANEQ(pd2.dst,
4707                                     &nk->addr[pd2.didx], pd2.af) ||
4708                                     nk->port[pd2.didx] != th.th_dport)
4709                                         pf_change_icmp(pd2.dst, &th.th_dport,
4710                                             NULL, /* XXX Inbound NAT? */
4711                                             &nk->addr[pd2.didx],
4712                                             nk->port[pd2.didx], NULL,
4713                                             pd2.ip_sum, icmpsum,
4714                                             pd->ip_sum, 0, pd2.af);
4715                                 copyback = 1;
4716                         }
4717
4718                         if (copyback) {
4719                                 switch (pd2.af) {
4720 #ifdef INET
4721                                 case AF_INET:
4722                                         m_copyback(m, off, ICMP_MINLEN,
4723                                             (caddr_t )pd->hdr.icmp);
4724                                         m_copyback(m, ipoff2, sizeof(h2),
4725                                             (caddr_t )&h2);
4726                                         break;
4727 #endif /* INET */
4728 #ifdef INET6
4729                                 case AF_INET6:
4730                                         m_copyback(m, off,
4731                                             sizeof(struct icmp6_hdr),
4732                                             (caddr_t )pd->hdr.icmp6);
4733                                         m_copyback(m, ipoff2, sizeof(h2_6),
4734                                             (caddr_t )&h2_6);
4735                                         break;
4736 #endif /* INET6 */
4737                                 }
4738                                 m_copyback(m, off2, 8, (caddr_t)&th);
4739                         }
4740
4741                         return (PF_PASS);
4742                         break;
4743                 }
4744                 case IPPROTO_UDP: {
4745                         struct udphdr           uh;
4746
4747                         if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
4748                             NULL, reason, pd2.af)) {
4749                                 DPFPRINTF(PF_DEBUG_MISC,
4750                                     ("pf: ICMP error message too short "
4751                                     "(udp)\n"));
4752                                 return (PF_DROP);
4753                         }
4754
4755                         key.af = pd2.af;
4756                         key.proto = IPPROTO_UDP;
4757                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4758                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4759                         key.port[pd2.sidx] = uh.uh_sport;
4760                         key.port[pd2.didx] = uh.uh_dport;
4761
4762                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4763
4764                         /* translate source/destination address, if necessary */
4765                         if ((*state)->key[PF_SK_WIRE] !=
4766                             (*state)->key[PF_SK_STACK]) {
4767                                 struct pf_state_key *nk =
4768                                     (*state)->key[pd->didx];
4769
4770                                 if (PF_ANEQ(pd2.src,
4771                                     &nk->addr[pd2.sidx], pd2.af) ||
4772                                     nk->port[pd2.sidx] != uh.uh_sport)
4773                                         pf_change_icmp(pd2.src, &uh.uh_sport,
4774                                             daddr, &nk->addr[pd2.sidx],
4775                                             nk->port[pd2.sidx], &uh.uh_sum,
4776                                             pd2.ip_sum, icmpsum,
4777                                             pd->ip_sum, 1, pd2.af);
4778
4779                                 if (PF_ANEQ(pd2.dst,
4780                                     &nk->addr[pd2.didx], pd2.af) ||
4781                                     nk->port[pd2.didx] != uh.uh_dport)
4782                                         pf_change_icmp(pd2.dst, &uh.uh_dport,
4783                                             NULL, /* XXX Inbound NAT? */
4784                                             &nk->addr[pd2.didx],
4785                                             nk->port[pd2.didx], &uh.uh_sum,
4786                                             pd2.ip_sum, icmpsum,
4787                                             pd->ip_sum, 1, pd2.af);
4788
4789                                 switch (pd2.af) {
4790 #ifdef INET
4791                                 case AF_INET:
4792                                         m_copyback(m, off, ICMP_MINLEN,
4793                                             (caddr_t )pd->hdr.icmp);
4794                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4795                                         break;
4796 #endif /* INET */
4797 #ifdef INET6
4798                                 case AF_INET6:
4799                                         m_copyback(m, off,
4800                                             sizeof(struct icmp6_hdr),
4801                                             (caddr_t )pd->hdr.icmp6);
4802                                         m_copyback(m, ipoff2, sizeof(h2_6),
4803                                             (caddr_t )&h2_6);
4804                                         break;
4805 #endif /* INET6 */
4806                                 }
4807                                 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
4808                         }
4809                         return (PF_PASS);
4810                         break;
4811                 }
4812 #ifdef INET
4813                 case IPPROTO_ICMP: {
4814                         struct icmp             iih;
4815
4816                         if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
4817                             NULL, reason, pd2.af)) {
4818                                 DPFPRINTF(PF_DEBUG_MISC,
4819                                     ("pf: ICMP error message too short i"
4820                                     "(icmp)\n"));
4821                                 return (PF_DROP);
4822                         }
4823
4824                         key.af = pd2.af;
4825                         key.proto = IPPROTO_ICMP;
4826                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4827                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4828                         key.port[0] = key.port[1] = iih.icmp_id;
4829
4830                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4831
4832                         /* translate source/destination address, if necessary */
4833                         if ((*state)->key[PF_SK_WIRE] !=
4834                             (*state)->key[PF_SK_STACK]) {
4835                                 struct pf_state_key *nk =
4836                                     (*state)->key[pd->didx];
4837
4838                                 if (PF_ANEQ(pd2.src,
4839                                     &nk->addr[pd2.sidx], pd2.af) ||
4840                                     nk->port[pd2.sidx] != iih.icmp_id)
4841                                         pf_change_icmp(pd2.src, &iih.icmp_id,
4842                                             daddr, &nk->addr[pd2.sidx],
4843                                             nk->port[pd2.sidx], NULL,
4844                                             pd2.ip_sum, icmpsum,
4845                                             pd->ip_sum, 0, AF_INET);
4846
4847                                 if (PF_ANEQ(pd2.dst,
4848                                     &nk->addr[pd2.didx], pd2.af) ||
4849                                     nk->port[pd2.didx] != iih.icmp_id)
4850                                         pf_change_icmp(pd2.dst, &iih.icmp_id,
4851                                             NULL, /* XXX Inbound NAT? */
4852                                             &nk->addr[pd2.didx],
4853                                             nk->port[pd2.didx], NULL,
4854                                             pd2.ip_sum, icmpsum,
4855                                             pd->ip_sum, 0, AF_INET);
4856
4857                                 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
4858                                 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4859                                 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
4860                         }
4861                         return (PF_PASS);
4862                         break;
4863                 }
4864 #endif /* INET */
4865 #ifdef INET6
4866                 case IPPROTO_ICMPV6: {
4867                         struct icmp6_hdr        iih;
4868
4869                         if (!pf_pull_hdr(m, off2, &iih,
4870                             sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
4871                                 DPFPRINTF(PF_DEBUG_MISC,
4872                                     ("pf: ICMP error message too short "
4873                                     "(icmp6)\n"));
4874                                 return (PF_DROP);
4875                         }
4876
4877                         key.af = pd2.af;
4878                         key.proto = IPPROTO_ICMPV6;
4879                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4880                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4881                         key.port[0] = key.port[1] = iih.icmp6_id;
4882
4883                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4884
4885                         /* translate source/destination address, if necessary */
4886                         if ((*state)->key[PF_SK_WIRE] !=
4887                             (*state)->key[PF_SK_STACK]) {
4888                                 struct pf_state_key *nk =
4889                                     (*state)->key[pd->didx];
4890
4891                                 if (PF_ANEQ(pd2.src,
4892                                     &nk->addr[pd2.sidx], pd2.af) ||
4893                                     nk->port[pd2.sidx] != iih.icmp6_id)
4894                                         pf_change_icmp(pd2.src, &iih.icmp6_id,
4895                                             daddr, &nk->addr[pd2.sidx],
4896                                             nk->port[pd2.sidx], NULL,
4897                                             pd2.ip_sum, icmpsum,
4898                                             pd->ip_sum, 0, AF_INET6);
4899
4900                                 if (PF_ANEQ(pd2.dst,
4901                                     &nk->addr[pd2.didx], pd2.af) ||
4902                                     nk->port[pd2.didx] != iih.icmp6_id)
4903                                         pf_change_icmp(pd2.dst, &iih.icmp6_id,
4904                                             NULL, /* XXX Inbound NAT? */
4905                                             &nk->addr[pd2.didx],
4906                                             nk->port[pd2.didx], NULL,
4907                                             pd2.ip_sum, icmpsum,
4908                                             pd->ip_sum, 0, AF_INET6);
4909
4910                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4911                                     (caddr_t)pd->hdr.icmp6);
4912                                 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
4913                                 m_copyback(m, off2, sizeof(struct icmp6_hdr),
4914                                     (caddr_t)&iih);
4915                         }
4916                         return (PF_PASS);
4917                         break;
4918                 }
4919 #endif /* INET6 */
4920                 default: {
4921                         key.af = pd2.af;
4922                         key.proto = pd2.proto;
4923                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4924                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4925                         key.port[0] = key.port[1] = 0;
4926
4927                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4928
4929                         /* translate source/destination address, if necessary */
4930                         if ((*state)->key[PF_SK_WIRE] !=
4931                             (*state)->key[PF_SK_STACK]) {
4932                                 struct pf_state_key *nk =
4933                                     (*state)->key[pd->didx];
4934
4935                                 if (PF_ANEQ(pd2.src,
4936                                     &nk->addr[pd2.sidx], pd2.af))
4937                                         pf_change_icmp(pd2.src, NULL, daddr,
4938                                             &nk->addr[pd2.sidx], 0, NULL,
4939                                             pd2.ip_sum, icmpsum,
4940                                             pd->ip_sum, 0, pd2.af);
4941
4942                                 if (PF_ANEQ(pd2.dst,
4943                                     &nk->addr[pd2.didx], pd2.af))
4944                                         pf_change_icmp(pd2.src, NULL,
4945                                             NULL, /* XXX Inbound NAT? */
4946                                             &nk->addr[pd2.didx], 0, NULL,
4947                                             pd2.ip_sum, icmpsum,
4948                                             pd->ip_sum, 0, pd2.af);
4949
4950                                 switch (pd2.af) {
4951 #ifdef INET
4952                                 case AF_INET:
4953                                         m_copyback(m, off, ICMP_MINLEN,
4954                                             (caddr_t)pd->hdr.icmp);
4955                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4956                                         break;
4957 #endif /* INET */
4958 #ifdef INET6
4959                                 case AF_INET6:
4960                                         m_copyback(m, off,
4961                                             sizeof(struct icmp6_hdr),
4962                                             (caddr_t )pd->hdr.icmp6);
4963                                         m_copyback(m, ipoff2, sizeof(h2_6),
4964                                             (caddr_t )&h2_6);
4965                                         break;
4966 #endif /* INET6 */
4967                                 }
4968                         }
4969                         return (PF_PASS);
4970                         break;
4971                 }
4972                 }
4973         }
4974 }
4975
4976 static int
4977 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
4978     struct mbuf *m, struct pf_pdesc *pd)
4979 {
4980         struct pf_state_peer    *src, *dst;
4981         struct pf_state_key_cmp  key;
4982
4983         bzero(&key, sizeof(key));
4984         key.af = pd->af;
4985         key.proto = pd->proto;
4986         if (direction == PF_IN) {
4987                 PF_ACPY(&key.addr[0], pd->src, key.af);
4988                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4989                 key.port[0] = key.port[1] = 0;
4990         } else {
4991                 PF_ACPY(&key.addr[1], pd->src, key.af);
4992                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4993                 key.port[1] = key.port[0] = 0;
4994         }
4995
4996         STATE_LOOKUP(kif, &key, direction, *state, pd);
4997
4998         if (direction == (*state)->direction) {
4999                 src = &(*state)->src;
5000                 dst = &(*state)->dst;
5001         } else {
5002                 src = &(*state)->dst;
5003                 dst = &(*state)->src;
5004         }
5005
5006         /* update states */
5007         if (src->state < PFOTHERS_SINGLE)
5008                 src->state = PFOTHERS_SINGLE;
5009         if (dst->state == PFOTHERS_SINGLE)
5010                 dst->state = PFOTHERS_MULTIPLE;
5011
5012         /* update expire time */
5013         (*state)->expire = time_uptime;
5014         if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5015                 (*state)->timeout = PFTM_OTHER_MULTIPLE;
5016         else
5017                 (*state)->timeout = PFTM_OTHER_SINGLE;
5018
5019         /* translate source/destination address, if necessary */
5020         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5021                 struct pf_state_key *nk = (*state)->key[pd->didx];
5022
5023                 KASSERT(nk, ("%s: nk is null", __func__));
5024                 KASSERT(pd, ("%s: pd is null", __func__));
5025                 KASSERT(pd->src, ("%s: pd->src is null", __func__));
5026                 KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5027                 switch (pd->af) {
5028 #ifdef INET
5029                 case AF_INET:
5030                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5031                                 pf_change_a(&pd->src->v4.s_addr,
5032                                     pd->ip_sum,
5033                                     nk->addr[pd->sidx].v4.s_addr,
5034                                     0);
5035
5036
5037                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5038                                 pf_change_a(&pd->dst->v4.s_addr,
5039                                     pd->ip_sum,
5040                                     nk->addr[pd->didx].v4.s_addr,
5041                                     0);
5042
5043                                 break;
5044 #endif /* INET */
5045 #ifdef INET6
5046                 case AF_INET6:
5047                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5048                                 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5049
5050                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5051                                 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5052 #endif /* INET6 */
5053                 }
5054         }
5055         return (PF_PASS);
5056 }
5057
5058 /*
5059  * ipoff and off are measured from the start of the mbuf chain.
5060  * h must be at "ipoff" on the mbuf chain.
5061  */
5062 void *
5063 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5064     u_short *actionp, u_short *reasonp, sa_family_t af)
5065 {
5066         switch (af) {
5067 #ifdef INET
5068         case AF_INET: {
5069                 struct ip       *h = mtod(m, struct ip *);
5070                 u_int16_t        fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5071
5072                 if (fragoff) {
5073                         if (fragoff >= len)
5074                                 ACTION_SET(actionp, PF_PASS);
5075                         else {
5076                                 ACTION_SET(actionp, PF_DROP);
5077                                 REASON_SET(reasonp, PFRES_FRAG);
5078                         }
5079                         return (NULL);
5080                 }
5081                 if (m->m_pkthdr.len < off + len ||
5082                     ntohs(h->ip_len) < off + len) {
5083                         ACTION_SET(actionp, PF_DROP);
5084                         REASON_SET(reasonp, PFRES_SHORT);
5085                         return (NULL);
5086                 }
5087                 break;
5088         }
5089 #endif /* INET */
5090 #ifdef INET6
5091         case AF_INET6: {
5092                 struct ip6_hdr  *h = mtod(m, struct ip6_hdr *);
5093
5094                 if (m->m_pkthdr.len < off + len ||
5095                     (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5096                     (unsigned)(off + len)) {
5097                         ACTION_SET(actionp, PF_DROP);
5098                         REASON_SET(reasonp, PFRES_SHORT);
5099                         return (NULL);
5100                 }
5101                 break;
5102         }
5103 #endif /* INET6 */
5104         }
5105         m_copydata(m, off, len, p);
5106         return (p);
5107 }
5108
5109 int
5110 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5111     int rtableid)
5112 {
5113 #ifdef RADIX_MPATH
5114         struct radix_node_head  *rnh;
5115 #endif
5116         struct sockaddr_in      *dst;
5117         int                      ret = 1;
5118         int                      check_mpath;
5119 #ifdef INET6
5120         struct sockaddr_in6     *dst6;
5121         struct route_in6         ro;
5122 #else
5123         struct route             ro;
5124 #endif
5125         struct radix_node       *rn;
5126         struct rtentry          *rt;
5127         struct ifnet            *ifp;
5128
5129         check_mpath = 0;
5130 #ifdef RADIX_MPATH
5131         /* XXX: stick to table 0 for now */
5132         rnh = rt_tables_get_rnh(0, af);
5133         if (rnh != NULL && rn_mpath_capable(rnh))
5134                 check_mpath = 1;
5135 #endif
5136         bzero(&ro, sizeof(ro));
5137         switch (af) {
5138         case AF_INET:
5139                 dst = satosin(&ro.ro_dst);
5140                 dst->sin_family = AF_INET;
5141                 dst->sin_len = sizeof(*dst);
5142                 dst->sin_addr = addr->v4;
5143                 break;
5144 #ifdef INET6
5145         case AF_INET6:
5146                 /*
5147                  * Skip check for addresses with embedded interface scope,
5148                  * as they would always match anyway.
5149                  */
5150                 if (IN6_IS_SCOPE_EMBED(&addr->v6))
5151                         goto out;
5152                 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5153                 dst6->sin6_family = AF_INET6;
5154                 dst6->sin6_len = sizeof(*dst6);
5155                 dst6->sin6_addr = addr->v6;
5156                 break;
5157 #endif /* INET6 */
5158         default:
5159                 return (0);
5160         }
5161
5162         /* Skip checks for ipsec interfaces */
5163         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5164                 goto out;
5165
5166         switch (af) {
5167 #ifdef INET6
5168         case AF_INET6:
5169                 in6_rtalloc_ign(&ro, 0, rtableid);
5170                 break;
5171 #endif
5172 #ifdef INET
5173         case AF_INET:
5174                 in_rtalloc_ign((struct route *)&ro, 0, rtableid);
5175                 break;
5176 #endif
5177         default:
5178                 rtalloc_ign((struct route *)&ro, 0);    /* No/default FIB. */
5179                 break;
5180         }
5181
5182         if (ro.ro_rt != NULL) {
5183                 /* No interface given, this is a no-route check */
5184                 if (kif == NULL)
5185                         goto out;
5186
5187                 if (kif->pfik_ifp == NULL) {
5188                         ret = 0;
5189                         goto out;
5190                 }
5191
5192                 /* Perform uRPF check if passed input interface */
5193                 ret = 0;
5194                 rn = (struct radix_node *)ro.ro_rt;
5195                 do {
5196                         rt = (struct rtentry *)rn;
5197                         ifp = rt->rt_ifp;
5198
5199                         if (kif->pfik_ifp == ifp)
5200                                 ret = 1;
5201 #ifdef RADIX_MPATH
5202                         rn = rn_mpath_next(rn);
5203 #endif
5204                 } while (check_mpath == 1 && rn != NULL && ret == 0);
5205         } else
5206                 ret = 0;
5207 out:
5208         if (ro.ro_rt != NULL)
5209                 RTFREE(ro.ro_rt);
5210         return (ret);
5211 }
5212
5213 #ifdef INET
5214 static void
5215 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5216     struct pf_state *s, struct pf_pdesc *pd)
5217 {
5218         struct mbuf             *m0, *m1;
5219         struct sockaddr_in      dst;
5220         struct ip               *ip;
5221         struct ifnet            *ifp = NULL;
5222         struct pf_addr           naddr;
5223         struct pf_src_node      *sn = NULL;
5224         int                      error = 0;
5225         uint16_t                 ip_len, ip_off;
5226
5227         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5228         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5229             __func__));
5230
5231         if ((pd->pf_mtag == NULL &&
5232             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5233             pd->pf_mtag->routed++ > 3) {
5234                 m0 = *m;
5235                 *m = NULL;
5236                 goto bad_locked;
5237         }
5238
5239         if (r->rt == PF_DUPTO) {
5240                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5241                         if (s)
5242                                 PF_STATE_UNLOCK(s);
5243                         return;
5244                 }
5245         } else {
5246                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5247                         if (s)
5248                                 PF_STATE_UNLOCK(s);
5249                         return;
5250                 }
5251                 m0 = *m;
5252         }
5253
5254         ip = mtod(m0, struct ip *);
5255
5256         bzero(&dst, sizeof(dst));
5257         dst.sin_family = AF_INET;
5258         dst.sin_len = sizeof(dst);
5259         dst.sin_addr = ip->ip_dst;
5260
5261         if (r->rt == PF_FASTROUTE) {
5262                 struct rtentry *rt;
5263
5264                 if (s)
5265                         PF_STATE_UNLOCK(s);
5266                 rt = rtalloc1_fib(sintosa(&dst), 0, 0, M_GETFIB(m0));
5267                 if (rt == NULL) {
5268                         KMOD_IPSTAT_INC(ips_noroute);
5269                         error = EHOSTUNREACH;
5270                         goto bad;
5271                 }
5272
5273                 ifp = rt->rt_ifp;
5274                 counter_u64_add(rt->rt_pksent, 1);
5275
5276                 if (rt->rt_flags & RTF_GATEWAY)
5277                         bcopy(satosin(rt->rt_gateway), &dst, sizeof(dst));
5278                 RTFREE_LOCKED(rt);
5279         } else {
5280                 if (TAILQ_EMPTY(&r->rpool.list)) {
5281                         DPFPRINTF(PF_DEBUG_URGENT,
5282                             ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5283                         goto bad_locked;
5284                 }
5285                 if (s == NULL) {
5286                         pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5287                             &naddr, NULL, &sn);
5288                         if (!PF_AZERO(&naddr, AF_INET))
5289                                 dst.sin_addr.s_addr = naddr.v4.s_addr;
5290                         ifp = r->rpool.cur->kif ?
5291                             r->rpool.cur->kif->pfik_ifp : NULL;
5292                 } else {
5293                         if (!PF_AZERO(&s->rt_addr, AF_INET))
5294                                 dst.sin_addr.s_addr =
5295                                     s->rt_addr.v4.s_addr;
5296                         ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5297                         PF_STATE_UNLOCK(s);
5298                 }
5299         }
5300         if (ifp == NULL)
5301                 goto bad;
5302
5303         if (oifp != ifp) {
5304                 if (pf_test(PF_OUT, ifp, &m0, NULL) != PF_PASS)
5305                         goto bad;
5306                 else if (m0 == NULL)
5307                         goto done;
5308                 if (m0->m_len < sizeof(struct ip)) {
5309                         DPFPRINTF(PF_DEBUG_URGENT,
5310                             ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5311                         goto bad;
5312                 }
5313                 ip = mtod(m0, struct ip *);
5314         }
5315
5316         if (ifp->if_flags & IFF_LOOPBACK)
5317                 m0->m_flags |= M_SKIP_FIREWALL;
5318
5319         ip_len = ntohs(ip->ip_len);
5320         ip_off = ntohs(ip->ip_off);
5321
5322         /* Copied from FreeBSD 10.0-CURRENT ip_output. */
5323         m0->m_pkthdr.csum_flags |= CSUM_IP;
5324         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5325                 in_delayed_cksum(m0);
5326                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5327         }
5328 #ifdef SCTP
5329         if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5330                 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
5331                 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5332         }
5333 #endif
5334
5335         /*
5336          * If small enough for interface, or the interface will take
5337          * care of the fragmentation for us, we can just send directly.
5338          */
5339         if (ip_len <= ifp->if_mtu ||
5340             (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 ||
5341             ((ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) {
5342                 ip->ip_sum = 0;
5343                 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5344                         ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5345                         m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5346                 }
5347                 m_clrprotoflags(m0);    /* Avoid confusing lower layers. */
5348                 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5349                 goto done;
5350         }
5351
5352         /* Balk when DF bit is set or the interface didn't support TSO. */
5353         if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5354                 error = EMSGSIZE;
5355                 KMOD_IPSTAT_INC(ips_cantfrag);
5356                 if (r->rt != PF_DUPTO) {
5357                         icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5358                             ifp->if_mtu);
5359                         goto done;
5360                 } else
5361                         goto bad;
5362         }
5363
5364         error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5365         if (error)
5366                 goto bad;
5367
5368         for (; m0; m0 = m1) {
5369                 m1 = m0->m_nextpkt;
5370                 m0->m_nextpkt = NULL;
5371                 if (error == 0) {
5372                         m_clrprotoflags(m0);
5373                         error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5374                 } else
5375                         m_freem(m0);
5376         }
5377
5378         if (error == 0)
5379                 KMOD_IPSTAT_INC(ips_fragmented);
5380
5381 done:
5382         if (r->rt != PF_DUPTO)
5383                 *m = NULL;
5384         return;
5385
5386 bad_locked:
5387         if (s)
5388                 PF_STATE_UNLOCK(s);
5389 bad:
5390         m_freem(m0);
5391         goto done;
5392 }
5393 #endif /* INET */
5394
5395 #ifdef INET6
5396 static void
5397 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5398     struct pf_state *s, struct pf_pdesc *pd)
5399 {
5400         struct mbuf             *m0;
5401         struct sockaddr_in6     dst;
5402         struct ip6_hdr          *ip6;
5403         struct ifnet            *ifp = NULL;
5404         struct pf_addr           naddr;
5405         struct pf_src_node      *sn = NULL;
5406
5407         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5408         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5409             __func__));
5410
5411         if ((pd->pf_mtag == NULL &&
5412             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5413             pd->pf_mtag->routed++ > 3) {
5414                 m0 = *m;
5415                 *m = NULL;
5416                 goto bad_locked;
5417         }
5418
5419         if (r->rt == PF_DUPTO) {
5420                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5421                         if (s)
5422                                 PF_STATE_UNLOCK(s);
5423                         return;
5424                 }
5425         } else {
5426                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5427                         if (s)
5428                                 PF_STATE_UNLOCK(s);
5429                         return;
5430                 }
5431                 m0 = *m;
5432         }
5433
5434         ip6 = mtod(m0, struct ip6_hdr *);
5435
5436         bzero(&dst, sizeof(dst));
5437         dst.sin6_family = AF_INET6;
5438         dst.sin6_len = sizeof(dst);
5439         dst.sin6_addr = ip6->ip6_dst;
5440
5441         /* Cheat. XXX why only in the v6 case??? */
5442         if (r->rt == PF_FASTROUTE) {
5443                 if (s)
5444                         PF_STATE_UNLOCK(s);
5445                 m0->m_flags |= M_SKIP_FIREWALL;
5446                 ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL);
5447                 return;
5448         }
5449
5450         if (TAILQ_EMPTY(&r->rpool.list)) {
5451                 DPFPRINTF(PF_DEBUG_URGENT,
5452                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5453                 goto bad_locked;
5454         }
5455         if (s == NULL) {
5456                 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5457                     &naddr, NULL, &sn);
5458                 if (!PF_AZERO(&naddr, AF_INET6))
5459                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5460                             &naddr, AF_INET6);
5461                 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5462         } else {
5463                 if (!PF_AZERO(&s->rt_addr, AF_INET6))
5464                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5465                             &s->rt_addr, AF_INET6);
5466                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5467         }
5468
5469         if (s)
5470                 PF_STATE_UNLOCK(s);
5471
5472         if (ifp == NULL)
5473                 goto bad;
5474
5475         if (oifp != ifp) {
5476                 if (pf_test6(PF_OUT, ifp, &m0, NULL) != PF_PASS)
5477                         goto bad;
5478                 else if (m0 == NULL)
5479                         goto done;
5480                 if (m0->m_len < sizeof(struct ip6_hdr)) {
5481                         DPFPRINTF(PF_DEBUG_URGENT,
5482                             ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5483                             __func__));
5484                         goto bad;
5485                 }
5486                 ip6 = mtod(m0, struct ip6_hdr *);
5487         }
5488
5489         if (ifp->if_flags & IFF_LOOPBACK)
5490                 m0->m_flags |= M_SKIP_FIREWALL;
5491
5492         /*
5493          * If the packet is too large for the outgoing interface,
5494          * send back an icmp6 error.
5495          */
5496         if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5497                 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5498         if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5499                 nd6_output(ifp, ifp, m0, &dst, NULL);
5500         else {
5501                 in6_ifstat_inc(ifp, ifs6_in_toobig);
5502                 if (r->rt != PF_DUPTO)
5503                         icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5504                 else
5505                         goto bad;
5506         }
5507
5508 done:
5509         if (r->rt != PF_DUPTO)
5510                 *m = NULL;
5511         return;
5512
5513 bad_locked:
5514         if (s)
5515                 PF_STATE_UNLOCK(s);
5516 bad:
5517         m_freem(m0);
5518         goto done;
5519 }
5520 #endif /* INET6 */
5521
5522 /*
5523  * FreeBSD supports cksum offloads for the following drivers.
5524  *  em(4), fxp(4), ixgb(4), lge(4), ndis(4), nge(4), re(4),
5525  *   ti(4), txp(4), xl(4)
5526  *
5527  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5528  *  network driver performed cksum including pseudo header, need to verify
5529  *   csum_data
5530  * CSUM_DATA_VALID :
5531  *  network driver performed cksum, needs to additional pseudo header
5532  *  cksum computation with partial csum_data(i.e. lack of H/W support for
5533  *  pseudo header, for instance hme(4), sk(4) and possibly gem(4))
5534  *
5535  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5536  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5537  * TCP/UDP layer.
5538  * Also, set csum_data to 0xffff to force cksum validation.
5539  */
5540 static int
5541 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5542 {
5543         u_int16_t sum = 0;
5544         int hw_assist = 0;
5545         struct ip *ip;
5546
5547         if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5548                 return (1);
5549         if (m->m_pkthdr.len < off + len)
5550                 return (1);
5551
5552         switch (p) {
5553         case IPPROTO_TCP:
5554                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5555                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5556                                 sum = m->m_pkthdr.csum_data;
5557                         } else {
5558                                 ip = mtod(m, struct ip *);
5559                                 sum = in_pseudo(ip->ip_src.s_addr,
5560                                 ip->ip_dst.s_addr, htonl((u_short)len +
5561                                 m->m_pkthdr.csum_data + IPPROTO_TCP));
5562                         }
5563                         sum ^= 0xffff;
5564                         ++hw_assist;
5565                 }
5566                 break;
5567         case IPPROTO_UDP:
5568                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5569                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5570                                 sum = m->m_pkthdr.csum_data;
5571                         } else {
5572                                 ip = mtod(m, struct ip *);
5573                                 sum = in_pseudo(ip->ip_src.s_addr,
5574                                 ip->ip_dst.s_addr, htonl((u_short)len +
5575                                 m->m_pkthdr.csum_data + IPPROTO_UDP));
5576                         }
5577                         sum ^= 0xffff;
5578                         ++hw_assist;
5579                 }
5580                 break;
5581         case IPPROTO_ICMP:
5582 #ifdef INET6
5583         case IPPROTO_ICMPV6:
5584 #endif /* INET6 */
5585                 break;
5586         default:
5587                 return (1);
5588         }
5589
5590         if (!hw_assist) {
5591                 switch (af) {
5592                 case AF_INET:
5593                         if (p == IPPROTO_ICMP) {
5594                                 if (m->m_len < off)
5595                                         return (1);
5596                                 m->m_data += off;
5597                                 m->m_len -= off;
5598                                 sum = in_cksum(m, len);
5599                                 m->m_data -= off;
5600                                 m->m_len += off;
5601                         } else {
5602                                 if (m->m_len < sizeof(struct ip))
5603                                         return (1);
5604                                 sum = in4_cksum(m, p, off, len);
5605                         }
5606                         break;
5607 #ifdef INET6
5608                 case AF_INET6:
5609                         if (m->m_len < sizeof(struct ip6_hdr))
5610                                 return (1);
5611                         sum = in6_cksum(m, p, off, len);
5612                         break;
5613 #endif /* INET6 */
5614                 default:
5615                         return (1);
5616                 }
5617         }
5618         if (sum) {
5619                 switch (p) {
5620                 case IPPROTO_TCP:
5621                     {
5622                         KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5623                         break;
5624                     }
5625                 case IPPROTO_UDP:
5626                     {
5627                         KMOD_UDPSTAT_INC(udps_badsum);
5628                         break;
5629                     }
5630 #ifdef INET
5631                 case IPPROTO_ICMP:
5632                     {
5633                         KMOD_ICMPSTAT_INC(icps_checksum);
5634                         break;
5635                     }
5636 #endif
5637 #ifdef INET6
5638                 case IPPROTO_ICMPV6:
5639                     {
5640                         KMOD_ICMP6STAT_INC(icp6s_checksum);
5641                         break;
5642                     }
5643 #endif /* INET6 */
5644                 }
5645                 return (1);
5646         } else {
5647                 if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5648                         m->m_pkthdr.csum_flags |=
5649                             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5650                         m->m_pkthdr.csum_data = 0xffff;
5651                 }
5652         }
5653         return (0);
5654 }
5655
5656
5657 #ifdef INET
5658 int
5659 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5660 {
5661         struct pfi_kif          *kif;
5662         u_short                  action, reason = 0, log = 0;
5663         struct mbuf             *m = *m0;
5664         struct ip               *h = NULL;
5665         struct m_tag            *ipfwtag;
5666         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5667         struct pf_state         *s = NULL;
5668         struct pf_ruleset       *ruleset = NULL;
5669         struct pf_pdesc          pd;
5670         int                      off, dirndx, pqid = 0;
5671
5672         M_ASSERTPKTHDR(m);
5673
5674         if (!V_pf_status.running)
5675                 return (PF_PASS);
5676
5677         memset(&pd, 0, sizeof(pd));
5678
5679         kif = (struct pfi_kif *)ifp->if_pf_kif;
5680
5681         if (kif == NULL) {
5682                 DPFPRINTF(PF_DEBUG_URGENT,
5683                     ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5684                 return (PF_DROP);
5685         }
5686         if (kif->pfik_flags & PFI_IFLAG_SKIP)
5687                 return (PF_PASS);
5688
5689         if (m->m_flags & M_SKIP_FIREWALL)
5690                 return (PF_PASS);
5691
5692         pd.pf_mtag = pf_find_mtag(m);
5693
5694         PF_RULES_RLOCK();
5695
5696         if (ip_divert_ptr != NULL &&
5697             ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5698                 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5699                 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5700                         if (pd.pf_mtag == NULL &&
5701                             ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5702                                 action = PF_DROP;
5703                                 goto done;
5704                         }
5705                         pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5706                         m_tag_delete(m, ipfwtag);
5707                 }
5708                 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5709                         m->m_flags |= M_FASTFWD_OURS;
5710                         pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5711                 }
5712         } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5713                 /* We do IP header normalization and packet reassembly here */
5714                 action = PF_DROP;
5715                 goto done;
5716         }
5717         m = *m0;        /* pf_normalize messes with m0 */
5718         h = mtod(m, struct ip *);
5719
5720         off = h->ip_hl << 2;
5721         if (off < (int)sizeof(struct ip)) {
5722                 action = PF_DROP;
5723                 REASON_SET(&reason, PFRES_SHORT);
5724                 log = 1;
5725                 goto done;
5726         }
5727
5728         pd.src = (struct pf_addr *)&h->ip_src;
5729         pd.dst = (struct pf_addr *)&h->ip_dst;
5730         pd.sport = pd.dport = NULL;
5731         pd.ip_sum = &h->ip_sum;
5732         pd.proto_sum = NULL;
5733         pd.proto = h->ip_p;
5734         pd.dir = dir;
5735         pd.sidx = (dir == PF_IN) ? 0 : 1;
5736         pd.didx = (dir == PF_IN) ? 1 : 0;
5737         pd.af = AF_INET;
5738         pd.tos = h->ip_tos;
5739         pd.tot_len = ntohs(h->ip_len);
5740
5741         /* handle fragments that didn't get reassembled by normalization */
5742         if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
5743                 action = pf_test_fragment(&r, dir, kif, m, h,
5744                     &pd, &a, &ruleset);
5745                 goto done;
5746         }
5747
5748         switch (h->ip_p) {
5749
5750         case IPPROTO_TCP: {
5751                 struct tcphdr   th;
5752
5753                 pd.hdr.tcp = &th;
5754                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
5755                     &action, &reason, AF_INET)) {
5756                         log = action != PF_PASS;
5757                         goto done;
5758                 }
5759                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
5760                 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
5761                         pqid = 1;
5762                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
5763                 if (action == PF_DROP)
5764                         goto done;
5765                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
5766                     &reason);
5767                 if (action == PF_PASS) {
5768                         if (pfsync_update_state_ptr != NULL)
5769                                 pfsync_update_state_ptr(s);
5770                         r = s->rule.ptr;
5771                         a = s->anchor.ptr;
5772                         log = s->log;
5773                 } else if (s == NULL)
5774                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5775                             &a, &ruleset, inp);
5776                 break;
5777         }
5778
5779         case IPPROTO_UDP: {
5780                 struct udphdr   uh;
5781
5782                 pd.hdr.udp = &uh;
5783                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
5784                     &action, &reason, AF_INET)) {
5785                         log = action != PF_PASS;
5786                         goto done;
5787                 }
5788                 if (uh.uh_dport == 0 ||
5789                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
5790                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
5791                         action = PF_DROP;
5792                         REASON_SET(&reason, PFRES_SHORT);
5793                         goto done;
5794                 }
5795                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
5796                 if (action == PF_PASS) {
5797                         if (pfsync_update_state_ptr != NULL)
5798                                 pfsync_update_state_ptr(s);
5799                         r = s->rule.ptr;
5800                         a = s->anchor.ptr;
5801                         log = s->log;
5802                 } else if (s == NULL)
5803                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5804                             &a, &ruleset, inp);
5805                 break;
5806         }
5807
5808         case IPPROTO_ICMP: {
5809                 struct icmp     ih;
5810
5811                 pd.hdr.icmp = &ih;
5812                 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
5813                     &action, &reason, AF_INET)) {
5814                         log = action != PF_PASS;
5815                         goto done;
5816                 }
5817                 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
5818                     &reason);
5819                 if (action == PF_PASS) {
5820                         if (pfsync_update_state_ptr != NULL)
5821                                 pfsync_update_state_ptr(s);
5822                         r = s->rule.ptr;
5823                         a = s->anchor.ptr;
5824                         log = s->log;
5825                 } else if (s == NULL)
5826                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5827                             &a, &ruleset, inp);
5828                 break;
5829         }
5830
5831 #ifdef INET6
5832         case IPPROTO_ICMPV6: {
5833                 action = PF_DROP;
5834                 DPFPRINTF(PF_DEBUG_MISC,
5835                     ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
5836                 goto done;
5837         }
5838 #endif
5839
5840         default:
5841                 action = pf_test_state_other(&s, dir, kif, m, &pd);
5842                 if (action == PF_PASS) {
5843                         if (pfsync_update_state_ptr != NULL)
5844                                 pfsync_update_state_ptr(s);
5845                         r = s->rule.ptr;
5846                         a = s->anchor.ptr;
5847                         log = s->log;
5848                 } else if (s == NULL)
5849                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5850                             &a, &ruleset, inp);
5851                 break;
5852         }
5853
5854 done:
5855         PF_RULES_RUNLOCK();
5856         if (action == PF_PASS && h->ip_hl > 5 &&
5857             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
5858                 action = PF_DROP;
5859                 REASON_SET(&reason, PFRES_IPOPTIONS);
5860                 log = 1;
5861                 DPFPRINTF(PF_DEBUG_MISC,
5862                     ("pf: dropping packet with ip options\n"));
5863         }
5864
5865         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
5866                 action = PF_DROP;
5867                 REASON_SET(&reason, PFRES_MEMORY);
5868         }
5869         if (r->rtableid >= 0)
5870                 M_SETFIB(m, r->rtableid);
5871
5872 #ifdef ALTQ
5873         if (action == PF_PASS && r->qid) {
5874                 if (pd.pf_mtag == NULL &&
5875                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5876                         action = PF_DROP;
5877                         REASON_SET(&reason, PFRES_MEMORY);
5878                 }
5879                 if (pqid || (pd.tos & IPTOS_LOWDELAY))
5880                         pd.pf_mtag->qid = r->pqid;
5881                 else
5882                         pd.pf_mtag->qid = r->qid;
5883                 /* add hints for ecn */
5884                 pd.pf_mtag->hdr = h;
5885
5886         }
5887 #endif /* ALTQ */
5888
5889         /*
5890          * connections redirected to loopback should not match sockets
5891          * bound specifically to loopback due to security implications,
5892          * see tcp_input() and in_pcblookup_listen().
5893          */
5894         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
5895             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
5896             (s->nat_rule.ptr->action == PF_RDR ||
5897             s->nat_rule.ptr->action == PF_BINAT) &&
5898             (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
5899                 m->m_flags |= M_SKIP_FIREWALL;
5900
5901         if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
5902             !PACKET_LOOPED(&pd)) {
5903
5904                 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
5905                     sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
5906                 if (ipfwtag != NULL) {
5907                         ((struct ipfw_rule_ref *)(ipfwtag+1))->info =
5908                             ntohs(r->divert.port);
5909                         ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
5910
5911                         if (s)
5912                                 PF_STATE_UNLOCK(s);
5913
5914                         m_tag_prepend(m, ipfwtag);
5915                         if (m->m_flags & M_FASTFWD_OURS) {
5916                                 if (pd.pf_mtag == NULL &&
5917                                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5918                                         action = PF_DROP;
5919                                         REASON_SET(&reason, PFRES_MEMORY);
5920                                         log = 1;
5921                                         DPFPRINTF(PF_DEBUG_MISC,
5922                                             ("pf: failed to allocate tag\n"));
5923                                 }
5924                                 pd.pf_mtag->flags |= PF_FASTFWD_OURS_PRESENT;
5925                                 m->m_flags &= ~M_FASTFWD_OURS;
5926                         }
5927                         ip_divert_ptr(*m0, dir ==  PF_IN ? DIR_IN : DIR_OUT);
5928                         *m0 = NULL;
5929
5930                         return (action);
5931                 } else {
5932                         /* XXX: ipfw has the same behaviour! */
5933                         action = PF_DROP;
5934                         REASON_SET(&reason, PFRES_MEMORY);
5935                         log = 1;
5936                         DPFPRINTF(PF_DEBUG_MISC,
5937                             ("pf: failed to allocate divert tag\n"));
5938                 }
5939         }
5940
5941         if (log) {
5942                 struct pf_rule *lr;
5943
5944                 if (s != NULL && s->nat_rule.ptr != NULL &&
5945                     s->nat_rule.ptr->log & PF_LOG_ALL)
5946                         lr = s->nat_rule.ptr;
5947                 else
5948                         lr = r;
5949                 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
5950                     (s == NULL));
5951         }
5952
5953         kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
5954         kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
5955
5956         if (action == PF_PASS || r->action == PF_DROP) {
5957                 dirndx = (dir == PF_OUT);
5958                 r->packets[dirndx]++;
5959                 r->bytes[dirndx] += pd.tot_len;
5960                 if (a != NULL) {
5961                         a->packets[dirndx]++;
5962                         a->bytes[dirndx] += pd.tot_len;
5963                 }
5964                 if (s != NULL) {
5965                         if (s->nat_rule.ptr != NULL) {
5966                                 s->nat_rule.ptr->packets[dirndx]++;
5967                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
5968                         }
5969                         if (s->src_node != NULL) {
5970                                 s->src_node->packets[dirndx]++;
5971                                 s->src_node->bytes[dirndx] += pd.tot_len;
5972                         }
5973                         if (s->nat_src_node != NULL) {
5974                                 s->nat_src_node->packets[dirndx]++;
5975                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
5976                         }
5977                         dirndx = (dir == s->direction) ? 0 : 1;
5978                         s->packets[dirndx]++;
5979                         s->bytes[dirndx] += pd.tot_len;
5980                 }
5981                 tr = r;
5982                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
5983                 if (nr != NULL && r == &V_pf_default_rule)
5984                         tr = nr;
5985                 if (tr->src.addr.type == PF_ADDR_TABLE)
5986                         pfr_update_stats(tr->src.addr.p.tbl,
5987                             (s == NULL) ? pd.src :
5988                             &s->key[(s->direction == PF_IN)]->
5989                                 addr[(s->direction == PF_OUT)],
5990                             pd.af, pd.tot_len, dir == PF_OUT,
5991                             r->action == PF_PASS, tr->src.neg);
5992                 if (tr->dst.addr.type == PF_ADDR_TABLE)
5993                         pfr_update_stats(tr->dst.addr.p.tbl,
5994                             (s == NULL) ? pd.dst :
5995                             &s->key[(s->direction == PF_IN)]->
5996                                 addr[(s->direction == PF_IN)],
5997                             pd.af, pd.tot_len, dir == PF_OUT,
5998                             r->action == PF_PASS, tr->dst.neg);
5999         }
6000
6001         switch (action) {
6002         case PF_SYNPROXY_DROP:
6003                 m_freem(*m0);
6004         case PF_DEFER:
6005                 *m0 = NULL;
6006                 action = PF_PASS;
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         default:
6385                 /* pf_route6() returns unlocked. */
6386                 if (r->rt) {
6387                         pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
6388                         return (action);
6389                 }
6390                 break;
6391         }
6392
6393         if (s)
6394                 PF_STATE_UNLOCK(s);
6395
6396         return (action);
6397 }
6398 #endif /* INET6 */