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 void              pf_set_rt_ifp(struct pf_state *,
 269                             struct pf_addr *);
 270 static int               pf_check_proto_cksum(struct mbuf *, int, int,
 271                             u_int8_t, sa_family_t);
 272 static void              pf_print_state_parts(struct pf_state *,
 273                             struct pf_state_key *, struct pf_state_key *);
 274 static int               pf_addr_wrap_neq(struct pf_addr_wrap *,
 275                             struct pf_addr_wrap *);
 276 static struct pf_state  *pf_find_state(struct pfi_kif *,
 277                             struct pf_state_key_cmp *, u_int);
 278 static int               pf_src_connlimit(struct pf_state **);
 279 static void              pf_overload_task(void *v, int pending);
 280 static int               pf_insert_src_node(struct pf_src_node **,
 281                             struct pf_rule *, struct pf_addr *, sa_family_t);
 282 static u_int             pf_purge_expired_states(u_int, int);
 283 static void              pf_purge_unlinked_rules(void);
 284 static int               pf_mtag_uminit(void *, int, int);
 285 static void              pf_mtag_free(struct m_tag *);
 286 #ifdef INET
 287 static void              pf_route(struct mbuf **, struct pf_rule *, int,
 288                             struct ifnet *, struct pf_state *,
 289                             struct pf_pdesc *);
 290 #endif /* INET */
 291 #ifdef INET6
 292 static void              pf_change_a6(struct pf_addr *, u_int16_t *,
 293                             struct pf_addr *, u_int8_t);
 294 static void              pf_route6(struct mbuf **, struct pf_rule *, int,
 295                             struct ifnet *, struct pf_state *,
 296                             struct pf_pdesc *);
 297 #endif /* INET6 */
 298
 299 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
 300
 301 VNET_DECLARE(int, pf_end_threads);
 302
 303 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
 304
 305 #define PACKET_LOOPED(pd)       ((pd)->pf_mtag &&                       \
 306                                  (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
 307
 308 #define STATE_LOOKUP(i, k, d, s, pd)                                    \
 309         do {                                                            \
 310                 (s) = pf_find_state((i), (k), (d));                     \
 311                 if ((s) == NULL)                                        \
 312                         return (PF_DROP);                               \
 313                 if (PACKET_LOOPED(pd))                                  \
 314                         return (PF_PASS);                               \
 315                 if ((d) == PF_OUT &&                                    \
 316                     (((s)->rule.ptr->rt == PF_ROUTETO &&                \
 317                     (s)->rule.ptr->direction == PF_OUT) ||              \
 318                     ((s)->rule.ptr->rt == PF_REPLYTO &&                 \
 319                     (s)->rule.ptr->direction == PF_IN)) &&              \
 320                     (s)->rt_kif != NULL &&                              \
 321                     (s)->rt_kif != (i))                                 \
 322                         return (PF_PASS);                               \
 323         } while (0)
 324
 325 #define BOUND_IFACE(r, k) \
 326         ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
 327
 328 #define STATE_INC_COUNTERS(s)                                           \
 329         do {                                                            \
 330                 counter_u64_add(s->rule.ptr->states_cur, 1);            \
 331                 counter_u64_add(s->rule.ptr->states_tot, 1);            \
 332                 if (s->anchor.ptr != NULL) {                            \
 333                         counter_u64_add(s->anchor.ptr->states_cur, 1);  \
 334                         counter_u64_add(s->anchor.ptr->states_tot, 1);  \
 335                 }                                                       \
 336                 if (s->nat_rule.ptr != NULL) {                          \
 337                         counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
 338                         counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
 339                 }                                                       \
 340         } while (0)
 341
 342 #define STATE_DEC_COUNTERS(s)                                           \
 343         do {                                                            \
 344                 if (s->nat_rule.ptr != NULL)                            \
 345                         counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
 346                 if (s->anchor.ptr != NULL)                              \
 347                         counter_u64_add(s->anchor.ptr->states_cur, -1); \
 348                 counter_u64_add(s->rule.ptr->states_cur, -1);           \
 349         } while (0)
 350
 351 static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
 352 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
 353 VNET_DEFINE(struct pf_idhash *, pf_idhash);
 354 VNET_DEFINE(u_long, pf_hashmask);
 355 VNET_DEFINE(struct pf_srchash *, pf_srchash);
 356 VNET_DEFINE(u_long, pf_srchashmask);
 357
 358 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)");
 359
 360 VNET_DEFINE(u_long, pf_hashsize);
 361 #define V_pf_hashsize   VNET(pf_hashsize)
 362 SYSCTL_VNET_UINT(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
 363     &VNET_NAME(pf_hashsize), 0, "Size of pf(4) states hashtable");
 364
 365 VNET_DEFINE(u_long, pf_srchashsize);
 366 #define V_pf_srchashsize        VNET(pf_srchashsize)
 367 SYSCTL_VNET_UINT(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
 368     &VNET_NAME(pf_srchashsize), 0, "Size of pf(4) source nodes hashtable");
 369
 370 VNET_DEFINE(void *, pf_swi_cookie);
 371
 372 VNET_DEFINE(uint32_t, pf_hashseed);
 373 #define V_pf_hashseed   VNET(pf_hashseed)
 374
 375 static __inline uint32_t
 376 pf_hashkey(struct pf_state_key *sk)
 377 {
 378         uint32_t h;
 379
 380         h = jenkins_hash32((uint32_t *)sk,
 381             sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
 382             V_pf_hashseed);
 383
 384         return (h & V_pf_hashmask);
 385 }
 386
 387 static __inline uint32_t
 388 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
 389 {
 390         uint32_t h;
 391
 392         switch (af) {
 393         case AF_INET:
 394                 h = jenkins_hash32((uint32_t *)&addr->v4,
 395                     sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
 396                 break;
 397         case AF_INET6:
 398                 h = jenkins_hash32((uint32_t *)&addr->v6,
 399                     sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
 400                 break;
 401         default:
 402                 panic("%s: unknown address family %u", __func__, af);
 403         }
 404
 405         return (h & V_pf_srchashmask);
 406 }
 407
 408 #ifdef INET6
 409 void
 410 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
 411 {
 412         switch (af) {
 413 #ifdef INET
 414         case AF_INET:
 415                 dst->addr32[0] = src->addr32[0];
 416                 break;
 417 #endif /* INET */
 418         case AF_INET6:
 419                 dst->addr32[0] = src->addr32[0];
 420                 dst->addr32[1] = src->addr32[1];
 421                 dst->addr32[2] = src->addr32[2];
 422                 dst->addr32[3] = src->addr32[3];
 423                 break;
 424         }
 425 }
 426 #endif /* INET6 */
 427
 428 static void
 429 pf_init_threshold(struct pf_threshold *threshold,
 430     u_int32_t limit, u_int32_t seconds)
 431 {
 432         threshold->limit = limit * PF_THRESHOLD_MULT;
 433         threshold->seconds = seconds;
 434         threshold->count = 0;
 435         threshold->last = time_uptime;
 436 }
 437
 438 static void
 439 pf_add_threshold(struct pf_threshold *threshold)
 440 {
 441         u_int32_t t = time_uptime, diff = t - threshold->last;
 442
 443         if (diff >= threshold->seconds)
 444                 threshold->count = 0;
 445         else
 446                 threshold->count -= threshold->count * diff /
 447                     threshold->seconds;
 448         threshold->count += PF_THRESHOLD_MULT;
 449         threshold->last = t;
 450 }
 451
 452 static int
 453 pf_check_threshold(struct pf_threshold *threshold)
 454 {
 455         return (threshold->count > threshold->limit);
 456 }
 457
 458 static int
 459 pf_src_connlimit(struct pf_state **state)
 460 {
 461         struct pf_overload_entry *pfoe;
 462         int bad = 0;
 463
 464         PF_STATE_LOCK_ASSERT(*state);
 465
 466         (*state)->src_node->conn++;
 467         (*state)->src.tcp_est = 1;
 468         pf_add_threshold(&(*state)->src_node->conn_rate);
 469
 470         if ((*state)->rule.ptr->max_src_conn &&
 471             (*state)->rule.ptr->max_src_conn <
 472             (*state)->src_node->conn) {
 473                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
 474                 bad++;
 475         }
 476
 477         if ((*state)->rule.ptr->max_src_conn_rate.limit &&
 478             pf_check_threshold(&(*state)->src_node->conn_rate)) {
 479                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
 480                 bad++;
 481         }
 482
 483         if (!bad)
 484                 return (0);
 485
 486         /* Kill this state. */
 487         (*state)->timeout = PFTM_PURGE;
 488         (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
 489
 490         if ((*state)->rule.ptr->overload_tbl == NULL)
 491                 return (1);
 492
 493         /* Schedule overloading and flushing task. */
 494         pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
 495         if (pfoe == NULL)
 496                 return (1);     /* too bad :( */
 497
 498         bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
 499         pfoe->af = (*state)->key[PF_SK_WIRE]->af;
 500         pfoe->rule = (*state)->rule.ptr;
 501         pfoe->dir = (*state)->direction;
 502         PF_OVERLOADQ_LOCK();
 503         SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
 504         PF_OVERLOADQ_UNLOCK();
 505         taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
 506
 507         return (1);
 508 }
 509
 510 static void
 511 pf_overload_task(void *v, int pending)
 512 {
 513         struct pf_overload_head queue;
 514         struct pfr_addr p;
 515         struct pf_overload_entry *pfoe, *pfoe1;
 516         uint32_t killed = 0;
 517
 518         CURVNET_SET((struct vnet *)v);
 519
 520         PF_OVERLOADQ_LOCK();
 521         queue = V_pf_overloadqueue;
 522         SLIST_INIT(&V_pf_overloadqueue);
 523         PF_OVERLOADQ_UNLOCK();
 524
 525         bzero(&p, sizeof(p));
 526         SLIST_FOREACH(pfoe, &queue, next) {
 527                 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
 528                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
 529                         printf("%s: blocking address ", __func__);
 530                         pf_print_host(&pfoe->addr, 0, pfoe->af);
 531                         printf("\n");
 532                 }
 533
 534                 p.pfra_af = pfoe->af;
 535                 switch (pfoe->af) {
 536 #ifdef INET
 537                 case AF_INET:
 538                         p.pfra_net = 32;
 539                         p.pfra_ip4addr = pfoe->addr.v4;
 540                         break;
 541 #endif
 542 #ifdef INET6
 543                 case AF_INET6:
 544                         p.pfra_net = 128;
 545                         p.pfra_ip6addr = pfoe->addr.v6;
 546                         break;
 547 #endif
 548                 }
 549
 550                 PF_RULES_WLOCK();
 551                 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
 552                 PF_RULES_WUNLOCK();
 553         }
 554
 555         /*
 556          * Remove those entries, that don't need flushing.
 557          */
 558         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 559                 if (pfoe->rule->flush == 0) {
 560                         SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
 561                         free(pfoe, M_PFTEMP);
 562                 } else
 563                         counter_u64_add(
 564                             V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
 565
 566         /* If nothing to flush, return. */
 567         if (SLIST_EMPTY(&queue)) {
 568                 CURVNET_RESTORE();
 569                 return;
 570         }
 571
 572         for (int i = 0; i <= V_pf_hashmask; i++) {
 573                 struct pf_idhash *ih = &V_pf_idhash[i];
 574                 struct pf_state_key *sk;
 575                 struct pf_state *s;
 576
 577                 PF_HASHROW_LOCK(ih);
 578                 LIST_FOREACH(s, &ih->states, entry) {
 579                     sk = s->key[PF_SK_WIRE];
 580                     SLIST_FOREACH(pfoe, &queue, next)
 581                         if (sk->af == pfoe->af &&
 582                             ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
 583                             pfoe->rule == s->rule.ptr) &&
 584                             ((pfoe->dir == PF_OUT &&
 585                             PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
 586                             (pfoe->dir == PF_IN &&
 587                             PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
 588                                 s->timeout = PFTM_PURGE;
 589                                 s->src.state = s->dst.state = TCPS_CLOSED;
 590                                 killed++;
 591                         }
 592                 }
 593                 PF_HASHROW_UNLOCK(ih);
 594         }
 595         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 596                 free(pfoe, M_PFTEMP);
 597         if (V_pf_status.debug >= PF_DEBUG_MISC)
 598                 printf("%s: %u states killed", __func__, killed);
 599
 600         CURVNET_RESTORE();
 601 }
 602
 603 /*
 604  * Can return locked on failure, so that we can consistently
 605  * allocate and insert a new one.
 606  */
 607 struct pf_src_node *
 608 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
 609         int returnlocked)
 610 {
 611         struct pf_srchash *sh;
 612         struct pf_src_node *n;
 613
 614         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
 615
 616         sh = &V_pf_srchash[pf_hashsrc(src, af)];
 617         PF_HASHROW_LOCK(sh);
 618         LIST_FOREACH(n, &sh->nodes, entry)
 619                 if (n->rule.ptr == rule && n->af == af &&
 620                     ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
 621                     (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
 622                         break;
 623         if (n != NULL || returnlocked == 0)
 624                 PF_HASHROW_UNLOCK(sh);
 625
 626         return (n);
 627 }
 628
 629 static int
 630 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
 631     struct pf_addr *src, sa_family_t af)
 632 {
 633
 634         KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
 635             rule->rpool.opts & PF_POOL_STICKYADDR),
 636             ("%s for non-tracking rule %p", __func__, rule));
 637
 638         if (*sn == NULL)
 639                 *sn = pf_find_src_node(src, rule, af, 1);
 640
 641         if (*sn == NULL) {
 642                 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
 643
 644                 PF_HASHROW_ASSERT(sh);
 645
 646                 if (!rule->max_src_nodes ||
 647                     counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
 648                         (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
 649                 else
 650                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
 651                             1);
 652                 if ((*sn) == NULL) {
 653                         PF_HASHROW_UNLOCK(sh);
 654                         return (-1);
 655                 }
 656
 657                 pf_init_threshold(&(*sn)->conn_rate,
 658                     rule->max_src_conn_rate.limit,
 659                     rule->max_src_conn_rate.seconds);
 660
 661                 (*sn)->af = af;
 662                 (*sn)->rule.ptr = rule;
 663                 PF_ACPY(&(*sn)->addr, src, af);
 664                 LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
 665                 (*sn)->creation = time_uptime;
 666                 (*sn)->ruletype = rule->action;
 667                 if ((*sn)->rule.ptr != NULL)
 668                         counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
 669                 PF_HASHROW_UNLOCK(sh);
 670                 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
 671         } else {
 672                 if (rule->max_src_states &&
 673                     (*sn)->states >= rule->max_src_states) {
 674                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
 675                             1);
 676                         return (-1);
 677                 }
 678         }
 679         return (0);
 680 }
 681
 682 void
 683 pf_unlink_src_node_locked(struct pf_src_node *src)
 684 {
 685 #ifdef INVARIANTS
 686         struct pf_srchash *sh;
 687
 688         sh = &V_pf_srchash[pf_hashsrc(&src->addr, src->af)];
 689         PF_HASHROW_ASSERT(sh);
 690 #endif
 691         LIST_REMOVE(src, entry);
 692         if (src->rule.ptr)
 693                 counter_u64_add(src->rule.ptr->src_nodes, -1);
 694         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
 695 }
 696
 697 void
 698 pf_unlink_src_node(struct pf_src_node *src)
 699 {
 700         struct pf_srchash *sh;
 701
 702         sh = &V_pf_srchash[pf_hashsrc(&src->addr, src->af)];
 703         PF_HASHROW_LOCK(sh);
 704         pf_unlink_src_node_locked(src);
 705         PF_HASHROW_UNLOCK(sh);
 706 }
 707
 708 static void
 709 pf_free_src_node(struct pf_src_node *sn)
 710 {
 711
 712         KASSERT(sn->states == 0, ("%s: %p has refs", __func__, sn));
 713         uma_zfree(V_pf_sources_z, sn);
 714 }
 715
 716 u_int
 717 pf_free_src_nodes(struct pf_src_node_list *head)
 718 {
 719         struct pf_src_node *sn, *tmp;
 720         u_int count = 0;
 721
 722         LIST_FOREACH_SAFE(sn, head, entry, tmp) {
 723                 pf_free_src_node(sn);
 724                 count++;
 725         }
 726
 727         return (count);
 728 }
 729
 730 void
 731 pf_mtag_initialize()
 732 {
 733
 734         pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
 735             sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
 736             UMA_ALIGN_PTR, 0);
 737 }
 738
 739 /* Per-vnet data storage structures initialization. */
 740 void
 741 pf_initialize()
 742 {
 743         struct pf_keyhash       *kh;
 744         struct pf_idhash        *ih;
 745         struct pf_srchash       *sh;
 746         u_int i;
 747
 748         TUNABLE_ULONG_FETCH("net.pf.states_hashsize", &V_pf_hashsize);
 749         if (V_pf_hashsize == 0 || !powerof2(V_pf_hashsize))
 750                 V_pf_hashsize = PF_HASHSIZ;
 751         TUNABLE_ULONG_FETCH("net.pf.source_nodes_hashsize", &V_pf_srchashsize);
 752         if (V_pf_srchashsize == 0 || !powerof2(V_pf_srchashsize))
 753                 V_pf_srchashsize = PF_HASHSIZ / 4;
 754
 755         V_pf_hashseed = arc4random();
 756
 757         /* States and state keys storage. */
 758         V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
 759             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 760         V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
 761         uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
 762         uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
 763
 764         V_pf_state_key_z = uma_zcreate("pf state keys",
 765             sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
 766             UMA_ALIGN_PTR, 0);
 767         V_pf_keyhash = malloc(V_pf_hashsize * sizeof(struct pf_keyhash),
 768             M_PFHASH, M_WAITOK | M_ZERO);
 769         V_pf_idhash = malloc(V_pf_hashsize * sizeof(struct pf_idhash),
 770             M_PFHASH, M_WAITOK | M_ZERO);
 771         V_pf_hashmask = V_pf_hashsize - 1;
 772         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
 773             i++, kh++, ih++) {
 774                 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
 775                 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
 776         }
 777
 778         /* Source nodes. */
 779         V_pf_sources_z = uma_zcreate("pf source nodes",
 780             sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
 781             0);
 782         V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
 783         uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
 784         uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
 785         V_pf_srchash = malloc(V_pf_srchashsize * sizeof(struct pf_srchash),
 786           M_PFHASH, M_WAITOK|M_ZERO);
 787         V_pf_srchashmask = V_pf_srchashsize - 1;
 788         for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++)
 789                 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
 790
 791         /* ALTQ */
 792         TAILQ_INIT(&V_pf_altqs[0]);
 793         TAILQ_INIT(&V_pf_altqs[1]);
 794         TAILQ_INIT(&V_pf_pabuf);
 795         V_pf_altqs_active = &V_pf_altqs[0];
 796         V_pf_altqs_inactive = &V_pf_altqs[1];
 797
 798
 799         /* Send & overload+flush queues. */
 800         STAILQ_INIT(&V_pf_sendqueue);
 801         SLIST_INIT(&V_pf_overloadqueue);
 802         TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
 803         mtx_init(&pf_sendqueue_mtx, "pf send queue", NULL, MTX_DEF);
 804         mtx_init(&pf_overloadqueue_mtx, "pf overload/flush queue", NULL,
 805             MTX_DEF);
 806
 807         /* Unlinked, but may be referenced rules. */
 808         TAILQ_INIT(&V_pf_unlinked_rules);
 809         mtx_init(&pf_unlnkdrules_mtx, "pf unlinked rules", NULL, MTX_DEF);
 810 }
 811
 812 void
 813 pf_mtag_cleanup()
 814 {
 815
 816         uma_zdestroy(pf_mtag_z);
 817 }
 818
 819 void
 820 pf_cleanup()
 821 {
 822         struct pf_keyhash       *kh;
 823         struct pf_idhash        *ih;
 824         struct pf_srchash       *sh;
 825         struct pf_send_entry    *pfse, *next;
 826         u_int i;
 827
 828         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= V_pf_hashmask;
 829             i++, kh++, ih++) {
 830                 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
 831                     __func__));
 832                 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
 833                     __func__));
 834                 mtx_destroy(&kh->lock);
 835                 mtx_destroy(&ih->lock);
 836         }
 837         free(V_pf_keyhash, M_PFHASH);
 838         free(V_pf_idhash, M_PFHASH);
 839
 840         for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
 841                 KASSERT(LIST_EMPTY(&sh->nodes),
 842                     ("%s: source node hash not empty", __func__));
 843                 mtx_destroy(&sh->lock);
 844         }
 845         free(V_pf_srchash, M_PFHASH);
 846
 847         STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
 848                 m_freem(pfse->pfse_m);
 849                 free(pfse, M_PFTEMP);
 850         }
 851
 852         mtx_destroy(&pf_sendqueue_mtx);
 853         mtx_destroy(&pf_overloadqueue_mtx);
 854         mtx_destroy(&pf_unlnkdrules_mtx);
 855
 856         uma_zdestroy(V_pf_sources_z);
 857         uma_zdestroy(V_pf_state_z);
 858         uma_zdestroy(V_pf_state_key_z);
 859 }
 860
 861 static int
 862 pf_mtag_uminit(void *mem, int size, int how)
 863 {
 864         struct m_tag *t;
 865
 866         t = (struct m_tag *)mem;
 867         t->m_tag_cookie = MTAG_ABI_COMPAT;
 868         t->m_tag_id = PACKET_TAG_PF;
 869         t->m_tag_len = sizeof(struct pf_mtag);
 870         t->m_tag_free = pf_mtag_free;
 871
 872         return (0);
 873 }
 874
 875 static void
 876 pf_mtag_free(struct m_tag *t)
 877 {
 878
 879         uma_zfree(pf_mtag_z, t);
 880 }
 881
 882 struct pf_mtag *
 883 pf_get_mtag(struct mbuf *m)
 884 {
 885         struct m_tag *mtag;
 886
 887         if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
 888                 return ((struct pf_mtag *)(mtag + 1));
 889
 890         mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
 891         if (mtag == NULL)
 892                 return (NULL);
 893         bzero(mtag + 1, sizeof(struct pf_mtag));
 894         m_tag_prepend(m, mtag);
 895
 896         return ((struct pf_mtag *)(mtag + 1));
 897 }
 898
 899 static int
 900 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
 901     struct pf_state *s)
 902 {
 903         struct pf_keyhash       *khs, *khw, *kh;
 904         struct pf_state_key     *sk, *cur;
 905         struct pf_state         *si, *olds = NULL;
 906         int idx;
 907
 908         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
 909         KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
 910         KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
 911
 912         /*
 913          * We need to lock hash slots of both keys. To avoid deadlock
 914          * we always lock the slot with lower address first. Unlock order
 915          * isn't important.
 916          *
 917          * We also need to lock ID hash slot before dropping key
 918          * locks. On success we return with ID hash slot locked.
 919          */
 920
 921         if (skw == sks) {
 922                 khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
 923                 PF_HASHROW_LOCK(khs);
 924         } else {
 925                 khs = &V_pf_keyhash[pf_hashkey(sks)];
 926                 khw = &V_pf_keyhash[pf_hashkey(skw)];
 927                 if (khs == khw) {
 928                         PF_HASHROW_LOCK(khs);
 929                 } else if (khs < khw) {
 930                         PF_HASHROW_LOCK(khs);
 931                         PF_HASHROW_LOCK(khw);
 932                 } else {
 933                         PF_HASHROW_LOCK(khw);
 934                         PF_HASHROW_LOCK(khs);
 935                 }
 936         }
 937
 938 #define KEYS_UNLOCK()   do {                    \
 939         if (khs != khw) {                       \
 940                 PF_HASHROW_UNLOCK(khs);         \
 941                 PF_HASHROW_UNLOCK(khw);         \
 942         } else                                  \
 943                 PF_HASHROW_UNLOCK(khs);         \
 944 } while (0)
 945
 946         /*
 947          * First run: start with wire key.
 948          */
 949         sk = skw;
 950         kh = khw;
 951         idx = PF_SK_WIRE;
 952
 953 keyattach:
 954         LIST_FOREACH(cur, &kh->keys, entry)
 955                 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
 956                         break;
 957
 958         if (cur != NULL) {
 959                 /* Key exists. Check for same kif, if none, add to key. */
 960                 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
 961                         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
 962
 963                         PF_HASHROW_LOCK(ih);
 964                         if (si->kif == s->kif &&
 965                             si->direction == s->direction) {
 966                                 if (sk->proto == IPPROTO_TCP &&
 967                                     si->src.state >= TCPS_FIN_WAIT_2 &&
 968                                     si->dst.state >= TCPS_FIN_WAIT_2) {
 969                                         /*
 970                                          * New state matches an old >FIN_WAIT_2
 971                                          * state. We can't drop key hash locks,
 972                                          * thus we can't unlink it properly.
 973                                          *
 974                                          * As a workaround we drop it into
 975                                          * TCPS_CLOSED state, schedule purge
 976                                          * ASAP and push it into the very end
 977                                          * of the slot TAILQ, so that it won't
 978                                          * conflict with our new state.
 979                                          */
 980                                         si->src.state = si->dst.state =
 981                                             TCPS_CLOSED;
 982                                         si->timeout = PFTM_PURGE;
 983                                         olds = si;
 984                                 } else {
 985                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
 986                                                 printf("pf: %s key attach "
 987                                                     "failed on %s: ",
 988                                                     (idx == PF_SK_WIRE) ?
 989                                                     "wire" : "stack",
 990                                                     s->kif->pfik_name);
 991                                                 pf_print_state_parts(s,
 992                                                     (idx == PF_SK_WIRE) ?
 993                                                     sk : NULL,
 994                                                     (idx == PF_SK_STACK) ?
 995                                                     sk : NULL);
 996                                                 printf(", existing: ");
 997                                                 pf_print_state_parts(si,
 998                                                     (idx == PF_SK_WIRE) ?
 999                                                     sk : NULL,
1000                                                     (idx == PF_SK_STACK) ?
1001                                                     sk : NULL);
1002                                                 printf("\n");
1003                                         }
1004                                         PF_HASHROW_UNLOCK(ih);
1005                                         KEYS_UNLOCK();
1006                                         uma_zfree(V_pf_state_key_z, sk);
1007                                         if (idx == PF_SK_STACK)
1008                                                 pf_detach_state(s);
1009                                         return (EEXIST); /* collision! */
1010                                 }
1011                         }
1012                         PF_HASHROW_UNLOCK(ih);
1013                 }
1014                 uma_zfree(V_pf_state_key_z, sk);
1015                 s->key[idx] = cur;
1016         } else {
1017                 LIST_INSERT_HEAD(&kh->keys, sk, entry);
1018                 s->key[idx] = sk;
1019         }
1020
1021 stateattach:
1022         /* List is sorted, if-bound states before floating. */
1023         if (s->kif == V_pfi_all)
1024                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1025         else
1026                 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1027
1028         if (olds) {
1029                 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1030                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1031                     key_list[idx]);
1032                 olds = NULL;
1033         }
1034
1035         /*
1036          * Attach done. See how should we (or should not?)
1037          * attach a second key.
1038          */
1039         if (sks == skw) {
1040                 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1041                 idx = PF_SK_STACK;
1042                 sks = NULL;
1043                 goto stateattach;
1044         } else if (sks != NULL) {
1045                 /*
1046                  * Continue attaching with stack key.
1047                  */
1048                 sk = sks;
1049                 kh = khs;
1050                 idx = PF_SK_STACK;
1051                 sks = NULL;
1052                 goto keyattach;
1053         }
1054
1055         PF_STATE_LOCK(s);
1056         KEYS_UNLOCK();
1057
1058         KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1059             ("%s failure", __func__));
1060
1061         return (0);
1062 #undef  KEYS_UNLOCK
1063 }
1064
1065 static void
1066 pf_detach_state(struct pf_state *s)
1067 {
1068         struct pf_state_key *sks = s->key[PF_SK_STACK];
1069         struct pf_keyhash *kh;
1070
1071         if (sks != NULL) {
1072                 kh = &V_pf_keyhash[pf_hashkey(sks)];
1073                 PF_HASHROW_LOCK(kh);
1074                 if (s->key[PF_SK_STACK] != NULL)
1075                         pf_state_key_detach(s, PF_SK_STACK);
1076                 /*
1077                  * If both point to same key, then we are done.
1078                  */
1079                 if (sks == s->key[PF_SK_WIRE]) {
1080                         pf_state_key_detach(s, PF_SK_WIRE);
1081                         PF_HASHROW_UNLOCK(kh);
1082                         return;
1083                 }
1084                 PF_HASHROW_UNLOCK(kh);
1085         }
1086
1087         if (s->key[PF_SK_WIRE] != NULL) {
1088                 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1089                 PF_HASHROW_LOCK(kh);
1090                 if (s->key[PF_SK_WIRE] != NULL)
1091                         pf_state_key_detach(s, PF_SK_WIRE);
1092                 PF_HASHROW_UNLOCK(kh);
1093         }
1094 }
1095
1096 static void
1097 pf_state_key_detach(struct pf_state *s, int idx)
1098 {
1099         struct pf_state_key *sk = s->key[idx];
1100 #ifdef INVARIANTS
1101         struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1102
1103         PF_HASHROW_ASSERT(kh);
1104 #endif
1105         TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1106         s->key[idx] = NULL;
1107
1108         if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1109                 LIST_REMOVE(sk, entry);
1110                 uma_zfree(V_pf_state_key_z, sk);
1111         }
1112 }
1113
1114 static int
1115 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1116 {
1117         struct pf_state_key *sk = mem;
1118
1119         bzero(sk, sizeof(struct pf_state_key_cmp));
1120         TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1121         TAILQ_INIT(&sk->states[PF_SK_STACK]);
1122
1123         return (0);
1124 }
1125
1126 struct pf_state_key *
1127 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1128         struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1129 {
1130         struct pf_state_key *sk;
1131
1132         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1133         if (sk == NULL)
1134                 return (NULL);
1135
1136         PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1137         PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1138         sk->port[pd->sidx] = sport;
1139         sk->port[pd->didx] = dport;
1140         sk->proto = pd->proto;
1141         sk->af = pd->af;
1142
1143         return (sk);
1144 }
1145
1146 struct pf_state_key *
1147 pf_state_key_clone(struct pf_state_key *orig)
1148 {
1149         struct pf_state_key *sk;
1150
1151         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1152         if (sk == NULL)
1153                 return (NULL);
1154
1155         bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1156
1157         return (sk);
1158 }
1159
1160 int
1161 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1162     struct pf_state_key *sks, struct pf_state *s)
1163 {
1164         struct pf_idhash *ih;
1165         struct pf_state *cur;
1166         int error;
1167
1168         KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1169             ("%s: sks not pristine", __func__));
1170         KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1171             ("%s: skw not pristine", __func__));
1172         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1173
1174         s->kif = kif;
1175
1176         if (s->id == 0 && s->creatorid == 0) {
1177                 /* XXX: should be atomic, but probability of collision low */
1178                 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1179                         V_pf_stateid[curcpu] = 1;
1180                 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1181                 s->id = htobe64(s->id);
1182                 s->creatorid = V_pf_status.hostid;
1183         }
1184
1185         /* Returns with ID locked on success. */
1186         if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1187                 return (error);
1188
1189         ih = &V_pf_idhash[PF_IDHASH(s)];
1190         PF_HASHROW_ASSERT(ih);
1191         LIST_FOREACH(cur, &ih->states, entry)
1192                 if (cur->id == s->id && cur->creatorid == s->creatorid)
1193                         break;
1194
1195         if (cur != NULL) {
1196                 PF_HASHROW_UNLOCK(ih);
1197                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1198                         printf("pf: state ID collision: "
1199                             "id: %016llx creatorid: %08x\n",
1200                             (unsigned long long)be64toh(s->id),
1201                             ntohl(s->creatorid));
1202                 }
1203                 pf_detach_state(s);
1204                 return (EEXIST);
1205         }
1206         LIST_INSERT_HEAD(&ih->states, s, entry);
1207         /* One for keys, one for ID hash. */
1208         refcount_init(&s->refs, 2);
1209
1210         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1211         if (pfsync_insert_state_ptr != NULL)
1212                 pfsync_insert_state_ptr(s);
1213
1214         /* Returns locked. */
1215         return (0);
1216 }
1217
1218 /*
1219  * Find state by ID: returns with locked row on success.
1220  */
1221 struct pf_state *
1222 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1223 {
1224         struct pf_idhash *ih;
1225         struct pf_state *s;
1226
1227         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1228
1229         ih = &V_pf_idhash[(be64toh(id) % (V_pf_hashmask + 1))];
1230
1231         PF_HASHROW_LOCK(ih);
1232         LIST_FOREACH(s, &ih->states, entry)
1233                 if (s->id == id && s->creatorid == creatorid)
1234                         break;
1235
1236         if (s == NULL)
1237                 PF_HASHROW_UNLOCK(ih);
1238
1239         return (s);
1240 }
1241
1242 /*
1243  * Find state by key.
1244  * Returns with ID hash slot locked on success.
1245  */
1246 static struct pf_state *
1247 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1248 {
1249         struct pf_keyhash       *kh;
1250         struct pf_state_key     *sk;
1251         struct pf_state         *s;
1252         int idx;
1253
1254         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1255
1256         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1257
1258         PF_HASHROW_LOCK(kh);
1259         LIST_FOREACH(sk, &kh->keys, entry)
1260                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1261                         break;
1262         if (sk == NULL) {
1263                 PF_HASHROW_UNLOCK(kh);
1264                 return (NULL);
1265         }
1266
1267         idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1268
1269         /* List is sorted, if-bound states before floating ones. */
1270         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1271                 if (s->kif == V_pfi_all || s->kif == kif) {
1272                         PF_STATE_LOCK(s);
1273                         PF_HASHROW_UNLOCK(kh);
1274                         if (s->timeout >= PFTM_MAX) {
1275                                 /*
1276                                  * State is either being processed by
1277                                  * pf_unlink_state() in an other thread, or
1278                                  * is scheduled for immediate expiry.
1279                                  */
1280                                 PF_STATE_UNLOCK(s);
1281                                 return (NULL);
1282                         }
1283                         return (s);
1284                 }
1285         PF_HASHROW_UNLOCK(kh);
1286
1287         return (NULL);
1288 }
1289
1290 struct pf_state *
1291 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1292 {
1293         struct pf_keyhash       *kh;
1294         struct pf_state_key     *sk;
1295         struct pf_state         *s, *ret = NULL;
1296         int                      idx, inout = 0;
1297
1298         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1299
1300         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1301
1302         PF_HASHROW_LOCK(kh);
1303         LIST_FOREACH(sk, &kh->keys, entry)
1304                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1305                         break;
1306         if (sk == NULL) {
1307                 PF_HASHROW_UNLOCK(kh);
1308                 return (NULL);
1309         }
1310         switch (dir) {
1311         case PF_IN:
1312                 idx = PF_SK_WIRE;
1313                 break;
1314         case PF_OUT:
1315                 idx = PF_SK_STACK;
1316                 break;
1317         case PF_INOUT:
1318                 idx = PF_SK_WIRE;
1319                 inout = 1;
1320                 break;
1321         default:
1322                 panic("%s: dir %u", __func__, dir);
1323         }
1324 second_run:
1325         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1326                 if (more == NULL) {
1327                         PF_HASHROW_UNLOCK(kh);
1328                         return (s);
1329                 }
1330
1331                 if (ret)
1332                         (*more)++;
1333                 else
1334                         ret = s;
1335         }
1336         if (inout == 1) {
1337                 inout = 0;
1338                 idx = PF_SK_STACK;
1339                 goto second_run;
1340         }
1341         PF_HASHROW_UNLOCK(kh);
1342
1343         return (ret);
1344 }
1345
1346 /* END state table stuff */
1347
1348 static void
1349 pf_send(struct pf_send_entry *pfse)
1350 {
1351
1352         PF_SENDQ_LOCK();
1353         STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1354         PF_SENDQ_UNLOCK();
1355         swi_sched(V_pf_swi_cookie, 0);
1356 }
1357
1358 void
1359 pf_intr(void *v)
1360 {
1361         struct pf_send_head queue;
1362         struct pf_send_entry *pfse, *next;
1363
1364         CURVNET_SET((struct vnet *)v);
1365
1366         PF_SENDQ_LOCK();
1367         queue = V_pf_sendqueue;
1368         STAILQ_INIT(&V_pf_sendqueue);
1369         PF_SENDQ_UNLOCK();
1370
1371         STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1372                 switch (pfse->pfse_type) {
1373 #ifdef INET
1374                 case PFSE_IP:
1375                         ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1376                         break;
1377                 case PFSE_ICMP:
1378                         icmp_error(pfse->pfse_m, pfse->pfse_icmp_type,
1379                             pfse->pfse_icmp_code, 0, pfse->pfse_icmp_mtu);
1380                         break;
1381 #endif /* INET */
1382 #ifdef INET6
1383                 case PFSE_IP6:
1384                         ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1385                             NULL);
1386                         break;
1387                 case PFSE_ICMP6:
1388                         icmp6_error(pfse->pfse_m, pfse->pfse_icmp_type,
1389                             pfse->pfse_icmp_code, pfse->pfse_icmp_mtu);
1390                         break;
1391 #endif /* INET6 */
1392                 default:
1393                         panic("%s: unknown type", __func__);
1394                 }
1395                 free(pfse, M_PFTEMP);
1396         }
1397         CURVNET_RESTORE();
1398 }
1399
1400 void
1401 pf_purge_thread(void *v)
1402 {
1403         u_int idx = 0;
1404
1405         CURVNET_SET((struct vnet *)v);
1406
1407         for (;;) {
1408                 PF_RULES_RLOCK();
1409                 rw_sleep(pf_purge_thread, &pf_rules_lock, 0, "pftm", hz / 10);
1410
1411                 if (V_pf_end_threads) {
1412                         /*
1413                          * To cleanse up all kifs and rules we need
1414                          * two runs: first one clears reference flags,
1415                          * then pf_purge_expired_states() doesn't
1416                          * raise them, and then second run frees.
1417                          */
1418                         PF_RULES_RUNLOCK();
1419                         pf_purge_unlinked_rules();
1420                         pfi_kif_purge();
1421
1422                         /*
1423                          * Now purge everything.
1424                          */
1425                         pf_purge_expired_states(0, V_pf_hashmask);
1426                         pf_purge_expired_fragments();
1427                         pf_purge_expired_src_nodes();
1428
1429                         /*
1430                          * Now all kifs & rules should be unreferenced,
1431                          * thus should be successfully freed.
1432                          */
1433                         pf_purge_unlinked_rules();
1434                         pfi_kif_purge();
1435
1436                         /*
1437                          * Announce success and exit.
1438                          */
1439                         PF_RULES_RLOCK();
1440                         V_pf_end_threads++;
1441                         PF_RULES_RUNLOCK();
1442                         wakeup(pf_purge_thread);
1443                         kproc_exit(0);
1444                 }
1445                 PF_RULES_RUNLOCK();
1446
1447                 /* Process 1/interval fraction of the state table every run. */
1448                 idx = pf_purge_expired_states(idx, V_pf_hashmask /
1449                             (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1450
1451                 /* Purge other expired types every PFTM_INTERVAL seconds. */
1452                 if (idx == 0) {
1453                         /*
1454                          * Order is important:
1455                          * - states and src nodes reference rules
1456                          * - states and rules reference kifs
1457                          */
1458                         pf_purge_expired_fragments();
1459                         pf_purge_expired_src_nodes();
1460                         pf_purge_unlinked_rules();
1461                         pfi_kif_purge();
1462                 }
1463         }
1464         /* not reached */
1465         CURVNET_RESTORE();
1466 }
1467
1468 u_int32_t
1469 pf_state_expires(const struct pf_state *state)
1470 {
1471         u_int32_t       timeout;
1472         u_int32_t       start;
1473         u_int32_t       end;
1474         u_int32_t       states;
1475
1476         /* handle all PFTM_* > PFTM_MAX here */
1477         if (state->timeout == PFTM_PURGE)
1478                 return (time_uptime);
1479         KASSERT(state->timeout != PFTM_UNLINKED,
1480             ("pf_state_expires: timeout == PFTM_UNLINKED"));
1481         KASSERT((state->timeout < PFTM_MAX),
1482             ("pf_state_expires: timeout > PFTM_MAX"));
1483         timeout = state->rule.ptr->timeout[state->timeout];
1484         if (!timeout)
1485                 timeout = V_pf_default_rule.timeout[state->timeout];
1486         start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1487         if (start) {
1488                 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1489                 states = counter_u64_fetch(state->rule.ptr->states_cur);
1490         } else {
1491                 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1492                 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1493                 states = V_pf_status.states;
1494         }
1495         if (end && states > start && start < end) {
1496                 if (states < end)
1497                         return (state->expire + timeout * (end - states) /
1498                             (end - start));
1499                 else
1500                         return (time_uptime);
1501         }
1502         return (state->expire + timeout);
1503 }
1504
1505 void
1506 pf_purge_expired_src_nodes()
1507 {
1508         struct pf_src_node_list  freelist;
1509         struct pf_srchash       *sh;
1510         struct pf_src_node      *cur, *next;
1511         int i;
1512
1513         LIST_INIT(&freelist);
1514         for (i = 0, sh = V_pf_srchash; i <= V_pf_srchashmask; i++, sh++) {
1515             PF_HASHROW_LOCK(sh);
1516             LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1517                 if (cur->states == 0 && cur->expire <= time_uptime) {
1518                         pf_unlink_src_node_locked(cur);
1519                         LIST_INSERT_HEAD(&freelist, cur, entry);
1520                 } else if (cur->rule.ptr != NULL)
1521                         cur->rule.ptr->rule_flag |= PFRULE_REFS;
1522             PF_HASHROW_UNLOCK(sh);
1523         }
1524
1525         pf_free_src_nodes(&freelist);
1526
1527         V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1528 }
1529
1530 static void
1531 pf_src_tree_remove_state(struct pf_state *s)
1532 {
1533         u_int32_t timeout;
1534
1535         if (s->src_node != NULL) {
1536                 if (s->src.tcp_est)
1537                         --s->src_node->conn;
1538                 if (--s->src_node->states == 0) {
1539                         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1540                         if (!timeout)
1541                                 timeout =
1542                                     V_pf_default_rule.timeout[PFTM_SRC_NODE];
1543                         s->src_node->expire = time_uptime + timeout;
1544                 }
1545         }
1546         if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1547                 if (--s->nat_src_node->states == 0) {
1548                         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE];
1549                         if (!timeout)
1550                                 timeout =
1551                                     V_pf_default_rule.timeout[PFTM_SRC_NODE];
1552                         s->nat_src_node->expire = time_uptime + timeout;
1553                 }
1554         }
1555         s->src_node = s->nat_src_node = NULL;
1556 }
1557
1558 /*
1559  * Unlink and potentilly free a state. Function may be
1560  * called with ID hash row locked, but always returns
1561  * unlocked, since it needs to go through key hash locking.
1562  */
1563 int
1564 pf_unlink_state(struct pf_state *s, u_int flags)
1565 {
1566         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1567
1568         if ((flags & PF_ENTER_LOCKED) == 0)
1569                 PF_HASHROW_LOCK(ih);
1570         else
1571                 PF_HASHROW_ASSERT(ih);
1572
1573         if (s->timeout == PFTM_UNLINKED) {
1574                 /*
1575                  * State is being processed
1576                  * by pf_unlink_state() in
1577                  * an other thread.
1578                  */
1579                 PF_HASHROW_UNLOCK(ih);
1580                 return (0);     /* XXXGL: undefined actually */
1581         }
1582
1583         if (s->src.state == PF_TCPS_PROXY_DST) {
1584                 /* XXX wire key the right one? */
1585                 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1586                     &s->key[PF_SK_WIRE]->addr[1],
1587                     &s->key[PF_SK_WIRE]->addr[0],
1588                     s->key[PF_SK_WIRE]->port[1],
1589                     s->key[PF_SK_WIRE]->port[0],
1590                     s->src.seqhi, s->src.seqlo + 1,
1591                     TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1592         }
1593
1594         LIST_REMOVE(s, entry);
1595         pf_src_tree_remove_state(s);
1596
1597         if (pfsync_delete_state_ptr != NULL)
1598                 pfsync_delete_state_ptr(s);
1599
1600         STATE_DEC_COUNTERS(s);
1601
1602         s->timeout = PFTM_UNLINKED;
1603
1604         PF_HASHROW_UNLOCK(ih);
1605
1606         pf_detach_state(s);
1607         refcount_release(&s->refs);
1608
1609         return (pf_release_state(s));
1610 }
1611
1612 void
1613 pf_free_state(struct pf_state *cur)
1614 {
1615
1616         KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1617         KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1618             cur->timeout));
1619
1620         pf_normalize_tcp_cleanup(cur);
1621         uma_zfree(V_pf_state_z, cur);
1622         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
1623 }
1624
1625 /*
1626  * Called only from pf_purge_thread(), thus serialized.
1627  */
1628 static u_int
1629 pf_purge_expired_states(u_int i, int maxcheck)
1630 {
1631         struct pf_idhash *ih;
1632         struct pf_state *s;
1633
1634         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1635
1636         /*
1637          * Go through hash and unlink states that expire now.
1638          */
1639         while (maxcheck > 0) {
1640
1641                 ih = &V_pf_idhash[i];
1642 relock:
1643                 PF_HASHROW_LOCK(ih);
1644                 LIST_FOREACH(s, &ih->states, entry) {
1645                         if (pf_state_expires(s) <= time_uptime) {
1646                                 V_pf_status.states -=
1647                                     pf_unlink_state(s, PF_ENTER_LOCKED);
1648                                 goto relock;
1649                         }
1650                         s->rule.ptr->rule_flag |= PFRULE_REFS;
1651                         if (s->nat_rule.ptr != NULL)
1652                                 s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1653                         if (s->anchor.ptr != NULL)
1654                                 s->anchor.ptr->rule_flag |= PFRULE_REFS;
1655                         s->kif->pfik_flags |= PFI_IFLAG_REFS;
1656                         if (s->rt_kif)
1657                                 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1658                 }
1659                 PF_HASHROW_UNLOCK(ih);
1660
1661                 /* Return when we hit end of hash. */
1662                 if (++i > V_pf_hashmask) {
1663                         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1664                         return (0);
1665                 }
1666
1667                 maxcheck--;
1668         }
1669
1670         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1671
1672         return (i);
1673 }
1674
1675 static void
1676 pf_purge_unlinked_rules()
1677 {
1678         struct pf_rulequeue tmpq;
1679         struct pf_rule *r, *r1;
1680
1681         /*
1682          * If we have overloading task pending, then we'd
1683          * better skip purging this time. There is a tiny
1684          * probability that overloading task references
1685          * an already unlinked rule.
1686          */
1687         PF_OVERLOADQ_LOCK();
1688         if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1689                 PF_OVERLOADQ_UNLOCK();
1690                 return;
1691         }
1692         PF_OVERLOADQ_UNLOCK();
1693
1694         /*
1695          * Do naive mark-and-sweep garbage collecting of old rules.
1696          * Reference flag is raised by pf_purge_expired_states()
1697          * and pf_purge_expired_src_nodes().
1698          *
1699          * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1700          * use a temporary queue.
1701          */
1702         TAILQ_INIT(&tmpq);
1703         PF_UNLNKDRULES_LOCK();
1704         TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1705                 if (!(r->rule_flag & PFRULE_REFS)) {
1706                         TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1707                         TAILQ_INSERT_TAIL(&tmpq, r, entries);
1708                 } else
1709                         r->rule_flag &= ~PFRULE_REFS;
1710         }
1711         PF_UNLNKDRULES_UNLOCK();
1712
1713         if (!TAILQ_EMPTY(&tmpq)) {
1714                 PF_RULES_WLOCK();
1715                 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1716                         TAILQ_REMOVE(&tmpq, r, entries);
1717                         pf_free_rule(r);
1718                 }
1719                 PF_RULES_WUNLOCK();
1720         }
1721 }
1722
1723 void
1724 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1725 {
1726         switch (af) {
1727 #ifdef INET
1728         case AF_INET: {
1729                 u_int32_t a = ntohl(addr->addr32[0]);
1730                 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1731                     (a>>8)&255, a&255);
1732                 if (p) {
1733                         p = ntohs(p);
1734                         printf(":%u", p);
1735                 }
1736                 break;
1737         }
1738 #endif /* INET */
1739 #ifdef INET6
1740         case AF_INET6: {
1741                 u_int16_t b;
1742                 u_int8_t i, curstart, curend, maxstart, maxend;
1743                 curstart = curend = maxstart = maxend = 255;
1744                 for (i = 0; i < 8; i++) {
1745                         if (!addr->addr16[i]) {
1746                                 if (curstart == 255)
1747                                         curstart = i;
1748                                 curend = i;
1749                         } else {
1750                                 if ((curend - curstart) >
1751                                     (maxend - maxstart)) {
1752                                         maxstart = curstart;
1753                                         maxend = curend;
1754                                 }
1755                                 curstart = curend = 255;
1756                         }
1757                 }
1758                 if ((curend - curstart) >
1759                     (maxend - maxstart)) {
1760                         maxstart = curstart;
1761                         maxend = curend;
1762                 }
1763                 for (i = 0; i < 8; i++) {
1764                         if (i >= maxstart && i <= maxend) {
1765                                 if (i == 0)
1766                                         printf(":");
1767                                 if (i == maxend)
1768                                         printf(":");
1769                         } else {
1770                                 b = ntohs(addr->addr16[i]);
1771                                 printf("%x", b);
1772                                 if (i < 7)
1773                                         printf(":");
1774                         }
1775                 }
1776                 if (p) {
1777                         p = ntohs(p);
1778                         printf("[%u]", p);
1779                 }
1780                 break;
1781         }
1782 #endif /* INET6 */
1783         }
1784 }
1785
1786 void
1787 pf_print_state(struct pf_state *s)
1788 {
1789         pf_print_state_parts(s, NULL, NULL);
1790 }
1791
1792 static void
1793 pf_print_state_parts(struct pf_state *s,
1794     struct pf_state_key *skwp, struct pf_state_key *sksp)
1795 {
1796         struct pf_state_key *skw, *sks;
1797         u_int8_t proto, dir;
1798
1799         /* Do our best to fill these, but they're skipped if NULL */
1800         skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1801         sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1802         proto = skw ? skw->proto : (sks ? sks->proto : 0);
1803         dir = s ? s->direction : 0;
1804
1805         switch (proto) {
1806         case IPPROTO_IPV4:
1807                 printf("IPv4");
1808                 break;
1809         case IPPROTO_IPV6:
1810                 printf("IPv6");
1811                 break;
1812         case IPPROTO_TCP:
1813                 printf("TCP");
1814                 break;
1815         case IPPROTO_UDP:
1816                 printf("UDP");
1817                 break;
1818         case IPPROTO_ICMP:
1819                 printf("ICMP");
1820                 break;
1821         case IPPROTO_ICMPV6:
1822                 printf("ICMPv6");
1823                 break;
1824         default:
1825                 printf("%u", skw->proto);
1826                 break;
1827         }
1828         switch (dir) {
1829         case PF_IN:
1830                 printf(" in");
1831                 break;
1832         case PF_OUT:
1833                 printf(" out");
1834                 break;
1835         }
1836         if (skw) {
1837                 printf(" wire: ");
1838                 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1839                 printf(" ");
1840                 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1841         }
1842         if (sks) {
1843                 printf(" stack: ");
1844                 if (sks != skw) {
1845                         pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1846                         printf(" ");
1847                         pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1848                 } else
1849                         printf("-");
1850         }
1851         if (s) {
1852                 if (proto == IPPROTO_TCP) {
1853                         printf(" [lo=%u high=%u win=%u modulator=%u",
1854                             s->src.seqlo, s->src.seqhi,
1855                             s->src.max_win, s->src.seqdiff);
1856                         if (s->src.wscale && s->dst.wscale)
1857                                 printf(" wscale=%u",
1858                                     s->src.wscale & PF_WSCALE_MASK);
1859                         printf("]");
1860                         printf(" [lo=%u high=%u win=%u modulator=%u",
1861                             s->dst.seqlo, s->dst.seqhi,
1862                             s->dst.max_win, s->dst.seqdiff);
1863                         if (s->src.wscale && s->dst.wscale)
1864                                 printf(" wscale=%u",
1865                                 s->dst.wscale & PF_WSCALE_MASK);
1866                         printf("]");
1867                 }
1868                 printf(" %u:%u", s->src.state, s->dst.state);
1869         }
1870 }
1871
1872 void
1873 pf_print_flags(u_int8_t f)
1874 {
1875         if (f)
1876                 printf(" ");
1877         if (f & TH_FIN)
1878                 printf("F");
1879         if (f & TH_SYN)
1880                 printf("S");
1881         if (f & TH_RST)
1882                 printf("R");
1883         if (f & TH_PUSH)
1884                 printf("P");
1885         if (f & TH_ACK)
1886                 printf("A");
1887         if (f & TH_URG)
1888                 printf("U");
1889         if (f & TH_ECE)
1890                 printf("E");
1891         if (f & TH_CWR)
1892                 printf("W");
1893 }
1894
1895 #define PF_SET_SKIP_STEPS(i)                                    \
1896         do {                                                    \
1897                 while (head[i] != cur) {                        \
1898                         head[i]->skip[i].ptr = cur;             \
1899                         head[i] = TAILQ_NEXT(head[i], entries); \
1900                 }                                               \
1901         } while (0)
1902
1903 void
1904 pf_calc_skip_steps(struct pf_rulequeue *rules)
1905 {
1906         struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1907         int i;
1908
1909         cur = TAILQ_FIRST(rules);
1910         prev = cur;
1911         for (i = 0; i < PF_SKIP_COUNT; ++i)
1912                 head[i] = cur;
1913         while (cur != NULL) {
1914
1915                 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1916                         PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1917                 if (cur->direction != prev->direction)
1918                         PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1919                 if (cur->af != prev->af)
1920                         PF_SET_SKIP_STEPS(PF_SKIP_AF);
1921                 if (cur->proto != prev->proto)
1922                         PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
1923                 if (cur->src.neg != prev->src.neg ||
1924                     pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
1925                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
1926                 if (cur->src.port[0] != prev->src.port[0] ||
1927                     cur->src.port[1] != prev->src.port[1] ||
1928                     cur->src.port_op != prev->src.port_op)
1929                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
1930                 if (cur->dst.neg != prev->dst.neg ||
1931                     pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
1932                         PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
1933                 if (cur->dst.port[0] != prev->dst.port[0] ||
1934                     cur->dst.port[1] != prev->dst.port[1] ||
1935                     cur->dst.port_op != prev->dst.port_op)
1936                         PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
1937
1938                 prev = cur;
1939                 cur = TAILQ_NEXT(cur, entries);
1940         }
1941         for (i = 0; i < PF_SKIP_COUNT; ++i)
1942                 PF_SET_SKIP_STEPS(i);
1943 }
1944
1945 static int
1946 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
1947 {
1948         if (aw1->type != aw2->type)
1949                 return (1);
1950         switch (aw1->type) {
1951         case PF_ADDR_ADDRMASK:
1952         case PF_ADDR_RANGE:
1953                 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0))
1954                         return (1);
1955                 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0))
1956                         return (1);
1957                 return (0);
1958         case PF_ADDR_DYNIFTL:
1959                 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
1960         case PF_ADDR_NOROUTE:
1961         case PF_ADDR_URPFFAILED:
1962                 return (0);
1963         case PF_ADDR_TABLE:
1964                 return (aw1->p.tbl != aw2->p.tbl);
1965         default:
1966                 printf("invalid address type: %d\n", aw1->type);
1967                 return (1);
1968         }
1969 }
1970
1971 u_int16_t
1972 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
1973 {
1974         u_int32_t       l;
1975
1976         if (udp && !cksum)
1977                 return (0x0000);
1978         l = cksum + old - new;
1979         l = (l >> 16) + (l & 65535);
1980         l = l & 65535;
1981         if (udp && !l)
1982                 return (0xFFFF);
1983         return (l);
1984 }
1985
1986 static void
1987 pf_change_ap(struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc,
1988     struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af)
1989 {
1990         struct pf_addr  ao;
1991         u_int16_t       po = *p;
1992
1993         PF_ACPY(&ao, a, af);
1994         PF_ACPY(a, an, af);
1995
1996         *p = pn;
1997
1998         switch (af) {
1999 #ifdef INET
2000         case AF_INET:
2001                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2002                     ao.addr16[0], an->addr16[0], 0),
2003                     ao.addr16[1], an->addr16[1], 0);
2004                 *p = pn;
2005                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
2006                     ao.addr16[0], an->addr16[0], u),
2007                     ao.addr16[1], an->addr16[1], u),
2008                     po, pn, u);
2009                 break;
2010 #endif /* INET */
2011 #ifdef INET6
2012         case AF_INET6:
2013                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2014                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2015                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(*pc,
2016                     ao.addr16[0], an->addr16[0], u),
2017                     ao.addr16[1], an->addr16[1], u),
2018                     ao.addr16[2], an->addr16[2], u),
2019                     ao.addr16[3], an->addr16[3], u),
2020                     ao.addr16[4], an->addr16[4], u),
2021                     ao.addr16[5], an->addr16[5], u),
2022                     ao.addr16[6], an->addr16[6], u),
2023                     ao.addr16[7], an->addr16[7], u),
2024                     po, pn, u);
2025                 break;
2026 #endif /* INET6 */
2027         }
2028 }
2029
2030
2031 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2032 void
2033 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2034 {
2035         u_int32_t       ao;
2036
2037         memcpy(&ao, a, sizeof(ao));
2038         memcpy(a, &an, sizeof(u_int32_t));
2039         *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2040             ao % 65536, an % 65536, u);
2041 }
2042
2043 #ifdef INET6
2044 static void
2045 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2046 {
2047         struct pf_addr  ao;
2048
2049         PF_ACPY(&ao, a, AF_INET6);
2050         PF_ACPY(a, an, AF_INET6);
2051
2052         *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2053             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2054             pf_cksum_fixup(pf_cksum_fixup(*c,
2055             ao.addr16[0], an->addr16[0], u),
2056             ao.addr16[1], an->addr16[1], u),
2057             ao.addr16[2], an->addr16[2], u),
2058             ao.addr16[3], an->addr16[3], u),
2059             ao.addr16[4], an->addr16[4], u),
2060             ao.addr16[5], an->addr16[5], u),
2061             ao.addr16[6], an->addr16[6], u),
2062             ao.addr16[7], an->addr16[7], u);
2063 }
2064 #endif /* INET6 */
2065
2066 static void
2067 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2068     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2069     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2070 {
2071         struct pf_addr  oia, ooa;
2072
2073         PF_ACPY(&oia, ia, af);
2074         if (oa)
2075                 PF_ACPY(&ooa, oa, af);
2076
2077         /* Change inner protocol port, fix inner protocol checksum. */
2078         if (ip != NULL) {
2079                 u_int16_t       oip = *ip;
2080                 u_int32_t       opc;
2081
2082                 if (pc != NULL)
2083                         opc = *pc;
2084                 *ip = np;
2085                 if (pc != NULL)
2086                         *pc = pf_cksum_fixup(*pc, oip, *ip, u);
2087                 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2088                 if (pc != NULL)
2089                         *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2090         }
2091         /* Change inner ip address, fix inner ip and icmp checksums. */
2092         PF_ACPY(ia, na, af);
2093         switch (af) {
2094 #ifdef INET
2095         case AF_INET: {
2096                 u_int32_t        oh2c = *h2c;
2097
2098                 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2099                     oia.addr16[0], ia->addr16[0], 0),
2100                     oia.addr16[1], ia->addr16[1], 0);
2101                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2102                     oia.addr16[0], ia->addr16[0], 0),
2103                     oia.addr16[1], ia->addr16[1], 0);
2104                 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2105                 break;
2106         }
2107 #endif /* INET */
2108 #ifdef INET6
2109         case AF_INET6:
2110                 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2111                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2112                     pf_cksum_fixup(pf_cksum_fixup(*ic,
2113                     oia.addr16[0], ia->addr16[0], u),
2114                     oia.addr16[1], ia->addr16[1], u),
2115                     oia.addr16[2], ia->addr16[2], u),
2116                     oia.addr16[3], ia->addr16[3], u),
2117                     oia.addr16[4], ia->addr16[4], u),
2118                     oia.addr16[5], ia->addr16[5], u),
2119                     oia.addr16[6], ia->addr16[6], u),
2120                     oia.addr16[7], ia->addr16[7], u);
2121                 break;
2122 #endif /* INET6 */
2123         }
2124         /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2125         if (oa) {
2126                 PF_ACPY(oa, na, af);
2127                 switch (af) {
2128 #ifdef INET
2129                 case AF_INET:
2130                         *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2131                             ooa.addr16[0], oa->addr16[0], 0),
2132                             ooa.addr16[1], oa->addr16[1], 0);
2133                         break;
2134 #endif /* INET */
2135 #ifdef INET6
2136                 case AF_INET6:
2137                         *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2138                             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2139                             pf_cksum_fixup(pf_cksum_fixup(*ic,
2140                             ooa.addr16[0], oa->addr16[0], u),
2141                             ooa.addr16[1], oa->addr16[1], u),
2142                             ooa.addr16[2], oa->addr16[2], u),
2143                             ooa.addr16[3], oa->addr16[3], u),
2144                             ooa.addr16[4], oa->addr16[4], u),
2145                             ooa.addr16[5], oa->addr16[5], u),
2146                             ooa.addr16[6], oa->addr16[6], u),
2147                             ooa.addr16[7], oa->addr16[7], u);
2148                         break;
2149 #endif /* INET6 */
2150                 }
2151         }
2152 }
2153
2154
2155 /*
2156  * Need to modulate the sequence numbers in the TCP SACK option
2157  * (credits to Krzysztof Pfaff for report and patch)
2158  */
2159 static int
2160 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2161     struct tcphdr *th, struct pf_state_peer *dst)
2162 {
2163         int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2164         u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2165         int copyback = 0, i, olen;
2166         struct sackblk sack;
2167
2168 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2169         if (hlen < TCPOLEN_SACKLEN ||
2170             !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2171                 return 0;
2172
2173         while (hlen >= TCPOLEN_SACKLEN) {
2174                 olen = opt[1];
2175                 switch (*opt) {
2176                 case TCPOPT_EOL:        /* FALLTHROUGH */
2177                 case TCPOPT_NOP:
2178                         opt++;
2179                         hlen--;
2180                         break;
2181                 case TCPOPT_SACK:
2182                         if (olen > hlen)
2183                                 olen = hlen;
2184                         if (olen >= TCPOLEN_SACKLEN) {
2185                                 for (i = 2; i + TCPOLEN_SACK <= olen;
2186                                     i += TCPOLEN_SACK) {
2187                                         memcpy(&sack, &opt[i], sizeof(sack));
2188                                         pf_change_a(&sack.start, &th->th_sum,
2189                                             htonl(ntohl(sack.start) -
2190                                             dst->seqdiff), 0);
2191                                         pf_change_a(&sack.end, &th->th_sum,
2192                                             htonl(ntohl(sack.end) -
2193                                             dst->seqdiff), 0);
2194                                         memcpy(&opt[i], &sack, sizeof(sack));
2195                                 }
2196                                 copyback = 1;
2197                         }
2198                         /* FALLTHROUGH */
2199                 default:
2200                         if (olen < 2)
2201                                 olen = 2;
2202                         hlen -= olen;
2203                         opt += olen;
2204                 }
2205         }
2206
2207         if (copyback)
2208                 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2209         return (copyback);
2210 }
2211
2212 static void
2213 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2214     const struct pf_addr *saddr, const struct pf_addr *daddr,
2215     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2216     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2217     u_int16_t rtag, struct ifnet *ifp)
2218 {
2219         struct pf_send_entry *pfse;
2220         struct mbuf     *m;
2221         int              len, tlen;
2222 #ifdef INET
2223         struct ip       *h = NULL;
2224 #endif /* INET */
2225 #ifdef INET6
2226         struct ip6_hdr  *h6 = NULL;
2227 #endif /* INET6 */
2228         struct tcphdr   *th;
2229         char            *opt;
2230         struct pf_mtag  *pf_mtag;
2231
2232         len = 0;
2233         th = NULL;
2234
2235         /* maximum segment size tcp option */
2236         tlen = sizeof(struct tcphdr);
2237         if (mss)
2238                 tlen += 4;
2239
2240         switch (af) {
2241 #ifdef INET
2242         case AF_INET:
2243                 len = sizeof(struct ip) + tlen;
2244                 break;
2245 #endif /* INET */
2246 #ifdef INET6
2247         case AF_INET6:
2248                 len = sizeof(struct ip6_hdr) + tlen;
2249                 break;
2250 #endif /* INET6 */
2251         default:
2252                 panic("%s: unsupported af %d", __func__, af);
2253         }
2254
2255         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2256         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2257         if (pfse == NULL)
2258                 return;
2259         m = m_gethdr(M_NOWAIT, MT_DATA);
2260         if (m == NULL) {
2261                 free(pfse, M_PFTEMP);
2262                 return;
2263         }
2264 #ifdef MAC
2265         mac_netinet_firewall_send(m);
2266 #endif
2267         if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2268                 free(pfse, M_PFTEMP);
2269                 m_freem(m);
2270                 return;
2271         }
2272         if (tag)
2273                 m->m_flags |= M_SKIP_FIREWALL;
2274         pf_mtag->tag = rtag;
2275
2276         if (r != NULL && r->rtableid >= 0)
2277                 M_SETFIB(m, r->rtableid);
2278
2279 #ifdef ALTQ
2280         if (r != NULL && r->qid) {
2281                 pf_mtag->qid = r->qid;
2282
2283                 /* add hints for ecn */
2284                 pf_mtag->hdr = mtod(m, struct ip *);
2285         }
2286 #endif /* ALTQ */
2287         m->m_data += max_linkhdr;
2288         m->m_pkthdr.len = m->m_len = len;
2289         m->m_pkthdr.rcvif = NULL;
2290         bzero(m->m_data, len);
2291         switch (af) {
2292 #ifdef INET
2293         case AF_INET:
2294                 h = mtod(m, struct ip *);
2295
2296                 /* IP header fields included in the TCP checksum */
2297                 h->ip_p = IPPROTO_TCP;
2298                 h->ip_len = htons(tlen);
2299                 h->ip_src.s_addr = saddr->v4.s_addr;
2300                 h->ip_dst.s_addr = daddr->v4.s_addr;
2301
2302                 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2303                 break;
2304 #endif /* INET */
2305 #ifdef INET6
2306         case AF_INET6:
2307                 h6 = mtod(m, struct ip6_hdr *);
2308
2309                 /* IP header fields included in the TCP checksum */
2310                 h6->ip6_nxt = IPPROTO_TCP;
2311                 h6->ip6_plen = htons(tlen);
2312                 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2313                 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2314
2315                 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2316                 break;
2317 #endif /* INET6 */
2318         }
2319
2320         /* TCP header */
2321         th->th_sport = sport;
2322         th->th_dport = dport;
2323         th->th_seq = htonl(seq);
2324         th->th_ack = htonl(ack);
2325         th->th_off = tlen >> 2;
2326         th->th_flags = flags;
2327         th->th_win = htons(win);
2328
2329         if (mss) {
2330                 opt = (char *)(th + 1);
2331                 opt[0] = TCPOPT_MAXSEG;
2332                 opt[1] = 4;
2333                 HTONS(mss);
2334                 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2335         }
2336
2337         switch (af) {
2338 #ifdef INET
2339         case AF_INET:
2340                 /* TCP checksum */
2341                 th->th_sum = in_cksum(m, len);
2342
2343                 /* Finish the IP header */
2344                 h->ip_v = 4;
2345                 h->ip_hl = sizeof(*h) >> 2;
2346                 h->ip_tos = IPTOS_LOWDELAY;
2347                 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2348                 h->ip_len = htons(len);
2349                 h->ip_ttl = ttl ? ttl : V_ip_defttl;
2350                 h->ip_sum = 0;
2351
2352                 pfse->pfse_type = PFSE_IP;
2353                 break;
2354 #endif /* INET */
2355 #ifdef INET6
2356         case AF_INET6:
2357                 /* TCP checksum */
2358                 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2359                     sizeof(struct ip6_hdr), tlen);
2360
2361                 h6->ip6_vfc |= IPV6_VERSION;
2362                 h6->ip6_hlim = IPV6_DEFHLIM;
2363
2364                 pfse->pfse_type = PFSE_IP6;
2365                 break;
2366 #endif /* INET6 */
2367         }
2368         pfse->pfse_m = m;
2369         pf_send(pfse);
2370 }
2371
2372 static void
2373 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2374     struct pf_rule *r)
2375 {
2376         struct pf_send_entry *pfse;
2377         struct mbuf *m0;
2378         struct pf_mtag *pf_mtag;
2379
2380         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2381         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2382         if (pfse == NULL)
2383                 return;
2384
2385         if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2386                 free(pfse, M_PFTEMP);
2387                 return;
2388         }
2389
2390         if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2391                 free(pfse, M_PFTEMP);
2392                 return;
2393         }
2394         /* XXX: revisit */
2395         m0->m_flags |= M_SKIP_FIREWALL;
2396
2397         if (r->rtableid >= 0)
2398                 M_SETFIB(m0, r->rtableid);
2399
2400 #ifdef ALTQ
2401         if (r->qid) {
2402                 pf_mtag->qid = r->qid;
2403                 /* add hints for ecn */
2404                 pf_mtag->hdr = mtod(m0, struct ip *);
2405         }
2406 #endif /* ALTQ */
2407
2408         switch (af) {
2409 #ifdef INET
2410         case AF_INET:
2411                 pfse->pfse_type = PFSE_ICMP;
2412                 break;
2413 #endif /* INET */
2414 #ifdef INET6
2415         case AF_INET6:
2416                 pfse->pfse_type = PFSE_ICMP6;
2417                 break;
2418 #endif /* INET6 */
2419         }
2420         pfse->pfse_m = m0;
2421         pfse->pfse_icmp_type = type;
2422         pfse->pfse_icmp_code = code;
2423         pf_send(pfse);
2424 }
2425
2426 /*
2427  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2428  * If n is 0, they match if they are equal. If n is != 0, they match if they
2429  * are different.
2430  */
2431 int
2432 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2433     struct pf_addr *b, sa_family_t af)
2434 {
2435         int     match = 0;
2436
2437         switch (af) {
2438 #ifdef INET
2439         case AF_INET:
2440                 if ((a->addr32[0] & m->addr32[0]) ==
2441                     (b->addr32[0] & m->addr32[0]))
2442                         match++;
2443                 break;
2444 #endif /* INET */
2445 #ifdef INET6
2446         case AF_INET6:
2447                 if (((a->addr32[0] & m->addr32[0]) ==
2448                      (b->addr32[0] & m->addr32[0])) &&
2449                     ((a->addr32[1] & m->addr32[1]) ==
2450                      (b->addr32[1] & m->addr32[1])) &&
2451                     ((a->addr32[2] & m->addr32[2]) ==
2452                      (b->addr32[2] & m->addr32[2])) &&
2453                     ((a->addr32[3] & m->addr32[3]) ==
2454                      (b->addr32[3] & m->addr32[3])))
2455                         match++;
2456                 break;
2457 #endif /* INET6 */
2458         }
2459         if (match) {
2460                 if (n)
2461                         return (0);
2462                 else
2463                         return (1);
2464         } else {
2465                 if (n)
2466                         return (1);
2467                 else
2468                         return (0);
2469         }
2470 }
2471
2472 /*
2473  * Return 1 if b <= a <= e, otherwise return 0.
2474  */
2475 int
2476 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2477     struct pf_addr *a, sa_family_t af)
2478 {
2479         switch (af) {
2480 #ifdef INET
2481         case AF_INET:
2482                 if ((a->addr32[0] < b->addr32[0]) ||
2483                     (a->addr32[0] > e->addr32[0]))
2484                         return (0);
2485                 break;
2486 #endif /* INET */
2487 #ifdef INET6
2488         case AF_INET6: {
2489                 int     i;
2490
2491                 /* check a >= b */
2492                 for (i = 0; i < 4; ++i)
2493                         if (a->addr32[i] > b->addr32[i])
2494                                 break;
2495                         else if (a->addr32[i] < b->addr32[i])
2496                                 return (0);
2497                 /* check a <= e */
2498                 for (i = 0; i < 4; ++i)
2499                         if (a->addr32[i] < e->addr32[i])
2500                                 break;
2501                         else if (a->addr32[i] > e->addr32[i])
2502                                 return (0);
2503                 break;
2504         }
2505 #endif /* INET6 */
2506         }
2507         return (1);
2508 }
2509
2510 static int
2511 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2512 {
2513         switch (op) {
2514         case PF_OP_IRG:
2515                 return ((p > a1) && (p < a2));
2516         case PF_OP_XRG:
2517                 return ((p < a1) || (p > a2));
2518         case PF_OP_RRG:
2519                 return ((p >= a1) && (p <= a2));
2520         case PF_OP_EQ:
2521                 return (p == a1);
2522         case PF_OP_NE:
2523                 return (p != a1);
2524         case PF_OP_LT:
2525                 return (p < a1);
2526         case PF_OP_LE:
2527                 return (p <= a1);
2528         case PF_OP_GT:
2529                 return (p > a1);
2530         case PF_OP_GE:
2531                 return (p >= a1);
2532         }
2533         return (0); /* never reached */
2534 }
2535
2536 int
2537 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2538 {
2539         NTOHS(a1);
2540         NTOHS(a2);
2541         NTOHS(p);
2542         return (pf_match(op, a1, a2, p));
2543 }
2544
2545 static int
2546 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2547 {
2548         if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2549                 return (0);
2550         return (pf_match(op, a1, a2, u));
2551 }
2552
2553 static int
2554 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2555 {
2556         if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2557                 return (0);
2558         return (pf_match(op, a1, a2, g));
2559 }
2560
2561 int
2562 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2563 {
2564         if (*tag == -1)
2565                 *tag = mtag;
2566
2567         return ((!r->match_tag_not && r->match_tag == *tag) ||
2568             (r->match_tag_not && r->match_tag != *tag));
2569 }
2570
2571 int
2572 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2573 {
2574
2575         KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2576
2577         if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2578                 return (ENOMEM);
2579
2580         pd->pf_mtag->tag = tag;
2581
2582         return (0);
2583 }
2584
2585 #define PF_ANCHOR_STACKSIZE     32
2586 struct pf_anchor_stackframe {
2587         struct pf_ruleset       *rs;
2588         struct pf_rule          *r;     /* XXX: + match bit */
2589         struct pf_anchor        *child;
2590 };
2591
2592 /*
2593  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2594  */
2595 #define PF_ANCHORSTACK_MATCH    0x00000001
2596 #define PF_ANCHORSTACK_MASK     (PF_ANCHORSTACK_MATCH)
2597
2598 #define PF_ANCHOR_MATCH(f)      ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2599 #define PF_ANCHOR_RULE(f)       (struct pf_rule *)                      \
2600                                 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2601 #define PF_ANCHOR_SET_MATCH(f)  do { (f)->r = (void *)                  \
2602                                 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2603 } while (0)
2604
2605 void
2606 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2607     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2608     int *match)
2609 {
2610         struct pf_anchor_stackframe     *f;
2611
2612         PF_RULES_RASSERT();
2613
2614         if (match)
2615                 *match = 0;
2616         if (*depth >= PF_ANCHOR_STACKSIZE) {
2617                 printf("%s: anchor stack overflow on %s\n",
2618                     __func__, (*r)->anchor->name);
2619                 *r = TAILQ_NEXT(*r, entries);
2620                 return;
2621         } else if (*depth == 0 && a != NULL)
2622                 *a = *r;
2623         f = stack + (*depth)++;
2624         f->rs = *rs;
2625         f->r = *r;
2626         if ((*r)->anchor_wildcard) {
2627                 struct pf_anchor_node *parent = &(*r)->anchor->children;
2628
2629                 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2630                         *r = NULL;
2631                         return;
2632                 }
2633                 *rs = &f->child->ruleset;
2634         } else {
2635                 f->child = NULL;
2636                 *rs = &(*r)->anchor->ruleset;
2637         }
2638         *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2639 }
2640
2641 int
2642 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2643     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2644     int *match)
2645 {
2646         struct pf_anchor_stackframe     *f;
2647         struct pf_rule *fr;
2648         int quick = 0;
2649
2650         PF_RULES_RASSERT();
2651
2652         do {
2653                 if (*depth <= 0)
2654                         break;
2655                 f = stack + *depth - 1;
2656                 fr = PF_ANCHOR_RULE(f);
2657                 if (f->child != NULL) {
2658                         struct pf_anchor_node *parent;
2659
2660                         /*
2661                          * This block traverses through
2662                          * a wildcard anchor.
2663                          */
2664                         parent = &fr->anchor->children;
2665                         if (match != NULL && *match) {
2666                                 /*
2667                                  * If any of "*" matched, then
2668                                  * "foo/ *" matched, mark frame
2669                                  * appropriately.
2670                                  */
2671                                 PF_ANCHOR_SET_MATCH(f);
2672                                 *match = 0;
2673                         }
2674                         f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2675                         if (f->child != NULL) {
2676                                 *rs = &f->child->ruleset;
2677                                 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2678                                 if (*r == NULL)
2679                                         continue;
2680                                 else
2681                                         break;
2682                         }
2683                 }
2684                 (*depth)--;
2685                 if (*depth == 0 && a != NULL)
2686                         *a = NULL;
2687                 *rs = f->rs;
2688                 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2689                         quick = fr->quick;
2690                 *r = TAILQ_NEXT(fr, entries);
2691         } while (*r == NULL);
2692
2693         return (quick);
2694 }
2695
2696 #ifdef INET6
2697 void
2698 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2699     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2700 {
2701         switch (af) {
2702 #ifdef INET
2703         case AF_INET:
2704                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2705                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2706                 break;
2707 #endif /* INET */
2708         case AF_INET6:
2709                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2710                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2711                 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2712                 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2713                 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2714                 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2715                 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2716                 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2717                 break;
2718         }
2719 }
2720
2721 void
2722 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2723 {
2724         switch (af) {
2725 #ifdef INET
2726         case AF_INET:
2727                 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2728                 break;
2729 #endif /* INET */
2730         case AF_INET6:
2731                 if (addr->addr32[3] == 0xffffffff) {
2732                         addr->addr32[3] = 0;
2733                         if (addr->addr32[2] == 0xffffffff) {
2734                                 addr->addr32[2] = 0;
2735                                 if (addr->addr32[1] == 0xffffffff) {
2736                                         addr->addr32[1] = 0;
2737                                         addr->addr32[0] =
2738                                             htonl(ntohl(addr->addr32[0]) + 1);
2739                                 } else
2740                                         addr->addr32[1] =
2741                                             htonl(ntohl(addr->addr32[1]) + 1);
2742                         } else
2743                                 addr->addr32[2] =
2744                                     htonl(ntohl(addr->addr32[2]) + 1);
2745                 } else
2746                         addr->addr32[3] =
2747                             htonl(ntohl(addr->addr32[3]) + 1);
2748                 break;
2749         }
2750 }
2751 #endif /* INET6 */
2752
2753 int
2754 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
2755 {
2756         struct pf_addr          *saddr, *daddr;
2757         u_int16_t                sport, dport;
2758         struct inpcbinfo        *pi;
2759         struct inpcb            *inp;
2760
2761         pd->lookup.uid = UID_MAX;
2762         pd->lookup.gid = GID_MAX;
2763
2764         switch (pd->proto) {
2765         case IPPROTO_TCP:
2766                 if (pd->hdr.tcp == NULL)
2767                         return (-1);
2768                 sport = pd->hdr.tcp->th_sport;
2769                 dport = pd->hdr.tcp->th_dport;
2770                 pi = &V_tcbinfo;
2771                 break;
2772         case IPPROTO_UDP:
2773                 if (pd->hdr.udp == NULL)
2774                         return (-1);
2775                 sport = pd->hdr.udp->uh_sport;
2776                 dport = pd->hdr.udp->uh_dport;
2777                 pi = &V_udbinfo;
2778                 break;
2779         default:
2780                 return (-1);
2781         }
2782         if (direction == PF_IN) {
2783                 saddr = pd->src;
2784                 daddr = pd->dst;
2785         } else {
2786                 u_int16_t       p;
2787
2788                 p = sport;
2789                 sport = dport;
2790                 dport = p;
2791                 saddr = pd->dst;
2792                 daddr = pd->src;
2793         }
2794         switch (pd->af) {
2795 #ifdef INET
2796         case AF_INET:
2797                 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
2798                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2799                 if (inp == NULL) {
2800                         inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
2801                            daddr->v4, dport, INPLOOKUP_WILDCARD |
2802                            INPLOOKUP_RLOCKPCB, NULL, m);
2803                         if (inp == NULL)
2804                                 return (-1);
2805                 }
2806                 break;
2807 #endif /* INET */
2808 #ifdef INET6
2809         case AF_INET6:
2810                 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
2811                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2812                 if (inp == NULL) {
2813                         inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
2814                             &daddr->v6, dport, INPLOOKUP_WILDCARD |
2815                             INPLOOKUP_RLOCKPCB, NULL, m);
2816                         if (inp == NULL)
2817                                 return (-1);
2818                 }
2819                 break;
2820 #endif /* INET6 */
2821
2822         default:
2823                 return (-1);
2824         }
2825         INP_RLOCK_ASSERT(inp);
2826         pd->lookup.uid = inp->inp_cred->cr_uid;
2827         pd->lookup.gid = inp->inp_cred->cr_groups[0];
2828         INP_RUNLOCK(inp);
2829
2830         return (1);
2831 }
2832
2833 static u_int8_t
2834 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2835 {
2836         int              hlen;
2837         u_int8_t         hdr[60];
2838         u_int8_t        *opt, optlen;
2839         u_int8_t         wscale = 0;
2840
2841         hlen = th_off << 2;             /* hlen <= sizeof(hdr) */
2842         if (hlen <= sizeof(struct tcphdr))
2843                 return (0);
2844         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2845                 return (0);
2846         opt = hdr + sizeof(struct tcphdr);
2847         hlen -= sizeof(struct tcphdr);
2848         while (hlen >= 3) {
2849                 switch (*opt) {
2850                 case TCPOPT_EOL:
2851                 case TCPOPT_NOP:
2852                         ++opt;
2853                         --hlen;
2854                         break;
2855                 case TCPOPT_WINDOW:
2856                         wscale = opt[2];
2857                         if (wscale > TCP_MAX_WINSHIFT)
2858                                 wscale = TCP_MAX_WINSHIFT;
2859                         wscale |= PF_WSCALE_FLAG;
2860                         /* FALLTHROUGH */
2861                 default:
2862                         optlen = opt[1];
2863                         if (optlen < 2)
2864                                 optlen = 2;
2865                         hlen -= optlen;
2866                         opt += optlen;
2867                         break;
2868                 }
2869         }
2870         return (wscale);
2871 }
2872
2873 static u_int16_t
2874 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2875 {
2876         int              hlen;
2877         u_int8_t         hdr[60];
2878         u_int8_t        *opt, optlen;
2879         u_int16_t        mss = V_tcp_mssdflt;
2880
2881         hlen = th_off << 2;     /* hlen <= sizeof(hdr) */
2882         if (hlen <= sizeof(struct tcphdr))
2883                 return (0);
2884         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2885                 return (0);
2886         opt = hdr + sizeof(struct tcphdr);
2887         hlen -= sizeof(struct tcphdr);
2888         while (hlen >= TCPOLEN_MAXSEG) {
2889                 switch (*opt) {
2890                 case TCPOPT_EOL:
2891                 case TCPOPT_NOP:
2892                         ++opt;
2893                         --hlen;
2894                         break;
2895                 case TCPOPT_MAXSEG:
2896                         bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
2897                         NTOHS(mss);
2898                         /* FALLTHROUGH */
2899                 default:
2900                         optlen = opt[1];
2901                         if (optlen < 2)
2902                                 optlen = 2;
2903                         hlen -= optlen;
2904                         opt += optlen;
2905                         break;
2906                 }
2907         }
2908         return (mss);
2909 }
2910
2911 static u_int16_t
2912 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
2913 {
2914 #ifdef INET
2915         struct sockaddr_in      *dst;
2916         struct route             ro;
2917 #endif /* INET */
2918 #ifdef INET6
2919         struct sockaddr_in6     *dst6;
2920         struct route_in6         ro6;
2921 #endif /* INET6 */
2922         struct rtentry          *rt = NULL;
2923         int                      hlen = 0;
2924         u_int16_t                mss = V_tcp_mssdflt;
2925
2926         switch (af) {
2927 #ifdef INET
2928         case AF_INET:
2929                 hlen = sizeof(struct ip);
2930                 bzero(&ro, sizeof(ro));
2931                 dst = (struct sockaddr_in *)&ro.ro_dst;
2932                 dst->sin_family = AF_INET;
2933                 dst->sin_len = sizeof(*dst);
2934                 dst->sin_addr = addr->v4;
2935                 in_rtalloc_ign(&ro, 0, rtableid);
2936                 rt = ro.ro_rt;
2937                 break;
2938 #endif /* INET */
2939 #ifdef INET6
2940         case AF_INET6:
2941                 hlen = sizeof(struct ip6_hdr);
2942                 bzero(&ro6, sizeof(ro6));
2943                 dst6 = (struct sockaddr_in6 *)&ro6.ro_dst;
2944                 dst6->sin6_family = AF_INET6;
2945                 dst6->sin6_len = sizeof(*dst6);
2946                 dst6->sin6_addr = addr->v6;
2947                 in6_rtalloc_ign(&ro6, 0, rtableid);
2948                 rt = ro6.ro_rt;
2949                 break;
2950 #endif /* INET6 */
2951         }
2952
2953         if (rt && rt->rt_ifp) {
2954                 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr);
2955                 mss = max(V_tcp_mssdflt, mss);
2956                 RTFREE(rt);
2957         }
2958         mss = min(mss, offer);
2959         mss = max(mss, 64);             /* sanity - at least max opt space */
2960         return (mss);
2961 }
2962
2963 static void
2964 pf_set_rt_ifp(struct pf_state *s, struct pf_addr *saddr)
2965 {
2966         struct pf_rule *r = s->rule.ptr;
2967         struct pf_src_node *sn = NULL;
2968
2969         s->rt_kif = NULL;
2970         if (!r->rt || r->rt == PF_FASTROUTE)
2971                 return;
2972         switch (s->key[PF_SK_WIRE]->af) {
2973 #ifdef INET
2974         case AF_INET:
2975                 pf_map_addr(AF_INET, r, saddr, &s->rt_addr, NULL, &sn);
2976                 s->rt_kif = r->rpool.cur->kif;
2977                 break;
2978 #endif /* INET */
2979 #ifdef INET6
2980         case AF_INET6:
2981                 pf_map_addr(AF_INET6, r, saddr, &s->rt_addr, NULL, &sn);
2982                 s->rt_kif = r->rpool.cur->kif;
2983                 break;
2984 #endif /* INET6 */
2985         }
2986 }
2987
2988 static u_int32_t
2989 pf_tcp_iss(struct pf_pdesc *pd)
2990 {
2991         MD5_CTX ctx;
2992         u_int32_t digest[4];
2993
2994         if (V_pf_tcp_secret_init == 0) {
2995                 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
2996                 MD5Init(&V_pf_tcp_secret_ctx);
2997                 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
2998                     sizeof(V_pf_tcp_secret));
2999                 V_pf_tcp_secret_init = 1;
3000         }
3001
3002         ctx = V_pf_tcp_secret_ctx;
3003
3004         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
3005         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
3006         if (pd->af == AF_INET6) {
3007                 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3008                 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3009         } else {
3010                 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3011                 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3012         }
3013         MD5Final((u_char *)digest, &ctx);
3014         V_pf_tcp_iss_off += 4096;
3015 #define ISN_RANDOM_INCREMENT (4096 - 1)
3016         return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
3017             V_pf_tcp_iss_off);
3018 #undef  ISN_RANDOM_INCREMENT
3019 }
3020
3021 static int
3022 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
3023     struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
3024     struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
3025 {
3026         struct pf_rule          *nr = NULL;
3027         struct pf_addr          * const saddr = pd->src;
3028         struct pf_addr          * const daddr = pd->dst;
3029         sa_family_t              af = pd->af;
3030         struct pf_rule          *r, *a = NULL;
3031         struct pf_ruleset       *ruleset = NULL;
3032         struct pf_src_node      *nsn = NULL;
3033         struct tcphdr           *th = pd->hdr.tcp;
3034         struct pf_state_key     *sk = NULL, *nk = NULL;
3035         u_short                  reason;
3036         int                      rewrite = 0, hdrlen = 0;
3037         int                      tag = -1, rtableid = -1;
3038         int                      asd = 0;
3039         int                      match = 0;
3040         int                      state_icmp = 0;
3041         u_int16_t                sport = 0, dport = 0;
3042         u_int16_t                bproto_sum = 0, bip_sum = 0;
3043         u_int8_t                 icmptype = 0, icmpcode = 0;
3044         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3045
3046         PF_RULES_RASSERT();
3047
3048         if (inp != NULL) {
3049                 INP_LOCK_ASSERT(inp);
3050                 pd->lookup.uid = inp->inp_cred->cr_uid;
3051                 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3052                 pd->lookup.done = 1;
3053         }
3054
3055         switch (pd->proto) {
3056         case IPPROTO_TCP:
3057                 sport = th->th_sport;
3058                 dport = th->th_dport;
3059                 hdrlen = sizeof(*th);
3060                 break;
3061         case IPPROTO_UDP:
3062                 sport = pd->hdr.udp->uh_sport;
3063                 dport = pd->hdr.udp->uh_dport;
3064                 hdrlen = sizeof(*pd->hdr.udp);
3065                 break;
3066 #ifdef INET
3067         case IPPROTO_ICMP:
3068                 if (pd->af != AF_INET)
3069                         break;
3070                 sport = dport = pd->hdr.icmp->icmp_id;
3071                 hdrlen = sizeof(*pd->hdr.icmp);
3072                 icmptype = pd->hdr.icmp->icmp_type;
3073                 icmpcode = pd->hdr.icmp->icmp_code;
3074
3075                 if (icmptype == ICMP_UNREACH ||
3076                     icmptype == ICMP_SOURCEQUENCH ||
3077                     icmptype == ICMP_REDIRECT ||
3078                     icmptype == ICMP_TIMXCEED ||
3079                     icmptype == ICMP_PARAMPROB)
3080                         state_icmp++;
3081                 break;
3082 #endif /* INET */
3083 #ifdef INET6
3084         case IPPROTO_ICMPV6:
3085                 if (af != AF_INET6)
3086                         break;
3087                 sport = dport = pd->hdr.icmp6->icmp6_id;
3088                 hdrlen = sizeof(*pd->hdr.icmp6);
3089                 icmptype = pd->hdr.icmp6->icmp6_type;
3090                 icmpcode = pd->hdr.icmp6->icmp6_code;
3091
3092                 if (icmptype == ICMP6_DST_UNREACH ||
3093                     icmptype == ICMP6_PACKET_TOO_BIG ||
3094                     icmptype == ICMP6_TIME_EXCEEDED ||
3095                     icmptype == ICMP6_PARAM_PROB)
3096                         state_icmp++;
3097                 break;
3098 #endif /* INET6 */
3099         default:
3100                 sport = dport = hdrlen = 0;
3101                 break;
3102         }
3103
3104         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3105
3106         /* check packet for BINAT/NAT/RDR */
3107         if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3108             &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3109                 KASSERT(sk != NULL, ("%s: null sk", __func__));
3110                 KASSERT(nk != NULL, ("%s: null nk", __func__));
3111
3112                 if (pd->ip_sum)
3113                         bip_sum = *pd->ip_sum;
3114
3115                 switch (pd->proto) {
3116                 case IPPROTO_TCP:
3117                         bproto_sum = th->th_sum;
3118                         pd->proto_sum = &th->th_sum;
3119
3120                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3121                             nk->port[pd->sidx] != sport) {
3122                                 pf_change_ap(saddr, &th->th_sport, pd->ip_sum,
3123                                     &th->th_sum, &nk->addr[pd->sidx],
3124                                     nk->port[pd->sidx], 0, af);
3125                                 pd->sport = &th->th_sport;
3126                                 sport = th->th_sport;
3127                         }
3128
3129                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3130                             nk->port[pd->didx] != dport) {
3131                                 pf_change_ap(daddr, &th->th_dport, pd->ip_sum,
3132                                     &th->th_sum, &nk->addr[pd->didx],
3133                                     nk->port[pd->didx], 0, af);
3134                                 dport = th->th_dport;
3135                                 pd->dport = &th->th_dport;
3136                         }
3137                         rewrite++;
3138                         break;
3139                 case IPPROTO_UDP:
3140                         bproto_sum = pd->hdr.udp->uh_sum;
3141                         pd->proto_sum = &pd->hdr.udp->uh_sum;
3142
3143                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3144                             nk->port[pd->sidx] != sport) {
3145                                 pf_change_ap(saddr, &pd->hdr.udp->uh_sport,
3146                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3147                                     &nk->addr[pd->sidx],
3148                                     nk->port[pd->sidx], 1, af);
3149                                 sport = pd->hdr.udp->uh_sport;
3150                                 pd->sport = &pd->hdr.udp->uh_sport;
3151                         }
3152
3153                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3154                             nk->port[pd->didx] != dport) {
3155                                 pf_change_ap(daddr, &pd->hdr.udp->uh_dport,
3156                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3157                                     &nk->addr[pd->didx],
3158                                     nk->port[pd->didx], 1, af);
3159                                 dport = pd->hdr.udp->uh_dport;
3160                                 pd->dport = &pd->hdr.udp->uh_dport;
3161                         }
3162                         rewrite++;
3163                         break;
3164 #ifdef INET
3165                 case IPPROTO_ICMP:
3166                         nk->port[0] = nk->port[1];
3167                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3168                                 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3169                                     nk->addr[pd->sidx].v4.s_addr, 0);
3170
3171                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3172                                 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3173                                     nk->addr[pd->didx].v4.s_addr, 0);
3174
3175                         if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3176                                 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3177                                     pd->hdr.icmp->icmp_cksum, sport,
3178                                     nk->port[1], 0);
3179                                 pd->hdr.icmp->icmp_id = nk->port[1];
3180                                 pd->sport = &pd->hdr.icmp->icmp_id;
3181                         }
3182                         m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3183                         break;
3184 #endif /* INET */
3185 #ifdef INET6
3186                 case IPPROTO_ICMPV6:
3187                         nk->port[0] = nk->port[1];
3188                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3189                                 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3190                                     &nk->addr[pd->sidx], 0);
3191
3192                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3193                                 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3194                                     &nk->addr[pd->didx], 0);
3195                         rewrite++;
3196                         break;
3197 #endif /* INET */
3198                 default:
3199                         switch (af) {
3200 #ifdef INET
3201                         case AF_INET:
3202                                 if (PF_ANEQ(saddr,
3203                                     &nk->addr[pd->sidx], AF_INET))
3204                                         pf_change_a(&saddr->v4.s_addr,
3205                                             pd->ip_sum,
3206                                             nk->addr[pd->sidx].v4.s_addr, 0);
3207
3208                                 if (PF_ANEQ(daddr,
3209                                     &nk->addr[pd->didx], AF_INET))
3210                                         pf_change_a(&daddr->v4.s_addr,
3211                                             pd->ip_sum,
3212                                             nk->addr[pd->didx].v4.s_addr, 0);
3213                                 break;
3214 #endif /* INET */
3215 #ifdef INET6
3216                         case AF_INET6:
3217                                 if (PF_ANEQ(saddr,
3218                                     &nk->addr[pd->sidx], AF_INET6))
3219                                         PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3220
3221                                 if (PF_ANEQ(daddr,
3222                                     &nk->addr[pd->didx], AF_INET6))
3223                                         PF_ACPY(saddr, &nk->addr[pd->didx], af);
3224                                 break;
3225 #endif /* INET */
3226                         }
3227                         break;
3228                 }
3229                 if (nr->natpass)
3230                         r = NULL;
3231                 pd->nat_rule = nr;
3232         }
3233
3234         while (r != NULL) {
3235                 r->evaluations++;
3236                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3237                         r = r->skip[PF_SKIP_IFP].ptr;
3238                 else if (r->direction && r->direction != direction)
3239                         r = r->skip[PF_SKIP_DIR].ptr;
3240                 else if (r->af && r->af != af)
3241                         r = r->skip[PF_SKIP_AF].ptr;
3242                 else if (r->proto && r->proto != pd->proto)
3243                         r = r->skip[PF_SKIP_PROTO].ptr;
3244                 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3245                     r->src.neg, kif, M_GETFIB(m)))
3246                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3247                 /* tcp/udp only. port_op always 0 in other cases */
3248                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3249                     r->src.port[0], r->src.port[1], sport))
3250                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
3251                 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3252                     r->dst.neg, NULL, M_GETFIB(m)))
3253                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3254                 /* tcp/udp only. port_op always 0 in other cases */
3255                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3256                     r->dst.port[0], r->dst.port[1], dport))
3257                         r = r->skip[PF_SKIP_DST_PORT].ptr;
3258                 /* icmp only. type always 0 in other cases */
3259                 else if (r->type && r->type != icmptype + 1)
3260                         r = TAILQ_NEXT(r, entries);
3261                 /* icmp only. type always 0 in other cases */
3262                 else if (r->code && r->code != icmpcode + 1)
3263                         r = TAILQ_NEXT(r, entries);
3264                 else if (r->tos && !(r->tos == pd->tos))
3265                         r = TAILQ_NEXT(r, entries);
3266                 else if (r->rule_flag & PFRULE_FRAGMENT)
3267                         r = TAILQ_NEXT(r, entries);
3268                 else if (pd->proto == IPPROTO_TCP &&
3269                     (r->flagset & th->th_flags) != r->flags)
3270                         r = TAILQ_NEXT(r, entries);
3271                 /* tcp/udp only. uid.op always 0 in other cases */
3272                 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3273                     pf_socket_lookup(direction, pd, m), 1)) &&
3274                     !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3275                     pd->lookup.uid))
3276                         r = TAILQ_NEXT(r, entries);
3277                 /* tcp/udp only. gid.op always 0 in other cases */
3278                 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3279                     pf_socket_lookup(direction, pd, m), 1)) &&
3280                     !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3281                     pd->lookup.gid))
3282                         r = TAILQ_NEXT(r, entries);
3283                 else if (r->prob &&
3284                     r->prob <= arc4random())
3285                         r = TAILQ_NEXT(r, entries);
3286                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3287                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3288                         r = TAILQ_NEXT(r, entries);
3289                 else if (r->os_fingerprint != PF_OSFP_ANY &&
3290                     (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3291                     pf_osfp_fingerprint(pd, m, off, th),
3292                     r->os_fingerprint)))
3293                         r = TAILQ_NEXT(r, entries);
3294                 else {
3295                         if (r->tag)
3296                                 tag = r->tag;
3297                         if (r->rtableid >= 0)
3298                                 rtableid = r->rtableid;
3299                         if (r->anchor == NULL) {
3300                                 match = 1;
3301                                 *rm = r;
3302                                 *am = a;
3303                                 *rsm = ruleset;
3304                                 if ((*rm)->quick)
3305                                         break;
3306                                 r = TAILQ_NEXT(r, entries);
3307                         } else
3308                                 pf_step_into_anchor(anchor_stack, &asd,
3309                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3310                                     &match);
3311                 }
3312                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3313                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3314                         break;
3315         }
3316         r = *rm;
3317         a = *am;
3318         ruleset = *rsm;
3319
3320         REASON_SET(&reason, PFRES_MATCH);
3321
3322         if (r->log || (nr != NULL && nr->log)) {
3323                 if (rewrite)
3324                         m_copyback(m, off, hdrlen, pd->hdr.any);
3325                 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3326                     ruleset, pd, 1);
3327         }
3328
3329         if ((r->action == PF_DROP) &&
3330             ((r->rule_flag & PFRULE_RETURNRST) ||
3331             (r->rule_flag & PFRULE_RETURNICMP) ||
3332             (r->rule_flag & PFRULE_RETURN))) {
3333                 /* undo NAT changes, if they have taken place */
3334                 if (nr != NULL) {
3335                         PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3336                         PF_ACPY(daddr, &sk->addr[pd->didx], af);
3337                         if (pd->sport)
3338                                 *pd->sport = sk->port[pd->sidx];
3339                         if (pd->dport)
3340                                 *pd->dport = sk->port[pd->didx];
3341                         if (pd->proto_sum)
3342                                 *pd->proto_sum = bproto_sum;
3343                         if (pd->ip_sum)
3344                                 *pd->ip_sum = bip_sum;
3345                         m_copyback(m, off, hdrlen, pd->hdr.any);
3346                 }
3347                 if (pd->proto == IPPROTO_TCP &&
3348                     ((r->rule_flag & PFRULE_RETURNRST) ||
3349                     (r->rule_flag & PFRULE_RETURN)) &&
3350                     !(th->th_flags & TH_RST)) {
3351                         u_int32_t        ack = ntohl(th->th_seq) + pd->p_len;
3352                         int              len = 0;
3353 #ifdef INET
3354                         struct ip       *h4;
3355 #endif
3356 #ifdef INET6
3357                         struct ip6_hdr  *h6;
3358 #endif
3359
3360                         switch (af) {
3361 #ifdef INET
3362                         case AF_INET:
3363                                 h4 = mtod(m, struct ip *);
3364                                 len = ntohs(h4->ip_len) - off;
3365                                 break;
3366 #endif
3367 #ifdef INET6
3368                         case AF_INET6:
3369                                 h6 = mtod(m, struct ip6_hdr *);
3370                                 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3371                                 break;
3372 #endif
3373                         }
3374
3375                         if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3376                                 REASON_SET(&reason, PFRES_PROTCKSUM);
3377                         else {
3378                                 if (th->th_flags & TH_SYN)
3379                                         ack++;
3380                                 if (th->th_flags & TH_FIN)
3381                                         ack++;
3382                                 pf_send_tcp(m, r, af, pd->dst,
3383                                     pd->src, th->th_dport, th->th_sport,
3384                                     ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3385                                     r->return_ttl, 1, 0, kif->pfik_ifp);
3386                         }
3387                 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3388                     r->return_icmp)
3389                         pf_send_icmp(m, r->return_icmp >> 8,
3390                             r->return_icmp & 255, af, r);
3391                 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3392                     r->return_icmp6)
3393                         pf_send_icmp(m, r->return_icmp6 >> 8,
3394                             r->return_icmp6 & 255, af, r);
3395         }
3396
3397         if (r->action == PF_DROP)
3398                 goto cleanup;
3399
3400         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3401                 REASON_SET(&reason, PFRES_MEMORY);
3402                 goto cleanup;
3403         }
3404         if (rtableid >= 0)
3405                 M_SETFIB(m, rtableid);
3406
3407         if (!state_icmp && (r->keep_state || nr != NULL ||
3408             (pd->flags & PFDESC_TCP_NORM))) {
3409                 int action;
3410                 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3411                     sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3412                     hdrlen);
3413                 if (action != PF_PASS)
3414                         return (action);
3415         } else {
3416                 if (sk != NULL)
3417                         uma_zfree(V_pf_state_key_z, sk);
3418                 if (nk != NULL)
3419                         uma_zfree(V_pf_state_key_z, nk);
3420         }
3421
3422         /* copy back packet headers if we performed NAT operations */
3423         if (rewrite)
3424                 m_copyback(m, off, hdrlen, pd->hdr.any);
3425
3426         if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3427             direction == PF_OUT &&
3428             pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m))
3429                 /*
3430                  * We want the state created, but we dont
3431                  * want to send this in case a partner
3432                  * firewall has to know about it to allow
3433                  * replies through it.
3434                  */
3435                 return (PF_DEFER);
3436
3437         return (PF_PASS);
3438
3439 cleanup:
3440         if (sk != NULL)
3441                 uma_zfree(V_pf_state_key_z, sk);
3442         if (nk != NULL)
3443                 uma_zfree(V_pf_state_key_z, nk);
3444         return (PF_DROP);
3445 }
3446
3447 static int
3448 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3449     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3450     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3451     u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3452     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3453 {
3454         struct pf_state         *s = NULL;
3455         struct pf_src_node      *sn = NULL;
3456         struct tcphdr           *th = pd->hdr.tcp;
3457         u_int16_t                mss = V_tcp_mssdflt;
3458         u_short                  reason;
3459
3460         /* check maximums */
3461         if (r->max_states &&
3462             (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3463                 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
3464                 REASON_SET(&reason, PFRES_MAXSTATES);
3465                 return (PF_DROP);
3466         }
3467         /* src node for filter rule */
3468         if ((r->rule_flag & PFRULE_SRCTRACK ||
3469             r->rpool.opts & PF_POOL_STICKYADDR) &&
3470             pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3471                 REASON_SET(&reason, PFRES_SRCLIMIT);
3472                 goto csfailed;
3473         }
3474         /* src node for translation rule */
3475         if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3476             pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3477                 REASON_SET(&reason, PFRES_SRCLIMIT);
3478                 goto csfailed;
3479         }
3480         s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3481         if (s == NULL) {
3482                 REASON_SET(&reason, PFRES_MEMORY);
3483                 goto csfailed;
3484         }
3485         s->rule.ptr = r;
3486         s->nat_rule.ptr = nr;
3487         s->anchor.ptr = a;
3488         STATE_INC_COUNTERS(s);
3489         if (r->allow_opts)
3490                 s->state_flags |= PFSTATE_ALLOWOPTS;
3491         if (r->rule_flag & PFRULE_STATESLOPPY)
3492                 s->state_flags |= PFSTATE_SLOPPY;
3493         s->log = r->log & PF_LOG_ALL;
3494         s->sync_state = PFSYNC_S_NONE;
3495         if (nr != NULL)
3496                 s->log |= nr->log & PF_LOG_ALL;
3497         switch (pd->proto) {
3498         case IPPROTO_TCP:
3499                 s->src.seqlo = ntohl(th->th_seq);
3500                 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3501                 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3502                     r->keep_state == PF_STATE_MODULATE) {
3503                         /* Generate sequence number modulator */
3504                         if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3505                             0)
3506                                 s->src.seqdiff = 1;
3507                         pf_change_a(&th->th_seq, &th->th_sum,
3508                             htonl(s->src.seqlo + s->src.seqdiff), 0);
3509                         *rewrite = 1;
3510                 } else
3511                         s->src.seqdiff = 0;
3512                 if (th->th_flags & TH_SYN) {
3513                         s->src.seqhi++;
3514                         s->src.wscale = pf_get_wscale(m, off,
3515                             th->th_off, pd->af);
3516                 }
3517                 s->src.max_win = MAX(ntohs(th->th_win), 1);
3518                 if (s->src.wscale & PF_WSCALE_MASK) {
3519                         /* Remove scale factor from initial window */
3520                         int win = s->src.max_win;
3521                         win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3522                         s->src.max_win = (win - 1) >>
3523                             (s->src.wscale & PF_WSCALE_MASK);
3524                 }
3525                 if (th->th_flags & TH_FIN)
3526                         s->src.seqhi++;
3527                 s->dst.seqhi = 1;
3528                 s->dst.max_win = 1;
3529                 s->src.state = TCPS_SYN_SENT;
3530                 s->dst.state = TCPS_CLOSED;
3531                 s->timeout = PFTM_TCP_FIRST_PACKET;
3532                 break;
3533         case IPPROTO_UDP:
3534                 s->src.state = PFUDPS_SINGLE;
3535                 s->dst.state = PFUDPS_NO_TRAFFIC;
3536                 s->timeout = PFTM_UDP_FIRST_PACKET;
3537                 break;
3538         case IPPROTO_ICMP:
3539 #ifdef INET6
3540         case IPPROTO_ICMPV6:
3541 #endif
3542                 s->timeout = PFTM_ICMP_FIRST_PACKET;
3543                 break;
3544         default:
3545                 s->src.state = PFOTHERS_SINGLE;
3546                 s->dst.state = PFOTHERS_NO_TRAFFIC;
3547                 s->timeout = PFTM_OTHER_FIRST_PACKET;
3548         }
3549
3550         s->creation = time_uptime;
3551         s->expire = time_uptime;
3552
3553         if (sn != NULL) {
3554                 s->src_node = sn;
3555                 s->src_node->states++;
3556         }
3557         if (nsn != NULL) {
3558                 /* XXX We only modify one side for now. */
3559                 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3560                 s->nat_src_node = nsn;
3561                 s->nat_src_node->states++;
3562         }
3563         if (pd->proto == IPPROTO_TCP) {
3564                 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3565                     off, pd, th, &s->src, &s->dst)) {
3566                         REASON_SET(&reason, PFRES_MEMORY);
3567                         pf_src_tree_remove_state(s);
3568                         STATE_DEC_COUNTERS(s);
3569                         uma_zfree(V_pf_state_z, s);
3570                         return (PF_DROP);
3571                 }
3572                 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3573                     pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3574                     &s->src, &s->dst, rewrite)) {
3575                         /* This really shouldn't happen!!! */
3576                         DPFPRINTF(PF_DEBUG_URGENT,
3577                             ("pf_normalize_tcp_stateful failed on first pkt"));
3578                         pf_normalize_tcp_cleanup(s);
3579                         pf_src_tree_remove_state(s);
3580                         STATE_DEC_COUNTERS(s);
3581                         uma_zfree(V_pf_state_z, s);
3582                         return (PF_DROP);
3583                 }
3584         }
3585         s->direction = pd->dir;
3586
3587         /*
3588          * sk/nk could already been setup by pf_get_translation().
3589          */
3590         if (nr == NULL) {
3591                 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3592                     __func__, nr, sk, nk));
3593                 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3594                 if (sk == NULL)
3595                         goto csfailed;
3596                 nk = sk;
3597         } else
3598                 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3599                     __func__, nr, sk, nk));
3600
3601         /* Swap sk/nk for PF_OUT. */
3602         if (pf_state_insert(BOUND_IFACE(r, kif),
3603             (pd->dir == PF_IN) ? sk : nk,
3604             (pd->dir == PF_IN) ? nk : sk, s)) {
3605                 if (pd->proto == IPPROTO_TCP)
3606                         pf_normalize_tcp_cleanup(s);
3607                 REASON_SET(&reason, PFRES_STATEINS);
3608                 pf_src_tree_remove_state(s);
3609                 STATE_DEC_COUNTERS(s);
3610                 uma_zfree(V_pf_state_z, s);
3611                 return (PF_DROP);
3612         } else
3613                 *sm = s;
3614
3615         pf_set_rt_ifp(s, pd->src);      /* needs s->state_key set */
3616         if (tag > 0)
3617                 s->tag = tag;
3618         if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3619             TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3620                 s->src.state = PF_TCPS_PROXY_SRC;
3621                 /* undo NAT changes, if they have taken place */
3622                 if (nr != NULL) {
3623                         struct pf_state_key *skt = s->key[PF_SK_WIRE];
3624                         if (pd->dir == PF_OUT)
3625                                 skt = s->key[PF_SK_STACK];
3626                         PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3627                         PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3628                         if (pd->sport)
3629                                 *pd->sport = skt->port[pd->sidx];
3630                         if (pd->dport)
3631                                 *pd->dport = skt->port[pd->didx];
3632                         if (pd->proto_sum)
3633                                 *pd->proto_sum = bproto_sum;
3634                         if (pd->ip_sum)
3635                                 *pd->ip_sum = bip_sum;
3636                         m_copyback(m, off, hdrlen, pd->hdr.any);
3637                 }
3638                 s->src.seqhi = htonl(arc4random());
3639                 /* Find mss option */
3640                 int rtid = M_GETFIB(m);
3641                 mss = pf_get_mss(m, off, th->th_off, pd->af);
3642                 mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3643                 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3644                 s->src.mss = mss;
3645                 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3646                     th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3647                     TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3648                 REASON_SET(&reason, PFRES_SYNPROXY);
3649                 return (PF_SYNPROXY_DROP);
3650         }
3651
3652         return (PF_PASS);
3653
3654 csfailed:
3655         if (sk != NULL)
3656                 uma_zfree(V_pf_state_key_z, sk);
3657         if (nk != NULL)
3658                 uma_zfree(V_pf_state_key_z, nk);
3659
3660         if (sn != NULL && sn->states == 0 && sn->expire == 0) {
3661                 pf_unlink_src_node(sn);
3662                 pf_free_src_node(sn);
3663         }
3664
3665         if (nsn != sn && nsn != NULL && nsn->states == 0 && nsn->expire == 0) {
3666                 pf_unlink_src_node(nsn);
3667                 pf_free_src_node(nsn);
3668         }
3669
3670         return (PF_DROP);
3671 }
3672
3673 static int
3674 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
3675     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
3676     struct pf_ruleset **rsm)
3677 {
3678         struct pf_rule          *r, *a = NULL;
3679         struct pf_ruleset       *ruleset = NULL;
3680         sa_family_t              af = pd->af;
3681         u_short                  reason;
3682         int                      tag = -1;
3683         int                      asd = 0;
3684         int                      match = 0;
3685         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3686
3687         PF_RULES_RASSERT();
3688
3689         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3690         while (r != NULL) {
3691                 r->evaluations++;
3692                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3693                         r = r->skip[PF_SKIP_IFP].ptr;
3694                 else if (r->direction && r->direction != direction)
3695                         r = r->skip[PF_SKIP_DIR].ptr;
3696                 else if (r->af && r->af != af)
3697                         r = r->skip[PF_SKIP_AF].ptr;
3698                 else if (r->proto && r->proto != pd->proto)
3699                         r = r->skip[PF_SKIP_PROTO].ptr;
3700                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3701                     r->src.neg, kif, M_GETFIB(m)))
3702                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3703                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3704                     r->dst.neg, NULL, M_GETFIB(m)))
3705                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3706                 else if (r->tos && !(r->tos == pd->tos))
3707                         r = TAILQ_NEXT(r, entries);
3708                 else if (r->os_fingerprint != PF_OSFP_ANY)
3709                         r = TAILQ_NEXT(r, entries);
3710                 else if (pd->proto == IPPROTO_UDP &&
3711                     (r->src.port_op || r->dst.port_op))
3712                         r = TAILQ_NEXT(r, entries);
3713                 else if (pd->proto == IPPROTO_TCP &&
3714                     (r->src.port_op || r->dst.port_op || r->flagset))
3715                         r = TAILQ_NEXT(r, entries);
3716                 else if ((pd->proto == IPPROTO_ICMP ||
3717                     pd->proto == IPPROTO_ICMPV6) &&
3718                     (r->type || r->code))
3719                         r = TAILQ_NEXT(r, entries);
3720                 else if (r->prob && r->prob <=
3721                     (arc4random() % (UINT_MAX - 1) + 1))
3722                         r = TAILQ_NEXT(r, entries);
3723                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3724                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3725                         r = TAILQ_NEXT(r, entries);
3726                 else {
3727                         if (r->anchor == NULL) {
3728                                 match = 1;
3729                                 *rm = r;
3730                                 *am = a;
3731                                 *rsm = ruleset;
3732                                 if ((*rm)->quick)
3733                                         break;
3734                                 r = TAILQ_NEXT(r, entries);
3735                         } else
3736                                 pf_step_into_anchor(anchor_stack, &asd,
3737                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3738                                     &match);
3739                 }
3740                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3741                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3742                         break;
3743         }
3744         r = *rm;
3745         a = *am;
3746         ruleset = *rsm;
3747
3748         REASON_SET(&reason, PFRES_MATCH);
3749
3750         if (r->log)
3751                 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
3752                     1);
3753
3754         if (r->action != PF_PASS)
3755                 return (PF_DROP);
3756
3757         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3758                 REASON_SET(&reason, PFRES_MEMORY);
3759                 return (PF_DROP);
3760         }
3761
3762         return (PF_PASS);
3763 }
3764
3765 static int
3766 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
3767         struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
3768         struct pf_pdesc *pd, u_short *reason, int *copyback)
3769 {
3770         struct tcphdr           *th = pd->hdr.tcp;
3771         u_int16_t                win = ntohs(th->th_win);
3772         u_int32_t                ack, end, seq, orig_seq;
3773         u_int8_t                 sws, dws;
3774         int                      ackskew;
3775
3776         if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
3777                 sws = src->wscale & PF_WSCALE_MASK;
3778                 dws = dst->wscale & PF_WSCALE_MASK;
3779         } else
3780                 sws = dws = 0;
3781
3782         /*
3783          * Sequence tracking algorithm from Guido van Rooij's paper:
3784          *   http://www.madison-gurkha.com/publications/tcp_filtering/
3785          *      tcp_filtering.ps
3786          */
3787
3788         orig_seq = seq = ntohl(th->th_seq);
3789         if (src->seqlo == 0) {
3790                 /* First packet from this end. Set its state */
3791
3792                 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
3793                     src->scrub == NULL) {
3794                         if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
3795                                 REASON_SET(reason, PFRES_MEMORY);
3796                                 return (PF_DROP);
3797                         }
3798                 }
3799
3800                 /* Deferred generation of sequence number modulator */
3801                 if (dst->seqdiff && !src->seqdiff) {
3802                         /* use random iss for the TCP server */
3803                         while ((src->seqdiff = arc4random() - seq) == 0)
3804                                 ;
3805                         ack = ntohl(th->th_ack) - dst->seqdiff;
3806                         pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
3807                             src->seqdiff), 0);
3808                         pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
3809                         *copyback = 1;
3810                 } else {
3811                         ack = ntohl(th->th_ack);
3812                 }
3813
3814                 end = seq + pd->p_len;
3815                 if (th->th_flags & TH_SYN) {
3816                         end++;
3817                         if (dst->wscale & PF_WSCALE_FLAG) {
3818                                 src->wscale = pf_get_wscale(m, off, th->th_off,
3819                                     pd->af);
3820                                 if (src->wscale & PF_WSCALE_FLAG) {
3821                                         /* Remove scale factor from initial
3822                                          * window */
3823                                         sws = src->wscale & PF_WSCALE_MASK;
3824                                         win = ((u_int32_t)win + (1 << sws) - 1)
3825                                             >> sws;
3826                                         dws = dst->wscale & PF_WSCALE_MASK;
3827                                 } else {
3828                                         /* fixup other window */
3829                                         dst->max_win <<= dst->wscale &
3830                                             PF_WSCALE_MASK;
3831                                         /* in case of a retrans SYN|ACK */
3832                                         dst->wscale = 0;
3833                                 }
3834                         }
3835                 }
3836                 if (th->th_flags & TH_FIN)
3837                         end++;
3838
3839                 src->seqlo = seq;
3840                 if (src->state < TCPS_SYN_SENT)
3841                         src->state = TCPS_SYN_SENT;
3842
3843                 /*
3844                  * May need to slide the window (seqhi may have been set by
3845                  * the crappy stack check or if we picked up the connection
3846                  * after establishment)
3847                  */
3848                 if (src->seqhi == 1 ||
3849                     SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
3850                         src->seqhi = end + MAX(1, dst->max_win << dws);
3851                 if (win > src->max_win)
3852                         src->max_win = win;
3853
3854         } else {
3855                 ack = ntohl(th->th_ack) - dst->seqdiff;
3856                 if (src->seqdiff) {
3857                         /* Modulate sequence numbers */
3858                         pf_change_a(&th->th_seq, &th->th_sum, htonl(seq +
3859                             src->seqdiff), 0);
3860                         pf_change_a(&th->th_ack, &th->th_sum, htonl(ack), 0);
3861                         *copyback = 1;
3862                 }
3863                 end = seq + pd->p_len;
3864                 if (th->th_flags & TH_SYN)
3865                         end++;
3866                 if (th->th_flags & TH_FIN)
3867                         end++;
3868         }
3869
3870         if ((th->th_flags & TH_ACK) == 0) {
3871                 /* Let it pass through the ack skew check */
3872                 ack = dst->seqlo;
3873         } else if ((ack == 0 &&
3874             (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
3875             /* broken tcp stacks do not set ack */
3876             (dst->state < TCPS_SYN_SENT)) {
3877                 /*
3878                  * Many stacks (ours included) will set the ACK number in an
3879                  * FIN|ACK if the SYN times out -- no sequence to ACK.
3880                  */
3881                 ack = dst->seqlo;
3882         }
3883
3884         if (seq == end) {
3885                 /* Ease sequencing restrictions on no data packets */
3886                 seq = src->seqlo;
3887                 end = seq;
3888         }
3889
3890         ackskew = dst->seqlo - ack;
3891
3892
3893         /*
3894          * Need to demodulate the sequence numbers in any TCP SACK options
3895          * (Selective ACK). We could optionally validate the SACK values
3896          * against the current ACK window, either forwards or backwards, but
3897          * I'm not confident that SACK has been implemented properly
3898          * everywhere. It wouldn't surprise me if several stacks accidently
3899          * SACK too far backwards of previously ACKed data. There really aren't
3900          * any security implications of bad SACKing unless the target stack
3901          * doesn't validate the option length correctly. Someone trying to
3902          * spoof into a TCP connection won't bother blindly sending SACK
3903          * options anyway.
3904          */
3905         if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
3906                 if (pf_modulate_sack(m, off, pd, th, dst))
3907                         *copyback = 1;
3908         }
3909
3910
3911 #define MAXACKWINDOW (0xffff + 1500)    /* 1500 is an arbitrary fudge factor */
3912         if (SEQ_GEQ(src->seqhi, end) &&
3913             /* Last octet inside other's window space */
3914             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
3915             /* Retrans: not more than one window back */
3916             (ackskew >= -MAXACKWINDOW) &&
3917             /* Acking not more than one reassembled fragment backwards */
3918             (ackskew <= (MAXACKWINDOW << sws)) &&
3919             /* Acking not more than one window forward */
3920             ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
3921             (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
3922             (pd->flags & PFDESC_IP_REAS) == 0)) {
3923             /* Require an exact/+1 sequence match on resets when possible */
3924
3925                 if (dst->scrub || src->scrub) {
3926                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
3927                             *state, src, dst, copyback))
3928                                 return (PF_DROP);
3929                 }
3930
3931                 /* update max window */
3932                 if (src->max_win < win)
3933                         src->max_win = win;
3934                 /* synchronize sequencing */
3935                 if (SEQ_GT(end, src->seqlo))
3936                         src->seqlo = end;
3937                 /* slide the window of what the other end can send */
3938                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
3939                         dst->seqhi = ack + MAX((win << sws), 1);
3940
3941
3942                 /* update states */
3943                 if (th->th_flags & TH_SYN)
3944                         if (src->state < TCPS_SYN_SENT)
3945                                 src->state = TCPS_SYN_SENT;
3946                 if (th->th_flags & TH_FIN)
3947                         if (src->state < TCPS_CLOSING)
3948                                 src->state = TCPS_CLOSING;
3949                 if (th->th_flags & TH_ACK) {
3950                         if (dst->state == TCPS_SYN_SENT) {
3951                                 dst->state = TCPS_ESTABLISHED;
3952                                 if (src->state == TCPS_ESTABLISHED &&
3953                                     (*state)->src_node != NULL &&
3954                                     pf_src_connlimit(state)) {
3955                                         REASON_SET(reason, PFRES_SRCLIMIT);
3956                                         return (PF_DROP);
3957                                 }
3958                         } else if (dst->state == TCPS_CLOSING)
3959                                 dst->state = TCPS_FIN_WAIT_2;
3960                 }
3961                 if (th->th_flags & TH_RST)
3962                         src->state = dst->state = TCPS_TIME_WAIT;
3963
3964                 /* update expire time */
3965                 (*state)->expire = time_uptime;
3966                 if (src->state >= TCPS_FIN_WAIT_2 &&
3967                     dst->state >= TCPS_FIN_WAIT_2)
3968                         (*state)->timeout = PFTM_TCP_CLOSED;
3969                 else if (src->state >= TCPS_CLOSING &&
3970                     dst->state >= TCPS_CLOSING)
3971                         (*state)->timeout = PFTM_TCP_FIN_WAIT;
3972                 else if (src->state < TCPS_ESTABLISHED ||
3973                     dst->state < TCPS_ESTABLISHED)
3974                         (*state)->timeout = PFTM_TCP_OPENING;
3975                 else if (src->state >= TCPS_CLOSING ||
3976                     dst->state >= TCPS_CLOSING)
3977                         (*state)->timeout = PFTM_TCP_CLOSING;
3978                 else
3979                         (*state)->timeout = PFTM_TCP_ESTABLISHED;
3980
3981                 /* Fall through to PASS packet */
3982
3983         } else if ((dst->state < TCPS_SYN_SENT ||
3984                 dst->state >= TCPS_FIN_WAIT_2 ||
3985                 src->state >= TCPS_FIN_WAIT_2) &&
3986             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
3987             /* Within a window forward of the originating packet */
3988             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
3989             /* Within a window backward of the originating packet */
3990
3991                 /*
3992                  * This currently handles three situations:
3993                  *  1) Stupid stacks will shotgun SYNs before their peer
3994                  *     replies.
3995                  *  2) When PF catches an already established stream (the
3996                  *     firewall rebooted, the state table was flushed, routes
3997                  *     changed...)
3998                  *  3) Packets get funky immediately after the connection
3999                  *     closes (this should catch Solaris spurious ACK|FINs
4000                  *     that web servers like to spew after a close)
4001                  *
4002                  * This must be a little more careful than the above code
4003                  * since packet floods will also be caught here. We don't
4004                  * update the TTL here to mitigate the damage of a packet
4005                  * flood and so the same code can handle awkward establishment
4006                  * and a loosened connection close.
4007                  * In the establishment case, a correct peer response will
4008                  * validate the connection, go through the normal state code
4009                  * and keep updating the state TTL.
4010                  */
4011
4012                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4013                         printf("pf: loose state match: ");
4014                         pf_print_state(*state);
4015                         pf_print_flags(th->th_flags);
4016                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4017                             "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
4018                             pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
4019                             (unsigned long long)(*state)->packets[1],
4020                             pd->dir == PF_IN ? "in" : "out",
4021                             pd->dir == (*state)->direction ? "fwd" : "rev");
4022                 }
4023
4024                 if (dst->scrub || src->scrub) {
4025                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4026                             *state, src, dst, copyback))
4027                                 return (PF_DROP);
4028                 }
4029
4030                 /* update max window */
4031                 if (src->max_win < win)
4032                         src->max_win = win;
4033                 /* synchronize sequencing */
4034                 if (SEQ_GT(end, src->seqlo))
4035                         src->seqlo = end;
4036                 /* slide the window of what the other end can send */
4037                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4038                         dst->seqhi = ack + MAX((win << sws), 1);
4039
4040                 /*
4041                  * Cannot set dst->seqhi here since this could be a shotgunned
4042                  * SYN and not an already established connection.
4043                  */
4044
4045                 if (th->th_flags & TH_FIN)
4046                         if (src->state < TCPS_CLOSING)
4047                                 src->state = TCPS_CLOSING;
4048                 if (th->th_flags & TH_RST)
4049                         src->state = dst->state = TCPS_TIME_WAIT;
4050
4051                 /* Fall through to PASS packet */
4052
4053         } else {
4054                 if ((*state)->dst.state == TCPS_SYN_SENT &&
4055                     (*state)->src.state == TCPS_SYN_SENT) {
4056                         /* Send RST for state mismatches during handshake */
4057                         if (!(th->th_flags & TH_RST))
4058                                 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4059                                     pd->dst, pd->src, th->th_dport,
4060                                     th->th_sport, ntohl(th->th_ack), 0,
4061                                     TH_RST, 0, 0,
4062                                     (*state)->rule.ptr->return_ttl, 1, 0,
4063                                     kif->pfik_ifp);
4064                         src->seqlo = 0;
4065                         src->seqhi = 1;
4066                         src->max_win = 1;
4067                 } else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4068                         printf("pf: BAD state: ");
4069                         pf_print_state(*state);
4070                         pf_print_flags(th->th_flags);
4071                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4072                             "pkts=%llu:%llu dir=%s,%s\n",
4073                             seq, orig_seq, ack, pd->p_len, ackskew,
4074                             (unsigned long long)(*state)->packets[0],
4075                             (unsigned long long)(*state)->packets[1],
4076                             pd->dir == PF_IN ? "in" : "out",
4077                             pd->dir == (*state)->direction ? "fwd" : "rev");
4078                         printf("pf: State failure on: %c %c %c %c | %c %c\n",
4079                             SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4080                             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4081                             ' ': '2',
4082                             (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4083                             (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4084                             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4085                             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4086                 }
4087                 REASON_SET(reason, PFRES_BADSTATE);
4088                 return (PF_DROP);
4089         }
4090
4091         return (PF_PASS);
4092 }
4093
4094 static int
4095 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4096         struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4097 {
4098         struct tcphdr           *th = pd->hdr.tcp;
4099
4100         if (th->th_flags & TH_SYN)
4101                 if (src->state < TCPS_SYN_SENT)
4102                         src->state = TCPS_SYN_SENT;
4103         if (th->th_flags & TH_FIN)
4104                 if (src->state < TCPS_CLOSING)
4105                         src->state = TCPS_CLOSING;
4106         if (th->th_flags & TH_ACK) {
4107                 if (dst->state == TCPS_SYN_SENT) {
4108                         dst->state = TCPS_ESTABLISHED;
4109                         if (src->state == TCPS_ESTABLISHED &&
4110                             (*state)->src_node != NULL &&
4111                             pf_src_connlimit(state)) {
4112                                 REASON_SET(reason, PFRES_SRCLIMIT);
4113                                 return (PF_DROP);
4114                         }
4115                 } else if (dst->state == TCPS_CLOSING) {
4116                         dst->state = TCPS_FIN_WAIT_2;
4117                 } else if (src->state == TCPS_SYN_SENT &&
4118                     dst->state < TCPS_SYN_SENT) {
4119                         /*
4120                          * Handle a special sloppy case where we only see one
4121                          * half of the connection. If there is a ACK after
4122                          * the initial SYN without ever seeing a packet from
4123                          * the destination, set the connection to established.
4124                          */
4125                         dst->state = src->state = TCPS_ESTABLISHED;
4126                         if ((*state)->src_node != NULL &&
4127                             pf_src_connlimit(state)) {
4128                                 REASON_SET(reason, PFRES_SRCLIMIT);
4129                                 return (PF_DROP);
4130                         }
4131                 } else if (src->state == TCPS_CLOSING &&
4132                     dst->state == TCPS_ESTABLISHED &&
4133                     dst->seqlo == 0) {
4134                         /*
4135                          * Handle the closing of half connections where we
4136                          * don't see the full bidirectional FIN/ACK+ACK
4137                          * handshake.
4138                          */
4139                         dst->state = TCPS_CLOSING;
4140                 }
4141         }
4142         if (th->th_flags & TH_RST)
4143                 src->state = dst->state = TCPS_TIME_WAIT;
4144
4145         /* update expire time */
4146         (*state)->expire = time_uptime;
4147         if (src->state >= TCPS_FIN_WAIT_2 &&
4148             dst->state >= TCPS_FIN_WAIT_2)
4149                 (*state)->timeout = PFTM_TCP_CLOSED;
4150         else if (src->state >= TCPS_CLOSING &&
4151             dst->state >= TCPS_CLOSING)
4152                 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4153         else if (src->state < TCPS_ESTABLISHED ||
4154             dst->state < TCPS_ESTABLISHED)
4155                 (*state)->timeout = PFTM_TCP_OPENING;
4156         else if (src->state >= TCPS_CLOSING ||
4157             dst->state >= TCPS_CLOSING)
4158                 (*state)->timeout = PFTM_TCP_CLOSING;
4159         else
4160                 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4161
4162         return (PF_PASS);
4163 }
4164
4165 static int
4166 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4167     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4168     u_short *reason)
4169 {
4170         struct pf_state_key_cmp  key;
4171         struct tcphdr           *th = pd->hdr.tcp;
4172         int                      copyback = 0;
4173         struct pf_state_peer    *src, *dst;
4174         struct pf_state_key     *sk;
4175
4176         bzero(&key, sizeof(key));
4177         key.af = pd->af;
4178         key.proto = IPPROTO_TCP;
4179         if (direction == PF_IN) {       /* wire side, straight */
4180                 PF_ACPY(&key.addr[0], pd->src, key.af);
4181                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4182                 key.port[0] = th->th_sport;
4183                 key.port[1] = th->th_dport;
4184         } else {                        /* stack side, reverse */
4185                 PF_ACPY(&key.addr[1], pd->src, key.af);
4186                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4187                 key.port[1] = th->th_sport;
4188                 key.port[0] = th->th_dport;
4189         }
4190
4191         STATE_LOOKUP(kif, &key, direction, *state, pd);
4192
4193         if (direction == (*state)->direction) {
4194                 src = &(*state)->src;
4195                 dst = &(*state)->dst;
4196         } else {
4197                 src = &(*state)->dst;
4198                 dst = &(*state)->src;
4199         }
4200
4201         sk = (*state)->key[pd->didx];
4202
4203         if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4204                 if (direction != (*state)->direction) {
4205                         REASON_SET(reason, PFRES_SYNPROXY);
4206                         return (PF_SYNPROXY_DROP);
4207                 }
4208                 if (th->th_flags & TH_SYN) {
4209                         if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4210                                 REASON_SET(reason, PFRES_SYNPROXY);
4211                                 return (PF_DROP);
4212                         }
4213                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4214                             pd->src, th->th_dport, th->th_sport,
4215                             (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4216                             TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4217                         REASON_SET(reason, PFRES_SYNPROXY);
4218                         return (PF_SYNPROXY_DROP);
4219                 } else if (!(th->th_flags & TH_ACK) ||
4220                     (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4221                     (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4222                         REASON_SET(reason, PFRES_SYNPROXY);
4223                         return (PF_DROP);
4224                 } else if ((*state)->src_node != NULL &&
4225                     pf_src_connlimit(state)) {
4226                         REASON_SET(reason, PFRES_SRCLIMIT);
4227                         return (PF_DROP);
4228                 } else
4229                         (*state)->src.state = PF_TCPS_PROXY_DST;
4230         }
4231         if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4232                 if (direction == (*state)->direction) {
4233                         if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4234                             (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4235                             (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4236                                 REASON_SET(reason, PFRES_SYNPROXY);
4237                                 return (PF_DROP);
4238                         }
4239                         (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4240                         if ((*state)->dst.seqhi == 1)
4241                                 (*state)->dst.seqhi = htonl(arc4random());
4242                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4243                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4244                             sk->port[pd->sidx], sk->port[pd->didx],
4245                             (*state)->dst.seqhi, 0, TH_SYN, 0,
4246                             (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4247                         REASON_SET(reason, PFRES_SYNPROXY);
4248                         return (PF_SYNPROXY_DROP);
4249                 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4250                     (TH_SYN|TH_ACK)) ||
4251                     (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4252                         REASON_SET(reason, PFRES_SYNPROXY);
4253                         return (PF_DROP);
4254                 } else {
4255                         (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4256                         (*state)->dst.seqlo = ntohl(th->th_seq);
4257                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4258                             pd->src, th->th_dport, th->th_sport,
4259                             ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4260                             TH_ACK, (*state)->src.max_win, 0, 0, 0,
4261                             (*state)->tag, NULL);
4262                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4263                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4264                             sk->port[pd->sidx], sk->port[pd->didx],
4265                             (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4266                             TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4267                         (*state)->src.seqdiff = (*state)->dst.seqhi -
4268                             (*state)->src.seqlo;
4269                         (*state)->dst.seqdiff = (*state)->src.seqhi -
4270                             (*state)->dst.seqlo;
4271                         (*state)->src.seqhi = (*state)->src.seqlo +
4272                             (*state)->dst.max_win;
4273                         (*state)->dst.seqhi = (*state)->dst.seqlo +
4274                             (*state)->src.max_win;
4275                         (*state)->src.wscale = (*state)->dst.wscale = 0;
4276                         (*state)->src.state = (*state)->dst.state =
4277                             TCPS_ESTABLISHED;
4278                         REASON_SET(reason, PFRES_SYNPROXY);
4279                         return (PF_SYNPROXY_DROP);
4280                 }
4281         }
4282
4283         if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4284             dst->state >= TCPS_FIN_WAIT_2 &&
4285             src->state >= TCPS_FIN_WAIT_2) {
4286                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4287                         printf("pf: state reuse ");
4288                         pf_print_state(*state);
4289                         pf_print_flags(th->th_flags);
4290                         printf("\n");
4291                 }
4292                 /* XXX make sure it's the same direction ?? */
4293                 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4294                 pf_unlink_state(*state, PF_ENTER_LOCKED);
4295                 *state = NULL;
4296                 return (PF_DROP);
4297         }
4298
4299         if ((*state)->state_flags & PFSTATE_SLOPPY) {
4300                 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4301                         return (PF_DROP);
4302         } else {
4303                 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4304                     &copyback) == PF_DROP)
4305                         return (PF_DROP);
4306         }
4307
4308         /* translate source/destination address, if necessary */
4309         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4310                 struct pf_state_key *nk = (*state)->key[pd->didx];
4311
4312                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4313                     nk->port[pd->sidx] != th->th_sport)
4314                         pf_change_ap(pd->src, &th->th_sport, pd->ip_sum,
4315                             &th->th_sum, &nk->addr[pd->sidx],
4316                             nk->port[pd->sidx], 0, pd->af);
4317
4318                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4319                     nk->port[pd->didx] != th->th_dport)
4320                         pf_change_ap(pd->dst, &th->th_dport, pd->ip_sum,
4321                             &th->th_sum, &nk->addr[pd->didx],
4322                             nk->port[pd->didx], 0, pd->af);
4323                 copyback = 1;
4324         }
4325
4326         /* Copyback sequence modulation or stateful scrub changes if needed */
4327         if (copyback)
4328                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
4329
4330         return (PF_PASS);
4331 }
4332
4333 static int
4334 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
4335     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4336 {
4337         struct pf_state_peer    *src, *dst;
4338         struct pf_state_key_cmp  key;
4339         struct udphdr           *uh = pd->hdr.udp;
4340
4341         bzero(&key, sizeof(key));
4342         key.af = pd->af;
4343         key.proto = IPPROTO_UDP;
4344         if (direction == PF_IN) {       /* wire side, straight */
4345                 PF_ACPY(&key.addr[0], pd->src, key.af);
4346                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4347                 key.port[0] = uh->uh_sport;
4348                 key.port[1] = uh->uh_dport;
4349         } else {                        /* stack side, reverse */
4350                 PF_ACPY(&key.addr[1], pd->src, key.af);
4351                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4352                 key.port[1] = uh->uh_sport;
4353                 key.port[0] = uh->uh_dport;
4354         }
4355
4356         STATE_LOOKUP(kif, &key, direction, *state, pd);
4357
4358         if (direction == (*state)->direction) {
4359                 src = &(*state)->src;
4360                 dst = &(*state)->dst;
4361         } else {
4362                 src = &(*state)->dst;
4363                 dst = &(*state)->src;
4364         }
4365
4366         /* update states */
4367         if (src->state < PFUDPS_SINGLE)
4368                 src->state = PFUDPS_SINGLE;
4369         if (dst->state == PFUDPS_SINGLE)
4370                 dst->state = PFUDPS_MULTIPLE;
4371
4372         /* update expire time */
4373         (*state)->expire = time_uptime;
4374         if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4375                 (*state)->timeout = PFTM_UDP_MULTIPLE;
4376         else
4377                 (*state)->timeout = PFTM_UDP_SINGLE;
4378
4379         /* translate source/destination address, if necessary */
4380         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4381                 struct pf_state_key *nk = (*state)->key[pd->didx];
4382
4383                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4384                     nk->port[pd->sidx] != uh->uh_sport)
4385                         pf_change_ap(pd->src, &uh->uh_sport, pd->ip_sum,
4386                             &uh->uh_sum, &nk->addr[pd->sidx],
4387                             nk->port[pd->sidx], 1, pd->af);
4388
4389                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4390                     nk->port[pd->didx] != uh->uh_dport)
4391                         pf_change_ap(pd->dst, &uh->uh_dport, pd->ip_sum,
4392                             &uh->uh_sum, &nk->addr[pd->didx],
4393                             nk->port[pd->didx], 1, pd->af);
4394                 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4395         }
4396
4397         return (PF_PASS);
4398 }
4399
4400 static int
4401 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
4402     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4403 {
4404         struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
4405         u_int16_t        icmpid = 0, *icmpsum;
4406         u_int8_t         icmptype;
4407         int              state_icmp = 0;
4408         struct pf_state_key_cmp key;
4409
4410         bzero(&key, sizeof(key));
4411         switch (pd->proto) {
4412 #ifdef INET
4413         case IPPROTO_ICMP:
4414                 icmptype = pd->hdr.icmp->icmp_type;
4415                 icmpid = pd->hdr.icmp->icmp_id;
4416                 icmpsum = &pd->hdr.icmp->icmp_cksum;
4417
4418                 if (icmptype == ICMP_UNREACH ||
4419                     icmptype == ICMP_SOURCEQUENCH ||
4420                     icmptype == ICMP_REDIRECT ||
4421                     icmptype == ICMP_TIMXCEED ||
4422                     icmptype == ICMP_PARAMPROB)
4423                         state_icmp++;
4424                 break;
4425 #endif /* INET */
4426 #ifdef INET6
4427         case IPPROTO_ICMPV6:
4428                 icmptype = pd->hdr.icmp6->icmp6_type;
4429                 icmpid = pd->hdr.icmp6->icmp6_id;
4430                 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4431
4432                 if (icmptype == ICMP6_DST_UNREACH ||
4433                     icmptype == ICMP6_PACKET_TOO_BIG ||
4434                     icmptype == ICMP6_TIME_EXCEEDED ||
4435                     icmptype == ICMP6_PARAM_PROB)
4436                         state_icmp++;
4437                 break;
4438 #endif /* INET6 */
4439         }
4440
4441         if (!state_icmp) {
4442
4443                 /*
4444                  * ICMP query/reply message not related to a TCP/UDP packet.
4445                  * Search for an ICMP state.
4446                  */
4447                 key.af = pd->af;
4448                 key.proto = pd->proto;
4449                 key.port[0] = key.port[1] = icmpid;
4450                 if (direction == PF_IN) {       /* wire side, straight */
4451                         PF_ACPY(&key.addr[0], pd->src, key.af);
4452                         PF_ACPY(&key.addr[1], pd->dst, key.af);
4453                 } else {                        /* stack side, reverse */
4454                         PF_ACPY(&key.addr[1], pd->src, key.af);
4455                         PF_ACPY(&key.addr[0], pd->dst, key.af);
4456                 }
4457
4458                 STATE_LOOKUP(kif, &key, direction, *state, pd);
4459
4460                 (*state)->expire = time_uptime;
4461                 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4462
4463                 /* translate source/destination address, if necessary */
4464                 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4465                         struct pf_state_key *nk = (*state)->key[pd->didx];
4466
4467                         switch (pd->af) {
4468 #ifdef INET
4469                         case AF_INET:
4470                                 if (PF_ANEQ(pd->src,
4471                                     &nk->addr[pd->sidx], AF_INET))
4472                                         pf_change_a(&saddr->v4.s_addr,
4473                                             pd->ip_sum,
4474                                             nk->addr[pd->sidx].v4.s_addr, 0);
4475
4476                                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4477                                     AF_INET))
4478                                         pf_change_a(&daddr->v4.s_addr,
4479                                             pd->ip_sum,
4480                                             nk->addr[pd->didx].v4.s_addr, 0);
4481
4482                                 if (nk->port[0] !=
4483                                     pd->hdr.icmp->icmp_id) {
4484                                         pd->hdr.icmp->icmp_cksum =
4485                                             pf_cksum_fixup(
4486                                             pd->hdr.icmp->icmp_cksum, icmpid,
4487                                             nk->port[pd->sidx], 0);
4488                                         pd->hdr.icmp->icmp_id =
4489                                             nk->port[pd->sidx];
4490                                 }
4491
4492                                 m_copyback(m, off, ICMP_MINLEN,
4493                                     (caddr_t )pd->hdr.icmp);
4494                                 break;
4495 #endif /* INET */
4496 #ifdef INET6
4497                         case AF_INET6:
4498                                 if (PF_ANEQ(pd->src,
4499                                     &nk->addr[pd->sidx], AF_INET6))
4500                                         pf_change_a6(saddr,
4501                                             &pd->hdr.icmp6->icmp6_cksum,
4502                                             &nk->addr[pd->sidx], 0);
4503
4504                                 if (PF_ANEQ(pd->dst,
4505                                     &nk->addr[pd->didx], AF_INET6))
4506                                         pf_change_a6(daddr,
4507                                             &pd->hdr.icmp6->icmp6_cksum,
4508                                             &nk->addr[pd->didx], 0);
4509
4510                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4511                                     (caddr_t )pd->hdr.icmp6);
4512                                 break;
4513 #endif /* INET6 */
4514                         }
4515                 }
4516                 return (PF_PASS);
4517
4518         } else {
4519                 /*
4520                  * ICMP error message in response to a TCP/UDP packet.
4521                  * Extract the inner TCP/UDP header and search for that state.
4522                  */
4523
4524                 struct pf_pdesc pd2;
4525                 bzero(&pd2, sizeof pd2);
4526 #ifdef INET
4527                 struct ip       h2;
4528 #endif /* INET */
4529 #ifdef INET6
4530                 struct ip6_hdr  h2_6;
4531                 int             terminal = 0;
4532 #endif /* INET6 */
4533                 int             ipoff2 = 0;
4534                 int             off2 = 0;
4535
4536                 pd2.af = pd->af;
4537                 /* Payload packet is from the opposite direction. */
4538                 pd2.sidx = (direction == PF_IN) ? 1 : 0;
4539                 pd2.didx = (direction == PF_IN) ? 0 : 1;
4540                 switch (pd->af) {
4541 #ifdef INET
4542                 case AF_INET:
4543                         /* offset of h2 in mbuf chain */
4544                         ipoff2 = off + ICMP_MINLEN;
4545
4546                         if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4547                             NULL, reason, pd2.af)) {
4548                                 DPFPRINTF(PF_DEBUG_MISC,
4549                                     ("pf: ICMP error message too short "
4550                                     "(ip)\n"));
4551                                 return (PF_DROP);
4552                         }
4553                         /*
4554                          * ICMP error messages don't refer to non-first
4555                          * fragments
4556                          */
4557                         if (h2.ip_off & htons(IP_OFFMASK)) {
4558                                 REASON_SET(reason, PFRES_FRAG);
4559                                 return (PF_DROP);
4560                         }
4561
4562                         /* offset of protocol header that follows h2 */
4563                         off2 = ipoff2 + (h2.ip_hl << 2);
4564
4565                         pd2.proto = h2.ip_p;
4566                         pd2.src = (struct pf_addr *)&h2.ip_src;
4567                         pd2.dst = (struct pf_addr *)&h2.ip_dst;
4568                         pd2.ip_sum = &h2.ip_sum;
4569                         break;
4570 #endif /* INET */
4571 #ifdef INET6
4572                 case AF_INET6:
4573                         ipoff2 = off + sizeof(struct icmp6_hdr);
4574
4575                         if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4576                             NULL, reason, pd2.af)) {
4577                                 DPFPRINTF(PF_DEBUG_MISC,
4578                                     ("pf: ICMP error message too short "
4579                                     "(ip6)\n"));
4580                                 return (PF_DROP);
4581                         }
4582                         pd2.proto = h2_6.ip6_nxt;
4583                         pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4584                         pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4585                         pd2.ip_sum = NULL;
4586                         off2 = ipoff2 + sizeof(h2_6);
4587                         do {
4588                                 switch (pd2.proto) {
4589                                 case IPPROTO_FRAGMENT:
4590                                         /*
4591                                          * ICMPv6 error messages for
4592                                          * non-first fragments
4593                                          */
4594                                         REASON_SET(reason, PFRES_FRAG);
4595                                         return (PF_DROP);
4596                                 case IPPROTO_AH:
4597                                 case IPPROTO_HOPOPTS:
4598                                 case IPPROTO_ROUTING:
4599                                 case IPPROTO_DSTOPTS: {
4600                                         /* get next header and header length */
4601                                         struct ip6_ext opt6;
4602
4603                                         if (!pf_pull_hdr(m, off2, &opt6,
4604                                             sizeof(opt6), NULL, reason,
4605                                             pd2.af)) {
4606                                                 DPFPRINTF(PF_DEBUG_MISC,
4607                                                     ("pf: ICMPv6 short opt\n"));
4608                                                 return (PF_DROP);
4609                                         }
4610                                         if (pd2.proto == IPPROTO_AH)
4611                                                 off2 += (opt6.ip6e_len + 2) * 4;
4612                                         else
4613                                                 off2 += (opt6.ip6e_len + 1) * 8;
4614                                         pd2.proto = opt6.ip6e_nxt;
4615                                         /* goto the next header */
4616                                         break;
4617                                 }
4618                                 default:
4619                                         terminal++;
4620                                         break;
4621                                 }
4622                         } while (!terminal);
4623                         break;
4624 #endif /* INET6 */
4625                 }
4626
4627                 switch (pd2.proto) {
4628                 case IPPROTO_TCP: {
4629                         struct tcphdr            th;
4630                         u_int32_t                seq;
4631                         struct pf_state_peer    *src, *dst;
4632                         u_int8_t                 dws;
4633                         int                      copyback = 0;
4634
4635                         /*
4636                          * Only the first 8 bytes of the TCP header can be
4637                          * expected. Don't access any TCP header fields after
4638                          * th_seq, an ackskew test is not possible.
4639                          */
4640                         if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4641                             pd2.af)) {
4642                                 DPFPRINTF(PF_DEBUG_MISC,
4643                                     ("pf: ICMP error message too short "
4644                                     "(tcp)\n"));
4645                                 return (PF_DROP);
4646                         }
4647
4648                         key.af = pd2.af;
4649                         key.proto = IPPROTO_TCP;
4650                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4651                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4652                         key.port[pd2.sidx] = th.th_sport;
4653                         key.port[pd2.didx] = th.th_dport;
4654
4655                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4656
4657                         if (direction == (*state)->direction) {
4658                                 src = &(*state)->dst;
4659                                 dst = &(*state)->src;
4660                         } else {
4661                                 src = &(*state)->src;
4662                                 dst = &(*state)->dst;
4663                         }
4664
4665                         if (src->wscale && dst->wscale)
4666                                 dws = dst->wscale & PF_WSCALE_MASK;
4667                         else
4668                                 dws = 0;
4669
4670                         /* Demodulate sequence number */
4671                         seq = ntohl(th.th_seq) - src->seqdiff;
4672                         if (src->seqdiff) {
4673                                 pf_change_a(&th.th_seq, icmpsum,
4674                                     htonl(seq), 0);
4675                                 copyback = 1;
4676                         }
4677
4678                         if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4679                             (!SEQ_GEQ(src->seqhi, seq) ||
4680                             !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4681                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4682                                         printf("pf: BAD ICMP %d:%d ",
4683                                             icmptype, pd->hdr.icmp->icmp_code);
4684                                         pf_print_host(pd->src, 0, pd->af);
4685                                         printf(" -> ");
4686                                         pf_print_host(pd->dst, 0, pd->af);
4687                                         printf(" state: ");
4688                                         pf_print_state(*state);
4689                                         printf(" seq=%u\n", seq);
4690                                 }
4691                                 REASON_SET(reason, PFRES_BADSTATE);
4692                                 return (PF_DROP);
4693                         } else {
4694                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4695                                         printf("pf: OK ICMP %d:%d ",
4696                                             icmptype, pd->hdr.icmp->icmp_code);
4697                                         pf_print_host(pd->src, 0, pd->af);
4698                                         printf(" -> ");
4699                                         pf_print_host(pd->dst, 0, pd->af);
4700                                         printf(" state: ");
4701                                         pf_print_state(*state);
4702                                         printf(" seq=%u\n", seq);
4703                                 }
4704                         }
4705
4706                         /* translate source/destination address, if necessary */
4707                         if ((*state)->key[PF_SK_WIRE] !=
4708                             (*state)->key[PF_SK_STACK]) {
4709                                 struct pf_state_key *nk =
4710                                     (*state)->key[pd->didx];
4711
4712                                 if (PF_ANEQ(pd2.src,
4713                                     &nk->addr[pd2.sidx], pd2.af) ||
4714                                     nk->port[pd2.sidx] != th.th_sport)
4715                                         pf_change_icmp(pd2.src, &th.th_sport,
4716                                             daddr, &nk->addr[pd2.sidx],
4717                                             nk->port[pd2.sidx], NULL,
4718                                             pd2.ip_sum, icmpsum,
4719                                             pd->ip_sum, 0, pd2.af);
4720
4721                                 if (PF_ANEQ(pd2.dst,
4722                                     &nk->addr[pd2.didx], pd2.af) ||
4723                                     nk->port[pd2.didx] != th.th_dport)
4724                                         pf_change_icmp(pd2.dst, &th.th_dport,
4725                                             NULL, /* XXX Inbound NAT? */
4726                                             &nk->addr[pd2.didx],
4727                                             nk->port[pd2.didx], NULL,
4728                                             pd2.ip_sum, icmpsum,
4729                                             pd->ip_sum, 0, pd2.af);
4730                                 copyback = 1;
4731                         }
4732
4733                         if (copyback) {
4734                                 switch (pd2.af) {
4735 #ifdef INET
4736                                 case AF_INET:
4737                                         m_copyback(m, off, ICMP_MINLEN,
4738                                             (caddr_t )pd->hdr.icmp);
4739                                         m_copyback(m, ipoff2, sizeof(h2),
4740                                             (caddr_t )&h2);
4741                                         break;
4742 #endif /* INET */
4743 #ifdef INET6
4744                                 case AF_INET6:
4745                                         m_copyback(m, off,
4746                                             sizeof(struct icmp6_hdr),
4747                                             (caddr_t )pd->hdr.icmp6);
4748                                         m_copyback(m, ipoff2, sizeof(h2_6),
4749                                             (caddr_t )&h2_6);
4750                                         break;
4751 #endif /* INET6 */
4752                                 }
4753                                 m_copyback(m, off2, 8, (caddr_t)&th);
4754                         }
4755
4756                         return (PF_PASS);
4757                         break;
4758                 }
4759                 case IPPROTO_UDP: {
4760                         struct udphdr           uh;
4761
4762                         if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
4763                             NULL, reason, pd2.af)) {
4764                                 DPFPRINTF(PF_DEBUG_MISC,
4765                                     ("pf: ICMP error message too short "
4766                                     "(udp)\n"));
4767                                 return (PF_DROP);
4768                         }
4769
4770                         key.af = pd2.af;
4771                         key.proto = IPPROTO_UDP;
4772                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4773                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4774                         key.port[pd2.sidx] = uh.uh_sport;
4775                         key.port[pd2.didx] = uh.uh_dport;
4776
4777                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4778
4779                         /* translate source/destination address, if necessary */
4780                         if ((*state)->key[PF_SK_WIRE] !=
4781                             (*state)->key[PF_SK_STACK]) {
4782                                 struct pf_state_key *nk =
4783                                     (*state)->key[pd->didx];
4784
4785                                 if (PF_ANEQ(pd2.src,
4786                                     &nk->addr[pd2.sidx], pd2.af) ||
4787                                     nk->port[pd2.sidx] != uh.uh_sport)
4788                                         pf_change_icmp(pd2.src, &uh.uh_sport,
4789                                             daddr, &nk->addr[pd2.sidx],
4790                                             nk->port[pd2.sidx], &uh.uh_sum,
4791                                             pd2.ip_sum, icmpsum,
4792                                             pd->ip_sum, 1, pd2.af);
4793
4794                                 if (PF_ANEQ(pd2.dst,
4795                                     &nk->addr[pd2.didx], pd2.af) ||
4796                                     nk->port[pd2.didx] != uh.uh_dport)
4797                                         pf_change_icmp(pd2.dst, &uh.uh_dport,
4798                                             NULL, /* XXX Inbound NAT? */
4799                                             &nk->addr[pd2.didx],
4800                                             nk->port[pd2.didx], &uh.uh_sum,
4801                                             pd2.ip_sum, icmpsum,
4802                                             pd->ip_sum, 1, pd2.af);
4803
4804                                 switch (pd2.af) {
4805 #ifdef INET
4806                                 case AF_INET:
4807                                         m_copyback(m, off, ICMP_MINLEN,
4808                                             (caddr_t )pd->hdr.icmp);
4809                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4810                                         break;
4811 #endif /* INET */
4812 #ifdef INET6
4813                                 case AF_INET6:
4814                                         m_copyback(m, off,
4815                                             sizeof(struct icmp6_hdr),
4816                                             (caddr_t )pd->hdr.icmp6);
4817                                         m_copyback(m, ipoff2, sizeof(h2_6),
4818                                             (caddr_t )&h2_6);
4819                                         break;
4820 #endif /* INET6 */
4821                                 }
4822                                 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
4823                         }
4824                         return (PF_PASS);
4825                         break;
4826                 }
4827 #ifdef INET
4828                 case IPPROTO_ICMP: {
4829                         struct icmp             iih;
4830
4831                         if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
4832                             NULL, reason, pd2.af)) {
4833                                 DPFPRINTF(PF_DEBUG_MISC,
4834                                     ("pf: ICMP error message too short i"
4835                                     "(icmp)\n"));
4836                                 return (PF_DROP);
4837                         }
4838
4839                         key.af = pd2.af;
4840                         key.proto = IPPROTO_ICMP;
4841                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4842                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4843                         key.port[0] = key.port[1] = iih.icmp_id;
4844
4845                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4846
4847                         /* translate source/destination address, if necessary */
4848                         if ((*state)->key[PF_SK_WIRE] !=
4849                             (*state)->key[PF_SK_STACK]) {
4850                                 struct pf_state_key *nk =
4851                                     (*state)->key[pd->didx];
4852
4853                                 if (PF_ANEQ(pd2.src,
4854                                     &nk->addr[pd2.sidx], pd2.af) ||
4855                                     nk->port[pd2.sidx] != iih.icmp_id)
4856                                         pf_change_icmp(pd2.src, &iih.icmp_id,
4857                                             daddr, &nk->addr[pd2.sidx],
4858                                             nk->port[pd2.sidx], NULL,
4859                                             pd2.ip_sum, icmpsum,
4860                                             pd->ip_sum, 0, AF_INET);
4861
4862                                 if (PF_ANEQ(pd2.dst,
4863                                     &nk->addr[pd2.didx], pd2.af) ||
4864                                     nk->port[pd2.didx] != iih.icmp_id)
4865                                         pf_change_icmp(pd2.dst, &iih.icmp_id,
4866                                             NULL, /* XXX Inbound NAT? */
4867                                             &nk->addr[pd2.didx],
4868                                             nk->port[pd2.didx], NULL,
4869                                             pd2.ip_sum, icmpsum,
4870                                             pd->ip_sum, 0, AF_INET);
4871
4872                                 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
4873                                 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4874                                 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
4875                         }
4876                         return (PF_PASS);
4877                         break;
4878                 }
4879 #endif /* INET */
4880 #ifdef INET6
4881                 case IPPROTO_ICMPV6: {
4882                         struct icmp6_hdr        iih;
4883
4884                         if (!pf_pull_hdr(m, off2, &iih,
4885                             sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
4886                                 DPFPRINTF(PF_DEBUG_MISC,
4887                                     ("pf: ICMP error message too short "
4888                                     "(icmp6)\n"));
4889                                 return (PF_DROP);
4890                         }
4891
4892                         key.af = pd2.af;
4893                         key.proto = IPPROTO_ICMPV6;
4894                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4895                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4896                         key.port[0] = key.port[1] = iih.icmp6_id;
4897
4898                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4899
4900                         /* translate source/destination address, if necessary */
4901                         if ((*state)->key[PF_SK_WIRE] !=
4902                             (*state)->key[PF_SK_STACK]) {
4903                                 struct pf_state_key *nk =
4904                                     (*state)->key[pd->didx];
4905
4906                                 if (PF_ANEQ(pd2.src,
4907                                     &nk->addr[pd2.sidx], pd2.af) ||
4908                                     nk->port[pd2.sidx] != iih.icmp6_id)
4909                                         pf_change_icmp(pd2.src, &iih.icmp6_id,
4910                                             daddr, &nk->addr[pd2.sidx],
4911                                             nk->port[pd2.sidx], NULL,
4912                                             pd2.ip_sum, icmpsum,
4913                                             pd->ip_sum, 0, AF_INET6);
4914
4915                                 if (PF_ANEQ(pd2.dst,
4916                                     &nk->addr[pd2.didx], pd2.af) ||
4917                                     nk->port[pd2.didx] != iih.icmp6_id)
4918                                         pf_change_icmp(pd2.dst, &iih.icmp6_id,
4919                                             NULL, /* XXX Inbound NAT? */
4920                                             &nk->addr[pd2.didx],
4921                                             nk->port[pd2.didx], NULL,
4922                                             pd2.ip_sum, icmpsum,
4923                                             pd->ip_sum, 0, AF_INET6);
4924
4925                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4926                                     (caddr_t)pd->hdr.icmp6);
4927                                 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
4928                                 m_copyback(m, off2, sizeof(struct icmp6_hdr),
4929                                     (caddr_t)&iih);
4930                         }
4931                         return (PF_PASS);
4932                         break;
4933                 }
4934 #endif /* INET6 */
4935                 default: {
4936                         key.af = pd2.af;
4937                         key.proto = pd2.proto;
4938                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4939                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4940                         key.port[0] = key.port[1] = 0;
4941
4942                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4943
4944                         /* translate source/destination address, if necessary */
4945                         if ((*state)->key[PF_SK_WIRE] !=
4946                             (*state)->key[PF_SK_STACK]) {
4947                                 struct pf_state_key *nk =
4948                                     (*state)->key[pd->didx];
4949
4950                                 if (PF_ANEQ(pd2.src,
4951                                     &nk->addr[pd2.sidx], pd2.af))
4952                                         pf_change_icmp(pd2.src, NULL, daddr,
4953                                             &nk->addr[pd2.sidx], 0, NULL,
4954                                             pd2.ip_sum, icmpsum,
4955                                             pd->ip_sum, 0, pd2.af);
4956
4957                                 if (PF_ANEQ(pd2.dst,
4958                                     &nk->addr[pd2.didx], pd2.af))
4959                                         pf_change_icmp(pd2.src, NULL,
4960                                             NULL, /* XXX Inbound NAT? */
4961                                             &nk->addr[pd2.didx], 0, NULL,
4962                                             pd2.ip_sum, icmpsum,
4963                                             pd->ip_sum, 0, pd2.af);
4964
4965                                 switch (pd2.af) {
4966 #ifdef INET
4967                                 case AF_INET:
4968                                         m_copyback(m, off, ICMP_MINLEN,
4969                                             (caddr_t)pd->hdr.icmp);
4970                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4971                                         break;
4972 #endif /* INET */
4973 #ifdef INET6
4974                                 case AF_INET6:
4975                                         m_copyback(m, off,
4976                                             sizeof(struct icmp6_hdr),
4977                                             (caddr_t )pd->hdr.icmp6);
4978                                         m_copyback(m, ipoff2, sizeof(h2_6),
4979                                             (caddr_t )&h2_6);
4980                                         break;
4981 #endif /* INET6 */
4982                                 }
4983                         }
4984                         return (PF_PASS);
4985                         break;
4986                 }
4987                 }
4988         }
4989 }
4990
4991 static int
4992 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
4993     struct mbuf *m, struct pf_pdesc *pd)
4994 {
4995         struct pf_state_peer    *src, *dst;
4996         struct pf_state_key_cmp  key;
4997
4998         bzero(&key, sizeof(key));
4999         key.af = pd->af;
5000         key.proto = pd->proto;
5001         if (direction == PF_IN) {
5002                 PF_ACPY(&key.addr[0], pd->src, key.af);
5003                 PF_ACPY(&key.addr[1], pd->dst, key.af);
5004                 key.port[0] = key.port[1] = 0;
5005         } else {
5006                 PF_ACPY(&key.addr[1], pd->src, key.af);
5007                 PF_ACPY(&key.addr[0], pd->dst, key.af);
5008                 key.port[1] = key.port[0] = 0;
5009         }
5010
5011         STATE_LOOKUP(kif, &key, direction, *state, pd);
5012
5013         if (direction == (*state)->direction) {
5014                 src = &(*state)->src;
5015                 dst = &(*state)->dst;
5016         } else {
5017                 src = &(*state)->dst;
5018                 dst = &(*state)->src;
5019         }
5020
5021         /* update states */
5022         if (src->state < PFOTHERS_SINGLE)
5023                 src->state = PFOTHERS_SINGLE;
5024         if (dst->state == PFOTHERS_SINGLE)
5025                 dst->state = PFOTHERS_MULTIPLE;
5026
5027         /* update expire time */
5028         (*state)->expire = time_uptime;
5029         if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5030                 (*state)->timeout = PFTM_OTHER_MULTIPLE;
5031         else
5032                 (*state)->timeout = PFTM_OTHER_SINGLE;
5033
5034         /* translate source/destination address, if necessary */
5035         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5036                 struct pf_state_key *nk = (*state)->key[pd->didx];
5037
5038                 KASSERT(nk, ("%s: nk is null", __func__));
5039                 KASSERT(pd, ("%s: pd is null", __func__));
5040                 KASSERT(pd->src, ("%s: pd->src is null", __func__));
5041                 KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5042                 switch (pd->af) {
5043 #ifdef INET
5044                 case AF_INET:
5045                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5046                                 pf_change_a(&pd->src->v4.s_addr,
5047                                     pd->ip_sum,
5048                                     nk->addr[pd->sidx].v4.s_addr,
5049                                     0);
5050
5051
5052                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5053                                 pf_change_a(&pd->dst->v4.s_addr,
5054                                     pd->ip_sum,
5055                                     nk->addr[pd->didx].v4.s_addr,
5056                                     0);
5057
5058                                 break;
5059 #endif /* INET */
5060 #ifdef INET6
5061                 case AF_INET6:
5062                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5063                                 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5064
5065                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5066                                 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5067 #endif /* INET6 */
5068                 }
5069         }
5070         return (PF_PASS);
5071 }
5072
5073 /*
5074  * ipoff and off are measured from the start of the mbuf chain.
5075  * h must be at "ipoff" on the mbuf chain.
5076  */
5077 void *
5078 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5079     u_short *actionp, u_short *reasonp, sa_family_t af)
5080 {
5081         switch (af) {
5082 #ifdef INET
5083         case AF_INET: {
5084                 struct ip       *h = mtod(m, struct ip *);
5085                 u_int16_t        fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5086
5087                 if (fragoff) {
5088                         if (fragoff >= len)
5089                                 ACTION_SET(actionp, PF_PASS);
5090                         else {
5091                                 ACTION_SET(actionp, PF_DROP);
5092                                 REASON_SET(reasonp, PFRES_FRAG);
5093                         }
5094                         return (NULL);
5095                 }
5096                 if (m->m_pkthdr.len < off + len ||
5097                     ntohs(h->ip_len) < off + len) {
5098                         ACTION_SET(actionp, PF_DROP);
5099                         REASON_SET(reasonp, PFRES_SHORT);
5100                         return (NULL);
5101                 }
5102                 break;
5103         }
5104 #endif /* INET */
5105 #ifdef INET6
5106         case AF_INET6: {
5107                 struct ip6_hdr  *h = mtod(m, struct ip6_hdr *);
5108
5109                 if (m->m_pkthdr.len < off + len ||
5110                     (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5111                     (unsigned)(off + len)) {
5112                         ACTION_SET(actionp, PF_DROP);
5113                         REASON_SET(reasonp, PFRES_SHORT);
5114                         return (NULL);
5115                 }
5116                 break;
5117         }
5118 #endif /* INET6 */
5119         }
5120         m_copydata(m, off, len, p);
5121         return (p);
5122 }
5123
5124 int
5125 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5126     int rtableid)
5127 {
5128 #ifdef RADIX_MPATH
5129         struct radix_node_head  *rnh;
5130 #endif
5131         struct sockaddr_in      *dst;
5132         int                      ret = 1;
5133         int                      check_mpath;
5134 #ifdef INET6
5135         struct sockaddr_in6     *dst6;
5136         struct route_in6         ro;
5137 #else
5138         struct route             ro;
5139 #endif
5140         struct radix_node       *rn;
5141         struct rtentry          *rt;
5142         struct ifnet            *ifp;
5143
5144         check_mpath = 0;
5145 #ifdef RADIX_MPATH
5146         /* XXX: stick to table 0 for now */
5147         rnh = rt_tables_get_rnh(0, af);
5148         if (rnh != NULL && rn_mpath_capable(rnh))
5149                 check_mpath = 1;
5150 #endif
5151         bzero(&ro, sizeof(ro));
5152         switch (af) {
5153         case AF_INET:
5154                 dst = satosin(&ro.ro_dst);
5155                 dst->sin_family = AF_INET;
5156                 dst->sin_len = sizeof(*dst);
5157                 dst->sin_addr = addr->v4;
5158                 break;
5159 #ifdef INET6
5160         case AF_INET6:
5161                 /*
5162                  * Skip check for addresses with embedded interface scope,
5163                  * as they would always match anyway.
5164                  */
5165                 if (IN6_IS_SCOPE_EMBED(&addr->v6))
5166                         goto out;
5167                 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5168                 dst6->sin6_family = AF_INET6;
5169                 dst6->sin6_len = sizeof(*dst6);
5170                 dst6->sin6_addr = addr->v6;
5171                 break;
5172 #endif /* INET6 */
5173         default:
5174                 return (0);
5175         }
5176
5177         /* Skip checks for ipsec interfaces */
5178         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5179                 goto out;
5180
5181         switch (af) {
5182 #ifdef INET6
5183         case AF_INET6:
5184                 in6_rtalloc_ign(&ro, 0, rtableid);
5185                 break;
5186 #endif
5187 #ifdef INET
5188         case AF_INET:
5189                 in_rtalloc_ign((struct route *)&ro, 0, rtableid);
5190                 break;
5191 #endif
5192         default:
5193                 rtalloc_ign((struct route *)&ro, 0);    /* No/default FIB. */
5194                 break;
5195         }
5196
5197         if (ro.ro_rt != NULL) {
5198                 /* No interface given, this is a no-route check */
5199                 if (kif == NULL)
5200                         goto out;
5201
5202                 if (kif->pfik_ifp == NULL) {
5203                         ret = 0;
5204                         goto out;
5205                 }
5206
5207                 /* Perform uRPF check if passed input interface */
5208                 ret = 0;
5209                 rn = (struct radix_node *)ro.ro_rt;
5210                 do {
5211                         rt = (struct rtentry *)rn;
5212                         ifp = rt->rt_ifp;
5213
5214                         if (kif->pfik_ifp == ifp)
5215                                 ret = 1;
5216 #ifdef RADIX_MPATH
5217                         rn = rn_mpath_next(rn);
5218 #endif
5219                 } while (check_mpath == 1 && rn != NULL && ret == 0);
5220         } else
5221                 ret = 0;
5222 out:
5223         if (ro.ro_rt != NULL)
5224                 RTFREE(ro.ro_rt);
5225         return (ret);
5226 }
5227
5228 #ifdef INET
5229 static void
5230 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5231     struct pf_state *s, struct pf_pdesc *pd)
5232 {
5233         struct mbuf             *m0, *m1;
5234         struct sockaddr_in      dst;
5235         struct ip               *ip;
5236         struct ifnet            *ifp = NULL;
5237         struct pf_addr           naddr;
5238         struct pf_src_node      *sn = NULL;
5239         int                      error = 0;
5240         uint16_t                 ip_len, ip_off;
5241
5242         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5243         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5244             __func__));
5245
5246         if ((pd->pf_mtag == NULL &&
5247             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5248             pd->pf_mtag->routed++ > 3) {
5249                 m0 = *m;
5250                 *m = NULL;
5251                 goto bad_locked;
5252         }
5253
5254         if (r->rt == PF_DUPTO) {
5255                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5256                         if (s)
5257                                 PF_STATE_UNLOCK(s);
5258                         return;
5259                 }
5260         } else {
5261                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5262                         if (s)
5263                                 PF_STATE_UNLOCK(s);
5264                         return;
5265                 }
5266                 m0 = *m;
5267         }
5268
5269         ip = mtod(m0, struct ip *);
5270
5271         bzero(&dst, sizeof(dst));
5272         dst.sin_family = AF_INET;
5273         dst.sin_len = sizeof(dst);
5274         dst.sin_addr = ip->ip_dst;
5275
5276         if (r->rt == PF_FASTROUTE) {
5277                 struct rtentry *rt;
5278
5279                 if (s)
5280                         PF_STATE_UNLOCK(s);
5281                 rt = rtalloc1_fib(sintosa(&dst), 0, 0, M_GETFIB(m0));
5282                 if (rt == NULL) {
5283                         KMOD_IPSTAT_INC(ips_noroute);
5284                         error = EHOSTUNREACH;
5285                         goto bad;
5286                 }
5287
5288                 ifp = rt->rt_ifp;
5289                 counter_u64_add(rt->rt_pksent, 1);
5290
5291                 if (rt->rt_flags & RTF_GATEWAY)
5292                         bcopy(satosin(rt->rt_gateway), &dst, sizeof(dst));
5293                 RTFREE_LOCKED(rt);
5294         } else {
5295                 if (TAILQ_EMPTY(&r->rpool.list)) {
5296                         DPFPRINTF(PF_DEBUG_URGENT,
5297                             ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5298                         goto bad_locked;
5299                 }
5300                 if (s == NULL) {
5301                         pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5302                             &naddr, NULL, &sn);
5303                         if (!PF_AZERO(&naddr, AF_INET))
5304                                 dst.sin_addr.s_addr = naddr.v4.s_addr;
5305                         ifp = r->rpool.cur->kif ?
5306                             r->rpool.cur->kif->pfik_ifp : NULL;
5307                 } else {
5308                         if (!PF_AZERO(&s->rt_addr, AF_INET))
5309                                 dst.sin_addr.s_addr =
5310                                     s->rt_addr.v4.s_addr;
5311                         ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5312                         PF_STATE_UNLOCK(s);
5313                 }
5314         }
5315         if (ifp == NULL)
5316                 goto bad;
5317
5318         if (oifp != ifp) {
5319                 if (pf_test(PF_OUT, ifp, &m0, NULL) != PF_PASS)
5320                         goto bad;
5321                 else if (m0 == NULL)
5322                         goto done;
5323                 if (m0->m_len < sizeof(struct ip)) {
5324                         DPFPRINTF(PF_DEBUG_URGENT,
5325                             ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5326                         goto bad;
5327                 }
5328                 ip = mtod(m0, struct ip *);
5329         }
5330
5331         if (ifp->if_flags & IFF_LOOPBACK)
5332                 m0->m_flags |= M_SKIP_FIREWALL;
5333
5334         ip_len = ntohs(ip->ip_len);
5335         ip_off = ntohs(ip->ip_off);
5336
5337         /* Copied from FreeBSD 10.0-CURRENT ip_output. */
5338         m0->m_pkthdr.csum_flags |= CSUM_IP;
5339         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5340                 in_delayed_cksum(m0);
5341                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5342         }
5343 #ifdef SCTP
5344         if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5345                 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
5346                 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5347         }
5348 #endif
5349
5350         /*
5351          * If small enough for interface, or the interface will take
5352          * care of the fragmentation for us, we can just send directly.
5353          */
5354         if (ip_len <= ifp->if_mtu ||
5355             (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 ||
5356             ((ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) {
5357                 ip->ip_sum = 0;
5358                 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5359                         ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5360                         m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5361                 }
5362                 m_clrprotoflags(m0);    /* Avoid confusing lower layers. */
5363                 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5364                 goto done;
5365         }
5366
5367         /* Balk when DF bit is set or the interface didn't support TSO. */
5368         if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5369                 error = EMSGSIZE;
5370                 KMOD_IPSTAT_INC(ips_cantfrag);
5371                 if (r->rt != PF_DUPTO) {
5372                         icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5373                             ifp->if_mtu);
5374                         goto done;
5375                 } else
5376                         goto bad;
5377         }
5378
5379         error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5380         if (error)
5381                 goto bad;
5382
5383         for (; m0; m0 = m1) {
5384                 m1 = m0->m_nextpkt;
5385                 m0->m_nextpkt = NULL;
5386                 if (error == 0) {
5387                         m_clrprotoflags(m0);
5388                         error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5389                 } else
5390                         m_freem(m0);
5391         }
5392
5393         if (error == 0)
5394                 KMOD_IPSTAT_INC(ips_fragmented);
5395
5396 done:
5397         if (r->rt != PF_DUPTO)
5398                 *m = NULL;
5399         return;
5400
5401 bad_locked:
5402         if (s)
5403                 PF_STATE_UNLOCK(s);
5404 bad:
5405         m_freem(m0);
5406         goto done;
5407 }
5408 #endif /* INET */
5409
5410 #ifdef INET6
5411 static void
5412 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5413     struct pf_state *s, struct pf_pdesc *pd)
5414 {
5415         struct mbuf             *m0;
5416         struct sockaddr_in6     dst;
5417         struct ip6_hdr          *ip6;
5418         struct ifnet            *ifp = NULL;
5419         struct pf_addr           naddr;
5420         struct pf_src_node      *sn = NULL;
5421
5422         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5423         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5424             __func__));
5425
5426         if ((pd->pf_mtag == NULL &&
5427             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5428             pd->pf_mtag->routed++ > 3) {
5429                 m0 = *m;
5430                 *m = NULL;
5431                 goto bad_locked;
5432         }
5433
5434         if (r->rt == PF_DUPTO) {
5435                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5436                         if (s)
5437                                 PF_STATE_UNLOCK(s);
5438                         return;
5439                 }
5440         } else {
5441                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5442                         if (s)
5443                                 PF_STATE_UNLOCK(s);
5444                         return;
5445                 }
5446                 m0 = *m;
5447         }
5448
5449         ip6 = mtod(m0, struct ip6_hdr *);
5450
5451         bzero(&dst, sizeof(dst));
5452         dst.sin6_family = AF_INET6;
5453         dst.sin6_len = sizeof(dst);
5454         dst.sin6_addr = ip6->ip6_dst;
5455
5456         /* Cheat. XXX why only in the v6 case??? */
5457         if (r->rt == PF_FASTROUTE) {
5458                 if (s)
5459                         PF_STATE_UNLOCK(s);
5460                 m0->m_flags |= M_SKIP_FIREWALL;
5461                 ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL);
5462                 return;
5463         }
5464
5465         if (TAILQ_EMPTY(&r->rpool.list)) {
5466                 DPFPRINTF(PF_DEBUG_URGENT,
5467                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5468                 goto bad_locked;
5469         }
5470         if (s == NULL) {
5471                 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5472                     &naddr, NULL, &sn);
5473                 if (!PF_AZERO(&naddr, AF_INET6))
5474                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5475                             &naddr, AF_INET6);
5476                 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5477         } else {
5478                 if (!PF_AZERO(&s->rt_addr, AF_INET6))
5479                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5480                             &s->rt_addr, AF_INET6);
5481                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5482         }
5483
5484         if (s)
5485                 PF_STATE_UNLOCK(s);
5486
5487         if (ifp == NULL)
5488                 goto bad;
5489
5490         if (oifp != ifp) {
5491                 if (pf_test6(PF_OUT, ifp, &m0, NULL) != PF_PASS)
5492                         goto bad;
5493                 else if (m0 == NULL)
5494                         goto done;
5495                 if (m0->m_len < sizeof(struct ip6_hdr)) {
5496                         DPFPRINTF(PF_DEBUG_URGENT,
5497                             ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5498                             __func__));
5499                         goto bad;
5500                 }
5501                 ip6 = mtod(m0, struct ip6_hdr *);
5502         }
5503
5504         if (ifp->if_flags & IFF_LOOPBACK)
5505                 m0->m_flags |= M_SKIP_FIREWALL;
5506
5507         /*
5508          * If the packet is too large for the outgoing interface,
5509          * send back an icmp6 error.
5510          */
5511         if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5512                 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5513         if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5514                 nd6_output(ifp, ifp, m0, &dst, NULL);
5515         else {
5516                 in6_ifstat_inc(ifp, ifs6_in_toobig);
5517                 if (r->rt != PF_DUPTO)
5518                         icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5519                 else
5520                         goto bad;
5521         }
5522
5523 done:
5524         if (r->rt != PF_DUPTO)
5525                 *m = NULL;
5526         return;
5527
5528 bad_locked:
5529         if (s)
5530                 PF_STATE_UNLOCK(s);
5531 bad:
5532         m_freem(m0);
5533         goto done;
5534 }
5535 #endif /* INET6 */
5536
5537 /*
5538  * FreeBSD supports cksum offloads for the following drivers.
5539  *  em(4), fxp(4), ixgb(4), lge(4), ndis(4), nge(4), re(4),
5540  *   ti(4), txp(4), xl(4)
5541  *
5542  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5543  *  network driver performed cksum including pseudo header, need to verify
5544  *   csum_data
5545  * CSUM_DATA_VALID :
5546  *  network driver performed cksum, needs to additional pseudo header
5547  *  cksum computation with partial csum_data(i.e. lack of H/W support for
5548  *  pseudo header, for instance hme(4), sk(4) and possibly gem(4))
5549  *
5550  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5551  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5552  * TCP/UDP layer.
5553  * Also, set csum_data to 0xffff to force cksum validation.
5554  */
5555 static int
5556 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5557 {
5558         u_int16_t sum = 0;
5559         int hw_assist = 0;
5560         struct ip *ip;
5561
5562         if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5563                 return (1);
5564         if (m->m_pkthdr.len < off + len)
5565                 return (1);
5566
5567         switch (p) {
5568         case IPPROTO_TCP:
5569                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5570                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5571                                 sum = m->m_pkthdr.csum_data;
5572                         } else {
5573                                 ip = mtod(m, struct ip *);
5574                                 sum = in_pseudo(ip->ip_src.s_addr,
5575                                 ip->ip_dst.s_addr, htonl((u_short)len +
5576                                 m->m_pkthdr.csum_data + IPPROTO_TCP));
5577                         }
5578                         sum ^= 0xffff;
5579                         ++hw_assist;
5580                 }
5581                 break;
5582         case IPPROTO_UDP:
5583                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5584                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5585                                 sum = m->m_pkthdr.csum_data;
5586                         } else {
5587                                 ip = mtod(m, struct ip *);
5588                                 sum = in_pseudo(ip->ip_src.s_addr,
5589                                 ip->ip_dst.s_addr, htonl((u_short)len +
5590                                 m->m_pkthdr.csum_data + IPPROTO_UDP));
5591                         }
5592                         sum ^= 0xffff;
5593                         ++hw_assist;
5594                 }
5595                 break;
5596         case IPPROTO_ICMP:
5597 #ifdef INET6
5598         case IPPROTO_ICMPV6:
5599 #endif /* INET6 */
5600                 break;
5601         default:
5602                 return (1);
5603         }
5604
5605         if (!hw_assist) {
5606                 switch (af) {
5607                 case AF_INET:
5608                         if (p == IPPROTO_ICMP) {
5609                                 if (m->m_len < off)
5610                                         return (1);
5611                                 m->m_data += off;
5612                                 m->m_len -= off;
5613                                 sum = in_cksum(m, len);
5614                                 m->m_data -= off;
5615                                 m->m_len += off;
5616                         } else {
5617                                 if (m->m_len < sizeof(struct ip))
5618                                         return (1);
5619                                 sum = in4_cksum(m, p, off, len);
5620                         }
5621                         break;
5622 #ifdef INET6
5623                 case AF_INET6:
5624                         if (m->m_len < sizeof(struct ip6_hdr))
5625                                 return (1);
5626                         sum = in6_cksum(m, p, off, len);
5627                         break;
5628 #endif /* INET6 */
5629                 default:
5630                         return (1);
5631                 }
5632         }
5633         if (sum) {
5634                 switch (p) {
5635                 case IPPROTO_TCP:
5636                     {
5637                         KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5638                         break;
5639                     }
5640                 case IPPROTO_UDP:
5641                     {
5642                         KMOD_UDPSTAT_INC(udps_badsum);
5643                         break;
5644                     }
5645 #ifdef INET
5646                 case IPPROTO_ICMP:
5647                     {
5648                         KMOD_ICMPSTAT_INC(icps_checksum);
5649                         break;
5650                     }
5651 #endif
5652 #ifdef INET6
5653                 case IPPROTO_ICMPV6:
5654                     {
5655                         KMOD_ICMP6STAT_INC(icp6s_checksum);
5656                         break;
5657                     }
5658 #endif /* INET6 */
5659                 }
5660                 return (1);
5661         } else {
5662                 if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5663                         m->m_pkthdr.csum_flags |=
5664                             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5665                         m->m_pkthdr.csum_data = 0xffff;
5666                 }
5667         }
5668         return (0);
5669 }
5670
5671
5672 #ifdef INET
5673 int
5674 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5675 {
5676         struct pfi_kif          *kif;
5677         u_short                  action, reason = 0, log = 0;
5678         struct mbuf             *m = *m0;
5679         struct ip               *h = NULL;
5680         struct m_tag            *ipfwtag;
5681         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5682         struct pf_state         *s = NULL;
5683         struct pf_ruleset       *ruleset = NULL;
5684         struct pf_pdesc          pd;
5685         int                      off, dirndx, pqid = 0;
5686
5687         M_ASSERTPKTHDR(m);
5688
5689         if (!V_pf_status.running)
5690                 return (PF_PASS);
5691
5692         memset(&pd, 0, sizeof(pd));
5693
5694         kif = (struct pfi_kif *)ifp->if_pf_kif;
5695
5696         if (kif == NULL) {
5697                 DPFPRINTF(PF_DEBUG_URGENT,
5698                     ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5699                 return (PF_DROP);
5700         }
5701         if (kif->pfik_flags & PFI_IFLAG_SKIP)
5702                 return (PF_PASS);
5703
5704         if (m->m_flags & M_SKIP_FIREWALL)
5705                 return (PF_PASS);
5706
5707         pd.pf_mtag = pf_find_mtag(m);
5708
5709         PF_RULES_RLOCK();
5710
5711         if (ip_divert_ptr != NULL &&
5712             ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5713                 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5714                 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5715                         if (pd.pf_mtag == NULL &&
5716                             ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5717                                 action = PF_DROP;
5718                                 goto done;
5719                         }
5720                         pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5721                         m_tag_delete(m, ipfwtag);
5722                 }
5723                 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5724                         m->m_flags |= M_FASTFWD_OURS;
5725                         pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5726                 }
5727         } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5728                 /* We do IP header normalization and packet reassembly here */
5729                 action = PF_DROP;
5730                 goto done;
5731         }
5732         m = *m0;        /* pf_normalize messes with m0 */
5733         h = mtod(m, struct ip *);
5734
5735         off = h->ip_hl << 2;
5736         if (off < (int)sizeof(struct ip)) {
5737                 action = PF_DROP;
5738                 REASON_SET(&reason, PFRES_SHORT);
5739                 log = 1;
5740                 goto done;
5741         }
5742
5743         pd.src = (struct pf_addr *)&h->ip_src;
5744         pd.dst = (struct pf_addr *)&h->ip_dst;
5745         pd.sport = pd.dport = NULL;
5746         pd.ip_sum = &h->ip_sum;
5747         pd.proto_sum = NULL;
5748         pd.proto = h->ip_p;
5749         pd.dir = dir;
5750         pd.sidx = (dir == PF_IN) ? 0 : 1;
5751         pd.didx = (dir == PF_IN) ? 1 : 0;
5752         pd.af = AF_INET;
5753         pd.tos = h->ip_tos;
5754         pd.tot_len = ntohs(h->ip_len);
5755
5756         /* handle fragments that didn't get reassembled by normalization */
5757         if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
5758                 action = pf_test_fragment(&r, dir, kif, m, h,
5759                     &pd, &a, &ruleset);
5760                 goto done;
5761         }
5762
5763         switch (h->ip_p) {
5764
5765         case IPPROTO_TCP: {
5766                 struct tcphdr   th;
5767
5768                 pd.hdr.tcp = &th;
5769                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
5770                     &action, &reason, AF_INET)) {
5771                         log = action != PF_PASS;
5772                         goto done;
5773                 }
5774                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
5775                 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
5776                         pqid = 1;
5777                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
5778                 if (action == PF_DROP)
5779                         goto done;
5780                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
5781                     &reason);
5782                 if (action == PF_PASS) {
5783                         if (pfsync_update_state_ptr != NULL)
5784                                 pfsync_update_state_ptr(s);
5785                         r = s->rule.ptr;
5786                         a = s->anchor.ptr;
5787                         log = s->log;
5788                 } else if (s == NULL)
5789                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5790                             &a, &ruleset, inp);
5791                 break;
5792         }
5793
5794         case IPPROTO_UDP: {
5795                 struct udphdr   uh;
5796
5797                 pd.hdr.udp = &uh;
5798                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
5799                     &action, &reason, AF_INET)) {
5800                         log = action != PF_PASS;
5801                         goto done;
5802                 }
5803                 if (uh.uh_dport == 0 ||
5804                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
5805                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
5806                         action = PF_DROP;
5807                         REASON_SET(&reason, PFRES_SHORT);
5808                         goto done;
5809                 }
5810                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
5811                 if (action == PF_PASS) {
5812                         if (pfsync_update_state_ptr != NULL)
5813                                 pfsync_update_state_ptr(s);
5814                         r = s->rule.ptr;
5815                         a = s->anchor.ptr;
5816                         log = s->log;
5817                 } else if (s == NULL)
5818                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5819                             &a, &ruleset, inp);
5820                 break;
5821         }
5822
5823         case IPPROTO_ICMP: {
5824                 struct icmp     ih;
5825
5826                 pd.hdr.icmp = &ih;
5827                 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
5828                     &action, &reason, AF_INET)) {
5829                         log = action != PF_PASS;
5830                         goto done;
5831                 }
5832                 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
5833                     &reason);
5834                 if (action == PF_PASS) {
5835                         if (pfsync_update_state_ptr != NULL)
5836                                 pfsync_update_state_ptr(s);
5837                         r = s->rule.ptr;
5838                         a = s->anchor.ptr;
5839                         log = s->log;
5840                 } else if (s == NULL)
5841                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5842                             &a, &ruleset, inp);
5843                 break;
5844         }
5845
5846 #ifdef INET6
5847         case IPPROTO_ICMPV6: {
5848                 action = PF_DROP;
5849                 DPFPRINTF(PF_DEBUG_MISC,
5850                     ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
5851                 goto done;
5852         }
5853 #endif
5854
5855         default:
5856                 action = pf_test_state_other(&s, dir, kif, m, &pd);
5857                 if (action == PF_PASS) {
5858                         if (pfsync_update_state_ptr != NULL)
5859                                 pfsync_update_state_ptr(s);
5860                         r = s->rule.ptr;
5861                         a = s->anchor.ptr;
5862                         log = s->log;
5863                 } else if (s == NULL)
5864                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5865                             &a, &ruleset, inp);
5866                 break;
5867         }
5868
5869 done:
5870         PF_RULES_RUNLOCK();
5871         if (action == PF_PASS && h->ip_hl > 5 &&
5872             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
5873                 action = PF_DROP;
5874                 REASON_SET(&reason, PFRES_IPOPTIONS);
5875                 log = 1;
5876                 DPFPRINTF(PF_DEBUG_MISC,
5877                     ("pf: dropping packet with ip options\n"));
5878         }
5879
5880         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
5881                 action = PF_DROP;
5882                 REASON_SET(&reason, PFRES_MEMORY);
5883         }
5884         if (r->rtableid >= 0)
5885                 M_SETFIB(m, r->rtableid);
5886
5887 #ifdef ALTQ
5888         if (action == PF_PASS && r->qid) {
5889                 if (pd.pf_mtag == NULL &&
5890                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5891                         action = PF_DROP;
5892                         REASON_SET(&reason, PFRES_MEMORY);
5893                 }
5894                 if (pqid || (pd.tos & IPTOS_LOWDELAY))
5895                         pd.pf_mtag->qid = r->pqid;
5896                 else
5897                         pd.pf_mtag->qid = r->qid;
5898                 /* add hints for ecn */
5899                 pd.pf_mtag->hdr = h;
5900
5901         }
5902 #endif /* ALTQ */
5903
5904         /*
5905          * connections redirected to loopback should not match sockets
5906          * bound specifically to loopback due to security implications,
5907          * see tcp_input() and in_pcblookup_listen().
5908          */
5909         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
5910             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
5911             (s->nat_rule.ptr->action == PF_RDR ||
5912             s->nat_rule.ptr->action == PF_BINAT) &&
5913             (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
5914                 m->m_flags |= M_SKIP_FIREWALL;
5915
5916         if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
5917             !PACKET_LOOPED(&pd)) {
5918
5919                 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
5920                     sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
5921                 if (ipfwtag != NULL) {
5922                         ((struct ipfw_rule_ref *)(ipfwtag+1))->info =
5923                             ntohs(r->divert.port);
5924                         ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
5925
5926                         if (s)
5927                                 PF_STATE_UNLOCK(s);
5928
5929                         m_tag_prepend(m, ipfwtag);
5930                         if (m->m_flags & M_FASTFWD_OURS) {
5931                                 if (pd.pf_mtag == NULL &&
5932                                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5933                                         action = PF_DROP;
5934                                         REASON_SET(&reason, PFRES_MEMORY);
5935                                         log = 1;
5936                                         DPFPRINTF(PF_DEBUG_MISC,
5937                                             ("pf: failed to allocate tag\n"));
5938                                 }
5939                                 pd.pf_mtag->flags |= PF_FASTFWD_OURS_PRESENT;
5940                                 m->m_flags &= ~M_FASTFWD_OURS;
5941                         }
5942                         ip_divert_ptr(*m0, dir ==  PF_IN ? DIR_IN : DIR_OUT);
5943                         *m0 = NULL;
5944
5945                         return (action);
5946                 } else {
5947                         /* XXX: ipfw has the same behaviour! */
5948                         action = PF_DROP;
5949                         REASON_SET(&reason, PFRES_MEMORY);
5950                         log = 1;
5951                         DPFPRINTF(PF_DEBUG_MISC,
5952                             ("pf: failed to allocate divert tag\n"));
5953                 }
5954         }
5955
5956         if (log) {
5957                 struct pf_rule *lr;
5958
5959                 if (s != NULL && s->nat_rule.ptr != NULL &&
5960                     s->nat_rule.ptr->log & PF_LOG_ALL)
5961                         lr = s->nat_rule.ptr;
5962                 else
5963                         lr = r;
5964                 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
5965                     (s == NULL));
5966         }
5967
5968         kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
5969         kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
5970
5971         if (action == PF_PASS || r->action == PF_DROP) {
5972                 dirndx = (dir == PF_OUT);
5973                 r->packets[dirndx]++;
5974                 r->bytes[dirndx] += pd.tot_len;
5975                 if (a != NULL) {
5976                         a->packets[dirndx]++;
5977                         a->bytes[dirndx] += pd.tot_len;
5978                 }
5979                 if (s != NULL) {
5980                         if (s->nat_rule.ptr != NULL) {
5981                                 s->nat_rule.ptr->packets[dirndx]++;
5982                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
5983                         }
5984                         if (s->src_node != NULL) {
5985                                 s->src_node->packets[dirndx]++;
5986                                 s->src_node->bytes[dirndx] += pd.tot_len;
5987                         }
5988                         if (s->nat_src_node != NULL) {
5989                                 s->nat_src_node->packets[dirndx]++;
5990                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
5991                         }
5992                         dirndx = (dir == s->direction) ? 0 : 1;
5993                         s->packets[dirndx]++;
5994                         s->bytes[dirndx] += pd.tot_len;
5995                 }
5996                 tr = r;
5997                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
5998                 if (nr != NULL && r == &V_pf_default_rule)
5999                         tr = nr;
6000                 if (tr->src.addr.type == PF_ADDR_TABLE)
6001                         pfr_update_stats(tr->src.addr.p.tbl,
6002                             (s == NULL) ? pd.src :
6003                             &s->key[(s->direction == PF_IN)]->
6004                                 addr[(s->direction == PF_OUT)],
6005                             pd.af, pd.tot_len, dir == PF_OUT,
6006                             r->action == PF_PASS, tr->src.neg);
6007                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6008                         pfr_update_stats(tr->dst.addr.p.tbl,
6009                             (s == NULL) ? pd.dst :
6010                             &s->key[(s->direction == PF_IN)]->
6011                                 addr[(s->direction == PF_IN)],
6012                             pd.af, pd.tot_len, dir == PF_OUT,
6013                             r->action == PF_PASS, tr->dst.neg);
6014         }
6015
6016         switch (action) {
6017         case PF_SYNPROXY_DROP:
6018                 m_freem(*m0);
6019         case PF_DEFER:
6020                 *m0 = NULL;
6021                 action = PF_PASS;
6022                 break;
6023         default:
6024                 /* pf_route() returns unlocked. */
6025                 if (r->rt) {
6026                         pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
6027                         return (action);
6028                 }
6029                 break;
6030         }
6031         if (s)
6032                 PF_STATE_UNLOCK(s);
6033
6034         return (action);
6035 }
6036 #endif /* INET */
6037
6038 #ifdef INET6
6039 int
6040 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6041 {
6042         struct pfi_kif          *kif;
6043         u_short                  action, reason = 0, log = 0;
6044         struct mbuf             *m = *m0, *n = NULL;
6045         struct ip6_hdr          *h = NULL;
6046         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6047         struct pf_state         *s = NULL;
6048         struct pf_ruleset       *ruleset = NULL;
6049         struct pf_pdesc          pd;
6050         int                      off, terminal = 0, dirndx, rh_cnt = 0;
6051
6052         M_ASSERTPKTHDR(m);
6053
6054         if (!V_pf_status.running)
6055                 return (PF_PASS);
6056
6057         memset(&pd, 0, sizeof(pd));
6058         pd.pf_mtag = pf_find_mtag(m);
6059
6060         if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6061                 return (PF_PASS);
6062
6063         kif = (struct pfi_kif *)ifp->if_pf_kif;
6064         if (kif == NULL) {
6065                 DPFPRINTF(PF_DEBUG_URGENT,
6066                     ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6067                 return (PF_DROP);
6068         }
6069         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6070                 return (PF_PASS);
6071
6072         if (m->m_flags & M_SKIP_FIREWALL)
6073                 return (PF_PASS);
6074
6075         PF_RULES_RLOCK();
6076
6077         /* We do IP header normalization and packet reassembly here */
6078         if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6079                 action = PF_DROP;
6080                 goto done;
6081         }
6082         m = *m0;        /* pf_normalize messes with m0 */
6083         h = mtod(m, struct ip6_hdr *);
6084
6085 #if 1
6086         /*
6087          * we do not support jumbogram yet.  if we keep going, zero ip6_plen
6088          * will do something bad, so drop the packet for now.
6089          */
6090         if (htons(h->ip6_plen) == 0) {
6091                 action = PF_DROP;
6092                 REASON_SET(&reason, PFRES_NORM);        /*XXX*/
6093                 goto done;
6094         }
6095 #endif
6096
6097         pd.src = (struct pf_addr *)&h->ip6_src;
6098         pd.dst = (struct pf_addr *)&h->ip6_dst;
6099         pd.sport = pd.dport = NULL;
6100         pd.ip_sum = NULL;
6101         pd.proto_sum = NULL;
6102         pd.dir = dir;
6103         pd.sidx = (dir == PF_IN) ? 0 : 1;
6104         pd.didx = (dir == PF_IN) ? 1 : 0;
6105         pd.af = AF_INET6;
6106         pd.tos = 0;
6107         pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6108
6109         off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6110         pd.proto = h->ip6_nxt;
6111         do {
6112                 switch (pd.proto) {
6113                 case IPPROTO_FRAGMENT:
6114                         action = pf_test_fragment(&r, dir, kif, m, h,
6115                             &pd, &a, &ruleset);
6116                         if (action == PF_DROP)
6117                                 REASON_SET(&reason, PFRES_FRAG);
6118                         goto done;
6119                 case IPPROTO_ROUTING: {
6120                         struct ip6_rthdr rthdr;
6121
6122                         if (rh_cnt++) {
6123                                 DPFPRINTF(PF_DEBUG_MISC,
6124                                     ("pf: IPv6 more than one rthdr\n"));
6125                                 action = PF_DROP;
6126                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6127                                 log = 1;
6128                                 goto done;
6129                         }
6130                         if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6131                             &reason, pd.af)) {
6132                                 DPFPRINTF(PF_DEBUG_MISC,
6133                                     ("pf: IPv6 short rthdr\n"));
6134                                 action = PF_DROP;
6135                                 REASON_SET(&reason, PFRES_SHORT);
6136                                 log = 1;
6137                                 goto done;
6138                         }
6139                         if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6140                                 DPFPRINTF(PF_DEBUG_MISC,
6141                                     ("pf: IPv6 rthdr0\n"));
6142                                 action = PF_DROP;
6143                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6144                                 log = 1;
6145                                 goto done;
6146                         }
6147                         /* FALLTHROUGH */
6148                 }
6149                 case IPPROTO_AH:
6150                 case IPPROTO_HOPOPTS:
6151                 case IPPROTO_DSTOPTS: {
6152                         /* get next header and header length */
6153                         struct ip6_ext  opt6;
6154
6155                         if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6156                             NULL, &reason, pd.af)) {
6157                                 DPFPRINTF(PF_DEBUG_MISC,
6158                                     ("pf: IPv6 short opt\n"));
6159                                 action = PF_DROP;
6160                                 log = 1;
6161                                 goto done;
6162                         }
6163                         if (pd.proto == IPPROTO_AH)
6164                                 off += (opt6.ip6e_len + 2) * 4;
6165                         else
6166                                 off += (opt6.ip6e_len + 1) * 8;
6167                         pd.proto = opt6.ip6e_nxt;
6168                         /* goto the next header */
6169                         break;
6170                 }
6171                 default:
6172                         terminal++;
6173                         break;
6174                 }
6175         } while (!terminal);
6176
6177         /* if there's no routing header, use unmodified mbuf for checksumming */
6178         if (!n)
6179                 n = m;
6180
6181         switch (pd.proto) {
6182
6183         case IPPROTO_TCP: {
6184                 struct tcphdr   th;
6185
6186                 pd.hdr.tcp = &th;
6187                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6188                     &action, &reason, AF_INET6)) {
6189                         log = action != PF_PASS;
6190                         goto done;
6191                 }
6192                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6193                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6194                 if (action == PF_DROP)
6195                         goto done;
6196                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6197                     &reason);
6198                 if (action == PF_PASS) {
6199                         if (pfsync_update_state_ptr != NULL)
6200                                 pfsync_update_state_ptr(s);
6201                         r = s->rule.ptr;
6202                         a = s->anchor.ptr;
6203                         log = s->log;
6204                 } else if (s == NULL)
6205                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6206                             &a, &ruleset, inp);
6207                 break;
6208         }
6209
6210         case IPPROTO_UDP: {
6211                 struct udphdr   uh;
6212
6213                 pd.hdr.udp = &uh;
6214                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6215                     &action, &reason, AF_INET6)) {
6216                         log = action != PF_PASS;
6217                         goto done;
6218                 }
6219                 if (uh.uh_dport == 0 ||
6220                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6221                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6222                         action = PF_DROP;
6223                         REASON_SET(&reason, PFRES_SHORT);
6224                         goto done;
6225                 }
6226                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6227                 if (action == PF_PASS) {
6228                         if (pfsync_update_state_ptr != NULL)
6229                                 pfsync_update_state_ptr(s);
6230                         r = s->rule.ptr;
6231                         a = s->anchor.ptr;
6232                         log = s->log;
6233                 } else if (s == NULL)
6234                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6235                             &a, &ruleset, inp);
6236                 break;
6237         }
6238
6239         case IPPROTO_ICMP: {
6240                 action = PF_DROP;
6241                 DPFPRINTF(PF_DEBUG_MISC,
6242                     ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6243                 goto done;
6244         }
6245
6246         case IPPROTO_ICMPV6: {
6247                 struct icmp6_hdr        ih;
6248
6249                 pd.hdr.icmp6 = &ih;
6250                 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6251                     &action, &reason, AF_INET6)) {
6252                         log = action != PF_PASS;
6253                         goto done;
6254                 }
6255                 action = pf_test_state_icmp(&s, dir, kif,
6256                     m, off, h, &pd, &reason);
6257                 if (action == PF_PASS) {
6258                         if (pfsync_update_state_ptr != NULL)
6259                                 pfsync_update_state_ptr(s);
6260                         r = s->rule.ptr;
6261                         a = s->anchor.ptr;
6262                         log = s->log;
6263                 } else if (s == NULL)
6264                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6265                             &a, &ruleset, inp);
6266                 break;
6267         }
6268
6269         default:
6270                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6271                 if (action == PF_PASS) {
6272                         if (pfsync_update_state_ptr != NULL)
6273                                 pfsync_update_state_ptr(s);
6274                         r = s->rule.ptr;
6275                         a = s->anchor.ptr;
6276                         log = s->log;
6277                 } else if (s == NULL)
6278                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6279                             &a, &ruleset, inp);
6280                 break;
6281         }
6282
6283 done:
6284         PF_RULES_RUNLOCK();
6285         if (n != m) {
6286                 m_freem(n);
6287                 n = NULL;
6288         }
6289
6290         /* handle dangerous IPv6 extension headers. */
6291         if (action == PF_PASS && rh_cnt &&
6292             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6293                 action = PF_DROP;
6294                 REASON_SET(&reason, PFRES_IPOPTIONS);
6295                 log = 1;
6296                 DPFPRINTF(PF_DEBUG_MISC,
6297                     ("pf: dropping packet with dangerous v6 headers\n"));
6298         }
6299
6300         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6301                 action = PF_DROP;
6302                 REASON_SET(&reason, PFRES_MEMORY);
6303         }
6304         if (r->rtableid >= 0)
6305                 M_SETFIB(m, r->rtableid);
6306
6307 #ifdef ALTQ
6308         if (action == PF_PASS && r->qid) {
6309                 if (pd.pf_mtag == NULL &&
6310                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6311                         action = PF_DROP;
6312                         REASON_SET(&reason, PFRES_MEMORY);
6313                 }
6314                 if (pd.tos & IPTOS_LOWDELAY)
6315                         pd.pf_mtag->qid = r->pqid;
6316                 else
6317                         pd.pf_mtag->qid = r->qid;
6318                 /* add hints for ecn */
6319                 pd.pf_mtag->hdr = h;
6320         }
6321 #endif /* ALTQ */
6322
6323         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6324             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6325             (s->nat_rule.ptr->action == PF_RDR ||
6326             s->nat_rule.ptr->action == PF_BINAT) &&
6327             IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6328                 m->m_flags |= M_SKIP_FIREWALL;
6329
6330         /* XXX: Anybody working on it?! */
6331         if (r->divert.port)
6332                 printf("pf: divert(9) is not supported for IPv6\n");
6333
6334         if (log) {
6335                 struct pf_rule *lr;
6336
6337                 if (s != NULL && s->nat_rule.ptr != NULL &&
6338                     s->nat_rule.ptr->log & PF_LOG_ALL)
6339                         lr = s->nat_rule.ptr;
6340                 else
6341                         lr = r;
6342                 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6343                     &pd, (s == NULL));
6344         }
6345
6346         kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6347         kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
6348
6349         if (action == PF_PASS || r->action == PF_DROP) {
6350                 dirndx = (dir == PF_OUT);
6351                 r->packets[dirndx]++;
6352                 r->bytes[dirndx] += pd.tot_len;
6353                 if (a != NULL) {
6354                         a->packets[dirndx]++;
6355                         a->bytes[dirndx] += pd.tot_len;
6356                 }
6357                 if (s != NULL) {
6358                         if (s->nat_rule.ptr != NULL) {
6359                                 s->nat_rule.ptr->packets[dirndx]++;
6360                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6361                         }
6362                         if (s->src_node != NULL) {
6363                                 s->src_node->packets[dirndx]++;
6364                                 s->src_node->bytes[dirndx] += pd.tot_len;
6365                         }
6366                         if (s->nat_src_node != NULL) {
6367                                 s->nat_src_node->packets[dirndx]++;
6368                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6369                         }
6370                         dirndx = (dir == s->direction) ? 0 : 1;
6371                         s->packets[dirndx]++;
6372                         s->bytes[dirndx] += pd.tot_len;
6373                 }
6374                 tr = r;
6375                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6376                 if (nr != NULL && r == &V_pf_default_rule)
6377                         tr = nr;
6378                 if (tr->src.addr.type == PF_ADDR_TABLE)
6379                         pfr_update_stats(tr->src.addr.p.tbl,
6380                             (s == NULL) ? pd.src :
6381                             &s->key[(s->direction == PF_IN)]->addr[0],
6382                             pd.af, pd.tot_len, dir == PF_OUT,
6383                             r->action == PF_PASS, tr->src.neg);
6384                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6385                         pfr_update_stats(tr->dst.addr.p.tbl,
6386                             (s == NULL) ? pd.dst :
6387                             &s->key[(s->direction == PF_IN)]->addr[1],
6388                             pd.af, pd.tot_len, dir == PF_OUT,
6389                             r->action == PF_PASS, tr->dst.neg);
6390         }
6391
6392         switch (action) {
6393         case PF_SYNPROXY_DROP:
6394                 m_freem(*m0);
6395         case PF_DEFER:
6396                 *m0 = NULL;
6397                 action = PF_PASS;
6398                 break;
6399         default:
6400                 /* pf_route6() returns unlocked. */
6401                 if (r->rt) {
6402                         pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
6403                         return (action);
6404                 }
6405                 break;
6406         }
6407
6408         if (s)
6409                 PF_STATE_UNLOCK(s);
6410
6411         return (action);
6412 }
6413 #endif /* INET6 */