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 mbuf *, struct pf_addr *, u_int16_t *,
 207                             u_int16_t *, u_int16_t *, struct pf_addr *,
 208                             u_int16_t, u_int8_t, sa_family_t);
 209 static int               pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
 210                             struct tcphdr *, struct pf_state_peer *);
 211 static void              pf_change_icmp(struct pf_addr *, u_int16_t *,
 212                             struct pf_addr *, struct pf_addr *, u_int16_t,
 213                             u_int16_t *, u_int16_t *, u_int16_t *,
 214                             u_int16_t *, u_int8_t, sa_family_t);
 215 static void              pf_send_tcp(struct mbuf *,
 216                             const struct pf_rule *, sa_family_t,
 217                             const struct pf_addr *, const struct pf_addr *,
 218                             u_int16_t, u_int16_t, u_int32_t, u_int32_t,
 219                             u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
 220                             u_int16_t, struct ifnet *);
 221 static void              pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
 222                             sa_family_t, struct pf_rule *);
 223 static void              pf_detach_state(struct pf_state *);
 224 static int               pf_state_key_attach(struct pf_state_key *,
 225                             struct pf_state_key *, struct pf_state *);
 226 static void              pf_state_key_detach(struct pf_state *, int);
 227 static int               pf_state_key_ctor(void *, int, void *, int);
 228 static u_int32_t         pf_tcp_iss(struct pf_pdesc *);
 229 static int               pf_test_rule(struct pf_rule **, struct pf_state **,
 230                             int, struct pfi_kif *, struct mbuf *, int,
 231                             struct pf_pdesc *, struct pf_rule **,
 232                             struct pf_ruleset **, struct inpcb *);
 233 static int               pf_create_state(struct pf_rule *, struct pf_rule *,
 234                             struct pf_rule *, struct pf_pdesc *,
 235                             struct pf_src_node *, struct pf_state_key *,
 236                             struct pf_state_key *, struct mbuf *, int,
 237                             u_int16_t, u_int16_t, int *, struct pfi_kif *,
 238                             struct pf_state **, int, u_int16_t, u_int16_t,
 239                             int);
 240 static int               pf_test_fragment(struct pf_rule **, int,
 241                             struct pfi_kif *, struct mbuf *, void *,
 242                             struct pf_pdesc *, struct pf_rule **,
 243                             struct pf_ruleset **);
 244 static int               pf_tcp_track_full(struct pf_state_peer *,
 245                             struct pf_state_peer *, struct pf_state **,
 246                             struct pfi_kif *, struct mbuf *, int,
 247                             struct pf_pdesc *, u_short *, int *);
 248 static int               pf_tcp_track_sloppy(struct pf_state_peer *,
 249                             struct pf_state_peer *, struct pf_state **,
 250                             struct pf_pdesc *, u_short *);
 251 static int               pf_test_state_tcp(struct pf_state **, int,
 252                             struct pfi_kif *, struct mbuf *, int,
 253                             void *, struct pf_pdesc *, u_short *);
 254 static int               pf_test_state_udp(struct pf_state **, int,
 255                             struct pfi_kif *, struct mbuf *, int,
 256                             void *, struct pf_pdesc *);
 257 static int               pf_test_state_icmp(struct pf_state **, int,
 258                             struct pfi_kif *, struct mbuf *, int,
 259                             void *, struct pf_pdesc *, u_short *);
 260 static int               pf_test_state_other(struct pf_state **, int,
 261                             struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
 262 static u_int8_t          pf_get_wscale(struct mbuf *, int, u_int16_t,
 263                             sa_family_t);
 264 static u_int16_t         pf_get_mss(struct mbuf *, int, u_int16_t,
 265                             sa_family_t);
 266 static u_int16_t         pf_calc_mss(struct pf_addr *, sa_family_t,
 267                                 int, u_int16_t);
 268 static int               pf_check_proto_cksum(struct mbuf *, int, int,
 269                             u_int8_t, sa_family_t);
 270 static void              pf_print_state_parts(struct pf_state *,
 271                             struct pf_state_key *, struct pf_state_key *);
 272 static int               pf_addr_wrap_neq(struct pf_addr_wrap *,
 273                             struct pf_addr_wrap *);
 274 static struct pf_state  *pf_find_state(struct pfi_kif *,
 275                             struct pf_state_key_cmp *, u_int);
 276 static int               pf_src_connlimit(struct pf_state **);
 277 static void              pf_overload_task(void *v, int pending);
 278 static int               pf_insert_src_node(struct pf_src_node **,
 279                             struct pf_rule *, struct pf_addr *, sa_family_t);
 280 static u_int             pf_purge_expired_states(u_int, int);
 281 static void              pf_purge_unlinked_rules(void);
 282 static int               pf_mtag_uminit(void *, int, int);
 283 static void              pf_mtag_free(struct m_tag *);
 284 #ifdef INET
 285 static void              pf_route(struct mbuf **, struct pf_rule *, int,
 286                             struct ifnet *, struct pf_state *,
 287                             struct pf_pdesc *);
 288 #endif /* INET */
 289 #ifdef INET6
 290 static void              pf_change_a6(struct pf_addr *, u_int16_t *,
 291                             struct pf_addr *, u_int8_t);
 292 static void              pf_route6(struct mbuf **, struct pf_rule *, int,
 293                             struct ifnet *, struct pf_state *,
 294                             struct pf_pdesc *);
 295 #endif /* INET6 */
 296
 297 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
 298
 299 VNET_DECLARE(int, pf_end_threads);
 300
 301 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
 302
 303 #define PACKET_LOOPED(pd)       ((pd)->pf_mtag &&                       \
 304                                  (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
 305
 306 #define STATE_LOOKUP(i, k, d, s, pd)                                    \
 307         do {                                                            \
 308                 (s) = pf_find_state((i), (k), (d));                     \
 309                 if ((s) == NULL)                                        \
 310                         return (PF_DROP);                               \
 311                 if (PACKET_LOOPED(pd))                                  \
 312                         return (PF_PASS);                               \
 313                 if ((d) == PF_OUT &&                                    \
 314                     (((s)->rule.ptr->rt == PF_ROUTETO &&                \
 315                     (s)->rule.ptr->direction == PF_OUT) ||              \
 316                     ((s)->rule.ptr->rt == PF_REPLYTO &&                 \
 317                     (s)->rule.ptr->direction == PF_IN)) &&              \
 318                     (s)->rt_kif != NULL &&                              \
 319                     (s)->rt_kif != (i))                                 \
 320                         return (PF_PASS);                               \
 321         } while (0)
 322
 323 #define BOUND_IFACE(r, k) \
 324         ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
 325
 326 #define STATE_INC_COUNTERS(s)                                           \
 327         do {                                                            \
 328                 counter_u64_add(s->rule.ptr->states_cur, 1);            \
 329                 counter_u64_add(s->rule.ptr->states_tot, 1);            \
 330                 if (s->anchor.ptr != NULL) {                            \
 331                         counter_u64_add(s->anchor.ptr->states_cur, 1);  \
 332                         counter_u64_add(s->anchor.ptr->states_tot, 1);  \
 333                 }                                                       \
 334                 if (s->nat_rule.ptr != NULL) {                          \
 335                         counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
 336                         counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
 337                 }                                                       \
 338         } while (0)
 339
 340 #define STATE_DEC_COUNTERS(s)                                           \
 341         do {                                                            \
 342                 if (s->nat_rule.ptr != NULL)                            \
 343                         counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
 344                 if (s->anchor.ptr != NULL)                              \
 345                         counter_u64_add(s->anchor.ptr->states_cur, -1); \
 346                 counter_u64_add(s->rule.ptr->states_cur, -1);           \
 347         } while (0)
 348
 349 static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
 350 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
 351 VNET_DEFINE(struct pf_idhash *, pf_idhash);
 352 VNET_DEFINE(struct pf_srchash *, pf_srchash);
 353
 354 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)");
 355
 356 u_long  pf_hashmask;
 357 u_long  pf_srchashmask;
 358 static u_long   pf_hashsize;
 359 static u_long   pf_srchashsize;
 360
 361 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
 362     &pf_hashsize, 0, "Size of pf(4) states hashtable");
 363 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
 364     &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
 365
 366 VNET_DEFINE(void *, pf_swi_cookie);
 367
 368 VNET_DEFINE(uint32_t, pf_hashseed);
 369 #define V_pf_hashseed   VNET(pf_hashseed)
 370
 371 int
 372 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
 373 {
 374
 375         switch (af) {
 376 #ifdef INET
 377         case AF_INET:
 378                 if (a->addr32[0] > b->addr32[0])
 379                         return (1);
 380                 if (a->addr32[0] < b->addr32[0])
 381                         return (-1);
 382                 break;
 383 #endif /* INET */
 384 #ifdef INET6
 385         case AF_INET6:
 386                 if (a->addr32[3] > b->addr32[3])
 387                         return (1);
 388                 if (a->addr32[3] < b->addr32[3])
 389                         return (-1);
 390                 if (a->addr32[2] > b->addr32[2])
 391                         return (1);
 392                 if (a->addr32[2] < b->addr32[2])
 393                         return (-1);
 394                 if (a->addr32[1] > b->addr32[1])
 395                         return (1);
 396                 if (a->addr32[1] < b->addr32[1])
 397                         return (-1);
 398                 if (a->addr32[0] > b->addr32[0])
 399                         return (1);
 400                 if (a->addr32[0] < b->addr32[0])
 401                         return (-1);
 402                 break;
 403 #endif /* INET6 */
 404         default:
 405                 panic("%s: unknown address family %u", __func__, af);
 406         }
 407         return (0);
 408 }
 409
 410 static __inline uint32_t
 411 pf_hashkey(struct pf_state_key *sk)
 412 {
 413         uint32_t h;
 414
 415         h = murmur3_aligned_32((uint32_t *)sk,
 416                                sizeof(struct pf_state_key_cmp),
 417                                V_pf_hashseed);
 418
 419         return (h & pf_hashmask);
 420 }
 421
 422 static __inline uint32_t
 423 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
 424 {
 425         uint32_t h;
 426
 427         switch (af) {
 428         case AF_INET:
 429                 h = murmur3_aligned_32((uint32_t *)&addr->v4,
 430                                        sizeof(addr->v4), V_pf_hashseed);
 431                 break;
 432         case AF_INET6:
 433                 h = murmur3_aligned_32((uint32_t *)&addr->v6,
 434                                        sizeof(addr->v6), V_pf_hashseed);
 435                 break;
 436         default:
 437                 panic("%s: unknown address family %u", __func__, af);
 438         }
 439
 440         return (h & pf_srchashmask);
 441 }
 442
 443 #ifdef INET6
 444 void
 445 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
 446 {
 447         switch (af) {
 448 #ifdef INET
 449         case AF_INET:
 450                 dst->addr32[0] = src->addr32[0];
 451                 break;
 452 #endif /* INET */
 453         case AF_INET6:
 454                 dst->addr32[0] = src->addr32[0];
 455                 dst->addr32[1] = src->addr32[1];
 456                 dst->addr32[2] = src->addr32[2];
 457                 dst->addr32[3] = src->addr32[3];
 458                 break;
 459         }
 460 }
 461 #endif /* INET6 */
 462
 463 static void
 464 pf_init_threshold(struct pf_threshold *threshold,
 465     u_int32_t limit, u_int32_t seconds)
 466 {
 467         threshold->limit = limit * PF_THRESHOLD_MULT;
 468         threshold->seconds = seconds;
 469         threshold->count = 0;
 470         threshold->last = time_uptime;
 471 }
 472
 473 static void
 474 pf_add_threshold(struct pf_threshold *threshold)
 475 {
 476         u_int32_t t = time_uptime, diff = t - threshold->last;
 477
 478         if (diff >= threshold->seconds)
 479                 threshold->count = 0;
 480         else
 481                 threshold->count -= threshold->count * diff /
 482                     threshold->seconds;
 483         threshold->count += PF_THRESHOLD_MULT;
 484         threshold->last = t;
 485 }
 486
 487 static int
 488 pf_check_threshold(struct pf_threshold *threshold)
 489 {
 490         return (threshold->count > threshold->limit);
 491 }
 492
 493 static int
 494 pf_src_connlimit(struct pf_state **state)
 495 {
 496         struct pf_overload_entry *pfoe;
 497         int bad = 0;
 498
 499         PF_STATE_LOCK_ASSERT(*state);
 500
 501         (*state)->src_node->conn++;
 502         (*state)->src.tcp_est = 1;
 503         pf_add_threshold(&(*state)->src_node->conn_rate);
 504
 505         if ((*state)->rule.ptr->max_src_conn &&
 506             (*state)->rule.ptr->max_src_conn <
 507             (*state)->src_node->conn) {
 508                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
 509                 bad++;
 510         }
 511
 512         if ((*state)->rule.ptr->max_src_conn_rate.limit &&
 513             pf_check_threshold(&(*state)->src_node->conn_rate)) {
 514                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
 515                 bad++;
 516         }
 517
 518         if (!bad)
 519                 return (0);
 520
 521         /* Kill this state. */
 522         (*state)->timeout = PFTM_PURGE;
 523         (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
 524
 525         if ((*state)->rule.ptr->overload_tbl == NULL)
 526                 return (1);
 527
 528         /* Schedule overloading and flushing task. */
 529         pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
 530         if (pfoe == NULL)
 531                 return (1);     /* too bad :( */
 532
 533         bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
 534         pfoe->af = (*state)->key[PF_SK_WIRE]->af;
 535         pfoe->rule = (*state)->rule.ptr;
 536         pfoe->dir = (*state)->direction;
 537         PF_OVERLOADQ_LOCK();
 538         SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
 539         PF_OVERLOADQ_UNLOCK();
 540         taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
 541
 542         return (1);
 543 }
 544
 545 static void
 546 pf_overload_task(void *v, int pending)
 547 {
 548         struct pf_overload_head queue;
 549         struct pfr_addr p;
 550         struct pf_overload_entry *pfoe, *pfoe1;
 551         uint32_t killed = 0;
 552
 553         CURVNET_SET((struct vnet *)v);
 554
 555         PF_OVERLOADQ_LOCK();
 556         queue = V_pf_overloadqueue;
 557         SLIST_INIT(&V_pf_overloadqueue);
 558         PF_OVERLOADQ_UNLOCK();
 559
 560         bzero(&p, sizeof(p));
 561         SLIST_FOREACH(pfoe, &queue, next) {
 562                 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
 563                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
 564                         printf("%s: blocking address ", __func__);
 565                         pf_print_host(&pfoe->addr, 0, pfoe->af);
 566                         printf("\n");
 567                 }
 568
 569                 p.pfra_af = pfoe->af;
 570                 switch (pfoe->af) {
 571 #ifdef INET
 572                 case AF_INET:
 573                         p.pfra_net = 32;
 574                         p.pfra_ip4addr = pfoe->addr.v4;
 575                         break;
 576 #endif
 577 #ifdef INET6
 578                 case AF_INET6:
 579                         p.pfra_net = 128;
 580                         p.pfra_ip6addr = pfoe->addr.v6;
 581                         break;
 582 #endif
 583                 }
 584
 585                 PF_RULES_WLOCK();
 586                 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
 587                 PF_RULES_WUNLOCK();
 588         }
 589
 590         /*
 591          * Remove those entries, that don't need flushing.
 592          */
 593         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 594                 if (pfoe->rule->flush == 0) {
 595                         SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
 596                         free(pfoe, M_PFTEMP);
 597                 } else
 598                         counter_u64_add(
 599                             V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
 600
 601         /* If nothing to flush, return. */
 602         if (SLIST_EMPTY(&queue)) {
 603                 CURVNET_RESTORE();
 604                 return;
 605         }
 606
 607         for (int i = 0; i <= pf_hashmask; i++) {
 608                 struct pf_idhash *ih = &V_pf_idhash[i];
 609                 struct pf_state_key *sk;
 610                 struct pf_state *s;
 611
 612                 PF_HASHROW_LOCK(ih);
 613                 LIST_FOREACH(s, &ih->states, entry) {
 614                     sk = s->key[PF_SK_WIRE];
 615                     SLIST_FOREACH(pfoe, &queue, next)
 616                         if (sk->af == pfoe->af &&
 617                             ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
 618                             pfoe->rule == s->rule.ptr) &&
 619                             ((pfoe->dir == PF_OUT &&
 620                             PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
 621                             (pfoe->dir == PF_IN &&
 622                             PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
 623                                 s->timeout = PFTM_PURGE;
 624                                 s->src.state = s->dst.state = TCPS_CLOSED;
 625                                 killed++;
 626                         }
 627                 }
 628                 PF_HASHROW_UNLOCK(ih);
 629         }
 630         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 631                 free(pfoe, M_PFTEMP);
 632         if (V_pf_status.debug >= PF_DEBUG_MISC)
 633                 printf("%s: %u states killed", __func__, killed);
 634
 635         CURVNET_RESTORE();
 636 }
 637
 638 /*
 639  * Can return locked on failure, so that we can consistently
 640  * allocate and insert a new one.
 641  */
 642 struct pf_src_node *
 643 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
 644         int returnlocked)
 645 {
 646         struct pf_srchash *sh;
 647         struct pf_src_node *n;
 648
 649         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
 650
 651         sh = &V_pf_srchash[pf_hashsrc(src, af)];
 652         PF_HASHROW_LOCK(sh);
 653         LIST_FOREACH(n, &sh->nodes, entry)
 654                 if (n->rule.ptr == rule && n->af == af &&
 655                     ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
 656                     (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
 657                         break;
 658         if (n != NULL) {
 659                 n->states++;
 660                 PF_HASHROW_UNLOCK(sh);
 661         } else if (returnlocked == 0)
 662                 PF_HASHROW_UNLOCK(sh);
 663
 664         return (n);
 665 }
 666
 667 static int
 668 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
 669     struct pf_addr *src, sa_family_t af)
 670 {
 671
 672         KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
 673             rule->rpool.opts & PF_POOL_STICKYADDR),
 674             ("%s for non-tracking rule %p", __func__, rule));
 675
 676         if (*sn == NULL)
 677                 *sn = pf_find_src_node(src, rule, af, 1);
 678
 679         if (*sn == NULL) {
 680                 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
 681
 682                 PF_HASHROW_ASSERT(sh);
 683
 684                 if (!rule->max_src_nodes ||
 685                     counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
 686                         (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
 687                 else
 688                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
 689                             1);
 690                 if ((*sn) == NULL) {
 691                         PF_HASHROW_UNLOCK(sh);
 692                         return (-1);
 693                 }
 694
 695                 pf_init_threshold(&(*sn)->conn_rate,
 696                     rule->max_src_conn_rate.limit,
 697                     rule->max_src_conn_rate.seconds);
 698
 699                 (*sn)->af = af;
 700                 (*sn)->rule.ptr = rule;
 701                 PF_ACPY(&(*sn)->addr, src, af);
 702                 LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
 703                 (*sn)->creation = time_uptime;
 704                 (*sn)->ruletype = rule->action;
 705                 (*sn)->states = 1;
 706                 if ((*sn)->rule.ptr != NULL)
 707                         counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
 708                 PF_HASHROW_UNLOCK(sh);
 709                 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
 710         } else {
 711                 if (rule->max_src_states &&
 712                     (*sn)->states >= rule->max_src_states) {
 713                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
 714                             1);
 715                         return (-1);
 716                 }
 717         }
 718         return (0);
 719 }
 720
 721 void
 722 pf_unlink_src_node(struct pf_src_node *src)
 723 {
 724
 725         PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]);
 726         LIST_REMOVE(src, entry);
 727         if (src->rule.ptr)
 728                 counter_u64_add(src->rule.ptr->src_nodes, -1);
 729 }
 730
 731 u_int
 732 pf_free_src_nodes(struct pf_src_node_list *head)
 733 {
 734         struct pf_src_node *sn, *tmp;
 735         u_int count = 0;
 736
 737         LIST_FOREACH_SAFE(sn, head, entry, tmp) {
 738                 uma_zfree(V_pf_sources_z, sn);
 739                 count++;
 740         }
 741
 742         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
 743
 744         return (count);
 745 }
 746
 747 void
 748 pf_mtag_initialize()
 749 {
 750
 751         pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
 752             sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
 753             UMA_ALIGN_PTR, 0);
 754 }
 755
 756 /* Per-vnet data storage structures initialization. */
 757 void
 758 pf_initialize()
 759 {
 760         struct pf_keyhash       *kh;
 761         struct pf_idhash        *ih;
 762         struct pf_srchash       *sh;
 763         u_int i;
 764
 765         TUNABLE_ULONG_FETCH("net.pf.states_hashsize", &pf_hashsize);
 766         if (pf_hashsize == 0 || !powerof2(pf_hashsize))
 767                 pf_hashsize = PF_HASHSIZ;
 768         TUNABLE_ULONG_FETCH("net.pf.source_nodes_hashsize", &pf_srchashsize);
 769         if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
 770                 pf_srchashsize = PF_HASHSIZ / 4;
 771
 772         V_pf_hashseed = arc4random();
 773
 774         /* States and state keys storage. */
 775         V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
 776             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 777         V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
 778         uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
 779         uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
 780
 781         V_pf_state_key_z = uma_zcreate("pf state keys",
 782             sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
 783             UMA_ALIGN_PTR, 0);
 784         V_pf_keyhash = malloc(pf_hashsize * sizeof(struct pf_keyhash),
 785             M_PFHASH, M_WAITOK | M_ZERO);
 786         V_pf_idhash = malloc(pf_hashsize * sizeof(struct pf_idhash),
 787             M_PFHASH, M_WAITOK | M_ZERO);
 788         pf_hashmask = pf_hashsize - 1;
 789         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 790             i++, kh++, ih++) {
 791                 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
 792                 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
 793         }
 794
 795         /* Source nodes. */
 796         V_pf_sources_z = uma_zcreate("pf source nodes",
 797             sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
 798             0);
 799         V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
 800         uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
 801         uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
 802         V_pf_srchash = malloc(pf_srchashsize * sizeof(struct pf_srchash),
 803           M_PFHASH, M_WAITOK|M_ZERO);
 804         pf_srchashmask = pf_srchashsize - 1;
 805         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
 806                 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
 807
 808         /* ALTQ */
 809         TAILQ_INIT(&V_pf_altqs[0]);
 810         TAILQ_INIT(&V_pf_altqs[1]);
 811         TAILQ_INIT(&V_pf_pabuf);
 812         V_pf_altqs_active = &V_pf_altqs[0];
 813         V_pf_altqs_inactive = &V_pf_altqs[1];
 814
 815
 816         /* Send & overload+flush queues. */
 817         STAILQ_INIT(&V_pf_sendqueue);
 818         SLIST_INIT(&V_pf_overloadqueue);
 819         TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
 820         mtx_init(&pf_sendqueue_mtx, "pf send queue", NULL, MTX_DEF);
 821         mtx_init(&pf_overloadqueue_mtx, "pf overload/flush queue", NULL,
 822             MTX_DEF);
 823
 824         /* Unlinked, but may be referenced rules. */
 825         TAILQ_INIT(&V_pf_unlinked_rules);
 826         mtx_init(&pf_unlnkdrules_mtx, "pf unlinked rules", NULL, MTX_DEF);
 827 }
 828
 829 void
 830 pf_mtag_cleanup()
 831 {
 832
 833         uma_zdestroy(pf_mtag_z);
 834 }
 835
 836 void
 837 pf_cleanup()
 838 {
 839         struct pf_keyhash       *kh;
 840         struct pf_idhash        *ih;
 841         struct pf_srchash       *sh;
 842         struct pf_send_entry    *pfse, *next;
 843         u_int i;
 844
 845         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 846             i++, kh++, ih++) {
 847                 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
 848                     __func__));
 849                 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
 850                     __func__));
 851                 mtx_destroy(&kh->lock);
 852                 mtx_destroy(&ih->lock);
 853         }
 854         free(V_pf_keyhash, M_PFHASH);
 855         free(V_pf_idhash, M_PFHASH);
 856
 857         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
 858                 KASSERT(LIST_EMPTY(&sh->nodes),
 859                     ("%s: source node hash not empty", __func__));
 860                 mtx_destroy(&sh->lock);
 861         }
 862         free(V_pf_srchash, M_PFHASH);
 863
 864         STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
 865                 m_freem(pfse->pfse_m);
 866                 free(pfse, M_PFTEMP);
 867         }
 868
 869         mtx_destroy(&pf_sendqueue_mtx);
 870         mtx_destroy(&pf_overloadqueue_mtx);
 871         mtx_destroy(&pf_unlnkdrules_mtx);
 872
 873         uma_zdestroy(V_pf_sources_z);
 874         uma_zdestroy(V_pf_state_z);
 875         uma_zdestroy(V_pf_state_key_z);
 876 }
 877
 878 static int
 879 pf_mtag_uminit(void *mem, int size, int how)
 880 {
 881         struct m_tag *t;
 882
 883         t = (struct m_tag *)mem;
 884         t->m_tag_cookie = MTAG_ABI_COMPAT;
 885         t->m_tag_id = PACKET_TAG_PF;
 886         t->m_tag_len = sizeof(struct pf_mtag);
 887         t->m_tag_free = pf_mtag_free;
 888
 889         return (0);
 890 }
 891
 892 static void
 893 pf_mtag_free(struct m_tag *t)
 894 {
 895
 896         uma_zfree(pf_mtag_z, t);
 897 }
 898
 899 struct pf_mtag *
 900 pf_get_mtag(struct mbuf *m)
 901 {
 902         struct m_tag *mtag;
 903
 904         if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
 905                 return ((struct pf_mtag *)(mtag + 1));
 906
 907         mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
 908         if (mtag == NULL)
 909                 return (NULL);
 910         bzero(mtag + 1, sizeof(struct pf_mtag));
 911         m_tag_prepend(m, mtag);
 912
 913         return ((struct pf_mtag *)(mtag + 1));
 914 }
 915
 916 static int
 917 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
 918     struct pf_state *s)
 919 {
 920         struct pf_keyhash       *khs, *khw, *kh;
 921         struct pf_state_key     *sk, *cur;
 922         struct pf_state         *si, *olds = NULL;
 923         int idx;
 924
 925         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
 926         KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
 927         KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
 928
 929         /*
 930          * We need to lock hash slots of both keys. To avoid deadlock
 931          * we always lock the slot with lower address first. Unlock order
 932          * isn't important.
 933          *
 934          * We also need to lock ID hash slot before dropping key
 935          * locks. On success we return with ID hash slot locked.
 936          */
 937
 938         if (skw == sks) {
 939                 khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
 940                 PF_HASHROW_LOCK(khs);
 941         } else {
 942                 khs = &V_pf_keyhash[pf_hashkey(sks)];
 943                 khw = &V_pf_keyhash[pf_hashkey(skw)];
 944                 if (khs == khw) {
 945                         PF_HASHROW_LOCK(khs);
 946                 } else if (khs < khw) {
 947                         PF_HASHROW_LOCK(khs);
 948                         PF_HASHROW_LOCK(khw);
 949                 } else {
 950                         PF_HASHROW_LOCK(khw);
 951                         PF_HASHROW_LOCK(khs);
 952                 }
 953         }
 954
 955 #define KEYS_UNLOCK()   do {                    \
 956         if (khs != khw) {                       \
 957                 PF_HASHROW_UNLOCK(khs);         \
 958                 PF_HASHROW_UNLOCK(khw);         \
 959         } else                                  \
 960                 PF_HASHROW_UNLOCK(khs);         \
 961 } while (0)
 962
 963         /*
 964          * First run: start with wire key.
 965          */
 966         sk = skw;
 967         kh = khw;
 968         idx = PF_SK_WIRE;
 969
 970 keyattach:
 971         LIST_FOREACH(cur, &kh->keys, entry)
 972                 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
 973                         break;
 974
 975         if (cur != NULL) {
 976                 /* Key exists. Check for same kif, if none, add to key. */
 977                 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
 978                         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
 979
 980                         PF_HASHROW_LOCK(ih);
 981                         if (si->kif == s->kif &&
 982                             si->direction == s->direction) {
 983                                 if (sk->proto == IPPROTO_TCP &&
 984                                     si->src.state >= TCPS_FIN_WAIT_2 &&
 985                                     si->dst.state >= TCPS_FIN_WAIT_2) {
 986                                         /*
 987                                          * New state matches an old >FIN_WAIT_2
 988                                          * state. We can't drop key hash locks,
 989                                          * thus we can't unlink it properly.
 990                                          *
 991                                          * As a workaround we drop it into
 992                                          * TCPS_CLOSED state, schedule purge
 993                                          * ASAP and push it into the very end
 994                                          * of the slot TAILQ, so that it won't
 995                                          * conflict with our new state.
 996                                          */
 997                                         si->src.state = si->dst.state =
 998                                             TCPS_CLOSED;
 999                                         si->timeout = PFTM_PURGE;
1000                                         olds = si;
1001                                 } else {
1002                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
1003                                                 printf("pf: %s key attach "
1004                                                     "failed on %s: ",
1005                                                     (idx == PF_SK_WIRE) ?
1006                                                     "wire" : "stack",
1007                                                     s->kif->pfik_name);
1008                                                 pf_print_state_parts(s,
1009                                                     (idx == PF_SK_WIRE) ?
1010                                                     sk : NULL,
1011                                                     (idx == PF_SK_STACK) ?
1012                                                     sk : NULL);
1013                                                 printf(", existing: ");
1014                                                 pf_print_state_parts(si,
1015                                                     (idx == PF_SK_WIRE) ?
1016                                                     sk : NULL,
1017                                                     (idx == PF_SK_STACK) ?
1018                                                     sk : NULL);
1019                                                 printf("\n");
1020                                         }
1021                                         PF_HASHROW_UNLOCK(ih);
1022                                         KEYS_UNLOCK();
1023                                         uma_zfree(V_pf_state_key_z, sk);
1024                                         if (idx == PF_SK_STACK)
1025                                                 pf_detach_state(s);
1026                                         return (EEXIST); /* collision! */
1027                                 }
1028                         }
1029                         PF_HASHROW_UNLOCK(ih);
1030                 }
1031                 uma_zfree(V_pf_state_key_z, sk);
1032                 s->key[idx] = cur;
1033         } else {
1034                 LIST_INSERT_HEAD(&kh->keys, sk, entry);
1035                 s->key[idx] = sk;
1036         }
1037
1038 stateattach:
1039         /* List is sorted, if-bound states before floating. */
1040         if (s->kif == V_pfi_all)
1041                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1042         else
1043                 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1044
1045         if (olds) {
1046                 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1047                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1048                     key_list[idx]);
1049                 olds = NULL;
1050         }
1051
1052         /*
1053          * Attach done. See how should we (or should not?)
1054          * attach a second key.
1055          */
1056         if (sks == skw) {
1057                 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1058                 idx = PF_SK_STACK;
1059                 sks = NULL;
1060                 goto stateattach;
1061         } else if (sks != NULL) {
1062                 /*
1063                  * Continue attaching with stack key.
1064                  */
1065                 sk = sks;
1066                 kh = khs;
1067                 idx = PF_SK_STACK;
1068                 sks = NULL;
1069                 goto keyattach;
1070         }
1071
1072         PF_STATE_LOCK(s);
1073         KEYS_UNLOCK();
1074
1075         KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1076             ("%s failure", __func__));
1077
1078         return (0);
1079 #undef  KEYS_UNLOCK
1080 }
1081
1082 static void
1083 pf_detach_state(struct pf_state *s)
1084 {
1085         struct pf_state_key *sks = s->key[PF_SK_STACK];
1086         struct pf_keyhash *kh;
1087
1088         if (sks != NULL) {
1089                 kh = &V_pf_keyhash[pf_hashkey(sks)];
1090                 PF_HASHROW_LOCK(kh);
1091                 if (s->key[PF_SK_STACK] != NULL)
1092                         pf_state_key_detach(s, PF_SK_STACK);
1093                 /*
1094                  * If both point to same key, then we are done.
1095                  */
1096                 if (sks == s->key[PF_SK_WIRE]) {
1097                         pf_state_key_detach(s, PF_SK_WIRE);
1098                         PF_HASHROW_UNLOCK(kh);
1099                         return;
1100                 }
1101                 PF_HASHROW_UNLOCK(kh);
1102         }
1103
1104         if (s->key[PF_SK_WIRE] != NULL) {
1105                 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1106                 PF_HASHROW_LOCK(kh);
1107                 if (s->key[PF_SK_WIRE] != NULL)
1108                         pf_state_key_detach(s, PF_SK_WIRE);
1109                 PF_HASHROW_UNLOCK(kh);
1110         }
1111 }
1112
1113 static void
1114 pf_state_key_detach(struct pf_state *s, int idx)
1115 {
1116         struct pf_state_key *sk = s->key[idx];
1117 #ifdef INVARIANTS
1118         struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1119
1120         PF_HASHROW_ASSERT(kh);
1121 #endif
1122         TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1123         s->key[idx] = NULL;
1124
1125         if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1126                 LIST_REMOVE(sk, entry);
1127                 uma_zfree(V_pf_state_key_z, sk);
1128         }
1129 }
1130
1131 static int
1132 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1133 {
1134         struct pf_state_key *sk = mem;
1135
1136         bzero(sk, sizeof(struct pf_state_key_cmp));
1137         TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1138         TAILQ_INIT(&sk->states[PF_SK_STACK]);
1139
1140         return (0);
1141 }
1142
1143 struct pf_state_key *
1144 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1145         struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1146 {
1147         struct pf_state_key *sk;
1148
1149         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1150         if (sk == NULL)
1151                 return (NULL);
1152
1153         PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1154         PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1155         sk->port[pd->sidx] = sport;
1156         sk->port[pd->didx] = dport;
1157         sk->proto = pd->proto;
1158         sk->af = pd->af;
1159
1160         return (sk);
1161 }
1162
1163 struct pf_state_key *
1164 pf_state_key_clone(struct pf_state_key *orig)
1165 {
1166         struct pf_state_key *sk;
1167
1168         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1169         if (sk == NULL)
1170                 return (NULL);
1171
1172         bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1173
1174         return (sk);
1175 }
1176
1177 int
1178 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1179     struct pf_state_key *sks, struct pf_state *s)
1180 {
1181         struct pf_idhash *ih;
1182         struct pf_state *cur;
1183         int error;
1184
1185         KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1186             ("%s: sks not pristine", __func__));
1187         KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1188             ("%s: skw not pristine", __func__));
1189         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1190
1191         s->kif = kif;
1192
1193         if (s->id == 0 && s->creatorid == 0) {
1194                 /* XXX: should be atomic, but probability of collision low */
1195                 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1196                         V_pf_stateid[curcpu] = 1;
1197                 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1198                 s->id = htobe64(s->id);
1199                 s->creatorid = V_pf_status.hostid;
1200         }
1201
1202         /* Returns with ID locked on success. */
1203         if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1204                 return (error);
1205
1206         ih = &V_pf_idhash[PF_IDHASH(s)];
1207         PF_HASHROW_ASSERT(ih);
1208         LIST_FOREACH(cur, &ih->states, entry)
1209                 if (cur->id == s->id && cur->creatorid == s->creatorid)
1210                         break;
1211
1212         if (cur != NULL) {
1213                 PF_HASHROW_UNLOCK(ih);
1214                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1215                         printf("pf: state ID collision: "
1216                             "id: %016llx creatorid: %08x\n",
1217                             (unsigned long long)be64toh(s->id),
1218                             ntohl(s->creatorid));
1219                 }
1220                 pf_detach_state(s);
1221                 return (EEXIST);
1222         }
1223         LIST_INSERT_HEAD(&ih->states, s, entry);
1224         /* One for keys, one for ID hash. */
1225         refcount_init(&s->refs, 2);
1226
1227         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1228         if (pfsync_insert_state_ptr != NULL)
1229                 pfsync_insert_state_ptr(s);
1230
1231         /* Returns locked. */
1232         return (0);
1233 }
1234
1235 /*
1236  * Find state by ID: returns with locked row on success.
1237  */
1238 struct pf_state *
1239 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1240 {
1241         struct pf_idhash *ih;
1242         struct pf_state *s;
1243
1244         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1245
1246         ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1247
1248         PF_HASHROW_LOCK(ih);
1249         LIST_FOREACH(s, &ih->states, entry)
1250                 if (s->id == id && s->creatorid == creatorid)
1251                         break;
1252
1253         if (s == NULL)
1254                 PF_HASHROW_UNLOCK(ih);
1255
1256         return (s);
1257 }
1258
1259 /*
1260  * Find state by key.
1261  * Returns with ID hash slot locked on success.
1262  */
1263 static struct pf_state *
1264 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1265 {
1266         struct pf_keyhash       *kh;
1267         struct pf_state_key     *sk;
1268         struct pf_state         *s;
1269         int idx;
1270
1271         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1272
1273         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1274
1275         PF_HASHROW_LOCK(kh);
1276         LIST_FOREACH(sk, &kh->keys, entry)
1277                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1278                         break;
1279         if (sk == NULL) {
1280                 PF_HASHROW_UNLOCK(kh);
1281                 return (NULL);
1282         }
1283
1284         idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1285
1286         /* List is sorted, if-bound states before floating ones. */
1287         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1288                 if (s->kif == V_pfi_all || s->kif == kif) {
1289                         PF_STATE_LOCK(s);
1290                         PF_HASHROW_UNLOCK(kh);
1291                         if (s->timeout >= PFTM_MAX) {
1292                                 /*
1293                                  * State is either being processed by
1294                                  * pf_unlink_state() in an other thread, or
1295                                  * is scheduled for immediate expiry.
1296                                  */
1297                                 PF_STATE_UNLOCK(s);
1298                                 return (NULL);
1299                         }
1300                         return (s);
1301                 }
1302         PF_HASHROW_UNLOCK(kh);
1303
1304         return (NULL);
1305 }
1306
1307 struct pf_state *
1308 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1309 {
1310         struct pf_keyhash       *kh;
1311         struct pf_state_key     *sk;
1312         struct pf_state         *s, *ret = NULL;
1313         int                      idx, inout = 0;
1314
1315         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1316
1317         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1318
1319         PF_HASHROW_LOCK(kh);
1320         LIST_FOREACH(sk, &kh->keys, entry)
1321                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1322                         break;
1323         if (sk == NULL) {
1324                 PF_HASHROW_UNLOCK(kh);
1325                 return (NULL);
1326         }
1327         switch (dir) {
1328         case PF_IN:
1329                 idx = PF_SK_WIRE;
1330                 break;
1331         case PF_OUT:
1332                 idx = PF_SK_STACK;
1333                 break;
1334         case PF_INOUT:
1335                 idx = PF_SK_WIRE;
1336                 inout = 1;
1337                 break;
1338         default:
1339                 panic("%s: dir %u", __func__, dir);
1340         }
1341 second_run:
1342         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1343                 if (more == NULL) {
1344                         PF_HASHROW_UNLOCK(kh);
1345                         return (s);
1346                 }
1347
1348                 if (ret)
1349                         (*more)++;
1350                 else
1351                         ret = s;
1352         }
1353         if (inout == 1) {
1354                 inout = 0;
1355                 idx = PF_SK_STACK;
1356                 goto second_run;
1357         }
1358         PF_HASHROW_UNLOCK(kh);
1359
1360         return (ret);
1361 }
1362
1363 /* END state table stuff */
1364
1365 static void
1366 pf_send(struct pf_send_entry *pfse)
1367 {
1368
1369         PF_SENDQ_LOCK();
1370         STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1371         PF_SENDQ_UNLOCK();
1372         swi_sched(V_pf_swi_cookie, 0);
1373 }
1374
1375 void
1376 pf_intr(void *v)
1377 {
1378         struct pf_send_head queue;
1379         struct pf_send_entry *pfse, *next;
1380
1381         CURVNET_SET((struct vnet *)v);
1382
1383         PF_SENDQ_LOCK();
1384         queue = V_pf_sendqueue;
1385         STAILQ_INIT(&V_pf_sendqueue);
1386         PF_SENDQ_UNLOCK();
1387
1388         STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1389                 switch (pfse->pfse_type) {
1390 #ifdef INET
1391                 case PFSE_IP:
1392                         ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1393                         break;
1394                 case PFSE_ICMP:
1395                         icmp_error(pfse->pfse_m, pfse->pfse_icmp_type,
1396                             pfse->pfse_icmp_code, 0, pfse->pfse_icmp_mtu);
1397                         break;
1398 #endif /* INET */
1399 #ifdef INET6
1400                 case PFSE_IP6:
1401                         ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1402                             NULL);
1403                         break;
1404                 case PFSE_ICMP6:
1405                         icmp6_error(pfse->pfse_m, pfse->pfse_icmp_type,
1406                             pfse->pfse_icmp_code, pfse->pfse_icmp_mtu);
1407                         break;
1408 #endif /* INET6 */
1409                 default:
1410                         panic("%s: unknown type", __func__);
1411                 }
1412                 free(pfse, M_PFTEMP);
1413         }
1414         CURVNET_RESTORE();
1415 }
1416
1417 void
1418 pf_purge_thread(void *v)
1419 {
1420         u_int idx = 0;
1421
1422         CURVNET_SET((struct vnet *)v);
1423
1424         for (;;) {
1425                 PF_RULES_RLOCK();
1426                 rw_sleep(pf_purge_thread, &pf_rules_lock, 0, "pftm", hz / 10);
1427
1428                 if (V_pf_end_threads) {
1429                         /*
1430                          * To cleanse up all kifs and rules we need
1431                          * two runs: first one clears reference flags,
1432                          * then pf_purge_expired_states() doesn't
1433                          * raise them, and then second run frees.
1434                          */
1435                         PF_RULES_RUNLOCK();
1436                         pf_purge_unlinked_rules();
1437                         pfi_kif_purge();
1438
1439                         /*
1440                          * Now purge everything.
1441                          */
1442                         pf_purge_expired_states(0, pf_hashmask);
1443                         pf_purge_expired_fragments();
1444                         pf_purge_expired_src_nodes();
1445
1446                         /*
1447                          * Now all kifs & rules should be unreferenced,
1448                          * thus should be successfully freed.
1449                          */
1450                         pf_purge_unlinked_rules();
1451                         pfi_kif_purge();
1452
1453                         /*
1454                          * Announce success and exit.
1455                          */
1456                         PF_RULES_RLOCK();
1457                         V_pf_end_threads++;
1458                         PF_RULES_RUNLOCK();
1459                         wakeup(pf_purge_thread);
1460                         kproc_exit(0);
1461                 }
1462                 PF_RULES_RUNLOCK();
1463
1464                 /* Process 1/interval fraction of the state table every run. */
1465                 idx = pf_purge_expired_states(idx, pf_hashmask /
1466                             (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1467
1468                 /* Purge other expired types every PFTM_INTERVAL seconds. */
1469                 if (idx == 0) {
1470                         /*
1471                          * Order is important:
1472                          * - states and src nodes reference rules
1473                          * - states and rules reference kifs
1474                          */
1475                         pf_purge_expired_fragments();
1476                         pf_purge_expired_src_nodes();
1477                         pf_purge_unlinked_rules();
1478                         pfi_kif_purge();
1479                 }
1480         }
1481         /* not reached */
1482         CURVNET_RESTORE();
1483 }
1484
1485 u_int32_t
1486 pf_state_expires(const struct pf_state *state)
1487 {
1488         u_int32_t       timeout;
1489         u_int32_t       start;
1490         u_int32_t       end;
1491         u_int32_t       states;
1492
1493         /* handle all PFTM_* > PFTM_MAX here */
1494         if (state->timeout == PFTM_PURGE)
1495                 return (time_uptime);
1496         KASSERT(state->timeout != PFTM_UNLINKED,
1497             ("pf_state_expires: timeout == PFTM_UNLINKED"));
1498         KASSERT((state->timeout < PFTM_MAX),
1499             ("pf_state_expires: timeout > PFTM_MAX"));
1500         timeout = state->rule.ptr->timeout[state->timeout];
1501         if (!timeout)
1502                 timeout = V_pf_default_rule.timeout[state->timeout];
1503         start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1504         if (start) {
1505                 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1506                 states = counter_u64_fetch(state->rule.ptr->states_cur);
1507         } else {
1508                 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1509                 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1510                 states = V_pf_status.states;
1511         }
1512         if (end && states > start && start < end) {
1513                 if (states < end)
1514                         return (state->expire + timeout * (end - states) /
1515                             (end - start));
1516                 else
1517                         return (time_uptime);
1518         }
1519         return (state->expire + timeout);
1520 }
1521
1522 void
1523 pf_purge_expired_src_nodes()
1524 {
1525         struct pf_src_node_list  freelist;
1526         struct pf_srchash       *sh;
1527         struct pf_src_node      *cur, *next;
1528         int i;
1529
1530         LIST_INIT(&freelist);
1531         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1532             PF_HASHROW_LOCK(sh);
1533             LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1534                 if (cur->states == 0 && cur->expire <= time_uptime) {
1535                         pf_unlink_src_node(cur);
1536                         LIST_INSERT_HEAD(&freelist, cur, entry);
1537                 } else if (cur->rule.ptr != NULL)
1538                         cur->rule.ptr->rule_flag |= PFRULE_REFS;
1539             PF_HASHROW_UNLOCK(sh);
1540         }
1541
1542         pf_free_src_nodes(&freelist);
1543
1544         V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1545 }
1546
1547 static void
1548 pf_src_tree_remove_state(struct pf_state *s)
1549 {
1550         struct pf_src_node *sn;
1551         struct pf_srchash *sh;
1552         uint32_t timeout;
1553
1554         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
1555             s->rule.ptr->timeout[PFTM_SRC_NODE] :
1556             V_pf_default_rule.timeout[PFTM_SRC_NODE];
1557
1558         if (s->src_node != NULL) {
1559                 sn = s->src_node;
1560                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1561                 PF_HASHROW_LOCK(sh);
1562                 if (s->src.tcp_est)
1563                         --sn->conn;
1564                 if (--sn->states == 0)
1565                         sn->expire = time_uptime + timeout;
1566                 PF_HASHROW_UNLOCK(sh);
1567         }
1568         if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1569                 sn = s->nat_src_node;
1570                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1571                 PF_HASHROW_LOCK(sh);
1572                 if (--sn->states == 0)
1573                         sn->expire = time_uptime + timeout;
1574                 PF_HASHROW_UNLOCK(sh);
1575         }
1576         s->src_node = s->nat_src_node = NULL;
1577 }
1578
1579 /*
1580  * Unlink and potentilly free a state. Function may be
1581  * called with ID hash row locked, but always returns
1582  * unlocked, since it needs to go through key hash locking.
1583  */
1584 int
1585 pf_unlink_state(struct pf_state *s, u_int flags)
1586 {
1587         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1588
1589         if ((flags & PF_ENTER_LOCKED) == 0)
1590                 PF_HASHROW_LOCK(ih);
1591         else
1592                 PF_HASHROW_ASSERT(ih);
1593
1594         if (s->timeout == PFTM_UNLINKED) {
1595                 /*
1596                  * State is being processed
1597                  * by pf_unlink_state() in
1598                  * an other thread.
1599                  */
1600                 PF_HASHROW_UNLOCK(ih);
1601                 return (0);     /* XXXGL: undefined actually */
1602         }
1603
1604         if (s->src.state == PF_TCPS_PROXY_DST) {
1605                 /* XXX wire key the right one? */
1606                 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1607                     &s->key[PF_SK_WIRE]->addr[1],
1608                     &s->key[PF_SK_WIRE]->addr[0],
1609                     s->key[PF_SK_WIRE]->port[1],
1610                     s->key[PF_SK_WIRE]->port[0],
1611                     s->src.seqhi, s->src.seqlo + 1,
1612                     TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1613         }
1614
1615         LIST_REMOVE(s, entry);
1616         pf_src_tree_remove_state(s);
1617
1618         if (pfsync_delete_state_ptr != NULL)
1619                 pfsync_delete_state_ptr(s);
1620
1621         STATE_DEC_COUNTERS(s);
1622
1623         s->timeout = PFTM_UNLINKED;
1624
1625         PF_HASHROW_UNLOCK(ih);
1626
1627         pf_detach_state(s);
1628         refcount_release(&s->refs);
1629
1630         return (pf_release_state(s));
1631 }
1632
1633 void
1634 pf_free_state(struct pf_state *cur)
1635 {
1636
1637         KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1638         KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1639             cur->timeout));
1640
1641         pf_normalize_tcp_cleanup(cur);
1642         uma_zfree(V_pf_state_z, cur);
1643         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
1644 }
1645
1646 /*
1647  * Called only from pf_purge_thread(), thus serialized.
1648  */
1649 static u_int
1650 pf_purge_expired_states(u_int i, int maxcheck)
1651 {
1652         struct pf_idhash *ih;
1653         struct pf_state *s;
1654
1655         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1656
1657         /*
1658          * Go through hash and unlink states that expire now.
1659          */
1660         while (maxcheck > 0) {
1661
1662                 ih = &V_pf_idhash[i];
1663 relock:
1664                 PF_HASHROW_LOCK(ih);
1665                 LIST_FOREACH(s, &ih->states, entry) {
1666                         if (pf_state_expires(s) <= time_uptime) {
1667                                 V_pf_status.states -=
1668                                     pf_unlink_state(s, PF_ENTER_LOCKED);
1669                                 goto relock;
1670                         }
1671                         s->rule.ptr->rule_flag |= PFRULE_REFS;
1672                         if (s->nat_rule.ptr != NULL)
1673                                 s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1674                         if (s->anchor.ptr != NULL)
1675                                 s->anchor.ptr->rule_flag |= PFRULE_REFS;
1676                         s->kif->pfik_flags |= PFI_IFLAG_REFS;
1677                         if (s->rt_kif)
1678                                 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1679                 }
1680                 PF_HASHROW_UNLOCK(ih);
1681
1682                 /* Return when we hit end of hash. */
1683                 if (++i > pf_hashmask) {
1684                         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1685                         return (0);
1686                 }
1687
1688                 maxcheck--;
1689         }
1690
1691         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1692
1693         return (i);
1694 }
1695
1696 static void
1697 pf_purge_unlinked_rules()
1698 {
1699         struct pf_rulequeue tmpq;
1700         struct pf_rule *r, *r1;
1701
1702         /*
1703          * If we have overloading task pending, then we'd
1704          * better skip purging this time. There is a tiny
1705          * probability that overloading task references
1706          * an already unlinked rule.
1707          */
1708         PF_OVERLOADQ_LOCK();
1709         if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1710                 PF_OVERLOADQ_UNLOCK();
1711                 return;
1712         }
1713         PF_OVERLOADQ_UNLOCK();
1714
1715         /*
1716          * Do naive mark-and-sweep garbage collecting of old rules.
1717          * Reference flag is raised by pf_purge_expired_states()
1718          * and pf_purge_expired_src_nodes().
1719          *
1720          * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1721          * use a temporary queue.
1722          */
1723         TAILQ_INIT(&tmpq);
1724         PF_UNLNKDRULES_LOCK();
1725         TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1726                 if (!(r->rule_flag & PFRULE_REFS)) {
1727                         TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1728                         TAILQ_INSERT_TAIL(&tmpq, r, entries);
1729                 } else
1730                         r->rule_flag &= ~PFRULE_REFS;
1731         }
1732         PF_UNLNKDRULES_UNLOCK();
1733
1734         if (!TAILQ_EMPTY(&tmpq)) {
1735                 PF_RULES_WLOCK();
1736                 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1737                         TAILQ_REMOVE(&tmpq, r, entries);
1738                         pf_free_rule(r);
1739                 }
1740                 PF_RULES_WUNLOCK();
1741         }
1742 }
1743
1744 void
1745 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1746 {
1747         switch (af) {
1748 #ifdef INET
1749         case AF_INET: {
1750                 u_int32_t a = ntohl(addr->addr32[0]);
1751                 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1752                     (a>>8)&255, a&255);
1753                 if (p) {
1754                         p = ntohs(p);
1755                         printf(":%u", p);
1756                 }
1757                 break;
1758         }
1759 #endif /* INET */
1760 #ifdef INET6
1761         case AF_INET6: {
1762                 u_int16_t b;
1763                 u_int8_t i, curstart, curend, maxstart, maxend;
1764                 curstart = curend = maxstart = maxend = 255;
1765                 for (i = 0; i < 8; i++) {
1766                         if (!addr->addr16[i]) {
1767                                 if (curstart == 255)
1768                                         curstart = i;
1769                                 curend = i;
1770                         } else {
1771                                 if ((curend - curstart) >
1772                                     (maxend - maxstart)) {
1773                                         maxstart = curstart;
1774                                         maxend = curend;
1775                                 }
1776                                 curstart = curend = 255;
1777                         }
1778                 }
1779                 if ((curend - curstart) >
1780                     (maxend - maxstart)) {
1781                         maxstart = curstart;
1782                         maxend = curend;
1783                 }
1784                 for (i = 0; i < 8; i++) {
1785                         if (i >= maxstart && i <= maxend) {
1786                                 if (i == 0)
1787                                         printf(":");
1788                                 if (i == maxend)
1789                                         printf(":");
1790                         } else {
1791                                 b = ntohs(addr->addr16[i]);
1792                                 printf("%x", b);
1793                                 if (i < 7)
1794                                         printf(":");
1795                         }
1796                 }
1797                 if (p) {
1798                         p = ntohs(p);
1799                         printf("[%u]", p);
1800                 }
1801                 break;
1802         }
1803 #endif /* INET6 */
1804         }
1805 }
1806
1807 void
1808 pf_print_state(struct pf_state *s)
1809 {
1810         pf_print_state_parts(s, NULL, NULL);
1811 }
1812
1813 static void
1814 pf_print_state_parts(struct pf_state *s,
1815     struct pf_state_key *skwp, struct pf_state_key *sksp)
1816 {
1817         struct pf_state_key *skw, *sks;
1818         u_int8_t proto, dir;
1819
1820         /* Do our best to fill these, but they're skipped if NULL */
1821         skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1822         sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1823         proto = skw ? skw->proto : (sks ? sks->proto : 0);
1824         dir = s ? s->direction : 0;
1825
1826         switch (proto) {
1827         case IPPROTO_IPV4:
1828                 printf("IPv4");
1829                 break;
1830         case IPPROTO_IPV6:
1831                 printf("IPv6");
1832                 break;
1833         case IPPROTO_TCP:
1834                 printf("TCP");
1835                 break;
1836         case IPPROTO_UDP:
1837                 printf("UDP");
1838                 break;
1839         case IPPROTO_ICMP:
1840                 printf("ICMP");
1841                 break;
1842         case IPPROTO_ICMPV6:
1843                 printf("ICMPv6");
1844                 break;
1845         default:
1846                 printf("%u", skw->proto);
1847                 break;
1848         }
1849         switch (dir) {
1850         case PF_IN:
1851                 printf(" in");
1852                 break;
1853         case PF_OUT:
1854                 printf(" out");
1855                 break;
1856         }
1857         if (skw) {
1858                 printf(" wire: ");
1859                 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1860                 printf(" ");
1861                 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1862         }
1863         if (sks) {
1864                 printf(" stack: ");
1865                 if (sks != skw) {
1866                         pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1867                         printf(" ");
1868                         pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1869                 } else
1870                         printf("-");
1871         }
1872         if (s) {
1873                 if (proto == IPPROTO_TCP) {
1874                         printf(" [lo=%u high=%u win=%u modulator=%u",
1875                             s->src.seqlo, s->src.seqhi,
1876                             s->src.max_win, s->src.seqdiff);
1877                         if (s->src.wscale && s->dst.wscale)
1878                                 printf(" wscale=%u",
1879                                     s->src.wscale & PF_WSCALE_MASK);
1880                         printf("]");
1881                         printf(" [lo=%u high=%u win=%u modulator=%u",
1882                             s->dst.seqlo, s->dst.seqhi,
1883                             s->dst.max_win, s->dst.seqdiff);
1884                         if (s->src.wscale && s->dst.wscale)
1885                                 printf(" wscale=%u",
1886                                 s->dst.wscale & PF_WSCALE_MASK);
1887                         printf("]");
1888                 }
1889                 printf(" %u:%u", s->src.state, s->dst.state);
1890         }
1891 }
1892
1893 void
1894 pf_print_flags(u_int8_t f)
1895 {
1896         if (f)
1897                 printf(" ");
1898         if (f & TH_FIN)
1899                 printf("F");
1900         if (f & TH_SYN)
1901                 printf("S");
1902         if (f & TH_RST)
1903                 printf("R");
1904         if (f & TH_PUSH)
1905                 printf("P");
1906         if (f & TH_ACK)
1907                 printf("A");
1908         if (f & TH_URG)
1909                 printf("U");
1910         if (f & TH_ECE)
1911                 printf("E");
1912         if (f & TH_CWR)
1913                 printf("W");
1914 }
1915
1916 #define PF_SET_SKIP_STEPS(i)                                    \
1917         do {                                                    \
1918                 while (head[i] != cur) {                        \
1919                         head[i]->skip[i].ptr = cur;             \
1920                         head[i] = TAILQ_NEXT(head[i], entries); \
1921                 }                                               \
1922         } while (0)
1923
1924 void
1925 pf_calc_skip_steps(struct pf_rulequeue *rules)
1926 {
1927         struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1928         int i;
1929
1930         cur = TAILQ_FIRST(rules);
1931         prev = cur;
1932         for (i = 0; i < PF_SKIP_COUNT; ++i)
1933                 head[i] = cur;
1934         while (cur != NULL) {
1935
1936                 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1937                         PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1938                 if (cur->direction != prev->direction)
1939                         PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1940                 if (cur->af != prev->af)
1941                         PF_SET_SKIP_STEPS(PF_SKIP_AF);
1942                 if (cur->proto != prev->proto)
1943                         PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
1944                 if (cur->src.neg != prev->src.neg ||
1945                     pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
1946                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
1947                 if (cur->src.port[0] != prev->src.port[0] ||
1948                     cur->src.port[1] != prev->src.port[1] ||
1949                     cur->src.port_op != prev->src.port_op)
1950                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
1951                 if (cur->dst.neg != prev->dst.neg ||
1952                     pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
1953                         PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
1954                 if (cur->dst.port[0] != prev->dst.port[0] ||
1955                     cur->dst.port[1] != prev->dst.port[1] ||
1956                     cur->dst.port_op != prev->dst.port_op)
1957                         PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
1958
1959                 prev = cur;
1960                 cur = TAILQ_NEXT(cur, entries);
1961         }
1962         for (i = 0; i < PF_SKIP_COUNT; ++i)
1963                 PF_SET_SKIP_STEPS(i);
1964 }
1965
1966 static int
1967 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
1968 {
1969         if (aw1->type != aw2->type)
1970                 return (1);
1971         switch (aw1->type) {
1972         case PF_ADDR_ADDRMASK:
1973         case PF_ADDR_RANGE:
1974                 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, 0))
1975                         return (1);
1976                 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, 0))
1977                         return (1);
1978                 return (0);
1979         case PF_ADDR_DYNIFTL:
1980                 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
1981         case PF_ADDR_NOROUTE:
1982         case PF_ADDR_URPFFAILED:
1983                 return (0);
1984         case PF_ADDR_TABLE:
1985                 return (aw1->p.tbl != aw2->p.tbl);
1986         default:
1987                 printf("invalid address type: %d\n", aw1->type);
1988                 return (1);
1989         }
1990 }
1991
1992 /**
1993  * Checksum updates are a little complicated because the checksum in the TCP/UDP
1994  * header isn't always a full checksum. In some cases (i.e. output) it's a
1995  * pseudo-header checksum, which is a partial checksum over src/dst IP
1996  * addresses, protocol number and length.
1997  *
1998  * That means we have the following cases:
1999  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2000  *      checksums, we need to update the checksum whenever we change anything.
2001  *  * Output (i.e. the checksum is a pseudo-header checksum):
2002  *      x The field being updated is src/dst address or affects the length of
2003  *      the packet. We need to update the pseudo-header checksum (note that this
2004  *      checksum is not ones' complement).
2005  *      x Some other field is being modified (e.g. src/dst port numbers): We
2006  *      don't have to update anything.
2007  **/
2008 u_int16_t
2009 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2010 {
2011         u_int32_t       l;
2012
2013         if (udp && !cksum)
2014                 return (0x0000);
2015         l = cksum + old - new;
2016         l = (l >> 16) + (l & 65535);
2017         l = l & 65535;
2018         if (udp && !l)
2019                 return (0xFFFF);
2020         return (l);
2021 }
2022
2023 u_int16_t
2024 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2025         u_int16_t new, u_int8_t udp)
2026 {
2027         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2028                 return (cksum);
2029
2030         return (pf_cksum_fixup(cksum, old, new, udp));
2031 }
2032
2033 static void
2034 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2035         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2036         sa_family_t af)
2037 {
2038         struct pf_addr  ao;
2039         u_int16_t       po = *p;
2040
2041         PF_ACPY(&ao, a, af);
2042         PF_ACPY(a, an, af);
2043
2044         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2045                 *pc = ~*pc;
2046
2047         *p = pn;
2048
2049         switch (af) {
2050 #ifdef INET
2051         case AF_INET:
2052                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2053                     ao.addr16[0], an->addr16[0], 0),
2054                     ao.addr16[1], an->addr16[1], 0);
2055                 *p = pn;
2056
2057                 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2058                     ao.addr16[0], an->addr16[0], u),
2059                     ao.addr16[1], an->addr16[1], u);
2060
2061                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2062                 break;
2063 #endif /* INET */
2064 #ifdef INET6
2065         case AF_INET6:
2066                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2067                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2068                     pf_cksum_fixup(pf_cksum_fixup(*pc,
2069                     ao.addr16[0], an->addr16[0], u),
2070                     ao.addr16[1], an->addr16[1], u),
2071                     ao.addr16[2], an->addr16[2], u),
2072                     ao.addr16[3], an->addr16[3], u),
2073                     ao.addr16[4], an->addr16[4], u),
2074                     ao.addr16[5], an->addr16[5], u),
2075                     ao.addr16[6], an->addr16[6], u),
2076                     ao.addr16[7], an->addr16[7], u);
2077
2078                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2079                 break;
2080 #endif /* INET6 */
2081         }
2082
2083         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2084             CSUM_DELAY_DATA_IPV6)) {
2085                 *pc = ~*pc;
2086                 if (! *pc)
2087                         *pc = 0xffff;
2088         }
2089 }
2090
2091 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2092 void
2093 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2094 {
2095         u_int32_t       ao;
2096
2097         memcpy(&ao, a, sizeof(ao));
2098         memcpy(a, &an, sizeof(u_int32_t));
2099         *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2100             ao % 65536, an % 65536, u);
2101 }
2102
2103 void
2104 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2105 {
2106         u_int32_t       ao;
2107
2108         memcpy(&ao, a, sizeof(ao));
2109         memcpy(a, &an, sizeof(u_int32_t));
2110
2111         *c = pf_proto_cksum_fixup(m,
2112             pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2113             ao % 65536, an % 65536, udp);
2114 }
2115
2116 #ifdef INET6
2117 static void
2118 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2119 {
2120         struct pf_addr  ao;
2121
2122         PF_ACPY(&ao, a, AF_INET6);
2123         PF_ACPY(a, an, AF_INET6);
2124
2125         *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2126             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2127             pf_cksum_fixup(pf_cksum_fixup(*c,
2128             ao.addr16[0], an->addr16[0], u),
2129             ao.addr16[1], an->addr16[1], u),
2130             ao.addr16[2], an->addr16[2], u),
2131             ao.addr16[3], an->addr16[3], u),
2132             ao.addr16[4], an->addr16[4], u),
2133             ao.addr16[5], an->addr16[5], u),
2134             ao.addr16[6], an->addr16[6], u),
2135             ao.addr16[7], an->addr16[7], u);
2136 }
2137 #endif /* INET6 */
2138
2139 static void
2140 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2141     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2142     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2143 {
2144         struct pf_addr  oia, ooa;
2145
2146         PF_ACPY(&oia, ia, af);
2147         if (oa)
2148                 PF_ACPY(&ooa, oa, af);
2149
2150         /* Change inner protocol port, fix inner protocol checksum. */
2151         if (ip != NULL) {
2152                 u_int16_t       oip = *ip;
2153                 u_int32_t       opc;
2154
2155                 if (pc != NULL)
2156                         opc = *pc;
2157                 *ip = np;
2158                 if (pc != NULL)
2159                         *pc = pf_cksum_fixup(*pc, oip, *ip, u);
2160                 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2161                 if (pc != NULL)
2162                         *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2163         }
2164         /* Change inner ip address, fix inner ip and icmp checksums. */
2165         PF_ACPY(ia, na, af);
2166         switch (af) {
2167 #ifdef INET
2168         case AF_INET: {
2169                 u_int32_t        oh2c = *h2c;
2170
2171                 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2172                     oia.addr16[0], ia->addr16[0], 0),
2173                     oia.addr16[1], ia->addr16[1], 0);
2174                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2175                     oia.addr16[0], ia->addr16[0], 0),
2176                     oia.addr16[1], ia->addr16[1], 0);
2177                 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2178                 break;
2179         }
2180 #endif /* INET */
2181 #ifdef INET6
2182         case AF_INET6:
2183                 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2184                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2185                     pf_cksum_fixup(pf_cksum_fixup(*ic,
2186                     oia.addr16[0], ia->addr16[0], u),
2187                     oia.addr16[1], ia->addr16[1], u),
2188                     oia.addr16[2], ia->addr16[2], u),
2189                     oia.addr16[3], ia->addr16[3], u),
2190                     oia.addr16[4], ia->addr16[4], u),
2191                     oia.addr16[5], ia->addr16[5], u),
2192                     oia.addr16[6], ia->addr16[6], u),
2193                     oia.addr16[7], ia->addr16[7], u);
2194                 break;
2195 #endif /* INET6 */
2196         }
2197         /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2198         if (oa) {
2199                 PF_ACPY(oa, na, af);
2200                 switch (af) {
2201 #ifdef INET
2202                 case AF_INET:
2203                         *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2204                             ooa.addr16[0], oa->addr16[0], 0),
2205                             ooa.addr16[1], oa->addr16[1], 0);
2206                         break;
2207 #endif /* INET */
2208 #ifdef INET6
2209                 case AF_INET6:
2210                         *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2211                             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2212                             pf_cksum_fixup(pf_cksum_fixup(*ic,
2213                             ooa.addr16[0], oa->addr16[0], u),
2214                             ooa.addr16[1], oa->addr16[1], u),
2215                             ooa.addr16[2], oa->addr16[2], u),
2216                             ooa.addr16[3], oa->addr16[3], u),
2217                             ooa.addr16[4], oa->addr16[4], u),
2218                             ooa.addr16[5], oa->addr16[5], u),
2219                             ooa.addr16[6], oa->addr16[6], u),
2220                             ooa.addr16[7], oa->addr16[7], u);
2221                         break;
2222 #endif /* INET6 */
2223                 }
2224         }
2225 }
2226
2227
2228 /*
2229  * Need to modulate the sequence numbers in the TCP SACK option
2230  * (credits to Krzysztof Pfaff for report and patch)
2231  */
2232 static int
2233 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2234     struct tcphdr *th, struct pf_state_peer *dst)
2235 {
2236         int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2237         u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2238         int copyback = 0, i, olen;
2239         struct sackblk sack;
2240
2241 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2242         if (hlen < TCPOLEN_SACKLEN ||
2243             !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2244                 return 0;
2245
2246         while (hlen >= TCPOLEN_SACKLEN) {
2247                 olen = opt[1];
2248                 switch (*opt) {
2249                 case TCPOPT_EOL:        /* FALLTHROUGH */
2250                 case TCPOPT_NOP:
2251                         opt++;
2252                         hlen--;
2253                         break;
2254                 case TCPOPT_SACK:
2255                         if (olen > hlen)
2256                                 olen = hlen;
2257                         if (olen >= TCPOLEN_SACKLEN) {
2258                                 for (i = 2; i + TCPOLEN_SACK <= olen;
2259                                     i += TCPOLEN_SACK) {
2260                                         memcpy(&sack, &opt[i], sizeof(sack));
2261                                         pf_change_proto_a(m, &sack.start, &th->th_sum,
2262                                             htonl(ntohl(sack.start) - dst->seqdiff), 0);
2263                                         pf_change_proto_a(m, &sack.end, &th->th_sum,
2264                                             htonl(ntohl(sack.end) - dst->seqdiff), 0);
2265                                         memcpy(&opt[i], &sack, sizeof(sack));
2266                                 }
2267                                 copyback = 1;
2268                         }
2269                         /* FALLTHROUGH */
2270                 default:
2271                         if (olen < 2)
2272                                 olen = 2;
2273                         hlen -= olen;
2274                         opt += olen;
2275                 }
2276         }
2277
2278         if (copyback)
2279                 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2280         return (copyback);
2281 }
2282
2283 static void
2284 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2285     const struct pf_addr *saddr, const struct pf_addr *daddr,
2286     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2287     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2288     u_int16_t rtag, struct ifnet *ifp)
2289 {
2290         struct pf_send_entry *pfse;
2291         struct mbuf     *m;
2292         int              len, tlen;
2293 #ifdef INET
2294         struct ip       *h = NULL;
2295 #endif /* INET */
2296 #ifdef INET6
2297         struct ip6_hdr  *h6 = NULL;
2298 #endif /* INET6 */
2299         struct tcphdr   *th;
2300         char            *opt;
2301         struct pf_mtag  *pf_mtag;
2302
2303         len = 0;
2304         th = NULL;
2305
2306         /* maximum segment size tcp option */
2307         tlen = sizeof(struct tcphdr);
2308         if (mss)
2309                 tlen += 4;
2310
2311         switch (af) {
2312 #ifdef INET
2313         case AF_INET:
2314                 len = sizeof(struct ip) + tlen;
2315                 break;
2316 #endif /* INET */
2317 #ifdef INET6
2318         case AF_INET6:
2319                 len = sizeof(struct ip6_hdr) + tlen;
2320                 break;
2321 #endif /* INET6 */
2322         default:
2323                 panic("%s: unsupported af %d", __func__, af);
2324         }
2325
2326         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2327         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2328         if (pfse == NULL)
2329                 return;
2330         m = m_gethdr(M_NOWAIT, MT_DATA);
2331         if (m == NULL) {
2332                 free(pfse, M_PFTEMP);
2333                 return;
2334         }
2335 #ifdef MAC
2336         mac_netinet_firewall_send(m);
2337 #endif
2338         if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2339                 free(pfse, M_PFTEMP);
2340                 m_freem(m);
2341                 return;
2342         }
2343         if (tag)
2344                 m->m_flags |= M_SKIP_FIREWALL;
2345         pf_mtag->tag = rtag;
2346
2347         if (r != NULL && r->rtableid >= 0)
2348                 M_SETFIB(m, r->rtableid);
2349
2350 #ifdef ALTQ
2351         if (r != NULL && r->qid) {
2352                 pf_mtag->qid = r->qid;
2353
2354                 /* add hints for ecn */
2355                 pf_mtag->hdr = mtod(m, struct ip *);
2356         }
2357 #endif /* ALTQ */
2358         m->m_data += max_linkhdr;
2359         m->m_pkthdr.len = m->m_len = len;
2360         m->m_pkthdr.rcvif = NULL;
2361         bzero(m->m_data, len);
2362         switch (af) {
2363 #ifdef INET
2364         case AF_INET:
2365                 h = mtod(m, struct ip *);
2366
2367                 /* IP header fields included in the TCP checksum */
2368                 h->ip_p = IPPROTO_TCP;
2369                 h->ip_len = htons(tlen);
2370                 h->ip_src.s_addr = saddr->v4.s_addr;
2371                 h->ip_dst.s_addr = daddr->v4.s_addr;
2372
2373                 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2374                 break;
2375 #endif /* INET */
2376 #ifdef INET6
2377         case AF_INET6:
2378                 h6 = mtod(m, struct ip6_hdr *);
2379
2380                 /* IP header fields included in the TCP checksum */
2381                 h6->ip6_nxt = IPPROTO_TCP;
2382                 h6->ip6_plen = htons(tlen);
2383                 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2384                 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2385
2386                 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2387                 break;
2388 #endif /* INET6 */
2389         }
2390
2391         /* TCP header */
2392         th->th_sport = sport;
2393         th->th_dport = dport;
2394         th->th_seq = htonl(seq);
2395         th->th_ack = htonl(ack);
2396         th->th_off = tlen >> 2;
2397         th->th_flags = flags;
2398         th->th_win = htons(win);
2399
2400         if (mss) {
2401                 opt = (char *)(th + 1);
2402                 opt[0] = TCPOPT_MAXSEG;
2403                 opt[1] = 4;
2404                 HTONS(mss);
2405                 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2406         }
2407
2408         switch (af) {
2409 #ifdef INET
2410         case AF_INET:
2411                 /* TCP checksum */
2412                 th->th_sum = in_cksum(m, len);
2413
2414                 /* Finish the IP header */
2415                 h->ip_v = 4;
2416                 h->ip_hl = sizeof(*h) >> 2;
2417                 h->ip_tos = IPTOS_LOWDELAY;
2418                 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2419                 h->ip_len = htons(len);
2420                 h->ip_ttl = ttl ? ttl : V_ip_defttl;
2421                 h->ip_sum = 0;
2422
2423                 pfse->pfse_type = PFSE_IP;
2424                 break;
2425 #endif /* INET */
2426 #ifdef INET6
2427         case AF_INET6:
2428                 /* TCP checksum */
2429                 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2430                     sizeof(struct ip6_hdr), tlen);
2431
2432                 h6->ip6_vfc |= IPV6_VERSION;
2433                 h6->ip6_hlim = IPV6_DEFHLIM;
2434
2435                 pfse->pfse_type = PFSE_IP6;
2436                 break;
2437 #endif /* INET6 */
2438         }
2439         pfse->pfse_m = m;
2440         pf_send(pfse);
2441 }
2442
2443 static void
2444 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2445     struct pf_rule *r)
2446 {
2447         struct pf_send_entry *pfse;
2448         struct mbuf *m0;
2449         struct pf_mtag *pf_mtag;
2450
2451         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2452         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2453         if (pfse == NULL)
2454                 return;
2455
2456         if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2457                 free(pfse, M_PFTEMP);
2458                 return;
2459         }
2460
2461         if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2462                 free(pfse, M_PFTEMP);
2463                 return;
2464         }
2465         /* XXX: revisit */
2466         m0->m_flags |= M_SKIP_FIREWALL;
2467
2468         if (r->rtableid >= 0)
2469                 M_SETFIB(m0, r->rtableid);
2470
2471 #ifdef ALTQ
2472         if (r->qid) {
2473                 pf_mtag->qid = r->qid;
2474                 /* add hints for ecn */
2475                 pf_mtag->hdr = mtod(m0, struct ip *);
2476         }
2477 #endif /* ALTQ */
2478
2479         switch (af) {
2480 #ifdef INET
2481         case AF_INET:
2482                 pfse->pfse_type = PFSE_ICMP;
2483                 break;
2484 #endif /* INET */
2485 #ifdef INET6
2486         case AF_INET6:
2487                 pfse->pfse_type = PFSE_ICMP6;
2488                 break;
2489 #endif /* INET6 */
2490         }
2491         pfse->pfse_m = m0;
2492         pfse->pfse_icmp_type = type;
2493         pfse->pfse_icmp_code = code;
2494         pf_send(pfse);
2495 }
2496
2497 /*
2498  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2499  * If n is 0, they match if they are equal. If n is != 0, they match if they
2500  * are different.
2501  */
2502 int
2503 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2504     struct pf_addr *b, sa_family_t af)
2505 {
2506         int     match = 0;
2507
2508         switch (af) {
2509 #ifdef INET
2510         case AF_INET:
2511                 if ((a->addr32[0] & m->addr32[0]) ==
2512                     (b->addr32[0] & m->addr32[0]))
2513                         match++;
2514                 break;
2515 #endif /* INET */
2516 #ifdef INET6
2517         case AF_INET6:
2518                 if (((a->addr32[0] & m->addr32[0]) ==
2519                      (b->addr32[0] & m->addr32[0])) &&
2520                     ((a->addr32[1] & m->addr32[1]) ==
2521                      (b->addr32[1] & m->addr32[1])) &&
2522                     ((a->addr32[2] & m->addr32[2]) ==
2523                      (b->addr32[2] & m->addr32[2])) &&
2524                     ((a->addr32[3] & m->addr32[3]) ==
2525                      (b->addr32[3] & m->addr32[3])))
2526                         match++;
2527                 break;
2528 #endif /* INET6 */
2529         }
2530         if (match) {
2531                 if (n)
2532                         return (0);
2533                 else
2534                         return (1);
2535         } else {
2536                 if (n)
2537                         return (1);
2538                 else
2539                         return (0);
2540         }
2541 }
2542
2543 /*
2544  * Return 1 if b <= a <= e, otherwise return 0.
2545  */
2546 int
2547 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2548     struct pf_addr *a, sa_family_t af)
2549 {
2550         switch (af) {
2551 #ifdef INET
2552         case AF_INET:
2553                 if ((a->addr32[0] < b->addr32[0]) ||
2554                     (a->addr32[0] > e->addr32[0]))
2555                         return (0);
2556                 break;
2557 #endif /* INET */
2558 #ifdef INET6
2559         case AF_INET6: {
2560                 int     i;
2561
2562                 /* check a >= b */
2563                 for (i = 0; i < 4; ++i)
2564                         if (a->addr32[i] > b->addr32[i])
2565                                 break;
2566                         else if (a->addr32[i] < b->addr32[i])
2567                                 return (0);
2568                 /* check a <= e */
2569                 for (i = 0; i < 4; ++i)
2570                         if (a->addr32[i] < e->addr32[i])
2571                                 break;
2572                         else if (a->addr32[i] > e->addr32[i])
2573                                 return (0);
2574                 break;
2575         }
2576 #endif /* INET6 */
2577         }
2578         return (1);
2579 }
2580
2581 static int
2582 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2583 {
2584         switch (op) {
2585         case PF_OP_IRG:
2586                 return ((p > a1) && (p < a2));
2587         case PF_OP_XRG:
2588                 return ((p < a1) || (p > a2));
2589         case PF_OP_RRG:
2590                 return ((p >= a1) && (p <= a2));
2591         case PF_OP_EQ:
2592                 return (p == a1);
2593         case PF_OP_NE:
2594                 return (p != a1);
2595         case PF_OP_LT:
2596                 return (p < a1);
2597         case PF_OP_LE:
2598                 return (p <= a1);
2599         case PF_OP_GT:
2600                 return (p > a1);
2601         case PF_OP_GE:
2602                 return (p >= a1);
2603         }
2604         return (0); /* never reached */
2605 }
2606
2607 int
2608 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2609 {
2610         NTOHS(a1);
2611         NTOHS(a2);
2612         NTOHS(p);
2613         return (pf_match(op, a1, a2, p));
2614 }
2615
2616 static int
2617 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2618 {
2619         if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2620                 return (0);
2621         return (pf_match(op, a1, a2, u));
2622 }
2623
2624 static int
2625 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2626 {
2627         if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2628                 return (0);
2629         return (pf_match(op, a1, a2, g));
2630 }
2631
2632 int
2633 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2634 {
2635         if (*tag == -1)
2636                 *tag = mtag;
2637
2638         return ((!r->match_tag_not && r->match_tag == *tag) ||
2639             (r->match_tag_not && r->match_tag != *tag));
2640 }
2641
2642 int
2643 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2644 {
2645
2646         KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2647
2648         if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2649                 return (ENOMEM);
2650
2651         pd->pf_mtag->tag = tag;
2652
2653         return (0);
2654 }
2655
2656 #define PF_ANCHOR_STACKSIZE     32
2657 struct pf_anchor_stackframe {
2658         struct pf_ruleset       *rs;
2659         struct pf_rule          *r;     /* XXX: + match bit */
2660         struct pf_anchor        *child;
2661 };
2662
2663 /*
2664  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2665  */
2666 #define PF_ANCHORSTACK_MATCH    0x00000001
2667 #define PF_ANCHORSTACK_MASK     (PF_ANCHORSTACK_MATCH)
2668
2669 #define PF_ANCHOR_MATCH(f)      ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2670 #define PF_ANCHOR_RULE(f)       (struct pf_rule *)                      \
2671                                 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2672 #define PF_ANCHOR_SET_MATCH(f)  do { (f)->r = (void *)                  \
2673                                 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2674 } while (0)
2675
2676 void
2677 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2678     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2679     int *match)
2680 {
2681         struct pf_anchor_stackframe     *f;
2682
2683         PF_RULES_RASSERT();
2684
2685         if (match)
2686                 *match = 0;
2687         if (*depth >= PF_ANCHOR_STACKSIZE) {
2688                 printf("%s: anchor stack overflow on %s\n",
2689                     __func__, (*r)->anchor->name);
2690                 *r = TAILQ_NEXT(*r, entries);
2691                 return;
2692         } else if (*depth == 0 && a != NULL)
2693                 *a = *r;
2694         f = stack + (*depth)++;
2695         f->rs = *rs;
2696         f->r = *r;
2697         if ((*r)->anchor_wildcard) {
2698                 struct pf_anchor_node *parent = &(*r)->anchor->children;
2699
2700                 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2701                         *r = NULL;
2702                         return;
2703                 }
2704                 *rs = &f->child->ruleset;
2705         } else {
2706                 f->child = NULL;
2707                 *rs = &(*r)->anchor->ruleset;
2708         }
2709         *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2710 }
2711
2712 int
2713 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2714     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2715     int *match)
2716 {
2717         struct pf_anchor_stackframe     *f;
2718         struct pf_rule *fr;
2719         int quick = 0;
2720
2721         PF_RULES_RASSERT();
2722
2723         do {
2724                 if (*depth <= 0)
2725                         break;
2726                 f = stack + *depth - 1;
2727                 fr = PF_ANCHOR_RULE(f);
2728                 if (f->child != NULL) {
2729                         struct pf_anchor_node *parent;
2730
2731                         /*
2732                          * This block traverses through
2733                          * a wildcard anchor.
2734                          */
2735                         parent = &fr->anchor->children;
2736                         if (match != NULL && *match) {
2737                                 /*
2738                                  * If any of "*" matched, then
2739                                  * "foo/ *" matched, mark frame
2740                                  * appropriately.
2741                                  */
2742                                 PF_ANCHOR_SET_MATCH(f);
2743                                 *match = 0;
2744                         }
2745                         f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2746                         if (f->child != NULL) {
2747                                 *rs = &f->child->ruleset;
2748                                 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2749                                 if (*r == NULL)
2750                                         continue;
2751                                 else
2752                                         break;
2753                         }
2754                 }
2755                 (*depth)--;
2756                 if (*depth == 0 && a != NULL)
2757                         *a = NULL;
2758                 *rs = f->rs;
2759                 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2760                         quick = fr->quick;
2761                 *r = TAILQ_NEXT(fr, entries);
2762         } while (*r == NULL);
2763
2764         return (quick);
2765 }
2766
2767 #ifdef INET6
2768 void
2769 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2770     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2771 {
2772         switch (af) {
2773 #ifdef INET
2774         case AF_INET:
2775                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2776                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2777                 break;
2778 #endif /* INET */
2779         case AF_INET6:
2780                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2781                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2782                 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2783                 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2784                 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2785                 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2786                 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2787                 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2788                 break;
2789         }
2790 }
2791
2792 void
2793 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2794 {
2795         switch (af) {
2796 #ifdef INET
2797         case AF_INET:
2798                 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2799                 break;
2800 #endif /* INET */
2801         case AF_INET6:
2802                 if (addr->addr32[3] == 0xffffffff) {
2803                         addr->addr32[3] = 0;
2804                         if (addr->addr32[2] == 0xffffffff) {
2805                                 addr->addr32[2] = 0;
2806                                 if (addr->addr32[1] == 0xffffffff) {
2807                                         addr->addr32[1] = 0;
2808                                         addr->addr32[0] =
2809                                             htonl(ntohl(addr->addr32[0]) + 1);
2810                                 } else
2811                                         addr->addr32[1] =
2812                                             htonl(ntohl(addr->addr32[1]) + 1);
2813                         } else
2814                                 addr->addr32[2] =
2815                                     htonl(ntohl(addr->addr32[2]) + 1);
2816                 } else
2817                         addr->addr32[3] =
2818                             htonl(ntohl(addr->addr32[3]) + 1);
2819                 break;
2820         }
2821 }
2822 #endif /* INET6 */
2823
2824 int
2825 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
2826 {
2827         struct pf_addr          *saddr, *daddr;
2828         u_int16_t                sport, dport;
2829         struct inpcbinfo        *pi;
2830         struct inpcb            *inp;
2831
2832         pd->lookup.uid = UID_MAX;
2833         pd->lookup.gid = GID_MAX;
2834
2835         switch (pd->proto) {
2836         case IPPROTO_TCP:
2837                 if (pd->hdr.tcp == NULL)
2838                         return (-1);
2839                 sport = pd->hdr.tcp->th_sport;
2840                 dport = pd->hdr.tcp->th_dport;
2841                 pi = &V_tcbinfo;
2842                 break;
2843         case IPPROTO_UDP:
2844                 if (pd->hdr.udp == NULL)
2845                         return (-1);
2846                 sport = pd->hdr.udp->uh_sport;
2847                 dport = pd->hdr.udp->uh_dport;
2848                 pi = &V_udbinfo;
2849                 break;
2850         default:
2851                 return (-1);
2852         }
2853         if (direction == PF_IN) {
2854                 saddr = pd->src;
2855                 daddr = pd->dst;
2856         } else {
2857                 u_int16_t       p;
2858
2859                 p = sport;
2860                 sport = dport;
2861                 dport = p;
2862                 saddr = pd->dst;
2863                 daddr = pd->src;
2864         }
2865         switch (pd->af) {
2866 #ifdef INET
2867         case AF_INET:
2868                 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
2869                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2870                 if (inp == NULL) {
2871                         inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
2872                            daddr->v4, dport, INPLOOKUP_WILDCARD |
2873                            INPLOOKUP_RLOCKPCB, NULL, m);
2874                         if (inp == NULL)
2875                                 return (-1);
2876                 }
2877                 break;
2878 #endif /* INET */
2879 #ifdef INET6
2880         case AF_INET6:
2881                 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
2882                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
2883                 if (inp == NULL) {
2884                         inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
2885                             &daddr->v6, dport, INPLOOKUP_WILDCARD |
2886                             INPLOOKUP_RLOCKPCB, NULL, m);
2887                         if (inp == NULL)
2888                                 return (-1);
2889                 }
2890                 break;
2891 #endif /* INET6 */
2892
2893         default:
2894                 return (-1);
2895         }
2896         INP_RLOCK_ASSERT(inp);
2897         pd->lookup.uid = inp->inp_cred->cr_uid;
2898         pd->lookup.gid = inp->inp_cred->cr_groups[0];
2899         INP_RUNLOCK(inp);
2900
2901         return (1);
2902 }
2903
2904 static u_int8_t
2905 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2906 {
2907         int              hlen;
2908         u_int8_t         hdr[60];
2909         u_int8_t        *opt, optlen;
2910         u_int8_t         wscale = 0;
2911
2912         hlen = th_off << 2;             /* hlen <= sizeof(hdr) */
2913         if (hlen <= sizeof(struct tcphdr))
2914                 return (0);
2915         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2916                 return (0);
2917         opt = hdr + sizeof(struct tcphdr);
2918         hlen -= sizeof(struct tcphdr);
2919         while (hlen >= 3) {
2920                 switch (*opt) {
2921                 case TCPOPT_EOL:
2922                 case TCPOPT_NOP:
2923                         ++opt;
2924                         --hlen;
2925                         break;
2926                 case TCPOPT_WINDOW:
2927                         wscale = opt[2];
2928                         if (wscale > TCP_MAX_WINSHIFT)
2929                                 wscale = TCP_MAX_WINSHIFT;
2930                         wscale |= PF_WSCALE_FLAG;
2931                         /* FALLTHROUGH */
2932                 default:
2933                         optlen = opt[1];
2934                         if (optlen < 2)
2935                                 optlen = 2;
2936                         hlen -= optlen;
2937                         opt += optlen;
2938                         break;
2939                 }
2940         }
2941         return (wscale);
2942 }
2943
2944 static u_int16_t
2945 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
2946 {
2947         int              hlen;
2948         u_int8_t         hdr[60];
2949         u_int8_t        *opt, optlen;
2950         u_int16_t        mss = V_tcp_mssdflt;
2951
2952         hlen = th_off << 2;     /* hlen <= sizeof(hdr) */
2953         if (hlen <= sizeof(struct tcphdr))
2954                 return (0);
2955         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
2956                 return (0);
2957         opt = hdr + sizeof(struct tcphdr);
2958         hlen -= sizeof(struct tcphdr);
2959         while (hlen >= TCPOLEN_MAXSEG) {
2960                 switch (*opt) {
2961                 case TCPOPT_EOL:
2962                 case TCPOPT_NOP:
2963                         ++opt;
2964                         --hlen;
2965                         break;
2966                 case TCPOPT_MAXSEG:
2967                         bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
2968                         NTOHS(mss);
2969                         /* FALLTHROUGH */
2970                 default:
2971                         optlen = opt[1];
2972                         if (optlen < 2)
2973                                 optlen = 2;
2974                         hlen -= optlen;
2975                         opt += optlen;
2976                         break;
2977                 }
2978         }
2979         return (mss);
2980 }
2981
2982 static u_int16_t
2983 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
2984 {
2985 #ifdef INET
2986         struct sockaddr_in      *dst;
2987         struct route             ro;
2988 #endif /* INET */
2989 #ifdef INET6
2990         struct sockaddr_in6     *dst6;
2991         struct route_in6         ro6;
2992 #endif /* INET6 */
2993         struct rtentry          *rt = NULL;
2994         int                      hlen = 0;
2995         u_int16_t                mss = V_tcp_mssdflt;
2996
2997         switch (af) {
2998 #ifdef INET
2999         case AF_INET:
3000                 hlen = sizeof(struct ip);
3001                 bzero(&ro, sizeof(ro));
3002                 dst = (struct sockaddr_in *)&ro.ro_dst;
3003                 dst->sin_family = AF_INET;
3004                 dst->sin_len = sizeof(*dst);
3005                 dst->sin_addr = addr->v4;
3006                 in_rtalloc_ign(&ro, 0, rtableid);
3007                 rt = ro.ro_rt;
3008                 break;
3009 #endif /* INET */
3010 #ifdef INET6
3011         case AF_INET6:
3012                 hlen = sizeof(struct ip6_hdr);
3013                 bzero(&ro6, sizeof(ro6));
3014                 dst6 = (struct sockaddr_in6 *)&ro6.ro_dst;
3015                 dst6->sin6_family = AF_INET6;
3016                 dst6->sin6_len = sizeof(*dst6);
3017                 dst6->sin6_addr = addr->v6;
3018                 in6_rtalloc_ign(&ro6, 0, rtableid);
3019                 rt = ro6.ro_rt;
3020                 break;
3021 #endif /* INET6 */
3022         }
3023
3024         if (rt && rt->rt_ifp) {
3025                 mss = rt->rt_ifp->if_mtu - hlen - sizeof(struct tcphdr);
3026                 mss = max(V_tcp_mssdflt, mss);
3027                 RTFREE(rt);
3028         }
3029         mss = min(mss, offer);
3030         mss = max(mss, 64);             /* sanity - at least max opt space */
3031         return (mss);
3032 }
3033
3034 static u_int32_t
3035 pf_tcp_iss(struct pf_pdesc *pd)
3036 {
3037         MD5_CTX ctx;
3038         u_int32_t digest[4];
3039
3040         if (V_pf_tcp_secret_init == 0) {
3041                 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
3042                 MD5Init(&V_pf_tcp_secret_ctx);
3043                 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
3044                     sizeof(V_pf_tcp_secret));
3045                 V_pf_tcp_secret_init = 1;
3046         }
3047
3048         ctx = V_pf_tcp_secret_ctx;
3049
3050         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
3051         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
3052         if (pd->af == AF_INET6) {
3053                 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3054                 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3055         } else {
3056                 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3057                 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3058         }
3059         MD5Final((u_char *)digest, &ctx);
3060         V_pf_tcp_iss_off += 4096;
3061 #define ISN_RANDOM_INCREMENT (4096 - 1)
3062         return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
3063             V_pf_tcp_iss_off);
3064 #undef  ISN_RANDOM_INCREMENT
3065 }
3066
3067 static int
3068 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
3069     struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
3070     struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
3071 {
3072         struct pf_rule          *nr = NULL;
3073         struct pf_addr          * const saddr = pd->src;
3074         struct pf_addr          * const daddr = pd->dst;
3075         sa_family_t              af = pd->af;
3076         struct pf_rule          *r, *a = NULL;
3077         struct pf_ruleset       *ruleset = NULL;
3078         struct pf_src_node      *nsn = NULL;
3079         struct tcphdr           *th = pd->hdr.tcp;
3080         struct pf_state_key     *sk = NULL, *nk = NULL;
3081         u_short                  reason;
3082         int                      rewrite = 0, hdrlen = 0;
3083         int                      tag = -1, rtableid = -1;
3084         int                      asd = 0;
3085         int                      match = 0;
3086         int                      state_icmp = 0;
3087         u_int16_t                sport = 0, dport = 0;
3088         u_int16_t                bproto_sum = 0, bip_sum = 0;
3089         u_int8_t                 icmptype = 0, icmpcode = 0;
3090         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3091
3092         PF_RULES_RASSERT();
3093
3094         if (inp != NULL) {
3095                 INP_LOCK_ASSERT(inp);
3096                 pd->lookup.uid = inp->inp_cred->cr_uid;
3097                 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3098                 pd->lookup.done = 1;
3099         }
3100
3101         switch (pd->proto) {
3102         case IPPROTO_TCP:
3103                 sport = th->th_sport;
3104                 dport = th->th_dport;
3105                 hdrlen = sizeof(*th);
3106                 break;
3107         case IPPROTO_UDP:
3108                 sport = pd->hdr.udp->uh_sport;
3109                 dport = pd->hdr.udp->uh_dport;
3110                 hdrlen = sizeof(*pd->hdr.udp);
3111                 break;
3112 #ifdef INET
3113         case IPPROTO_ICMP:
3114                 if (pd->af != AF_INET)
3115                         break;
3116                 sport = dport = pd->hdr.icmp->icmp_id;
3117                 hdrlen = sizeof(*pd->hdr.icmp);
3118                 icmptype = pd->hdr.icmp->icmp_type;
3119                 icmpcode = pd->hdr.icmp->icmp_code;
3120
3121                 if (icmptype == ICMP_UNREACH ||
3122                     icmptype == ICMP_SOURCEQUENCH ||
3123                     icmptype == ICMP_REDIRECT ||
3124                     icmptype == ICMP_TIMXCEED ||
3125                     icmptype == ICMP_PARAMPROB)
3126                         state_icmp++;
3127                 break;
3128 #endif /* INET */
3129 #ifdef INET6
3130         case IPPROTO_ICMPV6:
3131                 if (af != AF_INET6)
3132                         break;
3133                 sport = dport = pd->hdr.icmp6->icmp6_id;
3134                 hdrlen = sizeof(*pd->hdr.icmp6);
3135                 icmptype = pd->hdr.icmp6->icmp6_type;
3136                 icmpcode = pd->hdr.icmp6->icmp6_code;
3137
3138                 if (icmptype == ICMP6_DST_UNREACH ||
3139                     icmptype == ICMP6_PACKET_TOO_BIG ||
3140                     icmptype == ICMP6_TIME_EXCEEDED ||
3141                     icmptype == ICMP6_PARAM_PROB)
3142                         state_icmp++;
3143                 break;
3144 #endif /* INET6 */
3145         default:
3146                 sport = dport = hdrlen = 0;
3147                 break;
3148         }
3149
3150         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3151
3152         /* check packet for BINAT/NAT/RDR */
3153         if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3154             &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3155                 KASSERT(sk != NULL, ("%s: null sk", __func__));
3156                 KASSERT(nk != NULL, ("%s: null nk", __func__));
3157
3158                 if (pd->ip_sum)
3159                         bip_sum = *pd->ip_sum;
3160
3161                 switch (pd->proto) {
3162                 case IPPROTO_TCP:
3163                         bproto_sum = th->th_sum;
3164                         pd->proto_sum = &th->th_sum;
3165
3166                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3167                             nk->port[pd->sidx] != sport) {
3168                                 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
3169                                     &th->th_sum, &nk->addr[pd->sidx],
3170                                     nk->port[pd->sidx], 0, af);
3171                                 pd->sport = &th->th_sport;
3172                                 sport = th->th_sport;
3173                         }
3174
3175                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3176                             nk->port[pd->didx] != dport) {
3177                                 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
3178                                     &th->th_sum, &nk->addr[pd->didx],
3179                                     nk->port[pd->didx], 0, af);
3180                                 dport = th->th_dport;
3181                                 pd->dport = &th->th_dport;
3182                         }
3183                         rewrite++;
3184                         break;
3185                 case IPPROTO_UDP:
3186                         bproto_sum = pd->hdr.udp->uh_sum;
3187                         pd->proto_sum = &pd->hdr.udp->uh_sum;
3188
3189                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3190                             nk->port[pd->sidx] != sport) {
3191                                 pf_change_ap(m, saddr, &pd->hdr.udp->uh_sport,
3192                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3193                                     &nk->addr[pd->sidx],
3194                                     nk->port[pd->sidx], 1, af);
3195                                 sport = pd->hdr.udp->uh_sport;
3196                                 pd->sport = &pd->hdr.udp->uh_sport;
3197                         }
3198
3199                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3200                             nk->port[pd->didx] != dport) {
3201                                 pf_change_ap(m, daddr, &pd->hdr.udp->uh_dport,
3202                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3203                                     &nk->addr[pd->didx],
3204                                     nk->port[pd->didx], 1, af);
3205                                 dport = pd->hdr.udp->uh_dport;
3206                                 pd->dport = &pd->hdr.udp->uh_dport;
3207                         }
3208                         rewrite++;
3209                         break;
3210 #ifdef INET
3211                 case IPPROTO_ICMP:
3212                         nk->port[0] = nk->port[1];
3213                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3214                                 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3215                                     nk->addr[pd->sidx].v4.s_addr, 0);
3216
3217                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3218                                 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3219                                     nk->addr[pd->didx].v4.s_addr, 0);
3220
3221                         if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3222                                 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3223                                     pd->hdr.icmp->icmp_cksum, sport,
3224                                     nk->port[1], 0);
3225                                 pd->hdr.icmp->icmp_id = nk->port[1];
3226                                 pd->sport = &pd->hdr.icmp->icmp_id;
3227                         }
3228                         m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3229                         break;
3230 #endif /* INET */
3231 #ifdef INET6
3232                 case IPPROTO_ICMPV6:
3233                         nk->port[0] = nk->port[1];
3234                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3235                                 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3236                                     &nk->addr[pd->sidx], 0);
3237
3238                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3239                                 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3240                                     &nk->addr[pd->didx], 0);
3241                         rewrite++;
3242                         break;
3243 #endif /* INET */
3244                 default:
3245                         switch (af) {
3246 #ifdef INET
3247                         case AF_INET:
3248                                 if (PF_ANEQ(saddr,
3249                                     &nk->addr[pd->sidx], AF_INET))
3250                                         pf_change_a(&saddr->v4.s_addr,
3251                                             pd->ip_sum,
3252                                             nk->addr[pd->sidx].v4.s_addr, 0);
3253
3254                                 if (PF_ANEQ(daddr,
3255                                     &nk->addr[pd->didx], AF_INET))
3256                                         pf_change_a(&daddr->v4.s_addr,
3257                                             pd->ip_sum,
3258                                             nk->addr[pd->didx].v4.s_addr, 0);
3259                                 break;
3260 #endif /* INET */
3261 #ifdef INET6
3262                         case AF_INET6:
3263                                 if (PF_ANEQ(saddr,
3264                                     &nk->addr[pd->sidx], AF_INET6))
3265                                         PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3266
3267                                 if (PF_ANEQ(daddr,
3268                                     &nk->addr[pd->didx], AF_INET6))
3269                                         PF_ACPY(saddr, &nk->addr[pd->didx], af);
3270                                 break;
3271 #endif /* INET */
3272                         }
3273                         break;
3274                 }
3275                 if (nr->natpass)
3276                         r = NULL;
3277                 pd->nat_rule = nr;
3278         }
3279
3280         while (r != NULL) {
3281                 r->evaluations++;
3282                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3283                         r = r->skip[PF_SKIP_IFP].ptr;
3284                 else if (r->direction && r->direction != direction)
3285                         r = r->skip[PF_SKIP_DIR].ptr;
3286                 else if (r->af && r->af != af)
3287                         r = r->skip[PF_SKIP_AF].ptr;
3288                 else if (r->proto && r->proto != pd->proto)
3289                         r = r->skip[PF_SKIP_PROTO].ptr;
3290                 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3291                     r->src.neg, kif, M_GETFIB(m)))
3292                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3293                 /* tcp/udp only. port_op always 0 in other cases */
3294                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3295                     r->src.port[0], r->src.port[1], sport))
3296                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
3297                 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3298                     r->dst.neg, NULL, M_GETFIB(m)))
3299                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3300                 /* tcp/udp only. port_op always 0 in other cases */
3301                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3302                     r->dst.port[0], r->dst.port[1], dport))
3303                         r = r->skip[PF_SKIP_DST_PORT].ptr;
3304                 /* icmp only. type always 0 in other cases */
3305                 else if (r->type && r->type != icmptype + 1)
3306                         r = TAILQ_NEXT(r, entries);
3307                 /* icmp only. type always 0 in other cases */
3308                 else if (r->code && r->code != icmpcode + 1)
3309                         r = TAILQ_NEXT(r, entries);
3310                 else if (r->tos && !(r->tos == pd->tos))
3311                         r = TAILQ_NEXT(r, entries);
3312                 else if (r->rule_flag & PFRULE_FRAGMENT)
3313                         r = TAILQ_NEXT(r, entries);
3314                 else if (pd->proto == IPPROTO_TCP &&
3315                     (r->flagset & th->th_flags) != r->flags)
3316                         r = TAILQ_NEXT(r, entries);
3317                 /* tcp/udp only. uid.op always 0 in other cases */
3318                 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3319                     pf_socket_lookup(direction, pd, m), 1)) &&
3320                     !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3321                     pd->lookup.uid))
3322                         r = TAILQ_NEXT(r, entries);
3323                 /* tcp/udp only. gid.op always 0 in other cases */
3324                 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3325                     pf_socket_lookup(direction, pd, m), 1)) &&
3326                     !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3327                     pd->lookup.gid))
3328                         r = TAILQ_NEXT(r, entries);
3329                 else if (r->prob &&
3330                     r->prob <= arc4random())
3331                         r = TAILQ_NEXT(r, entries);
3332                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3333                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3334                         r = TAILQ_NEXT(r, entries);
3335                 else if (r->os_fingerprint != PF_OSFP_ANY &&
3336                     (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3337                     pf_osfp_fingerprint(pd, m, off, th),
3338                     r->os_fingerprint)))
3339                         r = TAILQ_NEXT(r, entries);
3340                 else {
3341                         if (r->tag)
3342                                 tag = r->tag;
3343                         if (r->rtableid >= 0)
3344                                 rtableid = r->rtableid;
3345                         if (r->anchor == NULL) {
3346                                 match = 1;
3347                                 *rm = r;
3348                                 *am = a;
3349                                 *rsm = ruleset;
3350                                 if ((*rm)->quick)
3351                                         break;
3352                                 r = TAILQ_NEXT(r, entries);
3353                         } else
3354                                 pf_step_into_anchor(anchor_stack, &asd,
3355                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3356                                     &match);
3357                 }
3358                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3359                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3360                         break;
3361         }
3362         r = *rm;
3363         a = *am;
3364         ruleset = *rsm;
3365
3366         REASON_SET(&reason, PFRES_MATCH);
3367
3368         if (r->log || (nr != NULL && nr->log)) {
3369                 if (rewrite)
3370                         m_copyback(m, off, hdrlen, pd->hdr.any);
3371                 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3372                     ruleset, pd, 1);
3373         }
3374
3375         if ((r->action == PF_DROP) &&
3376             ((r->rule_flag & PFRULE_RETURNRST) ||
3377             (r->rule_flag & PFRULE_RETURNICMP) ||
3378             (r->rule_flag & PFRULE_RETURN))) {
3379                 /* undo NAT changes, if they have taken place */
3380                 if (nr != NULL) {
3381                         PF_ACPY(saddr, &sk->addr[pd->sidx], af);
3382                         PF_ACPY(daddr, &sk->addr[pd->didx], af);
3383                         if (pd->sport)
3384                                 *pd->sport = sk->port[pd->sidx];
3385                         if (pd->dport)
3386                                 *pd->dport = sk->port[pd->didx];
3387                         if (pd->proto_sum)
3388                                 *pd->proto_sum = bproto_sum;
3389                         if (pd->ip_sum)
3390                                 *pd->ip_sum = bip_sum;
3391                         m_copyback(m, off, hdrlen, pd->hdr.any);
3392                 }
3393                 if (pd->proto == IPPROTO_TCP &&
3394                     ((r->rule_flag & PFRULE_RETURNRST) ||
3395                     (r->rule_flag & PFRULE_RETURN)) &&
3396                     !(th->th_flags & TH_RST)) {
3397                         u_int32_t        ack = ntohl(th->th_seq) + pd->p_len;
3398                         int              len = 0;
3399 #ifdef INET
3400                         struct ip       *h4;
3401 #endif
3402 #ifdef INET6
3403                         struct ip6_hdr  *h6;
3404 #endif
3405
3406                         switch (af) {
3407 #ifdef INET
3408                         case AF_INET:
3409                                 h4 = mtod(m, struct ip *);
3410                                 len = ntohs(h4->ip_len) - off;
3411                                 break;
3412 #endif
3413 #ifdef INET6
3414                         case AF_INET6:
3415                                 h6 = mtod(m, struct ip6_hdr *);
3416                                 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
3417                                 break;
3418 #endif
3419                         }
3420
3421                         if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
3422                                 REASON_SET(&reason, PFRES_PROTCKSUM);
3423                         else {
3424                                 if (th->th_flags & TH_SYN)
3425                                         ack++;
3426                                 if (th->th_flags & TH_FIN)
3427                                         ack++;
3428                                 pf_send_tcp(m, r, af, pd->dst,
3429                                     pd->src, th->th_dport, th->th_sport,
3430                                     ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
3431                                     r->return_ttl, 1, 0, kif->pfik_ifp);
3432                         }
3433                 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
3434                     r->return_icmp)
3435                         pf_send_icmp(m, r->return_icmp >> 8,
3436                             r->return_icmp & 255, af, r);
3437                 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
3438                     r->return_icmp6)
3439                         pf_send_icmp(m, r->return_icmp6 >> 8,
3440                             r->return_icmp6 & 255, af, r);
3441         }
3442
3443         if (r->action == PF_DROP)
3444                 goto cleanup;
3445
3446         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3447                 REASON_SET(&reason, PFRES_MEMORY);
3448                 goto cleanup;
3449         }
3450         if (rtableid >= 0)
3451                 M_SETFIB(m, rtableid);
3452
3453         if (!state_icmp && (r->keep_state || nr != NULL ||
3454             (pd->flags & PFDESC_TCP_NORM))) {
3455                 int action;
3456                 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3457                     sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3458                     hdrlen);
3459                 if (action != PF_PASS)
3460                         return (action);
3461         } else {
3462                 if (sk != NULL)
3463                         uma_zfree(V_pf_state_key_z, sk);
3464                 if (nk != NULL)
3465                         uma_zfree(V_pf_state_key_z, nk);
3466         }
3467
3468         /* copy back packet headers if we performed NAT operations */
3469         if (rewrite)
3470                 m_copyback(m, off, hdrlen, pd->hdr.any);
3471
3472         if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3473             direction == PF_OUT &&
3474             pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m))
3475                 /*
3476                  * We want the state created, but we dont
3477                  * want to send this in case a partner
3478                  * firewall has to know about it to allow
3479                  * replies through it.
3480                  */
3481                 return (PF_DEFER);
3482
3483         return (PF_PASS);
3484
3485 cleanup:
3486         if (sk != NULL)
3487                 uma_zfree(V_pf_state_key_z, sk);
3488         if (nk != NULL)
3489                 uma_zfree(V_pf_state_key_z, nk);
3490         return (PF_DROP);
3491 }
3492
3493 static int
3494 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3495     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3496     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3497     u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3498     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3499 {
3500         struct pf_state         *s = NULL;
3501         struct pf_src_node      *sn = NULL;
3502         struct tcphdr           *th = pd->hdr.tcp;
3503         u_int16_t                mss = V_tcp_mssdflt;
3504         u_short                  reason;
3505
3506         /* check maximums */
3507         if (r->max_states &&
3508             (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3509                 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
3510                 REASON_SET(&reason, PFRES_MAXSTATES);
3511                 return (PF_DROP);
3512         }
3513         /* src node for filter rule */
3514         if ((r->rule_flag & PFRULE_SRCTRACK ||
3515             r->rpool.opts & PF_POOL_STICKYADDR) &&
3516             pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3517                 REASON_SET(&reason, PFRES_SRCLIMIT);
3518                 goto csfailed;
3519         }
3520         /* src node for translation rule */
3521         if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3522             pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3523                 REASON_SET(&reason, PFRES_SRCLIMIT);
3524                 goto csfailed;
3525         }
3526         s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3527         if (s == NULL) {
3528                 REASON_SET(&reason, PFRES_MEMORY);
3529                 goto csfailed;
3530         }
3531         s->rule.ptr = r;
3532         s->nat_rule.ptr = nr;
3533         s->anchor.ptr = a;
3534         STATE_INC_COUNTERS(s);
3535         if (r->allow_opts)
3536                 s->state_flags |= PFSTATE_ALLOWOPTS;
3537         if (r->rule_flag & PFRULE_STATESLOPPY)
3538                 s->state_flags |= PFSTATE_SLOPPY;
3539         s->log = r->log & PF_LOG_ALL;
3540         s->sync_state = PFSYNC_S_NONE;
3541         if (nr != NULL)
3542                 s->log |= nr->log & PF_LOG_ALL;
3543         switch (pd->proto) {
3544         case IPPROTO_TCP:
3545                 s->src.seqlo = ntohl(th->th_seq);
3546                 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3547                 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3548                     r->keep_state == PF_STATE_MODULATE) {
3549                         /* Generate sequence number modulator */
3550                         if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3551                             0)
3552                                 s->src.seqdiff = 1;
3553                         pf_change_proto_a(m, &th->th_seq, &th->th_sum,
3554                             htonl(s->src.seqlo + s->src.seqdiff), 0);
3555                         *rewrite = 1;
3556                 } else
3557                         s->src.seqdiff = 0;
3558                 if (th->th_flags & TH_SYN) {
3559                         s->src.seqhi++;
3560                         s->src.wscale = pf_get_wscale(m, off,
3561                             th->th_off, pd->af);
3562                 }
3563                 s->src.max_win = MAX(ntohs(th->th_win), 1);
3564                 if (s->src.wscale & PF_WSCALE_MASK) {
3565                         /* Remove scale factor from initial window */
3566                         int win = s->src.max_win;
3567                         win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3568                         s->src.max_win = (win - 1) >>
3569                             (s->src.wscale & PF_WSCALE_MASK);
3570                 }
3571                 if (th->th_flags & TH_FIN)
3572                         s->src.seqhi++;
3573                 s->dst.seqhi = 1;
3574                 s->dst.max_win = 1;
3575                 s->src.state = TCPS_SYN_SENT;
3576                 s->dst.state = TCPS_CLOSED;
3577                 s->timeout = PFTM_TCP_FIRST_PACKET;
3578                 break;
3579         case IPPROTO_UDP:
3580                 s->src.state = PFUDPS_SINGLE;
3581                 s->dst.state = PFUDPS_NO_TRAFFIC;
3582                 s->timeout = PFTM_UDP_FIRST_PACKET;
3583                 break;
3584         case IPPROTO_ICMP:
3585 #ifdef INET6
3586         case IPPROTO_ICMPV6:
3587 #endif
3588                 s->timeout = PFTM_ICMP_FIRST_PACKET;
3589                 break;
3590         default:
3591                 s->src.state = PFOTHERS_SINGLE;
3592                 s->dst.state = PFOTHERS_NO_TRAFFIC;
3593                 s->timeout = PFTM_OTHER_FIRST_PACKET;
3594         }
3595
3596         if (r->rt && r->rt != PF_FASTROUTE) {
3597                 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) {
3598                         REASON_SET(&reason, PFRES_BADSTATE);
3599                         pf_src_tree_remove_state(s);
3600                         STATE_DEC_COUNTERS(s);
3601                         uma_zfree(V_pf_state_z, s);
3602                         goto csfailed;
3603                 }
3604                 s->rt_kif = r->rpool.cur->kif;
3605         }
3606
3607         s->creation = time_uptime;
3608         s->expire = time_uptime;
3609
3610         if (sn != NULL)
3611                 s->src_node = sn;
3612         if (nsn != NULL) {
3613                 /* XXX We only modify one side for now. */
3614                 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3615                 s->nat_src_node = nsn;
3616         }
3617         if (pd->proto == IPPROTO_TCP) {
3618                 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3619                     off, pd, th, &s->src, &s->dst)) {
3620                         REASON_SET(&reason, PFRES_MEMORY);
3621                         pf_src_tree_remove_state(s);
3622                         STATE_DEC_COUNTERS(s);
3623                         uma_zfree(V_pf_state_z, s);
3624                         return (PF_DROP);
3625                 }
3626                 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3627                     pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3628                     &s->src, &s->dst, rewrite)) {
3629                         /* This really shouldn't happen!!! */
3630                         DPFPRINTF(PF_DEBUG_URGENT,
3631                             ("pf_normalize_tcp_stateful failed on first pkt"));
3632                         pf_normalize_tcp_cleanup(s);
3633                         pf_src_tree_remove_state(s);
3634                         STATE_DEC_COUNTERS(s);
3635                         uma_zfree(V_pf_state_z, s);
3636                         return (PF_DROP);
3637                 }
3638         }
3639         s->direction = pd->dir;
3640
3641         /*
3642          * sk/nk could already been setup by pf_get_translation().
3643          */
3644         if (nr == NULL) {
3645                 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3646                     __func__, nr, sk, nk));
3647                 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3648                 if (sk == NULL)
3649                         goto csfailed;
3650                 nk = sk;
3651         } else
3652                 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3653                     __func__, nr, sk, nk));
3654
3655         /* Swap sk/nk for PF_OUT. */
3656         if (pf_state_insert(BOUND_IFACE(r, kif),
3657             (pd->dir == PF_IN) ? sk : nk,
3658             (pd->dir == PF_IN) ? nk : sk, s)) {
3659                 if (pd->proto == IPPROTO_TCP)
3660                         pf_normalize_tcp_cleanup(s);
3661                 REASON_SET(&reason, PFRES_STATEINS);
3662                 pf_src_tree_remove_state(s);
3663                 STATE_DEC_COUNTERS(s);
3664                 uma_zfree(V_pf_state_z, s);
3665                 return (PF_DROP);
3666         } else
3667                 *sm = s;
3668
3669         if (tag > 0)
3670                 s->tag = tag;
3671         if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3672             TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3673                 s->src.state = PF_TCPS_PROXY_SRC;
3674                 /* undo NAT changes, if they have taken place */
3675                 if (nr != NULL) {
3676                         struct pf_state_key *skt = s->key[PF_SK_WIRE];
3677                         if (pd->dir == PF_OUT)
3678                                 skt = s->key[PF_SK_STACK];
3679                         PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3680                         PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3681                         if (pd->sport)
3682                                 *pd->sport = skt->port[pd->sidx];
3683                         if (pd->dport)
3684                                 *pd->dport = skt->port[pd->didx];
3685                         if (pd->proto_sum)
3686                                 *pd->proto_sum = bproto_sum;
3687                         if (pd->ip_sum)
3688                                 *pd->ip_sum = bip_sum;
3689                         m_copyback(m, off, hdrlen, pd->hdr.any);
3690                 }
3691                 s->src.seqhi = htonl(arc4random());
3692                 /* Find mss option */
3693                 int rtid = M_GETFIB(m);
3694                 mss = pf_get_mss(m, off, th->th_off, pd->af);
3695                 mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3696                 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3697                 s->src.mss = mss;
3698                 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3699                     th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3700                     TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3701                 REASON_SET(&reason, PFRES_SYNPROXY);
3702                 return (PF_SYNPROXY_DROP);
3703         }
3704
3705         return (PF_PASS);
3706
3707 csfailed:
3708         if (sk != NULL)
3709                 uma_zfree(V_pf_state_key_z, sk);
3710         if (nk != NULL)
3711                 uma_zfree(V_pf_state_key_z, nk);
3712
3713         if (sn != NULL) {
3714                 struct pf_srchash *sh;
3715
3716                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
3717                 PF_HASHROW_LOCK(sh);
3718                 if (--sn->states == 0 && sn->expire == 0) {
3719                         pf_unlink_src_node(sn);
3720                         uma_zfree(V_pf_sources_z, sn);
3721                         counter_u64_add(
3722                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3723                 }
3724                 PF_HASHROW_UNLOCK(sh);
3725         }
3726
3727         if (nsn != sn && nsn != NULL) {
3728                 struct pf_srchash *sh;
3729
3730                 sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)];
3731                 PF_HASHROW_LOCK(sh);
3732                 if (--nsn->states == 0 && nsn->expire == 0) {
3733                         pf_unlink_src_node(nsn);
3734                         uma_zfree(V_pf_sources_z, nsn);
3735                         counter_u64_add(
3736                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3737                 }
3738                 PF_HASHROW_UNLOCK(sh);
3739         }
3740
3741         return (PF_DROP);
3742 }
3743
3744 static int
3745 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
3746     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
3747     struct pf_ruleset **rsm)
3748 {
3749         struct pf_rule          *r, *a = NULL;
3750         struct pf_ruleset       *ruleset = NULL;
3751         sa_family_t              af = pd->af;
3752         u_short                  reason;
3753         int                      tag = -1;
3754         int                      asd = 0;
3755         int                      match = 0;
3756         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3757
3758         PF_RULES_RASSERT();
3759
3760         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3761         while (r != NULL) {
3762                 r->evaluations++;
3763                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3764                         r = r->skip[PF_SKIP_IFP].ptr;
3765                 else if (r->direction && r->direction != direction)
3766                         r = r->skip[PF_SKIP_DIR].ptr;
3767                 else if (r->af && r->af != af)
3768                         r = r->skip[PF_SKIP_AF].ptr;
3769                 else if (r->proto && r->proto != pd->proto)
3770                         r = r->skip[PF_SKIP_PROTO].ptr;
3771                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3772                     r->src.neg, kif, M_GETFIB(m)))
3773                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3774                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3775                     r->dst.neg, NULL, M_GETFIB(m)))
3776                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3777                 else if (r->tos && !(r->tos == pd->tos))
3778                         r = TAILQ_NEXT(r, entries);
3779                 else if (r->os_fingerprint != PF_OSFP_ANY)
3780                         r = TAILQ_NEXT(r, entries);
3781                 else if (pd->proto == IPPROTO_UDP &&
3782                     (r->src.port_op || r->dst.port_op))
3783                         r = TAILQ_NEXT(r, entries);
3784                 else if (pd->proto == IPPROTO_TCP &&
3785                     (r->src.port_op || r->dst.port_op || r->flagset))
3786                         r = TAILQ_NEXT(r, entries);
3787                 else if ((pd->proto == IPPROTO_ICMP ||
3788                     pd->proto == IPPROTO_ICMPV6) &&
3789                     (r->type || r->code))
3790                         r = TAILQ_NEXT(r, entries);
3791                 else if (r->prob && r->prob <=
3792                     (arc4random() % (UINT_MAX - 1) + 1))
3793                         r = TAILQ_NEXT(r, entries);
3794                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3795                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3796                         r = TAILQ_NEXT(r, entries);
3797                 else {
3798                         if (r->anchor == NULL) {
3799                                 match = 1;
3800                                 *rm = r;
3801                                 *am = a;
3802                                 *rsm = ruleset;
3803                                 if ((*rm)->quick)
3804                                         break;
3805                                 r = TAILQ_NEXT(r, entries);
3806                         } else
3807                                 pf_step_into_anchor(anchor_stack, &asd,
3808                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3809                                     &match);
3810                 }
3811                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3812                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3813                         break;
3814         }
3815         r = *rm;
3816         a = *am;
3817         ruleset = *rsm;
3818
3819         REASON_SET(&reason, PFRES_MATCH);
3820
3821         if (r->log)
3822                 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
3823                     1);
3824
3825         if (r->action != PF_PASS)
3826                 return (PF_DROP);
3827
3828         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3829                 REASON_SET(&reason, PFRES_MEMORY);
3830                 return (PF_DROP);
3831         }
3832
3833         return (PF_PASS);
3834 }
3835
3836 static int
3837 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
3838         struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
3839         struct pf_pdesc *pd, u_short *reason, int *copyback)
3840 {
3841         struct tcphdr           *th = pd->hdr.tcp;
3842         u_int16_t                win = ntohs(th->th_win);
3843         u_int32_t                ack, end, seq, orig_seq;
3844         u_int8_t                 sws, dws;
3845         int                      ackskew;
3846
3847         if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
3848                 sws = src->wscale & PF_WSCALE_MASK;
3849                 dws = dst->wscale & PF_WSCALE_MASK;
3850         } else
3851                 sws = dws = 0;
3852
3853         /*
3854          * Sequence tracking algorithm from Guido van Rooij's paper:
3855          *   http://www.madison-gurkha.com/publications/tcp_filtering/
3856          *      tcp_filtering.ps
3857          */
3858
3859         orig_seq = seq = ntohl(th->th_seq);
3860         if (src->seqlo == 0) {
3861                 /* First packet from this end. Set its state */
3862
3863                 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
3864                     src->scrub == NULL) {
3865                         if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
3866                                 REASON_SET(reason, PFRES_MEMORY);
3867                                 return (PF_DROP);
3868                         }
3869                 }
3870
3871                 /* Deferred generation of sequence number modulator */
3872                 if (dst->seqdiff && !src->seqdiff) {
3873                         /* use random iss for the TCP server */
3874                         while ((src->seqdiff = arc4random() - seq) == 0)
3875                                 ;
3876                         ack = ntohl(th->th_ack) - dst->seqdiff;
3877                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
3878                             src->seqdiff), 0);
3879                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
3880                         *copyback = 1;
3881                 } else {
3882                         ack = ntohl(th->th_ack);
3883                 }
3884
3885                 end = seq + pd->p_len;
3886                 if (th->th_flags & TH_SYN) {
3887                         end++;
3888                         if (dst->wscale & PF_WSCALE_FLAG) {
3889                                 src->wscale = pf_get_wscale(m, off, th->th_off,
3890                                     pd->af);
3891                                 if (src->wscale & PF_WSCALE_FLAG) {
3892                                         /* Remove scale factor from initial
3893                                          * window */
3894                                         sws = src->wscale & PF_WSCALE_MASK;
3895                                         win = ((u_int32_t)win + (1 << sws) - 1)
3896                                             >> sws;
3897                                         dws = dst->wscale & PF_WSCALE_MASK;
3898                                 } else {
3899                                         /* fixup other window */
3900                                         dst->max_win <<= dst->wscale &
3901                                             PF_WSCALE_MASK;
3902                                         /* in case of a retrans SYN|ACK */
3903                                         dst->wscale = 0;
3904                                 }
3905                         }
3906                 }
3907                 if (th->th_flags & TH_FIN)
3908                         end++;
3909
3910                 src->seqlo = seq;
3911                 if (src->state < TCPS_SYN_SENT)
3912                         src->state = TCPS_SYN_SENT;
3913
3914                 /*
3915                  * May need to slide the window (seqhi may have been set by
3916                  * the crappy stack check or if we picked up the connection
3917                  * after establishment)
3918                  */
3919                 if (src->seqhi == 1 ||
3920                     SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
3921                         src->seqhi = end + MAX(1, dst->max_win << dws);
3922                 if (win > src->max_win)
3923                         src->max_win = win;
3924
3925         } else {
3926                 ack = ntohl(th->th_ack) - dst->seqdiff;
3927                 if (src->seqdiff) {
3928                         /* Modulate sequence numbers */
3929                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
3930                             src->seqdiff), 0);
3931                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
3932                         *copyback = 1;
3933                 }
3934                 end = seq + pd->p_len;
3935                 if (th->th_flags & TH_SYN)
3936                         end++;
3937                 if (th->th_flags & TH_FIN)
3938                         end++;
3939         }
3940
3941         if ((th->th_flags & TH_ACK) == 0) {
3942                 /* Let it pass through the ack skew check */
3943                 ack = dst->seqlo;
3944         } else if ((ack == 0 &&
3945             (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
3946             /* broken tcp stacks do not set ack */
3947             (dst->state < TCPS_SYN_SENT)) {
3948                 /*
3949                  * Many stacks (ours included) will set the ACK number in an
3950                  * FIN|ACK if the SYN times out -- no sequence to ACK.
3951                  */
3952                 ack = dst->seqlo;
3953         }
3954
3955         if (seq == end) {
3956                 /* Ease sequencing restrictions on no data packets */
3957                 seq = src->seqlo;
3958                 end = seq;
3959         }
3960
3961         ackskew = dst->seqlo - ack;
3962
3963
3964         /*
3965          * Need to demodulate the sequence numbers in any TCP SACK options
3966          * (Selective ACK). We could optionally validate the SACK values
3967          * against the current ACK window, either forwards or backwards, but
3968          * I'm not confident that SACK has been implemented properly
3969          * everywhere. It wouldn't surprise me if several stacks accidently
3970          * SACK too far backwards of previously ACKed data. There really aren't
3971          * any security implications of bad SACKing unless the target stack
3972          * doesn't validate the option length correctly. Someone trying to
3973          * spoof into a TCP connection won't bother blindly sending SACK
3974          * options anyway.
3975          */
3976         if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
3977                 if (pf_modulate_sack(m, off, pd, th, dst))
3978                         *copyback = 1;
3979         }
3980
3981
3982 #define MAXACKWINDOW (0xffff + 1500)    /* 1500 is an arbitrary fudge factor */
3983         if (SEQ_GEQ(src->seqhi, end) &&
3984             /* Last octet inside other's window space */
3985             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
3986             /* Retrans: not more than one window back */
3987             (ackskew >= -MAXACKWINDOW) &&
3988             /* Acking not more than one reassembled fragment backwards */
3989             (ackskew <= (MAXACKWINDOW << sws)) &&
3990             /* Acking not more than one window forward */
3991             ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
3992             (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
3993             (pd->flags & PFDESC_IP_REAS) == 0)) {
3994             /* Require an exact/+1 sequence match on resets when possible */
3995
3996                 if (dst->scrub || src->scrub) {
3997                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
3998                             *state, src, dst, copyback))
3999                                 return (PF_DROP);
4000                 }
4001
4002                 /* update max window */
4003                 if (src->max_win < win)
4004                         src->max_win = win;
4005                 /* synchronize sequencing */
4006                 if (SEQ_GT(end, src->seqlo))
4007                         src->seqlo = end;
4008                 /* slide the window of what the other end can send */
4009                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4010                         dst->seqhi = ack + MAX((win << sws), 1);
4011
4012
4013                 /* update states */
4014                 if (th->th_flags & TH_SYN)
4015                         if (src->state < TCPS_SYN_SENT)
4016                                 src->state = TCPS_SYN_SENT;
4017                 if (th->th_flags & TH_FIN)
4018                         if (src->state < TCPS_CLOSING)
4019                                 src->state = TCPS_CLOSING;
4020                 if (th->th_flags & TH_ACK) {
4021                         if (dst->state == TCPS_SYN_SENT) {
4022                                 dst->state = TCPS_ESTABLISHED;
4023                                 if (src->state == TCPS_ESTABLISHED &&
4024                                     (*state)->src_node != NULL &&
4025                                     pf_src_connlimit(state)) {
4026                                         REASON_SET(reason, PFRES_SRCLIMIT);
4027                                         return (PF_DROP);
4028                                 }
4029                         } else if (dst->state == TCPS_CLOSING)
4030                                 dst->state = TCPS_FIN_WAIT_2;
4031                 }
4032                 if (th->th_flags & TH_RST)
4033                         src->state = dst->state = TCPS_TIME_WAIT;
4034
4035                 /* update expire time */
4036                 (*state)->expire = time_uptime;
4037                 if (src->state >= TCPS_FIN_WAIT_2 &&
4038                     dst->state >= TCPS_FIN_WAIT_2)
4039                         (*state)->timeout = PFTM_TCP_CLOSED;
4040                 else if (src->state >= TCPS_CLOSING &&
4041                     dst->state >= TCPS_CLOSING)
4042                         (*state)->timeout = PFTM_TCP_FIN_WAIT;
4043                 else if (src->state < TCPS_ESTABLISHED ||
4044                     dst->state < TCPS_ESTABLISHED)
4045                         (*state)->timeout = PFTM_TCP_OPENING;
4046                 else if (src->state >= TCPS_CLOSING ||
4047                     dst->state >= TCPS_CLOSING)
4048                         (*state)->timeout = PFTM_TCP_CLOSING;
4049                 else
4050                         (*state)->timeout = PFTM_TCP_ESTABLISHED;
4051
4052                 /* Fall through to PASS packet */
4053
4054         } else if ((dst->state < TCPS_SYN_SENT ||
4055                 dst->state >= TCPS_FIN_WAIT_2 ||
4056                 src->state >= TCPS_FIN_WAIT_2) &&
4057             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4058             /* Within a window forward of the originating packet */
4059             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4060             /* Within a window backward of the originating packet */
4061
4062                 /*
4063                  * This currently handles three situations:
4064                  *  1) Stupid stacks will shotgun SYNs before their peer
4065                  *     replies.
4066                  *  2) When PF catches an already established stream (the
4067                  *     firewall rebooted, the state table was flushed, routes
4068                  *     changed...)
4069                  *  3) Packets get funky immediately after the connection
4070                  *     closes (this should catch Solaris spurious ACK|FINs
4071                  *     that web servers like to spew after a close)
4072                  *
4073                  * This must be a little more careful than the above code
4074                  * since packet floods will also be caught here. We don't
4075                  * update the TTL here to mitigate the damage of a packet
4076                  * flood and so the same code can handle awkward establishment
4077                  * and a loosened connection close.
4078                  * In the establishment case, a correct peer response will
4079                  * validate the connection, go through the normal state code
4080                  * and keep updating the state TTL.
4081                  */
4082
4083                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4084                         printf("pf: loose state match: ");
4085                         pf_print_state(*state);
4086                         pf_print_flags(th->th_flags);
4087                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4088                             "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
4089                             pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
4090                             (unsigned long long)(*state)->packets[1],
4091                             pd->dir == PF_IN ? "in" : "out",
4092                             pd->dir == (*state)->direction ? "fwd" : "rev");
4093                 }
4094
4095                 if (dst->scrub || src->scrub) {
4096                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4097                             *state, src, dst, copyback))
4098                                 return (PF_DROP);
4099                 }
4100
4101                 /* update max window */
4102                 if (src->max_win < win)
4103                         src->max_win = win;
4104                 /* synchronize sequencing */
4105                 if (SEQ_GT(end, src->seqlo))
4106                         src->seqlo = end;
4107                 /* slide the window of what the other end can send */
4108                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4109                         dst->seqhi = ack + MAX((win << sws), 1);
4110
4111                 /*
4112                  * Cannot set dst->seqhi here since this could be a shotgunned
4113                  * SYN and not an already established connection.
4114                  */
4115
4116                 if (th->th_flags & TH_FIN)
4117                         if (src->state < TCPS_CLOSING)
4118                                 src->state = TCPS_CLOSING;
4119                 if (th->th_flags & TH_RST)
4120                         src->state = dst->state = TCPS_TIME_WAIT;
4121
4122                 /* Fall through to PASS packet */
4123
4124         } else {
4125                 if ((*state)->dst.state == TCPS_SYN_SENT &&
4126                     (*state)->src.state == TCPS_SYN_SENT) {
4127                         /* Send RST for state mismatches during handshake */
4128                         if (!(th->th_flags & TH_RST))
4129                                 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4130                                     pd->dst, pd->src, th->th_dport,
4131                                     th->th_sport, ntohl(th->th_ack), 0,
4132                                     TH_RST, 0, 0,
4133                                     (*state)->rule.ptr->return_ttl, 1, 0,
4134                                     kif->pfik_ifp);
4135                         src->seqlo = 0;
4136                         src->seqhi = 1;
4137                         src->max_win = 1;
4138                 } else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4139                         printf("pf: BAD state: ");
4140                         pf_print_state(*state);
4141                         pf_print_flags(th->th_flags);
4142                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4143                             "pkts=%llu:%llu dir=%s,%s\n",
4144                             seq, orig_seq, ack, pd->p_len, ackskew,
4145                             (unsigned long long)(*state)->packets[0],
4146                             (unsigned long long)(*state)->packets[1],
4147                             pd->dir == PF_IN ? "in" : "out",
4148                             pd->dir == (*state)->direction ? "fwd" : "rev");
4149                         printf("pf: State failure on: %c %c %c %c | %c %c\n",
4150                             SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4151                             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4152                             ' ': '2',
4153                             (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4154                             (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4155                             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4156                             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4157                 }
4158                 REASON_SET(reason, PFRES_BADSTATE);
4159                 return (PF_DROP);
4160         }
4161
4162         return (PF_PASS);
4163 }
4164
4165 static int
4166 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4167         struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4168 {
4169         struct tcphdr           *th = pd->hdr.tcp;
4170
4171         if (th->th_flags & TH_SYN)
4172                 if (src->state < TCPS_SYN_SENT)
4173                         src->state = TCPS_SYN_SENT;
4174         if (th->th_flags & TH_FIN)
4175                 if (src->state < TCPS_CLOSING)
4176                         src->state = TCPS_CLOSING;
4177         if (th->th_flags & TH_ACK) {
4178                 if (dst->state == TCPS_SYN_SENT) {
4179                         dst->state = TCPS_ESTABLISHED;
4180                         if (src->state == TCPS_ESTABLISHED &&
4181                             (*state)->src_node != NULL &&
4182                             pf_src_connlimit(state)) {
4183                                 REASON_SET(reason, PFRES_SRCLIMIT);
4184                                 return (PF_DROP);
4185                         }
4186                 } else if (dst->state == TCPS_CLOSING) {
4187                         dst->state = TCPS_FIN_WAIT_2;
4188                 } else if (src->state == TCPS_SYN_SENT &&
4189                     dst->state < TCPS_SYN_SENT) {
4190                         /*
4191                          * Handle a special sloppy case where we only see one
4192                          * half of the connection. If there is a ACK after
4193                          * the initial SYN without ever seeing a packet from
4194                          * the destination, set the connection to established.
4195                          */
4196                         dst->state = src->state = TCPS_ESTABLISHED;
4197                         if ((*state)->src_node != NULL &&
4198                             pf_src_connlimit(state)) {
4199                                 REASON_SET(reason, PFRES_SRCLIMIT);
4200                                 return (PF_DROP);
4201                         }
4202                 } else if (src->state == TCPS_CLOSING &&
4203                     dst->state == TCPS_ESTABLISHED &&
4204                     dst->seqlo == 0) {
4205                         /*
4206                          * Handle the closing of half connections where we
4207                          * don't see the full bidirectional FIN/ACK+ACK
4208                          * handshake.
4209                          */
4210                         dst->state = TCPS_CLOSING;
4211                 }
4212         }
4213         if (th->th_flags & TH_RST)
4214                 src->state = dst->state = TCPS_TIME_WAIT;
4215
4216         /* update expire time */
4217         (*state)->expire = time_uptime;
4218         if (src->state >= TCPS_FIN_WAIT_2 &&
4219             dst->state >= TCPS_FIN_WAIT_2)
4220                 (*state)->timeout = PFTM_TCP_CLOSED;
4221         else if (src->state >= TCPS_CLOSING &&
4222             dst->state >= TCPS_CLOSING)
4223                 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4224         else if (src->state < TCPS_ESTABLISHED ||
4225             dst->state < TCPS_ESTABLISHED)
4226                 (*state)->timeout = PFTM_TCP_OPENING;
4227         else if (src->state >= TCPS_CLOSING ||
4228             dst->state >= TCPS_CLOSING)
4229                 (*state)->timeout = PFTM_TCP_CLOSING;
4230         else
4231                 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4232
4233         return (PF_PASS);
4234 }
4235
4236 static int
4237 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4238     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4239     u_short *reason)
4240 {
4241         struct pf_state_key_cmp  key;
4242         struct tcphdr           *th = pd->hdr.tcp;
4243         int                      copyback = 0;
4244         struct pf_state_peer    *src, *dst;
4245         struct pf_state_key     *sk;
4246
4247         bzero(&key, sizeof(key));
4248         key.af = pd->af;
4249         key.proto = IPPROTO_TCP;
4250         if (direction == PF_IN) {       /* wire side, straight */
4251                 PF_ACPY(&key.addr[0], pd->src, key.af);
4252                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4253                 key.port[0] = th->th_sport;
4254                 key.port[1] = th->th_dport;
4255         } else {                        /* stack side, reverse */
4256                 PF_ACPY(&key.addr[1], pd->src, key.af);
4257                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4258                 key.port[1] = th->th_sport;
4259                 key.port[0] = th->th_dport;
4260         }
4261
4262         STATE_LOOKUP(kif, &key, direction, *state, pd);
4263
4264         if (direction == (*state)->direction) {
4265                 src = &(*state)->src;
4266                 dst = &(*state)->dst;
4267         } else {
4268                 src = &(*state)->dst;
4269                 dst = &(*state)->src;
4270         }
4271
4272         sk = (*state)->key[pd->didx];
4273
4274         if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4275                 if (direction != (*state)->direction) {
4276                         REASON_SET(reason, PFRES_SYNPROXY);
4277                         return (PF_SYNPROXY_DROP);
4278                 }
4279                 if (th->th_flags & TH_SYN) {
4280                         if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4281                                 REASON_SET(reason, PFRES_SYNPROXY);
4282                                 return (PF_DROP);
4283                         }
4284                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4285                             pd->src, th->th_dport, th->th_sport,
4286                             (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4287                             TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4288                         REASON_SET(reason, PFRES_SYNPROXY);
4289                         return (PF_SYNPROXY_DROP);
4290                 } else if (!(th->th_flags & TH_ACK) ||
4291                     (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4292                     (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4293                         REASON_SET(reason, PFRES_SYNPROXY);
4294                         return (PF_DROP);
4295                 } else if ((*state)->src_node != NULL &&
4296                     pf_src_connlimit(state)) {
4297                         REASON_SET(reason, PFRES_SRCLIMIT);
4298                         return (PF_DROP);
4299                 } else
4300                         (*state)->src.state = PF_TCPS_PROXY_DST;
4301         }
4302         if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4303                 if (direction == (*state)->direction) {
4304                         if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4305                             (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4306                             (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4307                                 REASON_SET(reason, PFRES_SYNPROXY);
4308                                 return (PF_DROP);
4309                         }
4310                         (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4311                         if ((*state)->dst.seqhi == 1)
4312                                 (*state)->dst.seqhi = htonl(arc4random());
4313                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4314                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4315                             sk->port[pd->sidx], sk->port[pd->didx],
4316                             (*state)->dst.seqhi, 0, TH_SYN, 0,
4317                             (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4318                         REASON_SET(reason, PFRES_SYNPROXY);
4319                         return (PF_SYNPROXY_DROP);
4320                 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4321                     (TH_SYN|TH_ACK)) ||
4322                     (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4323                         REASON_SET(reason, PFRES_SYNPROXY);
4324                         return (PF_DROP);
4325                 } else {
4326                         (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4327                         (*state)->dst.seqlo = ntohl(th->th_seq);
4328                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4329                             pd->src, th->th_dport, th->th_sport,
4330                             ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4331                             TH_ACK, (*state)->src.max_win, 0, 0, 0,
4332                             (*state)->tag, NULL);
4333                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4334                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4335                             sk->port[pd->sidx], sk->port[pd->didx],
4336                             (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4337                             TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4338                         (*state)->src.seqdiff = (*state)->dst.seqhi -
4339                             (*state)->src.seqlo;
4340                         (*state)->dst.seqdiff = (*state)->src.seqhi -
4341                             (*state)->dst.seqlo;
4342                         (*state)->src.seqhi = (*state)->src.seqlo +
4343                             (*state)->dst.max_win;
4344                         (*state)->dst.seqhi = (*state)->dst.seqlo +
4345                             (*state)->src.max_win;
4346                         (*state)->src.wscale = (*state)->dst.wscale = 0;
4347                         (*state)->src.state = (*state)->dst.state =
4348                             TCPS_ESTABLISHED;
4349                         REASON_SET(reason, PFRES_SYNPROXY);
4350                         return (PF_SYNPROXY_DROP);
4351                 }
4352         }
4353
4354         if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4355             dst->state >= TCPS_FIN_WAIT_2 &&
4356             src->state >= TCPS_FIN_WAIT_2) {
4357                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4358                         printf("pf: state reuse ");
4359                         pf_print_state(*state);
4360                         pf_print_flags(th->th_flags);
4361                         printf("\n");
4362                 }
4363                 /* XXX make sure it's the same direction ?? */
4364                 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4365                 pf_unlink_state(*state, PF_ENTER_LOCKED);
4366                 *state = NULL;
4367                 return (PF_DROP);
4368         }
4369
4370         if ((*state)->state_flags & PFSTATE_SLOPPY) {
4371                 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4372                         return (PF_DROP);
4373         } else {
4374                 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4375                     &copyback) == PF_DROP)
4376                         return (PF_DROP);
4377         }
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] != th->th_sport)
4385                         pf_change_ap(m, pd->src, &th->th_sport,
4386                             pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
4387                             nk->port[pd->sidx], 0, pd->af);
4388
4389                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4390                     nk->port[pd->didx] != th->th_dport)
4391                         pf_change_ap(m, pd->dst, &th->th_dport,
4392                             pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
4393                             nk->port[pd->didx], 0, pd->af);
4394                 copyback = 1;
4395         }
4396
4397         /* Copyback sequence modulation or stateful scrub changes if needed */
4398         if (copyback)
4399                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
4400
4401         return (PF_PASS);
4402 }
4403
4404 static int
4405 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
4406     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4407 {
4408         struct pf_state_peer    *src, *dst;
4409         struct pf_state_key_cmp  key;
4410         struct udphdr           *uh = pd->hdr.udp;
4411
4412         bzero(&key, sizeof(key));
4413         key.af = pd->af;
4414         key.proto = IPPROTO_UDP;
4415         if (direction == PF_IN) {       /* wire side, straight */
4416                 PF_ACPY(&key.addr[0], pd->src, key.af);
4417                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4418                 key.port[0] = uh->uh_sport;
4419                 key.port[1] = uh->uh_dport;
4420         } else {                        /* stack side, reverse */
4421                 PF_ACPY(&key.addr[1], pd->src, key.af);
4422                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4423                 key.port[1] = uh->uh_sport;
4424                 key.port[0] = uh->uh_dport;
4425         }
4426
4427         STATE_LOOKUP(kif, &key, direction, *state, pd);
4428
4429         if (direction == (*state)->direction) {
4430                 src = &(*state)->src;
4431                 dst = &(*state)->dst;
4432         } else {
4433                 src = &(*state)->dst;
4434                 dst = &(*state)->src;
4435         }
4436
4437         /* update states */
4438         if (src->state < PFUDPS_SINGLE)
4439                 src->state = PFUDPS_SINGLE;
4440         if (dst->state == PFUDPS_SINGLE)
4441                 dst->state = PFUDPS_MULTIPLE;
4442
4443         /* update expire time */
4444         (*state)->expire = time_uptime;
4445         if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4446                 (*state)->timeout = PFTM_UDP_MULTIPLE;
4447         else
4448                 (*state)->timeout = PFTM_UDP_SINGLE;
4449
4450         /* translate source/destination address, if necessary */
4451         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4452                 struct pf_state_key *nk = (*state)->key[pd->didx];
4453
4454                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4455                     nk->port[pd->sidx] != uh->uh_sport)
4456                         pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
4457                             &uh->uh_sum, &nk->addr[pd->sidx],
4458                             nk->port[pd->sidx], 1, pd->af);
4459
4460                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4461                     nk->port[pd->didx] != uh->uh_dport)
4462                         pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
4463                             &uh->uh_sum, &nk->addr[pd->didx],
4464                             nk->port[pd->didx], 1, pd->af);
4465                 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4466         }
4467
4468         return (PF_PASS);
4469 }
4470
4471 static int
4472 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
4473     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4474 {
4475         struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
4476         u_int16_t        icmpid = 0, *icmpsum;
4477         u_int8_t         icmptype;
4478         int              state_icmp = 0;
4479         struct pf_state_key_cmp key;
4480
4481         bzero(&key, sizeof(key));
4482         switch (pd->proto) {
4483 #ifdef INET
4484         case IPPROTO_ICMP:
4485                 icmptype = pd->hdr.icmp->icmp_type;
4486                 icmpid = pd->hdr.icmp->icmp_id;
4487                 icmpsum = &pd->hdr.icmp->icmp_cksum;
4488
4489                 if (icmptype == ICMP_UNREACH ||
4490                     icmptype == ICMP_SOURCEQUENCH ||
4491                     icmptype == ICMP_REDIRECT ||
4492                     icmptype == ICMP_TIMXCEED ||
4493                     icmptype == ICMP_PARAMPROB)
4494                         state_icmp++;
4495                 break;
4496 #endif /* INET */
4497 #ifdef INET6
4498         case IPPROTO_ICMPV6:
4499                 icmptype = pd->hdr.icmp6->icmp6_type;
4500                 icmpid = pd->hdr.icmp6->icmp6_id;
4501                 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4502
4503                 if (icmptype == ICMP6_DST_UNREACH ||
4504                     icmptype == ICMP6_PACKET_TOO_BIG ||
4505                     icmptype == ICMP6_TIME_EXCEEDED ||
4506                     icmptype == ICMP6_PARAM_PROB)
4507                         state_icmp++;
4508                 break;
4509 #endif /* INET6 */
4510         }
4511
4512         if (!state_icmp) {
4513
4514                 /*
4515                  * ICMP query/reply message not related to a TCP/UDP packet.
4516                  * Search for an ICMP state.
4517                  */
4518                 key.af = pd->af;
4519                 key.proto = pd->proto;
4520                 key.port[0] = key.port[1] = icmpid;
4521                 if (direction == PF_IN) {       /* wire side, straight */
4522                         PF_ACPY(&key.addr[0], pd->src, key.af);
4523                         PF_ACPY(&key.addr[1], pd->dst, key.af);
4524                 } else {                        /* stack side, reverse */
4525                         PF_ACPY(&key.addr[1], pd->src, key.af);
4526                         PF_ACPY(&key.addr[0], pd->dst, key.af);
4527                 }
4528
4529                 STATE_LOOKUP(kif, &key, direction, *state, pd);
4530
4531                 (*state)->expire = time_uptime;
4532                 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4533
4534                 /* translate source/destination address, if necessary */
4535                 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4536                         struct pf_state_key *nk = (*state)->key[pd->didx];
4537
4538                         switch (pd->af) {
4539 #ifdef INET
4540                         case AF_INET:
4541                                 if (PF_ANEQ(pd->src,
4542                                     &nk->addr[pd->sidx], AF_INET))
4543                                         pf_change_a(&saddr->v4.s_addr,
4544                                             pd->ip_sum,
4545                                             nk->addr[pd->sidx].v4.s_addr, 0);
4546
4547                                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4548                                     AF_INET))
4549                                         pf_change_a(&daddr->v4.s_addr,
4550                                             pd->ip_sum,
4551                                             nk->addr[pd->didx].v4.s_addr, 0);
4552
4553                                 if (nk->port[0] !=
4554                                     pd->hdr.icmp->icmp_id) {
4555                                         pd->hdr.icmp->icmp_cksum =
4556                                             pf_cksum_fixup(
4557                                             pd->hdr.icmp->icmp_cksum, icmpid,
4558                                             nk->port[pd->sidx], 0);
4559                                         pd->hdr.icmp->icmp_id =
4560                                             nk->port[pd->sidx];
4561                                 }
4562
4563                                 m_copyback(m, off, ICMP_MINLEN,
4564                                     (caddr_t )pd->hdr.icmp);
4565                                 break;
4566 #endif /* INET */
4567 #ifdef INET6
4568                         case AF_INET6:
4569                                 if (PF_ANEQ(pd->src,
4570                                     &nk->addr[pd->sidx], AF_INET6))
4571                                         pf_change_a6(saddr,
4572                                             &pd->hdr.icmp6->icmp6_cksum,
4573                                             &nk->addr[pd->sidx], 0);
4574
4575                                 if (PF_ANEQ(pd->dst,
4576                                     &nk->addr[pd->didx], AF_INET6))
4577                                         pf_change_a6(daddr,
4578                                             &pd->hdr.icmp6->icmp6_cksum,
4579                                             &nk->addr[pd->didx], 0);
4580
4581                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4582                                     (caddr_t )pd->hdr.icmp6);
4583                                 break;
4584 #endif /* INET6 */
4585                         }
4586                 }
4587                 return (PF_PASS);
4588
4589         } else {
4590                 /*
4591                  * ICMP error message in response to a TCP/UDP packet.
4592                  * Extract the inner TCP/UDP header and search for that state.
4593                  */
4594
4595                 struct pf_pdesc pd2;
4596                 bzero(&pd2, sizeof pd2);
4597 #ifdef INET
4598                 struct ip       h2;
4599 #endif /* INET */
4600 #ifdef INET6
4601                 struct ip6_hdr  h2_6;
4602                 int             terminal = 0;
4603 #endif /* INET6 */
4604                 int             ipoff2 = 0;
4605                 int             off2 = 0;
4606
4607                 pd2.af = pd->af;
4608                 /* Payload packet is from the opposite direction. */
4609                 pd2.sidx = (direction == PF_IN) ? 1 : 0;
4610                 pd2.didx = (direction == PF_IN) ? 0 : 1;
4611                 switch (pd->af) {
4612 #ifdef INET
4613                 case AF_INET:
4614                         /* offset of h2 in mbuf chain */
4615                         ipoff2 = off + ICMP_MINLEN;
4616
4617                         if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4618                             NULL, reason, pd2.af)) {
4619                                 DPFPRINTF(PF_DEBUG_MISC,
4620                                     ("pf: ICMP error message too short "
4621                                     "(ip)\n"));
4622                                 return (PF_DROP);
4623                         }
4624                         /*
4625                          * ICMP error messages don't refer to non-first
4626                          * fragments
4627                          */
4628                         if (h2.ip_off & htons(IP_OFFMASK)) {
4629                                 REASON_SET(reason, PFRES_FRAG);
4630                                 return (PF_DROP);
4631                         }
4632
4633                         /* offset of protocol header that follows h2 */
4634                         off2 = ipoff2 + (h2.ip_hl << 2);
4635
4636                         pd2.proto = h2.ip_p;
4637                         pd2.src = (struct pf_addr *)&h2.ip_src;
4638                         pd2.dst = (struct pf_addr *)&h2.ip_dst;
4639                         pd2.ip_sum = &h2.ip_sum;
4640                         break;
4641 #endif /* INET */
4642 #ifdef INET6
4643                 case AF_INET6:
4644                         ipoff2 = off + sizeof(struct icmp6_hdr);
4645
4646                         if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4647                             NULL, reason, pd2.af)) {
4648                                 DPFPRINTF(PF_DEBUG_MISC,
4649                                     ("pf: ICMP error message too short "
4650                                     "(ip6)\n"));
4651                                 return (PF_DROP);
4652                         }
4653                         pd2.proto = h2_6.ip6_nxt;
4654                         pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4655                         pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4656                         pd2.ip_sum = NULL;
4657                         off2 = ipoff2 + sizeof(h2_6);
4658                         do {
4659                                 switch (pd2.proto) {
4660                                 case IPPROTO_FRAGMENT:
4661                                         /*
4662                                          * ICMPv6 error messages for
4663                                          * non-first fragments
4664                                          */
4665                                         REASON_SET(reason, PFRES_FRAG);
4666                                         return (PF_DROP);
4667                                 case IPPROTO_AH:
4668                                 case IPPROTO_HOPOPTS:
4669                                 case IPPROTO_ROUTING:
4670                                 case IPPROTO_DSTOPTS: {
4671                                         /* get next header and header length */
4672                                         struct ip6_ext opt6;
4673
4674                                         if (!pf_pull_hdr(m, off2, &opt6,
4675                                             sizeof(opt6), NULL, reason,
4676                                             pd2.af)) {
4677                                                 DPFPRINTF(PF_DEBUG_MISC,
4678                                                     ("pf: ICMPv6 short opt\n"));
4679                                                 return (PF_DROP);
4680                                         }
4681                                         if (pd2.proto == IPPROTO_AH)
4682                                                 off2 += (opt6.ip6e_len + 2) * 4;
4683                                         else
4684                                                 off2 += (opt6.ip6e_len + 1) * 8;
4685                                         pd2.proto = opt6.ip6e_nxt;
4686                                         /* goto the next header */
4687                                         break;
4688                                 }
4689                                 default:
4690                                         terminal++;
4691                                         break;
4692                                 }
4693                         } while (!terminal);
4694                         break;
4695 #endif /* INET6 */
4696                 }
4697
4698                 switch (pd2.proto) {
4699                 case IPPROTO_TCP: {
4700                         struct tcphdr            th;
4701                         u_int32_t                seq;
4702                         struct pf_state_peer    *src, *dst;
4703                         u_int8_t                 dws;
4704                         int                      copyback = 0;
4705
4706                         /*
4707                          * Only the first 8 bytes of the TCP header can be
4708                          * expected. Don't access any TCP header fields after
4709                          * th_seq, an ackskew test is not possible.
4710                          */
4711                         if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4712                             pd2.af)) {
4713                                 DPFPRINTF(PF_DEBUG_MISC,
4714                                     ("pf: ICMP error message too short "
4715                                     "(tcp)\n"));
4716                                 return (PF_DROP);
4717                         }
4718
4719                         key.af = pd2.af;
4720                         key.proto = IPPROTO_TCP;
4721                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4722                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4723                         key.port[pd2.sidx] = th.th_sport;
4724                         key.port[pd2.didx] = th.th_dport;
4725
4726                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4727
4728                         if (direction == (*state)->direction) {
4729                                 src = &(*state)->dst;
4730                                 dst = &(*state)->src;
4731                         } else {
4732                                 src = &(*state)->src;
4733                                 dst = &(*state)->dst;
4734                         }
4735
4736                         if (src->wscale && dst->wscale)
4737                                 dws = dst->wscale & PF_WSCALE_MASK;
4738                         else
4739                                 dws = 0;
4740
4741                         /* Demodulate sequence number */
4742                         seq = ntohl(th.th_seq) - src->seqdiff;
4743                         if (src->seqdiff) {
4744                                 pf_change_a(&th.th_seq, icmpsum,
4745                                     htonl(seq), 0);
4746                                 copyback = 1;
4747                         }
4748
4749                         if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4750                             (!SEQ_GEQ(src->seqhi, seq) ||
4751                             !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4752                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4753                                         printf("pf: BAD ICMP %d:%d ",
4754                                             icmptype, pd->hdr.icmp->icmp_code);
4755                                         pf_print_host(pd->src, 0, pd->af);
4756                                         printf(" -> ");
4757                                         pf_print_host(pd->dst, 0, pd->af);
4758                                         printf(" state: ");
4759                                         pf_print_state(*state);
4760                                         printf(" seq=%u\n", seq);
4761                                 }
4762                                 REASON_SET(reason, PFRES_BADSTATE);
4763                                 return (PF_DROP);
4764                         } else {
4765                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4766                                         printf("pf: OK ICMP %d:%d ",
4767                                             icmptype, pd->hdr.icmp->icmp_code);
4768                                         pf_print_host(pd->src, 0, pd->af);
4769                                         printf(" -> ");
4770                                         pf_print_host(pd->dst, 0, pd->af);
4771                                         printf(" state: ");
4772                                         pf_print_state(*state);
4773                                         printf(" seq=%u\n", seq);
4774                                 }
4775                         }
4776
4777                         /* translate source/destination address, if necessary */
4778                         if ((*state)->key[PF_SK_WIRE] !=
4779                             (*state)->key[PF_SK_STACK]) {
4780                                 struct pf_state_key *nk =
4781                                     (*state)->key[pd->didx];
4782
4783                                 if (PF_ANEQ(pd2.src,
4784                                     &nk->addr[pd2.sidx], pd2.af) ||
4785                                     nk->port[pd2.sidx] != th.th_sport)
4786                                         pf_change_icmp(pd2.src, &th.th_sport,
4787                                             daddr, &nk->addr[pd2.sidx],
4788                                             nk->port[pd2.sidx], NULL,
4789                                             pd2.ip_sum, icmpsum,
4790                                             pd->ip_sum, 0, pd2.af);
4791
4792                                 if (PF_ANEQ(pd2.dst,
4793                                     &nk->addr[pd2.didx], pd2.af) ||
4794                                     nk->port[pd2.didx] != th.th_dport)
4795                                         pf_change_icmp(pd2.dst, &th.th_dport,
4796                                             NULL, /* XXX Inbound NAT? */
4797                                             &nk->addr[pd2.didx],
4798                                             nk->port[pd2.didx], NULL,
4799                                             pd2.ip_sum, icmpsum,
4800                                             pd->ip_sum, 0, pd2.af);
4801                                 copyback = 1;
4802                         }
4803
4804                         if (copyback) {
4805                                 switch (pd2.af) {
4806 #ifdef INET
4807                                 case AF_INET:
4808                                         m_copyback(m, off, ICMP_MINLEN,
4809                                             (caddr_t )pd->hdr.icmp);
4810                                         m_copyback(m, ipoff2, sizeof(h2),
4811                                             (caddr_t )&h2);
4812                                         break;
4813 #endif /* INET */
4814 #ifdef INET6
4815                                 case AF_INET6:
4816                                         m_copyback(m, off,
4817                                             sizeof(struct icmp6_hdr),
4818                                             (caddr_t )pd->hdr.icmp6);
4819                                         m_copyback(m, ipoff2, sizeof(h2_6),
4820                                             (caddr_t )&h2_6);
4821                                         break;
4822 #endif /* INET6 */
4823                                 }
4824                                 m_copyback(m, off2, 8, (caddr_t)&th);
4825                         }
4826
4827                         return (PF_PASS);
4828                         break;
4829                 }
4830                 case IPPROTO_UDP: {
4831                         struct udphdr           uh;
4832
4833                         if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
4834                             NULL, reason, pd2.af)) {
4835                                 DPFPRINTF(PF_DEBUG_MISC,
4836                                     ("pf: ICMP error message too short "
4837                                     "(udp)\n"));
4838                                 return (PF_DROP);
4839                         }
4840
4841                         key.af = pd2.af;
4842                         key.proto = IPPROTO_UDP;
4843                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4844                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4845                         key.port[pd2.sidx] = uh.uh_sport;
4846                         key.port[pd2.didx] = uh.uh_dport;
4847
4848                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4849
4850                         /* translate source/destination address, if necessary */
4851                         if ((*state)->key[PF_SK_WIRE] !=
4852                             (*state)->key[PF_SK_STACK]) {
4853                                 struct pf_state_key *nk =
4854                                     (*state)->key[pd->didx];
4855
4856                                 if (PF_ANEQ(pd2.src,
4857                                     &nk->addr[pd2.sidx], pd2.af) ||
4858                                     nk->port[pd2.sidx] != uh.uh_sport)
4859                                         pf_change_icmp(pd2.src, &uh.uh_sport,
4860                                             daddr, &nk->addr[pd2.sidx],
4861                                             nk->port[pd2.sidx], &uh.uh_sum,
4862                                             pd2.ip_sum, icmpsum,
4863                                             pd->ip_sum, 1, pd2.af);
4864
4865                                 if (PF_ANEQ(pd2.dst,
4866                                     &nk->addr[pd2.didx], pd2.af) ||
4867                                     nk->port[pd2.didx] != uh.uh_dport)
4868                                         pf_change_icmp(pd2.dst, &uh.uh_dport,
4869                                             NULL, /* XXX Inbound NAT? */
4870                                             &nk->addr[pd2.didx],
4871                                             nk->port[pd2.didx], &uh.uh_sum,
4872                                             pd2.ip_sum, icmpsum,
4873                                             pd->ip_sum, 1, pd2.af);
4874
4875                                 switch (pd2.af) {
4876 #ifdef INET
4877                                 case AF_INET:
4878                                         m_copyback(m, off, ICMP_MINLEN,
4879                                             (caddr_t )pd->hdr.icmp);
4880                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4881                                         break;
4882 #endif /* INET */
4883 #ifdef INET6
4884                                 case AF_INET6:
4885                                         m_copyback(m, off,
4886                                             sizeof(struct icmp6_hdr),
4887                                             (caddr_t )pd->hdr.icmp6);
4888                                         m_copyback(m, ipoff2, sizeof(h2_6),
4889                                             (caddr_t )&h2_6);
4890                                         break;
4891 #endif /* INET6 */
4892                                 }
4893                                 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
4894                         }
4895                         return (PF_PASS);
4896                         break;
4897                 }
4898 #ifdef INET
4899                 case IPPROTO_ICMP: {
4900                         struct icmp             iih;
4901
4902                         if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
4903                             NULL, reason, pd2.af)) {
4904                                 DPFPRINTF(PF_DEBUG_MISC,
4905                                     ("pf: ICMP error message too short i"
4906                                     "(icmp)\n"));
4907                                 return (PF_DROP);
4908                         }
4909
4910                         key.af = pd2.af;
4911                         key.proto = IPPROTO_ICMP;
4912                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4913                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4914                         key.port[0] = key.port[1] = iih.icmp_id;
4915
4916                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4917
4918                         /* translate source/destination address, if necessary */
4919                         if ((*state)->key[PF_SK_WIRE] !=
4920                             (*state)->key[PF_SK_STACK]) {
4921                                 struct pf_state_key *nk =
4922                                     (*state)->key[pd->didx];
4923
4924                                 if (PF_ANEQ(pd2.src,
4925                                     &nk->addr[pd2.sidx], pd2.af) ||
4926                                     nk->port[pd2.sidx] != iih.icmp_id)
4927                                         pf_change_icmp(pd2.src, &iih.icmp_id,
4928                                             daddr, &nk->addr[pd2.sidx],
4929                                             nk->port[pd2.sidx], NULL,
4930                                             pd2.ip_sum, icmpsum,
4931                                             pd->ip_sum, 0, AF_INET);
4932
4933                                 if (PF_ANEQ(pd2.dst,
4934                                     &nk->addr[pd2.didx], pd2.af) ||
4935                                     nk->port[pd2.didx] != iih.icmp_id)
4936                                         pf_change_icmp(pd2.dst, &iih.icmp_id,
4937                                             NULL, /* XXX Inbound NAT? */
4938                                             &nk->addr[pd2.didx],
4939                                             nk->port[pd2.didx], NULL,
4940                                             pd2.ip_sum, icmpsum,
4941                                             pd->ip_sum, 0, AF_INET);
4942
4943                                 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
4944                                 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4945                                 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
4946                         }
4947                         return (PF_PASS);
4948                         break;
4949                 }
4950 #endif /* INET */
4951 #ifdef INET6
4952                 case IPPROTO_ICMPV6: {
4953                         struct icmp6_hdr        iih;
4954
4955                         if (!pf_pull_hdr(m, off2, &iih,
4956                             sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
4957                                 DPFPRINTF(PF_DEBUG_MISC,
4958                                     ("pf: ICMP error message too short "
4959                                     "(icmp6)\n"));
4960                                 return (PF_DROP);
4961                         }
4962
4963                         key.af = pd2.af;
4964                         key.proto = IPPROTO_ICMPV6;
4965                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4966                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4967                         key.port[0] = key.port[1] = iih.icmp6_id;
4968
4969                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4970
4971                         /* translate source/destination address, if necessary */
4972                         if ((*state)->key[PF_SK_WIRE] !=
4973                             (*state)->key[PF_SK_STACK]) {
4974                                 struct pf_state_key *nk =
4975                                     (*state)->key[pd->didx];
4976
4977                                 if (PF_ANEQ(pd2.src,
4978                                     &nk->addr[pd2.sidx], pd2.af) ||
4979                                     nk->port[pd2.sidx] != iih.icmp6_id)
4980                                         pf_change_icmp(pd2.src, &iih.icmp6_id,
4981                                             daddr, &nk->addr[pd2.sidx],
4982                                             nk->port[pd2.sidx], NULL,
4983                                             pd2.ip_sum, icmpsum,
4984                                             pd->ip_sum, 0, AF_INET6);
4985
4986                                 if (PF_ANEQ(pd2.dst,
4987                                     &nk->addr[pd2.didx], pd2.af) ||
4988                                     nk->port[pd2.didx] != iih.icmp6_id)
4989                                         pf_change_icmp(pd2.dst, &iih.icmp6_id,
4990                                             NULL, /* XXX Inbound NAT? */
4991                                             &nk->addr[pd2.didx],
4992                                             nk->port[pd2.didx], NULL,
4993                                             pd2.ip_sum, icmpsum,
4994                                             pd->ip_sum, 0, AF_INET6);
4995
4996                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4997                                     (caddr_t)pd->hdr.icmp6);
4998                                 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
4999                                 m_copyback(m, off2, sizeof(struct icmp6_hdr),
5000                                     (caddr_t)&iih);
5001                         }
5002                         return (PF_PASS);
5003                         break;
5004                 }
5005 #endif /* INET6 */
5006                 default: {
5007                         key.af = pd2.af;
5008                         key.proto = pd2.proto;
5009                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5010                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5011                         key.port[0] = key.port[1] = 0;
5012
5013                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5014
5015                         /* translate source/destination address, if necessary */
5016                         if ((*state)->key[PF_SK_WIRE] !=
5017                             (*state)->key[PF_SK_STACK]) {
5018                                 struct pf_state_key *nk =
5019                                     (*state)->key[pd->didx];
5020
5021                                 if (PF_ANEQ(pd2.src,
5022                                     &nk->addr[pd2.sidx], pd2.af))
5023                                         pf_change_icmp(pd2.src, NULL, daddr,
5024                                             &nk->addr[pd2.sidx], 0, NULL,
5025                                             pd2.ip_sum, icmpsum,
5026                                             pd->ip_sum, 0, pd2.af);
5027
5028                                 if (PF_ANEQ(pd2.dst,
5029                                     &nk->addr[pd2.didx], pd2.af))
5030                                         pf_change_icmp(pd2.src, NULL,
5031                                             NULL, /* XXX Inbound NAT? */
5032                                             &nk->addr[pd2.didx], 0, NULL,
5033                                             pd2.ip_sum, icmpsum,
5034                                             pd->ip_sum, 0, pd2.af);
5035
5036                                 switch (pd2.af) {
5037 #ifdef INET
5038                                 case AF_INET:
5039                                         m_copyback(m, off, ICMP_MINLEN,
5040                                             (caddr_t)pd->hdr.icmp);
5041                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5042                                         break;
5043 #endif /* INET */
5044 #ifdef INET6
5045                                 case AF_INET6:
5046                                         m_copyback(m, off,
5047                                             sizeof(struct icmp6_hdr),
5048                                             (caddr_t )pd->hdr.icmp6);
5049                                         m_copyback(m, ipoff2, sizeof(h2_6),
5050                                             (caddr_t )&h2_6);
5051                                         break;
5052 #endif /* INET6 */
5053                                 }
5054                         }
5055                         return (PF_PASS);
5056                         break;
5057                 }
5058                 }
5059         }
5060 }
5061
5062 static int
5063 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
5064     struct mbuf *m, struct pf_pdesc *pd)
5065 {
5066         struct pf_state_peer    *src, *dst;
5067         struct pf_state_key_cmp  key;
5068
5069         bzero(&key, sizeof(key));
5070         key.af = pd->af;
5071         key.proto = pd->proto;
5072         if (direction == PF_IN) {
5073                 PF_ACPY(&key.addr[0], pd->src, key.af);
5074                 PF_ACPY(&key.addr[1], pd->dst, key.af);
5075                 key.port[0] = key.port[1] = 0;
5076         } else {
5077                 PF_ACPY(&key.addr[1], pd->src, key.af);
5078                 PF_ACPY(&key.addr[0], pd->dst, key.af);
5079                 key.port[1] = key.port[0] = 0;
5080         }
5081
5082         STATE_LOOKUP(kif, &key, direction, *state, pd);
5083
5084         if (direction == (*state)->direction) {
5085                 src = &(*state)->src;
5086                 dst = &(*state)->dst;
5087         } else {
5088                 src = &(*state)->dst;
5089                 dst = &(*state)->src;
5090         }
5091
5092         /* update states */
5093         if (src->state < PFOTHERS_SINGLE)
5094                 src->state = PFOTHERS_SINGLE;
5095         if (dst->state == PFOTHERS_SINGLE)
5096                 dst->state = PFOTHERS_MULTIPLE;
5097
5098         /* update expire time */
5099         (*state)->expire = time_uptime;
5100         if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5101                 (*state)->timeout = PFTM_OTHER_MULTIPLE;
5102         else
5103                 (*state)->timeout = PFTM_OTHER_SINGLE;
5104
5105         /* translate source/destination address, if necessary */
5106         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5107                 struct pf_state_key *nk = (*state)->key[pd->didx];
5108
5109                 KASSERT(nk, ("%s: nk is null", __func__));
5110                 KASSERT(pd, ("%s: pd is null", __func__));
5111                 KASSERT(pd->src, ("%s: pd->src is null", __func__));
5112                 KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5113                 switch (pd->af) {
5114 #ifdef INET
5115                 case AF_INET:
5116                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5117                                 pf_change_a(&pd->src->v4.s_addr,
5118                                     pd->ip_sum,
5119                                     nk->addr[pd->sidx].v4.s_addr,
5120                                     0);
5121
5122
5123                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5124                                 pf_change_a(&pd->dst->v4.s_addr,
5125                                     pd->ip_sum,
5126                                     nk->addr[pd->didx].v4.s_addr,
5127                                     0);
5128
5129                                 break;
5130 #endif /* INET */
5131 #ifdef INET6
5132                 case AF_INET6:
5133                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5134                                 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5135
5136                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5137                                 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5138 #endif /* INET6 */
5139                 }
5140         }
5141         return (PF_PASS);
5142 }
5143
5144 /*
5145  * ipoff and off are measured from the start of the mbuf chain.
5146  * h must be at "ipoff" on the mbuf chain.
5147  */
5148 void *
5149 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5150     u_short *actionp, u_short *reasonp, sa_family_t af)
5151 {
5152         switch (af) {
5153 #ifdef INET
5154         case AF_INET: {
5155                 struct ip       *h = mtod(m, struct ip *);
5156                 u_int16_t        fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5157
5158                 if (fragoff) {
5159                         if (fragoff >= len)
5160                                 ACTION_SET(actionp, PF_PASS);
5161                         else {
5162                                 ACTION_SET(actionp, PF_DROP);
5163                                 REASON_SET(reasonp, PFRES_FRAG);
5164                         }
5165                         return (NULL);
5166                 }
5167                 if (m->m_pkthdr.len < off + len ||
5168                     ntohs(h->ip_len) < off + len) {
5169                         ACTION_SET(actionp, PF_DROP);
5170                         REASON_SET(reasonp, PFRES_SHORT);
5171                         return (NULL);
5172                 }
5173                 break;
5174         }
5175 #endif /* INET */
5176 #ifdef INET6
5177         case AF_INET6: {
5178                 struct ip6_hdr  *h = mtod(m, struct ip6_hdr *);
5179
5180                 if (m->m_pkthdr.len < off + len ||
5181                     (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5182                     (unsigned)(off + len)) {
5183                         ACTION_SET(actionp, PF_DROP);
5184                         REASON_SET(reasonp, PFRES_SHORT);
5185                         return (NULL);
5186                 }
5187                 break;
5188         }
5189 #endif /* INET6 */
5190         }
5191         m_copydata(m, off, len, p);
5192         return (p);
5193 }
5194
5195 int
5196 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5197     int rtableid)
5198 {
5199 #ifdef RADIX_MPATH
5200         struct radix_node_head  *rnh;
5201 #endif
5202         struct sockaddr_in      *dst;
5203         int                      ret = 1;
5204         int                      check_mpath;
5205 #ifdef INET6
5206         struct sockaddr_in6     *dst6;
5207         struct route_in6         ro;
5208 #else
5209         struct route             ro;
5210 #endif
5211         struct radix_node       *rn;
5212         struct rtentry          *rt;
5213         struct ifnet            *ifp;
5214
5215         check_mpath = 0;
5216 #ifdef RADIX_MPATH
5217         /* XXX: stick to table 0 for now */
5218         rnh = rt_tables_get_rnh(0, af);
5219         if (rnh != NULL && rn_mpath_capable(rnh))
5220                 check_mpath = 1;
5221 #endif
5222         bzero(&ro, sizeof(ro));
5223         switch (af) {
5224         case AF_INET:
5225                 dst = satosin(&ro.ro_dst);
5226                 dst->sin_family = AF_INET;
5227                 dst->sin_len = sizeof(*dst);
5228                 dst->sin_addr = addr->v4;
5229                 break;
5230 #ifdef INET6
5231         case AF_INET6:
5232                 /*
5233                  * Skip check for addresses with embedded interface scope,
5234                  * as they would always match anyway.
5235                  */
5236                 if (IN6_IS_SCOPE_EMBED(&addr->v6))
5237                         goto out;
5238                 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5239                 dst6->sin6_family = AF_INET6;
5240                 dst6->sin6_len = sizeof(*dst6);
5241                 dst6->sin6_addr = addr->v6;
5242                 break;
5243 #endif /* INET6 */
5244         default:
5245                 return (0);
5246         }
5247
5248         /* Skip checks for ipsec interfaces */
5249         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5250                 goto out;
5251
5252         switch (af) {
5253 #ifdef INET6
5254         case AF_INET6:
5255                 in6_rtalloc_ign(&ro, 0, rtableid);
5256                 break;
5257 #endif
5258 #ifdef INET
5259         case AF_INET:
5260                 in_rtalloc_ign((struct route *)&ro, 0, rtableid);
5261                 break;
5262 #endif
5263         default:
5264                 rtalloc_ign((struct route *)&ro, 0);    /* No/default FIB. */
5265                 break;
5266         }
5267
5268         if (ro.ro_rt != NULL) {
5269                 /* No interface given, this is a no-route check */
5270                 if (kif == NULL)
5271                         goto out;
5272
5273                 if (kif->pfik_ifp == NULL) {
5274                         ret = 0;
5275                         goto out;
5276                 }
5277
5278                 /* Perform uRPF check if passed input interface */
5279                 ret = 0;
5280                 rn = (struct radix_node *)ro.ro_rt;
5281                 do {
5282                         rt = (struct rtentry *)rn;
5283                         ifp = rt->rt_ifp;
5284
5285                         if (kif->pfik_ifp == ifp)
5286                                 ret = 1;
5287 #ifdef RADIX_MPATH
5288                         rn = rn_mpath_next(rn);
5289 #endif
5290                 } while (check_mpath == 1 && rn != NULL && ret == 0);
5291         } else
5292                 ret = 0;
5293 out:
5294         if (ro.ro_rt != NULL)
5295                 RTFREE(ro.ro_rt);
5296         return (ret);
5297 }
5298
5299 #ifdef INET
5300 static void
5301 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5302     struct pf_state *s, struct pf_pdesc *pd)
5303 {
5304         struct mbuf             *m0, *m1;
5305         struct sockaddr_in      dst;
5306         struct ip               *ip;
5307         struct ifnet            *ifp = NULL;
5308         struct pf_addr           naddr;
5309         struct pf_src_node      *sn = NULL;
5310         int                      error = 0;
5311         uint16_t                 ip_len, ip_off;
5312
5313         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5314         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5315             __func__));
5316
5317         if ((pd->pf_mtag == NULL &&
5318             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5319             pd->pf_mtag->routed++ > 3) {
5320                 m0 = *m;
5321                 *m = NULL;
5322                 goto bad_locked;
5323         }
5324
5325         if (r->rt == PF_DUPTO) {
5326                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5327                         if (s)
5328                                 PF_STATE_UNLOCK(s);
5329                         return;
5330                 }
5331         } else {
5332                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5333                         if (s)
5334                                 PF_STATE_UNLOCK(s);
5335                         return;
5336                 }
5337                 m0 = *m;
5338         }
5339
5340         ip = mtod(m0, struct ip *);
5341
5342         bzero(&dst, sizeof(dst));
5343         dst.sin_family = AF_INET;
5344         dst.sin_len = sizeof(dst);
5345         dst.sin_addr = ip->ip_dst;
5346
5347         if (r->rt == PF_FASTROUTE) {
5348                 struct rtentry *rt;
5349
5350                 if (s)
5351                         PF_STATE_UNLOCK(s);
5352                 rt = rtalloc1_fib(sintosa(&dst), 0, 0, M_GETFIB(m0));
5353                 if (rt == NULL) {
5354                         KMOD_IPSTAT_INC(ips_noroute);
5355                         error = EHOSTUNREACH;
5356                         goto bad;
5357                 }
5358
5359                 ifp = rt->rt_ifp;
5360                 counter_u64_add(rt->rt_pksent, 1);
5361
5362                 if (rt->rt_flags & RTF_GATEWAY)
5363                         bcopy(satosin(rt->rt_gateway), &dst, sizeof(dst));
5364                 RTFREE_LOCKED(rt);
5365         } else {
5366                 if (TAILQ_EMPTY(&r->rpool.list)) {
5367                         DPFPRINTF(PF_DEBUG_URGENT,
5368                             ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5369                         goto bad_locked;
5370                 }
5371                 if (s == NULL) {
5372                         pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5373                             &naddr, NULL, &sn);
5374                         if (!PF_AZERO(&naddr, AF_INET))
5375                                 dst.sin_addr.s_addr = naddr.v4.s_addr;
5376                         ifp = r->rpool.cur->kif ?
5377                             r->rpool.cur->kif->pfik_ifp : NULL;
5378                 } else {
5379                         if (!PF_AZERO(&s->rt_addr, AF_INET))
5380                                 dst.sin_addr.s_addr =
5381                                     s->rt_addr.v4.s_addr;
5382                         ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5383                         PF_STATE_UNLOCK(s);
5384                 }
5385         }
5386         if (ifp == NULL)
5387                 goto bad;
5388
5389         if (oifp != ifp) {
5390                 if (pf_test(PF_OUT, ifp, &m0, NULL) != PF_PASS)
5391                         goto bad;
5392                 else if (m0 == NULL)
5393                         goto done;
5394                 if (m0->m_len < sizeof(struct ip)) {
5395                         DPFPRINTF(PF_DEBUG_URGENT,
5396                             ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5397                         goto bad;
5398                 }
5399                 ip = mtod(m0, struct ip *);
5400         }
5401
5402         if (ifp->if_flags & IFF_LOOPBACK)
5403                 m0->m_flags |= M_SKIP_FIREWALL;
5404
5405         ip_len = ntohs(ip->ip_len);
5406         ip_off = ntohs(ip->ip_off);
5407
5408         /* Copied from FreeBSD 10.0-CURRENT ip_output. */
5409         m0->m_pkthdr.csum_flags |= CSUM_IP;
5410         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5411                 in_delayed_cksum(m0);
5412                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5413         }
5414 #ifdef SCTP
5415         if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5416                 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
5417                 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5418         }
5419 #endif
5420
5421         /*
5422          * If small enough for interface, or the interface will take
5423          * care of the fragmentation for us, we can just send directly.
5424          */
5425         if (ip_len <= ifp->if_mtu ||
5426             (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 ||
5427             ((ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) {
5428                 ip->ip_sum = 0;
5429                 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5430                         ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5431                         m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5432                 }
5433                 m_clrprotoflags(m0);    /* Avoid confusing lower layers. */
5434                 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5435                 goto done;
5436         }
5437
5438         /* Balk when DF bit is set or the interface didn't support TSO. */
5439         if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5440                 error = EMSGSIZE;
5441                 KMOD_IPSTAT_INC(ips_cantfrag);
5442                 if (r->rt != PF_DUPTO) {
5443                         icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5444                             ifp->if_mtu);
5445                         goto done;
5446                 } else
5447                         goto bad;
5448         }
5449
5450         error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5451         if (error)
5452                 goto bad;
5453
5454         for (; m0; m0 = m1) {
5455                 m1 = m0->m_nextpkt;
5456                 m0->m_nextpkt = NULL;
5457                 if (error == 0) {
5458                         m_clrprotoflags(m0);
5459                         error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5460                 } else
5461                         m_freem(m0);
5462         }
5463
5464         if (error == 0)
5465                 KMOD_IPSTAT_INC(ips_fragmented);
5466
5467 done:
5468         if (r->rt != PF_DUPTO)
5469                 *m = NULL;
5470         return;
5471
5472 bad_locked:
5473         if (s)
5474                 PF_STATE_UNLOCK(s);
5475 bad:
5476         m_freem(m0);
5477         goto done;
5478 }
5479 #endif /* INET */
5480
5481 #ifdef INET6
5482 static void
5483 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5484     struct pf_state *s, struct pf_pdesc *pd)
5485 {
5486         struct mbuf             *m0;
5487         struct sockaddr_in6     dst;
5488         struct ip6_hdr          *ip6;
5489         struct ifnet            *ifp = NULL;
5490         struct pf_addr           naddr;
5491         struct pf_src_node      *sn = NULL;
5492
5493         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5494         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5495             __func__));
5496
5497         if ((pd->pf_mtag == NULL &&
5498             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5499             pd->pf_mtag->routed++ > 3) {
5500                 m0 = *m;
5501                 *m = NULL;
5502                 goto bad_locked;
5503         }
5504
5505         if (r->rt == PF_DUPTO) {
5506                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5507                         if (s)
5508                                 PF_STATE_UNLOCK(s);
5509                         return;
5510                 }
5511         } else {
5512                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5513                         if (s)
5514                                 PF_STATE_UNLOCK(s);
5515                         return;
5516                 }
5517                 m0 = *m;
5518         }
5519
5520         ip6 = mtod(m0, struct ip6_hdr *);
5521
5522         bzero(&dst, sizeof(dst));
5523         dst.sin6_family = AF_INET6;
5524         dst.sin6_len = sizeof(dst);
5525         dst.sin6_addr = ip6->ip6_dst;
5526
5527         /* Cheat. XXX why only in the v6 case??? */
5528         if (r->rt == PF_FASTROUTE) {
5529                 if (s)
5530                         PF_STATE_UNLOCK(s);
5531                 m0->m_flags |= M_SKIP_FIREWALL;
5532                 ip6_output(m0, NULL, NULL, 0, NULL, NULL, NULL);
5533                 *m = NULL;
5534                 return;
5535         }
5536
5537         if (TAILQ_EMPTY(&r->rpool.list)) {
5538                 DPFPRINTF(PF_DEBUG_URGENT,
5539                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5540                 goto bad_locked;
5541         }
5542         if (s == NULL) {
5543                 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5544                     &naddr, NULL, &sn);
5545                 if (!PF_AZERO(&naddr, AF_INET6))
5546                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5547                             &naddr, AF_INET6);
5548                 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5549         } else {
5550                 if (!PF_AZERO(&s->rt_addr, AF_INET6))
5551                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5552                             &s->rt_addr, AF_INET6);
5553                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5554         }
5555
5556         if (s)
5557                 PF_STATE_UNLOCK(s);
5558
5559         if (ifp == NULL)
5560                 goto bad;
5561
5562         if (oifp != ifp) {
5563                 if (pf_test6(PF_FWD, ifp, &m0, NULL) != PF_PASS)
5564                         goto bad;
5565                 else if (m0 == NULL)
5566                         goto done;
5567                 if (m0->m_len < sizeof(struct ip6_hdr)) {
5568                         DPFPRINTF(PF_DEBUG_URGENT,
5569                             ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5570                             __func__));
5571                         goto bad;
5572                 }
5573                 ip6 = mtod(m0, struct ip6_hdr *);
5574         }
5575
5576         if (ifp->if_flags & IFF_LOOPBACK)
5577                 m0->m_flags |= M_SKIP_FIREWALL;
5578
5579         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
5580             ~ifp->if_hwassist) {
5581                 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
5582                 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
5583                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
5584         }
5585
5586         /*
5587          * If the packet is too large for the outgoing interface,
5588          * send back an icmp6 error.
5589          */
5590         if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5591                 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5592         if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5593                 nd6_output(ifp, ifp, m0, &dst, NULL);
5594         else {
5595                 in6_ifstat_inc(ifp, ifs6_in_toobig);
5596                 if (r->rt != PF_DUPTO)
5597                         icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5598                 else
5599                         goto bad;
5600         }
5601
5602 done:
5603         if (r->rt != PF_DUPTO)
5604                 *m = NULL;
5605         return;
5606
5607 bad_locked:
5608         if (s)
5609                 PF_STATE_UNLOCK(s);
5610 bad:
5611         m_freem(m0);
5612         goto done;
5613 }
5614 #endif /* INET6 */
5615
5616 /*
5617  * FreeBSD supports cksum offloads for the following drivers.
5618  *  em(4), fxp(4), ixgb(4), lge(4), ndis(4), nge(4), re(4),
5619  *   ti(4), txp(4), xl(4)
5620  *
5621  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5622  *  network driver performed cksum including pseudo header, need to verify
5623  *   csum_data
5624  * CSUM_DATA_VALID :
5625  *  network driver performed cksum, needs to additional pseudo header
5626  *  cksum computation with partial csum_data(i.e. lack of H/W support for
5627  *  pseudo header, for instance hme(4), sk(4) and possibly gem(4))
5628  *
5629  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5630  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5631  * TCP/UDP layer.
5632  * Also, set csum_data to 0xffff to force cksum validation.
5633  */
5634 static int
5635 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5636 {
5637         u_int16_t sum = 0;
5638         int hw_assist = 0;
5639         struct ip *ip;
5640
5641         if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5642                 return (1);
5643         if (m->m_pkthdr.len < off + len)
5644                 return (1);
5645
5646         switch (p) {
5647         case IPPROTO_TCP:
5648                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5649                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5650                                 sum = m->m_pkthdr.csum_data;
5651                         } else {
5652                                 ip = mtod(m, struct ip *);
5653                                 sum = in_pseudo(ip->ip_src.s_addr,
5654                                 ip->ip_dst.s_addr, htonl((u_short)len +
5655                                 m->m_pkthdr.csum_data + IPPROTO_TCP));
5656                         }
5657                         sum ^= 0xffff;
5658                         ++hw_assist;
5659                 }
5660                 break;
5661         case IPPROTO_UDP:
5662                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5663                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5664                                 sum = m->m_pkthdr.csum_data;
5665                         } else {
5666                                 ip = mtod(m, struct ip *);
5667                                 sum = in_pseudo(ip->ip_src.s_addr,
5668                                 ip->ip_dst.s_addr, htonl((u_short)len +
5669                                 m->m_pkthdr.csum_data + IPPROTO_UDP));
5670                         }
5671                         sum ^= 0xffff;
5672                         ++hw_assist;
5673                 }
5674                 break;
5675         case IPPROTO_ICMP:
5676 #ifdef INET6
5677         case IPPROTO_ICMPV6:
5678 #endif /* INET6 */
5679                 break;
5680         default:
5681                 return (1);
5682         }
5683
5684         if (!hw_assist) {
5685                 switch (af) {
5686                 case AF_INET:
5687                         if (p == IPPROTO_ICMP) {
5688                                 if (m->m_len < off)
5689                                         return (1);
5690                                 m->m_data += off;
5691                                 m->m_len -= off;
5692                                 sum = in_cksum(m, len);
5693                                 m->m_data -= off;
5694                                 m->m_len += off;
5695                         } else {
5696                                 if (m->m_len < sizeof(struct ip))
5697                                         return (1);
5698                                 sum = in4_cksum(m, p, off, len);
5699                         }
5700                         break;
5701 #ifdef INET6
5702                 case AF_INET6:
5703                         if (m->m_len < sizeof(struct ip6_hdr))
5704                                 return (1);
5705                         sum = in6_cksum(m, p, off, len);
5706                         break;
5707 #endif /* INET6 */
5708                 default:
5709                         return (1);
5710                 }
5711         }
5712         if (sum) {
5713                 switch (p) {
5714                 case IPPROTO_TCP:
5715                     {
5716                         KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5717                         break;
5718                     }
5719                 case IPPROTO_UDP:
5720                     {
5721                         KMOD_UDPSTAT_INC(udps_badsum);
5722                         break;
5723                     }
5724 #ifdef INET
5725                 case IPPROTO_ICMP:
5726                     {
5727                         KMOD_ICMPSTAT_INC(icps_checksum);
5728                         break;
5729                     }
5730 #endif
5731 #ifdef INET6
5732                 case IPPROTO_ICMPV6:
5733                     {
5734                         KMOD_ICMP6STAT_INC(icp6s_checksum);
5735                         break;
5736                     }
5737 #endif /* INET6 */
5738                 }
5739                 return (1);
5740         } else {
5741                 if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5742                         m->m_pkthdr.csum_flags |=
5743                             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5744                         m->m_pkthdr.csum_data = 0xffff;
5745                 }
5746         }
5747         return (0);
5748 }
5749
5750
5751 #ifdef INET
5752 int
5753 pf_test(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5754 {
5755         struct pfi_kif          *kif;
5756         u_short                  action, reason = 0, log = 0;
5757         struct mbuf             *m = *m0;
5758         struct ip               *h = NULL;
5759         struct m_tag            *ipfwtag;
5760         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5761         struct pf_state         *s = NULL;
5762         struct pf_ruleset       *ruleset = NULL;
5763         struct pf_pdesc          pd;
5764         int                      off, dirndx, pqid = 0;
5765
5766         M_ASSERTPKTHDR(m);
5767
5768         if (!V_pf_status.running)
5769                 return (PF_PASS);
5770
5771         memset(&pd, 0, sizeof(pd));
5772
5773         kif = (struct pfi_kif *)ifp->if_pf_kif;
5774
5775         if (kif == NULL) {
5776                 DPFPRINTF(PF_DEBUG_URGENT,
5777                     ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5778                 return (PF_DROP);
5779         }
5780         if (kif->pfik_flags & PFI_IFLAG_SKIP)
5781                 return (PF_PASS);
5782
5783         if (m->m_flags & M_SKIP_FIREWALL)
5784                 return (PF_PASS);
5785
5786         pd.pf_mtag = pf_find_mtag(m);
5787
5788         PF_RULES_RLOCK();
5789
5790         if (ip_divert_ptr != NULL &&
5791             ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5792                 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5793                 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5794                         if (pd.pf_mtag == NULL &&
5795                             ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5796                                 action = PF_DROP;
5797                                 goto done;
5798                         }
5799                         pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5800                         m_tag_delete(m, ipfwtag);
5801                 }
5802                 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5803                         m->m_flags |= M_FASTFWD_OURS;
5804                         pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5805                 }
5806         } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5807                 /* We do IP header normalization and packet reassembly here */
5808                 action = PF_DROP;
5809                 goto done;
5810         }
5811         m = *m0;        /* pf_normalize messes with m0 */
5812         h = mtod(m, struct ip *);
5813
5814         off = h->ip_hl << 2;
5815         if (off < (int)sizeof(struct ip)) {
5816                 action = PF_DROP;
5817                 REASON_SET(&reason, PFRES_SHORT);
5818                 log = 1;
5819                 goto done;
5820         }
5821
5822         pd.src = (struct pf_addr *)&h->ip_src;
5823         pd.dst = (struct pf_addr *)&h->ip_dst;
5824         pd.sport = pd.dport = NULL;
5825         pd.ip_sum = &h->ip_sum;
5826         pd.proto_sum = NULL;
5827         pd.proto = h->ip_p;
5828         pd.dir = dir;
5829         pd.sidx = (dir == PF_IN) ? 0 : 1;
5830         pd.didx = (dir == PF_IN) ? 1 : 0;
5831         pd.af = AF_INET;
5832         pd.tos = h->ip_tos;
5833         pd.tot_len = ntohs(h->ip_len);
5834
5835         /* handle fragments that didn't get reassembled by normalization */
5836         if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
5837                 action = pf_test_fragment(&r, dir, kif, m, h,
5838                     &pd, &a, &ruleset);
5839                 goto done;
5840         }
5841
5842         switch (h->ip_p) {
5843
5844         case IPPROTO_TCP: {
5845                 struct tcphdr   th;
5846
5847                 pd.hdr.tcp = &th;
5848                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
5849                     &action, &reason, AF_INET)) {
5850                         log = action != PF_PASS;
5851                         goto done;
5852                 }
5853                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
5854                 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
5855                         pqid = 1;
5856                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
5857                 if (action == PF_DROP)
5858                         goto done;
5859                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
5860                     &reason);
5861                 if (action == PF_PASS) {
5862                         if (pfsync_update_state_ptr != NULL)
5863                                 pfsync_update_state_ptr(s);
5864                         r = s->rule.ptr;
5865                         a = s->anchor.ptr;
5866                         log = s->log;
5867                 } else if (s == NULL)
5868                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5869                             &a, &ruleset, inp);
5870                 break;
5871         }
5872
5873         case IPPROTO_UDP: {
5874                 struct udphdr   uh;
5875
5876                 pd.hdr.udp = &uh;
5877                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
5878                     &action, &reason, AF_INET)) {
5879                         log = action != PF_PASS;
5880                         goto done;
5881                 }
5882                 if (uh.uh_dport == 0 ||
5883                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
5884                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
5885                         action = PF_DROP;
5886                         REASON_SET(&reason, PFRES_SHORT);
5887                         goto done;
5888                 }
5889                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
5890                 if (action == PF_PASS) {
5891                         if (pfsync_update_state_ptr != NULL)
5892                                 pfsync_update_state_ptr(s);
5893                         r = s->rule.ptr;
5894                         a = s->anchor.ptr;
5895                         log = s->log;
5896                 } else if (s == NULL)
5897                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5898                             &a, &ruleset, inp);
5899                 break;
5900         }
5901
5902         case IPPROTO_ICMP: {
5903                 struct icmp     ih;
5904
5905                 pd.hdr.icmp = &ih;
5906                 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
5907                     &action, &reason, AF_INET)) {
5908                         log = action != PF_PASS;
5909                         goto done;
5910                 }
5911                 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
5912                     &reason);
5913                 if (action == PF_PASS) {
5914                         if (pfsync_update_state_ptr != NULL)
5915                                 pfsync_update_state_ptr(s);
5916                         r = s->rule.ptr;
5917                         a = s->anchor.ptr;
5918                         log = s->log;
5919                 } else if (s == NULL)
5920                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5921                             &a, &ruleset, inp);
5922                 break;
5923         }
5924
5925 #ifdef INET6
5926         case IPPROTO_ICMPV6: {
5927                 action = PF_DROP;
5928                 DPFPRINTF(PF_DEBUG_MISC,
5929                     ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
5930                 goto done;
5931         }
5932 #endif
5933
5934         default:
5935                 action = pf_test_state_other(&s, dir, kif, m, &pd);
5936                 if (action == PF_PASS) {
5937                         if (pfsync_update_state_ptr != NULL)
5938                                 pfsync_update_state_ptr(s);
5939                         r = s->rule.ptr;
5940                         a = s->anchor.ptr;
5941                         log = s->log;
5942                 } else if (s == NULL)
5943                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5944                             &a, &ruleset, inp);
5945                 break;
5946         }
5947
5948 done:
5949         PF_RULES_RUNLOCK();
5950         if (action == PF_PASS && h->ip_hl > 5 &&
5951             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
5952                 action = PF_DROP;
5953                 REASON_SET(&reason, PFRES_IPOPTIONS);
5954                 log = 1;
5955                 DPFPRINTF(PF_DEBUG_MISC,
5956                     ("pf: dropping packet with ip options\n"));
5957         }
5958
5959         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
5960                 action = PF_DROP;
5961                 REASON_SET(&reason, PFRES_MEMORY);
5962         }
5963         if (r->rtableid >= 0)
5964                 M_SETFIB(m, r->rtableid);
5965
5966 #ifdef ALTQ
5967         if (action == PF_PASS && r->qid) {
5968                 if (pd.pf_mtag == NULL &&
5969                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5970                         action = PF_DROP;
5971                         REASON_SET(&reason, PFRES_MEMORY);
5972                 } else {
5973                         if (pqid || (pd.tos & IPTOS_LOWDELAY))
5974                                 pd.pf_mtag->qid = r->pqid;
5975                         else
5976                                 pd.pf_mtag->qid = r->qid;
5977                         /* Add hints for ecn. */
5978                         pd.pf_mtag->hdr = h;
5979                 }
5980
5981         }
5982 #endif /* ALTQ */
5983
5984         /*
5985          * connections redirected to loopback should not match sockets
5986          * bound specifically to loopback due to security implications,
5987          * see tcp_input() and in_pcblookup_listen().
5988          */
5989         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
5990             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
5991             (s->nat_rule.ptr->action == PF_RDR ||
5992             s->nat_rule.ptr->action == PF_BINAT) &&
5993             (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
5994                 m->m_flags |= M_SKIP_FIREWALL;
5995
5996         if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
5997             !PACKET_LOOPED(&pd)) {
5998
5999                 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
6000                     sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
6001                 if (ipfwtag != NULL) {
6002                         ((struct ipfw_rule_ref *)(ipfwtag+1))->info =
6003                             ntohs(r->divert.port);
6004                         ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
6005
6006                         if (s)
6007                                 PF_STATE_UNLOCK(s);
6008
6009                         m_tag_prepend(m, ipfwtag);
6010                         if (m->m_flags & M_FASTFWD_OURS) {
6011                                 if (pd.pf_mtag == NULL &&
6012                                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6013                                         action = PF_DROP;
6014                                         REASON_SET(&reason, PFRES_MEMORY);
6015                                         log = 1;
6016                                         DPFPRINTF(PF_DEBUG_MISC,
6017                                             ("pf: failed to allocate tag\n"));
6018                                 } else {
6019                                         pd.pf_mtag->flags |=
6020                                             PF_FASTFWD_OURS_PRESENT;
6021                                         m->m_flags &= ~M_FASTFWD_OURS;
6022                                 }
6023                         }
6024                         ip_divert_ptr(*m0, dir ==  PF_IN ? DIR_IN : DIR_OUT);
6025                         *m0 = NULL;
6026
6027                         return (action);
6028                 } else {
6029                         /* XXX: ipfw has the same behaviour! */
6030                         action = PF_DROP;
6031                         REASON_SET(&reason, PFRES_MEMORY);
6032                         log = 1;
6033                         DPFPRINTF(PF_DEBUG_MISC,
6034                             ("pf: failed to allocate divert tag\n"));
6035                 }
6036         }
6037
6038         if (log) {
6039                 struct pf_rule *lr;
6040
6041                 if (s != NULL && s->nat_rule.ptr != NULL &&
6042                     s->nat_rule.ptr->log & PF_LOG_ALL)
6043                         lr = s->nat_rule.ptr;
6044                 else
6045                         lr = r;
6046                 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
6047                     (s == NULL));
6048         }
6049
6050         kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6051         kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
6052
6053         if (action == PF_PASS || r->action == PF_DROP) {
6054                 dirndx = (dir == PF_OUT);
6055                 r->packets[dirndx]++;
6056                 r->bytes[dirndx] += pd.tot_len;
6057                 if (a != NULL) {
6058                         a->packets[dirndx]++;
6059                         a->bytes[dirndx] += pd.tot_len;
6060                 }
6061                 if (s != NULL) {
6062                         if (s->nat_rule.ptr != NULL) {
6063                                 s->nat_rule.ptr->packets[dirndx]++;
6064                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6065                         }
6066                         if (s->src_node != NULL) {
6067                                 s->src_node->packets[dirndx]++;
6068                                 s->src_node->bytes[dirndx] += pd.tot_len;
6069                         }
6070                         if (s->nat_src_node != NULL) {
6071                                 s->nat_src_node->packets[dirndx]++;
6072                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6073                         }
6074                         dirndx = (dir == s->direction) ? 0 : 1;
6075                         s->packets[dirndx]++;
6076                         s->bytes[dirndx] += pd.tot_len;
6077                 }
6078                 tr = r;
6079                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6080                 if (nr != NULL && r == &V_pf_default_rule)
6081                         tr = nr;
6082                 if (tr->src.addr.type == PF_ADDR_TABLE)
6083                         pfr_update_stats(tr->src.addr.p.tbl,
6084                             (s == NULL) ? pd.src :
6085                             &s->key[(s->direction == PF_IN)]->
6086                                 addr[(s->direction == PF_OUT)],
6087                             pd.af, pd.tot_len, dir == PF_OUT,
6088                             r->action == PF_PASS, tr->src.neg);
6089                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6090                         pfr_update_stats(tr->dst.addr.p.tbl,
6091                             (s == NULL) ? pd.dst :
6092                             &s->key[(s->direction == PF_IN)]->
6093                                 addr[(s->direction == PF_IN)],
6094                             pd.af, pd.tot_len, dir == PF_OUT,
6095                             r->action == PF_PASS, tr->dst.neg);
6096         }
6097
6098         switch (action) {
6099         case PF_SYNPROXY_DROP:
6100                 m_freem(*m0);
6101         case PF_DEFER:
6102                 *m0 = NULL;
6103                 action = PF_PASS;
6104                 break;
6105         case PF_DROP:
6106                 m_freem(*m0);
6107                 *m0 = NULL;
6108                 break;
6109         default:
6110                 /* pf_route() returns unlocked. */
6111                 if (r->rt) {
6112                         pf_route(m0, r, dir, kif->pfik_ifp, s, &pd);
6113                         return (action);
6114                 }
6115                 break;
6116         }
6117         if (s)
6118                 PF_STATE_UNLOCK(s);
6119
6120         return (action);
6121 }
6122 #endif /* INET */
6123
6124 #ifdef INET6
6125 int
6126 pf_test6(int dir, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6127 {
6128         struct pfi_kif          *kif;
6129         u_short                  action, reason = 0, log = 0;
6130         struct mbuf             *m = *m0, *n = NULL;
6131         struct m_tag            *mtag;
6132         struct ip6_hdr          *h = NULL;
6133         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6134         struct pf_state         *s = NULL;
6135         struct pf_ruleset       *ruleset = NULL;
6136         struct pf_pdesc          pd;
6137         int                      off, terminal = 0, dirndx, rh_cnt = 0;
6138         int                      fwdir = dir;
6139
6140         M_ASSERTPKTHDR(m);
6141
6142         if (dir == PF_OUT && m->m_pkthdr.rcvif && ifp != m->m_pkthdr.rcvif)
6143                 fwdir = PF_FWD;
6144
6145         if (!V_pf_status.running)
6146                 return (PF_PASS);
6147
6148         memset(&pd, 0, sizeof(pd));
6149         pd.pf_mtag = pf_find_mtag(m);
6150
6151         if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6152                 return (PF_PASS);
6153
6154         kif = (struct pfi_kif *)ifp->if_pf_kif;
6155         if (kif == NULL) {
6156                 DPFPRINTF(PF_DEBUG_URGENT,
6157                     ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6158                 return (PF_DROP);
6159         }
6160         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6161                 return (PF_PASS);
6162
6163         if (m->m_flags & M_SKIP_FIREWALL)
6164                 return (PF_PASS);
6165
6166         PF_RULES_RLOCK();
6167
6168         /* We do IP header normalization and packet reassembly here */
6169         if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6170                 action = PF_DROP;
6171                 goto done;
6172         }
6173         m = *m0;        /* pf_normalize messes with m0 */
6174         h = mtod(m, struct ip6_hdr *);
6175
6176 #if 1
6177         /*
6178          * we do not support jumbogram yet.  if we keep going, zero ip6_plen
6179          * will do something bad, so drop the packet for now.
6180          */
6181         if (htons(h->ip6_plen) == 0) {
6182                 action = PF_DROP;
6183                 REASON_SET(&reason, PFRES_NORM);        /*XXX*/
6184                 goto done;
6185         }
6186 #endif
6187
6188         pd.src = (struct pf_addr *)&h->ip6_src;
6189         pd.dst = (struct pf_addr *)&h->ip6_dst;
6190         pd.sport = pd.dport = NULL;
6191         pd.ip_sum = NULL;
6192         pd.proto_sum = NULL;
6193         pd.dir = dir;
6194         pd.sidx = (dir == PF_IN) ? 0 : 1;
6195         pd.didx = (dir == PF_IN) ? 1 : 0;
6196         pd.af = AF_INET6;
6197         pd.tos = 0;
6198         pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6199
6200         off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6201         pd.proto = h->ip6_nxt;
6202         do {
6203                 switch (pd.proto) {
6204                 case IPPROTO_FRAGMENT:
6205                         action = pf_test_fragment(&r, dir, kif, m, h,
6206                             &pd, &a, &ruleset);
6207                         if (action == PF_DROP)
6208                                 REASON_SET(&reason, PFRES_FRAG);
6209                         goto done;
6210                 case IPPROTO_ROUTING: {
6211                         struct ip6_rthdr rthdr;
6212
6213                         if (rh_cnt++) {
6214                                 DPFPRINTF(PF_DEBUG_MISC,
6215                                     ("pf: IPv6 more than one rthdr\n"));
6216                                 action = PF_DROP;
6217                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6218                                 log = 1;
6219                                 goto done;
6220                         }
6221                         if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6222                             &reason, pd.af)) {
6223                                 DPFPRINTF(PF_DEBUG_MISC,
6224                                     ("pf: IPv6 short rthdr\n"));
6225                                 action = PF_DROP;
6226                                 REASON_SET(&reason, PFRES_SHORT);
6227                                 log = 1;
6228                                 goto done;
6229                         }
6230                         if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6231                                 DPFPRINTF(PF_DEBUG_MISC,
6232                                     ("pf: IPv6 rthdr0\n"));
6233                                 action = PF_DROP;
6234                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6235                                 log = 1;
6236                                 goto done;
6237                         }
6238                         /* FALLTHROUGH */
6239                 }
6240                 case IPPROTO_AH:
6241                 case IPPROTO_HOPOPTS:
6242                 case IPPROTO_DSTOPTS: {
6243                         /* get next header and header length */
6244                         struct ip6_ext  opt6;
6245
6246                         if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6247                             NULL, &reason, pd.af)) {
6248                                 DPFPRINTF(PF_DEBUG_MISC,
6249                                     ("pf: IPv6 short opt\n"));
6250                                 action = PF_DROP;
6251                                 log = 1;
6252                                 goto done;
6253                         }
6254                         if (pd.proto == IPPROTO_AH)
6255                                 off += (opt6.ip6e_len + 2) * 4;
6256                         else
6257                                 off += (opt6.ip6e_len + 1) * 8;
6258                         pd.proto = opt6.ip6e_nxt;
6259                         /* goto the next header */
6260                         break;
6261                 }
6262                 default:
6263                         terminal++;
6264                         break;
6265                 }
6266         } while (!terminal);
6267
6268         /* if there's no routing header, use unmodified mbuf for checksumming */
6269         if (!n)
6270                 n = m;
6271
6272         switch (pd.proto) {
6273
6274         case IPPROTO_TCP: {
6275                 struct tcphdr   th;
6276
6277                 pd.hdr.tcp = &th;
6278                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6279                     &action, &reason, AF_INET6)) {
6280                         log = action != PF_PASS;
6281                         goto done;
6282                 }
6283                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6284                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6285                 if (action == PF_DROP)
6286                         goto done;
6287                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6288                     &reason);
6289                 if (action == PF_PASS) {
6290                         if (pfsync_update_state_ptr != NULL)
6291                                 pfsync_update_state_ptr(s);
6292                         r = s->rule.ptr;
6293                         a = s->anchor.ptr;
6294                         log = s->log;
6295                 } else if (s == NULL)
6296                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6297                             &a, &ruleset, inp);
6298                 break;
6299         }
6300
6301         case IPPROTO_UDP: {
6302                 struct udphdr   uh;
6303
6304                 pd.hdr.udp = &uh;
6305                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6306                     &action, &reason, AF_INET6)) {
6307                         log = action != PF_PASS;
6308                         goto done;
6309                 }
6310                 if (uh.uh_dport == 0 ||
6311                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6312                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6313                         action = PF_DROP;
6314                         REASON_SET(&reason, PFRES_SHORT);
6315                         goto done;
6316                 }
6317                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6318                 if (action == PF_PASS) {
6319                         if (pfsync_update_state_ptr != NULL)
6320                                 pfsync_update_state_ptr(s);
6321                         r = s->rule.ptr;
6322                         a = s->anchor.ptr;
6323                         log = s->log;
6324                 } else if (s == NULL)
6325                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6326                             &a, &ruleset, inp);
6327                 break;
6328         }
6329
6330         case IPPROTO_ICMP: {
6331                 action = PF_DROP;
6332                 DPFPRINTF(PF_DEBUG_MISC,
6333                     ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6334                 goto done;
6335         }
6336
6337         case IPPROTO_ICMPV6: {
6338                 struct icmp6_hdr        ih;
6339
6340                 pd.hdr.icmp6 = &ih;
6341                 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6342                     &action, &reason, AF_INET6)) {
6343                         log = action != PF_PASS;
6344                         goto done;
6345                 }
6346                 action = pf_test_state_icmp(&s, dir, kif,
6347                     m, off, h, &pd, &reason);
6348                 if (action == PF_PASS) {
6349                         if (pfsync_update_state_ptr != NULL)
6350                                 pfsync_update_state_ptr(s);
6351                         r = s->rule.ptr;
6352                         a = s->anchor.ptr;
6353                         log = s->log;
6354                 } else if (s == NULL)
6355                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6356                             &a, &ruleset, inp);
6357                 break;
6358         }
6359
6360         default:
6361                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6362                 if (action == PF_PASS) {
6363                         if (pfsync_update_state_ptr != NULL)
6364                                 pfsync_update_state_ptr(s);
6365                         r = s->rule.ptr;
6366                         a = s->anchor.ptr;
6367                         log = s->log;
6368                 } else if (s == NULL)
6369                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6370                             &a, &ruleset, inp);
6371                 break;
6372         }
6373
6374 done:
6375         PF_RULES_RUNLOCK();
6376         if (n != m) {
6377                 m_freem(n);
6378                 n = NULL;
6379         }
6380
6381         /* handle dangerous IPv6 extension headers. */
6382         if (action == PF_PASS && rh_cnt &&
6383             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6384                 action = PF_DROP;
6385                 REASON_SET(&reason, PFRES_IPOPTIONS);
6386                 log = 1;
6387                 DPFPRINTF(PF_DEBUG_MISC,
6388                     ("pf: dropping packet with dangerous v6 headers\n"));
6389         }
6390
6391         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6392                 action = PF_DROP;
6393                 REASON_SET(&reason, PFRES_MEMORY);
6394         }
6395         if (r->rtableid >= 0)
6396                 M_SETFIB(m, r->rtableid);
6397
6398 #ifdef ALTQ
6399         if (action == PF_PASS && r->qid) {
6400                 if (pd.pf_mtag == NULL &&
6401                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6402                         action = PF_DROP;
6403                         REASON_SET(&reason, PFRES_MEMORY);
6404                 } else {
6405                         if (pd.tos & IPTOS_LOWDELAY)
6406                                 pd.pf_mtag->qid = r->pqid;
6407                         else
6408                                 pd.pf_mtag->qid = r->qid;
6409                         /* Add hints for ecn. */
6410                         pd.pf_mtag->hdr = h;
6411                 }
6412         }
6413 #endif /* ALTQ */
6414
6415         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6416             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6417             (s->nat_rule.ptr->action == PF_RDR ||
6418             s->nat_rule.ptr->action == PF_BINAT) &&
6419             IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6420                 m->m_flags |= M_SKIP_FIREWALL;
6421
6422         /* XXX: Anybody working on it?! */
6423         if (r->divert.port)
6424                 printf("pf: divert(9) is not supported for IPv6\n");
6425
6426         if (log) {
6427                 struct pf_rule *lr;
6428
6429                 if (s != NULL && s->nat_rule.ptr != NULL &&
6430                     s->nat_rule.ptr->log & PF_LOG_ALL)
6431                         lr = s->nat_rule.ptr;
6432                 else
6433                         lr = r;
6434                 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6435                     &pd, (s == NULL));
6436         }
6437
6438         kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6439         kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
6440
6441         if (action == PF_PASS || r->action == PF_DROP) {
6442                 dirndx = (dir == PF_OUT);
6443                 r->packets[dirndx]++;
6444                 r->bytes[dirndx] += pd.tot_len;
6445                 if (a != NULL) {
6446                         a->packets[dirndx]++;
6447                         a->bytes[dirndx] += pd.tot_len;
6448                 }
6449                 if (s != NULL) {
6450                         if (s->nat_rule.ptr != NULL) {
6451                                 s->nat_rule.ptr->packets[dirndx]++;
6452                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6453                         }
6454                         if (s->src_node != NULL) {
6455                                 s->src_node->packets[dirndx]++;
6456                                 s->src_node->bytes[dirndx] += pd.tot_len;
6457                         }
6458                         if (s->nat_src_node != NULL) {
6459                                 s->nat_src_node->packets[dirndx]++;
6460                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6461                         }
6462                         dirndx = (dir == s->direction) ? 0 : 1;
6463                         s->packets[dirndx]++;
6464                         s->bytes[dirndx] += pd.tot_len;
6465                 }
6466                 tr = r;
6467                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6468                 if (nr != NULL && r == &V_pf_default_rule)
6469                         tr = nr;
6470                 if (tr->src.addr.type == PF_ADDR_TABLE)
6471                         pfr_update_stats(tr->src.addr.p.tbl,
6472                             (s == NULL) ? pd.src :
6473                             &s->key[(s->direction == PF_IN)]->addr[0],
6474                             pd.af, pd.tot_len, dir == PF_OUT,
6475                             r->action == PF_PASS, tr->src.neg);
6476                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6477                         pfr_update_stats(tr->dst.addr.p.tbl,
6478                             (s == NULL) ? pd.dst :
6479                             &s->key[(s->direction == PF_IN)]->addr[1],
6480                             pd.af, pd.tot_len, dir == PF_OUT,
6481                             r->action == PF_PASS, tr->dst.neg);
6482         }
6483
6484         switch (action) {
6485         case PF_SYNPROXY_DROP:
6486                 m_freem(*m0);
6487         case PF_DEFER:
6488                 *m0 = NULL;
6489                 action = PF_PASS;
6490                 break;
6491         case PF_DROP:
6492                 m_freem(*m0);
6493                 *m0 = NULL;
6494                 break;
6495         default:
6496                 /* pf_route6() returns unlocked. */
6497                 if (r->rt) {
6498                         pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd);
6499                         return (action);
6500                 }
6501                 break;
6502         }
6503
6504         if (s)
6505                 PF_STATE_UNLOCK(s);
6506
6507         /* If reassembled packet passed, create new fragments. */
6508         if (action == PF_PASS && *m0 && fwdir == PF_FWD &&
6509             (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL)
6510                 action = pf_refragment6(ifp, m0, mtag);
6511
6512         return (action);
6513 }
6514 #endif /* INET6 */