sys/netpfil/pf/pf.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause
   3  *
   4  * Copyright (c) 2001 Daniel Hartmeier
   5  * Copyright (c) 2002 - 2008 Henning Brauer
   6  * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org>
   7  * All rights reserved.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  *
  13  *    - Redistributions of source code must retain the above copyright
  14  *      notice, this list of conditions and the following disclaimer.
  15  *    - Redistributions in binary form must reproduce the above
  16  *      copyright notice, this list of conditions and the following
  17  *      disclaimer in the documentation and/or other materials provided
  18  *      with the distribution.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  21  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
  24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
  26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
  28  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
  30  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  31  * POSSIBILITY OF SUCH DAMAGE.
  32  *
  33  * Effort sponsored in part by the Defense Advanced Research Projects
  34  * Agency (DARPA) and Air Force Research Laboratory, Air Force
  35  * Materiel Command, USAF, under agreement number F30602-01-2-0537.
  36  *
  37  *      $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $
  38  */
  39
  40 #include <sys/cdefs.h>
  41 __FBSDID("$FreeBSD$");
  42
  43 #include "opt_bpf.h"
  44 #include "opt_inet.h"
  45 #include "opt_inet6.h"
  46 #include "opt_pf.h"
  47 #include "opt_sctp.h"
  48
  49 #include <sys/param.h>
  50 #include <sys/bus.h>
  51 #include <sys/endian.h>
  52 #include <sys/gsb_crc32.h>
  53 #include <sys/hash.h>
  54 #include <sys/interrupt.h>
  55 #include <sys/kernel.h>
  56 #include <sys/kthread.h>
  57 #include <sys/limits.h>
  58 #include <sys/mbuf.h>
  59 #include <sys/md5.h>
  60 #include <sys/random.h>
  61 #include <sys/refcount.h>
  62 #include <sys/socket.h>
  63 #include <sys/sysctl.h>
  64 #include <sys/taskqueue.h>
  65 #include <sys/ucred.h>
  66
  67 #include <net/if.h>
  68 #include <net/if_var.h>
  69 #include <net/if_types.h>
  70 #include <net/if_vlan_var.h>
  71 #include <net/route.h>
  72 #include <net/route/nhop.h>
  73 #include <net/vnet.h>
  74
  75 #include <net/pfil.h>
  76 #include <net/pfvar.h>
  77 #include <net/if_pflog.h>
  78 #include <net/if_pfsync.h>
  79
  80 #include <netinet/in_pcb.h>
  81 #include <netinet/in_var.h>
  82 #include <netinet/in_fib.h>
  83 #include <netinet/ip.h>
  84 #include <netinet/ip_fw.h>
  85 #include <netinet/ip_icmp.h>
  86 #include <netinet/icmp_var.h>
  87 #include <netinet/ip_var.h>
  88 #include <netinet/tcp.h>
  89 #include <netinet/tcp_fsm.h>
  90 #include <netinet/tcp_seq.h>
  91 #include <netinet/tcp_timer.h>
  92 #include <netinet/tcp_var.h>
  93 #include <netinet/udp.h>
  94 #include <netinet/udp_var.h>
  95
  96 #ifdef INET6
  97 #include <netinet/ip6.h>
  98 #include <netinet/icmp6.h>
  99 #include <netinet6/nd6.h>
 100 #include <netinet6/ip6_var.h>
 101 #include <netinet6/in6_pcb.h>
 102 #include <netinet6/in6_fib.h>
 103 #include <netinet6/scope6_var.h>
 104 #endif /* INET6 */
 105
 106 #if defined(SCTP) || defined(SCTP_SUPPORT)
 107 #include <netinet/sctp_crc32.h>
 108 #endif
 109
 110 #include <machine/in_cksum.h>
 111 #include <security/mac/mac_framework.h>
 112
 113 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
 114
 115 /*
 116  * Global variables
 117  */
 118
 119 /* state tables */
 120 VNET_DEFINE(struct pf_altqqueue,         pf_altqs[4]);
 121 VNET_DEFINE(struct pf_palist,            pf_pabuf);
 122 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_active);
 123 VNET_DEFINE(struct pf_altqqueue *,       pf_altq_ifs_active);
 124 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_inactive);
 125 VNET_DEFINE(struct pf_altqqueue *,       pf_altq_ifs_inactive);
 126 VNET_DEFINE(struct pf_kstatus,           pf_status);
 127
 128 VNET_DEFINE(u_int32_t,                   ticket_altqs_active);
 129 VNET_DEFINE(u_int32_t,                   ticket_altqs_inactive);
 130 VNET_DEFINE(int,                         altqs_inactive_open);
 131 VNET_DEFINE(u_int32_t,                   ticket_pabuf);
 132
 133 VNET_DEFINE(MD5_CTX,                     pf_tcp_secret_ctx);
 134 #define V_pf_tcp_secret_ctx              VNET(pf_tcp_secret_ctx)
 135 VNET_DEFINE(u_char,                      pf_tcp_secret[16]);
 136 #define V_pf_tcp_secret                  VNET(pf_tcp_secret)
 137 VNET_DEFINE(int,                         pf_tcp_secret_init);
 138 #define V_pf_tcp_secret_init             VNET(pf_tcp_secret_init)
 139 VNET_DEFINE(int,                         pf_tcp_iss_off);
 140 #define V_pf_tcp_iss_off                 VNET(pf_tcp_iss_off)
 141 VNET_DECLARE(int,                        pf_vnet_active);
 142 #define V_pf_vnet_active                 VNET(pf_vnet_active)
 143
 144 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx);
 145 #define V_pf_purge_idx  VNET(pf_purge_idx)
 146
 147 /*
 148  * Queue for pf_intr() sends.
 149  */
 150 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
 151 struct pf_send_entry {
 152         STAILQ_ENTRY(pf_send_entry)     pfse_next;
 153         struct mbuf                     *pfse_m;
 154         enum {
 155                 PFSE_IP,
 156                 PFSE_IP6,
 157                 PFSE_ICMP,
 158                 PFSE_ICMP6,
 159         }                               pfse_type;
 160         struct {
 161                 int             type;
 162                 int             code;
 163                 int             mtu;
 164         } icmpopts;
 165 };
 166
 167 STAILQ_HEAD(pf_send_head, pf_send_entry);
 168 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue);
 169 #define V_pf_sendqueue  VNET(pf_sendqueue)
 170
 171 static struct mtx pf_sendqueue_mtx;
 172 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
 173 #define PF_SENDQ_LOCK()         mtx_lock(&pf_sendqueue_mtx)
 174 #define PF_SENDQ_UNLOCK()       mtx_unlock(&pf_sendqueue_mtx)
 175
 176 /*
 177  * Queue for pf_overload_task() tasks.
 178  */
 179 struct pf_overload_entry {
 180         SLIST_ENTRY(pf_overload_entry)  next;
 181         struct pf_addr                  addr;
 182         sa_family_t                     af;
 183         uint8_t                         dir;
 184         struct pf_rule                  *rule;
 185 };
 186
 187 SLIST_HEAD(pf_overload_head, pf_overload_entry);
 188 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue);
 189 #define V_pf_overloadqueue      VNET(pf_overloadqueue)
 190 VNET_DEFINE_STATIC(struct task, pf_overloadtask);
 191 #define V_pf_overloadtask       VNET(pf_overloadtask)
 192
 193 static struct mtx pf_overloadqueue_mtx;
 194 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
 195     "pf overload/flush queue", MTX_DEF);
 196 #define PF_OVERLOADQ_LOCK()     mtx_lock(&pf_overloadqueue_mtx)
 197 #define PF_OVERLOADQ_UNLOCK()   mtx_unlock(&pf_overloadqueue_mtx)
 198
 199 VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules);
 200 struct mtx pf_unlnkdrules_mtx;
 201 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
 202     MTX_DEF);
 203
 204 VNET_DEFINE_STATIC(uma_zone_t,  pf_sources_z);
 205 #define V_pf_sources_z  VNET(pf_sources_z)
 206 uma_zone_t              pf_mtag_z;
 207 VNET_DEFINE(uma_zone_t,  pf_state_z);
 208 VNET_DEFINE(uma_zone_t,  pf_state_key_z);
 209
 210 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
 211 #define PFID_CPUBITS    8
 212 #define PFID_CPUSHIFT   (sizeof(uint64_t) * NBBY - PFID_CPUBITS)
 213 #define PFID_CPUMASK    ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT)
 214 #define PFID_MAXID      (~PFID_CPUMASK)
 215 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU);
 216
 217 static void              pf_src_tree_remove_state(struct pf_state *);
 218 static void              pf_init_threshold(struct pf_threshold *, u_int32_t,
 219                             u_int32_t);
 220 static void              pf_add_threshold(struct pf_threshold *);
 221 static int               pf_check_threshold(struct pf_threshold *);
 222
 223 static void              pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
 224                             u_int16_t *, u_int16_t *, struct pf_addr *,
 225                             u_int16_t, u_int8_t, sa_family_t);
 226 static int               pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
 227                             struct tcphdr *, struct pf_state_peer *);
 228 static void              pf_change_icmp(struct pf_addr *, u_int16_t *,
 229                             struct pf_addr *, struct pf_addr *, u_int16_t,
 230                             u_int16_t *, u_int16_t *, u_int16_t *,
 231                             u_int16_t *, u_int8_t, sa_family_t);
 232 static void              pf_send_tcp(struct mbuf *,
 233                             const struct pf_rule *, sa_family_t,
 234                             const struct pf_addr *, const struct pf_addr *,
 235                             u_int16_t, u_int16_t, u_int32_t, u_int32_t,
 236                             u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
 237                             u_int16_t, struct ifnet *);
 238 static void              pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
 239                             sa_family_t, struct pf_rule *);
 240 static void              pf_detach_state(struct pf_state *);
 241 static int               pf_state_key_attach(struct pf_state_key *,
 242                             struct pf_state_key *, struct pf_state *);
 243 static void              pf_state_key_detach(struct pf_state *, int);
 244 static int               pf_state_key_ctor(void *, int, void *, int);
 245 static u_int32_t         pf_tcp_iss(struct pf_pdesc *);
 246 static int               pf_test_rule(struct pf_rule **, struct pf_state **,
 247                             int, struct pfi_kif *, struct mbuf *, int,
 248                             struct pf_pdesc *, struct pf_rule **,
 249                             struct pf_ruleset **, struct inpcb *);
 250 static int               pf_create_state(struct pf_rule *, struct pf_rule *,
 251                             struct pf_rule *, struct pf_pdesc *,
 252                             struct pf_src_node *, struct pf_state_key *,
 253                             struct pf_state_key *, struct mbuf *, int,
 254                             u_int16_t, u_int16_t, int *, struct pfi_kif *,
 255                             struct pf_state **, int, u_int16_t, u_int16_t,
 256                             int);
 257 static int               pf_test_fragment(struct pf_rule **, int,
 258                             struct pfi_kif *, struct mbuf *, void *,
 259                             struct pf_pdesc *, struct pf_rule **,
 260                             struct pf_ruleset **);
 261 static int               pf_tcp_track_full(struct pf_state_peer *,
 262                             struct pf_state_peer *, struct pf_state **,
 263                             struct pfi_kif *, struct mbuf *, int,
 264                             struct pf_pdesc *, u_short *, int *);
 265 static int               pf_tcp_track_sloppy(struct pf_state_peer *,
 266                             struct pf_state_peer *, struct pf_state **,
 267                             struct pf_pdesc *, u_short *);
 268 static int               pf_test_state_tcp(struct pf_state **, int,
 269                             struct pfi_kif *, struct mbuf *, int,
 270                             void *, struct pf_pdesc *, u_short *);
 271 static int               pf_test_state_udp(struct pf_state **, int,
 272                             struct pfi_kif *, struct mbuf *, int,
 273                             void *, struct pf_pdesc *);
 274 static int               pf_test_state_icmp(struct pf_state **, int,
 275                             struct pfi_kif *, struct mbuf *, int,
 276                             void *, struct pf_pdesc *, u_short *);
 277 static int               pf_test_state_other(struct pf_state **, int,
 278                             struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
 279 static u_int8_t          pf_get_wscale(struct mbuf *, int, u_int16_t,
 280                             sa_family_t);
 281 static u_int16_t         pf_get_mss(struct mbuf *, int, u_int16_t,
 282                             sa_family_t);
 283 static u_int16_t         pf_calc_mss(struct pf_addr *, sa_family_t,
 284                                 int, u_int16_t);
 285 static int               pf_check_proto_cksum(struct mbuf *, int, int,
 286                             u_int8_t, sa_family_t);
 287 static void              pf_print_state_parts(struct pf_state *,
 288                             struct pf_state_key *, struct pf_state_key *);
 289 static int               pf_addr_wrap_neq(struct pf_addr_wrap *,
 290                             struct pf_addr_wrap *);
 291 static struct pf_state  *pf_find_state(struct pfi_kif *,
 292                             struct pf_state_key_cmp *, u_int);
 293 static int               pf_src_connlimit(struct pf_state **);
 294 static void              pf_overload_task(void *v, int pending);
 295 static int               pf_insert_src_node(struct pf_src_node **,
 296                             struct pf_rule *, struct pf_addr *, sa_family_t);
 297 static u_int             pf_purge_expired_states(u_int, int);
 298 static void              pf_purge_unlinked_rules(void);
 299 static int               pf_mtag_uminit(void *, int, int);
 300 static void              pf_mtag_free(struct m_tag *);
 301 #ifdef INET
 302 static void              pf_route(struct mbuf **, struct pf_rule *, int,
 303                             struct ifnet *, struct pf_state *,
 304                             struct pf_pdesc *, struct inpcb *);
 305 #endif /* INET */
 306 #ifdef INET6
 307 static void              pf_change_a6(struct pf_addr *, u_int16_t *,
 308                             struct pf_addr *, u_int8_t);
 309 static void              pf_route6(struct mbuf **, struct pf_rule *, int,
 310                             struct ifnet *, struct pf_state *,
 311                             struct pf_pdesc *, struct inpcb *);
 312 #endif /* INET6 */
 313
 314 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
 315
 316 extern int pf_end_threads;
 317 extern struct proc *pf_purge_proc;
 318
 319 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
 320
 321 #define PACKET_LOOPED(pd)       ((pd)->pf_mtag &&                       \
 322                                  (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
 323
 324 #define STATE_LOOKUP(i, k, d, s, pd)                                    \
 325         do {                                                            \
 326                 (s) = pf_find_state((i), (k), (d));                     \
 327                 if ((s) == NULL)                                        \
 328                         return (PF_DROP);                               \
 329                 if (PACKET_LOOPED(pd))                                  \
 330                         return (PF_PASS);                               \
 331                 if ((d) == PF_OUT &&                                    \
 332                     (((s)->rule.ptr->rt == PF_ROUTETO &&                \
 333                     (s)->rule.ptr->direction == PF_OUT) ||              \
 334                     ((s)->rule.ptr->rt == PF_REPLYTO &&                 \
 335                     (s)->rule.ptr->direction == PF_IN)) &&              \
 336                     (s)->rt_kif != NULL &&                              \
 337                     (s)->rt_kif != (i))                                 \
 338                         return (PF_PASS);                               \
 339         } while (0)
 340
 341 #define BOUND_IFACE(r, k) \
 342         ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
 343
 344 #define STATE_INC_COUNTERS(s)                                           \
 345         do {                                                            \
 346                 counter_u64_add(s->rule.ptr->states_cur, 1);            \
 347                 counter_u64_add(s->rule.ptr->states_tot, 1);            \
 348                 if (s->anchor.ptr != NULL) {                            \
 349                         counter_u64_add(s->anchor.ptr->states_cur, 1);  \
 350                         counter_u64_add(s->anchor.ptr->states_tot, 1);  \
 351                 }                                                       \
 352                 if (s->nat_rule.ptr != NULL) {                          \
 353                         counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
 354                         counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
 355                 }                                                       \
 356         } while (0)
 357
 358 #define STATE_DEC_COUNTERS(s)                                           \
 359         do {                                                            \
 360                 if (s->nat_rule.ptr != NULL)                            \
 361                         counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
 362                 if (s->anchor.ptr != NULL)                              \
 363                         counter_u64_add(s->anchor.ptr->states_cur, -1); \
 364                 counter_u64_add(s->rule.ptr->states_cur, -1);           \
 365         } while (0)
 366
 367 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
 368 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
 369 VNET_DEFINE(struct pf_idhash *, pf_idhash);
 370 VNET_DEFINE(struct pf_srchash *, pf_srchash);
 371
 372 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 373     "pf(4)");
 374
 375 u_long  pf_hashmask;
 376 u_long  pf_srchashmask;
 377 static u_long   pf_hashsize;
 378 static u_long   pf_srchashsize;
 379 u_long  pf_ioctl_maxcount = 65535;
 380
 381 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
 382     &pf_hashsize, 0, "Size of pf(4) states hashtable");
 383 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
 384     &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
 385 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RW,
 386     &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call");
 387
 388 VNET_DEFINE(void *, pf_swi_cookie);
 389
 390 VNET_DEFINE(uint32_t, pf_hashseed);
 391 #define V_pf_hashseed   VNET(pf_hashseed)
 392
 393 int
 394 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
 395 {
 396
 397         switch (af) {
 398 #ifdef INET
 399         case AF_INET:
 400                 if (a->addr32[0] > b->addr32[0])
 401                         return (1);
 402                 if (a->addr32[0] < b->addr32[0])
 403                         return (-1);
 404                 break;
 405 #endif /* INET */
 406 #ifdef INET6
 407         case AF_INET6:
 408                 if (a->addr32[3] > b->addr32[3])
 409                         return (1);
 410                 if (a->addr32[3] < b->addr32[3])
 411                         return (-1);
 412                 if (a->addr32[2] > b->addr32[2])
 413                         return (1);
 414                 if (a->addr32[2] < b->addr32[2])
 415                         return (-1);
 416                 if (a->addr32[1] > b->addr32[1])
 417                         return (1);
 418                 if (a->addr32[1] < b->addr32[1])
 419                         return (-1);
 420                 if (a->addr32[0] > b->addr32[0])
 421                         return (1);
 422                 if (a->addr32[0] < b->addr32[0])
 423                         return (-1);
 424                 break;
 425 #endif /* INET6 */
 426         default:
 427                 panic("%s: unknown address family %u", __func__, af);
 428         }
 429         return (0);
 430 }
 431
 432 static __inline uint32_t
 433 pf_hashkey(struct pf_state_key *sk)
 434 {
 435         uint32_t h;
 436
 437         h = murmur3_32_hash32((uint32_t *)sk,
 438             sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
 439             V_pf_hashseed);
 440
 441         return (h & pf_hashmask);
 442 }
 443
 444 static __inline uint32_t
 445 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
 446 {
 447         uint32_t h;
 448
 449         switch (af) {
 450         case AF_INET:
 451                 h = murmur3_32_hash32((uint32_t *)&addr->v4,
 452                     sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
 453                 break;
 454         case AF_INET6:
 455                 h = murmur3_32_hash32((uint32_t *)&addr->v6,
 456                     sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
 457                 break;
 458         default:
 459                 panic("%s: unknown address family %u", __func__, af);
 460         }
 461
 462         return (h & pf_srchashmask);
 463 }
 464
 465 #ifdef ALTQ
 466 static int
 467 pf_state_hash(struct pf_state *s)
 468 {
 469         u_int32_t hv = (intptr_t)s / sizeof(*s);
 470
 471         hv ^= crc32(&s->src, sizeof(s->src));
 472         hv ^= crc32(&s->dst, sizeof(s->dst));
 473         if (hv == 0)
 474                 hv = 1;
 475         return (hv);
 476 }
 477 #endif
 478
 479 #ifdef INET6
 480 void
 481 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
 482 {
 483         switch (af) {
 484 #ifdef INET
 485         case AF_INET:
 486                 dst->addr32[0] = src->addr32[0];
 487                 break;
 488 #endif /* INET */
 489         case AF_INET6:
 490                 dst->addr32[0] = src->addr32[0];
 491                 dst->addr32[1] = src->addr32[1];
 492                 dst->addr32[2] = src->addr32[2];
 493                 dst->addr32[3] = src->addr32[3];
 494                 break;
 495         }
 496 }
 497 #endif /* INET6 */
 498
 499 static void
 500 pf_init_threshold(struct pf_threshold *threshold,
 501     u_int32_t limit, u_int32_t seconds)
 502 {
 503         threshold->limit = limit * PF_THRESHOLD_MULT;
 504         threshold->seconds = seconds;
 505         threshold->count = 0;
 506         threshold->last = time_uptime;
 507 }
 508
 509 static void
 510 pf_add_threshold(struct pf_threshold *threshold)
 511 {
 512         u_int32_t t = time_uptime, diff = t - threshold->last;
 513
 514         if (diff >= threshold->seconds)
 515                 threshold->count = 0;
 516         else
 517                 threshold->count -= threshold->count * diff /
 518                     threshold->seconds;
 519         threshold->count += PF_THRESHOLD_MULT;
 520         threshold->last = t;
 521 }
 522
 523 static int
 524 pf_check_threshold(struct pf_threshold *threshold)
 525 {
 526         return (threshold->count > threshold->limit);
 527 }
 528
 529 static int
 530 pf_src_connlimit(struct pf_state **state)
 531 {
 532         struct pf_overload_entry *pfoe;
 533         int bad = 0;
 534
 535         PF_STATE_LOCK_ASSERT(*state);
 536
 537         (*state)->src_node->conn++;
 538         (*state)->src.tcp_est = 1;
 539         pf_add_threshold(&(*state)->src_node->conn_rate);
 540
 541         if ((*state)->rule.ptr->max_src_conn &&
 542             (*state)->rule.ptr->max_src_conn <
 543             (*state)->src_node->conn) {
 544                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
 545                 bad++;
 546         }
 547
 548         if ((*state)->rule.ptr->max_src_conn_rate.limit &&
 549             pf_check_threshold(&(*state)->src_node->conn_rate)) {
 550                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
 551                 bad++;
 552         }
 553
 554         if (!bad)
 555                 return (0);
 556
 557         /* Kill this state. */
 558         (*state)->timeout = PFTM_PURGE;
 559         (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
 560
 561         if ((*state)->rule.ptr->overload_tbl == NULL)
 562                 return (1);
 563
 564         /* Schedule overloading and flushing task. */
 565         pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
 566         if (pfoe == NULL)
 567                 return (1);     /* too bad :( */
 568
 569         bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
 570         pfoe->af = (*state)->key[PF_SK_WIRE]->af;
 571         pfoe->rule = (*state)->rule.ptr;
 572         pfoe->dir = (*state)->direction;
 573         PF_OVERLOADQ_LOCK();
 574         SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
 575         PF_OVERLOADQ_UNLOCK();
 576         taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
 577
 578         return (1);
 579 }
 580
 581 static void
 582 pf_overload_task(void *v, int pending)
 583 {
 584         struct pf_overload_head queue;
 585         struct pfr_addr p;
 586         struct pf_overload_entry *pfoe, *pfoe1;
 587         uint32_t killed = 0;
 588
 589         CURVNET_SET((struct vnet *)v);
 590
 591         PF_OVERLOADQ_LOCK();
 592         queue = V_pf_overloadqueue;
 593         SLIST_INIT(&V_pf_overloadqueue);
 594         PF_OVERLOADQ_UNLOCK();
 595
 596         bzero(&p, sizeof(p));
 597         SLIST_FOREACH(pfoe, &queue, next) {
 598                 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
 599                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
 600                         printf("%s: blocking address ", __func__);
 601                         pf_print_host(&pfoe->addr, 0, pfoe->af);
 602                         printf("\n");
 603                 }
 604
 605                 p.pfra_af = pfoe->af;
 606                 switch (pfoe->af) {
 607 #ifdef INET
 608                 case AF_INET:
 609                         p.pfra_net = 32;
 610                         p.pfra_ip4addr = pfoe->addr.v4;
 611                         break;
 612 #endif
 613 #ifdef INET6
 614                 case AF_INET6:
 615                         p.pfra_net = 128;
 616                         p.pfra_ip6addr = pfoe->addr.v6;
 617                         break;
 618 #endif
 619                 }
 620
 621                 PF_RULES_WLOCK();
 622                 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
 623                 PF_RULES_WUNLOCK();
 624         }
 625
 626         /*
 627          * Remove those entries, that don't need flushing.
 628          */
 629         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 630                 if (pfoe->rule->flush == 0) {
 631                         SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
 632                         free(pfoe, M_PFTEMP);
 633                 } else
 634                         counter_u64_add(
 635                             V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
 636
 637         /* If nothing to flush, return. */
 638         if (SLIST_EMPTY(&queue)) {
 639                 CURVNET_RESTORE();
 640                 return;
 641         }
 642
 643         for (int i = 0; i <= pf_hashmask; i++) {
 644                 struct pf_idhash *ih = &V_pf_idhash[i];
 645                 struct pf_state_key *sk;
 646                 struct pf_state *s;
 647
 648                 PF_HASHROW_LOCK(ih);
 649                 LIST_FOREACH(s, &ih->states, entry) {
 650                     sk = s->key[PF_SK_WIRE];
 651                     SLIST_FOREACH(pfoe, &queue, next)
 652                         if (sk->af == pfoe->af &&
 653                             ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
 654                             pfoe->rule == s->rule.ptr) &&
 655                             ((pfoe->dir == PF_OUT &&
 656                             PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
 657                             (pfoe->dir == PF_IN &&
 658                             PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
 659                                 s->timeout = PFTM_PURGE;
 660                                 s->src.state = s->dst.state = TCPS_CLOSED;
 661                                 killed++;
 662                         }
 663                 }
 664                 PF_HASHROW_UNLOCK(ih);
 665         }
 666         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 667                 free(pfoe, M_PFTEMP);
 668         if (V_pf_status.debug >= PF_DEBUG_MISC)
 669                 printf("%s: %u states killed", __func__, killed);
 670
 671         CURVNET_RESTORE();
 672 }
 673
 674 /*
 675  * Can return locked on failure, so that we can consistently
 676  * allocate and insert a new one.
 677  */
 678 struct pf_src_node *
 679 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
 680         int returnlocked)
 681 {
 682         struct pf_srchash *sh;
 683         struct pf_src_node *n;
 684
 685         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
 686
 687         sh = &V_pf_srchash[pf_hashsrc(src, af)];
 688         PF_HASHROW_LOCK(sh);
 689         LIST_FOREACH(n, &sh->nodes, entry)
 690                 if (n->rule.ptr == rule && n->af == af &&
 691                     ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
 692                     (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
 693                         break;
 694         if (n != NULL) {
 695                 n->states++;
 696                 PF_HASHROW_UNLOCK(sh);
 697         } else if (returnlocked == 0)
 698                 PF_HASHROW_UNLOCK(sh);
 699
 700         return (n);
 701 }
 702
 703 static int
 704 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
 705     struct pf_addr *src, sa_family_t af)
 706 {
 707
 708         KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
 709             rule->rpool.opts & PF_POOL_STICKYADDR),
 710             ("%s for non-tracking rule %p", __func__, rule));
 711
 712         if (*sn == NULL)
 713                 *sn = pf_find_src_node(src, rule, af, 1);
 714
 715         if (*sn == NULL) {
 716                 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
 717
 718                 PF_HASHROW_ASSERT(sh);
 719
 720                 if (!rule->max_src_nodes ||
 721                     counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
 722                         (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
 723                 else
 724                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
 725                             1);
 726                 if ((*sn) == NULL) {
 727                         PF_HASHROW_UNLOCK(sh);
 728                         return (-1);
 729                 }
 730
 731                 pf_init_threshold(&(*sn)->conn_rate,
 732                     rule->max_src_conn_rate.limit,
 733                     rule->max_src_conn_rate.seconds);
 734
 735                 (*sn)->af = af;
 736                 (*sn)->rule.ptr = rule;
 737                 PF_ACPY(&(*sn)->addr, src, af);
 738                 LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
 739                 (*sn)->creation = time_uptime;
 740                 (*sn)->ruletype = rule->action;
 741                 (*sn)->states = 1;
 742                 if ((*sn)->rule.ptr != NULL)
 743                         counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
 744                 PF_HASHROW_UNLOCK(sh);
 745                 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
 746         } else {
 747                 if (rule->max_src_states &&
 748                     (*sn)->states >= rule->max_src_states) {
 749                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
 750                             1);
 751                         return (-1);
 752                 }
 753         }
 754         return (0);
 755 }
 756
 757 void
 758 pf_unlink_src_node(struct pf_src_node *src)
 759 {
 760
 761         PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]);
 762         LIST_REMOVE(src, entry);
 763         if (src->rule.ptr)
 764                 counter_u64_add(src->rule.ptr->src_nodes, -1);
 765 }
 766
 767 u_int
 768 pf_free_src_nodes(struct pf_src_node_list *head)
 769 {
 770         struct pf_src_node *sn, *tmp;
 771         u_int count = 0;
 772
 773         LIST_FOREACH_SAFE(sn, head, entry, tmp) {
 774                 uma_zfree(V_pf_sources_z, sn);
 775                 count++;
 776         }
 777
 778         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
 779
 780         return (count);
 781 }
 782
 783 void
 784 pf_mtag_initialize()
 785 {
 786
 787         pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
 788             sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
 789             UMA_ALIGN_PTR, 0);
 790 }
 791
 792 /* Per-vnet data storage structures initialization. */
 793 void
 794 pf_initialize()
 795 {
 796         struct pf_keyhash       *kh;
 797         struct pf_idhash        *ih;
 798         struct pf_srchash       *sh;
 799         u_int i;
 800
 801         if (pf_hashsize == 0 || !powerof2(pf_hashsize))
 802                 pf_hashsize = PF_HASHSIZ;
 803         if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
 804                 pf_srchashsize = PF_SRCHASHSIZ;
 805
 806         V_pf_hashseed = arc4random();
 807
 808         /* States and state keys storage. */
 809         V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
 810             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 811         V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
 812         uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
 813         uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
 814
 815         V_pf_state_key_z = uma_zcreate("pf state keys",
 816             sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
 817             UMA_ALIGN_PTR, 0);
 818
 819         V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash),
 820             M_PFHASH, M_NOWAIT | M_ZERO);
 821         V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash),
 822             M_PFHASH, M_NOWAIT | M_ZERO);
 823         if (V_pf_keyhash == NULL || V_pf_idhash == NULL) {
 824                 printf("pf: Unable to allocate memory for "
 825                     "state_hashsize %lu.\n", pf_hashsize);
 826
 827                 free(V_pf_keyhash, M_PFHASH);
 828                 free(V_pf_idhash, M_PFHASH);
 829
 830                 pf_hashsize = PF_HASHSIZ;
 831                 V_pf_keyhash = mallocarray(pf_hashsize,
 832                     sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO);
 833                 V_pf_idhash = mallocarray(pf_hashsize,
 834                     sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO);
 835         }
 836
 837         pf_hashmask = pf_hashsize - 1;
 838         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 839             i++, kh++, ih++) {
 840                 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
 841                 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
 842         }
 843
 844         /* Source nodes. */
 845         V_pf_sources_z = uma_zcreate("pf source nodes",
 846             sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
 847             0);
 848         V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
 849         uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
 850         uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
 851
 852         V_pf_srchash = mallocarray(pf_srchashsize,
 853             sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO);
 854         if (V_pf_srchash == NULL) {
 855                 printf("pf: Unable to allocate memory for "
 856                     "source_hashsize %lu.\n", pf_srchashsize);
 857
 858                 pf_srchashsize = PF_SRCHASHSIZ;
 859                 V_pf_srchash = mallocarray(pf_srchashsize,
 860                     sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO);
 861         }
 862
 863         pf_srchashmask = pf_srchashsize - 1;
 864         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
 865                 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
 866
 867         /* ALTQ */
 868         TAILQ_INIT(&V_pf_altqs[0]);
 869         TAILQ_INIT(&V_pf_altqs[1]);
 870         TAILQ_INIT(&V_pf_altqs[2]);
 871         TAILQ_INIT(&V_pf_altqs[3]);
 872         TAILQ_INIT(&V_pf_pabuf);
 873         V_pf_altqs_active = &V_pf_altqs[0];
 874         V_pf_altq_ifs_active = &V_pf_altqs[1];
 875         V_pf_altqs_inactive = &V_pf_altqs[2];
 876         V_pf_altq_ifs_inactive = &V_pf_altqs[3];
 877
 878         /* Send & overload+flush queues. */
 879         STAILQ_INIT(&V_pf_sendqueue);
 880         SLIST_INIT(&V_pf_overloadqueue);
 881         TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
 882
 883         /* Unlinked, but may be referenced rules. */
 884         TAILQ_INIT(&V_pf_unlinked_rules);
 885 }
 886
 887 void
 888 pf_mtag_cleanup()
 889 {
 890
 891         uma_zdestroy(pf_mtag_z);
 892 }
 893
 894 void
 895 pf_cleanup()
 896 {
 897         struct pf_keyhash       *kh;
 898         struct pf_idhash        *ih;
 899         struct pf_srchash       *sh;
 900         struct pf_send_entry    *pfse, *next;
 901         u_int i;
 902
 903         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 904             i++, kh++, ih++) {
 905                 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
 906                     __func__));
 907                 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
 908                     __func__));
 909                 mtx_destroy(&kh->lock);
 910                 mtx_destroy(&ih->lock);
 911         }
 912         free(V_pf_keyhash, M_PFHASH);
 913         free(V_pf_idhash, M_PFHASH);
 914
 915         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
 916                 KASSERT(LIST_EMPTY(&sh->nodes),
 917                     ("%s: source node hash not empty", __func__));
 918                 mtx_destroy(&sh->lock);
 919         }
 920         free(V_pf_srchash, M_PFHASH);
 921
 922         STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
 923                 m_freem(pfse->pfse_m);
 924                 free(pfse, M_PFTEMP);
 925         }
 926
 927         uma_zdestroy(V_pf_sources_z);
 928         uma_zdestroy(V_pf_state_z);
 929         uma_zdestroy(V_pf_state_key_z);
 930 }
 931
 932 static int
 933 pf_mtag_uminit(void *mem, int size, int how)
 934 {
 935         struct m_tag *t;
 936
 937         t = (struct m_tag *)mem;
 938         t->m_tag_cookie = MTAG_ABI_COMPAT;
 939         t->m_tag_id = PACKET_TAG_PF;
 940         t->m_tag_len = sizeof(struct pf_mtag);
 941         t->m_tag_free = pf_mtag_free;
 942
 943         return (0);
 944 }
 945
 946 static void
 947 pf_mtag_free(struct m_tag *t)
 948 {
 949
 950         uma_zfree(pf_mtag_z, t);
 951 }
 952
 953 struct pf_mtag *
 954 pf_get_mtag(struct mbuf *m)
 955 {
 956         struct m_tag *mtag;
 957
 958         if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
 959                 return ((struct pf_mtag *)(mtag + 1));
 960
 961         mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
 962         if (mtag == NULL)
 963                 return (NULL);
 964         bzero(mtag + 1, sizeof(struct pf_mtag));
 965         m_tag_prepend(m, mtag);
 966
 967         return ((struct pf_mtag *)(mtag + 1));
 968 }
 969
 970 static int
 971 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
 972     struct pf_state *s)
 973 {
 974         struct pf_keyhash       *khs, *khw, *kh;
 975         struct pf_state_key     *sk, *cur;
 976         struct pf_state         *si, *olds = NULL;
 977         int idx;
 978
 979         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
 980         KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
 981         KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
 982
 983         /*
 984          * We need to lock hash slots of both keys. To avoid deadlock
 985          * we always lock the slot with lower address first. Unlock order
 986          * isn't important.
 987          *
 988          * We also need to lock ID hash slot before dropping key
 989          * locks. On success we return with ID hash slot locked.
 990          */
 991
 992         if (skw == sks) {
 993                 khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
 994                 PF_HASHROW_LOCK(khs);
 995         } else {
 996                 khs = &V_pf_keyhash[pf_hashkey(sks)];
 997                 khw = &V_pf_keyhash[pf_hashkey(skw)];
 998                 if (khs == khw) {
 999                         PF_HASHROW_LOCK(khs);
1000                 } else if (khs < khw) {
1001                         PF_HASHROW_LOCK(khs);
1002                         PF_HASHROW_LOCK(khw);
1003                 } else {
1004                         PF_HASHROW_LOCK(khw);
1005                         PF_HASHROW_LOCK(khs);
1006                 }
1007         }
1008
1009 #define KEYS_UNLOCK()   do {                    \
1010         if (khs != khw) {                       \
1011                 PF_HASHROW_UNLOCK(khs);         \
1012                 PF_HASHROW_UNLOCK(khw);         \
1013         } else                                  \
1014                 PF_HASHROW_UNLOCK(khs);         \
1015 } while (0)
1016
1017         /*
1018          * First run: start with wire key.
1019          */
1020         sk = skw;
1021         kh = khw;
1022         idx = PF_SK_WIRE;
1023
1024 keyattach:
1025         LIST_FOREACH(cur, &kh->keys, entry)
1026                 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
1027                         break;
1028
1029         if (cur != NULL) {
1030                 /* Key exists. Check for same kif, if none, add to key. */
1031                 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
1032                         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
1033
1034                         PF_HASHROW_LOCK(ih);
1035                         if (si->kif == s->kif &&
1036                             si->direction == s->direction) {
1037                                 if (sk->proto == IPPROTO_TCP &&
1038                                     si->src.state >= TCPS_FIN_WAIT_2 &&
1039                                     si->dst.state >= TCPS_FIN_WAIT_2) {
1040                                         /*
1041                                          * New state matches an old >FIN_WAIT_2
1042                                          * state. We can't drop key hash locks,
1043                                          * thus we can't unlink it properly.
1044                                          *
1045                                          * As a workaround we drop it into
1046                                          * TCPS_CLOSED state, schedule purge
1047                                          * ASAP and push it into the very end
1048                                          * of the slot TAILQ, so that it won't
1049                                          * conflict with our new state.
1050                                          */
1051                                         si->src.state = si->dst.state =
1052                                             TCPS_CLOSED;
1053                                         si->timeout = PFTM_PURGE;
1054                                         olds = si;
1055                                 } else {
1056                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
1057                                                 printf("pf: %s key attach "
1058                                                     "failed on %s: ",
1059                                                     (idx == PF_SK_WIRE) ?
1060                                                     "wire" : "stack",
1061                                                     s->kif->pfik_name);
1062                                                 pf_print_state_parts(s,
1063                                                     (idx == PF_SK_WIRE) ?
1064                                                     sk : NULL,
1065                                                     (idx == PF_SK_STACK) ?
1066                                                     sk : NULL);
1067                                                 printf(", existing: ");
1068                                                 pf_print_state_parts(si,
1069                                                     (idx == PF_SK_WIRE) ?
1070                                                     sk : NULL,
1071                                                     (idx == PF_SK_STACK) ?
1072                                                     sk : NULL);
1073                                                 printf("\n");
1074                                         }
1075                                         PF_HASHROW_UNLOCK(ih);
1076                                         KEYS_UNLOCK();
1077                                         uma_zfree(V_pf_state_key_z, sk);
1078                                         if (idx == PF_SK_STACK)
1079                                                 pf_detach_state(s);
1080                                         return (EEXIST); /* collision! */
1081                                 }
1082                         }
1083                         PF_HASHROW_UNLOCK(ih);
1084                 }
1085                 uma_zfree(V_pf_state_key_z, sk);
1086                 s->key[idx] = cur;
1087         } else {
1088                 LIST_INSERT_HEAD(&kh->keys, sk, entry);
1089                 s->key[idx] = sk;
1090         }
1091
1092 stateattach:
1093         /* List is sorted, if-bound states before floating. */
1094         if (s->kif == V_pfi_all)
1095                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1096         else
1097                 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1098
1099         if (olds) {
1100                 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1101                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1102                     key_list[idx]);
1103                 olds = NULL;
1104         }
1105
1106         /*
1107          * Attach done. See how should we (or should not?)
1108          * attach a second key.
1109          */
1110         if (sks == skw) {
1111                 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1112                 idx = PF_SK_STACK;
1113                 sks = NULL;
1114                 goto stateattach;
1115         } else if (sks != NULL) {
1116                 /*
1117                  * Continue attaching with stack key.
1118                  */
1119                 sk = sks;
1120                 kh = khs;
1121                 idx = PF_SK_STACK;
1122                 sks = NULL;
1123                 goto keyattach;
1124         }
1125
1126         PF_STATE_LOCK(s);
1127         KEYS_UNLOCK();
1128
1129         KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1130             ("%s failure", __func__));
1131
1132         return (0);
1133 #undef  KEYS_UNLOCK
1134 }
1135
1136 static void
1137 pf_detach_state(struct pf_state *s)
1138 {
1139         struct pf_state_key *sks = s->key[PF_SK_STACK];
1140         struct pf_keyhash *kh;
1141
1142         if (sks != NULL) {
1143                 kh = &V_pf_keyhash[pf_hashkey(sks)];
1144                 PF_HASHROW_LOCK(kh);
1145                 if (s->key[PF_SK_STACK] != NULL)
1146                         pf_state_key_detach(s, PF_SK_STACK);
1147                 /*
1148                  * If both point to same key, then we are done.
1149                  */
1150                 if (sks == s->key[PF_SK_WIRE]) {
1151                         pf_state_key_detach(s, PF_SK_WIRE);
1152                         PF_HASHROW_UNLOCK(kh);
1153                         return;
1154                 }
1155                 PF_HASHROW_UNLOCK(kh);
1156         }
1157
1158         if (s->key[PF_SK_WIRE] != NULL) {
1159                 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1160                 PF_HASHROW_LOCK(kh);
1161                 if (s->key[PF_SK_WIRE] != NULL)
1162                         pf_state_key_detach(s, PF_SK_WIRE);
1163                 PF_HASHROW_UNLOCK(kh);
1164         }
1165 }
1166
1167 static void
1168 pf_state_key_detach(struct pf_state *s, int idx)
1169 {
1170         struct pf_state_key *sk = s->key[idx];
1171 #ifdef INVARIANTS
1172         struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1173
1174         PF_HASHROW_ASSERT(kh);
1175 #endif
1176         TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1177         s->key[idx] = NULL;
1178
1179         if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1180                 LIST_REMOVE(sk, entry);
1181                 uma_zfree(V_pf_state_key_z, sk);
1182         }
1183 }
1184
1185 static int
1186 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1187 {
1188         struct pf_state_key *sk = mem;
1189
1190         bzero(sk, sizeof(struct pf_state_key_cmp));
1191         TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1192         TAILQ_INIT(&sk->states[PF_SK_STACK]);
1193
1194         return (0);
1195 }
1196
1197 struct pf_state_key *
1198 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1199         struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1200 {
1201         struct pf_state_key *sk;
1202
1203         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1204         if (sk == NULL)
1205                 return (NULL);
1206
1207         PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1208         PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1209         sk->port[pd->sidx] = sport;
1210         sk->port[pd->didx] = dport;
1211         sk->proto = pd->proto;
1212         sk->af = pd->af;
1213
1214         return (sk);
1215 }
1216
1217 struct pf_state_key *
1218 pf_state_key_clone(struct pf_state_key *orig)
1219 {
1220         struct pf_state_key *sk;
1221
1222         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1223         if (sk == NULL)
1224                 return (NULL);
1225
1226         bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1227
1228         return (sk);
1229 }
1230
1231 int
1232 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1233     struct pf_state_key *sks, struct pf_state *s)
1234 {
1235         struct pf_idhash *ih;
1236         struct pf_state *cur;
1237         int error;
1238
1239         KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1240             ("%s: sks not pristine", __func__));
1241         KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1242             ("%s: skw not pristine", __func__));
1243         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1244
1245         s->kif = kif;
1246
1247         if (s->id == 0 && s->creatorid == 0) {
1248                 /* XXX: should be atomic, but probability of collision low */
1249                 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1250                         V_pf_stateid[curcpu] = 1;
1251                 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1252                 s->id = htobe64(s->id);
1253                 s->creatorid = V_pf_status.hostid;
1254         }
1255
1256         /* Returns with ID locked on success. */
1257         if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1258                 return (error);
1259
1260         ih = &V_pf_idhash[PF_IDHASH(s)];
1261         PF_HASHROW_ASSERT(ih);
1262         LIST_FOREACH(cur, &ih->states, entry)
1263                 if (cur->id == s->id && cur->creatorid == s->creatorid)
1264                         break;
1265
1266         if (cur != NULL) {
1267                 PF_HASHROW_UNLOCK(ih);
1268                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1269                         printf("pf: state ID collision: "
1270                             "id: %016llx creatorid: %08x\n",
1271                             (unsigned long long)be64toh(s->id),
1272                             ntohl(s->creatorid));
1273                 }
1274                 pf_detach_state(s);
1275                 return (EEXIST);
1276         }
1277         LIST_INSERT_HEAD(&ih->states, s, entry);
1278         /* One for keys, one for ID hash. */
1279         refcount_init(&s->refs, 2);
1280
1281         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1282         if (V_pfsync_insert_state_ptr != NULL)
1283                 V_pfsync_insert_state_ptr(s);
1284
1285         /* Returns locked. */
1286         return (0);
1287 }
1288
1289 /*
1290  * Find state by ID: returns with locked row on success.
1291  */
1292 struct pf_state *
1293 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1294 {
1295         struct pf_idhash *ih;
1296         struct pf_state *s;
1297
1298         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1299
1300         ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1301
1302         PF_HASHROW_LOCK(ih);
1303         LIST_FOREACH(s, &ih->states, entry)
1304                 if (s->id == id && s->creatorid == creatorid)
1305                         break;
1306
1307         if (s == NULL)
1308                 PF_HASHROW_UNLOCK(ih);
1309
1310         return (s);
1311 }
1312
1313 /*
1314  * Find state by key.
1315  * Returns with ID hash slot locked on success.
1316  */
1317 static struct pf_state *
1318 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1319 {
1320         struct pf_keyhash       *kh;
1321         struct pf_state_key     *sk;
1322         struct pf_state         *s;
1323         int idx;
1324
1325         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1326
1327         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1328
1329         PF_HASHROW_LOCK(kh);
1330         LIST_FOREACH(sk, &kh->keys, entry)
1331                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1332                         break;
1333         if (sk == NULL) {
1334                 PF_HASHROW_UNLOCK(kh);
1335                 return (NULL);
1336         }
1337
1338         idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1339
1340         /* List is sorted, if-bound states before floating ones. */
1341         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1342                 if (s->kif == V_pfi_all || s->kif == kif) {
1343                         PF_STATE_LOCK(s);
1344                         PF_HASHROW_UNLOCK(kh);
1345                         if (s->timeout >= PFTM_MAX) {
1346                                 /*
1347                                  * State is either being processed by
1348                                  * pf_unlink_state() in an other thread, or
1349                                  * is scheduled for immediate expiry.
1350                                  */
1351                                 PF_STATE_UNLOCK(s);
1352                                 return (NULL);
1353                         }
1354                         return (s);
1355                 }
1356         PF_HASHROW_UNLOCK(kh);
1357
1358         return (NULL);
1359 }
1360
1361 struct pf_state *
1362 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1363 {
1364         struct pf_keyhash       *kh;
1365         struct pf_state_key     *sk;
1366         struct pf_state         *s, *ret = NULL;
1367         int                      idx, inout = 0;
1368
1369         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1370
1371         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1372
1373         PF_HASHROW_LOCK(kh);
1374         LIST_FOREACH(sk, &kh->keys, entry)
1375                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1376                         break;
1377         if (sk == NULL) {
1378                 PF_HASHROW_UNLOCK(kh);
1379                 return (NULL);
1380         }
1381         switch (dir) {
1382         case PF_IN:
1383                 idx = PF_SK_WIRE;
1384                 break;
1385         case PF_OUT:
1386                 idx = PF_SK_STACK;
1387                 break;
1388         case PF_INOUT:
1389                 idx = PF_SK_WIRE;
1390                 inout = 1;
1391                 break;
1392         default:
1393                 panic("%s: dir %u", __func__, dir);
1394         }
1395 second_run:
1396         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1397                 if (more == NULL) {
1398                         PF_HASHROW_UNLOCK(kh);
1399                         return (s);
1400                 }
1401
1402                 if (ret)
1403                         (*more)++;
1404                 else
1405                         ret = s;
1406         }
1407         if (inout == 1) {
1408                 inout = 0;
1409                 idx = PF_SK_STACK;
1410                 goto second_run;
1411         }
1412         PF_HASHROW_UNLOCK(kh);
1413
1414         return (ret);
1415 }
1416
1417 /* END state table stuff */
1418
1419 static void
1420 pf_send(struct pf_send_entry *pfse)
1421 {
1422
1423         PF_SENDQ_LOCK();
1424         STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1425         PF_SENDQ_UNLOCK();
1426         swi_sched(V_pf_swi_cookie, 0);
1427 }
1428
1429 void
1430 pf_intr(void *v)
1431 {
1432         struct epoch_tracker et;
1433         struct pf_send_head queue;
1434         struct pf_send_entry *pfse, *next;
1435
1436         CURVNET_SET((struct vnet *)v);
1437
1438         PF_SENDQ_LOCK();
1439         queue = V_pf_sendqueue;
1440         STAILQ_INIT(&V_pf_sendqueue);
1441         PF_SENDQ_UNLOCK();
1442
1443         NET_EPOCH_ENTER(et);
1444
1445         STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1446                 switch (pfse->pfse_type) {
1447 #ifdef INET
1448                 case PFSE_IP:
1449                         ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1450                         break;
1451                 case PFSE_ICMP:
1452                         icmp_error(pfse->pfse_m, pfse->icmpopts.type,
1453                             pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
1454                         break;
1455 #endif /* INET */
1456 #ifdef INET6
1457                 case PFSE_IP6:
1458                         ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1459                             NULL);
1460                         break;
1461                 case PFSE_ICMP6:
1462                         icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
1463                             pfse->icmpopts.code, pfse->icmpopts.mtu);
1464                         break;
1465 #endif /* INET6 */
1466                 default:
1467                         panic("%s: unknown type", __func__);
1468                 }
1469                 free(pfse, M_PFTEMP);
1470         }
1471         NET_EPOCH_EXIT(et);
1472         CURVNET_RESTORE();
1473 }
1474
1475 void
1476 pf_purge_thread(void *unused __unused)
1477 {
1478         VNET_ITERATOR_DECL(vnet_iter);
1479
1480         sx_xlock(&pf_end_lock);
1481         while (pf_end_threads == 0) {
1482                 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", hz / 10);
1483
1484                 VNET_LIST_RLOCK();
1485                 VNET_FOREACH(vnet_iter) {
1486                         CURVNET_SET(vnet_iter);
1487
1488                         /* Wait until V_pf_default_rule is initialized. */
1489                         if (V_pf_vnet_active == 0) {
1490                                 CURVNET_RESTORE();
1491                                 continue;
1492                         }
1493
1494                         /*
1495                          *  Process 1/interval fraction of the state
1496                          * table every run.
1497                          */
1498                         V_pf_purge_idx =
1499                             pf_purge_expired_states(V_pf_purge_idx, pf_hashmask /
1500                             (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1501
1502                         /*
1503                          * Purge other expired types every
1504                          * PFTM_INTERVAL seconds.
1505                          */
1506                         if (V_pf_purge_idx == 0) {
1507                                 /*
1508                                  * Order is important:
1509                                  * - states and src nodes reference rules
1510                                  * - states and rules reference kifs
1511                                  */
1512                                 pf_purge_expired_fragments();
1513                                 pf_purge_expired_src_nodes();
1514                                 pf_purge_unlinked_rules();
1515                                 pfi_kif_purge();
1516                         }
1517                         CURVNET_RESTORE();
1518                 }
1519                 VNET_LIST_RUNLOCK();
1520         }
1521
1522         pf_end_threads++;
1523         sx_xunlock(&pf_end_lock);
1524         kproc_exit(0);
1525 }
1526
1527 void
1528 pf_unload_vnet_purge(void)
1529 {
1530
1531         /*
1532          * To cleanse up all kifs and rules we need
1533          * two runs: first one clears reference flags,
1534          * then pf_purge_expired_states() doesn't
1535          * raise them, and then second run frees.
1536          */
1537         pf_purge_unlinked_rules();
1538         pfi_kif_purge();
1539
1540         /*
1541          * Now purge everything.
1542          */
1543         pf_purge_expired_states(0, pf_hashmask);
1544         pf_purge_fragments(UINT_MAX);
1545         pf_purge_expired_src_nodes();
1546
1547         /*
1548          * Now all kifs & rules should be unreferenced,
1549          * thus should be successfully freed.
1550          */
1551         pf_purge_unlinked_rules();
1552         pfi_kif_purge();
1553 }
1554
1555 u_int32_t
1556 pf_state_expires(const struct pf_state *state)
1557 {
1558         u_int32_t       timeout;
1559         u_int32_t       start;
1560         u_int32_t       end;
1561         u_int32_t       states;
1562
1563         /* handle all PFTM_* > PFTM_MAX here */
1564         if (state->timeout == PFTM_PURGE)
1565                 return (time_uptime);
1566         KASSERT(state->timeout != PFTM_UNLINKED,
1567             ("pf_state_expires: timeout == PFTM_UNLINKED"));
1568         KASSERT((state->timeout < PFTM_MAX),
1569             ("pf_state_expires: timeout > PFTM_MAX"));
1570         timeout = state->rule.ptr->timeout[state->timeout];
1571         if (!timeout)
1572                 timeout = V_pf_default_rule.timeout[state->timeout];
1573         start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1574         if (start && state->rule.ptr != &V_pf_default_rule) {
1575                 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1576                 states = counter_u64_fetch(state->rule.ptr->states_cur);
1577         } else {
1578                 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1579                 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1580                 states = V_pf_status.states;
1581         }
1582         if (end && states > start && start < end) {
1583                 if (states < end) {
1584                         timeout = (u_int64_t)timeout * (end - states) /
1585                             (end - start);
1586                         return (state->expire + timeout);
1587                 }
1588                 else
1589                         return (time_uptime);
1590         }
1591         return (state->expire + timeout);
1592 }
1593
1594 void
1595 pf_purge_expired_src_nodes()
1596 {
1597         struct pf_src_node_list  freelist;
1598         struct pf_srchash       *sh;
1599         struct pf_src_node      *cur, *next;
1600         int i;
1601
1602         LIST_INIT(&freelist);
1603         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1604             PF_HASHROW_LOCK(sh);
1605             LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1606                 if (cur->states == 0 && cur->expire <= time_uptime) {
1607                         pf_unlink_src_node(cur);
1608                         LIST_INSERT_HEAD(&freelist, cur, entry);
1609                 } else if (cur->rule.ptr != NULL)
1610                         cur->rule.ptr->rule_flag |= PFRULE_REFS;
1611             PF_HASHROW_UNLOCK(sh);
1612         }
1613
1614         pf_free_src_nodes(&freelist);
1615
1616         V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1617 }
1618
1619 static void
1620 pf_src_tree_remove_state(struct pf_state *s)
1621 {
1622         struct pf_src_node *sn;
1623         struct pf_srchash *sh;
1624         uint32_t timeout;
1625
1626         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
1627             s->rule.ptr->timeout[PFTM_SRC_NODE] :
1628             V_pf_default_rule.timeout[PFTM_SRC_NODE];
1629
1630         if (s->src_node != NULL) {
1631                 sn = s->src_node;
1632                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1633                 PF_HASHROW_LOCK(sh);
1634                 if (s->src.tcp_est)
1635                         --sn->conn;
1636                 if (--sn->states == 0)
1637                         sn->expire = time_uptime + timeout;
1638                 PF_HASHROW_UNLOCK(sh);
1639         }
1640         if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1641                 sn = s->nat_src_node;
1642                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1643                 PF_HASHROW_LOCK(sh);
1644                 if (--sn->states == 0)
1645                         sn->expire = time_uptime + timeout;
1646                 PF_HASHROW_UNLOCK(sh);
1647         }
1648         s->src_node = s->nat_src_node = NULL;
1649 }
1650
1651 /*
1652  * Unlink and potentilly free a state. Function may be
1653  * called with ID hash row locked, but always returns
1654  * unlocked, since it needs to go through key hash locking.
1655  */
1656 int
1657 pf_unlink_state(struct pf_state *s, u_int flags)
1658 {
1659         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1660
1661         if ((flags & PF_ENTER_LOCKED) == 0)
1662                 PF_HASHROW_LOCK(ih);
1663         else
1664                 PF_HASHROW_ASSERT(ih);
1665
1666         if (s->timeout == PFTM_UNLINKED) {
1667                 /*
1668                  * State is being processed
1669                  * by pf_unlink_state() in
1670                  * an other thread.
1671                  */
1672                 PF_HASHROW_UNLOCK(ih);
1673                 return (0);     /* XXXGL: undefined actually */
1674         }
1675
1676         if (s->src.state == PF_TCPS_PROXY_DST) {
1677                 /* XXX wire key the right one? */
1678                 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1679                     &s->key[PF_SK_WIRE]->addr[1],
1680                     &s->key[PF_SK_WIRE]->addr[0],
1681                     s->key[PF_SK_WIRE]->port[1],
1682                     s->key[PF_SK_WIRE]->port[0],
1683                     s->src.seqhi, s->src.seqlo + 1,
1684                     TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1685         }
1686
1687         LIST_REMOVE(s, entry);
1688         pf_src_tree_remove_state(s);
1689
1690         if (V_pfsync_delete_state_ptr != NULL)
1691                 V_pfsync_delete_state_ptr(s);
1692
1693         STATE_DEC_COUNTERS(s);
1694
1695         s->timeout = PFTM_UNLINKED;
1696
1697         PF_HASHROW_UNLOCK(ih);
1698
1699         pf_detach_state(s);
1700         /* pf_state_insert() initialises refs to 2, so we can never release the
1701          * last reference here, only in pf_release_state(). */
1702         (void)refcount_release(&s->refs);
1703
1704         return (pf_release_state(s));
1705 }
1706
1707 void
1708 pf_free_state(struct pf_state *cur)
1709 {
1710
1711         KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1712         KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1713             cur->timeout));
1714
1715         pf_normalize_tcp_cleanup(cur);
1716         uma_zfree(V_pf_state_z, cur);
1717         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
1718 }
1719
1720 /*
1721  * Called only from pf_purge_thread(), thus serialized.
1722  */
1723 static u_int
1724 pf_purge_expired_states(u_int i, int maxcheck)
1725 {
1726         struct pf_idhash *ih;
1727         struct pf_state *s;
1728
1729         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1730
1731         /*
1732          * Go through hash and unlink states that expire now.
1733          */
1734         while (maxcheck > 0) {
1735                 ih = &V_pf_idhash[i];
1736
1737                 /* only take the lock if we expect to do work */
1738                 if (!LIST_EMPTY(&ih->states)) {
1739 relock:
1740                         PF_HASHROW_LOCK(ih);
1741                         LIST_FOREACH(s, &ih->states, entry) {
1742                                 if (pf_state_expires(s) <= time_uptime) {
1743                                         V_pf_status.states -=
1744                                             pf_unlink_state(s, PF_ENTER_LOCKED);
1745                                         goto relock;
1746                                 }
1747                                 s->rule.ptr->rule_flag |= PFRULE_REFS;
1748                                 if (s->nat_rule.ptr != NULL)
1749                                         s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1750                                 if (s->anchor.ptr != NULL)
1751                                         s->anchor.ptr->rule_flag |= PFRULE_REFS;
1752                                 s->kif->pfik_flags |= PFI_IFLAG_REFS;
1753                                 if (s->rt_kif)
1754                                         s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1755                         }
1756                         PF_HASHROW_UNLOCK(ih);
1757                 }
1758
1759                 /* Return when we hit end of hash. */
1760                 if (++i > pf_hashmask) {
1761                         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1762                         return (0);
1763                 }
1764
1765                 maxcheck--;
1766         }
1767
1768         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1769
1770         return (i);
1771 }
1772
1773 static void
1774 pf_purge_unlinked_rules()
1775 {
1776         struct pf_rulequeue tmpq;
1777         struct pf_rule *r, *r1;
1778
1779         /*
1780          * If we have overloading task pending, then we'd
1781          * better skip purging this time. There is a tiny
1782          * probability that overloading task references
1783          * an already unlinked rule.
1784          */
1785         PF_OVERLOADQ_LOCK();
1786         if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1787                 PF_OVERLOADQ_UNLOCK();
1788                 return;
1789         }
1790         PF_OVERLOADQ_UNLOCK();
1791
1792         /*
1793          * Do naive mark-and-sweep garbage collecting of old rules.
1794          * Reference flag is raised by pf_purge_expired_states()
1795          * and pf_purge_expired_src_nodes().
1796          *
1797          * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1798          * use a temporary queue.
1799          */
1800         TAILQ_INIT(&tmpq);
1801         PF_UNLNKDRULES_LOCK();
1802         TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1803                 if (!(r->rule_flag & PFRULE_REFS)) {
1804                         TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1805                         TAILQ_INSERT_TAIL(&tmpq, r, entries);
1806                 } else
1807                         r->rule_flag &= ~PFRULE_REFS;
1808         }
1809         PF_UNLNKDRULES_UNLOCK();
1810
1811         if (!TAILQ_EMPTY(&tmpq)) {
1812                 PF_RULES_WLOCK();
1813                 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1814                         TAILQ_REMOVE(&tmpq, r, entries);
1815                         pf_free_rule(r);
1816                 }
1817                 PF_RULES_WUNLOCK();
1818         }
1819 }
1820
1821 void
1822 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1823 {
1824         switch (af) {
1825 #ifdef INET
1826         case AF_INET: {
1827                 u_int32_t a = ntohl(addr->addr32[0]);
1828                 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1829                     (a>>8)&255, a&255);
1830                 if (p) {
1831                         p = ntohs(p);
1832                         printf(":%u", p);
1833                 }
1834                 break;
1835         }
1836 #endif /* INET */
1837 #ifdef INET6
1838         case AF_INET6: {
1839                 u_int16_t b;
1840                 u_int8_t i, curstart, curend, maxstart, maxend;
1841                 curstart = curend = maxstart = maxend = 255;
1842                 for (i = 0; i < 8; i++) {
1843                         if (!addr->addr16[i]) {
1844                                 if (curstart == 255)
1845                                         curstart = i;
1846                                 curend = i;
1847                         } else {
1848                                 if ((curend - curstart) >
1849                                     (maxend - maxstart)) {
1850                                         maxstart = curstart;
1851                                         maxend = curend;
1852                                 }
1853                                 curstart = curend = 255;
1854                         }
1855                 }
1856                 if ((curend - curstart) >
1857                     (maxend - maxstart)) {
1858                         maxstart = curstart;
1859                         maxend = curend;
1860                 }
1861                 for (i = 0; i < 8; i++) {
1862                         if (i >= maxstart && i <= maxend) {
1863                                 if (i == 0)
1864                                         printf(":");
1865                                 if (i == maxend)
1866                                         printf(":");
1867                         } else {
1868                                 b = ntohs(addr->addr16[i]);
1869                                 printf("%x", b);
1870                                 if (i < 7)
1871                                         printf(":");
1872                         }
1873                 }
1874                 if (p) {
1875                         p = ntohs(p);
1876                         printf("[%u]", p);
1877                 }
1878                 break;
1879         }
1880 #endif /* INET6 */
1881         }
1882 }
1883
1884 void
1885 pf_print_state(struct pf_state *s)
1886 {
1887         pf_print_state_parts(s, NULL, NULL);
1888 }
1889
1890 static void
1891 pf_print_state_parts(struct pf_state *s,
1892     struct pf_state_key *skwp, struct pf_state_key *sksp)
1893 {
1894         struct pf_state_key *skw, *sks;
1895         u_int8_t proto, dir;
1896
1897         /* Do our best to fill these, but they're skipped if NULL */
1898         skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1899         sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1900         proto = skw ? skw->proto : (sks ? sks->proto : 0);
1901         dir = s ? s->direction : 0;
1902
1903         switch (proto) {
1904         case IPPROTO_IPV4:
1905                 printf("IPv4");
1906                 break;
1907         case IPPROTO_IPV6:
1908                 printf("IPv6");
1909                 break;
1910         case IPPROTO_TCP:
1911                 printf("TCP");
1912                 break;
1913         case IPPROTO_UDP:
1914                 printf("UDP");
1915                 break;
1916         case IPPROTO_ICMP:
1917                 printf("ICMP");
1918                 break;
1919         case IPPROTO_ICMPV6:
1920                 printf("ICMPv6");
1921                 break;
1922         default:
1923                 printf("%u", proto);
1924                 break;
1925         }
1926         switch (dir) {
1927         case PF_IN:
1928                 printf(" in");
1929                 break;
1930         case PF_OUT:
1931                 printf(" out");
1932                 break;
1933         }
1934         if (skw) {
1935                 printf(" wire: ");
1936                 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1937                 printf(" ");
1938                 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1939         }
1940         if (sks) {
1941                 printf(" stack: ");
1942                 if (sks != skw) {
1943                         pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1944                         printf(" ");
1945                         pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1946                 } else
1947                         printf("-");
1948         }
1949         if (s) {
1950                 if (proto == IPPROTO_TCP) {
1951                         printf(" [lo=%u high=%u win=%u modulator=%u",
1952                             s->src.seqlo, s->src.seqhi,
1953                             s->src.max_win, s->src.seqdiff);
1954                         if (s->src.wscale && s->dst.wscale)
1955                                 printf(" wscale=%u",
1956                                     s->src.wscale & PF_WSCALE_MASK);
1957                         printf("]");
1958                         printf(" [lo=%u high=%u win=%u modulator=%u",
1959                             s->dst.seqlo, s->dst.seqhi,
1960                             s->dst.max_win, s->dst.seqdiff);
1961                         if (s->src.wscale && s->dst.wscale)
1962                                 printf(" wscale=%u",
1963                                 s->dst.wscale & PF_WSCALE_MASK);
1964                         printf("]");
1965                 }
1966                 printf(" %u:%u", s->src.state, s->dst.state);
1967         }
1968 }
1969
1970 void
1971 pf_print_flags(u_int8_t f)
1972 {
1973         if (f)
1974                 printf(" ");
1975         if (f & TH_FIN)
1976                 printf("F");
1977         if (f & TH_SYN)
1978                 printf("S");
1979         if (f & TH_RST)
1980                 printf("R");
1981         if (f & TH_PUSH)
1982                 printf("P");
1983         if (f & TH_ACK)
1984                 printf("A");
1985         if (f & TH_URG)
1986                 printf("U");
1987         if (f & TH_ECE)
1988                 printf("E");
1989         if (f & TH_CWR)
1990                 printf("W");
1991 }
1992
1993 #define PF_SET_SKIP_STEPS(i)                                    \
1994         do {                                                    \
1995                 while (head[i] != cur) {                        \
1996                         head[i]->skip[i].ptr = cur;             \
1997                         head[i] = TAILQ_NEXT(head[i], entries); \
1998                 }                                               \
1999         } while (0)
2000
2001 void
2002 pf_calc_skip_steps(struct pf_rulequeue *rules)
2003 {
2004         struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
2005         int i;
2006
2007         cur = TAILQ_FIRST(rules);
2008         prev = cur;
2009         for (i = 0; i < PF_SKIP_COUNT; ++i)
2010                 head[i] = cur;
2011         while (cur != NULL) {
2012                 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
2013                         PF_SET_SKIP_STEPS(PF_SKIP_IFP);
2014                 if (cur->direction != prev->direction)
2015                         PF_SET_SKIP_STEPS(PF_SKIP_DIR);
2016                 if (cur->af != prev->af)
2017                         PF_SET_SKIP_STEPS(PF_SKIP_AF);
2018                 if (cur->proto != prev->proto)
2019                         PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
2020                 if (cur->src.neg != prev->src.neg ||
2021                     pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
2022                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
2023                 if (cur->src.port[0] != prev->src.port[0] ||
2024                     cur->src.port[1] != prev->src.port[1] ||
2025                     cur->src.port_op != prev->src.port_op)
2026                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
2027                 if (cur->dst.neg != prev->dst.neg ||
2028                     pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
2029                         PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
2030                 if (cur->dst.port[0] != prev->dst.port[0] ||
2031                     cur->dst.port[1] != prev->dst.port[1] ||
2032                     cur->dst.port_op != prev->dst.port_op)
2033                         PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
2034
2035                 prev = cur;
2036                 cur = TAILQ_NEXT(cur, entries);
2037         }
2038         for (i = 0; i < PF_SKIP_COUNT; ++i)
2039                 PF_SET_SKIP_STEPS(i);
2040 }
2041
2042 static int
2043 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
2044 {
2045         if (aw1->type != aw2->type)
2046                 return (1);
2047         switch (aw1->type) {
2048         case PF_ADDR_ADDRMASK:
2049         case PF_ADDR_RANGE:
2050                 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
2051                         return (1);
2052                 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
2053                         return (1);
2054                 return (0);
2055         case PF_ADDR_DYNIFTL:
2056                 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
2057         case PF_ADDR_NOROUTE:
2058         case PF_ADDR_URPFFAILED:
2059                 return (0);
2060         case PF_ADDR_TABLE:
2061                 return (aw1->p.tbl != aw2->p.tbl);
2062         default:
2063                 printf("invalid address type: %d\n", aw1->type);
2064                 return (1);
2065         }
2066 }
2067
2068 /**
2069  * Checksum updates are a little complicated because the checksum in the TCP/UDP
2070  * header isn't always a full checksum. In some cases (i.e. output) it's a
2071  * pseudo-header checksum, which is a partial checksum over src/dst IP
2072  * addresses, protocol number and length.
2073  *
2074  * That means we have the following cases:
2075  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2076  *      checksums, we need to update the checksum whenever we change anything.
2077  *  * Output (i.e. the checksum is a pseudo-header checksum):
2078  *      x The field being updated is src/dst address or affects the length of
2079  *      the packet. We need to update the pseudo-header checksum (note that this
2080  *      checksum is not ones' complement).
2081  *      x Some other field is being modified (e.g. src/dst port numbers): We
2082  *      don't have to update anything.
2083  **/
2084 u_int16_t
2085 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2086 {
2087         u_int32_t       l;
2088
2089         if (udp && !cksum)
2090                 return (0x0000);
2091         l = cksum + old - new;
2092         l = (l >> 16) + (l & 65535);
2093         l = l & 65535;
2094         if (udp && !l)
2095                 return (0xFFFF);
2096         return (l);
2097 }
2098
2099 u_int16_t
2100 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2101         u_int16_t new, u_int8_t udp)
2102 {
2103         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2104                 return (cksum);
2105
2106         return (pf_cksum_fixup(cksum, old, new, udp));
2107 }
2108
2109 static void
2110 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2111         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2112         sa_family_t af)
2113 {
2114         struct pf_addr  ao;
2115         u_int16_t       po = *p;
2116
2117         PF_ACPY(&ao, a, af);
2118         PF_ACPY(a, an, af);
2119
2120         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2121                 *pc = ~*pc;
2122
2123         *p = pn;
2124
2125         switch (af) {
2126 #ifdef INET
2127         case AF_INET:
2128                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2129                     ao.addr16[0], an->addr16[0], 0),
2130                     ao.addr16[1], an->addr16[1], 0);
2131                 *p = pn;
2132
2133                 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2134                     ao.addr16[0], an->addr16[0], u),
2135                     ao.addr16[1], an->addr16[1], u);
2136
2137                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2138                 break;
2139 #endif /* INET */
2140 #ifdef INET6
2141         case AF_INET6:
2142                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2143                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2144                     pf_cksum_fixup(pf_cksum_fixup(*pc,
2145                     ao.addr16[0], an->addr16[0], u),
2146                     ao.addr16[1], an->addr16[1], u),
2147                     ao.addr16[2], an->addr16[2], u),
2148                     ao.addr16[3], an->addr16[3], u),
2149                     ao.addr16[4], an->addr16[4], u),
2150                     ao.addr16[5], an->addr16[5], u),
2151                     ao.addr16[6], an->addr16[6], u),
2152                     ao.addr16[7], an->addr16[7], u);
2153
2154                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2155                 break;
2156 #endif /* INET6 */
2157         }
2158
2159         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2160             CSUM_DELAY_DATA_IPV6)) {
2161                 *pc = ~*pc;
2162                 if (! *pc)
2163                         *pc = 0xffff;
2164         }
2165 }
2166
2167 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2168 void
2169 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2170 {
2171         u_int32_t       ao;
2172
2173         memcpy(&ao, a, sizeof(ao));
2174         memcpy(a, &an, sizeof(u_int32_t));
2175         *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2176             ao % 65536, an % 65536, u);
2177 }
2178
2179 void
2180 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2181 {
2182         u_int32_t       ao;
2183
2184         memcpy(&ao, a, sizeof(ao));
2185         memcpy(a, &an, sizeof(u_int32_t));
2186
2187         *c = pf_proto_cksum_fixup(m,
2188             pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2189             ao % 65536, an % 65536, udp);
2190 }
2191
2192 #ifdef INET6
2193 static void
2194 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2195 {
2196         struct pf_addr  ao;
2197
2198         PF_ACPY(&ao, a, AF_INET6);
2199         PF_ACPY(a, an, AF_INET6);
2200
2201         *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2202             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2203             pf_cksum_fixup(pf_cksum_fixup(*c,
2204             ao.addr16[0], an->addr16[0], u),
2205             ao.addr16[1], an->addr16[1], u),
2206             ao.addr16[2], an->addr16[2], u),
2207             ao.addr16[3], an->addr16[3], u),
2208             ao.addr16[4], an->addr16[4], u),
2209             ao.addr16[5], an->addr16[5], u),
2210             ao.addr16[6], an->addr16[6], u),
2211             ao.addr16[7], an->addr16[7], u);
2212 }
2213 #endif /* INET6 */
2214
2215 static void
2216 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2217     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2218     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2219 {
2220         struct pf_addr  oia, ooa;
2221
2222         PF_ACPY(&oia, ia, af);
2223         if (oa)
2224                 PF_ACPY(&ooa, oa, af);
2225
2226         /* Change inner protocol port, fix inner protocol checksum. */
2227         if (ip != NULL) {
2228                 u_int16_t       oip = *ip;
2229                 u_int32_t       opc;
2230
2231                 if (pc != NULL)
2232                         opc = *pc;
2233                 *ip = np;
2234                 if (pc != NULL)
2235                         *pc = pf_cksum_fixup(*pc, oip, *ip, u);
2236                 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2237                 if (pc != NULL)
2238                         *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2239         }
2240         /* Change inner ip address, fix inner ip and icmp checksums. */
2241         PF_ACPY(ia, na, af);
2242         switch (af) {
2243 #ifdef INET
2244         case AF_INET: {
2245                 u_int32_t        oh2c = *h2c;
2246
2247                 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2248                     oia.addr16[0], ia->addr16[0], 0),
2249                     oia.addr16[1], ia->addr16[1], 0);
2250                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2251                     oia.addr16[0], ia->addr16[0], 0),
2252                     oia.addr16[1], ia->addr16[1], 0);
2253                 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2254                 break;
2255         }
2256 #endif /* INET */
2257 #ifdef INET6
2258         case AF_INET6:
2259                 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2260                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2261                     pf_cksum_fixup(pf_cksum_fixup(*ic,
2262                     oia.addr16[0], ia->addr16[0], u),
2263                     oia.addr16[1], ia->addr16[1], u),
2264                     oia.addr16[2], ia->addr16[2], u),
2265                     oia.addr16[3], ia->addr16[3], u),
2266                     oia.addr16[4], ia->addr16[4], u),
2267                     oia.addr16[5], ia->addr16[5], u),
2268                     oia.addr16[6], ia->addr16[6], u),
2269                     oia.addr16[7], ia->addr16[7], u);
2270                 break;
2271 #endif /* INET6 */
2272         }
2273         /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2274         if (oa) {
2275                 PF_ACPY(oa, na, af);
2276                 switch (af) {
2277 #ifdef INET
2278                 case AF_INET:
2279                         *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2280                             ooa.addr16[0], oa->addr16[0], 0),
2281                             ooa.addr16[1], oa->addr16[1], 0);
2282                         break;
2283 #endif /* INET */
2284 #ifdef INET6
2285                 case AF_INET6:
2286                         *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2287                             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2288                             pf_cksum_fixup(pf_cksum_fixup(*ic,
2289                             ooa.addr16[0], oa->addr16[0], u),
2290                             ooa.addr16[1], oa->addr16[1], u),
2291                             ooa.addr16[2], oa->addr16[2], u),
2292                             ooa.addr16[3], oa->addr16[3], u),
2293                             ooa.addr16[4], oa->addr16[4], u),
2294                             ooa.addr16[5], oa->addr16[5], u),
2295                             ooa.addr16[6], oa->addr16[6], u),
2296                             ooa.addr16[7], oa->addr16[7], u);
2297                         break;
2298 #endif /* INET6 */
2299                 }
2300         }
2301 }
2302
2303 /*
2304  * Need to modulate the sequence numbers in the TCP SACK option
2305  * (credits to Krzysztof Pfaff for report and patch)
2306  */
2307 static int
2308 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2309     struct tcphdr *th, struct pf_state_peer *dst)
2310 {
2311         int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2312         u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2313         int copyback = 0, i, olen;
2314         struct sackblk sack;
2315
2316 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2317         if (hlen < TCPOLEN_SACKLEN ||
2318             !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2319                 return 0;
2320
2321         while (hlen >= TCPOLEN_SACKLEN) {
2322                 olen = opt[1];
2323                 switch (*opt) {
2324                 case TCPOPT_EOL:        /* FALLTHROUGH */
2325                 case TCPOPT_NOP:
2326                         opt++;
2327                         hlen--;
2328                         break;
2329                 case TCPOPT_SACK:
2330                         if (olen > hlen)
2331                                 olen = hlen;
2332                         if (olen >= TCPOLEN_SACKLEN) {
2333                                 for (i = 2; i + TCPOLEN_SACK <= olen;
2334                                     i += TCPOLEN_SACK) {
2335                                         memcpy(&sack, &opt[i], sizeof(sack));
2336                                         pf_change_proto_a(m, &sack.start, &th->th_sum,
2337                                             htonl(ntohl(sack.start) - dst->seqdiff), 0);
2338                                         pf_change_proto_a(m, &sack.end, &th->th_sum,
2339                                             htonl(ntohl(sack.end) - dst->seqdiff), 0);
2340                                         memcpy(&opt[i], &sack, sizeof(sack));
2341                                 }
2342                                 copyback = 1;
2343                         }
2344                         /* FALLTHROUGH */
2345                 default:
2346                         if (olen < 2)
2347                                 olen = 2;
2348                         hlen -= olen;
2349                         opt += olen;
2350                 }
2351         }
2352
2353         if (copyback)
2354                 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2355         return (copyback);
2356 }
2357
2358 static void
2359 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2360     const struct pf_addr *saddr, const struct pf_addr *daddr,
2361     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2362     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2363     u_int16_t rtag, struct ifnet *ifp)
2364 {
2365         struct pf_send_entry *pfse;
2366         struct mbuf     *m;
2367         int              len, tlen;
2368 #ifdef INET
2369         struct ip       *h = NULL;
2370 #endif /* INET */
2371 #ifdef INET6
2372         struct ip6_hdr  *h6 = NULL;
2373 #endif /* INET6 */
2374         struct tcphdr   *th;
2375         char            *opt;
2376         struct pf_mtag  *pf_mtag;
2377
2378         len = 0;
2379         th = NULL;
2380
2381         /* maximum segment size tcp option */
2382         tlen = sizeof(struct tcphdr);
2383         if (mss)
2384                 tlen += 4;
2385
2386         switch (af) {
2387 #ifdef INET
2388         case AF_INET:
2389                 len = sizeof(struct ip) + tlen;
2390                 break;
2391 #endif /* INET */
2392 #ifdef INET6
2393         case AF_INET6:
2394                 len = sizeof(struct ip6_hdr) + tlen;
2395                 break;
2396 #endif /* INET6 */
2397         default:
2398                 panic("%s: unsupported af %d", __func__, af);
2399         }
2400
2401         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2402         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2403         if (pfse == NULL)
2404                 return;
2405         m = m_gethdr(M_NOWAIT, MT_DATA);
2406         if (m == NULL) {
2407                 free(pfse, M_PFTEMP);
2408                 return;
2409         }
2410 #ifdef MAC
2411         mac_netinet_firewall_send(m);
2412 #endif
2413         if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2414                 free(pfse, M_PFTEMP);
2415                 m_freem(m);
2416                 return;
2417         }
2418         if (tag)
2419                 m->m_flags |= M_SKIP_FIREWALL;
2420         pf_mtag->tag = rtag;
2421
2422         if (r != NULL && r->rtableid >= 0)
2423                 M_SETFIB(m, r->rtableid);
2424
2425 #ifdef ALTQ
2426         if (r != NULL && r->qid) {
2427                 pf_mtag->qid = r->qid;
2428
2429                 /* add hints for ecn */
2430                 pf_mtag->hdr = mtod(m, struct ip *);
2431         }
2432 #endif /* ALTQ */
2433         m->m_data += max_linkhdr;
2434         m->m_pkthdr.len = m->m_len = len;
2435         m->m_pkthdr.rcvif = NULL;
2436         bzero(m->m_data, len);
2437         switch (af) {
2438 #ifdef INET
2439         case AF_INET:
2440                 h = mtod(m, struct ip *);
2441
2442                 /* IP header fields included in the TCP checksum */
2443                 h->ip_p = IPPROTO_TCP;
2444                 h->ip_len = htons(tlen);
2445                 h->ip_src.s_addr = saddr->v4.s_addr;
2446                 h->ip_dst.s_addr = daddr->v4.s_addr;
2447
2448                 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2449                 break;
2450 #endif /* INET */
2451 #ifdef INET6
2452         case AF_INET6:
2453                 h6 = mtod(m, struct ip6_hdr *);
2454
2455                 /* IP header fields included in the TCP checksum */
2456                 h6->ip6_nxt = IPPROTO_TCP;
2457                 h6->ip6_plen = htons(tlen);
2458                 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2459                 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2460
2461                 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2462                 break;
2463 #endif /* INET6 */
2464         }
2465
2466         /* TCP header */
2467         th->th_sport = sport;
2468         th->th_dport = dport;
2469         th->th_seq = htonl(seq);
2470         th->th_ack = htonl(ack);
2471         th->th_off = tlen >> 2;
2472         th->th_flags = flags;
2473         th->th_win = htons(win);
2474
2475         if (mss) {
2476                 opt = (char *)(th + 1);
2477                 opt[0] = TCPOPT_MAXSEG;
2478                 opt[1] = 4;
2479                 HTONS(mss);
2480                 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2481         }
2482
2483         switch (af) {
2484 #ifdef INET
2485         case AF_INET:
2486                 /* TCP checksum */
2487                 th->th_sum = in_cksum(m, len);
2488
2489                 /* Finish the IP header */
2490                 h->ip_v = 4;
2491                 h->ip_hl = sizeof(*h) >> 2;
2492                 h->ip_tos = IPTOS_LOWDELAY;
2493                 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2494                 h->ip_len = htons(len);
2495                 h->ip_ttl = ttl ? ttl : V_ip_defttl;
2496                 h->ip_sum = 0;
2497
2498                 pfse->pfse_type = PFSE_IP;
2499                 break;
2500 #endif /* INET */
2501 #ifdef INET6
2502         case AF_INET6:
2503                 /* TCP checksum */
2504                 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2505                     sizeof(struct ip6_hdr), tlen);
2506
2507                 h6->ip6_vfc |= IPV6_VERSION;
2508                 h6->ip6_hlim = IPV6_DEFHLIM;
2509
2510                 pfse->pfse_type = PFSE_IP6;
2511                 break;
2512 #endif /* INET6 */
2513         }
2514         pfse->pfse_m = m;
2515         pf_send(pfse);
2516 }
2517
2518 static void
2519 pf_return(struct pf_rule *r, struct pf_rule *nr, struct pf_pdesc *pd,
2520     struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th,
2521     struct pfi_kif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
2522     u_short *reason)
2523 {
2524         struct pf_addr  * const saddr = pd->src;
2525         struct pf_addr  * const daddr = pd->dst;
2526         sa_family_t      af = pd->af;
2527
2528         /* undo NAT changes, if they have taken place */
2529         if (nr != NULL) {
2530                 PF_ACPY(saddr, &sk->addr[pd->sidx], af);
2531                 PF_ACPY(daddr, &sk->addr[pd->didx], af);
2532                 if (pd->sport)
2533                         *pd->sport = sk->port[pd->sidx];
2534                 if (pd->dport)
2535                         *pd->dport = sk->port[pd->didx];
2536                 if (pd->proto_sum)
2537                         *pd->proto_sum = bproto_sum;
2538                 if (pd->ip_sum)
2539                         *pd->ip_sum = bip_sum;
2540                 m_copyback(m, off, hdrlen, pd->hdr.any);
2541         }
2542         if (pd->proto == IPPROTO_TCP &&
2543             ((r->rule_flag & PFRULE_RETURNRST) ||
2544             (r->rule_flag & PFRULE_RETURN)) &&
2545             !(th->th_flags & TH_RST)) {
2546                 u_int32_t        ack = ntohl(th->th_seq) + pd->p_len;
2547                 int              len = 0;
2548 #ifdef INET
2549                 struct ip       *h4;
2550 #endif
2551 #ifdef INET6
2552                 struct ip6_hdr  *h6;
2553 #endif
2554
2555                 switch (af) {
2556 #ifdef INET
2557                 case AF_INET:
2558                         h4 = mtod(m, struct ip *);
2559                         len = ntohs(h4->ip_len) - off;
2560                         break;
2561 #endif
2562 #ifdef INET6
2563                 case AF_INET6:
2564                         h6 = mtod(m, struct ip6_hdr *);
2565                         len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
2566                         break;
2567 #endif
2568                 }
2569
2570                 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
2571                         REASON_SET(reason, PFRES_PROTCKSUM);
2572                 else {
2573                         if (th->th_flags & TH_SYN)
2574                                 ack++;
2575                         if (th->th_flags & TH_FIN)
2576                                 ack++;
2577                         pf_send_tcp(m, r, af, pd->dst,
2578                                 pd->src, th->th_dport, th->th_sport,
2579                                 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
2580                                 r->return_ttl, 1, 0, kif->pfik_ifp);
2581                 }
2582         } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
2583                 r->return_icmp)
2584                 pf_send_icmp(m, r->return_icmp >> 8,
2585                         r->return_icmp & 255, af, r);
2586         else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
2587                 r->return_icmp6)
2588                 pf_send_icmp(m, r->return_icmp6 >> 8,
2589                         r->return_icmp6 & 255, af, r);
2590 }
2591
2592 static int
2593 pf_ieee8021q_setpcp(struct mbuf *m, u_int8_t prio)
2594 {
2595         struct m_tag *mtag;
2596
2597         KASSERT(prio <= PF_PRIO_MAX,
2598             ("%s with invalid pcp", __func__));
2599
2600         mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_OUT, NULL);
2601         if (mtag == NULL) {
2602                 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_OUT,
2603                     sizeof(uint8_t), M_NOWAIT);
2604                 if (mtag == NULL)
2605                         return (ENOMEM);
2606                 m_tag_prepend(m, mtag);
2607         }
2608
2609         *(uint8_t *)(mtag + 1) = prio;
2610         return (0);
2611 }
2612
2613 static int
2614 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
2615 {
2616         struct m_tag *mtag;
2617         u_int8_t mpcp;
2618
2619         mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
2620         if (mtag == NULL)
2621                 return (0);
2622
2623         if (prio == PF_PRIO_ZERO)
2624                 prio = 0;
2625
2626         mpcp = *(uint8_t *)(mtag + 1);
2627
2628         return (mpcp == prio);
2629 }
2630
2631 static void
2632 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2633     struct pf_rule *r)
2634 {
2635         struct pf_send_entry *pfse;
2636         struct mbuf *m0;
2637         struct pf_mtag *pf_mtag;
2638
2639         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2640         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2641         if (pfse == NULL)
2642                 return;
2643
2644         if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2645                 free(pfse, M_PFTEMP);
2646                 return;
2647         }
2648
2649         if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2650                 free(pfse, M_PFTEMP);
2651                 return;
2652         }
2653         /* XXX: revisit */
2654         m0->m_flags |= M_SKIP_FIREWALL;
2655
2656         if (r->rtableid >= 0)
2657                 M_SETFIB(m0, r->rtableid);
2658
2659 #ifdef ALTQ
2660         if (r->qid) {
2661                 pf_mtag->qid = r->qid;
2662                 /* add hints for ecn */
2663                 pf_mtag->hdr = mtod(m0, struct ip *);
2664         }
2665 #endif /* ALTQ */
2666
2667         switch (af) {
2668 #ifdef INET
2669         case AF_INET:
2670                 pfse->pfse_type = PFSE_ICMP;
2671                 break;
2672 #endif /* INET */
2673 #ifdef INET6
2674         case AF_INET6:
2675                 pfse->pfse_type = PFSE_ICMP6;
2676                 break;
2677 #endif /* INET6 */
2678         }
2679         pfse->pfse_m = m0;
2680         pfse->icmpopts.type = type;
2681         pfse->icmpopts.code = code;
2682         pf_send(pfse);
2683 }
2684
2685 /*
2686  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2687  * If n is 0, they match if they are equal. If n is != 0, they match if they
2688  * are different.
2689  */
2690 int
2691 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2692     struct pf_addr *b, sa_family_t af)
2693 {
2694         int     match = 0;
2695
2696         switch (af) {
2697 #ifdef INET
2698         case AF_INET:
2699                 if ((a->addr32[0] & m->addr32[0]) ==
2700                     (b->addr32[0] & m->addr32[0]))
2701                         match++;
2702                 break;
2703 #endif /* INET */
2704 #ifdef INET6
2705         case AF_INET6:
2706                 if (((a->addr32[0] & m->addr32[0]) ==
2707                      (b->addr32[0] & m->addr32[0])) &&
2708                     ((a->addr32[1] & m->addr32[1]) ==
2709                      (b->addr32[1] & m->addr32[1])) &&
2710                     ((a->addr32[2] & m->addr32[2]) ==
2711                      (b->addr32[2] & m->addr32[2])) &&
2712                     ((a->addr32[3] & m->addr32[3]) ==
2713                      (b->addr32[3] & m->addr32[3])))
2714                         match++;
2715                 break;
2716 #endif /* INET6 */
2717         }
2718         if (match) {
2719                 if (n)
2720                         return (0);
2721                 else
2722                         return (1);
2723         } else {
2724                 if (n)
2725                         return (1);
2726                 else
2727                         return (0);
2728         }
2729 }
2730
2731 /*
2732  * Return 1 if b <= a <= e, otherwise return 0.
2733  */
2734 int
2735 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2736     struct pf_addr *a, sa_family_t af)
2737 {
2738         switch (af) {
2739 #ifdef INET
2740         case AF_INET:
2741                 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
2742                     (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
2743                         return (0);
2744                 break;
2745 #endif /* INET */
2746 #ifdef INET6
2747         case AF_INET6: {
2748                 int     i;
2749
2750                 /* check a >= b */
2751                 for (i = 0; i < 4; ++i)
2752                         if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
2753                                 break;
2754                         else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
2755                                 return (0);
2756                 /* check a <= e */
2757                 for (i = 0; i < 4; ++i)
2758                         if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
2759                                 break;
2760                         else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
2761                                 return (0);
2762                 break;
2763         }
2764 #endif /* INET6 */
2765         }
2766         return (1);
2767 }
2768
2769 static int
2770 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2771 {
2772         switch (op) {
2773         case PF_OP_IRG:
2774                 return ((p > a1) && (p < a2));
2775         case PF_OP_XRG:
2776                 return ((p < a1) || (p > a2));
2777         case PF_OP_RRG:
2778                 return ((p >= a1) && (p <= a2));
2779         case PF_OP_EQ:
2780                 return (p == a1);
2781         case PF_OP_NE:
2782                 return (p != a1);
2783         case PF_OP_LT:
2784                 return (p < a1);
2785         case PF_OP_LE:
2786                 return (p <= a1);
2787         case PF_OP_GT:
2788                 return (p > a1);
2789         case PF_OP_GE:
2790                 return (p >= a1);
2791         }
2792         return (0); /* never reached */
2793 }
2794
2795 int
2796 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2797 {
2798         NTOHS(a1);
2799         NTOHS(a2);
2800         NTOHS(p);
2801         return (pf_match(op, a1, a2, p));
2802 }
2803
2804 static int
2805 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2806 {
2807         if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2808                 return (0);
2809         return (pf_match(op, a1, a2, u));
2810 }
2811
2812 static int
2813 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2814 {
2815         if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2816                 return (0);
2817         return (pf_match(op, a1, a2, g));
2818 }
2819
2820 int
2821 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2822 {
2823         if (*tag == -1)
2824                 *tag = mtag;
2825
2826         return ((!r->match_tag_not && r->match_tag == *tag) ||
2827             (r->match_tag_not && r->match_tag != *tag));
2828 }
2829
2830 int
2831 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2832 {
2833
2834         KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2835
2836         if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2837                 return (ENOMEM);
2838
2839         pd->pf_mtag->tag = tag;
2840
2841         return (0);
2842 }
2843
2844 #define PF_ANCHOR_STACKSIZE     32
2845 struct pf_anchor_stackframe {
2846         struct pf_ruleset       *rs;
2847         struct pf_rule          *r;     /* XXX: + match bit */
2848         struct pf_anchor        *child;
2849 };
2850
2851 /*
2852  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2853  */
2854 #define PF_ANCHORSTACK_MATCH    0x00000001
2855 #define PF_ANCHORSTACK_MASK     (PF_ANCHORSTACK_MATCH)
2856
2857 #define PF_ANCHOR_MATCH(f)      ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2858 #define PF_ANCHOR_RULE(f)       (struct pf_rule *)                      \
2859                                 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2860 #define PF_ANCHOR_SET_MATCH(f)  do { (f)->r = (void *)                  \
2861                                 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2862 } while (0)
2863
2864 void
2865 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2866     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2867     int *match)
2868 {
2869         struct pf_anchor_stackframe     *f;
2870
2871         PF_RULES_RASSERT();
2872
2873         if (match)
2874                 *match = 0;
2875         if (*depth >= PF_ANCHOR_STACKSIZE) {
2876                 printf("%s: anchor stack overflow on %s\n",
2877                     __func__, (*r)->anchor->name);
2878                 *r = TAILQ_NEXT(*r, entries);
2879                 return;
2880         } else if (*depth == 0 && a != NULL)
2881                 *a = *r;
2882         f = stack + (*depth)++;
2883         f->rs = *rs;
2884         f->r = *r;
2885         if ((*r)->anchor_wildcard) {
2886                 struct pf_anchor_node *parent = &(*r)->anchor->children;
2887
2888                 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2889                         *r = NULL;
2890                         return;
2891                 }
2892                 *rs = &f->child->ruleset;
2893         } else {
2894                 f->child = NULL;
2895                 *rs = &(*r)->anchor->ruleset;
2896         }
2897         *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2898 }
2899
2900 int
2901 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2902     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2903     int *match)
2904 {
2905         struct pf_anchor_stackframe     *f;
2906         struct pf_rule *fr;
2907         int quick = 0;
2908
2909         PF_RULES_RASSERT();
2910
2911         do {
2912                 if (*depth <= 0)
2913                         break;
2914                 f = stack + *depth - 1;
2915                 fr = PF_ANCHOR_RULE(f);
2916                 if (f->child != NULL) {
2917                         struct pf_anchor_node *parent;
2918
2919                         /*
2920                          * This block traverses through
2921                          * a wildcard anchor.
2922                          */
2923                         parent = &fr->anchor->children;
2924                         if (match != NULL && *match) {
2925                                 /*
2926                                  * If any of "*" matched, then
2927                                  * "foo/ *" matched, mark frame
2928                                  * appropriately.
2929                                  */
2930                                 PF_ANCHOR_SET_MATCH(f);
2931                                 *match = 0;
2932                         }
2933                         f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2934                         if (f->child != NULL) {
2935                                 *rs = &f->child->ruleset;
2936                                 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2937                                 if (*r == NULL)
2938                                         continue;
2939                                 else
2940                                         break;
2941                         }
2942                 }
2943                 (*depth)--;
2944                 if (*depth == 0 && a != NULL)
2945                         *a = NULL;
2946                 *rs = f->rs;
2947                 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2948                         quick = fr->quick;
2949                 *r = TAILQ_NEXT(fr, entries);
2950         } while (*r == NULL);
2951
2952         return (quick);
2953 }
2954
2955 #ifdef INET6
2956 void
2957 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2958     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2959 {
2960         switch (af) {
2961 #ifdef INET
2962         case AF_INET:
2963                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2964                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2965                 break;
2966 #endif /* INET */
2967         case AF_INET6:
2968                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2969                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2970                 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2971                 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2972                 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2973                 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2974                 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2975                 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2976                 break;
2977         }
2978 }
2979
2980 void
2981 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2982 {
2983         switch (af) {
2984 #ifdef INET
2985         case AF_INET:
2986                 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2987                 break;
2988 #endif /* INET */
2989         case AF_INET6:
2990                 if (addr->addr32[3] == 0xffffffff) {
2991                         addr->addr32[3] = 0;
2992                         if (addr->addr32[2] == 0xffffffff) {
2993                                 addr->addr32[2] = 0;
2994                                 if (addr->addr32[1] == 0xffffffff) {
2995                                         addr->addr32[1] = 0;
2996                                         addr->addr32[0] =
2997                                             htonl(ntohl(addr->addr32[0]) + 1);
2998                                 } else
2999                                         addr->addr32[1] =
3000                                             htonl(ntohl(addr->addr32[1]) + 1);
3001                         } else
3002                                 addr->addr32[2] =
3003                                     htonl(ntohl(addr->addr32[2]) + 1);
3004                 } else
3005                         addr->addr32[3] =
3006                             htonl(ntohl(addr->addr32[3]) + 1);
3007                 break;
3008         }
3009 }
3010 #endif /* INET6 */
3011
3012 int
3013 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
3014 {
3015         struct pf_addr          *saddr, *daddr;
3016         u_int16_t                sport, dport;
3017         struct inpcbinfo        *pi;
3018         struct inpcb            *inp;
3019
3020         pd->lookup.uid = UID_MAX;
3021         pd->lookup.gid = GID_MAX;
3022
3023         switch (pd->proto) {
3024         case IPPROTO_TCP:
3025                 if (pd->hdr.tcp == NULL)
3026                         return (-1);
3027                 sport = pd->hdr.tcp->th_sport;
3028                 dport = pd->hdr.tcp->th_dport;
3029                 pi = &V_tcbinfo;
3030                 break;
3031         case IPPROTO_UDP:
3032                 if (pd->hdr.udp == NULL)
3033                         return (-1);
3034                 sport = pd->hdr.udp->uh_sport;
3035                 dport = pd->hdr.udp->uh_dport;
3036                 pi = &V_udbinfo;
3037                 break;
3038         default:
3039                 return (-1);
3040         }
3041         if (direction == PF_IN) {
3042                 saddr = pd->src;
3043                 daddr = pd->dst;
3044         } else {
3045                 u_int16_t       p;
3046
3047                 p = sport;
3048                 sport = dport;
3049                 dport = p;
3050                 saddr = pd->dst;
3051                 daddr = pd->src;
3052         }
3053         switch (pd->af) {
3054 #ifdef INET
3055         case AF_INET:
3056                 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
3057                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
3058                 if (inp == NULL) {
3059                         inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
3060                            daddr->v4, dport, INPLOOKUP_WILDCARD |
3061                            INPLOOKUP_RLOCKPCB, NULL, m);
3062                         if (inp == NULL)
3063                                 return (-1);
3064                 }
3065                 break;
3066 #endif /* INET */
3067 #ifdef INET6
3068         case AF_INET6:
3069                 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
3070                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
3071                 if (inp == NULL) {
3072                         inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
3073                             &daddr->v6, dport, INPLOOKUP_WILDCARD |
3074                             INPLOOKUP_RLOCKPCB, NULL, m);
3075                         if (inp == NULL)
3076                                 return (-1);
3077                 }
3078                 break;
3079 #endif /* INET6 */
3080
3081         default:
3082                 return (-1);
3083         }
3084         INP_RLOCK_ASSERT(inp);
3085         pd->lookup.uid = inp->inp_cred->cr_uid;
3086         pd->lookup.gid = inp->inp_cred->cr_groups[0];
3087         INP_RUNLOCK(inp);
3088
3089         return (1);
3090 }
3091
3092 static u_int8_t
3093 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3094 {
3095         int              hlen;
3096         u_int8_t         hdr[60];
3097         u_int8_t        *opt, optlen;
3098         u_int8_t         wscale = 0;
3099
3100         hlen = th_off << 2;             /* hlen <= sizeof(hdr) */
3101         if (hlen <= sizeof(struct tcphdr))
3102                 return (0);
3103         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3104                 return (0);
3105         opt = hdr + sizeof(struct tcphdr);
3106         hlen -= sizeof(struct tcphdr);
3107         while (hlen >= 3) {
3108                 switch (*opt) {
3109                 case TCPOPT_EOL:
3110                 case TCPOPT_NOP:
3111                         ++opt;
3112                         --hlen;
3113                         break;
3114                 case TCPOPT_WINDOW:
3115                         wscale = opt[2];
3116                         if (wscale > TCP_MAX_WINSHIFT)
3117                                 wscale = TCP_MAX_WINSHIFT;
3118                         wscale |= PF_WSCALE_FLAG;
3119                         /* FALLTHROUGH */
3120                 default:
3121                         optlen = opt[1];
3122                         if (optlen < 2)
3123                                 optlen = 2;
3124                         hlen -= optlen;
3125                         opt += optlen;
3126                         break;
3127                 }
3128         }
3129         return (wscale);
3130 }
3131
3132 static u_int16_t
3133 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3134 {
3135         int              hlen;
3136         u_int8_t         hdr[60];
3137         u_int8_t        *opt, optlen;
3138         u_int16_t        mss = V_tcp_mssdflt;
3139
3140         hlen = th_off << 2;     /* hlen <= sizeof(hdr) */
3141         if (hlen <= sizeof(struct tcphdr))
3142                 return (0);
3143         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3144                 return (0);
3145         opt = hdr + sizeof(struct tcphdr);
3146         hlen -= sizeof(struct tcphdr);
3147         while (hlen >= TCPOLEN_MAXSEG) {
3148                 switch (*opt) {
3149                 case TCPOPT_EOL:
3150                 case TCPOPT_NOP:
3151                         ++opt;
3152                         --hlen;
3153                         break;
3154                 case TCPOPT_MAXSEG:
3155                         bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
3156                         NTOHS(mss);
3157                         /* FALLTHROUGH */
3158                 default:
3159                         optlen = opt[1];
3160                         if (optlen < 2)
3161                                 optlen = 2;
3162                         hlen -= optlen;
3163                         opt += optlen;
3164                         break;
3165                 }
3166         }
3167         return (mss);
3168 }
3169
3170 static u_int16_t
3171 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
3172 {
3173         struct nhop_object *nh;
3174 #ifdef INET6
3175         struct in6_addr         dst6;
3176         uint32_t                scopeid;
3177 #endif /* INET6 */
3178         int                      hlen = 0;
3179         uint16_t                 mss = 0;
3180
3181         NET_EPOCH_ASSERT();
3182
3183         switch (af) {
3184 #ifdef INET
3185         case AF_INET:
3186                 hlen = sizeof(struct ip);
3187                 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0);
3188                 if (nh != NULL)
3189                         mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
3190                 break;
3191 #endif /* INET */
3192 #ifdef INET6
3193         case AF_INET6:
3194                 hlen = sizeof(struct ip6_hdr);
3195                 in6_splitscope(&addr->v6, &dst6, &scopeid);
3196                 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0);
3197                 if (nh != NULL)
3198                         mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
3199                 break;
3200 #endif /* INET6 */
3201         }
3202
3203         mss = max(V_tcp_mssdflt, mss);
3204         mss = min(mss, offer);
3205         mss = max(mss, 64);             /* sanity - at least max opt space */
3206         return (mss);
3207 }
3208
3209 static u_int32_t
3210 pf_tcp_iss(struct pf_pdesc *pd)
3211 {
3212         MD5_CTX ctx;
3213         u_int32_t digest[4];
3214
3215         if (V_pf_tcp_secret_init == 0) {
3216                 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
3217                 MD5Init(&V_pf_tcp_secret_ctx);
3218                 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
3219                     sizeof(V_pf_tcp_secret));
3220                 V_pf_tcp_secret_init = 1;
3221         }
3222
3223         ctx = V_pf_tcp_secret_ctx;
3224
3225         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
3226         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
3227         if (pd->af == AF_INET6) {
3228                 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3229                 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3230         } else {
3231                 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3232                 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3233         }
3234         MD5Final((u_char *)digest, &ctx);
3235         V_pf_tcp_iss_off += 4096;
3236 #define ISN_RANDOM_INCREMENT (4096 - 1)
3237         return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
3238             V_pf_tcp_iss_off);
3239 #undef  ISN_RANDOM_INCREMENT
3240 }
3241
3242 static int
3243 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
3244     struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
3245     struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
3246 {
3247         struct pf_rule          *nr = NULL;
3248         struct pf_addr          * const saddr = pd->src;
3249         struct pf_addr          * const daddr = pd->dst;
3250         sa_family_t              af = pd->af;
3251         struct pf_rule          *r, *a = NULL;
3252         struct pf_ruleset       *ruleset = NULL;
3253         struct pf_src_node      *nsn = NULL;
3254         struct tcphdr           *th = pd->hdr.tcp;
3255         struct pf_state_key     *sk = NULL, *nk = NULL;
3256         u_short                  reason;
3257         int                      rewrite = 0, hdrlen = 0;
3258         int                      tag = -1, rtableid = -1;
3259         int                      asd = 0;
3260         int                      match = 0;
3261         int                      state_icmp = 0;
3262         u_int16_t                sport = 0, dport = 0;
3263         u_int16_t                bproto_sum = 0, bip_sum = 0;
3264         u_int8_t                 icmptype = 0, icmpcode = 0;
3265         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3266
3267         PF_RULES_RASSERT();
3268
3269         if (inp != NULL) {
3270                 INP_LOCK_ASSERT(inp);
3271                 pd->lookup.uid = inp->inp_cred->cr_uid;
3272                 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3273                 pd->lookup.done = 1;
3274         }
3275
3276         switch (pd->proto) {
3277         case IPPROTO_TCP:
3278                 sport = th->th_sport;
3279                 dport = th->th_dport;
3280                 hdrlen = sizeof(*th);
3281                 break;
3282         case IPPROTO_UDP:
3283                 sport = pd->hdr.udp->uh_sport;
3284                 dport = pd->hdr.udp->uh_dport;
3285                 hdrlen = sizeof(*pd->hdr.udp);
3286                 break;
3287 #ifdef INET
3288         case IPPROTO_ICMP:
3289                 if (pd->af != AF_INET)
3290                         break;
3291                 sport = dport = pd->hdr.icmp->icmp_id;
3292                 hdrlen = sizeof(*pd->hdr.icmp);
3293                 icmptype = pd->hdr.icmp->icmp_type;
3294                 icmpcode = pd->hdr.icmp->icmp_code;
3295
3296                 if (icmptype == ICMP_UNREACH ||
3297                     icmptype == ICMP_SOURCEQUENCH ||
3298                     icmptype == ICMP_REDIRECT ||
3299                     icmptype == ICMP_TIMXCEED ||
3300                     icmptype == ICMP_PARAMPROB)
3301                         state_icmp++;
3302                 break;
3303 #endif /* INET */
3304 #ifdef INET6
3305         case IPPROTO_ICMPV6:
3306                 if (af != AF_INET6)
3307                         break;
3308                 sport = dport = pd->hdr.icmp6->icmp6_id;
3309                 hdrlen = sizeof(*pd->hdr.icmp6);
3310                 icmptype = pd->hdr.icmp6->icmp6_type;
3311                 icmpcode = pd->hdr.icmp6->icmp6_code;
3312
3313                 if (icmptype == ICMP6_DST_UNREACH ||
3314                     icmptype == ICMP6_PACKET_TOO_BIG ||
3315                     icmptype == ICMP6_TIME_EXCEEDED ||
3316                     icmptype == ICMP6_PARAM_PROB)
3317                         state_icmp++;
3318                 break;
3319 #endif /* INET6 */
3320         default:
3321                 sport = dport = hdrlen = 0;
3322                 break;
3323         }
3324
3325         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3326
3327         /* check packet for BINAT/NAT/RDR */
3328         if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3329             &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3330                 KASSERT(sk != NULL, ("%s: null sk", __func__));
3331                 KASSERT(nk != NULL, ("%s: null nk", __func__));
3332
3333                 if (pd->ip_sum)
3334                         bip_sum = *pd->ip_sum;
3335
3336                 switch (pd->proto) {
3337                 case IPPROTO_TCP:
3338                         bproto_sum = th->th_sum;
3339                         pd->proto_sum = &th->th_sum;
3340
3341                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3342                             nk->port[pd->sidx] != sport) {
3343                                 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
3344                                     &th->th_sum, &nk->addr[pd->sidx],
3345                                     nk->port[pd->sidx], 0, af);
3346                                 pd->sport = &th->th_sport;
3347                                 sport = th->th_sport;
3348                         }
3349
3350                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3351                             nk->port[pd->didx] != dport) {
3352                                 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
3353                                     &th->th_sum, &nk->addr[pd->didx],
3354                                     nk->port[pd->didx], 0, af);
3355                                 dport = th->th_dport;
3356                                 pd->dport = &th->th_dport;
3357                         }
3358                         rewrite++;
3359                         break;
3360                 case IPPROTO_UDP:
3361                         bproto_sum = pd->hdr.udp->uh_sum;
3362                         pd->proto_sum = &pd->hdr.udp->uh_sum;
3363
3364                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3365                             nk->port[pd->sidx] != sport) {
3366                                 pf_change_ap(m, saddr, &pd->hdr.udp->uh_sport,
3367                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3368                                     &nk->addr[pd->sidx],
3369                                     nk->port[pd->sidx], 1, af);
3370                                 sport = pd->hdr.udp->uh_sport;
3371                                 pd->sport = &pd->hdr.udp->uh_sport;
3372                         }
3373
3374                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3375                             nk->port[pd->didx] != dport) {
3376                                 pf_change_ap(m, daddr, &pd->hdr.udp->uh_dport,
3377                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3378                                     &nk->addr[pd->didx],
3379                                     nk->port[pd->didx], 1, af);
3380                                 dport = pd->hdr.udp->uh_dport;
3381                                 pd->dport = &pd->hdr.udp->uh_dport;
3382                         }
3383                         rewrite++;
3384                         break;
3385 #ifdef INET
3386                 case IPPROTO_ICMP:
3387                         nk->port[0] = nk->port[1];
3388                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3389                                 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3390                                     nk->addr[pd->sidx].v4.s_addr, 0);
3391
3392                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3393                                 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3394                                     nk->addr[pd->didx].v4.s_addr, 0);
3395
3396                         if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3397                                 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3398                                     pd->hdr.icmp->icmp_cksum, sport,
3399                                     nk->port[1], 0);
3400                                 pd->hdr.icmp->icmp_id = nk->port[1];
3401                                 pd->sport = &pd->hdr.icmp->icmp_id;
3402                         }
3403                         m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3404                         break;
3405 #endif /* INET */
3406 #ifdef INET6
3407                 case IPPROTO_ICMPV6:
3408                         nk->port[0] = nk->port[1];
3409                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3410                                 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3411                                     &nk->addr[pd->sidx], 0);
3412
3413                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3414                                 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3415                                     &nk->addr[pd->didx], 0);
3416                         rewrite++;
3417                         break;
3418 #endif /* INET */
3419                 default:
3420                         switch (af) {
3421 #ifdef INET
3422                         case AF_INET:
3423                                 if (PF_ANEQ(saddr,
3424                                     &nk->addr[pd->sidx], AF_INET))
3425                                         pf_change_a(&saddr->v4.s_addr,
3426                                             pd->ip_sum,
3427                                             nk->addr[pd->sidx].v4.s_addr, 0);
3428
3429                                 if (PF_ANEQ(daddr,
3430                                     &nk->addr[pd->didx], AF_INET))
3431                                         pf_change_a(&daddr->v4.s_addr,
3432                                             pd->ip_sum,
3433                                             nk->addr[pd->didx].v4.s_addr, 0);
3434                                 break;
3435 #endif /* INET */
3436 #ifdef INET6
3437                         case AF_INET6:
3438                                 if (PF_ANEQ(saddr,
3439                                     &nk->addr[pd->sidx], AF_INET6))
3440                                         PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3441
3442                                 if (PF_ANEQ(daddr,
3443                                     &nk->addr[pd->didx], AF_INET6))
3444                                         PF_ACPY(daddr, &nk->addr[pd->didx], af);
3445                                 break;
3446 #endif /* INET */
3447                         }
3448                         break;
3449                 }
3450                 if (nr->natpass)
3451                         r = NULL;
3452                 pd->nat_rule = nr;
3453         }
3454
3455         while (r != NULL) {
3456                 r->evaluations++;
3457                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3458                         r = r->skip[PF_SKIP_IFP].ptr;
3459                 else if (r->direction && r->direction != direction)
3460                         r = r->skip[PF_SKIP_DIR].ptr;
3461                 else if (r->af && r->af != af)
3462                         r = r->skip[PF_SKIP_AF].ptr;
3463                 else if (r->proto && r->proto != pd->proto)
3464                         r = r->skip[PF_SKIP_PROTO].ptr;
3465                 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3466                     r->src.neg, kif, M_GETFIB(m)))
3467                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3468                 /* tcp/udp only. port_op always 0 in other cases */
3469                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3470                     r->src.port[0], r->src.port[1], sport))
3471                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
3472                 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3473                     r->dst.neg, NULL, M_GETFIB(m)))
3474                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3475                 /* tcp/udp only. port_op always 0 in other cases */
3476                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3477                     r->dst.port[0], r->dst.port[1], dport))
3478                         r = r->skip[PF_SKIP_DST_PORT].ptr;
3479                 /* icmp only. type always 0 in other cases */
3480                 else if (r->type && r->type != icmptype + 1)
3481                         r = TAILQ_NEXT(r, entries);
3482                 /* icmp only. type always 0 in other cases */
3483                 else if (r->code && r->code != icmpcode + 1)
3484                         r = TAILQ_NEXT(r, entries);
3485                 else if (r->tos && !(r->tos == pd->tos))
3486                         r = TAILQ_NEXT(r, entries);
3487                 else if (r->rule_flag & PFRULE_FRAGMENT)
3488                         r = TAILQ_NEXT(r, entries);
3489                 else if (pd->proto == IPPROTO_TCP &&
3490                     (r->flagset & th->th_flags) != r->flags)
3491                         r = TAILQ_NEXT(r, entries);
3492                 /* tcp/udp only. uid.op always 0 in other cases */
3493                 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3494                     pf_socket_lookup(direction, pd, m), 1)) &&
3495                     !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3496                     pd->lookup.uid))
3497                         r = TAILQ_NEXT(r, entries);
3498                 /* tcp/udp only. gid.op always 0 in other cases */
3499                 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3500                     pf_socket_lookup(direction, pd, m), 1)) &&
3501                     !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3502                     pd->lookup.gid))
3503                         r = TAILQ_NEXT(r, entries);
3504                 else if (r->prio &&
3505                     !pf_match_ieee8021q_pcp(r->prio, m))
3506                         r = TAILQ_NEXT(r, entries);
3507                 else if (r->prob &&
3508                     r->prob <= arc4random())
3509                         r = TAILQ_NEXT(r, entries);
3510                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3511                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3512                         r = TAILQ_NEXT(r, entries);
3513                 else if (r->os_fingerprint != PF_OSFP_ANY &&
3514                     (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3515                     pf_osfp_fingerprint(pd, m, off, th),
3516                     r->os_fingerprint)))
3517                         r = TAILQ_NEXT(r, entries);
3518                 else {
3519                         if (r->tag)
3520                                 tag = r->tag;
3521                         if (r->rtableid >= 0)
3522                                 rtableid = r->rtableid;
3523                         if (r->anchor == NULL) {
3524                                 match = 1;
3525                                 *rm = r;
3526                                 *am = a;
3527                                 *rsm = ruleset;
3528                                 if ((*rm)->quick)
3529                                         break;
3530                                 r = TAILQ_NEXT(r, entries);
3531                         } else
3532                                 pf_step_into_anchor(anchor_stack, &asd,
3533                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3534                                     &match);
3535                 }
3536                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3537                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3538                         break;
3539         }
3540         r = *rm;
3541         a = *am;
3542         ruleset = *rsm;
3543
3544         REASON_SET(&reason, PFRES_MATCH);
3545
3546         if (r->log || (nr != NULL && nr->log)) {
3547                 if (rewrite)
3548                         m_copyback(m, off, hdrlen, pd->hdr.any);
3549                 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3550                     ruleset, pd, 1);
3551         }
3552
3553         if ((r->action == PF_DROP) &&
3554             ((r->rule_flag & PFRULE_RETURNRST) ||
3555             (r->rule_flag & PFRULE_RETURNICMP) ||
3556             (r->rule_flag & PFRULE_RETURN))) {
3557                 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum,
3558                     bip_sum, hdrlen, &reason);
3559         }
3560
3561         if (r->action == PF_DROP)
3562                 goto cleanup;
3563
3564         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3565                 REASON_SET(&reason, PFRES_MEMORY);
3566                 goto cleanup;
3567         }
3568         if (rtableid >= 0)
3569                 M_SETFIB(m, rtableid);
3570
3571         if (!state_icmp && (r->keep_state || nr != NULL ||
3572             (pd->flags & PFDESC_TCP_NORM))) {
3573                 int action;
3574                 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3575                     sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3576                     hdrlen);
3577                 if (action != PF_PASS) {
3578                         if (action == PF_DROP &&
3579                             (r->rule_flag & PFRULE_RETURN))
3580                                 pf_return(r, nr, pd, sk, off, m, th, kif,
3581                                     bproto_sum, bip_sum, hdrlen, &reason);
3582                         return (action);
3583                 }
3584         } else {
3585                 if (sk != NULL)
3586                         uma_zfree(V_pf_state_key_z, sk);
3587                 if (nk != NULL)
3588                         uma_zfree(V_pf_state_key_z, nk);
3589         }
3590
3591         /* copy back packet headers if we performed NAT operations */
3592         if (rewrite)
3593                 m_copyback(m, off, hdrlen, pd->hdr.any);
3594
3595         if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3596             direction == PF_OUT &&
3597             V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m))
3598                 /*
3599                  * We want the state created, but we dont
3600                  * want to send this in case a partner
3601                  * firewall has to know about it to allow
3602                  * replies through it.
3603                  */
3604                 return (PF_DEFER);
3605
3606         return (PF_PASS);
3607
3608 cleanup:
3609         if (sk != NULL)
3610                 uma_zfree(V_pf_state_key_z, sk);
3611         if (nk != NULL)
3612                 uma_zfree(V_pf_state_key_z, nk);
3613         return (PF_DROP);
3614 }
3615
3616 static int
3617 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3618     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3619     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3620     u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3621     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3622 {
3623         struct pf_state         *s = NULL;
3624         struct pf_src_node      *sn = NULL;
3625         struct tcphdr           *th = pd->hdr.tcp;
3626         u_int16_t                mss = V_tcp_mssdflt;
3627         u_short                  reason;
3628
3629         /* check maximums */
3630         if (r->max_states &&
3631             (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3632                 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
3633                 REASON_SET(&reason, PFRES_MAXSTATES);
3634                 goto csfailed;
3635         }
3636         /* src node for filter rule */
3637         if ((r->rule_flag & PFRULE_SRCTRACK ||
3638             r->rpool.opts & PF_POOL_STICKYADDR) &&
3639             pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3640                 REASON_SET(&reason, PFRES_SRCLIMIT);
3641                 goto csfailed;
3642         }
3643         /* src node for translation rule */
3644         if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3645             pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3646                 REASON_SET(&reason, PFRES_SRCLIMIT);
3647                 goto csfailed;
3648         }
3649         s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3650         if (s == NULL) {
3651                 REASON_SET(&reason, PFRES_MEMORY);
3652                 goto csfailed;
3653         }
3654         s->rule.ptr = r;
3655         s->nat_rule.ptr = nr;
3656         s->anchor.ptr = a;
3657         STATE_INC_COUNTERS(s);
3658         if (r->allow_opts)
3659                 s->state_flags |= PFSTATE_ALLOWOPTS;
3660         if (r->rule_flag & PFRULE_STATESLOPPY)
3661                 s->state_flags |= PFSTATE_SLOPPY;
3662         s->log = r->log & PF_LOG_ALL;
3663         s->sync_state = PFSYNC_S_NONE;
3664         if (nr != NULL)
3665                 s->log |= nr->log & PF_LOG_ALL;
3666         switch (pd->proto) {
3667         case IPPROTO_TCP:
3668                 s->src.seqlo = ntohl(th->th_seq);
3669                 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3670                 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3671                     r->keep_state == PF_STATE_MODULATE) {
3672                         /* Generate sequence number modulator */
3673                         if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3674                             0)
3675                                 s->src.seqdiff = 1;
3676                         pf_change_proto_a(m, &th->th_seq, &th->th_sum,
3677                             htonl(s->src.seqlo + s->src.seqdiff), 0);
3678                         *rewrite = 1;
3679                 } else
3680                         s->src.seqdiff = 0;
3681                 if (th->th_flags & TH_SYN) {
3682                         s->src.seqhi++;
3683                         s->src.wscale = pf_get_wscale(m, off,
3684                             th->th_off, pd->af);
3685                 }
3686                 s->src.max_win = MAX(ntohs(th->th_win), 1);
3687                 if (s->src.wscale & PF_WSCALE_MASK) {
3688                         /* Remove scale factor from initial window */
3689                         int win = s->src.max_win;
3690                         win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3691                         s->src.max_win = (win - 1) >>
3692                             (s->src.wscale & PF_WSCALE_MASK);
3693                 }
3694                 if (th->th_flags & TH_FIN)
3695                         s->src.seqhi++;
3696                 s->dst.seqhi = 1;
3697                 s->dst.max_win = 1;
3698                 s->src.state = TCPS_SYN_SENT;
3699                 s->dst.state = TCPS_CLOSED;
3700                 s->timeout = PFTM_TCP_FIRST_PACKET;
3701                 break;
3702         case IPPROTO_UDP:
3703                 s->src.state = PFUDPS_SINGLE;
3704                 s->dst.state = PFUDPS_NO_TRAFFIC;
3705                 s->timeout = PFTM_UDP_FIRST_PACKET;
3706                 break;
3707         case IPPROTO_ICMP:
3708 #ifdef INET6
3709         case IPPROTO_ICMPV6:
3710 #endif
3711                 s->timeout = PFTM_ICMP_FIRST_PACKET;
3712                 break;
3713         default:
3714                 s->src.state = PFOTHERS_SINGLE;
3715                 s->dst.state = PFOTHERS_NO_TRAFFIC;
3716                 s->timeout = PFTM_OTHER_FIRST_PACKET;
3717         }
3718
3719         if (r->rt) {
3720                 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) {
3721                         REASON_SET(&reason, PFRES_MAPFAILED);
3722                         pf_src_tree_remove_state(s);
3723                         STATE_DEC_COUNTERS(s);
3724                         uma_zfree(V_pf_state_z, s);
3725                         goto csfailed;
3726                 }
3727                 s->rt_kif = r->rpool.cur->kif;
3728         }
3729
3730         s->creation = time_uptime;
3731         s->expire = time_uptime;
3732
3733         if (sn != NULL)
3734                 s->src_node = sn;
3735         if (nsn != NULL) {
3736                 /* XXX We only modify one side for now. */
3737                 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3738                 s->nat_src_node = nsn;
3739         }
3740         if (pd->proto == IPPROTO_TCP) {
3741                 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3742                     off, pd, th, &s->src, &s->dst)) {
3743                         REASON_SET(&reason, PFRES_MEMORY);
3744                         pf_src_tree_remove_state(s);
3745                         STATE_DEC_COUNTERS(s);
3746                         uma_zfree(V_pf_state_z, s);
3747                         return (PF_DROP);
3748                 }
3749                 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3750                     pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3751                     &s->src, &s->dst, rewrite)) {
3752                         /* This really shouldn't happen!!! */
3753                         DPFPRINTF(PF_DEBUG_URGENT,
3754                             ("pf_normalize_tcp_stateful failed on first "
3755                              "pkt\n"));
3756                         pf_normalize_tcp_cleanup(s);
3757                         pf_src_tree_remove_state(s);
3758                         STATE_DEC_COUNTERS(s);
3759                         uma_zfree(V_pf_state_z, s);
3760                         return (PF_DROP);
3761                 }
3762         }
3763         s->direction = pd->dir;
3764
3765         /*
3766          * sk/nk could already been setup by pf_get_translation().
3767          */
3768         if (nr == NULL) {
3769                 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3770                     __func__, nr, sk, nk));
3771                 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3772                 if (sk == NULL)
3773                         goto csfailed;
3774                 nk = sk;
3775         } else
3776                 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3777                     __func__, nr, sk, nk));
3778
3779         /* Swap sk/nk for PF_OUT. */
3780         if (pf_state_insert(BOUND_IFACE(r, kif),
3781             (pd->dir == PF_IN) ? sk : nk,
3782             (pd->dir == PF_IN) ? nk : sk, s)) {
3783                 if (pd->proto == IPPROTO_TCP)
3784                         pf_normalize_tcp_cleanup(s);
3785                 REASON_SET(&reason, PFRES_STATEINS);
3786                 pf_src_tree_remove_state(s);
3787                 STATE_DEC_COUNTERS(s);
3788                 uma_zfree(V_pf_state_z, s);
3789                 return (PF_DROP);
3790         } else
3791                 *sm = s;
3792
3793         if (tag > 0)
3794                 s->tag = tag;
3795         if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3796             TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3797                 s->src.state = PF_TCPS_PROXY_SRC;
3798                 /* undo NAT changes, if they have taken place */
3799                 if (nr != NULL) {
3800                         struct pf_state_key *skt = s->key[PF_SK_WIRE];
3801                         if (pd->dir == PF_OUT)
3802                                 skt = s->key[PF_SK_STACK];
3803                         PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3804                         PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3805                         if (pd->sport)
3806                                 *pd->sport = skt->port[pd->sidx];
3807                         if (pd->dport)
3808                                 *pd->dport = skt->port[pd->didx];
3809                         if (pd->proto_sum)
3810                                 *pd->proto_sum = bproto_sum;
3811                         if (pd->ip_sum)
3812                                 *pd->ip_sum = bip_sum;
3813                         m_copyback(m, off, hdrlen, pd->hdr.any);
3814                 }
3815                 s->src.seqhi = htonl(arc4random());
3816                 /* Find mss option */
3817                 int rtid = M_GETFIB(m);
3818                 mss = pf_get_mss(m, off, th->th_off, pd->af);
3819                 mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3820                 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3821                 s->src.mss = mss;
3822                 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3823                     th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3824                     TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3825                 REASON_SET(&reason, PFRES_SYNPROXY);
3826                 return (PF_SYNPROXY_DROP);
3827         }
3828
3829         return (PF_PASS);
3830
3831 csfailed:
3832         if (sk != NULL)
3833                 uma_zfree(V_pf_state_key_z, sk);
3834         if (nk != NULL)
3835                 uma_zfree(V_pf_state_key_z, nk);
3836
3837         if (sn != NULL) {
3838                 struct pf_srchash *sh;
3839
3840                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
3841                 PF_HASHROW_LOCK(sh);
3842                 if (--sn->states == 0 && sn->expire == 0) {
3843                         pf_unlink_src_node(sn);
3844                         uma_zfree(V_pf_sources_z, sn);
3845                         counter_u64_add(
3846                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3847                 }
3848                 PF_HASHROW_UNLOCK(sh);
3849         }
3850
3851         if (nsn != sn && nsn != NULL) {
3852                 struct pf_srchash *sh;
3853
3854                 sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)];
3855                 PF_HASHROW_LOCK(sh);
3856                 if (--nsn->states == 0 && nsn->expire == 0) {
3857                         pf_unlink_src_node(nsn);
3858                         uma_zfree(V_pf_sources_z, nsn);
3859                         counter_u64_add(
3860                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3861                 }
3862                 PF_HASHROW_UNLOCK(sh);
3863         }
3864
3865         return (PF_DROP);
3866 }
3867
3868 static int
3869 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
3870     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
3871     struct pf_ruleset **rsm)
3872 {
3873         struct pf_rule          *r, *a = NULL;
3874         struct pf_ruleset       *ruleset = NULL;
3875         sa_family_t              af = pd->af;
3876         u_short                  reason;
3877         int                      tag = -1;
3878         int                      asd = 0;
3879         int                      match = 0;
3880         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3881
3882         PF_RULES_RASSERT();
3883
3884         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3885         while (r != NULL) {
3886                 r->evaluations++;
3887                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3888                         r = r->skip[PF_SKIP_IFP].ptr;
3889                 else if (r->direction && r->direction != direction)
3890                         r = r->skip[PF_SKIP_DIR].ptr;
3891                 else if (r->af && r->af != af)
3892                         r = r->skip[PF_SKIP_AF].ptr;
3893                 else if (r->proto && r->proto != pd->proto)
3894                         r = r->skip[PF_SKIP_PROTO].ptr;
3895                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3896                     r->src.neg, kif, M_GETFIB(m)))
3897                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3898                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3899                     r->dst.neg, NULL, M_GETFIB(m)))
3900                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3901                 else if (r->tos && !(r->tos == pd->tos))
3902                         r = TAILQ_NEXT(r, entries);
3903                 else if (r->os_fingerprint != PF_OSFP_ANY)
3904                         r = TAILQ_NEXT(r, entries);
3905                 else if (pd->proto == IPPROTO_UDP &&
3906                     (r->src.port_op || r->dst.port_op))
3907                         r = TAILQ_NEXT(r, entries);
3908                 else if (pd->proto == IPPROTO_TCP &&
3909                     (r->src.port_op || r->dst.port_op || r->flagset))
3910                         r = TAILQ_NEXT(r, entries);
3911                 else if ((pd->proto == IPPROTO_ICMP ||
3912                     pd->proto == IPPROTO_ICMPV6) &&
3913                     (r->type || r->code))
3914                         r = TAILQ_NEXT(r, entries);
3915                 else if (r->prio &&
3916                     !pf_match_ieee8021q_pcp(r->prio, m))
3917                         r = TAILQ_NEXT(r, entries);
3918                 else if (r->prob && r->prob <=
3919                     (arc4random() % (UINT_MAX - 1) + 1))
3920                         r = TAILQ_NEXT(r, entries);
3921                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3922                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3923                         r = TAILQ_NEXT(r, entries);
3924                 else {
3925                         if (r->anchor == NULL) {
3926                                 match = 1;
3927                                 *rm = r;
3928                                 *am = a;
3929                                 *rsm = ruleset;
3930                                 if ((*rm)->quick)
3931                                         break;
3932                                 r = TAILQ_NEXT(r, entries);
3933                         } else
3934                                 pf_step_into_anchor(anchor_stack, &asd,
3935                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3936                                     &match);
3937                 }
3938                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3939                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3940                         break;
3941         }
3942         r = *rm;
3943         a = *am;
3944         ruleset = *rsm;
3945
3946         REASON_SET(&reason, PFRES_MATCH);
3947
3948         if (r->log)
3949                 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
3950                     1);
3951
3952         if (r->action != PF_PASS)
3953                 return (PF_DROP);
3954
3955         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3956                 REASON_SET(&reason, PFRES_MEMORY);
3957                 return (PF_DROP);
3958         }
3959
3960         return (PF_PASS);
3961 }
3962
3963 static int
3964 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
3965         struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
3966         struct pf_pdesc *pd, u_short *reason, int *copyback)
3967 {
3968         struct tcphdr           *th = pd->hdr.tcp;
3969         u_int16_t                win = ntohs(th->th_win);
3970         u_int32_t                ack, end, seq, orig_seq;
3971         u_int8_t                 sws, dws;
3972         int                      ackskew;
3973
3974         if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
3975                 sws = src->wscale & PF_WSCALE_MASK;
3976                 dws = dst->wscale & PF_WSCALE_MASK;
3977         } else
3978                 sws = dws = 0;
3979
3980         /*
3981          * Sequence tracking algorithm from Guido van Rooij's paper:
3982          *   http://www.madison-gurkha.com/publications/tcp_filtering/
3983          *      tcp_filtering.ps
3984          */
3985
3986         orig_seq = seq = ntohl(th->th_seq);
3987         if (src->seqlo == 0) {
3988                 /* First packet from this end. Set its state */
3989
3990                 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
3991                     src->scrub == NULL) {
3992                         if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
3993                                 REASON_SET(reason, PFRES_MEMORY);
3994                                 return (PF_DROP);
3995                         }
3996                 }
3997
3998                 /* Deferred generation of sequence number modulator */
3999                 if (dst->seqdiff && !src->seqdiff) {
4000                         /* use random iss for the TCP server */
4001                         while ((src->seqdiff = arc4random() - seq) == 0)
4002                                 ;
4003                         ack = ntohl(th->th_ack) - dst->seqdiff;
4004                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
4005                             src->seqdiff), 0);
4006                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
4007                         *copyback = 1;
4008                 } else {
4009                         ack = ntohl(th->th_ack);
4010                 }
4011
4012                 end = seq + pd->p_len;
4013                 if (th->th_flags & TH_SYN) {
4014                         end++;
4015                         if (dst->wscale & PF_WSCALE_FLAG) {
4016                                 src->wscale = pf_get_wscale(m, off, th->th_off,
4017                                     pd->af);
4018                                 if (src->wscale & PF_WSCALE_FLAG) {
4019                                         /* Remove scale factor from initial
4020                                          * window */
4021                                         sws = src->wscale & PF_WSCALE_MASK;
4022                                         win = ((u_int32_t)win + (1 << sws) - 1)
4023                                             >> sws;
4024                                         dws = dst->wscale & PF_WSCALE_MASK;
4025                                 } else {
4026                                         /* fixup other window */
4027                                         dst->max_win <<= dst->wscale &
4028                                             PF_WSCALE_MASK;
4029                                         /* in case of a retrans SYN|ACK */
4030                                         dst->wscale = 0;
4031                                 }
4032                         }
4033                 }
4034                 if (th->th_flags & TH_FIN)
4035                         end++;
4036
4037                 src->seqlo = seq;
4038                 if (src->state < TCPS_SYN_SENT)
4039                         src->state = TCPS_SYN_SENT;
4040
4041                 /*
4042                  * May need to slide the window (seqhi may have been set by
4043                  * the crappy stack check or if we picked up the connection
4044                  * after establishment)
4045                  */
4046                 if (src->seqhi == 1 ||
4047                     SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
4048                         src->seqhi = end + MAX(1, dst->max_win << dws);
4049                 if (win > src->max_win)
4050                         src->max_win = win;
4051
4052         } else {
4053                 ack = ntohl(th->th_ack) - dst->seqdiff;
4054                 if (src->seqdiff) {
4055                         /* Modulate sequence numbers */
4056                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
4057                             src->seqdiff), 0);
4058                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
4059                         *copyback = 1;
4060                 }
4061                 end = seq + pd->p_len;
4062                 if (th->th_flags & TH_SYN)
4063                         end++;
4064                 if (th->th_flags & TH_FIN)
4065                         end++;
4066         }
4067
4068         if ((th->th_flags & TH_ACK) == 0) {
4069                 /* Let it pass through the ack skew check */
4070                 ack = dst->seqlo;
4071         } else if ((ack == 0 &&
4072             (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
4073             /* broken tcp stacks do not set ack */
4074             (dst->state < TCPS_SYN_SENT)) {
4075                 /*
4076                  * Many stacks (ours included) will set the ACK number in an
4077                  * FIN|ACK if the SYN times out -- no sequence to ACK.
4078                  */
4079                 ack = dst->seqlo;
4080         }
4081
4082         if (seq == end) {
4083                 /* Ease sequencing restrictions on no data packets */
4084                 seq = src->seqlo;
4085                 end = seq;
4086         }
4087
4088         ackskew = dst->seqlo - ack;
4089
4090         /*
4091          * Need to demodulate the sequence numbers in any TCP SACK options
4092          * (Selective ACK). We could optionally validate the SACK values
4093          * against the current ACK window, either forwards or backwards, but
4094          * I'm not confident that SACK has been implemented properly
4095          * everywhere. It wouldn't surprise me if several stacks accidentally
4096          * SACK too far backwards of previously ACKed data. There really aren't
4097          * any security implications of bad SACKing unless the target stack
4098          * doesn't validate the option length correctly. Someone trying to
4099          * spoof into a TCP connection won't bother blindly sending SACK
4100          * options anyway.
4101          */
4102         if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
4103                 if (pf_modulate_sack(m, off, pd, th, dst))
4104                         *copyback = 1;
4105         }
4106
4107 #define MAXACKWINDOW (0xffff + 1500)    /* 1500 is an arbitrary fudge factor */
4108         if (SEQ_GEQ(src->seqhi, end) &&
4109             /* Last octet inside other's window space */
4110             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
4111             /* Retrans: not more than one window back */
4112             (ackskew >= -MAXACKWINDOW) &&
4113             /* Acking not more than one reassembled fragment backwards */
4114             (ackskew <= (MAXACKWINDOW << sws)) &&
4115             /* Acking not more than one window forward */
4116             ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
4117             (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
4118             (pd->flags & PFDESC_IP_REAS) == 0)) {
4119             /* Require an exact/+1 sequence match on resets when possible */
4120
4121                 if (dst->scrub || src->scrub) {
4122                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4123                             *state, src, dst, copyback))
4124                                 return (PF_DROP);
4125                 }
4126
4127                 /* update max window */
4128                 if (src->max_win < win)
4129                         src->max_win = win;
4130                 /* synchronize sequencing */
4131                 if (SEQ_GT(end, src->seqlo))
4132                         src->seqlo = end;
4133                 /* slide the window of what the other end can send */
4134                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4135                         dst->seqhi = ack + MAX((win << sws), 1);
4136
4137                 /* update states */
4138                 if (th->th_flags & TH_SYN)
4139                         if (src->state < TCPS_SYN_SENT)
4140                                 src->state = TCPS_SYN_SENT;
4141                 if (th->th_flags & TH_FIN)
4142                         if (src->state < TCPS_CLOSING)
4143                                 src->state = TCPS_CLOSING;
4144                 if (th->th_flags & TH_ACK) {
4145                         if (dst->state == TCPS_SYN_SENT) {
4146                                 dst->state = TCPS_ESTABLISHED;
4147                                 if (src->state == TCPS_ESTABLISHED &&
4148                                     (*state)->src_node != NULL &&
4149                                     pf_src_connlimit(state)) {
4150                                         REASON_SET(reason, PFRES_SRCLIMIT);
4151                                         return (PF_DROP);
4152                                 }
4153                         } else if (dst->state == TCPS_CLOSING)
4154                                 dst->state = TCPS_FIN_WAIT_2;
4155                 }
4156                 if (th->th_flags & TH_RST)
4157                         src->state = dst->state = TCPS_TIME_WAIT;
4158
4159                 /* update expire time */
4160                 (*state)->expire = time_uptime;
4161                 if (src->state >= TCPS_FIN_WAIT_2 &&
4162                     dst->state >= TCPS_FIN_WAIT_2)
4163                         (*state)->timeout = PFTM_TCP_CLOSED;
4164                 else if (src->state >= TCPS_CLOSING &&
4165                     dst->state >= TCPS_CLOSING)
4166                         (*state)->timeout = PFTM_TCP_FIN_WAIT;
4167                 else if (src->state < TCPS_ESTABLISHED ||
4168                     dst->state < TCPS_ESTABLISHED)
4169                         (*state)->timeout = PFTM_TCP_OPENING;
4170                 else if (src->state >= TCPS_CLOSING ||
4171                     dst->state >= TCPS_CLOSING)
4172                         (*state)->timeout = PFTM_TCP_CLOSING;
4173                 else
4174                         (*state)->timeout = PFTM_TCP_ESTABLISHED;
4175
4176                 /* Fall through to PASS packet */
4177
4178         } else if ((dst->state < TCPS_SYN_SENT ||
4179                 dst->state >= TCPS_FIN_WAIT_2 ||
4180                 src->state >= TCPS_FIN_WAIT_2) &&
4181             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4182             /* Within a window forward of the originating packet */
4183             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4184             /* Within a window backward of the originating packet */
4185
4186                 /*
4187                  * This currently handles three situations:
4188                  *  1) Stupid stacks will shotgun SYNs before their peer
4189                  *     replies.
4190                  *  2) When PF catches an already established stream (the
4191                  *     firewall rebooted, the state table was flushed, routes
4192                  *     changed...)
4193                  *  3) Packets get funky immediately after the connection
4194                  *     closes (this should catch Solaris spurious ACK|FINs
4195                  *     that web servers like to spew after a close)
4196                  *
4197                  * This must be a little more careful than the above code
4198                  * since packet floods will also be caught here. We don't
4199                  * update the TTL here to mitigate the damage of a packet
4200                  * flood and so the same code can handle awkward establishment
4201                  * and a loosened connection close.
4202                  * In the establishment case, a correct peer response will
4203                  * validate the connection, go through the normal state code
4204                  * and keep updating the state TTL.
4205                  */
4206
4207                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4208                         printf("pf: loose state match: ");
4209                         pf_print_state(*state);
4210                         pf_print_flags(th->th_flags);
4211                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4212                             "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
4213                             pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
4214                             (unsigned long long)(*state)->packets[1],
4215                             pd->dir == PF_IN ? "in" : "out",
4216                             pd->dir == (*state)->direction ? "fwd" : "rev");
4217                 }
4218
4219                 if (dst->scrub || src->scrub) {
4220                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4221                             *state, src, dst, copyback))
4222                                 return (PF_DROP);
4223                 }
4224
4225                 /* update max window */
4226                 if (src->max_win < win)
4227                         src->max_win = win;
4228                 /* synchronize sequencing */
4229                 if (SEQ_GT(end, src->seqlo))
4230                         src->seqlo = end;
4231                 /* slide the window of what the other end can send */
4232                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4233                         dst->seqhi = ack + MAX((win << sws), 1);
4234
4235                 /*
4236                  * Cannot set dst->seqhi here since this could be a shotgunned
4237                  * SYN and not an already established connection.
4238                  */
4239
4240                 if (th->th_flags & TH_FIN)
4241                         if (src->state < TCPS_CLOSING)
4242                                 src->state = TCPS_CLOSING;
4243                 if (th->th_flags & TH_RST)
4244                         src->state = dst->state = TCPS_TIME_WAIT;
4245
4246                 /* Fall through to PASS packet */
4247
4248         } else {
4249                 if ((*state)->dst.state == TCPS_SYN_SENT &&
4250                     (*state)->src.state == TCPS_SYN_SENT) {
4251                         /* Send RST for state mismatches during handshake */
4252                         if (!(th->th_flags & TH_RST))
4253                                 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4254                                     pd->dst, pd->src, th->th_dport,
4255                                     th->th_sport, ntohl(th->th_ack), 0,
4256                                     TH_RST, 0, 0,
4257                                     (*state)->rule.ptr->return_ttl, 1, 0,
4258                                     kif->pfik_ifp);
4259                         src->seqlo = 0;
4260                         src->seqhi = 1;
4261                         src->max_win = 1;
4262                 } else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4263                         printf("pf: BAD state: ");
4264                         pf_print_state(*state);
4265                         pf_print_flags(th->th_flags);
4266                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4267                             "pkts=%llu:%llu dir=%s,%s\n",
4268                             seq, orig_seq, ack, pd->p_len, ackskew,
4269                             (unsigned long long)(*state)->packets[0],
4270                             (unsigned long long)(*state)->packets[1],
4271                             pd->dir == PF_IN ? "in" : "out",
4272                             pd->dir == (*state)->direction ? "fwd" : "rev");
4273                         printf("pf: State failure on: %c %c %c %c | %c %c\n",
4274                             SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4275                             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4276                             ' ': '2',
4277                             (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4278                             (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4279                             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4280                             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4281                 }
4282                 REASON_SET(reason, PFRES_BADSTATE);
4283                 return (PF_DROP);
4284         }
4285
4286         return (PF_PASS);
4287 }
4288
4289 static int
4290 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4291         struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4292 {
4293         struct tcphdr           *th = pd->hdr.tcp;
4294
4295         if (th->th_flags & TH_SYN)
4296                 if (src->state < TCPS_SYN_SENT)
4297                         src->state = TCPS_SYN_SENT;
4298         if (th->th_flags & TH_FIN)
4299                 if (src->state < TCPS_CLOSING)
4300                         src->state = TCPS_CLOSING;
4301         if (th->th_flags & TH_ACK) {
4302                 if (dst->state == TCPS_SYN_SENT) {
4303                         dst->state = TCPS_ESTABLISHED;
4304                         if (src->state == TCPS_ESTABLISHED &&
4305                             (*state)->src_node != NULL &&
4306                             pf_src_connlimit(state)) {
4307                                 REASON_SET(reason, PFRES_SRCLIMIT);
4308                                 return (PF_DROP);
4309                         }
4310                 } else if (dst->state == TCPS_CLOSING) {
4311                         dst->state = TCPS_FIN_WAIT_2;
4312                 } else if (src->state == TCPS_SYN_SENT &&
4313                     dst->state < TCPS_SYN_SENT) {
4314                         /*
4315                          * Handle a special sloppy case where we only see one
4316                          * half of the connection. If there is a ACK after
4317                          * the initial SYN without ever seeing a packet from
4318                          * the destination, set the connection to established.
4319                          */
4320                         dst->state = src->state = TCPS_ESTABLISHED;
4321                         if ((*state)->src_node != NULL &&
4322                             pf_src_connlimit(state)) {
4323                                 REASON_SET(reason, PFRES_SRCLIMIT);
4324                                 return (PF_DROP);
4325                         }
4326                 } else if (src->state == TCPS_CLOSING &&
4327                     dst->state == TCPS_ESTABLISHED &&
4328                     dst->seqlo == 0) {
4329                         /*
4330                          * Handle the closing of half connections where we
4331                          * don't see the full bidirectional FIN/ACK+ACK
4332                          * handshake.
4333                          */
4334                         dst->state = TCPS_CLOSING;
4335                 }
4336         }
4337         if (th->th_flags & TH_RST)
4338                 src->state = dst->state = TCPS_TIME_WAIT;
4339
4340         /* update expire time */
4341         (*state)->expire = time_uptime;
4342         if (src->state >= TCPS_FIN_WAIT_2 &&
4343             dst->state >= TCPS_FIN_WAIT_2)
4344                 (*state)->timeout = PFTM_TCP_CLOSED;
4345         else if (src->state >= TCPS_CLOSING &&
4346             dst->state >= TCPS_CLOSING)
4347                 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4348         else if (src->state < TCPS_ESTABLISHED ||
4349             dst->state < TCPS_ESTABLISHED)
4350                 (*state)->timeout = PFTM_TCP_OPENING;
4351         else if (src->state >= TCPS_CLOSING ||
4352             dst->state >= TCPS_CLOSING)
4353                 (*state)->timeout = PFTM_TCP_CLOSING;
4354         else
4355                 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4356
4357         return (PF_PASS);
4358 }
4359
4360 static int
4361 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4362     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4363     u_short *reason)
4364 {
4365         struct pf_state_key_cmp  key;
4366         struct tcphdr           *th = pd->hdr.tcp;
4367         int                      copyback = 0;
4368         struct pf_state_peer    *src, *dst;
4369         struct pf_state_key     *sk;
4370
4371         bzero(&key, sizeof(key));
4372         key.af = pd->af;
4373         key.proto = IPPROTO_TCP;
4374         if (direction == PF_IN) {       /* wire side, straight */
4375                 PF_ACPY(&key.addr[0], pd->src, key.af);
4376                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4377                 key.port[0] = th->th_sport;
4378                 key.port[1] = th->th_dport;
4379         } else {                        /* stack side, reverse */
4380                 PF_ACPY(&key.addr[1], pd->src, key.af);
4381                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4382                 key.port[1] = th->th_sport;
4383                 key.port[0] = th->th_dport;
4384         }
4385
4386         STATE_LOOKUP(kif, &key, direction, *state, pd);
4387
4388         if (direction == (*state)->direction) {
4389                 src = &(*state)->src;
4390                 dst = &(*state)->dst;
4391         } else {
4392                 src = &(*state)->dst;
4393                 dst = &(*state)->src;
4394         }
4395
4396         sk = (*state)->key[pd->didx];
4397
4398         if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4399                 if (direction != (*state)->direction) {
4400                         REASON_SET(reason, PFRES_SYNPROXY);
4401                         return (PF_SYNPROXY_DROP);
4402                 }
4403                 if (th->th_flags & TH_SYN) {
4404                         if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4405                                 REASON_SET(reason, PFRES_SYNPROXY);
4406                                 return (PF_DROP);
4407                         }
4408                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4409                             pd->src, th->th_dport, th->th_sport,
4410                             (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4411                             TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4412                         REASON_SET(reason, PFRES_SYNPROXY);
4413                         return (PF_SYNPROXY_DROP);
4414                 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK ||
4415                     (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4416                     (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4417                         REASON_SET(reason, PFRES_SYNPROXY);
4418                         return (PF_DROP);
4419                 } else if ((*state)->src_node != NULL &&
4420                     pf_src_connlimit(state)) {
4421                         REASON_SET(reason, PFRES_SRCLIMIT);
4422                         return (PF_DROP);
4423                 } else
4424                         (*state)->src.state = PF_TCPS_PROXY_DST;
4425         }
4426         if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4427                 if (direction == (*state)->direction) {
4428                         if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4429                             (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4430                             (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4431                                 REASON_SET(reason, PFRES_SYNPROXY);
4432                                 return (PF_DROP);
4433                         }
4434                         (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4435                         if ((*state)->dst.seqhi == 1)
4436                                 (*state)->dst.seqhi = htonl(arc4random());
4437                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4438                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4439                             sk->port[pd->sidx], sk->port[pd->didx],
4440                             (*state)->dst.seqhi, 0, TH_SYN, 0,
4441                             (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4442                         REASON_SET(reason, PFRES_SYNPROXY);
4443                         return (PF_SYNPROXY_DROP);
4444                 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4445                     (TH_SYN|TH_ACK)) ||
4446                     (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4447                         REASON_SET(reason, PFRES_SYNPROXY);
4448                         return (PF_DROP);
4449                 } else {
4450                         (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4451                         (*state)->dst.seqlo = ntohl(th->th_seq);
4452                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4453                             pd->src, th->th_dport, th->th_sport,
4454                             ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4455                             TH_ACK, (*state)->src.max_win, 0, 0, 0,
4456                             (*state)->tag, NULL);
4457                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4458                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4459                             sk->port[pd->sidx], sk->port[pd->didx],
4460                             (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4461                             TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4462                         (*state)->src.seqdiff = (*state)->dst.seqhi -
4463                             (*state)->src.seqlo;
4464                         (*state)->dst.seqdiff = (*state)->src.seqhi -
4465                             (*state)->dst.seqlo;
4466                         (*state)->src.seqhi = (*state)->src.seqlo +
4467                             (*state)->dst.max_win;
4468                         (*state)->dst.seqhi = (*state)->dst.seqlo +
4469                             (*state)->src.max_win;
4470                         (*state)->src.wscale = (*state)->dst.wscale = 0;
4471                         (*state)->src.state = (*state)->dst.state =
4472                             TCPS_ESTABLISHED;
4473                         REASON_SET(reason, PFRES_SYNPROXY);
4474                         return (PF_SYNPROXY_DROP);
4475                 }
4476         }
4477
4478         if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4479             dst->state >= TCPS_FIN_WAIT_2 &&
4480             src->state >= TCPS_FIN_WAIT_2) {
4481                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4482                         printf("pf: state reuse ");
4483                         pf_print_state(*state);
4484                         pf_print_flags(th->th_flags);
4485                         printf("\n");
4486                 }
4487                 /* XXX make sure it's the same direction ?? */
4488                 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4489                 pf_unlink_state(*state, PF_ENTER_LOCKED);
4490                 *state = NULL;
4491                 return (PF_DROP);
4492         }
4493
4494         if ((*state)->state_flags & PFSTATE_SLOPPY) {
4495                 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4496                         return (PF_DROP);
4497         } else {
4498                 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4499                     &copyback) == PF_DROP)
4500                         return (PF_DROP);
4501         }
4502
4503         /* translate source/destination address, if necessary */
4504         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4505                 struct pf_state_key *nk = (*state)->key[pd->didx];
4506
4507                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4508                     nk->port[pd->sidx] != th->th_sport)
4509                         pf_change_ap(m, pd->src, &th->th_sport,
4510                             pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
4511                             nk->port[pd->sidx], 0, pd->af);
4512
4513                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4514                     nk->port[pd->didx] != th->th_dport)
4515                         pf_change_ap(m, pd->dst, &th->th_dport,
4516                             pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
4517                             nk->port[pd->didx], 0, pd->af);
4518                 copyback = 1;
4519         }
4520
4521         /* Copyback sequence modulation or stateful scrub changes if needed */
4522         if (copyback)
4523                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
4524
4525         return (PF_PASS);
4526 }
4527
4528 static int
4529 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
4530     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4531 {
4532         struct pf_state_peer    *src, *dst;
4533         struct pf_state_key_cmp  key;
4534         struct udphdr           *uh = pd->hdr.udp;
4535
4536         bzero(&key, sizeof(key));
4537         key.af = pd->af;
4538         key.proto = IPPROTO_UDP;
4539         if (direction == PF_IN) {       /* wire side, straight */
4540                 PF_ACPY(&key.addr[0], pd->src, key.af);
4541                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4542                 key.port[0] = uh->uh_sport;
4543                 key.port[1] = uh->uh_dport;
4544         } else {                        /* stack side, reverse */
4545                 PF_ACPY(&key.addr[1], pd->src, key.af);
4546                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4547                 key.port[1] = uh->uh_sport;
4548                 key.port[0] = uh->uh_dport;
4549         }
4550
4551         STATE_LOOKUP(kif, &key, direction, *state, pd);
4552
4553         if (direction == (*state)->direction) {
4554                 src = &(*state)->src;
4555                 dst = &(*state)->dst;
4556         } else {
4557                 src = &(*state)->dst;
4558                 dst = &(*state)->src;
4559         }
4560
4561         /* update states */
4562         if (src->state < PFUDPS_SINGLE)
4563                 src->state = PFUDPS_SINGLE;
4564         if (dst->state == PFUDPS_SINGLE)
4565                 dst->state = PFUDPS_MULTIPLE;
4566
4567         /* update expire time */
4568         (*state)->expire = time_uptime;
4569         if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4570                 (*state)->timeout = PFTM_UDP_MULTIPLE;
4571         else
4572                 (*state)->timeout = PFTM_UDP_SINGLE;
4573
4574         /* translate source/destination address, if necessary */
4575         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4576                 struct pf_state_key *nk = (*state)->key[pd->didx];
4577
4578                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4579                     nk->port[pd->sidx] != uh->uh_sport)
4580                         pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
4581                             &uh->uh_sum, &nk->addr[pd->sidx],
4582                             nk->port[pd->sidx], 1, pd->af);
4583
4584                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4585                     nk->port[pd->didx] != uh->uh_dport)
4586                         pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
4587                             &uh->uh_sum, &nk->addr[pd->didx],
4588                             nk->port[pd->didx], 1, pd->af);
4589                 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4590         }
4591
4592         return (PF_PASS);
4593 }
4594
4595 static int
4596 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
4597     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4598 {
4599         struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
4600         u_int16_t        icmpid = 0, *icmpsum;
4601         u_int8_t         icmptype, icmpcode;
4602         int              state_icmp = 0;
4603         struct pf_state_key_cmp key;
4604
4605         bzero(&key, sizeof(key));
4606         switch (pd->proto) {
4607 #ifdef INET
4608         case IPPROTO_ICMP:
4609                 icmptype = pd->hdr.icmp->icmp_type;
4610                 icmpcode = pd->hdr.icmp->icmp_code;
4611                 icmpid = pd->hdr.icmp->icmp_id;
4612                 icmpsum = &pd->hdr.icmp->icmp_cksum;
4613
4614                 if (icmptype == ICMP_UNREACH ||
4615                     icmptype == ICMP_SOURCEQUENCH ||
4616                     icmptype == ICMP_REDIRECT ||
4617                     icmptype == ICMP_TIMXCEED ||
4618                     icmptype == ICMP_PARAMPROB)
4619                         state_icmp++;
4620                 break;
4621 #endif /* INET */
4622 #ifdef INET6
4623         case IPPROTO_ICMPV6:
4624                 icmptype = pd->hdr.icmp6->icmp6_type;
4625                 icmpcode = pd->hdr.icmp6->icmp6_code;
4626                 icmpid = pd->hdr.icmp6->icmp6_id;
4627                 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4628
4629                 if (icmptype == ICMP6_DST_UNREACH ||
4630                     icmptype == ICMP6_PACKET_TOO_BIG ||
4631                     icmptype == ICMP6_TIME_EXCEEDED ||
4632                     icmptype == ICMP6_PARAM_PROB)
4633                         state_icmp++;
4634                 break;
4635 #endif /* INET6 */
4636         }
4637
4638         if (!state_icmp) {
4639                 /*
4640                  * ICMP query/reply message not related to a TCP/UDP packet.
4641                  * Search for an ICMP state.
4642                  */
4643                 key.af = pd->af;
4644                 key.proto = pd->proto;
4645                 key.port[0] = key.port[1] = icmpid;
4646                 if (direction == PF_IN) {       /* wire side, straight */
4647                         PF_ACPY(&key.addr[0], pd->src, key.af);
4648                         PF_ACPY(&key.addr[1], pd->dst, key.af);
4649                 } else {                        /* stack side, reverse */
4650                         PF_ACPY(&key.addr[1], pd->src, key.af);
4651                         PF_ACPY(&key.addr[0], pd->dst, key.af);
4652                 }
4653
4654                 STATE_LOOKUP(kif, &key, direction, *state, pd);
4655
4656                 (*state)->expire = time_uptime;
4657                 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4658
4659                 /* translate source/destination address, if necessary */
4660                 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4661                         struct pf_state_key *nk = (*state)->key[pd->didx];
4662
4663                         switch (pd->af) {
4664 #ifdef INET
4665                         case AF_INET:
4666                                 if (PF_ANEQ(pd->src,
4667                                     &nk->addr[pd->sidx], AF_INET))
4668                                         pf_change_a(&saddr->v4.s_addr,
4669                                             pd->ip_sum,
4670                                             nk->addr[pd->sidx].v4.s_addr, 0);
4671
4672                                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4673                                     AF_INET))
4674                                         pf_change_a(&daddr->v4.s_addr,
4675                                             pd->ip_sum,
4676                                             nk->addr[pd->didx].v4.s_addr, 0);
4677
4678                                 if (nk->port[0] !=
4679                                     pd->hdr.icmp->icmp_id) {
4680                                         pd->hdr.icmp->icmp_cksum =
4681                                             pf_cksum_fixup(
4682                                             pd->hdr.icmp->icmp_cksum, icmpid,
4683                                             nk->port[pd->sidx], 0);
4684                                         pd->hdr.icmp->icmp_id =
4685                                             nk->port[pd->sidx];
4686                                 }
4687
4688                                 m_copyback(m, off, ICMP_MINLEN,
4689                                     (caddr_t )pd->hdr.icmp);
4690                                 break;
4691 #endif /* INET */
4692 #ifdef INET6
4693                         case AF_INET6:
4694                                 if (PF_ANEQ(pd->src,
4695                                     &nk->addr[pd->sidx], AF_INET6))
4696                                         pf_change_a6(saddr,
4697                                             &pd->hdr.icmp6->icmp6_cksum,
4698                                             &nk->addr[pd->sidx], 0);
4699
4700                                 if (PF_ANEQ(pd->dst,
4701                                     &nk->addr[pd->didx], AF_INET6))
4702                                         pf_change_a6(daddr,
4703                                             &pd->hdr.icmp6->icmp6_cksum,
4704                                             &nk->addr[pd->didx], 0);
4705
4706                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4707                                     (caddr_t )pd->hdr.icmp6);
4708                                 break;
4709 #endif /* INET6 */
4710                         }
4711                 }
4712                 return (PF_PASS);
4713
4714         } else {
4715                 /*
4716                  * ICMP error message in response to a TCP/UDP packet.
4717                  * Extract the inner TCP/UDP header and search for that state.
4718                  */
4719
4720                 struct pf_pdesc pd2;
4721                 bzero(&pd2, sizeof pd2);
4722 #ifdef INET
4723                 struct ip       h2;
4724 #endif /* INET */
4725 #ifdef INET6
4726                 struct ip6_hdr  h2_6;
4727                 int             terminal = 0;
4728 #endif /* INET6 */
4729                 int             ipoff2 = 0;
4730                 int             off2 = 0;
4731
4732                 pd2.af = pd->af;
4733                 /* Payload packet is from the opposite direction. */
4734                 pd2.sidx = (direction == PF_IN) ? 1 : 0;
4735                 pd2.didx = (direction == PF_IN) ? 0 : 1;
4736                 switch (pd->af) {
4737 #ifdef INET
4738                 case AF_INET:
4739                         /* offset of h2 in mbuf chain */
4740                         ipoff2 = off + ICMP_MINLEN;
4741
4742                         if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4743                             NULL, reason, pd2.af)) {
4744                                 DPFPRINTF(PF_DEBUG_MISC,
4745                                     ("pf: ICMP error message too short "
4746                                     "(ip)\n"));
4747                                 return (PF_DROP);
4748                         }
4749                         /*
4750                          * ICMP error messages don't refer to non-first
4751                          * fragments
4752                          */
4753                         if (h2.ip_off & htons(IP_OFFMASK)) {
4754                                 REASON_SET(reason, PFRES_FRAG);
4755                                 return (PF_DROP);
4756                         }
4757
4758                         /* offset of protocol header that follows h2 */
4759                         off2 = ipoff2 + (h2.ip_hl << 2);
4760
4761                         pd2.proto = h2.ip_p;
4762                         pd2.src = (struct pf_addr *)&h2.ip_src;
4763                         pd2.dst = (struct pf_addr *)&h2.ip_dst;
4764                         pd2.ip_sum = &h2.ip_sum;
4765                         break;
4766 #endif /* INET */
4767 #ifdef INET6
4768                 case AF_INET6:
4769                         ipoff2 = off + sizeof(struct icmp6_hdr);
4770
4771                         if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4772                             NULL, reason, pd2.af)) {
4773                                 DPFPRINTF(PF_DEBUG_MISC,
4774                                     ("pf: ICMP error message too short "
4775                                     "(ip6)\n"));
4776                                 return (PF_DROP);
4777                         }
4778                         pd2.proto = h2_6.ip6_nxt;
4779                         pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4780                         pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4781                         pd2.ip_sum = NULL;
4782                         off2 = ipoff2 + sizeof(h2_6);
4783                         do {
4784                                 switch (pd2.proto) {
4785                                 case IPPROTO_FRAGMENT:
4786                                         /*
4787                                          * ICMPv6 error messages for
4788                                          * non-first fragments
4789                                          */
4790                                         REASON_SET(reason, PFRES_FRAG);
4791                                         return (PF_DROP);
4792                                 case IPPROTO_AH:
4793                                 case IPPROTO_HOPOPTS:
4794                                 case IPPROTO_ROUTING:
4795                                 case IPPROTO_DSTOPTS: {
4796                                         /* get next header and header length */
4797                                         struct ip6_ext opt6;
4798
4799                                         if (!pf_pull_hdr(m, off2, &opt6,
4800                                             sizeof(opt6), NULL, reason,
4801                                             pd2.af)) {
4802                                                 DPFPRINTF(PF_DEBUG_MISC,
4803                                                     ("pf: ICMPv6 short opt\n"));
4804                                                 return (PF_DROP);
4805                                         }
4806                                         if (pd2.proto == IPPROTO_AH)
4807                                                 off2 += (opt6.ip6e_len + 2) * 4;
4808                                         else
4809                                                 off2 += (opt6.ip6e_len + 1) * 8;
4810                                         pd2.proto = opt6.ip6e_nxt;
4811                                         /* goto the next header */
4812                                         break;
4813                                 }
4814                                 default:
4815                                         terminal++;
4816                                         break;
4817                                 }
4818                         } while (!terminal);
4819                         break;
4820 #endif /* INET6 */
4821                 }
4822
4823                 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) {
4824                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
4825                                 printf("pf: BAD ICMP %d:%d outer dst: ",
4826                                     icmptype, icmpcode);
4827                                 pf_print_host(pd->src, 0, pd->af);
4828                                 printf(" -> ");
4829                                 pf_print_host(pd->dst, 0, pd->af);
4830                                 printf(" inner src: ");
4831                                 pf_print_host(pd2.src, 0, pd2.af);
4832                                 printf(" -> ");
4833                                 pf_print_host(pd2.dst, 0, pd2.af);
4834                                 printf("\n");
4835                         }
4836                         REASON_SET(reason, PFRES_BADSTATE);
4837                         return (PF_DROP);
4838                 }
4839
4840                 switch (pd2.proto) {
4841                 case IPPROTO_TCP: {
4842                         struct tcphdr            th;
4843                         u_int32_t                seq;
4844                         struct pf_state_peer    *src, *dst;
4845                         u_int8_t                 dws;
4846                         int                      copyback = 0;
4847
4848                         /*
4849                          * Only the first 8 bytes of the TCP header can be
4850                          * expected. Don't access any TCP header fields after
4851                          * th_seq, an ackskew test is not possible.
4852                          */
4853                         if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4854                             pd2.af)) {
4855                                 DPFPRINTF(PF_DEBUG_MISC,
4856                                     ("pf: ICMP error message too short "
4857                                     "(tcp)\n"));
4858                                 return (PF_DROP);
4859                         }
4860
4861                         key.af = pd2.af;
4862                         key.proto = IPPROTO_TCP;
4863                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4864                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4865                         key.port[pd2.sidx] = th.th_sport;
4866                         key.port[pd2.didx] = th.th_dport;
4867
4868                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4869
4870                         if (direction == (*state)->direction) {
4871                                 src = &(*state)->dst;
4872                                 dst = &(*state)->src;
4873                         } else {
4874                                 src = &(*state)->src;
4875                                 dst = &(*state)->dst;
4876                         }
4877
4878                         if (src->wscale && dst->wscale)
4879                                 dws = dst->wscale & PF_WSCALE_MASK;
4880                         else
4881                                 dws = 0;
4882
4883                         /* Demodulate sequence number */
4884                         seq = ntohl(th.th_seq) - src->seqdiff;
4885                         if (src->seqdiff) {
4886                                 pf_change_a(&th.th_seq, icmpsum,
4887                                     htonl(seq), 0);
4888                                 copyback = 1;
4889                         }
4890
4891                         if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4892                             (!SEQ_GEQ(src->seqhi, seq) ||
4893                             !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4894                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4895                                         printf("pf: BAD ICMP %d:%d ",
4896                                             icmptype, icmpcode);
4897                                         pf_print_host(pd->src, 0, pd->af);
4898                                         printf(" -> ");
4899                                         pf_print_host(pd->dst, 0, pd->af);
4900                                         printf(" state: ");
4901                                         pf_print_state(*state);
4902                                         printf(" seq=%u\n", seq);
4903                                 }
4904                                 REASON_SET(reason, PFRES_BADSTATE);
4905                                 return (PF_DROP);
4906                         } else {
4907                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4908                                         printf("pf: OK ICMP %d:%d ",
4909                                             icmptype, icmpcode);
4910                                         pf_print_host(pd->src, 0, pd->af);
4911                                         printf(" -> ");
4912                                         pf_print_host(pd->dst, 0, pd->af);
4913                                         printf(" state: ");
4914                                         pf_print_state(*state);
4915                                         printf(" seq=%u\n", seq);
4916                                 }
4917                         }
4918
4919                         /* translate source/destination address, if necessary */
4920                         if ((*state)->key[PF_SK_WIRE] !=
4921                             (*state)->key[PF_SK_STACK]) {
4922                                 struct pf_state_key *nk =
4923                                     (*state)->key[pd->didx];
4924
4925                                 if (PF_ANEQ(pd2.src,
4926                                     &nk->addr[pd2.sidx], pd2.af) ||
4927                                     nk->port[pd2.sidx] != th.th_sport)
4928                                         pf_change_icmp(pd2.src, &th.th_sport,
4929                                             daddr, &nk->addr[pd2.sidx],
4930                                             nk->port[pd2.sidx], NULL,
4931                                             pd2.ip_sum, icmpsum,
4932                                             pd->ip_sum, 0, pd2.af);
4933
4934                                 if (PF_ANEQ(pd2.dst,
4935                                     &nk->addr[pd2.didx], pd2.af) ||
4936                                     nk->port[pd2.didx] != th.th_dport)
4937                                         pf_change_icmp(pd2.dst, &th.th_dport,
4938                                             saddr, &nk->addr[pd2.didx],
4939                                             nk->port[pd2.didx], NULL,
4940                                             pd2.ip_sum, icmpsum,
4941                                             pd->ip_sum, 0, pd2.af);
4942                                 copyback = 1;
4943                         }
4944
4945                         if (copyback) {
4946                                 switch (pd2.af) {
4947 #ifdef INET
4948                                 case AF_INET:
4949                                         m_copyback(m, off, ICMP_MINLEN,
4950                                             (caddr_t )pd->hdr.icmp);
4951                                         m_copyback(m, ipoff2, sizeof(h2),
4952                                             (caddr_t )&h2);
4953                                         break;
4954 #endif /* INET */
4955 #ifdef INET6
4956                                 case AF_INET6:
4957                                         m_copyback(m, off,
4958                                             sizeof(struct icmp6_hdr),
4959                                             (caddr_t )pd->hdr.icmp6);
4960                                         m_copyback(m, ipoff2, sizeof(h2_6),
4961                                             (caddr_t )&h2_6);
4962                                         break;
4963 #endif /* INET6 */
4964                                 }
4965                                 m_copyback(m, off2, 8, (caddr_t)&th);
4966                         }
4967
4968                         return (PF_PASS);
4969                         break;
4970                 }
4971                 case IPPROTO_UDP: {
4972                         struct udphdr           uh;
4973
4974                         if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
4975                             NULL, reason, pd2.af)) {
4976                                 DPFPRINTF(PF_DEBUG_MISC,
4977                                     ("pf: ICMP error message too short "
4978                                     "(udp)\n"));
4979                                 return (PF_DROP);
4980                         }
4981
4982                         key.af = pd2.af;
4983                         key.proto = IPPROTO_UDP;
4984                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4985                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4986                         key.port[pd2.sidx] = uh.uh_sport;
4987                         key.port[pd2.didx] = uh.uh_dport;
4988
4989                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4990
4991                         /* translate source/destination address, if necessary */
4992                         if ((*state)->key[PF_SK_WIRE] !=
4993                             (*state)->key[PF_SK_STACK]) {
4994                                 struct pf_state_key *nk =
4995                                     (*state)->key[pd->didx];
4996
4997                                 if (PF_ANEQ(pd2.src,
4998                                     &nk->addr[pd2.sidx], pd2.af) ||
4999                                     nk->port[pd2.sidx] != uh.uh_sport)
5000                                         pf_change_icmp(pd2.src, &uh.uh_sport,
5001                                             daddr, &nk->addr[pd2.sidx],
5002                                             nk->port[pd2.sidx], &uh.uh_sum,
5003                                             pd2.ip_sum, icmpsum,
5004                                             pd->ip_sum, 1, pd2.af);
5005
5006                                 if (PF_ANEQ(pd2.dst,
5007                                     &nk->addr[pd2.didx], pd2.af) ||
5008                                     nk->port[pd2.didx] != uh.uh_dport)
5009                                         pf_change_icmp(pd2.dst, &uh.uh_dport,
5010                                             saddr, &nk->addr[pd2.didx],
5011                                             nk->port[pd2.didx], &uh.uh_sum,
5012                                             pd2.ip_sum, icmpsum,
5013                                             pd->ip_sum, 1, pd2.af);
5014
5015                                 switch (pd2.af) {
5016 #ifdef INET
5017                                 case AF_INET:
5018                                         m_copyback(m, off, ICMP_MINLEN,
5019                                             (caddr_t )pd->hdr.icmp);
5020                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5021                                         break;
5022 #endif /* INET */
5023 #ifdef INET6
5024                                 case AF_INET6:
5025                                         m_copyback(m, off,
5026                                             sizeof(struct icmp6_hdr),
5027                                             (caddr_t )pd->hdr.icmp6);
5028                                         m_copyback(m, ipoff2, sizeof(h2_6),
5029                                             (caddr_t )&h2_6);
5030                                         break;
5031 #endif /* INET6 */
5032                                 }
5033                                 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
5034                         }
5035                         return (PF_PASS);
5036                         break;
5037                 }
5038 #ifdef INET
5039                 case IPPROTO_ICMP: {
5040                         struct icmp             iih;
5041
5042                         if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
5043                             NULL, reason, pd2.af)) {
5044                                 DPFPRINTF(PF_DEBUG_MISC,
5045                                     ("pf: ICMP error message too short i"
5046                                     "(icmp)\n"));
5047                                 return (PF_DROP);
5048                         }
5049
5050                         key.af = pd2.af;
5051                         key.proto = IPPROTO_ICMP;
5052                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5053                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5054                         key.port[0] = key.port[1] = iih.icmp_id;
5055
5056                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5057
5058                         /* translate source/destination address, if necessary */
5059                         if ((*state)->key[PF_SK_WIRE] !=
5060                             (*state)->key[PF_SK_STACK]) {
5061                                 struct pf_state_key *nk =
5062                                     (*state)->key[pd->didx];
5063
5064                                 if (PF_ANEQ(pd2.src,
5065                                     &nk->addr[pd2.sidx], pd2.af) ||
5066                                     nk->port[pd2.sidx] != iih.icmp_id)
5067                                         pf_change_icmp(pd2.src, &iih.icmp_id,
5068                                             daddr, &nk->addr[pd2.sidx],
5069                                             nk->port[pd2.sidx], NULL,
5070                                             pd2.ip_sum, icmpsum,
5071                                             pd->ip_sum, 0, AF_INET);
5072
5073                                 if (PF_ANEQ(pd2.dst,
5074                                     &nk->addr[pd2.didx], pd2.af) ||
5075                                     nk->port[pd2.didx] != iih.icmp_id)
5076                                         pf_change_icmp(pd2.dst, &iih.icmp_id,
5077                                             saddr, &nk->addr[pd2.didx],
5078                                             nk->port[pd2.didx], NULL,
5079                                             pd2.ip_sum, icmpsum,
5080                                             pd->ip_sum, 0, AF_INET);
5081
5082                                 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
5083                                 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5084                                 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
5085                         }
5086                         return (PF_PASS);
5087                         break;
5088                 }
5089 #endif /* INET */
5090 #ifdef INET6
5091                 case IPPROTO_ICMPV6: {
5092                         struct icmp6_hdr        iih;
5093
5094                         if (!pf_pull_hdr(m, off2, &iih,
5095                             sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
5096                                 DPFPRINTF(PF_DEBUG_MISC,
5097                                     ("pf: ICMP error message too short "
5098                                     "(icmp6)\n"));
5099                                 return (PF_DROP);
5100                         }
5101
5102                         key.af = pd2.af;
5103                         key.proto = IPPROTO_ICMPV6;
5104                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5105                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5106                         key.port[0] = key.port[1] = iih.icmp6_id;
5107
5108                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5109
5110                         /* translate source/destination address, if necessary */
5111                         if ((*state)->key[PF_SK_WIRE] !=
5112                             (*state)->key[PF_SK_STACK]) {
5113                                 struct pf_state_key *nk =
5114                                     (*state)->key[pd->didx];
5115
5116                                 if (PF_ANEQ(pd2.src,
5117                                     &nk->addr[pd2.sidx], pd2.af) ||
5118                                     nk->port[pd2.sidx] != iih.icmp6_id)
5119                                         pf_change_icmp(pd2.src, &iih.icmp6_id,
5120                                             daddr, &nk->addr[pd2.sidx],
5121                                             nk->port[pd2.sidx], NULL,
5122                                             pd2.ip_sum, icmpsum,
5123                                             pd->ip_sum, 0, AF_INET6);
5124
5125                                 if (PF_ANEQ(pd2.dst,
5126                                     &nk->addr[pd2.didx], pd2.af) ||
5127                                     nk->port[pd2.didx] != iih.icmp6_id)
5128                                         pf_change_icmp(pd2.dst, &iih.icmp6_id,
5129                                             saddr, &nk->addr[pd2.didx],
5130                                             nk->port[pd2.didx], NULL,
5131                                             pd2.ip_sum, icmpsum,
5132                                             pd->ip_sum, 0, AF_INET6);
5133
5134                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
5135                                     (caddr_t)pd->hdr.icmp6);
5136                                 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
5137                                 m_copyback(m, off2, sizeof(struct icmp6_hdr),
5138                                     (caddr_t)&iih);
5139                         }
5140                         return (PF_PASS);
5141                         break;
5142                 }
5143 #endif /* INET6 */
5144                 default: {
5145                         key.af = pd2.af;
5146                         key.proto = pd2.proto;
5147                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5148                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5149                         key.port[0] = key.port[1] = 0;
5150
5151                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5152
5153                         /* translate source/destination address, if necessary */
5154                         if ((*state)->key[PF_SK_WIRE] !=
5155                             (*state)->key[PF_SK_STACK]) {
5156                                 struct pf_state_key *nk =
5157                                     (*state)->key[pd->didx];
5158
5159                                 if (PF_ANEQ(pd2.src,
5160                                     &nk->addr[pd2.sidx], pd2.af))
5161                                         pf_change_icmp(pd2.src, NULL, daddr,
5162                                             &nk->addr[pd2.sidx], 0, NULL,
5163                                             pd2.ip_sum, icmpsum,
5164                                             pd->ip_sum, 0, pd2.af);
5165
5166                                 if (PF_ANEQ(pd2.dst,
5167                                     &nk->addr[pd2.didx], pd2.af))
5168                                         pf_change_icmp(pd2.dst, NULL, saddr,
5169                                             &nk->addr[pd2.didx], 0, NULL,
5170                                             pd2.ip_sum, icmpsum,
5171                                             pd->ip_sum, 0, pd2.af);
5172
5173                                 switch (pd2.af) {
5174 #ifdef INET
5175                                 case AF_INET:
5176                                         m_copyback(m, off, ICMP_MINLEN,
5177                                             (caddr_t)pd->hdr.icmp);
5178                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5179                                         break;
5180 #endif /* INET */
5181 #ifdef INET6
5182                                 case AF_INET6:
5183                                         m_copyback(m, off,
5184                                             sizeof(struct icmp6_hdr),
5185                                             (caddr_t )pd->hdr.icmp6);
5186                                         m_copyback(m, ipoff2, sizeof(h2_6),
5187                                             (caddr_t )&h2_6);
5188                                         break;
5189 #endif /* INET6 */
5190                                 }
5191                         }
5192                         return (PF_PASS);
5193                         break;
5194                 }
5195                 }
5196         }
5197 }
5198
5199 static int
5200 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
5201     struct mbuf *m, struct pf_pdesc *pd)
5202 {
5203         struct pf_state_peer    *src, *dst;
5204         struct pf_state_key_cmp  key;
5205
5206         bzero(&key, sizeof(key));
5207         key.af = pd->af;
5208         key.proto = pd->proto;
5209         if (direction == PF_IN) {
5210                 PF_ACPY(&key.addr[0], pd->src, key.af);
5211                 PF_ACPY(&key.addr[1], pd->dst, key.af);
5212                 key.port[0] = key.port[1] = 0;
5213         } else {
5214                 PF_ACPY(&key.addr[1], pd->src, key.af);
5215                 PF_ACPY(&key.addr[0], pd->dst, key.af);
5216                 key.port[1] = key.port[0] = 0;
5217         }
5218
5219         STATE_LOOKUP(kif, &key, direction, *state, pd);
5220
5221         if (direction == (*state)->direction) {
5222                 src = &(*state)->src;
5223                 dst = &(*state)->dst;
5224         } else {
5225                 src = &(*state)->dst;
5226                 dst = &(*state)->src;
5227         }
5228
5229         /* update states */
5230         if (src->state < PFOTHERS_SINGLE)
5231                 src->state = PFOTHERS_SINGLE;
5232         if (dst->state == PFOTHERS_SINGLE)
5233                 dst->state = PFOTHERS_MULTIPLE;
5234
5235         /* update expire time */
5236         (*state)->expire = time_uptime;
5237         if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5238                 (*state)->timeout = PFTM_OTHER_MULTIPLE;
5239         else
5240                 (*state)->timeout = PFTM_OTHER_SINGLE;
5241
5242         /* translate source/destination address, if necessary */
5243         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5244                 struct pf_state_key *nk = (*state)->key[pd->didx];
5245
5246                 KASSERT(nk, ("%s: nk is null", __func__));
5247                 KASSERT(pd, ("%s: pd is null", __func__));
5248                 KASSERT(pd->src, ("%s: pd->src is null", __func__));
5249                 KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5250                 switch (pd->af) {
5251 #ifdef INET
5252                 case AF_INET:
5253                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5254                                 pf_change_a(&pd->src->v4.s_addr,
5255                                     pd->ip_sum,
5256                                     nk->addr[pd->sidx].v4.s_addr,
5257                                     0);
5258
5259                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5260                                 pf_change_a(&pd->dst->v4.s_addr,
5261                                     pd->ip_sum,
5262                                     nk->addr[pd->didx].v4.s_addr,
5263                                     0);
5264
5265                         break;
5266 #endif /* INET */
5267 #ifdef INET6
5268                 case AF_INET6:
5269                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5270                                 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5271
5272                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5273                                 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5274 #endif /* INET6 */
5275                 }
5276         }
5277         return (PF_PASS);
5278 }
5279
5280 /*
5281  * ipoff and off are measured from the start of the mbuf chain.
5282  * h must be at "ipoff" on the mbuf chain.
5283  */
5284 void *
5285 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5286     u_short *actionp, u_short *reasonp, sa_family_t af)
5287 {
5288         switch (af) {
5289 #ifdef INET
5290         case AF_INET: {
5291                 struct ip       *h = mtod(m, struct ip *);
5292                 u_int16_t        fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5293
5294                 if (fragoff) {
5295                         if (fragoff >= len)
5296                                 ACTION_SET(actionp, PF_PASS);
5297                         else {
5298                                 ACTION_SET(actionp, PF_DROP);
5299                                 REASON_SET(reasonp, PFRES_FRAG);
5300                         }
5301                         return (NULL);
5302                 }
5303                 if (m->m_pkthdr.len < off + len ||
5304                     ntohs(h->ip_len) < off + len) {
5305                         ACTION_SET(actionp, PF_DROP);
5306                         REASON_SET(reasonp, PFRES_SHORT);
5307                         return (NULL);
5308                 }
5309                 break;
5310         }
5311 #endif /* INET */
5312 #ifdef INET6
5313         case AF_INET6: {
5314                 struct ip6_hdr  *h = mtod(m, struct ip6_hdr *);
5315
5316                 if (m->m_pkthdr.len < off + len ||
5317                     (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5318                     (unsigned)(off + len)) {
5319                         ACTION_SET(actionp, PF_DROP);
5320                         REASON_SET(reasonp, PFRES_SHORT);
5321                         return (NULL);
5322                 }
5323                 break;
5324         }
5325 #endif /* INET6 */
5326         }
5327         m_copydata(m, off, len, p);
5328         return (p);
5329 }
5330
5331 int
5332 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5333     int rtableid)
5334 {
5335         struct ifnet            *ifp;
5336
5337         /*
5338          * Skip check for addresses with embedded interface scope,
5339          * as they would always match anyway.
5340          */
5341         if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
5342                 return (1);
5343
5344         if (af != AF_INET && af != AF_INET6)
5345                 return (0);
5346
5347         /* Skip checks for ipsec interfaces */
5348         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5349                 return (1);
5350
5351         ifp = (kif != NULL) ? kif->pfik_ifp : NULL;
5352
5353         switch (af) {
5354 #ifdef INET6
5355         case AF_INET6:
5356                 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE,
5357                     ifp));
5358 #endif
5359 #ifdef INET
5360         case AF_INET:
5361                 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE,
5362                     ifp));
5363 #endif
5364         }
5365
5366         return (0);
5367 }
5368
5369 #ifdef INET
5370 static void
5371 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5372     struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp)
5373 {
5374         struct mbuf             *m0, *m1;
5375         struct sockaddr_in      dst;
5376         struct ip               *ip;
5377         struct ifnet            *ifp = NULL;
5378         struct pf_addr           naddr;
5379         struct pf_src_node      *sn = NULL;
5380         int                      error = 0;
5381         uint16_t                 ip_len, ip_off;
5382
5383         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5384         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5385             __func__));
5386
5387         if ((pd->pf_mtag == NULL &&
5388             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5389             pd->pf_mtag->routed++ > 3) {
5390                 m0 = *m;
5391                 *m = NULL;
5392                 goto bad_locked;
5393         }
5394
5395         if (r->rt == PF_DUPTO) {
5396                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5397                         if (s)
5398                                 PF_STATE_UNLOCK(s);
5399                         return;
5400                 }
5401         } else {
5402                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5403                         if (s)
5404                                 PF_STATE_UNLOCK(s);
5405                         return;
5406                 }
5407                 m0 = *m;
5408         }
5409
5410         ip = mtod(m0, struct ip *);
5411
5412         bzero(&dst, sizeof(dst));
5413         dst.sin_family = AF_INET;
5414         dst.sin_len = sizeof(dst);
5415         dst.sin_addr = ip->ip_dst;
5416
5417         bzero(&naddr, sizeof(naddr));
5418
5419         if (TAILQ_EMPTY(&r->rpool.list)) {
5420                 DPFPRINTF(PF_DEBUG_URGENT,
5421                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5422                 goto bad_locked;
5423         }
5424         if (s == NULL) {
5425                 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5426                     &naddr, NULL, &sn);
5427                 if (!PF_AZERO(&naddr, AF_INET))
5428                         dst.sin_addr.s_addr = naddr.v4.s_addr;
5429                 ifp = r->rpool.cur->kif ?
5430                     r->rpool.cur->kif->pfik_ifp : NULL;
5431         } else {
5432                 if (!PF_AZERO(&s->rt_addr, AF_INET))
5433                         dst.sin_addr.s_addr =
5434                             s->rt_addr.v4.s_addr;
5435                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5436                 PF_STATE_UNLOCK(s);
5437         }
5438         if (ifp == NULL)
5439                 goto bad;
5440
5441         if (oifp != ifp) {
5442                 if (pf_test(PF_OUT, 0, ifp, &m0, inp) != PF_PASS)
5443                         goto bad;
5444                 else if (m0 == NULL)
5445                         goto done;
5446                 if (m0->m_len < sizeof(struct ip)) {
5447                         DPFPRINTF(PF_DEBUG_URGENT,
5448                             ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5449                         goto bad;
5450                 }
5451                 ip = mtod(m0, struct ip *);
5452         }
5453
5454         if (ifp->if_flags & IFF_LOOPBACK)
5455                 m0->m_flags |= M_SKIP_FIREWALL;
5456
5457         ip_len = ntohs(ip->ip_len);
5458         ip_off = ntohs(ip->ip_off);
5459
5460         /* Copied from FreeBSD 10.0-CURRENT ip_output. */
5461         m0->m_pkthdr.csum_flags |= CSUM_IP;
5462         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5463                 in_delayed_cksum(m0);
5464                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5465         }
5466 #if defined(SCTP) || defined(SCTP_SUPPORT)
5467         if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5468                 sctp_delayed_cksum(m0, (uint32_t)(ip->ip_hl << 2));
5469                 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5470         }
5471 #endif
5472
5473         /*
5474          * If small enough for interface, or the interface will take
5475          * care of the fragmentation for us, we can just send directly.
5476          */
5477         if (ip_len <= ifp->if_mtu ||
5478             (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
5479                 ip->ip_sum = 0;
5480                 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5481                         ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5482                         m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5483                 }
5484                 m_clrprotoflags(m0);    /* Avoid confusing lower layers. */
5485                 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5486                 goto done;
5487         }
5488
5489         /* Balk when DF bit is set or the interface didn't support TSO. */
5490         if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5491                 error = EMSGSIZE;
5492                 KMOD_IPSTAT_INC(ips_cantfrag);
5493                 if (r->rt != PF_DUPTO) {
5494                         icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5495                             ifp->if_mtu);
5496                         goto done;
5497                 } else
5498                         goto bad;
5499         }
5500
5501         error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5502         if (error)
5503                 goto bad;
5504
5505         for (; m0; m0 = m1) {
5506                 m1 = m0->m_nextpkt;
5507                 m0->m_nextpkt = NULL;
5508                 if (error == 0) {
5509                         m_clrprotoflags(m0);
5510                         error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5511                 } else
5512                         m_freem(m0);
5513         }
5514
5515         if (error == 0)
5516                 KMOD_IPSTAT_INC(ips_fragmented);
5517
5518 done:
5519         if (r->rt != PF_DUPTO)
5520                 *m = NULL;
5521         return;
5522
5523 bad_locked:
5524         if (s)
5525                 PF_STATE_UNLOCK(s);
5526 bad:
5527         m_freem(m0);
5528         goto done;
5529 }
5530 #endif /* INET */
5531
5532 #ifdef INET6
5533 static void
5534 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5535     struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp)
5536 {
5537         struct mbuf             *m0;
5538         struct sockaddr_in6     dst;
5539         struct ip6_hdr          *ip6;
5540         struct ifnet            *ifp = NULL;
5541         struct pf_addr           naddr;
5542         struct pf_src_node      *sn = NULL;
5543
5544         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5545         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5546             __func__));
5547
5548         if ((pd->pf_mtag == NULL &&
5549             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5550             pd->pf_mtag->routed++ > 3) {
5551                 m0 = *m;
5552                 *m = NULL;
5553                 goto bad_locked;
5554         }
5555
5556         if (r->rt == PF_DUPTO) {
5557                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5558                         if (s)
5559                                 PF_STATE_UNLOCK(s);
5560                         return;
5561                 }
5562         } else {
5563                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5564                         if (s)
5565                                 PF_STATE_UNLOCK(s);
5566                         return;
5567                 }
5568                 m0 = *m;
5569         }
5570
5571         ip6 = mtod(m0, struct ip6_hdr *);
5572
5573         bzero(&dst, sizeof(dst));
5574         dst.sin6_family = AF_INET6;
5575         dst.sin6_len = sizeof(dst);
5576         dst.sin6_addr = ip6->ip6_dst;
5577
5578         bzero(&naddr, sizeof(naddr));
5579
5580         if (TAILQ_EMPTY(&r->rpool.list)) {
5581                 DPFPRINTF(PF_DEBUG_URGENT,
5582                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5583                 goto bad_locked;
5584         }
5585         if (s == NULL) {
5586                 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5587                     &naddr, NULL, &sn);
5588                 if (!PF_AZERO(&naddr, AF_INET6))
5589                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5590                             &naddr, AF_INET6);
5591                 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5592         } else {
5593                 if (!PF_AZERO(&s->rt_addr, AF_INET6))
5594                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5595                             &s->rt_addr, AF_INET6);
5596                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5597         }
5598
5599         if (s)
5600                 PF_STATE_UNLOCK(s);
5601
5602         if (ifp == NULL)
5603                 goto bad;
5604
5605         if (oifp != ifp) {
5606                 if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp) != PF_PASS)
5607                         goto bad;
5608                 else if (m0 == NULL)
5609                         goto done;
5610                 if (m0->m_len < sizeof(struct ip6_hdr)) {
5611                         DPFPRINTF(PF_DEBUG_URGENT,
5612                             ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5613                             __func__));
5614                         goto bad;
5615                 }
5616                 ip6 = mtod(m0, struct ip6_hdr *);
5617         }
5618
5619         if (ifp->if_flags & IFF_LOOPBACK)
5620                 m0->m_flags |= M_SKIP_FIREWALL;
5621
5622         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
5623             ~ifp->if_hwassist) {
5624                 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
5625                 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
5626                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
5627         }
5628
5629         /*
5630          * If the packet is too large for the outgoing interface,
5631          * send back an icmp6 error.
5632          */
5633         if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5634                 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5635         if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5636                 nd6_output_ifp(ifp, ifp, m0, &dst, NULL);
5637         else {
5638                 in6_ifstat_inc(ifp, ifs6_in_toobig);
5639                 if (r->rt != PF_DUPTO)
5640                         icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5641                 else
5642                         goto bad;
5643         }
5644
5645 done:
5646         if (r->rt != PF_DUPTO)
5647                 *m = NULL;
5648         return;
5649
5650 bad_locked:
5651         if (s)
5652                 PF_STATE_UNLOCK(s);
5653 bad:
5654         m_freem(m0);
5655         goto done;
5656 }
5657 #endif /* INET6 */
5658
5659 /*
5660  * FreeBSD supports cksum offloads for the following drivers.
5661  *  em(4), fxp(4), lge(4), ndis(4), nge(4), re(4), ti(4), txp(4), xl(4)
5662  *
5663  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5664  *  network driver performed cksum including pseudo header, need to verify
5665  *   csum_data
5666  * CSUM_DATA_VALID :
5667  *  network driver performed cksum, needs to additional pseudo header
5668  *  cksum computation with partial csum_data(i.e. lack of H/W support for
5669  *  pseudo header, for instance hme(4), sk(4) and possibly gem(4))
5670  *
5671  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5672  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5673  * TCP/UDP layer.
5674  * Also, set csum_data to 0xffff to force cksum validation.
5675  */
5676 static int
5677 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5678 {
5679         u_int16_t sum = 0;
5680         int hw_assist = 0;
5681         struct ip *ip;
5682
5683         if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5684                 return (1);
5685         if (m->m_pkthdr.len < off + len)
5686                 return (1);
5687
5688         switch (p) {
5689         case IPPROTO_TCP:
5690                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5691                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5692                                 sum = m->m_pkthdr.csum_data;
5693                         } else {
5694                                 ip = mtod(m, struct ip *);
5695                                 sum = in_pseudo(ip->ip_src.s_addr,
5696                                 ip->ip_dst.s_addr, htonl((u_short)len +
5697                                 m->m_pkthdr.csum_data + IPPROTO_TCP));
5698                         }
5699                         sum ^= 0xffff;
5700                         ++hw_assist;
5701                 }
5702                 break;
5703         case IPPROTO_UDP:
5704                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5705                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5706                                 sum = m->m_pkthdr.csum_data;
5707                         } else {
5708                                 ip = mtod(m, struct ip *);
5709                                 sum = in_pseudo(ip->ip_src.s_addr,
5710                                 ip->ip_dst.s_addr, htonl((u_short)len +
5711                                 m->m_pkthdr.csum_data + IPPROTO_UDP));
5712                         }
5713                         sum ^= 0xffff;
5714                         ++hw_assist;
5715                 }
5716                 break;
5717         case IPPROTO_ICMP:
5718 #ifdef INET6
5719         case IPPROTO_ICMPV6:
5720 #endif /* INET6 */
5721                 break;
5722         default:
5723                 return (1);
5724         }
5725
5726         if (!hw_assist) {
5727                 switch (af) {
5728                 case AF_INET:
5729                         if (p == IPPROTO_ICMP) {
5730                                 if (m->m_len < off)
5731                                         return (1);
5732                                 m->m_data += off;
5733                                 m->m_len -= off;
5734                                 sum = in_cksum(m, len);
5735                                 m->m_data -= off;
5736                                 m->m_len += off;
5737                         } else {
5738                                 if (m->m_len < sizeof(struct ip))
5739                                         return (1);
5740                                 sum = in4_cksum(m, p, off, len);
5741                         }
5742                         break;
5743 #ifdef INET6
5744                 case AF_INET6:
5745                         if (m->m_len < sizeof(struct ip6_hdr))
5746                                 return (1);
5747                         sum = in6_cksum(m, p, off, len);
5748                         break;
5749 #endif /* INET6 */
5750                 default:
5751                         return (1);
5752                 }
5753         }
5754         if (sum) {
5755                 switch (p) {
5756                 case IPPROTO_TCP:
5757                     {
5758                         KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5759                         break;
5760                     }
5761                 case IPPROTO_UDP:
5762                     {
5763                         KMOD_UDPSTAT_INC(udps_badsum);
5764                         break;
5765                     }
5766 #ifdef INET
5767                 case IPPROTO_ICMP:
5768                     {
5769                         KMOD_ICMPSTAT_INC(icps_checksum);
5770                         break;
5771                     }
5772 #endif
5773 #ifdef INET6
5774                 case IPPROTO_ICMPV6:
5775                     {
5776                         KMOD_ICMP6STAT_INC(icp6s_checksum);
5777                         break;
5778                     }
5779 #endif /* INET6 */
5780                 }
5781                 return (1);
5782         } else {
5783                 if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5784                         m->m_pkthdr.csum_flags |=
5785                             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5786                         m->m_pkthdr.csum_data = 0xffff;
5787                 }
5788         }
5789         return (0);
5790 }
5791
5792 #ifdef INET
5793 int
5794 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5795 {
5796         struct pfi_kif          *kif;
5797         u_short                  action, reason = 0, log = 0;
5798         struct mbuf             *m = *m0;
5799         struct ip               *h = NULL;
5800         struct m_tag            *ipfwtag;
5801         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5802         struct pf_state         *s = NULL;
5803         struct pf_ruleset       *ruleset = NULL;
5804         struct pf_pdesc          pd;
5805         int                      off, dirndx, pqid = 0;
5806
5807         PF_RULES_RLOCK_TRACKER;
5808
5809         M_ASSERTPKTHDR(m);
5810
5811         if (!V_pf_status.running)
5812                 return (PF_PASS);
5813
5814         memset(&pd, 0, sizeof(pd));
5815
5816         kif = (struct pfi_kif *)ifp->if_pf_kif;
5817
5818         if (kif == NULL) {
5819                 DPFPRINTF(PF_DEBUG_URGENT,
5820                     ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5821                 return (PF_DROP);
5822         }
5823         if (kif->pfik_flags & PFI_IFLAG_SKIP)
5824                 return (PF_PASS);
5825
5826         if (m->m_flags & M_SKIP_FIREWALL)
5827                 return (PF_PASS);
5828
5829         pd.pf_mtag = pf_find_mtag(m);
5830
5831         PF_RULES_RLOCK();
5832
5833         if (ip_divert_ptr != NULL &&
5834             ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5835                 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5836                 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5837                         if (pd.pf_mtag == NULL &&
5838                             ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5839                                 action = PF_DROP;
5840                                 goto done;
5841                         }
5842                         pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5843                         m_tag_delete(m, ipfwtag);
5844                 }
5845                 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5846                         m->m_flags |= M_FASTFWD_OURS;
5847                         pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5848                 }
5849         } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5850                 /* We do IP header normalization and packet reassembly here */
5851                 action = PF_DROP;
5852                 goto done;
5853         }
5854         m = *m0;        /* pf_normalize messes with m0 */
5855         h = mtod(m, struct ip *);
5856
5857         off = h->ip_hl << 2;
5858         if (off < (int)sizeof(struct ip)) {
5859                 action = PF_DROP;
5860                 REASON_SET(&reason, PFRES_SHORT);
5861                 log = 1;
5862                 goto done;
5863         }
5864
5865         pd.src = (struct pf_addr *)&h->ip_src;
5866         pd.dst = (struct pf_addr *)&h->ip_dst;
5867         pd.sport = pd.dport = NULL;
5868         pd.ip_sum = &h->ip_sum;
5869         pd.proto_sum = NULL;
5870         pd.proto = h->ip_p;
5871         pd.dir = dir;
5872         pd.sidx = (dir == PF_IN) ? 0 : 1;
5873         pd.didx = (dir == PF_IN) ? 1 : 0;
5874         pd.af = AF_INET;
5875         pd.tos = h->ip_tos & ~IPTOS_ECN_MASK;
5876         pd.tot_len = ntohs(h->ip_len);
5877
5878         /* handle fragments that didn't get reassembled by normalization */
5879         if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
5880                 action = pf_test_fragment(&r, dir, kif, m, h,
5881                     &pd, &a, &ruleset);
5882                 goto done;
5883         }
5884
5885         switch (h->ip_p) {
5886         case IPPROTO_TCP: {
5887                 struct tcphdr   th;
5888
5889                 pd.hdr.tcp = &th;
5890                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
5891                     &action, &reason, AF_INET)) {
5892                         log = action != PF_PASS;
5893                         goto done;
5894                 }
5895                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
5896                 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
5897                         pqid = 1;
5898                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
5899                 if (action == PF_DROP)
5900                         goto done;
5901                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
5902                     &reason);
5903                 if (action == PF_PASS) {
5904                         if (V_pfsync_update_state_ptr != NULL)
5905                                 V_pfsync_update_state_ptr(s);
5906                         r = s->rule.ptr;
5907                         a = s->anchor.ptr;
5908                         log = s->log;
5909                 } else if (s == NULL)
5910                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5911                             &a, &ruleset, inp);
5912                 break;
5913         }
5914
5915         case IPPROTO_UDP: {
5916                 struct udphdr   uh;
5917
5918                 pd.hdr.udp = &uh;
5919                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
5920                     &action, &reason, AF_INET)) {
5921                         log = action != PF_PASS;
5922                         goto done;
5923                 }
5924                 if (uh.uh_dport == 0 ||
5925                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
5926                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
5927                         action = PF_DROP;
5928                         REASON_SET(&reason, PFRES_SHORT);
5929                         goto done;
5930                 }
5931                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
5932                 if (action == PF_PASS) {
5933                         if (V_pfsync_update_state_ptr != NULL)
5934                                 V_pfsync_update_state_ptr(s);
5935                         r = s->rule.ptr;
5936                         a = s->anchor.ptr;
5937                         log = s->log;
5938                 } else if (s == NULL)
5939                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5940                             &a, &ruleset, inp);
5941                 break;
5942         }
5943
5944         case IPPROTO_ICMP: {
5945                 struct icmp     ih;
5946
5947                 pd.hdr.icmp = &ih;
5948                 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
5949                     &action, &reason, AF_INET)) {
5950                         log = action != PF_PASS;
5951                         goto done;
5952                 }
5953                 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
5954                     &reason);
5955                 if (action == PF_PASS) {
5956                         if (V_pfsync_update_state_ptr != NULL)
5957                                 V_pfsync_update_state_ptr(s);
5958                         r = s->rule.ptr;
5959                         a = s->anchor.ptr;
5960                         log = s->log;
5961                 } else if (s == NULL)
5962                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5963                             &a, &ruleset, inp);
5964                 break;
5965         }
5966
5967 #ifdef INET6
5968         case IPPROTO_ICMPV6: {
5969                 action = PF_DROP;
5970                 DPFPRINTF(PF_DEBUG_MISC,
5971                     ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
5972                 goto done;
5973         }
5974 #endif
5975
5976         default:
5977                 action = pf_test_state_other(&s, dir, kif, m, &pd);
5978                 if (action == PF_PASS) {
5979                         if (V_pfsync_update_state_ptr != NULL)
5980                                 V_pfsync_update_state_ptr(s);
5981                         r = s->rule.ptr;
5982                         a = s->anchor.ptr;
5983                         log = s->log;
5984                 } else if (s == NULL)
5985                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
5986                             &a, &ruleset, inp);
5987                 break;
5988         }
5989
5990 done:
5991         PF_RULES_RUNLOCK();
5992         if (action == PF_PASS && h->ip_hl > 5 &&
5993             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
5994                 action = PF_DROP;
5995                 REASON_SET(&reason, PFRES_IPOPTIONS);
5996                 log = r->log;
5997                 DPFPRINTF(PF_DEBUG_MISC,
5998                     ("pf: dropping packet with ip options\n"));
5999         }
6000
6001         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6002                 action = PF_DROP;
6003                 REASON_SET(&reason, PFRES_MEMORY);
6004         }
6005         if (r->rtableid >= 0)
6006                 M_SETFIB(m, r->rtableid);
6007
6008         if (r->scrub_flags & PFSTATE_SETPRIO) {
6009                 if (pd.tos & IPTOS_LOWDELAY)
6010                         pqid = 1;
6011                 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6012                         action = PF_DROP;
6013                         REASON_SET(&reason, PFRES_MEMORY);
6014                         log = 1;
6015                         DPFPRINTF(PF_DEBUG_MISC,
6016                             ("pf: failed to allocate 802.1q mtag\n"));
6017                 }
6018         }
6019
6020 #ifdef ALTQ
6021         if (action == PF_PASS && r->qid) {
6022                 if (pd.pf_mtag == NULL &&
6023                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6024                         action = PF_DROP;
6025                         REASON_SET(&reason, PFRES_MEMORY);
6026                 } else {
6027                         if (s != NULL)
6028                                 pd.pf_mtag->qid_hash = pf_state_hash(s);
6029                         if (pqid || (pd.tos & IPTOS_LOWDELAY))
6030                                 pd.pf_mtag->qid = r->pqid;
6031                         else
6032                                 pd.pf_mtag->qid = r->qid;
6033                         /* Add hints for ecn. */
6034                         pd.pf_mtag->hdr = h;
6035                 }
6036         }
6037 #endif /* ALTQ */
6038
6039         /*
6040          * connections redirected to loopback should not match sockets
6041          * bound specifically to loopback due to security implications,
6042          * see tcp_input() and in_pcblookup_listen().
6043          */
6044         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6045             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6046             (s->nat_rule.ptr->action == PF_RDR ||
6047             s->nat_rule.ptr->action == PF_BINAT) &&
6048             IN_LOOPBACK(ntohl(pd.dst->v4.s_addr)))
6049                 m->m_flags |= M_SKIP_FIREWALL;
6050
6051         if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
6052             !PACKET_LOOPED(&pd)) {
6053                 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
6054                     sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
6055                 if (ipfwtag != NULL) {
6056                         ((struct ipfw_rule_ref *)(ipfwtag+1))->info =
6057                             ntohs(r->divert.port);
6058                         ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
6059
6060                         if (s)
6061                                 PF_STATE_UNLOCK(s);
6062
6063                         m_tag_prepend(m, ipfwtag);
6064                         if (m->m_flags & M_FASTFWD_OURS) {
6065                                 if (pd.pf_mtag == NULL &&
6066                                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6067                                         action = PF_DROP;
6068                                         REASON_SET(&reason, PFRES_MEMORY);
6069                                         log = 1;
6070                                         DPFPRINTF(PF_DEBUG_MISC,
6071                                             ("pf: failed to allocate tag\n"));
6072                                 } else {
6073                                         pd.pf_mtag->flags |=
6074                                             PF_FASTFWD_OURS_PRESENT;
6075                                         m->m_flags &= ~M_FASTFWD_OURS;
6076                                 }
6077                         }
6078                         ip_divert_ptr(*m0, dir == PF_IN);
6079                         *m0 = NULL;
6080
6081                         return (action);
6082                 } else {
6083                         /* XXX: ipfw has the same behaviour! */
6084                         action = PF_DROP;
6085                         REASON_SET(&reason, PFRES_MEMORY);
6086                         log = 1;
6087                         DPFPRINTF(PF_DEBUG_MISC,
6088                             ("pf: failed to allocate divert tag\n"));
6089                 }
6090         }
6091
6092         if (log) {
6093                 struct pf_rule *lr;
6094
6095                 if (s != NULL && s->nat_rule.ptr != NULL &&
6096                     s->nat_rule.ptr->log & PF_LOG_ALL)
6097                         lr = s->nat_rule.ptr;
6098                 else
6099                         lr = r;
6100                 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
6101                     (s == NULL));
6102         }
6103
6104         kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6105         kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
6106
6107         if (action == PF_PASS || r->action == PF_DROP) {
6108                 dirndx = (dir == PF_OUT);
6109                 r->packets[dirndx]++;
6110                 r->bytes[dirndx] += pd.tot_len;
6111                 if (a != NULL) {
6112                         a->packets[dirndx]++;
6113                         a->bytes[dirndx] += pd.tot_len;
6114                 }
6115                 if (s != NULL) {
6116                         if (s->nat_rule.ptr != NULL) {
6117                                 s->nat_rule.ptr->packets[dirndx]++;
6118                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6119                         }
6120                         if (s->src_node != NULL) {
6121                                 s->src_node->packets[dirndx]++;
6122                                 s->src_node->bytes[dirndx] += pd.tot_len;
6123                         }
6124                         if (s->nat_src_node != NULL) {
6125                                 s->nat_src_node->packets[dirndx]++;
6126                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6127                         }
6128                         dirndx = (dir == s->direction) ? 0 : 1;
6129                         s->packets[dirndx]++;
6130                         s->bytes[dirndx] += pd.tot_len;
6131                 }
6132                 tr = r;
6133                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6134                 if (nr != NULL && r == &V_pf_default_rule)
6135                         tr = nr;
6136                 if (tr->src.addr.type == PF_ADDR_TABLE)
6137                         pfr_update_stats(tr->src.addr.p.tbl,
6138                             (s == NULL) ? pd.src :
6139                             &s->key[(s->direction == PF_IN)]->
6140                                 addr[(s->direction == PF_OUT)],
6141                             pd.af, pd.tot_len, dir == PF_OUT,
6142                             r->action == PF_PASS, tr->src.neg);
6143                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6144                         pfr_update_stats(tr->dst.addr.p.tbl,
6145                             (s == NULL) ? pd.dst :
6146                             &s->key[(s->direction == PF_IN)]->
6147                                 addr[(s->direction == PF_IN)],
6148                             pd.af, pd.tot_len, dir == PF_OUT,
6149                             r->action == PF_PASS, tr->dst.neg);
6150         }
6151
6152         switch (action) {
6153         case PF_SYNPROXY_DROP:
6154                 m_freem(*m0);
6155         case PF_DEFER:
6156                 *m0 = NULL;
6157                 action = PF_PASS;
6158                 break;
6159         case PF_DROP:
6160                 m_freem(*m0);
6161                 *m0 = NULL;
6162                 break;
6163         default:
6164                 /* pf_route() returns unlocked. */
6165                 if (r->rt) {
6166                         pf_route(m0, r, dir, kif->pfik_ifp, s, &pd, inp);
6167                         return (action);
6168                 }
6169                 break;
6170         }
6171         if (s)
6172                 PF_STATE_UNLOCK(s);
6173
6174         return (action);
6175 }
6176 #endif /* INET */
6177
6178 #ifdef INET6
6179 int
6180 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6181 {
6182         struct pfi_kif          *kif;
6183         u_short                  action, reason = 0, log = 0;
6184         struct mbuf             *m = *m0, *n = NULL;
6185         struct m_tag            *mtag;
6186         struct ip6_hdr          *h = NULL;
6187         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6188         struct pf_state         *s = NULL;
6189         struct pf_ruleset       *ruleset = NULL;
6190         struct pf_pdesc          pd;
6191         int                      off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0;
6192
6193         PF_RULES_RLOCK_TRACKER;
6194         M_ASSERTPKTHDR(m);
6195
6196         if (!V_pf_status.running)
6197                 return (PF_PASS);
6198
6199         memset(&pd, 0, sizeof(pd));
6200         pd.pf_mtag = pf_find_mtag(m);
6201
6202         if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6203                 return (PF_PASS);
6204
6205         kif = (struct pfi_kif *)ifp->if_pf_kif;
6206         if (kif == NULL) {
6207                 DPFPRINTF(PF_DEBUG_URGENT,
6208                     ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6209                 return (PF_DROP);
6210         }
6211         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6212                 return (PF_PASS);
6213
6214         if (m->m_flags & M_SKIP_FIREWALL)
6215                 return (PF_PASS);
6216
6217         PF_RULES_RLOCK();
6218
6219         /* We do IP header normalization and packet reassembly here */
6220         if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6221                 action = PF_DROP;
6222                 goto done;
6223         }
6224         m = *m0;        /* pf_normalize messes with m0 */
6225         h = mtod(m, struct ip6_hdr *);
6226
6227         /*
6228          * we do not support jumbogram.  if we keep going, zero ip6_plen
6229          * will do something bad, so drop the packet for now.
6230          */
6231         if (htons(h->ip6_plen) == 0) {
6232                 action = PF_DROP;
6233                 REASON_SET(&reason, PFRES_NORM);        /*XXX*/
6234                 goto done;
6235         }
6236
6237         pd.src = (struct pf_addr *)&h->ip6_src;
6238         pd.dst = (struct pf_addr *)&h->ip6_dst;
6239         pd.sport = pd.dport = NULL;
6240         pd.ip_sum = NULL;
6241         pd.proto_sum = NULL;
6242         pd.dir = dir;
6243         pd.sidx = (dir == PF_IN) ? 0 : 1;
6244         pd.didx = (dir == PF_IN) ? 1 : 0;
6245         pd.af = AF_INET6;
6246         pd.tos = 0;
6247         pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6248
6249         off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6250         pd.proto = h->ip6_nxt;
6251         do {
6252                 switch (pd.proto) {
6253                 case IPPROTO_FRAGMENT:
6254                         action = pf_test_fragment(&r, dir, kif, m, h,
6255                             &pd, &a, &ruleset);
6256                         if (action == PF_DROP)
6257                                 REASON_SET(&reason, PFRES_FRAG);
6258                         goto done;
6259                 case IPPROTO_ROUTING: {
6260                         struct ip6_rthdr rthdr;
6261
6262                         if (rh_cnt++) {
6263                                 DPFPRINTF(PF_DEBUG_MISC,
6264                                     ("pf: IPv6 more than one rthdr\n"));
6265                                 action = PF_DROP;
6266                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6267                                 log = 1;
6268                                 goto done;
6269                         }
6270                         if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6271                             &reason, pd.af)) {
6272                                 DPFPRINTF(PF_DEBUG_MISC,
6273                                     ("pf: IPv6 short rthdr\n"));
6274                                 action = PF_DROP;
6275                                 REASON_SET(&reason, PFRES_SHORT);
6276                                 log = 1;
6277                                 goto done;
6278                         }
6279                         if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6280                                 DPFPRINTF(PF_DEBUG_MISC,
6281                                     ("pf: IPv6 rthdr0\n"));
6282                                 action = PF_DROP;
6283                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6284                                 log = 1;
6285                                 goto done;
6286                         }
6287                         /* FALLTHROUGH */
6288                 }
6289                 case IPPROTO_AH:
6290                 case IPPROTO_HOPOPTS:
6291                 case IPPROTO_DSTOPTS: {
6292                         /* get next header and header length */
6293                         struct ip6_ext  opt6;
6294
6295                         if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6296                             NULL, &reason, pd.af)) {
6297                                 DPFPRINTF(PF_DEBUG_MISC,
6298                                     ("pf: IPv6 short opt\n"));
6299                                 action = PF_DROP;
6300                                 log = 1;
6301                                 goto done;
6302                         }
6303                         if (pd.proto == IPPROTO_AH)
6304                                 off += (opt6.ip6e_len + 2) * 4;
6305                         else
6306                                 off += (opt6.ip6e_len + 1) * 8;
6307                         pd.proto = opt6.ip6e_nxt;
6308                         /* goto the next header */
6309                         break;
6310                 }
6311                 default:
6312                         terminal++;
6313                         break;
6314                 }
6315         } while (!terminal);
6316
6317         /* if there's no routing header, use unmodified mbuf for checksumming */
6318         if (!n)
6319                 n = m;
6320
6321         switch (pd.proto) {
6322         case IPPROTO_TCP: {
6323                 struct tcphdr   th;
6324
6325                 pd.hdr.tcp = &th;
6326                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6327                     &action, &reason, AF_INET6)) {
6328                         log = action != PF_PASS;
6329                         goto done;
6330                 }
6331                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6332                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6333                 if (action == PF_DROP)
6334                         goto done;
6335                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6336                     &reason);
6337                 if (action == PF_PASS) {
6338                         if (V_pfsync_update_state_ptr != NULL)
6339                                 V_pfsync_update_state_ptr(s);
6340                         r = s->rule.ptr;
6341                         a = s->anchor.ptr;
6342                         log = s->log;
6343                 } else if (s == NULL)
6344                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6345                             &a, &ruleset, inp);
6346                 break;
6347         }
6348
6349         case IPPROTO_UDP: {
6350                 struct udphdr   uh;
6351
6352                 pd.hdr.udp = &uh;
6353                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6354                     &action, &reason, AF_INET6)) {
6355                         log = action != PF_PASS;
6356                         goto done;
6357                 }
6358                 if (uh.uh_dport == 0 ||
6359                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6360                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6361                         action = PF_DROP;
6362                         REASON_SET(&reason, PFRES_SHORT);
6363                         goto done;
6364                 }
6365                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6366                 if (action == PF_PASS) {
6367                         if (V_pfsync_update_state_ptr != NULL)
6368                                 V_pfsync_update_state_ptr(s);
6369                         r = s->rule.ptr;
6370                         a = s->anchor.ptr;
6371                         log = s->log;
6372                 } else if (s == NULL)
6373                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6374                             &a, &ruleset, inp);
6375                 break;
6376         }
6377
6378         case IPPROTO_ICMP: {
6379                 action = PF_DROP;
6380                 DPFPRINTF(PF_DEBUG_MISC,
6381                     ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6382                 goto done;
6383         }
6384
6385         case IPPROTO_ICMPV6: {
6386                 struct icmp6_hdr        ih;
6387
6388                 pd.hdr.icmp6 = &ih;
6389                 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6390                     &action, &reason, AF_INET6)) {
6391                         log = action != PF_PASS;
6392                         goto done;
6393                 }
6394                 action = pf_test_state_icmp(&s, dir, kif,
6395                     m, off, h, &pd, &reason);
6396                 if (action == PF_PASS) {
6397                         if (V_pfsync_update_state_ptr != NULL)
6398                                 V_pfsync_update_state_ptr(s);
6399                         r = s->rule.ptr;
6400                         a = s->anchor.ptr;
6401                         log = s->log;
6402                 } else if (s == NULL)
6403                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6404                             &a, &ruleset, inp);
6405                 break;
6406         }
6407
6408         default:
6409                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6410                 if (action == PF_PASS) {
6411                         if (V_pfsync_update_state_ptr != NULL)
6412                                 V_pfsync_update_state_ptr(s);
6413                         r = s->rule.ptr;
6414                         a = s->anchor.ptr;
6415                         log = s->log;
6416                 } else if (s == NULL)
6417                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6418                             &a, &ruleset, inp);
6419                 break;
6420         }
6421
6422 done:
6423         PF_RULES_RUNLOCK();
6424         if (n != m) {
6425                 m_freem(n);
6426                 n = NULL;
6427         }
6428
6429         /* handle dangerous IPv6 extension headers. */
6430         if (action == PF_PASS && rh_cnt &&
6431             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6432                 action = PF_DROP;
6433                 REASON_SET(&reason, PFRES_IPOPTIONS);
6434                 log = r->log;
6435                 DPFPRINTF(PF_DEBUG_MISC,
6436                     ("pf: dropping packet with dangerous v6 headers\n"));
6437         }
6438
6439         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6440                 action = PF_DROP;
6441                 REASON_SET(&reason, PFRES_MEMORY);
6442         }
6443         if (r->rtableid >= 0)
6444                 M_SETFIB(m, r->rtableid);
6445
6446         if (r->scrub_flags & PFSTATE_SETPRIO) {
6447                 if (pd.tos & IPTOS_LOWDELAY)
6448                         pqid = 1;
6449                 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6450                         action = PF_DROP;
6451                         REASON_SET(&reason, PFRES_MEMORY);
6452                         log = 1;
6453                         DPFPRINTF(PF_DEBUG_MISC,
6454                             ("pf: failed to allocate 802.1q mtag\n"));
6455                 }
6456         }
6457
6458 #ifdef ALTQ
6459         if (action == PF_PASS && r->qid) {
6460                 if (pd.pf_mtag == NULL &&
6461                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6462                         action = PF_DROP;
6463                         REASON_SET(&reason, PFRES_MEMORY);
6464                 } else {
6465                         if (s != NULL)
6466                                 pd.pf_mtag->qid_hash = pf_state_hash(s);
6467                         if (pd.tos & IPTOS_LOWDELAY)
6468                                 pd.pf_mtag->qid = r->pqid;
6469                         else
6470                                 pd.pf_mtag->qid = r->qid;
6471                         /* Add hints for ecn. */
6472                         pd.pf_mtag->hdr = h;
6473                 }
6474         }
6475 #endif /* ALTQ */
6476
6477         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6478             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6479             (s->nat_rule.ptr->action == PF_RDR ||
6480             s->nat_rule.ptr->action == PF_BINAT) &&
6481             IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6482                 m->m_flags |= M_SKIP_FIREWALL;
6483
6484         /* XXX: Anybody working on it?! */
6485         if (r->divert.port)
6486                 printf("pf: divert(9) is not supported for IPv6\n");
6487
6488         if (log) {
6489                 struct pf_rule *lr;
6490
6491                 if (s != NULL && s->nat_rule.ptr != NULL &&
6492                     s->nat_rule.ptr->log & PF_LOG_ALL)
6493                         lr = s->nat_rule.ptr;
6494                 else
6495                         lr = r;
6496                 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6497                     &pd, (s == NULL));
6498         }
6499
6500         kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6501         kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
6502
6503         if (action == PF_PASS || r->action == PF_DROP) {
6504                 dirndx = (dir == PF_OUT);
6505                 r->packets[dirndx]++;
6506                 r->bytes[dirndx] += pd.tot_len;
6507                 if (a != NULL) {
6508                         a->packets[dirndx]++;
6509                         a->bytes[dirndx] += pd.tot_len;
6510                 }
6511                 if (s != NULL) {
6512                         if (s->nat_rule.ptr != NULL) {
6513                                 s->nat_rule.ptr->packets[dirndx]++;
6514                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6515                         }
6516                         if (s->src_node != NULL) {
6517                                 s->src_node->packets[dirndx]++;
6518                                 s->src_node->bytes[dirndx] += pd.tot_len;
6519                         }
6520                         if (s->nat_src_node != NULL) {
6521                                 s->nat_src_node->packets[dirndx]++;
6522                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6523                         }
6524                         dirndx = (dir == s->direction) ? 0 : 1;
6525                         s->packets[dirndx]++;
6526                         s->bytes[dirndx] += pd.tot_len;
6527                 }
6528                 tr = r;
6529                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6530                 if (nr != NULL && r == &V_pf_default_rule)
6531                         tr = nr;
6532                 if (tr->src.addr.type == PF_ADDR_TABLE)
6533                         pfr_update_stats(tr->src.addr.p.tbl,
6534                             (s == NULL) ? pd.src :
6535                             &s->key[(s->direction == PF_IN)]->addr[0],
6536                             pd.af, pd.tot_len, dir == PF_OUT,
6537                             r->action == PF_PASS, tr->src.neg);
6538                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6539                         pfr_update_stats(tr->dst.addr.p.tbl,
6540                             (s == NULL) ? pd.dst :
6541                             &s->key[(s->direction == PF_IN)]->addr[1],
6542                             pd.af, pd.tot_len, dir == PF_OUT,
6543                             r->action == PF_PASS, tr->dst.neg);
6544         }
6545
6546         switch (action) {
6547         case PF_SYNPROXY_DROP:
6548                 m_freem(*m0);
6549         case PF_DEFER:
6550                 *m0 = NULL;
6551                 action = PF_PASS;
6552                 break;
6553         case PF_DROP:
6554                 m_freem(*m0);
6555                 *m0 = NULL;
6556                 break;
6557         default:
6558                 /* pf_route6() returns unlocked. */
6559                 if (r->rt) {
6560                         pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd, inp);
6561                         return (action);
6562                 }
6563                 break;
6564         }
6565
6566         if (s)
6567                 PF_STATE_UNLOCK(s);
6568
6569         /* If reassembled packet passed, create new fragments. */
6570         if (action == PF_PASS && *m0 && (pflags & PFIL_FWD) &&
6571             (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL)
6572                 action = pf_refragment6(ifp, m0, mtag);
6573
6574         return (action);
6575 }
6576 #endif /* INET6 */