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/sdt.h>
  63 #include <sys/socket.h>
  64 #include <sys/sysctl.h>
  65 #include <sys/taskqueue.h>
  66 #include <sys/ucred.h>
  67
  68 #include <net/if.h>
  69 #include <net/if_var.h>
  70 #include <net/if_types.h>
  71 #include <net/if_vlan_var.h>
  72 #include <net/route.h>
  73 #include <net/route/nhop.h>
  74 #include <net/vnet.h>
  75
  76 #include <net/pfil.h>
  77 #include <net/pfvar.h>
  78 #include <net/if_pflog.h>
  79 #include <net/if_pfsync.h>
  80
  81 #include <netinet/in_pcb.h>
  82 #include <netinet/in_var.h>
  83 #include <netinet/in_fib.h>
  84 #include <netinet/ip.h>
  85 #include <netinet/ip_fw.h>
  86 #include <netinet/ip_icmp.h>
  87 #include <netinet/icmp_var.h>
  88 #include <netinet/ip_var.h>
  89 #include <netinet/tcp.h>
  90 #include <netinet/tcp_fsm.h>
  91 #include <netinet/tcp_seq.h>
  92 #include <netinet/tcp_timer.h>
  93 #include <netinet/tcp_var.h>
  94 #include <netinet/udp.h>
  95 #include <netinet/udp_var.h>
  96
  97 #ifdef INET6
  98 #include <netinet/ip6.h>
  99 #include <netinet/icmp6.h>
 100 #include <netinet6/nd6.h>
 101 #include <netinet6/ip6_var.h>
 102 #include <netinet6/in6_pcb.h>
 103 #include <netinet6/in6_fib.h>
 104 #include <netinet6/scope6_var.h>
 105 #endif /* INET6 */
 106
 107 #if defined(SCTP) || defined(SCTP_SUPPORT)
 108 #include <netinet/sctp_crc32.h>
 109 #endif
 110
 111 #include <machine/in_cksum.h>
 112 #include <security/mac/mac_framework.h>
 113
 114 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
 115
 116 SDT_PROVIDER_DEFINE(pf);
 117 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *",
 118     "struct pf_state *");
 119 SDT_PROBE_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *",
 120     "struct pf_state *");
 121 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *",
 122     "struct pf_state_key_cmp *", "int", "struct pf_pdesc *",
 123     "struct pf_state *");
 124
 125 /*
 126  * Global variables
 127  */
 128
 129 /* state tables */
 130 VNET_DEFINE(struct pf_altqqueue,         pf_altqs[4]);
 131 VNET_DEFINE(struct pf_kpalist,           pf_pabuf);
 132 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_active);
 133 VNET_DEFINE(struct pf_altqqueue *,       pf_altq_ifs_active);
 134 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_inactive);
 135 VNET_DEFINE(struct pf_altqqueue *,       pf_altq_ifs_inactive);
 136 VNET_DEFINE(struct pf_kstatus,           pf_status);
 137
 138 VNET_DEFINE(u_int32_t,                   ticket_altqs_active);
 139 VNET_DEFINE(u_int32_t,                   ticket_altqs_inactive);
 140 VNET_DEFINE(int,                         altqs_inactive_open);
 141 VNET_DEFINE(u_int32_t,                   ticket_pabuf);
 142
 143 VNET_DEFINE(MD5_CTX,                     pf_tcp_secret_ctx);
 144 #define V_pf_tcp_secret_ctx              VNET(pf_tcp_secret_ctx)
 145 VNET_DEFINE(u_char,                      pf_tcp_secret[16]);
 146 #define V_pf_tcp_secret                  VNET(pf_tcp_secret)
 147 VNET_DEFINE(int,                         pf_tcp_secret_init);
 148 #define V_pf_tcp_secret_init             VNET(pf_tcp_secret_init)
 149 VNET_DEFINE(int,                         pf_tcp_iss_off);
 150 #define V_pf_tcp_iss_off                 VNET(pf_tcp_iss_off)
 151 VNET_DECLARE(int,                        pf_vnet_active);
 152 #define V_pf_vnet_active                 VNET(pf_vnet_active)
 153
 154 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx);
 155 #define V_pf_purge_idx  VNET(pf_purge_idx)
 156
 157 /*
 158  * Queue for pf_intr() sends.
 159  */
 160 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
 161 struct pf_send_entry {
 162         STAILQ_ENTRY(pf_send_entry)     pfse_next;
 163         struct mbuf                     *pfse_m;
 164         enum {
 165                 PFSE_IP,
 166                 PFSE_IP6,
 167                 PFSE_ICMP,
 168                 PFSE_ICMP6,
 169         }                               pfse_type;
 170         struct {
 171                 int             type;
 172                 int             code;
 173                 int             mtu;
 174         } icmpopts;
 175 };
 176
 177 STAILQ_HEAD(pf_send_head, pf_send_entry);
 178 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue);
 179 #define V_pf_sendqueue  VNET(pf_sendqueue)
 180
 181 static struct mtx pf_sendqueue_mtx;
 182 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
 183 #define PF_SENDQ_LOCK()         mtx_lock(&pf_sendqueue_mtx)
 184 #define PF_SENDQ_UNLOCK()       mtx_unlock(&pf_sendqueue_mtx)
 185
 186 /*
 187  * Queue for pf_overload_task() tasks.
 188  */
 189 struct pf_overload_entry {
 190         SLIST_ENTRY(pf_overload_entry)  next;
 191         struct pf_addr                  addr;
 192         sa_family_t                     af;
 193         uint8_t                         dir;
 194         struct pf_krule                 *rule;
 195 };
 196
 197 SLIST_HEAD(pf_overload_head, pf_overload_entry);
 198 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue);
 199 #define V_pf_overloadqueue      VNET(pf_overloadqueue)
 200 VNET_DEFINE_STATIC(struct task, pf_overloadtask);
 201 #define V_pf_overloadtask       VNET(pf_overloadtask)
 202
 203 static struct mtx pf_overloadqueue_mtx;
 204 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
 205     "pf overload/flush queue", MTX_DEF);
 206 #define PF_OVERLOADQ_LOCK()     mtx_lock(&pf_overloadqueue_mtx)
 207 #define PF_OVERLOADQ_UNLOCK()   mtx_unlock(&pf_overloadqueue_mtx)
 208
 209 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules);
 210 struct mtx pf_unlnkdrules_mtx;
 211 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
 212     MTX_DEF);
 213
 214 VNET_DEFINE_STATIC(uma_zone_t,  pf_sources_z);
 215 #define V_pf_sources_z  VNET(pf_sources_z)
 216 uma_zone_t              pf_mtag_z;
 217 VNET_DEFINE(uma_zone_t,  pf_state_z);
 218 VNET_DEFINE(uma_zone_t,  pf_state_key_z);
 219
 220 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
 221 #define PFID_CPUBITS    8
 222 #define PFID_CPUSHIFT   (sizeof(uint64_t) * NBBY - PFID_CPUBITS)
 223 #define PFID_CPUMASK    ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT)
 224 #define PFID_MAXID      (~PFID_CPUMASK)
 225 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU);
 226
 227 static void              pf_src_tree_remove_state(struct pf_state *);
 228 static void              pf_init_threshold(struct pf_threshold *, u_int32_t,
 229                             u_int32_t);
 230 static void              pf_add_threshold(struct pf_threshold *);
 231 static int               pf_check_threshold(struct pf_threshold *);
 232
 233 static void              pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
 234                             u_int16_t *, u_int16_t *, struct pf_addr *,
 235                             u_int16_t, u_int8_t, sa_family_t);
 236 static int               pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
 237                             struct tcphdr *, struct pf_state_peer *);
 238 static void              pf_change_icmp(struct pf_addr *, u_int16_t *,
 239                             struct pf_addr *, struct pf_addr *, u_int16_t,
 240                             u_int16_t *, u_int16_t *, u_int16_t *,
 241                             u_int16_t *, u_int8_t, sa_family_t);
 242 static void              pf_send_tcp(struct mbuf *,
 243                             const struct pf_krule *, sa_family_t,
 244                             const struct pf_addr *, const struct pf_addr *,
 245                             u_int16_t, u_int16_t, u_int32_t, u_int32_t,
 246                             u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
 247                             u_int16_t, struct ifnet *);
 248 static void              pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
 249                             sa_family_t, struct pf_krule *);
 250 static void              pf_detach_state(struct pf_state *);
 251 static int               pf_state_key_attach(struct pf_state_key *,
 252                             struct pf_state_key *, struct pf_state *);
 253 static void              pf_state_key_detach(struct pf_state *, int);
 254 static int               pf_state_key_ctor(void *, int, void *, int);
 255 static u_int32_t         pf_tcp_iss(struct pf_pdesc *);
 256 static int               pf_test_rule(struct pf_krule **, struct pf_state **,
 257                             int, struct pfi_kkif *, struct mbuf *, int,
 258                             struct pf_pdesc *, struct pf_krule **,
 259                             struct pf_kruleset **, struct inpcb *);
 260 static int               pf_create_state(struct pf_krule *, struct pf_krule *,
 261                             struct pf_krule *, struct pf_pdesc *,
 262                             struct pf_ksrc_node *, struct pf_state_key *,
 263                             struct pf_state_key *, struct mbuf *, int,
 264                             u_int16_t, u_int16_t, int *, struct pfi_kkif *,
 265                             struct pf_state **, int, u_int16_t, u_int16_t,
 266                             int);
 267 static int               pf_test_fragment(struct pf_krule **, int,
 268                             struct pfi_kkif *, struct mbuf *, void *,
 269                             struct pf_pdesc *, struct pf_krule **,
 270                             struct pf_kruleset **);
 271 static int               pf_tcp_track_full(struct pf_state_peer *,
 272                             struct pf_state_peer *, struct pf_state **,
 273                             struct pfi_kkif *, struct mbuf *, int,
 274                             struct pf_pdesc *, u_short *, int *);
 275 static int               pf_tcp_track_sloppy(struct pf_state_peer *,
 276                             struct pf_state_peer *, struct pf_state **,
 277                             struct pf_pdesc *, u_short *);
 278 static int               pf_test_state_tcp(struct pf_state **, int,
 279                             struct pfi_kkif *, struct mbuf *, int,
 280                             void *, struct pf_pdesc *, u_short *);
 281 static int               pf_test_state_udp(struct pf_state **, int,
 282                             struct pfi_kkif *, struct mbuf *, int,
 283                             void *, struct pf_pdesc *);
 284 static int               pf_test_state_icmp(struct pf_state **, int,
 285                             struct pfi_kkif *, struct mbuf *, int,
 286                             void *, struct pf_pdesc *, u_short *);
 287 static int               pf_test_state_other(struct pf_state **, int,
 288                             struct pfi_kkif *, struct mbuf *, struct pf_pdesc *);
 289 static u_int8_t          pf_get_wscale(struct mbuf *, int, u_int16_t,
 290                             sa_family_t);
 291 static u_int16_t         pf_get_mss(struct mbuf *, int, u_int16_t,
 292                             sa_family_t);
 293 static u_int16_t         pf_calc_mss(struct pf_addr *, sa_family_t,
 294                                 int, u_int16_t);
 295 static int               pf_check_proto_cksum(struct mbuf *, int, int,
 296                             u_int8_t, sa_family_t);
 297 static void              pf_print_state_parts(struct pf_state *,
 298                             struct pf_state_key *, struct pf_state_key *);
 299 static int               pf_addr_wrap_neq(struct pf_addr_wrap *,
 300                             struct pf_addr_wrap *);
 301 static void              pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t,
 302                             bool, u_int8_t);
 303 static struct pf_state  *pf_find_state(struct pfi_kkif *,
 304                             struct pf_state_key_cmp *, u_int);
 305 static int               pf_src_connlimit(struct pf_state **);
 306 static void              pf_overload_task(void *v, int pending);
 307 static int               pf_insert_src_node(struct pf_ksrc_node **,
 308                             struct pf_krule *, struct pf_addr *, sa_family_t);
 309 static u_int             pf_purge_expired_states(u_int, int);
 310 static void              pf_purge_unlinked_rules(void);
 311 static int               pf_mtag_uminit(void *, int, int);
 312 static void              pf_mtag_free(struct m_tag *);
 313 #ifdef INET
 314 static void              pf_route(struct mbuf **, struct pf_krule *, int,
 315                             struct ifnet *, struct pf_state *,
 316                             struct pf_pdesc *, struct inpcb *);
 317 #endif /* INET */
 318 #ifdef INET6
 319 static void              pf_change_a6(struct pf_addr *, u_int16_t *,
 320                             struct pf_addr *, u_int8_t);
 321 static void              pf_route6(struct mbuf **, struct pf_krule *, int,
 322                             struct ifnet *, struct pf_state *,
 323                             struct pf_pdesc *, struct inpcb *);
 324 #endif /* INET6 */
 325
 326 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
 327
 328 extern int pf_end_threads;
 329 extern struct proc *pf_purge_proc;
 330
 331 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
 332
 333 #define PACKET_LOOPED(pd)       ((pd)->pf_mtag &&                       \
 334                                  (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
 335
 336 #define STATE_LOOKUP(i, k, d, s, pd)                                    \
 337         do {                                                            \
 338                 (s) = pf_find_state((i), (k), (d));                     \
 339                 SDT_PROBE5(pf, ip, state, lookup, i, k, d, pd, (s));    \
 340                 if ((s) == NULL)                                        \
 341                         return (PF_DROP);                               \
 342                 if (PACKET_LOOPED(pd))                                  \
 343                         return (PF_PASS);                               \
 344                 if ((d) == PF_OUT &&                                    \
 345                     (s)->rule.ptr->rt == PF_ROUTETO &&                  \
 346                     (s)->rule.ptr->direction == PF_OUT &&               \
 347                     (s)->rt_kif != NULL &&                              \
 348                     (s)->rt_kif != (i))                                 \
 349                         return (PF_PASS);                               \
 350         } while (0)
 351
 352 #define BOUND_IFACE(r, k) \
 353         ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
 354
 355 #define STATE_INC_COUNTERS(s)                                           \
 356         do {                                                            \
 357                 counter_u64_add(s->rule.ptr->states_cur, 1);            \
 358                 counter_u64_add(s->rule.ptr->states_tot, 1);            \
 359                 if (s->anchor.ptr != NULL) {                            \
 360                         counter_u64_add(s->anchor.ptr->states_cur, 1);  \
 361                         counter_u64_add(s->anchor.ptr->states_tot, 1);  \
 362                 }                                                       \
 363                 if (s->nat_rule.ptr != NULL) {                          \
 364                         counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
 365                         counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
 366                 }                                                       \
 367         } while (0)
 368
 369 #define STATE_DEC_COUNTERS(s)                                           \
 370         do {                                                            \
 371                 if (s->nat_rule.ptr != NULL)                            \
 372                         counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
 373                 if (s->anchor.ptr != NULL)                              \
 374                         counter_u64_add(s->anchor.ptr->states_cur, -1); \
 375                 counter_u64_add(s->rule.ptr->states_cur, -1);           \
 376         } while (0)
 377
 378 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
 379 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
 380 VNET_DEFINE(struct pf_idhash *, pf_idhash);
 381 VNET_DEFINE(struct pf_srchash *, pf_srchash);
 382
 383 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 384     "pf(4)");
 385
 386 u_long  pf_hashmask;
 387 u_long  pf_srchashmask;
 388 static u_long   pf_hashsize;
 389 static u_long   pf_srchashsize;
 390 u_long  pf_ioctl_maxcount = 65535;
 391
 392 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
 393     &pf_hashsize, 0, "Size of pf(4) states hashtable");
 394 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
 395     &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
 396 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN,
 397     &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call");
 398
 399 VNET_DEFINE(void *, pf_swi_cookie);
 400 VNET_DEFINE(struct intr_event *, pf_swi_ie);
 401
 402 VNET_DEFINE(uint32_t, pf_hashseed);
 403 #define V_pf_hashseed   VNET(pf_hashseed)
 404
 405 int
 406 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
 407 {
 408
 409         switch (af) {
 410 #ifdef INET
 411         case AF_INET:
 412                 if (a->addr32[0] > b->addr32[0])
 413                         return (1);
 414                 if (a->addr32[0] < b->addr32[0])
 415                         return (-1);
 416                 break;
 417 #endif /* INET */
 418 #ifdef INET6
 419         case AF_INET6:
 420                 if (a->addr32[3] > b->addr32[3])
 421                         return (1);
 422                 if (a->addr32[3] < b->addr32[3])
 423                         return (-1);
 424                 if (a->addr32[2] > b->addr32[2])
 425                         return (1);
 426                 if (a->addr32[2] < b->addr32[2])
 427                         return (-1);
 428                 if (a->addr32[1] > b->addr32[1])
 429                         return (1);
 430                 if (a->addr32[1] < b->addr32[1])
 431                         return (-1);
 432                 if (a->addr32[0] > b->addr32[0])
 433                         return (1);
 434                 if (a->addr32[0] < b->addr32[0])
 435                         return (-1);
 436                 break;
 437 #endif /* INET6 */
 438         default:
 439                 panic("%s: unknown address family %u", __func__, af);
 440         }
 441         return (0);
 442 }
 443
 444 static __inline uint32_t
 445 pf_hashkey(struct pf_state_key *sk)
 446 {
 447         uint32_t h;
 448
 449         h = murmur3_32_hash32((uint32_t *)sk,
 450             sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
 451             V_pf_hashseed);
 452
 453         return (h & pf_hashmask);
 454 }
 455
 456 static __inline uint32_t
 457 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
 458 {
 459         uint32_t h;
 460
 461         switch (af) {
 462         case AF_INET:
 463                 h = murmur3_32_hash32((uint32_t *)&addr->v4,
 464                     sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
 465                 break;
 466         case AF_INET6:
 467                 h = murmur3_32_hash32((uint32_t *)&addr->v6,
 468                     sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
 469                 break;
 470         default:
 471                 panic("%s: unknown address family %u", __func__, af);
 472         }
 473
 474         return (h & pf_srchashmask);
 475 }
 476
 477 #ifdef ALTQ
 478 static int
 479 pf_state_hash(struct pf_state *s)
 480 {
 481         u_int32_t hv = (intptr_t)s / sizeof(*s);
 482
 483         hv ^= crc32(&s->src, sizeof(s->src));
 484         hv ^= crc32(&s->dst, sizeof(s->dst));
 485         if (hv == 0)
 486                 hv = 1;
 487         return (hv);
 488 }
 489 #endif
 490
 491 #ifdef INET6
 492 void
 493 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
 494 {
 495         switch (af) {
 496 #ifdef INET
 497         case AF_INET:
 498                 dst->addr32[0] = src->addr32[0];
 499                 break;
 500 #endif /* INET */
 501         case AF_INET6:
 502                 dst->addr32[0] = src->addr32[0];
 503                 dst->addr32[1] = src->addr32[1];
 504                 dst->addr32[2] = src->addr32[2];
 505                 dst->addr32[3] = src->addr32[3];
 506                 break;
 507         }
 508 }
 509 #endif /* INET6 */
 510
 511 static void
 512 pf_init_threshold(struct pf_threshold *threshold,
 513     u_int32_t limit, u_int32_t seconds)
 514 {
 515         threshold->limit = limit * PF_THRESHOLD_MULT;
 516         threshold->seconds = seconds;
 517         threshold->count = 0;
 518         threshold->last = time_uptime;
 519 }
 520
 521 static void
 522 pf_add_threshold(struct pf_threshold *threshold)
 523 {
 524         u_int32_t t = time_uptime, diff = t - threshold->last;
 525
 526         if (diff >= threshold->seconds)
 527                 threshold->count = 0;
 528         else
 529                 threshold->count -= threshold->count * diff /
 530                     threshold->seconds;
 531         threshold->count += PF_THRESHOLD_MULT;
 532         threshold->last = t;
 533 }
 534
 535 static int
 536 pf_check_threshold(struct pf_threshold *threshold)
 537 {
 538         return (threshold->count > threshold->limit);
 539 }
 540
 541 static int
 542 pf_src_connlimit(struct pf_state **state)
 543 {
 544         struct pf_overload_entry *pfoe;
 545         int bad = 0;
 546
 547         PF_STATE_LOCK_ASSERT(*state);
 548
 549         (*state)->src_node->conn++;
 550         (*state)->src.tcp_est = 1;
 551         pf_add_threshold(&(*state)->src_node->conn_rate);
 552
 553         if ((*state)->rule.ptr->max_src_conn &&
 554             (*state)->rule.ptr->max_src_conn <
 555             (*state)->src_node->conn) {
 556                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
 557                 bad++;
 558         }
 559
 560         if ((*state)->rule.ptr->max_src_conn_rate.limit &&
 561             pf_check_threshold(&(*state)->src_node->conn_rate)) {
 562                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
 563                 bad++;
 564         }
 565
 566         if (!bad)
 567                 return (0);
 568
 569         /* Kill this state. */
 570         (*state)->timeout = PFTM_PURGE;
 571         (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
 572
 573         if ((*state)->rule.ptr->overload_tbl == NULL)
 574                 return (1);
 575
 576         /* Schedule overloading and flushing task. */
 577         pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
 578         if (pfoe == NULL)
 579                 return (1);     /* too bad :( */
 580
 581         bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
 582         pfoe->af = (*state)->key[PF_SK_WIRE]->af;
 583         pfoe->rule = (*state)->rule.ptr;
 584         pfoe->dir = (*state)->direction;
 585         PF_OVERLOADQ_LOCK();
 586         SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
 587         PF_OVERLOADQ_UNLOCK();
 588         taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
 589
 590         return (1);
 591 }
 592
 593 static void
 594 pf_overload_task(void *v, int pending)
 595 {
 596         struct pf_overload_head queue;
 597         struct pfr_addr p;
 598         struct pf_overload_entry *pfoe, *pfoe1;
 599         uint32_t killed = 0;
 600
 601         CURVNET_SET((struct vnet *)v);
 602
 603         PF_OVERLOADQ_LOCK();
 604         queue = V_pf_overloadqueue;
 605         SLIST_INIT(&V_pf_overloadqueue);
 606         PF_OVERLOADQ_UNLOCK();
 607
 608         bzero(&p, sizeof(p));
 609         SLIST_FOREACH(pfoe, &queue, next) {
 610                 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
 611                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
 612                         printf("%s: blocking address ", __func__);
 613                         pf_print_host(&pfoe->addr, 0, pfoe->af);
 614                         printf("\n");
 615                 }
 616
 617                 p.pfra_af = pfoe->af;
 618                 switch (pfoe->af) {
 619 #ifdef INET
 620                 case AF_INET:
 621                         p.pfra_net = 32;
 622                         p.pfra_ip4addr = pfoe->addr.v4;
 623                         break;
 624 #endif
 625 #ifdef INET6
 626                 case AF_INET6:
 627                         p.pfra_net = 128;
 628                         p.pfra_ip6addr = pfoe->addr.v6;
 629                         break;
 630 #endif
 631                 }
 632
 633                 PF_RULES_WLOCK();
 634                 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
 635                 PF_RULES_WUNLOCK();
 636         }
 637
 638         /*
 639          * Remove those entries, that don't need flushing.
 640          */
 641         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 642                 if (pfoe->rule->flush == 0) {
 643                         SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
 644                         free(pfoe, M_PFTEMP);
 645                 } else
 646                         counter_u64_add(
 647                             V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
 648
 649         /* If nothing to flush, return. */
 650         if (SLIST_EMPTY(&queue)) {
 651                 CURVNET_RESTORE();
 652                 return;
 653         }
 654
 655         for (int i = 0; i <= pf_hashmask; i++) {
 656                 struct pf_idhash *ih = &V_pf_idhash[i];
 657                 struct pf_state_key *sk;
 658                 struct pf_state *s;
 659
 660                 PF_HASHROW_LOCK(ih);
 661                 LIST_FOREACH(s, &ih->states, entry) {
 662                     sk = s->key[PF_SK_WIRE];
 663                     SLIST_FOREACH(pfoe, &queue, next)
 664                         if (sk->af == pfoe->af &&
 665                             ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
 666                             pfoe->rule == s->rule.ptr) &&
 667                             ((pfoe->dir == PF_OUT &&
 668                             PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
 669                             (pfoe->dir == PF_IN &&
 670                             PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
 671                                 s->timeout = PFTM_PURGE;
 672                                 s->src.state = s->dst.state = TCPS_CLOSED;
 673                                 killed++;
 674                         }
 675                 }
 676                 PF_HASHROW_UNLOCK(ih);
 677         }
 678         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 679                 free(pfoe, M_PFTEMP);
 680         if (V_pf_status.debug >= PF_DEBUG_MISC)
 681                 printf("%s: %u states killed", __func__, killed);
 682
 683         CURVNET_RESTORE();
 684 }
 685
 686 /*
 687  * Can return locked on failure, so that we can consistently
 688  * allocate and insert a new one.
 689  */
 690 struct pf_ksrc_node *
 691 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af,
 692         int returnlocked)
 693 {
 694         struct pf_srchash *sh;
 695         struct pf_ksrc_node *n;
 696
 697         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
 698
 699         sh = &V_pf_srchash[pf_hashsrc(src, af)];
 700         PF_HASHROW_LOCK(sh);
 701         LIST_FOREACH(n, &sh->nodes, entry)
 702                 if (n->rule.ptr == rule && n->af == af &&
 703                     ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
 704                     (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
 705                         break;
 706         if (n != NULL) {
 707                 n->states++;
 708                 PF_HASHROW_UNLOCK(sh);
 709         } else if (returnlocked == 0)
 710                 PF_HASHROW_UNLOCK(sh);
 711
 712         return (n);
 713 }
 714
 715 static void
 716 pf_free_src_node(struct pf_ksrc_node *sn)
 717 {
 718
 719         for (int i = 0; i < 2; i++) {
 720                 counter_u64_free(sn->bytes[i]);
 721                 counter_u64_free(sn->packets[i]);
 722         }
 723         uma_zfree(V_pf_sources_z, sn);
 724 }
 725
 726 static int
 727 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule,
 728     struct pf_addr *src, sa_family_t af)
 729 {
 730
 731         KASSERT((rule->rule_flag & PFRULE_SRCTRACK ||
 732             rule->rpool.opts & PF_POOL_STICKYADDR),
 733             ("%s for non-tracking rule %p", __func__, rule));
 734
 735         if (*sn == NULL)
 736                 *sn = pf_find_src_node(src, rule, af, 1);
 737
 738         if (*sn == NULL) {
 739                 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
 740
 741                 PF_HASHROW_ASSERT(sh);
 742
 743                 if (!rule->max_src_nodes ||
 744                     counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
 745                         (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
 746                 else
 747                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
 748                             1);
 749                 if ((*sn) == NULL) {
 750                         PF_HASHROW_UNLOCK(sh);
 751                         return (-1);
 752                 }
 753
 754                 for (int i = 0; i < 2; i++) {
 755                         (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT);
 756                         (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT);
 757
 758                         if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) {
 759                                 pf_free_src_node(*sn);
 760                                 PF_HASHROW_UNLOCK(sh);
 761                                 return (-1);
 762                         }
 763                 }
 764
 765                 pf_init_threshold(&(*sn)->conn_rate,
 766                     rule->max_src_conn_rate.limit,
 767                     rule->max_src_conn_rate.seconds);
 768
 769                 (*sn)->af = af;
 770                 (*sn)->rule.ptr = rule;
 771                 PF_ACPY(&(*sn)->addr, src, af);
 772                 LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
 773                 (*sn)->creation = time_uptime;
 774                 (*sn)->ruletype = rule->action;
 775                 (*sn)->states = 1;
 776                 if ((*sn)->rule.ptr != NULL)
 777                         counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
 778                 PF_HASHROW_UNLOCK(sh);
 779                 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
 780         } else {
 781                 if (rule->max_src_states &&
 782                     (*sn)->states >= rule->max_src_states) {
 783                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
 784                             1);
 785                         return (-1);
 786                 }
 787         }
 788         return (0);
 789 }
 790
 791 void
 792 pf_unlink_src_node(struct pf_ksrc_node *src)
 793 {
 794
 795         PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]);
 796         LIST_REMOVE(src, entry);
 797         if (src->rule.ptr)
 798                 counter_u64_add(src->rule.ptr->src_nodes, -1);
 799 }
 800
 801 u_int
 802 pf_free_src_nodes(struct pf_ksrc_node_list *head)
 803 {
 804         struct pf_ksrc_node *sn, *tmp;
 805         u_int count = 0;
 806
 807         LIST_FOREACH_SAFE(sn, head, entry, tmp) {
 808                 pf_free_src_node(sn);
 809                 count++;
 810         }
 811
 812         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
 813
 814         return (count);
 815 }
 816
 817 void
 818 pf_mtag_initialize()
 819 {
 820
 821         pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
 822             sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
 823             UMA_ALIGN_PTR, 0);
 824 }
 825
 826 /* Per-vnet data storage structures initialization. */
 827 void
 828 pf_initialize()
 829 {
 830         struct pf_keyhash       *kh;
 831         struct pf_idhash        *ih;
 832         struct pf_srchash       *sh;
 833         u_int i;
 834
 835         if (pf_hashsize == 0 || !powerof2(pf_hashsize))
 836                 pf_hashsize = PF_HASHSIZ;
 837         if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
 838                 pf_srchashsize = PF_SRCHASHSIZ;
 839
 840         V_pf_hashseed = arc4random();
 841
 842         /* States and state keys storage. */
 843         V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
 844             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 845         V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
 846         uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
 847         uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
 848
 849         V_pf_state_key_z = uma_zcreate("pf state keys",
 850             sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
 851             UMA_ALIGN_PTR, 0);
 852
 853         V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash),
 854             M_PFHASH, M_NOWAIT | M_ZERO);
 855         V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash),
 856             M_PFHASH, M_NOWAIT | M_ZERO);
 857         if (V_pf_keyhash == NULL || V_pf_idhash == NULL) {
 858                 printf("pf: Unable to allocate memory for "
 859                     "state_hashsize %lu.\n", pf_hashsize);
 860
 861                 free(V_pf_keyhash, M_PFHASH);
 862                 free(V_pf_idhash, M_PFHASH);
 863
 864                 pf_hashsize = PF_HASHSIZ;
 865                 V_pf_keyhash = mallocarray(pf_hashsize,
 866                     sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO);
 867                 V_pf_idhash = mallocarray(pf_hashsize,
 868                     sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO);
 869         }
 870
 871         pf_hashmask = pf_hashsize - 1;
 872         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 873             i++, kh++, ih++) {
 874                 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
 875                 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
 876         }
 877
 878         /* Source nodes. */
 879         V_pf_sources_z = uma_zcreate("pf source nodes",
 880             sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
 881             0);
 882         V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
 883         uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
 884         uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
 885
 886         V_pf_srchash = mallocarray(pf_srchashsize,
 887             sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO);
 888         if (V_pf_srchash == NULL) {
 889                 printf("pf: Unable to allocate memory for "
 890                     "source_hashsize %lu.\n", pf_srchashsize);
 891
 892                 pf_srchashsize = PF_SRCHASHSIZ;
 893                 V_pf_srchash = mallocarray(pf_srchashsize,
 894                     sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO);
 895         }
 896
 897         pf_srchashmask = pf_srchashsize - 1;
 898         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
 899                 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
 900
 901         /* ALTQ */
 902         TAILQ_INIT(&V_pf_altqs[0]);
 903         TAILQ_INIT(&V_pf_altqs[1]);
 904         TAILQ_INIT(&V_pf_altqs[2]);
 905         TAILQ_INIT(&V_pf_altqs[3]);
 906         TAILQ_INIT(&V_pf_pabuf);
 907         V_pf_altqs_active = &V_pf_altqs[0];
 908         V_pf_altq_ifs_active = &V_pf_altqs[1];
 909         V_pf_altqs_inactive = &V_pf_altqs[2];
 910         V_pf_altq_ifs_inactive = &V_pf_altqs[3];
 911
 912         /* Send & overload+flush queues. */
 913         STAILQ_INIT(&V_pf_sendqueue);
 914         SLIST_INIT(&V_pf_overloadqueue);
 915         TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
 916
 917         /* Unlinked, but may be referenced rules. */
 918         TAILQ_INIT(&V_pf_unlinked_rules);
 919 }
 920
 921 void
 922 pf_mtag_cleanup()
 923 {
 924
 925         uma_zdestroy(pf_mtag_z);
 926 }
 927
 928 void
 929 pf_cleanup()
 930 {
 931         struct pf_keyhash       *kh;
 932         struct pf_idhash        *ih;
 933         struct pf_srchash       *sh;
 934         struct pf_send_entry    *pfse, *next;
 935         u_int i;
 936
 937         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 938             i++, kh++, ih++) {
 939                 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
 940                     __func__));
 941                 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
 942                     __func__));
 943                 mtx_destroy(&kh->lock);
 944                 mtx_destroy(&ih->lock);
 945         }
 946         free(V_pf_keyhash, M_PFHASH);
 947         free(V_pf_idhash, M_PFHASH);
 948
 949         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
 950                 KASSERT(LIST_EMPTY(&sh->nodes),
 951                     ("%s: source node hash not empty", __func__));
 952                 mtx_destroy(&sh->lock);
 953         }
 954         free(V_pf_srchash, M_PFHASH);
 955
 956         STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
 957                 m_freem(pfse->pfse_m);
 958                 free(pfse, M_PFTEMP);
 959         }
 960
 961         uma_zdestroy(V_pf_sources_z);
 962         uma_zdestroy(V_pf_state_z);
 963         uma_zdestroy(V_pf_state_key_z);
 964 }
 965
 966 static int
 967 pf_mtag_uminit(void *mem, int size, int how)
 968 {
 969         struct m_tag *t;
 970
 971         t = (struct m_tag *)mem;
 972         t->m_tag_cookie = MTAG_ABI_COMPAT;
 973         t->m_tag_id = PACKET_TAG_PF;
 974         t->m_tag_len = sizeof(struct pf_mtag);
 975         t->m_tag_free = pf_mtag_free;
 976
 977         return (0);
 978 }
 979
 980 static void
 981 pf_mtag_free(struct m_tag *t)
 982 {
 983
 984         uma_zfree(pf_mtag_z, t);
 985 }
 986
 987 struct pf_mtag *
 988 pf_get_mtag(struct mbuf *m)
 989 {
 990         struct m_tag *mtag;
 991
 992         if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
 993                 return ((struct pf_mtag *)(mtag + 1));
 994
 995         mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
 996         if (mtag == NULL)
 997                 return (NULL);
 998         bzero(mtag + 1, sizeof(struct pf_mtag));
 999         m_tag_prepend(m, mtag);
1000
1001         return ((struct pf_mtag *)(mtag + 1));
1002 }
1003
1004 static int
1005 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
1006     struct pf_state *s)
1007 {
1008         struct pf_keyhash       *khs, *khw, *kh;
1009         struct pf_state_key     *sk, *cur;
1010         struct pf_state         *si, *olds = NULL;
1011         int idx;
1012
1013         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1014         KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
1015         KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
1016
1017         /*
1018          * We need to lock hash slots of both keys. To avoid deadlock
1019          * we always lock the slot with lower address first. Unlock order
1020          * isn't important.
1021          *
1022          * We also need to lock ID hash slot before dropping key
1023          * locks. On success we return with ID hash slot locked.
1024          */
1025
1026         if (skw == sks) {
1027                 khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
1028                 PF_HASHROW_LOCK(khs);
1029         } else {
1030                 khs = &V_pf_keyhash[pf_hashkey(sks)];
1031                 khw = &V_pf_keyhash[pf_hashkey(skw)];
1032                 if (khs == khw) {
1033                         PF_HASHROW_LOCK(khs);
1034                 } else if (khs < khw) {
1035                         PF_HASHROW_LOCK(khs);
1036                         PF_HASHROW_LOCK(khw);
1037                 } else {
1038                         PF_HASHROW_LOCK(khw);
1039                         PF_HASHROW_LOCK(khs);
1040                 }
1041         }
1042
1043 #define KEYS_UNLOCK()   do {                    \
1044         if (khs != khw) {                       \
1045                 PF_HASHROW_UNLOCK(khs);         \
1046                 PF_HASHROW_UNLOCK(khw);         \
1047         } else                                  \
1048                 PF_HASHROW_UNLOCK(khs);         \
1049 } while (0)
1050
1051         /*
1052          * First run: start with wire key.
1053          */
1054         sk = skw;
1055         kh = khw;
1056         idx = PF_SK_WIRE;
1057
1058 keyattach:
1059         LIST_FOREACH(cur, &kh->keys, entry)
1060                 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
1061                         break;
1062
1063         if (cur != NULL) {
1064                 /* Key exists. Check for same kif, if none, add to key. */
1065                 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
1066                         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
1067
1068                         PF_HASHROW_LOCK(ih);
1069                         if (si->kif == s->kif &&
1070                             si->direction == s->direction) {
1071                                 if (sk->proto == IPPROTO_TCP &&
1072                                     si->src.state >= TCPS_FIN_WAIT_2 &&
1073                                     si->dst.state >= TCPS_FIN_WAIT_2) {
1074                                         /*
1075                                          * New state matches an old >FIN_WAIT_2
1076                                          * state. We can't drop key hash locks,
1077                                          * thus we can't unlink it properly.
1078                                          *
1079                                          * As a workaround we drop it into
1080                                          * TCPS_CLOSED state, schedule purge
1081                                          * ASAP and push it into the very end
1082                                          * of the slot TAILQ, so that it won't
1083                                          * conflict with our new state.
1084                                          */
1085                                         si->src.state = si->dst.state =
1086                                             TCPS_CLOSED;
1087                                         si->timeout = PFTM_PURGE;
1088                                         olds = si;
1089                                 } else {
1090                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
1091                                                 printf("pf: %s key attach "
1092                                                     "failed on %s: ",
1093                                                     (idx == PF_SK_WIRE) ?
1094                                                     "wire" : "stack",
1095                                                     s->kif->pfik_name);
1096                                                 pf_print_state_parts(s,
1097                                                     (idx == PF_SK_WIRE) ?
1098                                                     sk : NULL,
1099                                                     (idx == PF_SK_STACK) ?
1100                                                     sk : NULL);
1101                                                 printf(", existing: ");
1102                                                 pf_print_state_parts(si,
1103                                                     (idx == PF_SK_WIRE) ?
1104                                                     sk : NULL,
1105                                                     (idx == PF_SK_STACK) ?
1106                                                     sk : NULL);
1107                                                 printf("\n");
1108                                         }
1109                                         PF_HASHROW_UNLOCK(ih);
1110                                         KEYS_UNLOCK();
1111                                         uma_zfree(V_pf_state_key_z, sk);
1112                                         if (idx == PF_SK_STACK)
1113                                                 pf_detach_state(s);
1114                                         return (EEXIST); /* collision! */
1115                                 }
1116                         }
1117                         PF_HASHROW_UNLOCK(ih);
1118                 }
1119                 uma_zfree(V_pf_state_key_z, sk);
1120                 s->key[idx] = cur;
1121         } else {
1122                 LIST_INSERT_HEAD(&kh->keys, sk, entry);
1123                 s->key[idx] = sk;
1124         }
1125
1126 stateattach:
1127         /* List is sorted, if-bound states before floating. */
1128         if (s->kif == V_pfi_all)
1129                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1130         else
1131                 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1132
1133         if (olds) {
1134                 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1135                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1136                     key_list[idx]);
1137                 olds = NULL;
1138         }
1139
1140         /*
1141          * Attach done. See how should we (or should not?)
1142          * attach a second key.
1143          */
1144         if (sks == skw) {
1145                 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1146                 idx = PF_SK_STACK;
1147                 sks = NULL;
1148                 goto stateattach;
1149         } else if (sks != NULL) {
1150                 /*
1151                  * Continue attaching with stack key.
1152                  */
1153                 sk = sks;
1154                 kh = khs;
1155                 idx = PF_SK_STACK;
1156                 sks = NULL;
1157                 goto keyattach;
1158         }
1159
1160         PF_STATE_LOCK(s);
1161         KEYS_UNLOCK();
1162
1163         KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1164             ("%s failure", __func__));
1165
1166         return (0);
1167 #undef  KEYS_UNLOCK
1168 }
1169
1170 static void
1171 pf_detach_state(struct pf_state *s)
1172 {
1173         struct pf_state_key *sks = s->key[PF_SK_STACK];
1174         struct pf_keyhash *kh;
1175
1176         if (sks != NULL) {
1177                 kh = &V_pf_keyhash[pf_hashkey(sks)];
1178                 PF_HASHROW_LOCK(kh);
1179                 if (s->key[PF_SK_STACK] != NULL)
1180                         pf_state_key_detach(s, PF_SK_STACK);
1181                 /*
1182                  * If both point to same key, then we are done.
1183                  */
1184                 if (sks == s->key[PF_SK_WIRE]) {
1185                         pf_state_key_detach(s, PF_SK_WIRE);
1186                         PF_HASHROW_UNLOCK(kh);
1187                         return;
1188                 }
1189                 PF_HASHROW_UNLOCK(kh);
1190         }
1191
1192         if (s->key[PF_SK_WIRE] != NULL) {
1193                 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1194                 PF_HASHROW_LOCK(kh);
1195                 if (s->key[PF_SK_WIRE] != NULL)
1196                         pf_state_key_detach(s, PF_SK_WIRE);
1197                 PF_HASHROW_UNLOCK(kh);
1198         }
1199 }
1200
1201 static void
1202 pf_state_key_detach(struct pf_state *s, int idx)
1203 {
1204         struct pf_state_key *sk = s->key[idx];
1205 #ifdef INVARIANTS
1206         struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1207
1208         PF_HASHROW_ASSERT(kh);
1209 #endif
1210         TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1211         s->key[idx] = NULL;
1212
1213         if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1214                 LIST_REMOVE(sk, entry);
1215                 uma_zfree(V_pf_state_key_z, sk);
1216         }
1217 }
1218
1219 static int
1220 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1221 {
1222         struct pf_state_key *sk = mem;
1223
1224         bzero(sk, sizeof(struct pf_state_key_cmp));
1225         TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1226         TAILQ_INIT(&sk->states[PF_SK_STACK]);
1227
1228         return (0);
1229 }
1230
1231 struct pf_state_key *
1232 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1233         struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1234 {
1235         struct pf_state_key *sk;
1236
1237         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1238         if (sk == NULL)
1239                 return (NULL);
1240
1241         PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1242         PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1243         sk->port[pd->sidx] = sport;
1244         sk->port[pd->didx] = dport;
1245         sk->proto = pd->proto;
1246         sk->af = pd->af;
1247
1248         return (sk);
1249 }
1250
1251 struct pf_state_key *
1252 pf_state_key_clone(struct pf_state_key *orig)
1253 {
1254         struct pf_state_key *sk;
1255
1256         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1257         if (sk == NULL)
1258                 return (NULL);
1259
1260         bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1261
1262         return (sk);
1263 }
1264
1265 int
1266 pf_state_insert(struct pfi_kkif *kif, struct pf_state_key *skw,
1267     struct pf_state_key *sks, struct pf_state *s)
1268 {
1269         struct pf_idhash *ih;
1270         struct pf_state *cur;
1271         int error;
1272
1273         KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1274             ("%s: sks not pristine", __func__));
1275         KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1276             ("%s: skw not pristine", __func__));
1277         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1278
1279         s->kif = kif;
1280
1281         if (s->id == 0 && s->creatorid == 0) {
1282                 /* XXX: should be atomic, but probability of collision low */
1283                 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1284                         V_pf_stateid[curcpu] = 1;
1285                 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1286                 s->id = htobe64(s->id);
1287                 s->creatorid = V_pf_status.hostid;
1288         }
1289
1290         /* Returns with ID locked on success. */
1291         if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1292                 return (error);
1293
1294         ih = &V_pf_idhash[PF_IDHASH(s)];
1295         PF_HASHROW_ASSERT(ih);
1296         LIST_FOREACH(cur, &ih->states, entry)
1297                 if (cur->id == s->id && cur->creatorid == s->creatorid)
1298                         break;
1299
1300         if (cur != NULL) {
1301                 PF_HASHROW_UNLOCK(ih);
1302                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1303                         printf("pf: state ID collision: "
1304                             "id: %016llx creatorid: %08x\n",
1305                             (unsigned long long)be64toh(s->id),
1306                             ntohl(s->creatorid));
1307                 }
1308                 pf_detach_state(s);
1309                 return (EEXIST);
1310         }
1311         LIST_INSERT_HEAD(&ih->states, s, entry);
1312         /* One for keys, one for ID hash. */
1313         refcount_init(&s->refs, 2);
1314
1315         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1316         if (V_pfsync_insert_state_ptr != NULL)
1317                 V_pfsync_insert_state_ptr(s);
1318
1319         /* Returns locked. */
1320         return (0);
1321 }
1322
1323 /*
1324  * Find state by ID: returns with locked row on success.
1325  */
1326 struct pf_state *
1327 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1328 {
1329         struct pf_idhash *ih;
1330         struct pf_state *s;
1331
1332         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1333
1334         ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1335
1336         PF_HASHROW_LOCK(ih);
1337         LIST_FOREACH(s, &ih->states, entry)
1338                 if (s->id == id && s->creatorid == creatorid)
1339                         break;
1340
1341         if (s == NULL)
1342                 PF_HASHROW_UNLOCK(ih);
1343
1344         return (s);
1345 }
1346
1347 /*
1348  * Find state by key.
1349  * Returns with ID hash slot locked on success.
1350  */
1351 static struct pf_state *
1352 pf_find_state(struct pfi_kkif *kif, struct pf_state_key_cmp *key, u_int dir)
1353 {
1354         struct pf_keyhash       *kh;
1355         struct pf_state_key     *sk;
1356         struct pf_state         *s;
1357         int idx;
1358
1359         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1360
1361         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1362
1363         PF_HASHROW_LOCK(kh);
1364         LIST_FOREACH(sk, &kh->keys, entry)
1365                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1366                         break;
1367         if (sk == NULL) {
1368                 PF_HASHROW_UNLOCK(kh);
1369                 return (NULL);
1370         }
1371
1372         idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1373
1374         /* List is sorted, if-bound states before floating ones. */
1375         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1376                 if (s->kif == V_pfi_all || s->kif == kif) {
1377                         PF_STATE_LOCK(s);
1378                         PF_HASHROW_UNLOCK(kh);
1379                         if (s->timeout >= PFTM_MAX) {
1380                                 /*
1381                                  * State is either being processed by
1382                                  * pf_unlink_state() in an other thread, or
1383                                  * is scheduled for immediate expiry.
1384                                  */
1385                                 PF_STATE_UNLOCK(s);
1386                                 return (NULL);
1387                         }
1388                         return (s);
1389                 }
1390         PF_HASHROW_UNLOCK(kh);
1391
1392         return (NULL);
1393 }
1394
1395 struct pf_state *
1396 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1397 {
1398         struct pf_keyhash       *kh;
1399         struct pf_state_key     *sk;
1400         struct pf_state         *s, *ret = NULL;
1401         int                      idx, inout = 0;
1402
1403         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1404
1405         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1406
1407         PF_HASHROW_LOCK(kh);
1408         LIST_FOREACH(sk, &kh->keys, entry)
1409                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1410                         break;
1411         if (sk == NULL) {
1412                 PF_HASHROW_UNLOCK(kh);
1413                 return (NULL);
1414         }
1415         switch (dir) {
1416         case PF_IN:
1417                 idx = PF_SK_WIRE;
1418                 break;
1419         case PF_OUT:
1420                 idx = PF_SK_STACK;
1421                 break;
1422         case PF_INOUT:
1423                 idx = PF_SK_WIRE;
1424                 inout = 1;
1425                 break;
1426         default:
1427                 panic("%s: dir %u", __func__, dir);
1428         }
1429 second_run:
1430         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1431                 if (more == NULL) {
1432                         PF_HASHROW_UNLOCK(kh);
1433                         return (s);
1434                 }
1435
1436                 if (ret)
1437                         (*more)++;
1438                 else
1439                         ret = s;
1440         }
1441         if (inout == 1) {
1442                 inout = 0;
1443                 idx = PF_SK_STACK;
1444                 goto second_run;
1445         }
1446         PF_HASHROW_UNLOCK(kh);
1447
1448         return (ret);
1449 }
1450
1451 /* END state table stuff */
1452
1453 static void
1454 pf_send(struct pf_send_entry *pfse)
1455 {
1456
1457         PF_SENDQ_LOCK();
1458         STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1459         PF_SENDQ_UNLOCK();
1460         swi_sched(V_pf_swi_cookie, 0);
1461 }
1462
1463 void
1464 pf_intr(void *v)
1465 {
1466         struct epoch_tracker et;
1467         struct pf_send_head queue;
1468         struct pf_send_entry *pfse, *next;
1469
1470         CURVNET_SET((struct vnet *)v);
1471
1472         PF_SENDQ_LOCK();
1473         queue = V_pf_sendqueue;
1474         STAILQ_INIT(&V_pf_sendqueue);
1475         PF_SENDQ_UNLOCK();
1476
1477         NET_EPOCH_ENTER(et);
1478
1479         STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1480                 switch (pfse->pfse_type) {
1481 #ifdef INET
1482                 case PFSE_IP:
1483                         ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1484                         break;
1485                 case PFSE_ICMP:
1486                         icmp_error(pfse->pfse_m, pfse->icmpopts.type,
1487                             pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
1488                         break;
1489 #endif /* INET */
1490 #ifdef INET6
1491                 case PFSE_IP6:
1492                         ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1493                             NULL);
1494                         break;
1495                 case PFSE_ICMP6:
1496                         icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
1497                             pfse->icmpopts.code, pfse->icmpopts.mtu);
1498                         break;
1499 #endif /* INET6 */
1500                 default:
1501                         panic("%s: unknown type", __func__);
1502                 }
1503                 free(pfse, M_PFTEMP);
1504         }
1505         NET_EPOCH_EXIT(et);
1506         CURVNET_RESTORE();
1507 }
1508
1509 void
1510 pf_purge_thread(void *unused __unused)
1511 {
1512         VNET_ITERATOR_DECL(vnet_iter);
1513
1514         sx_xlock(&pf_end_lock);
1515         while (pf_end_threads == 0) {
1516                 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", hz / 10);
1517
1518                 VNET_LIST_RLOCK();
1519                 VNET_FOREACH(vnet_iter) {
1520                         CURVNET_SET(vnet_iter);
1521
1522                         /* Wait until V_pf_default_rule is initialized. */
1523                         if (V_pf_vnet_active == 0) {
1524                                 CURVNET_RESTORE();
1525                                 continue;
1526                         }
1527
1528                         /*
1529                          *  Process 1/interval fraction of the state
1530                          * table every run.
1531                          */
1532                         V_pf_purge_idx =
1533                             pf_purge_expired_states(V_pf_purge_idx, pf_hashmask /
1534                             (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1535
1536                         /*
1537                          * Purge other expired types every
1538                          * PFTM_INTERVAL seconds.
1539                          */
1540                         if (V_pf_purge_idx == 0) {
1541                                 /*
1542                                  * Order is important:
1543                                  * - states and src nodes reference rules
1544                                  * - states and rules reference kifs
1545                                  */
1546                                 pf_purge_expired_fragments();
1547                                 pf_purge_expired_src_nodes();
1548                                 pf_purge_unlinked_rules();
1549                                 pfi_kkif_purge();
1550                         }
1551                         CURVNET_RESTORE();
1552                 }
1553                 VNET_LIST_RUNLOCK();
1554         }
1555
1556         pf_end_threads++;
1557         sx_xunlock(&pf_end_lock);
1558         kproc_exit(0);
1559 }
1560
1561 void
1562 pf_unload_vnet_purge(void)
1563 {
1564
1565         /*
1566          * To cleanse up all kifs and rules we need
1567          * two runs: first one clears reference flags,
1568          * then pf_purge_expired_states() doesn't
1569          * raise them, and then second run frees.
1570          */
1571         pf_purge_unlinked_rules();
1572         pfi_kkif_purge();
1573
1574         /*
1575          * Now purge everything.
1576          */
1577         pf_purge_expired_states(0, pf_hashmask);
1578         pf_purge_fragments(UINT_MAX);
1579         pf_purge_expired_src_nodes();
1580
1581         /*
1582          * Now all kifs & rules should be unreferenced,
1583          * thus should be successfully freed.
1584          */
1585         pf_purge_unlinked_rules();
1586         pfi_kkif_purge();
1587 }
1588
1589 u_int32_t
1590 pf_state_expires(const struct pf_state *state)
1591 {
1592         u_int32_t       timeout;
1593         u_int32_t       start;
1594         u_int32_t       end;
1595         u_int32_t       states;
1596
1597         /* handle all PFTM_* > PFTM_MAX here */
1598         if (state->timeout == PFTM_PURGE)
1599                 return (time_uptime);
1600         KASSERT(state->timeout != PFTM_UNLINKED,
1601             ("pf_state_expires: timeout == PFTM_UNLINKED"));
1602         KASSERT((state->timeout < PFTM_MAX),
1603             ("pf_state_expires: timeout > PFTM_MAX"));
1604         timeout = state->rule.ptr->timeout[state->timeout];
1605         if (!timeout)
1606                 timeout = V_pf_default_rule.timeout[state->timeout];
1607         start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1608         if (start && state->rule.ptr != &V_pf_default_rule) {
1609                 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1610                 states = counter_u64_fetch(state->rule.ptr->states_cur);
1611         } else {
1612                 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1613                 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1614                 states = V_pf_status.states;
1615         }
1616         if (end && states > start && start < end) {
1617                 if (states < end) {
1618                         timeout = (u_int64_t)timeout * (end - states) /
1619                             (end - start);
1620                         return (state->expire + timeout);
1621                 }
1622                 else
1623                         return (time_uptime);
1624         }
1625         return (state->expire + timeout);
1626 }
1627
1628 void
1629 pf_purge_expired_src_nodes()
1630 {
1631         struct pf_ksrc_node_list         freelist;
1632         struct pf_srchash       *sh;
1633         struct pf_ksrc_node     *cur, *next;
1634         int i;
1635
1636         LIST_INIT(&freelist);
1637         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1638             PF_HASHROW_LOCK(sh);
1639             LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1640                 if (cur->states == 0 && cur->expire <= time_uptime) {
1641                         pf_unlink_src_node(cur);
1642                         LIST_INSERT_HEAD(&freelist, cur, entry);
1643                 } else if (cur->rule.ptr != NULL)
1644                         cur->rule.ptr->rule_flag |= PFRULE_REFS;
1645             PF_HASHROW_UNLOCK(sh);
1646         }
1647
1648         pf_free_src_nodes(&freelist);
1649
1650         V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1651 }
1652
1653 static void
1654 pf_src_tree_remove_state(struct pf_state *s)
1655 {
1656         struct pf_ksrc_node *sn;
1657         struct pf_srchash *sh;
1658         uint32_t timeout;
1659
1660         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
1661             s->rule.ptr->timeout[PFTM_SRC_NODE] :
1662             V_pf_default_rule.timeout[PFTM_SRC_NODE];
1663
1664         if (s->src_node != NULL) {
1665                 sn = s->src_node;
1666                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1667                 PF_HASHROW_LOCK(sh);
1668                 if (s->src.tcp_est)
1669                         --sn->conn;
1670                 if (--sn->states == 0)
1671                         sn->expire = time_uptime + timeout;
1672                 PF_HASHROW_UNLOCK(sh);
1673         }
1674         if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1675                 sn = s->nat_src_node;
1676                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1677                 PF_HASHROW_LOCK(sh);
1678                 if (--sn->states == 0)
1679                         sn->expire = time_uptime + timeout;
1680                 PF_HASHROW_UNLOCK(sh);
1681         }
1682         s->src_node = s->nat_src_node = NULL;
1683 }
1684
1685 /*
1686  * Unlink and potentilly free a state. Function may be
1687  * called with ID hash row locked, but always returns
1688  * unlocked, since it needs to go through key hash locking.
1689  */
1690 int
1691 pf_unlink_state(struct pf_state *s, u_int flags)
1692 {
1693         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1694
1695         if ((flags & PF_ENTER_LOCKED) == 0)
1696                 PF_HASHROW_LOCK(ih);
1697         else
1698                 PF_HASHROW_ASSERT(ih);
1699
1700         if (s->timeout == PFTM_UNLINKED) {
1701                 /*
1702                  * State is being processed
1703                  * by pf_unlink_state() in
1704                  * an other thread.
1705                  */
1706                 PF_HASHROW_UNLOCK(ih);
1707                 return (0);     /* XXXGL: undefined actually */
1708         }
1709
1710         if (s->src.state == PF_TCPS_PROXY_DST) {
1711                 /* XXX wire key the right one? */
1712                 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1713                     &s->key[PF_SK_WIRE]->addr[1],
1714                     &s->key[PF_SK_WIRE]->addr[0],
1715                     s->key[PF_SK_WIRE]->port[1],
1716                     s->key[PF_SK_WIRE]->port[0],
1717                     s->src.seqhi, s->src.seqlo + 1,
1718                     TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1719         }
1720
1721         LIST_REMOVE(s, entry);
1722         pf_src_tree_remove_state(s);
1723
1724         if (V_pfsync_delete_state_ptr != NULL)
1725                 V_pfsync_delete_state_ptr(s);
1726
1727         STATE_DEC_COUNTERS(s);
1728
1729         s->timeout = PFTM_UNLINKED;
1730
1731         PF_HASHROW_UNLOCK(ih);
1732
1733         pf_detach_state(s);
1734         /* pf_state_insert() initialises refs to 2, so we can never release the
1735          * last reference here, only in pf_release_state(). */
1736         (void)refcount_release(&s->refs);
1737
1738         return (pf_release_state(s));
1739 }
1740
1741 void
1742 pf_free_state(struct pf_state *cur)
1743 {
1744
1745         KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1746         KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1747             cur->timeout));
1748
1749         for (int i = 0; i < 2; i++) {
1750                 counter_u64_free(cur->bytes[i]);
1751                 counter_u64_free(cur->packets[i]);
1752         }
1753
1754         pf_normalize_tcp_cleanup(cur);
1755         uma_zfree(V_pf_state_z, cur);
1756         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
1757 }
1758
1759 /*
1760  * Called only from pf_purge_thread(), thus serialized.
1761  */
1762 static u_int
1763 pf_purge_expired_states(u_int i, int maxcheck)
1764 {
1765         struct pf_idhash *ih;
1766         struct pf_state *s;
1767
1768         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1769
1770         /*
1771          * Go through hash and unlink states that expire now.
1772          */
1773         while (maxcheck > 0) {
1774                 ih = &V_pf_idhash[i];
1775
1776                 /* only take the lock if we expect to do work */
1777                 if (!LIST_EMPTY(&ih->states)) {
1778 relock:
1779                         PF_HASHROW_LOCK(ih);
1780                         LIST_FOREACH(s, &ih->states, entry) {
1781                                 if (pf_state_expires(s) <= time_uptime) {
1782                                         V_pf_status.states -=
1783                                             pf_unlink_state(s, PF_ENTER_LOCKED);
1784                                         goto relock;
1785                                 }
1786                                 s->rule.ptr->rule_flag |= PFRULE_REFS;
1787                                 if (s->nat_rule.ptr != NULL)
1788                                         s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1789                                 if (s->anchor.ptr != NULL)
1790                                         s->anchor.ptr->rule_flag |= PFRULE_REFS;
1791                                 s->kif->pfik_flags |= PFI_IFLAG_REFS;
1792                                 if (s->rt_kif)
1793                                         s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1794                         }
1795                         PF_HASHROW_UNLOCK(ih);
1796                 }
1797
1798                 /* Return when we hit end of hash. */
1799                 if (++i > pf_hashmask) {
1800                         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1801                         return (0);
1802                 }
1803
1804                 maxcheck--;
1805         }
1806
1807         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1808
1809         return (i);
1810 }
1811
1812 static void
1813 pf_purge_unlinked_rules()
1814 {
1815         struct pf_krulequeue tmpq;
1816         struct pf_krule *r, *r1;
1817
1818         /*
1819          * If we have overloading task pending, then we'd
1820          * better skip purging this time. There is a tiny
1821          * probability that overloading task references
1822          * an already unlinked rule.
1823          */
1824         PF_OVERLOADQ_LOCK();
1825         if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1826                 PF_OVERLOADQ_UNLOCK();
1827                 return;
1828         }
1829         PF_OVERLOADQ_UNLOCK();
1830
1831         /*
1832          * Do naive mark-and-sweep garbage collecting of old rules.
1833          * Reference flag is raised by pf_purge_expired_states()
1834          * and pf_purge_expired_src_nodes().
1835          *
1836          * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1837          * use a temporary queue.
1838          */
1839         TAILQ_INIT(&tmpq);
1840         PF_UNLNKDRULES_LOCK();
1841         TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1842                 if (!(r->rule_flag & PFRULE_REFS)) {
1843                         TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1844                         TAILQ_INSERT_TAIL(&tmpq, r, entries);
1845                 } else
1846                         r->rule_flag &= ~PFRULE_REFS;
1847         }
1848         PF_UNLNKDRULES_UNLOCK();
1849
1850         if (!TAILQ_EMPTY(&tmpq)) {
1851                 PF_RULES_WLOCK();
1852                 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1853                         TAILQ_REMOVE(&tmpq, r, entries);
1854                         pf_free_rule(r);
1855                 }
1856                 PF_RULES_WUNLOCK();
1857         }
1858 }
1859
1860 void
1861 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1862 {
1863         switch (af) {
1864 #ifdef INET
1865         case AF_INET: {
1866                 u_int32_t a = ntohl(addr->addr32[0]);
1867                 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1868                     (a>>8)&255, a&255);
1869                 if (p) {
1870                         p = ntohs(p);
1871                         printf(":%u", p);
1872                 }
1873                 break;
1874         }
1875 #endif /* INET */
1876 #ifdef INET6
1877         case AF_INET6: {
1878                 u_int16_t b;
1879                 u_int8_t i, curstart, curend, maxstart, maxend;
1880                 curstart = curend = maxstart = maxend = 255;
1881                 for (i = 0; i < 8; i++) {
1882                         if (!addr->addr16[i]) {
1883                                 if (curstart == 255)
1884                                         curstart = i;
1885                                 curend = i;
1886                         } else {
1887                                 if ((curend - curstart) >
1888                                     (maxend - maxstart)) {
1889                                         maxstart = curstart;
1890                                         maxend = curend;
1891                                 }
1892                                 curstart = curend = 255;
1893                         }
1894                 }
1895                 if ((curend - curstart) >
1896                     (maxend - maxstart)) {
1897                         maxstart = curstart;
1898                         maxend = curend;
1899                 }
1900                 for (i = 0; i < 8; i++) {
1901                         if (i >= maxstart && i <= maxend) {
1902                                 if (i == 0)
1903                                         printf(":");
1904                                 if (i == maxend)
1905                                         printf(":");
1906                         } else {
1907                                 b = ntohs(addr->addr16[i]);
1908                                 printf("%x", b);
1909                                 if (i < 7)
1910                                         printf(":");
1911                         }
1912                 }
1913                 if (p) {
1914                         p = ntohs(p);
1915                         printf("[%u]", p);
1916                 }
1917                 break;
1918         }
1919 #endif /* INET6 */
1920         }
1921 }
1922
1923 void
1924 pf_print_state(struct pf_state *s)
1925 {
1926         pf_print_state_parts(s, NULL, NULL);
1927 }
1928
1929 static void
1930 pf_print_state_parts(struct pf_state *s,
1931     struct pf_state_key *skwp, struct pf_state_key *sksp)
1932 {
1933         struct pf_state_key *skw, *sks;
1934         u_int8_t proto, dir;
1935
1936         /* Do our best to fill these, but they're skipped if NULL */
1937         skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1938         sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1939         proto = skw ? skw->proto : (sks ? sks->proto : 0);
1940         dir = s ? s->direction : 0;
1941
1942         switch (proto) {
1943         case IPPROTO_IPV4:
1944                 printf("IPv4");
1945                 break;
1946         case IPPROTO_IPV6:
1947                 printf("IPv6");
1948                 break;
1949         case IPPROTO_TCP:
1950                 printf("TCP");
1951                 break;
1952         case IPPROTO_UDP:
1953                 printf("UDP");
1954                 break;
1955         case IPPROTO_ICMP:
1956                 printf("ICMP");
1957                 break;
1958         case IPPROTO_ICMPV6:
1959                 printf("ICMPv6");
1960                 break;
1961         default:
1962                 printf("%u", proto);
1963                 break;
1964         }
1965         switch (dir) {
1966         case PF_IN:
1967                 printf(" in");
1968                 break;
1969         case PF_OUT:
1970                 printf(" out");
1971                 break;
1972         }
1973         if (skw) {
1974                 printf(" wire: ");
1975                 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1976                 printf(" ");
1977                 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1978         }
1979         if (sks) {
1980                 printf(" stack: ");
1981                 if (sks != skw) {
1982                         pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1983                         printf(" ");
1984                         pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1985                 } else
1986                         printf("-");
1987         }
1988         if (s) {
1989                 if (proto == IPPROTO_TCP) {
1990                         printf(" [lo=%u high=%u win=%u modulator=%u",
1991                             s->src.seqlo, s->src.seqhi,
1992                             s->src.max_win, s->src.seqdiff);
1993                         if (s->src.wscale && s->dst.wscale)
1994                                 printf(" wscale=%u",
1995                                     s->src.wscale & PF_WSCALE_MASK);
1996                         printf("]");
1997                         printf(" [lo=%u high=%u win=%u modulator=%u",
1998                             s->dst.seqlo, s->dst.seqhi,
1999                             s->dst.max_win, s->dst.seqdiff);
2000                         if (s->src.wscale && s->dst.wscale)
2001                                 printf(" wscale=%u",
2002                                 s->dst.wscale & PF_WSCALE_MASK);
2003                         printf("]");
2004                 }
2005                 printf(" %u:%u", s->src.state, s->dst.state);
2006         }
2007 }
2008
2009 void
2010 pf_print_flags(u_int8_t f)
2011 {
2012         if (f)
2013                 printf(" ");
2014         if (f & TH_FIN)
2015                 printf("F");
2016         if (f & TH_SYN)
2017                 printf("S");
2018         if (f & TH_RST)
2019                 printf("R");
2020         if (f & TH_PUSH)
2021                 printf("P");
2022         if (f & TH_ACK)
2023                 printf("A");
2024         if (f & TH_URG)
2025                 printf("U");
2026         if (f & TH_ECE)
2027                 printf("E");
2028         if (f & TH_CWR)
2029                 printf("W");
2030 }
2031
2032 #define PF_SET_SKIP_STEPS(i)                                    \
2033         do {                                                    \
2034                 while (head[i] != cur) {                        \
2035                         head[i]->skip[i].ptr = cur;             \
2036                         head[i] = TAILQ_NEXT(head[i], entries); \
2037                 }                                               \
2038         } while (0)
2039
2040 void
2041 pf_calc_skip_steps(struct pf_krulequeue *rules)
2042 {
2043         struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT];
2044         int i;
2045
2046         cur = TAILQ_FIRST(rules);
2047         prev = cur;
2048         for (i = 0; i < PF_SKIP_COUNT; ++i)
2049                 head[i] = cur;
2050         while (cur != NULL) {
2051                 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
2052                         PF_SET_SKIP_STEPS(PF_SKIP_IFP);
2053                 if (cur->direction != prev->direction)
2054                         PF_SET_SKIP_STEPS(PF_SKIP_DIR);
2055                 if (cur->af != prev->af)
2056                         PF_SET_SKIP_STEPS(PF_SKIP_AF);
2057                 if (cur->proto != prev->proto)
2058                         PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
2059                 if (cur->src.neg != prev->src.neg ||
2060                     pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
2061                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
2062                 if (cur->src.port[0] != prev->src.port[0] ||
2063                     cur->src.port[1] != prev->src.port[1] ||
2064                     cur->src.port_op != prev->src.port_op)
2065                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
2066                 if (cur->dst.neg != prev->dst.neg ||
2067                     pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
2068                         PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
2069                 if (cur->dst.port[0] != prev->dst.port[0] ||
2070                     cur->dst.port[1] != prev->dst.port[1] ||
2071                     cur->dst.port_op != prev->dst.port_op)
2072                         PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
2073
2074                 prev = cur;
2075                 cur = TAILQ_NEXT(cur, entries);
2076         }
2077         for (i = 0; i < PF_SKIP_COUNT; ++i)
2078                 PF_SET_SKIP_STEPS(i);
2079 }
2080
2081 static int
2082 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
2083 {
2084         if (aw1->type != aw2->type)
2085                 return (1);
2086         switch (aw1->type) {
2087         case PF_ADDR_ADDRMASK:
2088         case PF_ADDR_RANGE:
2089                 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
2090                         return (1);
2091                 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
2092                         return (1);
2093                 return (0);
2094         case PF_ADDR_DYNIFTL:
2095                 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
2096         case PF_ADDR_NOROUTE:
2097         case PF_ADDR_URPFFAILED:
2098                 return (0);
2099         case PF_ADDR_TABLE:
2100                 return (aw1->p.tbl != aw2->p.tbl);
2101         default:
2102                 printf("invalid address type: %d\n", aw1->type);
2103                 return (1);
2104         }
2105 }
2106
2107 /**
2108  * Checksum updates are a little complicated because the checksum in the TCP/UDP
2109  * header isn't always a full checksum. In some cases (i.e. output) it's a
2110  * pseudo-header checksum, which is a partial checksum over src/dst IP
2111  * addresses, protocol number and length.
2112  *
2113  * That means we have the following cases:
2114  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2115  *      checksums, we need to update the checksum whenever we change anything.
2116  *  * Output (i.e. the checksum is a pseudo-header checksum):
2117  *      x The field being updated is src/dst address or affects the length of
2118  *      the packet. We need to update the pseudo-header checksum (note that this
2119  *      checksum is not ones' complement).
2120  *      x Some other field is being modified (e.g. src/dst port numbers): We
2121  *      don't have to update anything.
2122  **/
2123 u_int16_t
2124 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2125 {
2126         u_int32_t x;
2127
2128         x = cksum + old - new;
2129         x = (x + (x >> 16)) & 0xffff;
2130
2131         /* optimise: eliminate a branch when not udp */
2132         if (udp && cksum == 0x0000)
2133                 return cksum;
2134         if (udp && x == 0x0000)
2135                 x = 0xffff;
2136
2137         return (u_int16_t)(x);
2138 }
2139
2140 static void
2141 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi,
2142     u_int8_t udp)
2143 {
2144         u_int16_t old = htons(hi ? (*f << 8) : *f);
2145         u_int16_t new = htons(hi ? ( v << 8) :  v);
2146
2147         if (*f == v)
2148                 return;
2149
2150         *f = v;
2151
2152         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2153                 return;
2154
2155         *cksum = pf_cksum_fixup(*cksum, old, new, udp);
2156 }
2157
2158 void
2159 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v,
2160     bool hi, u_int8_t udp)
2161 {
2162         u_int8_t *fb = (u_int8_t *)f;
2163         u_int8_t *vb = (u_int8_t *)&v;
2164
2165         pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2166         pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2167 }
2168
2169 void
2170 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v,
2171     bool hi, u_int8_t udp)
2172 {
2173         u_int8_t *fb = (u_int8_t *)f;
2174         u_int8_t *vb = (u_int8_t *)&v;
2175
2176         pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2177         pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2178         pf_patch_8(m, cksum, fb++, *vb++, hi, udp);
2179         pf_patch_8(m, cksum, fb++, *vb++, !hi, udp);
2180 }
2181
2182 u_int16_t
2183 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2184         u_int16_t new, u_int8_t udp)
2185 {
2186         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2187                 return (cksum);
2188
2189         return (pf_cksum_fixup(cksum, old, new, udp));
2190 }
2191
2192 static void
2193 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2194         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2195         sa_family_t af)
2196 {
2197         struct pf_addr  ao;
2198         u_int16_t       po = *p;
2199
2200         PF_ACPY(&ao, a, af);
2201         PF_ACPY(a, an, af);
2202
2203         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2204                 *pc = ~*pc;
2205
2206         *p = pn;
2207
2208         switch (af) {
2209 #ifdef INET
2210         case AF_INET:
2211                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2212                     ao.addr16[0], an->addr16[0], 0),
2213                     ao.addr16[1], an->addr16[1], 0);
2214                 *p = pn;
2215
2216                 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2217                     ao.addr16[0], an->addr16[0], u),
2218                     ao.addr16[1], an->addr16[1], u);
2219
2220                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2221                 break;
2222 #endif /* INET */
2223 #ifdef INET6
2224         case AF_INET6:
2225                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2226                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2227                     pf_cksum_fixup(pf_cksum_fixup(*pc,
2228                     ao.addr16[0], an->addr16[0], u),
2229                     ao.addr16[1], an->addr16[1], u),
2230                     ao.addr16[2], an->addr16[2], u),
2231                     ao.addr16[3], an->addr16[3], u),
2232                     ao.addr16[4], an->addr16[4], u),
2233                     ao.addr16[5], an->addr16[5], u),
2234                     ao.addr16[6], an->addr16[6], u),
2235                     ao.addr16[7], an->addr16[7], u);
2236
2237                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2238                 break;
2239 #endif /* INET6 */
2240         }
2241
2242         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2243             CSUM_DELAY_DATA_IPV6)) {
2244                 *pc = ~*pc;
2245                 if (! *pc)
2246                         *pc = 0xffff;
2247         }
2248 }
2249
2250 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2251 void
2252 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2253 {
2254         u_int32_t       ao;
2255
2256         memcpy(&ao, a, sizeof(ao));
2257         memcpy(a, &an, sizeof(u_int32_t));
2258         *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2259             ao % 65536, an % 65536, u);
2260 }
2261
2262 void
2263 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2264 {
2265         u_int32_t       ao;
2266
2267         memcpy(&ao, a, sizeof(ao));
2268         memcpy(a, &an, sizeof(u_int32_t));
2269
2270         *c = pf_proto_cksum_fixup(m,
2271             pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2272             ao % 65536, an % 65536, udp);
2273 }
2274
2275 #ifdef INET6
2276 static void
2277 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2278 {
2279         struct pf_addr  ao;
2280
2281         PF_ACPY(&ao, a, AF_INET6);
2282         PF_ACPY(a, an, AF_INET6);
2283
2284         *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2285             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2286             pf_cksum_fixup(pf_cksum_fixup(*c,
2287             ao.addr16[0], an->addr16[0], u),
2288             ao.addr16[1], an->addr16[1], u),
2289             ao.addr16[2], an->addr16[2], u),
2290             ao.addr16[3], an->addr16[3], u),
2291             ao.addr16[4], an->addr16[4], u),
2292             ao.addr16[5], an->addr16[5], u),
2293             ao.addr16[6], an->addr16[6], u),
2294             ao.addr16[7], an->addr16[7], u);
2295 }
2296 #endif /* INET6 */
2297
2298 static void
2299 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2300     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2301     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2302 {
2303         struct pf_addr  oia, ooa;
2304
2305         PF_ACPY(&oia, ia, af);
2306         if (oa)
2307                 PF_ACPY(&ooa, oa, af);
2308
2309         /* Change inner protocol port, fix inner protocol checksum. */
2310         if (ip != NULL) {
2311                 u_int16_t       oip = *ip;
2312                 u_int32_t       opc;
2313
2314                 if (pc != NULL)
2315                         opc = *pc;
2316                 *ip = np;
2317                 if (pc != NULL)
2318                         *pc = pf_cksum_fixup(*pc, oip, *ip, u);
2319                 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2320                 if (pc != NULL)
2321                         *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2322         }
2323         /* Change inner ip address, fix inner ip and icmp checksums. */
2324         PF_ACPY(ia, na, af);
2325         switch (af) {
2326 #ifdef INET
2327         case AF_INET: {
2328                 u_int32_t        oh2c = *h2c;
2329
2330                 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2331                     oia.addr16[0], ia->addr16[0], 0),
2332                     oia.addr16[1], ia->addr16[1], 0);
2333                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2334                     oia.addr16[0], ia->addr16[0], 0),
2335                     oia.addr16[1], ia->addr16[1], 0);
2336                 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2337                 break;
2338         }
2339 #endif /* INET */
2340 #ifdef INET6
2341         case AF_INET6:
2342                 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2343                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2344                     pf_cksum_fixup(pf_cksum_fixup(*ic,
2345                     oia.addr16[0], ia->addr16[0], u),
2346                     oia.addr16[1], ia->addr16[1], u),
2347                     oia.addr16[2], ia->addr16[2], u),
2348                     oia.addr16[3], ia->addr16[3], u),
2349                     oia.addr16[4], ia->addr16[4], u),
2350                     oia.addr16[5], ia->addr16[5], u),
2351                     oia.addr16[6], ia->addr16[6], u),
2352                     oia.addr16[7], ia->addr16[7], u);
2353                 break;
2354 #endif /* INET6 */
2355         }
2356         /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2357         if (oa) {
2358                 PF_ACPY(oa, na, af);
2359                 switch (af) {
2360 #ifdef INET
2361                 case AF_INET:
2362                         *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2363                             ooa.addr16[0], oa->addr16[0], 0),
2364                             ooa.addr16[1], oa->addr16[1], 0);
2365                         break;
2366 #endif /* INET */
2367 #ifdef INET6
2368                 case AF_INET6:
2369                         *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2370                             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2371                             pf_cksum_fixup(pf_cksum_fixup(*ic,
2372                             ooa.addr16[0], oa->addr16[0], u),
2373                             ooa.addr16[1], oa->addr16[1], u),
2374                             ooa.addr16[2], oa->addr16[2], u),
2375                             ooa.addr16[3], oa->addr16[3], u),
2376                             ooa.addr16[4], oa->addr16[4], u),
2377                             ooa.addr16[5], oa->addr16[5], u),
2378                             ooa.addr16[6], oa->addr16[6], u),
2379                             ooa.addr16[7], oa->addr16[7], u);
2380                         break;
2381 #endif /* INET6 */
2382                 }
2383         }
2384 }
2385
2386 /*
2387  * Need to modulate the sequence numbers in the TCP SACK option
2388  * (credits to Krzysztof Pfaff for report and patch)
2389  */
2390 static int
2391 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2392     struct tcphdr *th, struct pf_state_peer *dst)
2393 {
2394         int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2395         u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2396         int copyback = 0, i, olen;
2397         struct sackblk sack;
2398
2399 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2400         if (hlen < TCPOLEN_SACKLEN ||
2401             !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2402                 return 0;
2403
2404         while (hlen >= TCPOLEN_SACKLEN) {
2405                 size_t startoff = opt - opts;
2406                 olen = opt[1];
2407                 switch (*opt) {
2408                 case TCPOPT_EOL:        /* FALLTHROUGH */
2409                 case TCPOPT_NOP:
2410                         opt++;
2411                         hlen--;
2412                         break;
2413                 case TCPOPT_SACK:
2414                         if (olen > hlen)
2415                                 olen = hlen;
2416                         if (olen >= TCPOLEN_SACKLEN) {
2417                                 for (i = 2; i + TCPOLEN_SACK <= olen;
2418                                     i += TCPOLEN_SACK) {
2419                                         memcpy(&sack, &opt[i], sizeof(sack));
2420                                         pf_patch_32_unaligned(m,
2421                                             &th->th_sum, &sack.start,
2422                                             htonl(ntohl(sack.start) - dst->seqdiff),
2423                                             PF_ALGNMNT(startoff),
2424                                             0);
2425                                         pf_patch_32_unaligned(m, &th->th_sum,
2426                                             &sack.end,
2427                                             htonl(ntohl(sack.end) - dst->seqdiff),
2428                                             PF_ALGNMNT(startoff),
2429                                             0);
2430                                         memcpy(&opt[i], &sack, sizeof(sack));
2431                                 }
2432                                 copyback = 1;
2433                         }
2434                         /* FALLTHROUGH */
2435                 default:
2436                         if (olen < 2)
2437                                 olen = 2;
2438                         hlen -= olen;
2439                         opt += olen;
2440                 }
2441         }
2442
2443         if (copyback)
2444                 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2445         return (copyback);
2446 }
2447
2448 static void
2449 pf_send_tcp(struct mbuf *replyto, const struct pf_krule *r, sa_family_t af,
2450     const struct pf_addr *saddr, const struct pf_addr *daddr,
2451     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2452     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2453     u_int16_t rtag, struct ifnet *ifp)
2454 {
2455         struct pf_send_entry *pfse;
2456         struct mbuf     *m;
2457         int              len, tlen;
2458 #ifdef INET
2459         struct ip       *h = NULL;
2460 #endif /* INET */
2461 #ifdef INET6
2462         struct ip6_hdr  *h6 = NULL;
2463 #endif /* INET6 */
2464         struct tcphdr   *th;
2465         char            *opt;
2466         struct pf_mtag  *pf_mtag;
2467
2468         len = 0;
2469         th = NULL;
2470
2471         /* maximum segment size tcp option */
2472         tlen = sizeof(struct tcphdr);
2473         if (mss)
2474                 tlen += 4;
2475
2476         switch (af) {
2477 #ifdef INET
2478         case AF_INET:
2479                 len = sizeof(struct ip) + tlen;
2480                 break;
2481 #endif /* INET */
2482 #ifdef INET6
2483         case AF_INET6:
2484                 len = sizeof(struct ip6_hdr) + tlen;
2485                 break;
2486 #endif /* INET6 */
2487         default:
2488                 panic("%s: unsupported af %d", __func__, af);
2489         }
2490
2491         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2492         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2493         if (pfse == NULL)
2494                 return;
2495         m = m_gethdr(M_NOWAIT, MT_DATA);
2496         if (m == NULL) {
2497                 free(pfse, M_PFTEMP);
2498                 return;
2499         }
2500 #ifdef MAC
2501         mac_netinet_firewall_send(m);
2502 #endif
2503         if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2504                 free(pfse, M_PFTEMP);
2505                 m_freem(m);
2506                 return;
2507         }
2508         if (tag)
2509                 m->m_flags |= M_SKIP_FIREWALL;
2510         pf_mtag->tag = rtag;
2511
2512         if (r != NULL && r->rtableid >= 0)
2513                 M_SETFIB(m, r->rtableid);
2514
2515 #ifdef ALTQ
2516         if (r != NULL && r->qid) {
2517                 pf_mtag->qid = r->qid;
2518
2519                 /* add hints for ecn */
2520                 pf_mtag->hdr = mtod(m, struct ip *);
2521         }
2522 #endif /* ALTQ */
2523         m->m_data += max_linkhdr;
2524         m->m_pkthdr.len = m->m_len = len;
2525         m->m_pkthdr.rcvif = NULL;
2526         bzero(m->m_data, len);
2527         switch (af) {
2528 #ifdef INET
2529         case AF_INET:
2530                 h = mtod(m, struct ip *);
2531
2532                 /* IP header fields included in the TCP checksum */
2533                 h->ip_p = IPPROTO_TCP;
2534                 h->ip_len = htons(tlen);
2535                 h->ip_src.s_addr = saddr->v4.s_addr;
2536                 h->ip_dst.s_addr = daddr->v4.s_addr;
2537
2538                 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2539                 break;
2540 #endif /* INET */
2541 #ifdef INET6
2542         case AF_INET6:
2543                 h6 = mtod(m, struct ip6_hdr *);
2544
2545                 /* IP header fields included in the TCP checksum */
2546                 h6->ip6_nxt = IPPROTO_TCP;
2547                 h6->ip6_plen = htons(tlen);
2548                 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2549                 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2550
2551                 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2552                 break;
2553 #endif /* INET6 */
2554         }
2555
2556         /* TCP header */
2557         th->th_sport = sport;
2558         th->th_dport = dport;
2559         th->th_seq = htonl(seq);
2560         th->th_ack = htonl(ack);
2561         th->th_off = tlen >> 2;
2562         th->th_flags = flags;
2563         th->th_win = htons(win);
2564
2565         if (mss) {
2566                 opt = (char *)(th + 1);
2567                 opt[0] = TCPOPT_MAXSEG;
2568                 opt[1] = 4;
2569                 HTONS(mss);
2570                 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2571         }
2572
2573         switch (af) {
2574 #ifdef INET
2575         case AF_INET:
2576                 /* TCP checksum */
2577                 th->th_sum = in_cksum(m, len);
2578
2579                 /* Finish the IP header */
2580                 h->ip_v = 4;
2581                 h->ip_hl = sizeof(*h) >> 2;
2582                 h->ip_tos = IPTOS_LOWDELAY;
2583                 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2584                 h->ip_len = htons(len);
2585                 h->ip_ttl = ttl ? ttl : V_ip_defttl;
2586                 h->ip_sum = 0;
2587
2588                 pfse->pfse_type = PFSE_IP;
2589                 break;
2590 #endif /* INET */
2591 #ifdef INET6
2592         case AF_INET6:
2593                 /* TCP checksum */
2594                 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2595                     sizeof(struct ip6_hdr), tlen);
2596
2597                 h6->ip6_vfc |= IPV6_VERSION;
2598                 h6->ip6_hlim = IPV6_DEFHLIM;
2599
2600                 pfse->pfse_type = PFSE_IP6;
2601                 break;
2602 #endif /* INET6 */
2603         }
2604         pfse->pfse_m = m;
2605         pf_send(pfse);
2606 }
2607
2608 static void
2609 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd,
2610     struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th,
2611     struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
2612     u_short *reason)
2613 {
2614         struct pf_addr  * const saddr = pd->src;
2615         struct pf_addr  * const daddr = pd->dst;
2616         sa_family_t      af = pd->af;
2617
2618         /* undo NAT changes, if they have taken place */
2619         if (nr != NULL) {
2620                 PF_ACPY(saddr, &sk->addr[pd->sidx], af);
2621                 PF_ACPY(daddr, &sk->addr[pd->didx], af);
2622                 if (pd->sport)
2623                         *pd->sport = sk->port[pd->sidx];
2624                 if (pd->dport)
2625                         *pd->dport = sk->port[pd->didx];
2626                 if (pd->proto_sum)
2627                         *pd->proto_sum = bproto_sum;
2628                 if (pd->ip_sum)
2629                         *pd->ip_sum = bip_sum;
2630                 m_copyback(m, off, hdrlen, pd->hdr.any);
2631         }
2632         if (pd->proto == IPPROTO_TCP &&
2633             ((r->rule_flag & PFRULE_RETURNRST) ||
2634             (r->rule_flag & PFRULE_RETURN)) &&
2635             !(th->th_flags & TH_RST)) {
2636                 u_int32_t        ack = ntohl(th->th_seq) + pd->p_len;
2637                 int              len = 0;
2638 #ifdef INET
2639                 struct ip       *h4;
2640 #endif
2641 #ifdef INET6
2642                 struct ip6_hdr  *h6;
2643 #endif
2644
2645                 switch (af) {
2646 #ifdef INET
2647                 case AF_INET:
2648                         h4 = mtod(m, struct ip *);
2649                         len = ntohs(h4->ip_len) - off;
2650                         break;
2651 #endif
2652 #ifdef INET6
2653                 case AF_INET6:
2654                         h6 = mtod(m, struct ip6_hdr *);
2655                         len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
2656                         break;
2657 #endif
2658                 }
2659
2660                 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
2661                         REASON_SET(reason, PFRES_PROTCKSUM);
2662                 else {
2663                         if (th->th_flags & TH_SYN)
2664                                 ack++;
2665                         if (th->th_flags & TH_FIN)
2666                                 ack++;
2667                         pf_send_tcp(m, r, af, pd->dst,
2668                                 pd->src, th->th_dport, th->th_sport,
2669                                 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
2670                                 r->return_ttl, 1, 0, kif->pfik_ifp);
2671                 }
2672         } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
2673                 r->return_icmp)
2674                 pf_send_icmp(m, r->return_icmp >> 8,
2675                         r->return_icmp & 255, af, r);
2676         else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
2677                 r->return_icmp6)
2678                 pf_send_icmp(m, r->return_icmp6 >> 8,
2679                         r->return_icmp6 & 255, af, r);
2680 }
2681
2682 static int
2683 pf_ieee8021q_setpcp(struct mbuf *m, u_int8_t prio)
2684 {
2685         struct m_tag *mtag;
2686
2687         KASSERT(prio <= PF_PRIO_MAX,
2688             ("%s with invalid pcp", __func__));
2689
2690         mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_OUT, NULL);
2691         if (mtag == NULL) {
2692                 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_OUT,
2693                     sizeof(uint8_t), M_NOWAIT);
2694                 if (mtag == NULL)
2695                         return (ENOMEM);
2696                 m_tag_prepend(m, mtag);
2697         }
2698
2699         *(uint8_t *)(mtag + 1) = prio;
2700         return (0);
2701 }
2702
2703 static int
2704 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
2705 {
2706         struct m_tag *mtag;
2707         u_int8_t mpcp;
2708
2709         mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
2710         if (mtag == NULL)
2711                 return (0);
2712
2713         if (prio == PF_PRIO_ZERO)
2714                 prio = 0;
2715
2716         mpcp = *(uint8_t *)(mtag + 1);
2717
2718         return (mpcp == prio);
2719 }
2720
2721 static void
2722 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2723     struct pf_krule *r)
2724 {
2725         struct pf_send_entry *pfse;
2726         struct mbuf *m0;
2727         struct pf_mtag *pf_mtag;
2728
2729         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2730         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2731         if (pfse == NULL)
2732                 return;
2733
2734         if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2735                 free(pfse, M_PFTEMP);
2736                 return;
2737         }
2738
2739         if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2740                 free(pfse, M_PFTEMP);
2741                 return;
2742         }
2743         /* XXX: revisit */
2744         m0->m_flags |= M_SKIP_FIREWALL;
2745
2746         if (r->rtableid >= 0)
2747                 M_SETFIB(m0, r->rtableid);
2748
2749 #ifdef ALTQ
2750         if (r->qid) {
2751                 pf_mtag->qid = r->qid;
2752                 /* add hints for ecn */
2753                 pf_mtag->hdr = mtod(m0, struct ip *);
2754         }
2755 #endif /* ALTQ */
2756
2757         switch (af) {
2758 #ifdef INET
2759         case AF_INET:
2760                 pfse->pfse_type = PFSE_ICMP;
2761                 break;
2762 #endif /* INET */
2763 #ifdef INET6
2764         case AF_INET6:
2765                 pfse->pfse_type = PFSE_ICMP6;
2766                 break;
2767 #endif /* INET6 */
2768         }
2769         pfse->pfse_m = m0;
2770         pfse->icmpopts.type = type;
2771         pfse->icmpopts.code = code;
2772         pf_send(pfse);
2773 }
2774
2775 /*
2776  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2777  * If n is 0, they match if they are equal. If n is != 0, they match if they
2778  * are different.
2779  */
2780 int
2781 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2782     struct pf_addr *b, sa_family_t af)
2783 {
2784         int     match = 0;
2785
2786         switch (af) {
2787 #ifdef INET
2788         case AF_INET:
2789                 if ((a->addr32[0] & m->addr32[0]) ==
2790                     (b->addr32[0] & m->addr32[0]))
2791                         match++;
2792                 break;
2793 #endif /* INET */
2794 #ifdef INET6
2795         case AF_INET6:
2796                 if (((a->addr32[0] & m->addr32[0]) ==
2797                      (b->addr32[0] & m->addr32[0])) &&
2798                     ((a->addr32[1] & m->addr32[1]) ==
2799                      (b->addr32[1] & m->addr32[1])) &&
2800                     ((a->addr32[2] & m->addr32[2]) ==
2801                      (b->addr32[2] & m->addr32[2])) &&
2802                     ((a->addr32[3] & m->addr32[3]) ==
2803                      (b->addr32[3] & m->addr32[3])))
2804                         match++;
2805                 break;
2806 #endif /* INET6 */
2807         }
2808         if (match) {
2809                 if (n)
2810                         return (0);
2811                 else
2812                         return (1);
2813         } else {
2814                 if (n)
2815                         return (1);
2816                 else
2817                         return (0);
2818         }
2819 }
2820
2821 /*
2822  * Return 1 if b <= a <= e, otherwise return 0.
2823  */
2824 int
2825 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2826     struct pf_addr *a, sa_family_t af)
2827 {
2828         switch (af) {
2829 #ifdef INET
2830         case AF_INET:
2831                 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
2832                     (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
2833                         return (0);
2834                 break;
2835 #endif /* INET */
2836 #ifdef INET6
2837         case AF_INET6: {
2838                 int     i;
2839
2840                 /* check a >= b */
2841                 for (i = 0; i < 4; ++i)
2842                         if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
2843                                 break;
2844                         else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
2845                                 return (0);
2846                 /* check a <= e */
2847                 for (i = 0; i < 4; ++i)
2848                         if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
2849                                 break;
2850                         else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
2851                                 return (0);
2852                 break;
2853         }
2854 #endif /* INET6 */
2855         }
2856         return (1);
2857 }
2858
2859 static int
2860 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2861 {
2862         switch (op) {
2863         case PF_OP_IRG:
2864                 return ((p > a1) && (p < a2));
2865         case PF_OP_XRG:
2866                 return ((p < a1) || (p > a2));
2867         case PF_OP_RRG:
2868                 return ((p >= a1) && (p <= a2));
2869         case PF_OP_EQ:
2870                 return (p == a1);
2871         case PF_OP_NE:
2872                 return (p != a1);
2873         case PF_OP_LT:
2874                 return (p < a1);
2875         case PF_OP_LE:
2876                 return (p <= a1);
2877         case PF_OP_GT:
2878                 return (p > a1);
2879         case PF_OP_GE:
2880                 return (p >= a1);
2881         }
2882         return (0); /* never reached */
2883 }
2884
2885 int
2886 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2887 {
2888         NTOHS(a1);
2889         NTOHS(a2);
2890         NTOHS(p);
2891         return (pf_match(op, a1, a2, p));
2892 }
2893
2894 static int
2895 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2896 {
2897         if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2898                 return (0);
2899         return (pf_match(op, a1, a2, u));
2900 }
2901
2902 static int
2903 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2904 {
2905         if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2906                 return (0);
2907         return (pf_match(op, a1, a2, g));
2908 }
2909
2910 int
2911 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag)
2912 {
2913         if (*tag == -1)
2914                 *tag = mtag;
2915
2916         return ((!r->match_tag_not && r->match_tag == *tag) ||
2917             (r->match_tag_not && r->match_tag != *tag));
2918 }
2919
2920 int
2921 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2922 {
2923
2924         KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2925
2926         if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2927                 return (ENOMEM);
2928
2929         pd->pf_mtag->tag = tag;
2930
2931         return (0);
2932 }
2933
2934 #define PF_ANCHOR_STACKSIZE     32
2935 struct pf_kanchor_stackframe {
2936         struct pf_kruleset      *rs;
2937         struct pf_krule         *r;     /* XXX: + match bit */
2938         struct pf_kanchor       *child;
2939 };
2940
2941 /*
2942  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2943  */
2944 #define PF_ANCHORSTACK_MATCH    0x00000001
2945 #define PF_ANCHORSTACK_MASK     (PF_ANCHORSTACK_MATCH)
2946
2947 #define PF_ANCHOR_MATCH(f)      ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2948 #define PF_ANCHOR_RULE(f)       (struct pf_krule *)                     \
2949                                 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2950 #define PF_ANCHOR_SET_MATCH(f)  do { (f)->r = (void *)                  \
2951                                 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2952 } while (0)
2953
2954 void
2955 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth,
2956     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
2957     int *match)
2958 {
2959         struct pf_kanchor_stackframe    *f;
2960
2961         PF_RULES_RASSERT();
2962
2963         if (match)
2964                 *match = 0;
2965         if (*depth >= PF_ANCHOR_STACKSIZE) {
2966                 printf("%s: anchor stack overflow on %s\n",
2967                     __func__, (*r)->anchor->name);
2968                 *r = TAILQ_NEXT(*r, entries);
2969                 return;
2970         } else if (*depth == 0 && a != NULL)
2971                 *a = *r;
2972         f = stack + (*depth)++;
2973         f->rs = *rs;
2974         f->r = *r;
2975         if ((*r)->anchor_wildcard) {
2976                 struct pf_kanchor_node *parent = &(*r)->anchor->children;
2977
2978                 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) {
2979                         *r = NULL;
2980                         return;
2981                 }
2982                 *rs = &f->child->ruleset;
2983         } else {
2984                 f->child = NULL;
2985                 *rs = &(*r)->anchor->ruleset;
2986         }
2987         *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2988 }
2989
2990 int
2991 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth,
2992     struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a,
2993     int *match)
2994 {
2995         struct pf_kanchor_stackframe    *f;
2996         struct pf_krule *fr;
2997         int quick = 0;
2998
2999         PF_RULES_RASSERT();
3000
3001         do {
3002                 if (*depth <= 0)
3003                         break;
3004                 f = stack + *depth - 1;
3005                 fr = PF_ANCHOR_RULE(f);
3006                 if (f->child != NULL) {
3007                         struct pf_kanchor_node *parent;
3008
3009                         /*
3010                          * This block traverses through
3011                          * a wildcard anchor.
3012                          */
3013                         parent = &fr->anchor->children;
3014                         if (match != NULL && *match) {
3015                                 /*
3016                                  * If any of "*" matched, then
3017                                  * "foo/ *" matched, mark frame
3018                                  * appropriately.
3019                                  */
3020                                 PF_ANCHOR_SET_MATCH(f);
3021                                 *match = 0;
3022                         }
3023                         f->child = RB_NEXT(pf_kanchor_node, parent, f->child);
3024                         if (f->child != NULL) {
3025                                 *rs = &f->child->ruleset;
3026                                 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
3027                                 if (*r == NULL)
3028                                         continue;
3029                                 else
3030                                         break;
3031                         }
3032                 }
3033                 (*depth)--;
3034                 if (*depth == 0 && a != NULL)
3035                         *a = NULL;
3036                 *rs = f->rs;
3037                 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
3038                         quick = fr->quick;
3039                 *r = TAILQ_NEXT(fr, entries);
3040         } while (*r == NULL);
3041
3042         return (quick);
3043 }
3044
3045 #ifdef INET6
3046 void
3047 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
3048     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
3049 {
3050         switch (af) {
3051 #ifdef INET
3052         case AF_INET:
3053                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
3054                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
3055                 break;
3056 #endif /* INET */
3057         case AF_INET6:
3058                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
3059                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
3060                 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
3061                 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
3062                 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
3063                 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
3064                 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
3065                 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
3066                 break;
3067         }
3068 }
3069
3070 void
3071 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
3072 {
3073         switch (af) {
3074 #ifdef INET
3075         case AF_INET:
3076                 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
3077                 break;
3078 #endif /* INET */
3079         case AF_INET6:
3080                 if (addr->addr32[3] == 0xffffffff) {
3081                         addr->addr32[3] = 0;
3082                         if (addr->addr32[2] == 0xffffffff) {
3083                                 addr->addr32[2] = 0;
3084                                 if (addr->addr32[1] == 0xffffffff) {
3085                                         addr->addr32[1] = 0;
3086                                         addr->addr32[0] =
3087                                             htonl(ntohl(addr->addr32[0]) + 1);
3088                                 } else
3089                                         addr->addr32[1] =
3090                                             htonl(ntohl(addr->addr32[1]) + 1);
3091                         } else
3092                                 addr->addr32[2] =
3093                                     htonl(ntohl(addr->addr32[2]) + 1);
3094                 } else
3095                         addr->addr32[3] =
3096                             htonl(ntohl(addr->addr32[3]) + 1);
3097                 break;
3098         }
3099 }
3100 #endif /* INET6 */
3101
3102 int
3103 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
3104 {
3105         struct pf_addr          *saddr, *daddr;
3106         u_int16_t                sport, dport;
3107         struct inpcbinfo        *pi;
3108         struct inpcb            *inp;
3109
3110         pd->lookup.uid = UID_MAX;
3111         pd->lookup.gid = GID_MAX;
3112
3113         switch (pd->proto) {
3114         case IPPROTO_TCP:
3115                 if (pd->hdr.tcp == NULL)
3116                         return (-1);
3117                 sport = pd->hdr.tcp->th_sport;
3118                 dport = pd->hdr.tcp->th_dport;
3119                 pi = &V_tcbinfo;
3120                 break;
3121         case IPPROTO_UDP:
3122                 if (pd->hdr.udp == NULL)
3123                         return (-1);
3124                 sport = pd->hdr.udp->uh_sport;
3125                 dport = pd->hdr.udp->uh_dport;
3126                 pi = &V_udbinfo;
3127                 break;
3128         default:
3129                 return (-1);
3130         }
3131         if (direction == PF_IN) {
3132                 saddr = pd->src;
3133                 daddr = pd->dst;
3134         } else {
3135                 u_int16_t       p;
3136
3137                 p = sport;
3138                 sport = dport;
3139                 dport = p;
3140                 saddr = pd->dst;
3141                 daddr = pd->src;
3142         }
3143         switch (pd->af) {
3144 #ifdef INET
3145         case AF_INET:
3146                 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
3147                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
3148                 if (inp == NULL) {
3149                         inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
3150                            daddr->v4, dport, INPLOOKUP_WILDCARD |
3151                            INPLOOKUP_RLOCKPCB, NULL, m);
3152                         if (inp == NULL)
3153                                 return (-1);
3154                 }
3155                 break;
3156 #endif /* INET */
3157 #ifdef INET6
3158         case AF_INET6:
3159                 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
3160                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
3161                 if (inp == NULL) {
3162                         inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
3163                             &daddr->v6, dport, INPLOOKUP_WILDCARD |
3164                             INPLOOKUP_RLOCKPCB, NULL, m);
3165                         if (inp == NULL)
3166                                 return (-1);
3167                 }
3168                 break;
3169 #endif /* INET6 */
3170
3171         default:
3172                 return (-1);
3173         }
3174         INP_RLOCK_ASSERT(inp);
3175         pd->lookup.uid = inp->inp_cred->cr_uid;
3176         pd->lookup.gid = inp->inp_cred->cr_groups[0];
3177         INP_RUNLOCK(inp);
3178
3179         return (1);
3180 }
3181
3182 static u_int8_t
3183 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3184 {
3185         int              hlen;
3186         u_int8_t         hdr[60];
3187         u_int8_t        *opt, optlen;
3188         u_int8_t         wscale = 0;
3189
3190         hlen = th_off << 2;             /* hlen <= sizeof(hdr) */
3191         if (hlen <= sizeof(struct tcphdr))
3192                 return (0);
3193         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3194                 return (0);
3195         opt = hdr + sizeof(struct tcphdr);
3196         hlen -= sizeof(struct tcphdr);
3197         while (hlen >= 3) {
3198                 switch (*opt) {
3199                 case TCPOPT_EOL:
3200                 case TCPOPT_NOP:
3201                         ++opt;
3202                         --hlen;
3203                         break;
3204                 case TCPOPT_WINDOW:
3205                         wscale = opt[2];
3206                         if (wscale > TCP_MAX_WINSHIFT)
3207                                 wscale = TCP_MAX_WINSHIFT;
3208                         wscale |= PF_WSCALE_FLAG;
3209                         /* FALLTHROUGH */
3210                 default:
3211                         optlen = opt[1];
3212                         if (optlen < 2)
3213                                 optlen = 2;
3214                         hlen -= optlen;
3215                         opt += optlen;
3216                         break;
3217                 }
3218         }
3219         return (wscale);
3220 }
3221
3222 static u_int16_t
3223 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3224 {
3225         int              hlen;
3226         u_int8_t         hdr[60];
3227         u_int8_t        *opt, optlen;
3228         u_int16_t        mss = V_tcp_mssdflt;
3229
3230         hlen = th_off << 2;     /* hlen <= sizeof(hdr) */
3231         if (hlen <= sizeof(struct tcphdr))
3232                 return (0);
3233         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3234                 return (0);
3235         opt = hdr + sizeof(struct tcphdr);
3236         hlen -= sizeof(struct tcphdr);
3237         while (hlen >= TCPOLEN_MAXSEG) {
3238                 switch (*opt) {
3239                 case TCPOPT_EOL:
3240                 case TCPOPT_NOP:
3241                         ++opt;
3242                         --hlen;
3243                         break;
3244                 case TCPOPT_MAXSEG:
3245                         bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
3246                         NTOHS(mss);
3247                         /* FALLTHROUGH */
3248                 default:
3249                         optlen = opt[1];
3250                         if (optlen < 2)
3251                                 optlen = 2;
3252                         hlen -= optlen;
3253                         opt += optlen;
3254                         break;
3255                 }
3256         }
3257         return (mss);
3258 }
3259
3260 static u_int16_t
3261 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
3262 {
3263         struct nhop_object *nh;
3264 #ifdef INET6
3265         struct in6_addr         dst6;
3266         uint32_t                scopeid;
3267 #endif /* INET6 */
3268         int                      hlen = 0;
3269         uint16_t                 mss = 0;
3270
3271         NET_EPOCH_ASSERT();
3272
3273         switch (af) {
3274 #ifdef INET
3275         case AF_INET:
3276                 hlen = sizeof(struct ip);
3277                 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0);
3278                 if (nh != NULL)
3279                         mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
3280                 break;
3281 #endif /* INET */
3282 #ifdef INET6
3283         case AF_INET6:
3284                 hlen = sizeof(struct ip6_hdr);
3285                 in6_splitscope(&addr->v6, &dst6, &scopeid);
3286                 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0);
3287                 if (nh != NULL)
3288                         mss = nh->nh_mtu - hlen - sizeof(struct tcphdr);
3289                 break;
3290 #endif /* INET6 */
3291         }
3292
3293         mss = max(V_tcp_mssdflt, mss);
3294         mss = min(mss, offer);
3295         mss = max(mss, 64);             /* sanity - at least max opt space */
3296         return (mss);
3297 }
3298
3299 static u_int32_t
3300 pf_tcp_iss(struct pf_pdesc *pd)
3301 {
3302         MD5_CTX ctx;
3303         u_int32_t digest[4];
3304
3305         if (V_pf_tcp_secret_init == 0) {
3306                 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
3307                 MD5Init(&V_pf_tcp_secret_ctx);
3308                 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
3309                     sizeof(V_pf_tcp_secret));
3310                 V_pf_tcp_secret_init = 1;
3311         }
3312
3313         ctx = V_pf_tcp_secret_ctx;
3314
3315         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
3316         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
3317         if (pd->af == AF_INET6) {
3318                 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3319                 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3320         } else {
3321                 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3322                 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3323         }
3324         MD5Final((u_char *)digest, &ctx);
3325         V_pf_tcp_iss_off += 4096;
3326 #define ISN_RANDOM_INCREMENT (4096 - 1)
3327         return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
3328             V_pf_tcp_iss_off);
3329 #undef  ISN_RANDOM_INCREMENT
3330 }
3331
3332 static int
3333 pf_test_rule(struct pf_krule **rm, struct pf_state **sm, int direction,
3334     struct pfi_kkif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
3335     struct pf_krule **am, struct pf_kruleset **rsm, struct inpcb *inp)
3336 {
3337         struct pf_krule         *nr = NULL;
3338         struct pf_addr          * const saddr = pd->src;
3339         struct pf_addr          * const daddr = pd->dst;
3340         sa_family_t              af = pd->af;
3341         struct pf_krule         *r, *a = NULL;
3342         struct pf_kruleset      *ruleset = NULL;
3343         struct pf_ksrc_node     *nsn = NULL;
3344         struct tcphdr           *th = pd->hdr.tcp;
3345         struct pf_state_key     *sk = NULL, *nk = NULL;
3346         u_short                  reason;
3347         int                      rewrite = 0, hdrlen = 0;
3348         int                      tag = -1, rtableid = -1;
3349         int                      asd = 0;
3350         int                      match = 0;
3351         int                      state_icmp = 0;
3352         u_int16_t                sport = 0, dport = 0;
3353         u_int16_t                bproto_sum = 0, bip_sum = 0;
3354         u_int8_t                 icmptype = 0, icmpcode = 0;
3355         struct pf_kanchor_stackframe    anchor_stack[PF_ANCHOR_STACKSIZE];
3356
3357         PF_RULES_RASSERT();
3358
3359         if (inp != NULL) {
3360                 INP_LOCK_ASSERT(inp);
3361                 pd->lookup.uid = inp->inp_cred->cr_uid;
3362                 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3363                 pd->lookup.done = 1;
3364         }
3365
3366         switch (pd->proto) {
3367         case IPPROTO_TCP:
3368                 sport = th->th_sport;
3369                 dport = th->th_dport;
3370                 hdrlen = sizeof(*th);
3371                 break;
3372         case IPPROTO_UDP:
3373                 sport = pd->hdr.udp->uh_sport;
3374                 dport = pd->hdr.udp->uh_dport;
3375                 hdrlen = sizeof(*pd->hdr.udp);
3376                 break;
3377 #ifdef INET
3378         case IPPROTO_ICMP:
3379                 if (pd->af != AF_INET)
3380                         break;
3381                 sport = dport = pd->hdr.icmp->icmp_id;
3382                 hdrlen = sizeof(*pd->hdr.icmp);
3383                 icmptype = pd->hdr.icmp->icmp_type;
3384                 icmpcode = pd->hdr.icmp->icmp_code;
3385
3386                 if (icmptype == ICMP_UNREACH ||
3387                     icmptype == ICMP_SOURCEQUENCH ||
3388                     icmptype == ICMP_REDIRECT ||
3389                     icmptype == ICMP_TIMXCEED ||
3390                     icmptype == ICMP_PARAMPROB)
3391                         state_icmp++;
3392                 break;
3393 #endif /* INET */
3394 #ifdef INET6
3395         case IPPROTO_ICMPV6:
3396                 if (af != AF_INET6)
3397                         break;
3398                 sport = dport = pd->hdr.icmp6->icmp6_id;
3399                 hdrlen = sizeof(*pd->hdr.icmp6);
3400                 icmptype = pd->hdr.icmp6->icmp6_type;
3401                 icmpcode = pd->hdr.icmp6->icmp6_code;
3402
3403                 if (icmptype == ICMP6_DST_UNREACH ||
3404                     icmptype == ICMP6_PACKET_TOO_BIG ||
3405                     icmptype == ICMP6_TIME_EXCEEDED ||
3406                     icmptype == ICMP6_PARAM_PROB)
3407                         state_icmp++;
3408                 break;
3409 #endif /* INET6 */
3410         default:
3411                 sport = dport = hdrlen = 0;
3412                 break;
3413         }
3414
3415         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3416
3417         /* check packet for BINAT/NAT/RDR */
3418         if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3419             &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3420                 KASSERT(sk != NULL, ("%s: null sk", __func__));
3421                 KASSERT(nk != NULL, ("%s: null nk", __func__));
3422
3423                 if (pd->ip_sum)
3424                         bip_sum = *pd->ip_sum;
3425
3426                 switch (pd->proto) {
3427                 case IPPROTO_TCP:
3428                         bproto_sum = th->th_sum;
3429                         pd->proto_sum = &th->th_sum;
3430
3431                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3432                             nk->port[pd->sidx] != sport) {
3433                                 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
3434                                     &th->th_sum, &nk->addr[pd->sidx],
3435                                     nk->port[pd->sidx], 0, af);
3436                                 pd->sport = &th->th_sport;
3437                                 sport = th->th_sport;
3438                         }
3439
3440                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3441                             nk->port[pd->didx] != dport) {
3442                                 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
3443                                     &th->th_sum, &nk->addr[pd->didx],
3444                                     nk->port[pd->didx], 0, af);
3445                                 dport = th->th_dport;
3446                                 pd->dport = &th->th_dport;
3447                         }
3448                         rewrite++;
3449                         break;
3450                 case IPPROTO_UDP:
3451                         bproto_sum = pd->hdr.udp->uh_sum;
3452                         pd->proto_sum = &pd->hdr.udp->uh_sum;
3453
3454                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3455                             nk->port[pd->sidx] != sport) {
3456                                 pf_change_ap(m, saddr, &pd->hdr.udp->uh_sport,
3457                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3458                                     &nk->addr[pd->sidx],
3459                                     nk->port[pd->sidx], 1, af);
3460                                 sport = pd->hdr.udp->uh_sport;
3461                                 pd->sport = &pd->hdr.udp->uh_sport;
3462                         }
3463
3464                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3465                             nk->port[pd->didx] != dport) {
3466                                 pf_change_ap(m, daddr, &pd->hdr.udp->uh_dport,
3467                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3468                                     &nk->addr[pd->didx],
3469                                     nk->port[pd->didx], 1, af);
3470                                 dport = pd->hdr.udp->uh_dport;
3471                                 pd->dport = &pd->hdr.udp->uh_dport;
3472                         }
3473                         rewrite++;
3474                         break;
3475 #ifdef INET
3476                 case IPPROTO_ICMP:
3477                         nk->port[0] = nk->port[1];
3478                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3479                                 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3480                                     nk->addr[pd->sidx].v4.s_addr, 0);
3481
3482                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3483                                 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3484                                     nk->addr[pd->didx].v4.s_addr, 0);
3485
3486                         if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3487                                 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3488                                     pd->hdr.icmp->icmp_cksum, sport,
3489                                     nk->port[1], 0);
3490                                 pd->hdr.icmp->icmp_id = nk->port[1];
3491                                 pd->sport = &pd->hdr.icmp->icmp_id;
3492                         }
3493                         m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3494                         break;
3495 #endif /* INET */
3496 #ifdef INET6
3497                 case IPPROTO_ICMPV6:
3498                         nk->port[0] = nk->port[1];
3499                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3500                                 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3501                                     &nk->addr[pd->sidx], 0);
3502
3503                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3504                                 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3505                                     &nk->addr[pd->didx], 0);
3506                         rewrite++;
3507                         break;
3508 #endif /* INET */
3509                 default:
3510                         switch (af) {
3511 #ifdef INET
3512                         case AF_INET:
3513                                 if (PF_ANEQ(saddr,
3514                                     &nk->addr[pd->sidx], AF_INET))
3515                                         pf_change_a(&saddr->v4.s_addr,
3516                                             pd->ip_sum,
3517                                             nk->addr[pd->sidx].v4.s_addr, 0);
3518
3519                                 if (PF_ANEQ(daddr,
3520                                     &nk->addr[pd->didx], AF_INET))
3521                                         pf_change_a(&daddr->v4.s_addr,
3522                                             pd->ip_sum,
3523                                             nk->addr[pd->didx].v4.s_addr, 0);
3524                                 break;
3525 #endif /* INET */
3526 #ifdef INET6
3527                         case AF_INET6:
3528                                 if (PF_ANEQ(saddr,
3529                                     &nk->addr[pd->sidx], AF_INET6))
3530                                         PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3531
3532                                 if (PF_ANEQ(daddr,
3533                                     &nk->addr[pd->didx], AF_INET6))
3534                                         PF_ACPY(daddr, &nk->addr[pd->didx], af);
3535                                 break;
3536 #endif /* INET */
3537                         }
3538                         break;
3539                 }
3540                 if (nr->natpass)
3541                         r = NULL;
3542                 pd->nat_rule = nr;
3543         }
3544
3545         while (r != NULL) {
3546                 counter_u64_add(r->evaluations, 1);
3547                 if (pfi_kkif_match(r->kif, kif) == r->ifnot)
3548                         r = r->skip[PF_SKIP_IFP].ptr;
3549                 else if (r->direction && r->direction != direction)
3550                         r = r->skip[PF_SKIP_DIR].ptr;
3551                 else if (r->af && r->af != af)
3552                         r = r->skip[PF_SKIP_AF].ptr;
3553                 else if (r->proto && r->proto != pd->proto)
3554                         r = r->skip[PF_SKIP_PROTO].ptr;
3555                 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3556                     r->src.neg, kif, M_GETFIB(m)))
3557                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3558                 /* tcp/udp only. port_op always 0 in other cases */
3559                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3560                     r->src.port[0], r->src.port[1], sport))
3561                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
3562                 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3563                     r->dst.neg, NULL, M_GETFIB(m)))
3564                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3565                 /* tcp/udp only. port_op always 0 in other cases */
3566                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3567                     r->dst.port[0], r->dst.port[1], dport))
3568                         r = r->skip[PF_SKIP_DST_PORT].ptr;
3569                 /* icmp only. type always 0 in other cases */
3570                 else if (r->type && r->type != icmptype + 1)
3571                         r = TAILQ_NEXT(r, entries);
3572                 /* icmp only. type always 0 in other cases */
3573                 else if (r->code && r->code != icmpcode + 1)
3574                         r = TAILQ_NEXT(r, entries);
3575                 else if (r->tos && !(r->tos == pd->tos))
3576                         r = TAILQ_NEXT(r, entries);
3577                 else if (r->rule_flag & PFRULE_FRAGMENT)
3578                         r = TAILQ_NEXT(r, entries);
3579                 else if (pd->proto == IPPROTO_TCP &&
3580                     (r->flagset & th->th_flags) != r->flags)
3581                         r = TAILQ_NEXT(r, entries);
3582                 /* tcp/udp only. uid.op always 0 in other cases */
3583                 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3584                     pf_socket_lookup(direction, pd, m), 1)) &&
3585                     !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3586                     pd->lookup.uid))
3587                         r = TAILQ_NEXT(r, entries);
3588                 /* tcp/udp only. gid.op always 0 in other cases */
3589                 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3590                     pf_socket_lookup(direction, pd, m), 1)) &&
3591                     !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3592                     pd->lookup.gid))
3593                         r = TAILQ_NEXT(r, entries);
3594                 else if (r->prio &&
3595                     !pf_match_ieee8021q_pcp(r->prio, m))
3596                         r = TAILQ_NEXT(r, entries);
3597                 else if (r->prob &&
3598                     r->prob <= arc4random())
3599                         r = TAILQ_NEXT(r, entries);
3600                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3601                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3602                         r = TAILQ_NEXT(r, entries);
3603                 else if (r->os_fingerprint != PF_OSFP_ANY &&
3604                     (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3605                     pf_osfp_fingerprint(pd, m, off, th),
3606                     r->os_fingerprint)))
3607                         r = TAILQ_NEXT(r, entries);
3608                 else {
3609                         if (r->tag)
3610                                 tag = r->tag;
3611                         if (r->rtableid >= 0)
3612                                 rtableid = r->rtableid;
3613                         if (r->anchor == NULL) {
3614                                 match = 1;
3615                                 *rm = r;
3616                                 *am = a;
3617                                 *rsm = ruleset;
3618                                 if ((*rm)->quick)
3619                                         break;
3620                                 r = TAILQ_NEXT(r, entries);
3621                         } else
3622                                 pf_step_into_anchor(anchor_stack, &asd,
3623                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3624                                     &match);
3625                 }
3626                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3627                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3628                         break;
3629         }
3630         r = *rm;
3631         a = *am;
3632         ruleset = *rsm;
3633
3634         REASON_SET(&reason, PFRES_MATCH);
3635
3636         if (r->log || (nr != NULL && nr->log)) {
3637                 if (rewrite)
3638                         m_copyback(m, off, hdrlen, pd->hdr.any);
3639                 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3640                     ruleset, pd, 1);
3641         }
3642
3643         if ((r->action == PF_DROP) &&
3644             ((r->rule_flag & PFRULE_RETURNRST) ||
3645             (r->rule_flag & PFRULE_RETURNICMP) ||
3646             (r->rule_flag & PFRULE_RETURN))) {
3647                 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum,
3648                     bip_sum, hdrlen, &reason);
3649         }
3650
3651         if (r->action == PF_DROP)
3652                 goto cleanup;
3653
3654         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3655                 REASON_SET(&reason, PFRES_MEMORY);
3656                 goto cleanup;
3657         }
3658         if (rtableid >= 0)
3659                 M_SETFIB(m, rtableid);
3660
3661         if (!state_icmp && (r->keep_state || nr != NULL ||
3662             (pd->flags & PFDESC_TCP_NORM))) {
3663                 int action;
3664                 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3665                     sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3666                     hdrlen);
3667                 if (action != PF_PASS) {
3668                         if (action == PF_DROP &&
3669                             (r->rule_flag & PFRULE_RETURN))
3670                                 pf_return(r, nr, pd, sk, off, m, th, kif,
3671                                     bproto_sum, bip_sum, hdrlen, &reason);
3672                         return (action);
3673                 }
3674         } else {
3675                 if (sk != NULL)
3676                         uma_zfree(V_pf_state_key_z, sk);
3677                 if (nk != NULL)
3678                         uma_zfree(V_pf_state_key_z, nk);
3679         }
3680
3681         /* copy back packet headers if we performed NAT operations */
3682         if (rewrite)
3683                 m_copyback(m, off, hdrlen, pd->hdr.any);
3684
3685         if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3686             direction == PF_OUT &&
3687             V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m))
3688                 /*
3689                  * We want the state created, but we dont
3690                  * want to send this in case a partner
3691                  * firewall has to know about it to allow
3692                  * replies through it.
3693                  */
3694                 return (PF_DEFER);
3695
3696         return (PF_PASS);
3697
3698 cleanup:
3699         if (sk != NULL)
3700                 uma_zfree(V_pf_state_key_z, sk);
3701         if (nk != NULL)
3702                 uma_zfree(V_pf_state_key_z, nk);
3703         return (PF_DROP);
3704 }
3705
3706 static int
3707 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a,
3708     struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk,
3709     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3710     u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_state **sm,
3711     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3712 {
3713         struct pf_state         *s = NULL;
3714         struct pf_ksrc_node     *sn = NULL;
3715         struct tcphdr           *th = pd->hdr.tcp;
3716         u_int16_t                mss = V_tcp_mssdflt;
3717         u_short                  reason;
3718
3719         /* check maximums */
3720         if (r->max_states &&
3721             (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3722                 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
3723                 REASON_SET(&reason, PFRES_MAXSTATES);
3724                 goto csfailed;
3725         }
3726         /* src node for filter rule */
3727         if ((r->rule_flag & PFRULE_SRCTRACK ||
3728             r->rpool.opts & PF_POOL_STICKYADDR) &&
3729             pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3730                 REASON_SET(&reason, PFRES_SRCLIMIT);
3731                 goto csfailed;
3732         }
3733         /* src node for translation rule */
3734         if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3735             pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3736                 REASON_SET(&reason, PFRES_SRCLIMIT);
3737                 goto csfailed;
3738         }
3739         s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3740         if (s == NULL) {
3741                 REASON_SET(&reason, PFRES_MEMORY);
3742                 goto csfailed;
3743         }
3744         for (int i = 0; i < 2; i++) {
3745                 s->bytes[i] = counter_u64_alloc(M_NOWAIT);
3746                 s->packets[i] = counter_u64_alloc(M_NOWAIT);
3747
3748                 if (s->bytes[i] == NULL || s->packets[i] == NULL) {
3749                         pf_free_state(s);
3750                         REASON_SET(&reason, PFRES_MEMORY);
3751                         goto csfailed;
3752                 }
3753         }
3754         s->rule.ptr = r;
3755         s->nat_rule.ptr = nr;
3756         s->anchor.ptr = a;
3757         STATE_INC_COUNTERS(s);
3758         if (r->allow_opts)
3759                 s->state_flags |= PFSTATE_ALLOWOPTS;
3760         if (r->rule_flag & PFRULE_STATESLOPPY)
3761                 s->state_flags |= PFSTATE_SLOPPY;
3762         s->log = r->log & PF_LOG_ALL;
3763         s->sync_state = PFSYNC_S_NONE;
3764         if (nr != NULL)
3765                 s->log |= nr->log & PF_LOG_ALL;
3766         switch (pd->proto) {
3767         case IPPROTO_TCP:
3768                 s->src.seqlo = ntohl(th->th_seq);
3769                 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3770                 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3771                     r->keep_state == PF_STATE_MODULATE) {
3772                         /* Generate sequence number modulator */
3773                         if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3774                             0)
3775                                 s->src.seqdiff = 1;
3776                         pf_change_proto_a(m, &th->th_seq, &th->th_sum,
3777                             htonl(s->src.seqlo + s->src.seqdiff), 0);
3778                         *rewrite = 1;
3779                 } else
3780                         s->src.seqdiff = 0;
3781                 if (th->th_flags & TH_SYN) {
3782                         s->src.seqhi++;
3783                         s->src.wscale = pf_get_wscale(m, off,
3784                             th->th_off, pd->af);
3785                 }
3786                 s->src.max_win = MAX(ntohs(th->th_win), 1);
3787                 if (s->src.wscale & PF_WSCALE_MASK) {
3788                         /* Remove scale factor from initial window */
3789                         int win = s->src.max_win;
3790                         win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3791                         s->src.max_win = (win - 1) >>
3792                             (s->src.wscale & PF_WSCALE_MASK);
3793                 }
3794                 if (th->th_flags & TH_FIN)
3795                         s->src.seqhi++;
3796                 s->dst.seqhi = 1;
3797                 s->dst.max_win = 1;
3798                 s->src.state = TCPS_SYN_SENT;
3799                 s->dst.state = TCPS_CLOSED;
3800                 s->timeout = PFTM_TCP_FIRST_PACKET;
3801                 break;
3802         case IPPROTO_UDP:
3803                 s->src.state = PFUDPS_SINGLE;
3804                 s->dst.state = PFUDPS_NO_TRAFFIC;
3805                 s->timeout = PFTM_UDP_FIRST_PACKET;
3806                 break;
3807         case IPPROTO_ICMP:
3808 #ifdef INET6
3809         case IPPROTO_ICMPV6:
3810 #endif
3811                 s->timeout = PFTM_ICMP_FIRST_PACKET;
3812                 break;
3813         default:
3814                 s->src.state = PFOTHERS_SINGLE;
3815                 s->dst.state = PFOTHERS_NO_TRAFFIC;
3816                 s->timeout = PFTM_OTHER_FIRST_PACKET;
3817         }
3818
3819         if (r->rt) {
3820                 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) {
3821                         REASON_SET(&reason, PFRES_MAPFAILED);
3822                         pf_src_tree_remove_state(s);
3823                         STATE_DEC_COUNTERS(s);
3824                         uma_zfree(V_pf_state_z, s);
3825                         goto csfailed;
3826                 }
3827                 s->rt_kif = r->rpool.cur->kif;
3828         }
3829
3830         s->creation = time_uptime;
3831         s->expire = time_uptime;
3832
3833         if (sn != NULL)
3834                 s->src_node = sn;
3835         if (nsn != NULL) {
3836                 /* XXX We only modify one side for now. */
3837                 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3838                 s->nat_src_node = nsn;
3839         }
3840         if (pd->proto == IPPROTO_TCP) {
3841                 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3842                     off, pd, th, &s->src, &s->dst)) {
3843                         REASON_SET(&reason, PFRES_MEMORY);
3844                         pf_src_tree_remove_state(s);
3845                         STATE_DEC_COUNTERS(s);
3846                         uma_zfree(V_pf_state_z, s);
3847                         return (PF_DROP);
3848                 }
3849                 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3850                     pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3851                     &s->src, &s->dst, rewrite)) {
3852                         /* This really shouldn't happen!!! */
3853                         DPFPRINTF(PF_DEBUG_URGENT,
3854                             ("pf_normalize_tcp_stateful failed on first "
3855                              "pkt\n"));
3856                         pf_normalize_tcp_cleanup(s);
3857                         pf_src_tree_remove_state(s);
3858                         STATE_DEC_COUNTERS(s);
3859                         uma_zfree(V_pf_state_z, s);
3860                         return (PF_DROP);
3861                 }
3862         }
3863         s->direction = pd->dir;
3864
3865         /*
3866          * sk/nk could already been setup by pf_get_translation().
3867          */
3868         if (nr == NULL) {
3869                 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3870                     __func__, nr, sk, nk));
3871                 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3872                 if (sk == NULL)
3873                         goto csfailed;
3874                 nk = sk;
3875         } else
3876                 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3877                     __func__, nr, sk, nk));
3878
3879         /* Swap sk/nk for PF_OUT. */
3880         if (pf_state_insert(BOUND_IFACE(r, kif),
3881             (pd->dir == PF_IN) ? sk : nk,
3882             (pd->dir == PF_IN) ? nk : sk, s)) {
3883                 if (pd->proto == IPPROTO_TCP)
3884                         pf_normalize_tcp_cleanup(s);
3885                 REASON_SET(&reason, PFRES_STATEINS);
3886                 pf_src_tree_remove_state(s);
3887                 STATE_DEC_COUNTERS(s);
3888                 uma_zfree(V_pf_state_z, s);
3889                 return (PF_DROP);
3890         } else
3891                 *sm = s;
3892
3893         if (tag > 0)
3894                 s->tag = tag;
3895         if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3896             TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3897                 s->src.state = PF_TCPS_PROXY_SRC;
3898                 /* undo NAT changes, if they have taken place */
3899                 if (nr != NULL) {
3900                         struct pf_state_key *skt = s->key[PF_SK_WIRE];
3901                         if (pd->dir == PF_OUT)
3902                                 skt = s->key[PF_SK_STACK];
3903                         PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3904                         PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3905                         if (pd->sport)
3906                                 *pd->sport = skt->port[pd->sidx];
3907                         if (pd->dport)
3908                                 *pd->dport = skt->port[pd->didx];
3909                         if (pd->proto_sum)
3910                                 *pd->proto_sum = bproto_sum;
3911                         if (pd->ip_sum)
3912                                 *pd->ip_sum = bip_sum;
3913                         m_copyback(m, off, hdrlen, pd->hdr.any);
3914                 }
3915                 s->src.seqhi = htonl(arc4random());
3916                 /* Find mss option */
3917                 int rtid = M_GETFIB(m);
3918                 mss = pf_get_mss(m, off, th->th_off, pd->af);
3919                 mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3920                 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3921                 s->src.mss = mss;
3922                 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3923                     th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3924                     TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3925                 REASON_SET(&reason, PFRES_SYNPROXY);
3926                 return (PF_SYNPROXY_DROP);
3927         }
3928
3929         return (PF_PASS);
3930
3931 csfailed:
3932         if (sk != NULL)
3933                 uma_zfree(V_pf_state_key_z, sk);
3934         if (nk != NULL)
3935                 uma_zfree(V_pf_state_key_z, nk);
3936
3937         if (sn != NULL) {
3938                 struct pf_srchash *sh;
3939
3940                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
3941                 PF_HASHROW_LOCK(sh);
3942                 if (--sn->states == 0 && sn->expire == 0) {
3943                         pf_unlink_src_node(sn);
3944                         uma_zfree(V_pf_sources_z, sn);
3945                         counter_u64_add(
3946                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3947                 }
3948                 PF_HASHROW_UNLOCK(sh);
3949         }
3950
3951         if (nsn != sn && nsn != NULL) {
3952                 struct pf_srchash *sh;
3953
3954                 sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)];
3955                 PF_HASHROW_LOCK(sh);
3956                 if (--nsn->states == 0 && nsn->expire == 0) {
3957                         pf_unlink_src_node(nsn);
3958                         uma_zfree(V_pf_sources_z, nsn);
3959                         counter_u64_add(
3960                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3961                 }
3962                 PF_HASHROW_UNLOCK(sh);
3963         }
3964
3965         return (PF_DROP);
3966 }
3967
3968 static int
3969 pf_test_fragment(struct pf_krule **rm, int direction, struct pfi_kkif *kif,
3970     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am,
3971     struct pf_kruleset **rsm)
3972 {
3973         struct pf_krule         *r, *a = NULL;
3974         struct pf_kruleset      *ruleset = NULL;
3975         sa_family_t              af = pd->af;
3976         u_short                  reason;
3977         int                      tag = -1;
3978         int                      asd = 0;
3979         int                      match = 0;
3980         struct pf_kanchor_stackframe    anchor_stack[PF_ANCHOR_STACKSIZE];
3981
3982         PF_RULES_RASSERT();
3983
3984         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3985         while (r != NULL) {
3986                 counter_u64_add(r->evaluations, 1);
3987                 if (pfi_kkif_match(r->kif, kif) == r->ifnot)
3988                         r = r->skip[PF_SKIP_IFP].ptr;
3989                 else if (r->direction && r->direction != direction)
3990                         r = r->skip[PF_SKIP_DIR].ptr;
3991                 else if (r->af && r->af != af)
3992                         r = r->skip[PF_SKIP_AF].ptr;
3993                 else if (r->proto && r->proto != pd->proto)
3994                         r = r->skip[PF_SKIP_PROTO].ptr;
3995                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3996                     r->src.neg, kif, M_GETFIB(m)))
3997                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3998                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3999                     r->dst.neg, NULL, M_GETFIB(m)))
4000                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
4001                 else if (r->tos && !(r->tos == pd->tos))
4002                         r = TAILQ_NEXT(r, entries);
4003                 else if (r->os_fingerprint != PF_OSFP_ANY)
4004                         r = TAILQ_NEXT(r, entries);
4005                 else if (pd->proto == IPPROTO_UDP &&
4006                     (r->src.port_op || r->dst.port_op))
4007                         r = TAILQ_NEXT(r, entries);
4008                 else if (pd->proto == IPPROTO_TCP &&
4009                     (r->src.port_op || r->dst.port_op || r->flagset))
4010                         r = TAILQ_NEXT(r, entries);
4011                 else if ((pd->proto == IPPROTO_ICMP ||
4012                     pd->proto == IPPROTO_ICMPV6) &&
4013                     (r->type || r->code))
4014                         r = TAILQ_NEXT(r, entries);
4015                 else if (r->prio &&
4016                     !pf_match_ieee8021q_pcp(r->prio, m))
4017                         r = TAILQ_NEXT(r, entries);
4018                 else if (r->prob && r->prob <=
4019                     (arc4random() % (UINT_MAX - 1) + 1))
4020                         r = TAILQ_NEXT(r, entries);
4021                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
4022                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
4023                         r = TAILQ_NEXT(r, entries);
4024                 else {
4025                         if (r->anchor == NULL) {
4026                                 match = 1;
4027                                 *rm = r;
4028                                 *am = a;
4029                                 *rsm = ruleset;
4030                                 if ((*rm)->quick)
4031                                         break;
4032                                 r = TAILQ_NEXT(r, entries);
4033                         } else
4034                                 pf_step_into_anchor(anchor_stack, &asd,
4035                                     &ruleset, PF_RULESET_FILTER, &r, &a,
4036                                     &match);
4037                 }
4038                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
4039                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
4040                         break;
4041         }
4042         r = *rm;
4043         a = *am;
4044         ruleset = *rsm;
4045
4046         REASON_SET(&reason, PFRES_MATCH);
4047
4048         if (r->log)
4049                 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
4050                     1);
4051
4052         if (r->action != PF_PASS)
4053                 return (PF_DROP);
4054
4055         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
4056                 REASON_SET(&reason, PFRES_MEMORY);
4057                 return (PF_DROP);
4058         }
4059
4060         return (PF_PASS);
4061 }
4062
4063 static int
4064 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
4065         struct pf_state **state, struct pfi_kkif *kif, struct mbuf *m, int off,
4066         struct pf_pdesc *pd, u_short *reason, int *copyback)
4067 {
4068         struct tcphdr           *th = pd->hdr.tcp;
4069         u_int16_t                win = ntohs(th->th_win);
4070         u_int32_t                ack, end, seq, orig_seq;
4071         u_int8_t                 sws, dws;
4072         int                      ackskew;
4073
4074         if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
4075                 sws = src->wscale & PF_WSCALE_MASK;
4076                 dws = dst->wscale & PF_WSCALE_MASK;
4077         } else
4078                 sws = dws = 0;
4079
4080         /*
4081          * Sequence tracking algorithm from Guido van Rooij's paper:
4082          *   http://www.madison-gurkha.com/publications/tcp_filtering/
4083          *      tcp_filtering.ps
4084          */
4085
4086         orig_seq = seq = ntohl(th->th_seq);
4087         if (src->seqlo == 0) {
4088                 /* First packet from this end. Set its state */
4089
4090                 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
4091                     src->scrub == NULL) {
4092                         if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
4093                                 REASON_SET(reason, PFRES_MEMORY);
4094                                 return (PF_DROP);
4095                         }
4096                 }
4097
4098                 /* Deferred generation of sequence number modulator */
4099                 if (dst->seqdiff && !src->seqdiff) {
4100                         /* use random iss for the TCP server */
4101                         while ((src->seqdiff = arc4random() - seq) == 0)
4102                                 ;
4103                         ack = ntohl(th->th_ack) - dst->seqdiff;
4104                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
4105                             src->seqdiff), 0);
4106                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
4107                         *copyback = 1;
4108                 } else {
4109                         ack = ntohl(th->th_ack);
4110                 }
4111
4112                 end = seq + pd->p_len;
4113                 if (th->th_flags & TH_SYN) {
4114                         end++;
4115                         if (dst->wscale & PF_WSCALE_FLAG) {
4116                                 src->wscale = pf_get_wscale(m, off, th->th_off,
4117                                     pd->af);
4118                                 if (src->wscale & PF_WSCALE_FLAG) {
4119                                         /* Remove scale factor from initial
4120                                          * window */
4121                                         sws = src->wscale & PF_WSCALE_MASK;
4122                                         win = ((u_int32_t)win + (1 << sws) - 1)
4123                                             >> sws;
4124                                         dws = dst->wscale & PF_WSCALE_MASK;
4125                                 } else {
4126                                         /* fixup other window */
4127                                         dst->max_win <<= dst->wscale &
4128                                             PF_WSCALE_MASK;
4129                                         /* in case of a retrans SYN|ACK */
4130                                         dst->wscale = 0;
4131                                 }
4132                         }
4133                 }
4134                 if (th->th_flags & TH_FIN)
4135                         end++;
4136
4137                 src->seqlo = seq;
4138                 if (src->state < TCPS_SYN_SENT)
4139                         src->state = TCPS_SYN_SENT;
4140
4141                 /*
4142                  * May need to slide the window (seqhi may have been set by
4143                  * the crappy stack check or if we picked up the connection
4144                  * after establishment)
4145                  */
4146                 if (src->seqhi == 1 ||
4147                     SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
4148                         src->seqhi = end + MAX(1, dst->max_win << dws);
4149                 if (win > src->max_win)
4150                         src->max_win = win;
4151
4152         } else {
4153                 ack = ntohl(th->th_ack) - dst->seqdiff;
4154                 if (src->seqdiff) {
4155                         /* Modulate sequence numbers */
4156                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
4157                             src->seqdiff), 0);
4158                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
4159                         *copyback = 1;
4160                 }
4161                 end = seq + pd->p_len;
4162                 if (th->th_flags & TH_SYN)
4163                         end++;
4164                 if (th->th_flags & TH_FIN)
4165                         end++;
4166         }
4167
4168         if ((th->th_flags & TH_ACK) == 0) {
4169                 /* Let it pass through the ack skew check */
4170                 ack = dst->seqlo;
4171         } else if ((ack == 0 &&
4172             (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
4173             /* broken tcp stacks do not set ack */
4174             (dst->state < TCPS_SYN_SENT)) {
4175                 /*
4176                  * Many stacks (ours included) will set the ACK number in an
4177                  * FIN|ACK if the SYN times out -- no sequence to ACK.
4178                  */
4179                 ack = dst->seqlo;
4180         }
4181
4182         if (seq == end) {
4183                 /* Ease sequencing restrictions on no data packets */
4184                 seq = src->seqlo;
4185                 end = seq;
4186         }
4187
4188         ackskew = dst->seqlo - ack;
4189
4190         /*
4191          * Need to demodulate the sequence numbers in any TCP SACK options
4192          * (Selective ACK). We could optionally validate the SACK values
4193          * against the current ACK window, either forwards or backwards, but
4194          * I'm not confident that SACK has been implemented properly
4195          * everywhere. It wouldn't surprise me if several stacks accidentally
4196          * SACK too far backwards of previously ACKed data. There really aren't
4197          * any security implications of bad SACKing unless the target stack
4198          * doesn't validate the option length correctly. Someone trying to
4199          * spoof into a TCP connection won't bother blindly sending SACK
4200          * options anyway.
4201          */
4202         if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
4203                 if (pf_modulate_sack(m, off, pd, th, dst))
4204                         *copyback = 1;
4205         }
4206
4207 #define MAXACKWINDOW (0xffff + 1500)    /* 1500 is an arbitrary fudge factor */
4208         if (SEQ_GEQ(src->seqhi, end) &&
4209             /* Last octet inside other's window space */
4210             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
4211             /* Retrans: not more than one window back */
4212             (ackskew >= -MAXACKWINDOW) &&
4213             /* Acking not more than one reassembled fragment backwards */
4214             (ackskew <= (MAXACKWINDOW << sws)) &&
4215             /* Acking not more than one window forward */
4216             ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
4217             (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
4218             (pd->flags & PFDESC_IP_REAS) == 0)) {
4219             /* Require an exact/+1 sequence match on resets when possible */
4220
4221                 if (dst->scrub || src->scrub) {
4222                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4223                             *state, src, dst, copyback))
4224                                 return (PF_DROP);
4225                 }
4226
4227                 /* update max window */
4228                 if (src->max_win < win)
4229                         src->max_win = win;
4230                 /* synchronize sequencing */
4231                 if (SEQ_GT(end, src->seqlo))
4232                         src->seqlo = end;
4233                 /* slide the window of what the other end can send */
4234                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4235                         dst->seqhi = ack + MAX((win << sws), 1);
4236
4237                 /* update states */
4238                 if (th->th_flags & TH_SYN)
4239                         if (src->state < TCPS_SYN_SENT)
4240                                 src->state = TCPS_SYN_SENT;
4241                 if (th->th_flags & TH_FIN)
4242                         if (src->state < TCPS_CLOSING)
4243                                 src->state = TCPS_CLOSING;
4244                 if (th->th_flags & TH_ACK) {
4245                         if (dst->state == TCPS_SYN_SENT) {
4246                                 dst->state = TCPS_ESTABLISHED;
4247                                 if (src->state == TCPS_ESTABLISHED &&
4248                                     (*state)->src_node != NULL &&
4249                                     pf_src_connlimit(state)) {
4250                                         REASON_SET(reason, PFRES_SRCLIMIT);
4251                                         return (PF_DROP);
4252                                 }
4253                         } else if (dst->state == TCPS_CLOSING)
4254                                 dst->state = TCPS_FIN_WAIT_2;
4255                 }
4256                 if (th->th_flags & TH_RST)
4257                         src->state = dst->state = TCPS_TIME_WAIT;
4258
4259                 /* update expire time */
4260                 (*state)->expire = time_uptime;
4261                 if (src->state >= TCPS_FIN_WAIT_2 &&
4262                     dst->state >= TCPS_FIN_WAIT_2)
4263                         (*state)->timeout = PFTM_TCP_CLOSED;
4264                 else if (src->state >= TCPS_CLOSING &&
4265                     dst->state >= TCPS_CLOSING)
4266                         (*state)->timeout = PFTM_TCP_FIN_WAIT;
4267                 else if (src->state < TCPS_ESTABLISHED ||
4268                     dst->state < TCPS_ESTABLISHED)
4269                         (*state)->timeout = PFTM_TCP_OPENING;
4270                 else if (src->state >= TCPS_CLOSING ||
4271                     dst->state >= TCPS_CLOSING)
4272                         (*state)->timeout = PFTM_TCP_CLOSING;
4273                 else
4274                         (*state)->timeout = PFTM_TCP_ESTABLISHED;
4275
4276                 /* Fall through to PASS packet */
4277
4278         } else if ((dst->state < TCPS_SYN_SENT ||
4279                 dst->state >= TCPS_FIN_WAIT_2 ||
4280                 src->state >= TCPS_FIN_WAIT_2) &&
4281             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4282             /* Within a window forward of the originating packet */
4283             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4284             /* Within a window backward of the originating packet */
4285
4286                 /*
4287                  * This currently handles three situations:
4288                  *  1) Stupid stacks will shotgun SYNs before their peer
4289                  *     replies.
4290                  *  2) When PF catches an already established stream (the
4291                  *     firewall rebooted, the state table was flushed, routes
4292                  *     changed...)
4293                  *  3) Packets get funky immediately after the connection
4294                  *     closes (this should catch Solaris spurious ACK|FINs
4295                  *     that web servers like to spew after a close)
4296                  *
4297                  * This must be a little more careful than the above code
4298                  * since packet floods will also be caught here. We don't
4299                  * update the TTL here to mitigate the damage of a packet
4300                  * flood and so the same code can handle awkward establishment
4301                  * and a loosened connection close.
4302                  * In the establishment case, a correct peer response will
4303                  * validate the connection, go through the normal state code
4304                  * and keep updating the state TTL.
4305                  */
4306
4307                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4308                         printf("pf: loose state match: ");
4309                         pf_print_state(*state);
4310                         pf_print_flags(th->th_flags);
4311                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4312                             "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
4313                             pd->p_len, ackskew,
4314                             (unsigned long long)counter_u64_fetch((*state)->packets[0]),
4315                             (unsigned long long)counter_u64_fetch((*state)->packets[1]),
4316                             pd->dir == PF_IN ? "in" : "out",
4317                             pd->dir == (*state)->direction ? "fwd" : "rev");
4318                 }
4319
4320                 if (dst->scrub || src->scrub) {
4321                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4322                             *state, src, dst, copyback))
4323                                 return (PF_DROP);
4324                 }
4325
4326                 /* update max window */
4327                 if (src->max_win < win)
4328                         src->max_win = win;
4329                 /* synchronize sequencing */
4330                 if (SEQ_GT(end, src->seqlo))
4331                         src->seqlo = end;
4332                 /* slide the window of what the other end can send */
4333                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4334                         dst->seqhi = ack + MAX((win << sws), 1);
4335
4336                 /*
4337                  * Cannot set dst->seqhi here since this could be a shotgunned
4338                  * SYN and not an already established connection.
4339                  */
4340
4341                 if (th->th_flags & TH_FIN)
4342                         if (src->state < TCPS_CLOSING)
4343                                 src->state = TCPS_CLOSING;
4344                 if (th->th_flags & TH_RST)
4345                         src->state = dst->state = TCPS_TIME_WAIT;
4346
4347                 /* Fall through to PASS packet */
4348
4349         } else {
4350                 if ((*state)->dst.state == TCPS_SYN_SENT &&
4351                     (*state)->src.state == TCPS_SYN_SENT) {
4352                         /* Send RST for state mismatches during handshake */
4353                         if (!(th->th_flags & TH_RST))
4354                                 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4355                                     pd->dst, pd->src, th->th_dport,
4356                                     th->th_sport, ntohl(th->th_ack), 0,
4357                                     TH_RST, 0, 0,
4358                                     (*state)->rule.ptr->return_ttl, 1, 0,
4359                                     kif->pfik_ifp);
4360                         src->seqlo = 0;
4361                         src->seqhi = 1;
4362                         src->max_win = 1;
4363                 } else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4364                         printf("pf: BAD state: ");
4365                         pf_print_state(*state);
4366                         pf_print_flags(th->th_flags);
4367                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4368                             "pkts=%llu:%llu dir=%s,%s\n",
4369                             seq, orig_seq, ack, pd->p_len, ackskew,
4370                             (unsigned long long)counter_u64_fetch((*state)->packets[0]),
4371                             (unsigned long long)counter_u64_fetch((*state)->packets[1]),
4372                             pd->dir == PF_IN ? "in" : "out",
4373                             pd->dir == (*state)->direction ? "fwd" : "rev");
4374                         printf("pf: State failure on: %c %c %c %c | %c %c\n",
4375                             SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4376                             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4377                             ' ': '2',
4378                             (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4379                             (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4380                             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4381                             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4382                 }
4383                 REASON_SET(reason, PFRES_BADSTATE);
4384                 return (PF_DROP);
4385         }
4386
4387         return (PF_PASS);
4388 }
4389
4390 static int
4391 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4392         struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4393 {
4394         struct tcphdr           *th = pd->hdr.tcp;
4395
4396         if (th->th_flags & TH_SYN)
4397                 if (src->state < TCPS_SYN_SENT)
4398                         src->state = TCPS_SYN_SENT;
4399         if (th->th_flags & TH_FIN)
4400                 if (src->state < TCPS_CLOSING)
4401                         src->state = TCPS_CLOSING;
4402         if (th->th_flags & TH_ACK) {
4403                 if (dst->state == TCPS_SYN_SENT) {
4404                         dst->state = TCPS_ESTABLISHED;
4405                         if (src->state == TCPS_ESTABLISHED &&
4406                             (*state)->src_node != NULL &&
4407                             pf_src_connlimit(state)) {
4408                                 REASON_SET(reason, PFRES_SRCLIMIT);
4409                                 return (PF_DROP);
4410                         }
4411                 } else if (dst->state == TCPS_CLOSING) {
4412                         dst->state = TCPS_FIN_WAIT_2;
4413                 } else if (src->state == TCPS_SYN_SENT &&
4414                     dst->state < TCPS_SYN_SENT) {
4415                         /*
4416                          * Handle a special sloppy case where we only see one
4417                          * half of the connection. If there is a ACK after
4418                          * the initial SYN without ever seeing a packet from
4419                          * the destination, set the connection to established.
4420                          */
4421                         dst->state = src->state = TCPS_ESTABLISHED;
4422                         if ((*state)->src_node != NULL &&
4423                             pf_src_connlimit(state)) {
4424                                 REASON_SET(reason, PFRES_SRCLIMIT);
4425                                 return (PF_DROP);
4426                         }
4427                 } else if (src->state == TCPS_CLOSING &&
4428                     dst->state == TCPS_ESTABLISHED &&
4429                     dst->seqlo == 0) {
4430                         /*
4431                          * Handle the closing of half connections where we
4432                          * don't see the full bidirectional FIN/ACK+ACK
4433                          * handshake.
4434                          */
4435                         dst->state = TCPS_CLOSING;
4436                 }
4437         }
4438         if (th->th_flags & TH_RST)
4439                 src->state = dst->state = TCPS_TIME_WAIT;
4440
4441         /* update expire time */
4442         (*state)->expire = time_uptime;
4443         if (src->state >= TCPS_FIN_WAIT_2 &&
4444             dst->state >= TCPS_FIN_WAIT_2)
4445                 (*state)->timeout = PFTM_TCP_CLOSED;
4446         else if (src->state >= TCPS_CLOSING &&
4447             dst->state >= TCPS_CLOSING)
4448                 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4449         else if (src->state < TCPS_ESTABLISHED ||
4450             dst->state < TCPS_ESTABLISHED)
4451                 (*state)->timeout = PFTM_TCP_OPENING;
4452         else if (src->state >= TCPS_CLOSING ||
4453             dst->state >= TCPS_CLOSING)
4454                 (*state)->timeout = PFTM_TCP_CLOSING;
4455         else
4456                 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4457
4458         return (PF_PASS);
4459 }
4460
4461 static int
4462 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kkif *kif,
4463     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4464     u_short *reason)
4465 {
4466         struct pf_state_key_cmp  key;
4467         struct tcphdr           *th = pd->hdr.tcp;
4468         int                      copyback = 0;
4469         struct pf_state_peer    *src, *dst;
4470         struct pf_state_key     *sk;
4471
4472         bzero(&key, sizeof(key));
4473         key.af = pd->af;
4474         key.proto = IPPROTO_TCP;
4475         if (direction == PF_IN) {       /* wire side, straight */
4476                 PF_ACPY(&key.addr[0], pd->src, key.af);
4477                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4478                 key.port[0] = th->th_sport;
4479                 key.port[1] = th->th_dport;
4480         } else {                        /* stack side, reverse */
4481                 PF_ACPY(&key.addr[1], pd->src, key.af);
4482                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4483                 key.port[1] = th->th_sport;
4484                 key.port[0] = th->th_dport;
4485         }
4486
4487         STATE_LOOKUP(kif, &key, direction, *state, pd);
4488
4489         if (direction == (*state)->direction) {
4490                 src = &(*state)->src;
4491                 dst = &(*state)->dst;
4492         } else {
4493                 src = &(*state)->dst;
4494                 dst = &(*state)->src;
4495         }
4496
4497         sk = (*state)->key[pd->didx];
4498
4499         if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4500                 if (direction != (*state)->direction) {
4501                         REASON_SET(reason, PFRES_SYNPROXY);
4502                         return (PF_SYNPROXY_DROP);
4503                 }
4504                 if (th->th_flags & TH_SYN) {
4505                         if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4506                                 REASON_SET(reason, PFRES_SYNPROXY);
4507                                 return (PF_DROP);
4508                         }
4509                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4510                             pd->src, th->th_dport, th->th_sport,
4511                             (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4512                             TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4513                         REASON_SET(reason, PFRES_SYNPROXY);
4514                         return (PF_SYNPROXY_DROP);
4515                 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK ||
4516                     (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4517                     (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4518                         REASON_SET(reason, PFRES_SYNPROXY);
4519                         return (PF_DROP);
4520                 } else if ((*state)->src_node != NULL &&
4521                     pf_src_connlimit(state)) {
4522                         REASON_SET(reason, PFRES_SRCLIMIT);
4523                         return (PF_DROP);
4524                 } else
4525                         (*state)->src.state = PF_TCPS_PROXY_DST;
4526         }
4527         if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4528                 if (direction == (*state)->direction) {
4529                         if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4530                             (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4531                             (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4532                                 REASON_SET(reason, PFRES_SYNPROXY);
4533                                 return (PF_DROP);
4534                         }
4535                         (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4536                         if ((*state)->dst.seqhi == 1)
4537                                 (*state)->dst.seqhi = htonl(arc4random());
4538                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4539                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4540                             sk->port[pd->sidx], sk->port[pd->didx],
4541                             (*state)->dst.seqhi, 0, TH_SYN, 0,
4542                             (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4543                         REASON_SET(reason, PFRES_SYNPROXY);
4544                         return (PF_SYNPROXY_DROP);
4545                 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4546                     (TH_SYN|TH_ACK)) ||
4547                     (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4548                         REASON_SET(reason, PFRES_SYNPROXY);
4549                         return (PF_DROP);
4550                 } else {
4551                         (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4552                         (*state)->dst.seqlo = ntohl(th->th_seq);
4553                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4554                             pd->src, th->th_dport, th->th_sport,
4555                             ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4556                             TH_ACK, (*state)->src.max_win, 0, 0, 0,
4557                             (*state)->tag, NULL);
4558                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4559                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4560                             sk->port[pd->sidx], sk->port[pd->didx],
4561                             (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4562                             TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4563                         (*state)->src.seqdiff = (*state)->dst.seqhi -
4564                             (*state)->src.seqlo;
4565                         (*state)->dst.seqdiff = (*state)->src.seqhi -
4566                             (*state)->dst.seqlo;
4567                         (*state)->src.seqhi = (*state)->src.seqlo +
4568                             (*state)->dst.max_win;
4569                         (*state)->dst.seqhi = (*state)->dst.seqlo +
4570                             (*state)->src.max_win;
4571                         (*state)->src.wscale = (*state)->dst.wscale = 0;
4572                         (*state)->src.state = (*state)->dst.state =
4573                             TCPS_ESTABLISHED;
4574                         REASON_SET(reason, PFRES_SYNPROXY);
4575                         return (PF_SYNPROXY_DROP);
4576                 }
4577         }
4578
4579         if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4580             dst->state >= TCPS_FIN_WAIT_2 &&
4581             src->state >= TCPS_FIN_WAIT_2) {
4582                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4583                         printf("pf: state reuse ");
4584                         pf_print_state(*state);
4585                         pf_print_flags(th->th_flags);
4586                         printf("\n");
4587                 }
4588                 /* XXX make sure it's the same direction ?? */
4589                 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4590                 pf_unlink_state(*state, PF_ENTER_LOCKED);
4591                 *state = NULL;
4592                 return (PF_DROP);
4593         }
4594
4595         if ((*state)->state_flags & PFSTATE_SLOPPY) {
4596                 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4597                         return (PF_DROP);
4598         } else {
4599                 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4600                     &copyback) == PF_DROP)
4601                         return (PF_DROP);
4602         }
4603
4604         /* translate source/destination address, if necessary */
4605         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4606                 struct pf_state_key *nk = (*state)->key[pd->didx];
4607
4608                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4609                     nk->port[pd->sidx] != th->th_sport)
4610                         pf_change_ap(m, pd->src, &th->th_sport,
4611                             pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
4612                             nk->port[pd->sidx], 0, pd->af);
4613
4614                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4615                     nk->port[pd->didx] != th->th_dport)
4616                         pf_change_ap(m, pd->dst, &th->th_dport,
4617                             pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
4618                             nk->port[pd->didx], 0, pd->af);
4619                 copyback = 1;
4620         }
4621
4622         /* Copyback sequence modulation or stateful scrub changes if needed */
4623         if (copyback)
4624                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
4625
4626         return (PF_PASS);
4627 }
4628
4629 static int
4630 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kkif *kif,
4631     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4632 {
4633         struct pf_state_peer    *src, *dst;
4634         struct pf_state_key_cmp  key;
4635         struct udphdr           *uh = pd->hdr.udp;
4636
4637         bzero(&key, sizeof(key));
4638         key.af = pd->af;
4639         key.proto = IPPROTO_UDP;
4640         if (direction == PF_IN) {       /* wire side, straight */
4641                 PF_ACPY(&key.addr[0], pd->src, key.af);
4642                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4643                 key.port[0] = uh->uh_sport;
4644                 key.port[1] = uh->uh_dport;
4645         } else {                        /* stack side, reverse */
4646                 PF_ACPY(&key.addr[1], pd->src, key.af);
4647                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4648                 key.port[1] = uh->uh_sport;
4649                 key.port[0] = uh->uh_dport;
4650         }
4651
4652         STATE_LOOKUP(kif, &key, direction, *state, pd);
4653
4654         if (direction == (*state)->direction) {
4655                 src = &(*state)->src;
4656                 dst = &(*state)->dst;
4657         } else {
4658                 src = &(*state)->dst;
4659                 dst = &(*state)->src;
4660         }
4661
4662         /* update states */
4663         if (src->state < PFUDPS_SINGLE)
4664                 src->state = PFUDPS_SINGLE;
4665         if (dst->state == PFUDPS_SINGLE)
4666                 dst->state = PFUDPS_MULTIPLE;
4667
4668         /* update expire time */
4669         (*state)->expire = time_uptime;
4670         if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4671                 (*state)->timeout = PFTM_UDP_MULTIPLE;
4672         else
4673                 (*state)->timeout = PFTM_UDP_SINGLE;
4674
4675         /* translate source/destination address, if necessary */
4676         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4677                 struct pf_state_key *nk = (*state)->key[pd->didx];
4678
4679                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4680                     nk->port[pd->sidx] != uh->uh_sport)
4681                         pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
4682                             &uh->uh_sum, &nk->addr[pd->sidx],
4683                             nk->port[pd->sidx], 1, pd->af);
4684
4685                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4686                     nk->port[pd->didx] != uh->uh_dport)
4687                         pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
4688                             &uh->uh_sum, &nk->addr[pd->didx],
4689                             nk->port[pd->didx], 1, pd->af);
4690                 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4691         }
4692
4693         return (PF_PASS);
4694 }
4695
4696 static int
4697 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kkif *kif,
4698     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4699 {
4700         struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
4701         u_int16_t        icmpid = 0, *icmpsum;
4702         u_int8_t         icmptype, icmpcode;
4703         int              state_icmp = 0;
4704         struct pf_state_key_cmp key;
4705
4706         bzero(&key, sizeof(key));
4707         switch (pd->proto) {
4708 #ifdef INET
4709         case IPPROTO_ICMP:
4710                 icmptype = pd->hdr.icmp->icmp_type;
4711                 icmpcode = pd->hdr.icmp->icmp_code;
4712                 icmpid = pd->hdr.icmp->icmp_id;
4713                 icmpsum = &pd->hdr.icmp->icmp_cksum;
4714
4715                 if (icmptype == ICMP_UNREACH ||
4716                     icmptype == ICMP_SOURCEQUENCH ||
4717                     icmptype == ICMP_REDIRECT ||
4718                     icmptype == ICMP_TIMXCEED ||
4719                     icmptype == ICMP_PARAMPROB)
4720                         state_icmp++;
4721                 break;
4722 #endif /* INET */
4723 #ifdef INET6
4724         case IPPROTO_ICMPV6:
4725                 icmptype = pd->hdr.icmp6->icmp6_type;
4726                 icmpcode = pd->hdr.icmp6->icmp6_code;
4727                 icmpid = pd->hdr.icmp6->icmp6_id;
4728                 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4729
4730                 if (icmptype == ICMP6_DST_UNREACH ||
4731                     icmptype == ICMP6_PACKET_TOO_BIG ||
4732                     icmptype == ICMP6_TIME_EXCEEDED ||
4733                     icmptype == ICMP6_PARAM_PROB)
4734                         state_icmp++;
4735                 break;
4736 #endif /* INET6 */
4737         }
4738
4739         if (!state_icmp) {
4740                 /*
4741                  * ICMP query/reply message not related to a TCP/UDP packet.
4742                  * Search for an ICMP state.
4743                  */
4744                 key.af = pd->af;
4745                 key.proto = pd->proto;
4746                 key.port[0] = key.port[1] = icmpid;
4747                 if (direction == PF_IN) {       /* wire side, straight */
4748                         PF_ACPY(&key.addr[0], pd->src, key.af);
4749                         PF_ACPY(&key.addr[1], pd->dst, key.af);
4750                 } else {                        /* stack side, reverse */
4751                         PF_ACPY(&key.addr[1], pd->src, key.af);
4752                         PF_ACPY(&key.addr[0], pd->dst, key.af);
4753                 }
4754
4755                 STATE_LOOKUP(kif, &key, direction, *state, pd);
4756
4757                 (*state)->expire = time_uptime;
4758                 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4759
4760                 /* translate source/destination address, if necessary */
4761                 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4762                         struct pf_state_key *nk = (*state)->key[pd->didx];
4763
4764                         switch (pd->af) {
4765 #ifdef INET
4766                         case AF_INET:
4767                                 if (PF_ANEQ(pd->src,
4768                                     &nk->addr[pd->sidx], AF_INET))
4769                                         pf_change_a(&saddr->v4.s_addr,
4770                                             pd->ip_sum,
4771                                             nk->addr[pd->sidx].v4.s_addr, 0);
4772
4773                                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4774                                     AF_INET))
4775                                         pf_change_a(&daddr->v4.s_addr,
4776                                             pd->ip_sum,
4777                                             nk->addr[pd->didx].v4.s_addr, 0);
4778
4779                                 if (nk->port[0] !=
4780                                     pd->hdr.icmp->icmp_id) {
4781                                         pd->hdr.icmp->icmp_cksum =
4782                                             pf_cksum_fixup(
4783                                             pd->hdr.icmp->icmp_cksum, icmpid,
4784                                             nk->port[pd->sidx], 0);
4785                                         pd->hdr.icmp->icmp_id =
4786                                             nk->port[pd->sidx];
4787                                 }
4788
4789                                 m_copyback(m, off, ICMP_MINLEN,
4790                                     (caddr_t )pd->hdr.icmp);
4791                                 break;
4792 #endif /* INET */
4793 #ifdef INET6
4794                         case AF_INET6:
4795                                 if (PF_ANEQ(pd->src,
4796                                     &nk->addr[pd->sidx], AF_INET6))
4797                                         pf_change_a6(saddr,
4798                                             &pd->hdr.icmp6->icmp6_cksum,
4799                                             &nk->addr[pd->sidx], 0);
4800
4801                                 if (PF_ANEQ(pd->dst,
4802                                     &nk->addr[pd->didx], AF_INET6))
4803                                         pf_change_a6(daddr,
4804                                             &pd->hdr.icmp6->icmp6_cksum,
4805                                             &nk->addr[pd->didx], 0);
4806
4807                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4808                                     (caddr_t )pd->hdr.icmp6);
4809                                 break;
4810 #endif /* INET6 */
4811                         }
4812                 }
4813                 return (PF_PASS);
4814
4815         } else {
4816                 /*
4817                  * ICMP error message in response to a TCP/UDP packet.
4818                  * Extract the inner TCP/UDP header and search for that state.
4819                  */
4820
4821                 struct pf_pdesc pd2;
4822                 bzero(&pd2, sizeof pd2);
4823 #ifdef INET
4824                 struct ip       h2;
4825 #endif /* INET */
4826 #ifdef INET6
4827                 struct ip6_hdr  h2_6;
4828                 int             terminal = 0;
4829 #endif /* INET6 */
4830                 int             ipoff2 = 0;
4831                 int             off2 = 0;
4832
4833                 pd2.af = pd->af;
4834                 /* Payload packet is from the opposite direction. */
4835                 pd2.sidx = (direction == PF_IN) ? 1 : 0;
4836                 pd2.didx = (direction == PF_IN) ? 0 : 1;
4837                 switch (pd->af) {
4838 #ifdef INET
4839                 case AF_INET:
4840                         /* offset of h2 in mbuf chain */
4841                         ipoff2 = off + ICMP_MINLEN;
4842
4843                         if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4844                             NULL, reason, pd2.af)) {
4845                                 DPFPRINTF(PF_DEBUG_MISC,
4846                                     ("pf: ICMP error message too short "
4847                                     "(ip)\n"));
4848                                 return (PF_DROP);
4849                         }
4850                         /*
4851                          * ICMP error messages don't refer to non-first
4852                          * fragments
4853                          */
4854                         if (h2.ip_off & htons(IP_OFFMASK)) {
4855                                 REASON_SET(reason, PFRES_FRAG);
4856                                 return (PF_DROP);
4857                         }
4858
4859                         /* offset of protocol header that follows h2 */
4860                         off2 = ipoff2 + (h2.ip_hl << 2);
4861
4862                         pd2.proto = h2.ip_p;
4863                         pd2.src = (struct pf_addr *)&h2.ip_src;
4864                         pd2.dst = (struct pf_addr *)&h2.ip_dst;
4865                         pd2.ip_sum = &h2.ip_sum;
4866                         break;
4867 #endif /* INET */
4868 #ifdef INET6
4869                 case AF_INET6:
4870                         ipoff2 = off + sizeof(struct icmp6_hdr);
4871
4872                         if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4873                             NULL, reason, pd2.af)) {
4874                                 DPFPRINTF(PF_DEBUG_MISC,
4875                                     ("pf: ICMP error message too short "
4876                                     "(ip6)\n"));
4877                                 return (PF_DROP);
4878                         }
4879                         pd2.proto = h2_6.ip6_nxt;
4880                         pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4881                         pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4882                         pd2.ip_sum = NULL;
4883                         off2 = ipoff2 + sizeof(h2_6);
4884                         do {
4885                                 switch (pd2.proto) {
4886                                 case IPPROTO_FRAGMENT:
4887                                         /*
4888                                          * ICMPv6 error messages for
4889                                          * non-first fragments
4890                                          */
4891                                         REASON_SET(reason, PFRES_FRAG);
4892                                         return (PF_DROP);
4893                                 case IPPROTO_AH:
4894                                 case IPPROTO_HOPOPTS:
4895                                 case IPPROTO_ROUTING:
4896                                 case IPPROTO_DSTOPTS: {
4897                                         /* get next header and header length */
4898                                         struct ip6_ext opt6;
4899
4900                                         if (!pf_pull_hdr(m, off2, &opt6,
4901                                             sizeof(opt6), NULL, reason,
4902                                             pd2.af)) {
4903                                                 DPFPRINTF(PF_DEBUG_MISC,
4904                                                     ("pf: ICMPv6 short opt\n"));
4905                                                 return (PF_DROP);
4906                                         }
4907                                         if (pd2.proto == IPPROTO_AH)
4908                                                 off2 += (opt6.ip6e_len + 2) * 4;
4909                                         else
4910                                                 off2 += (opt6.ip6e_len + 1) * 8;
4911                                         pd2.proto = opt6.ip6e_nxt;
4912                                         /* goto the next header */
4913                                         break;
4914                                 }
4915                                 default:
4916                                         terminal++;
4917                                         break;
4918                                 }
4919                         } while (!terminal);
4920                         break;
4921 #endif /* INET6 */
4922                 }
4923
4924                 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) {
4925                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
4926                                 printf("pf: BAD ICMP %d:%d outer dst: ",
4927                                     icmptype, icmpcode);
4928                                 pf_print_host(pd->src, 0, pd->af);
4929                                 printf(" -> ");
4930                                 pf_print_host(pd->dst, 0, pd->af);
4931                                 printf(" inner src: ");
4932                                 pf_print_host(pd2.src, 0, pd2.af);
4933                                 printf(" -> ");
4934                                 pf_print_host(pd2.dst, 0, pd2.af);
4935                                 printf("\n");
4936                         }
4937                         REASON_SET(reason, PFRES_BADSTATE);
4938                         return (PF_DROP);
4939                 }
4940
4941                 switch (pd2.proto) {
4942                 case IPPROTO_TCP: {
4943                         struct tcphdr            th;
4944                         u_int32_t                seq;
4945                         struct pf_state_peer    *src, *dst;
4946                         u_int8_t                 dws;
4947                         int                      copyback = 0;
4948
4949                         /*
4950                          * Only the first 8 bytes of the TCP header can be
4951                          * expected. Don't access any TCP header fields after
4952                          * th_seq, an ackskew test is not possible.
4953                          */
4954                         if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4955                             pd2.af)) {
4956                                 DPFPRINTF(PF_DEBUG_MISC,
4957                                     ("pf: ICMP error message too short "
4958                                     "(tcp)\n"));
4959                                 return (PF_DROP);
4960                         }
4961
4962                         key.af = pd2.af;
4963                         key.proto = IPPROTO_TCP;
4964                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4965                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4966                         key.port[pd2.sidx] = th.th_sport;
4967                         key.port[pd2.didx] = th.th_dport;
4968
4969                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4970
4971                         if (direction == (*state)->direction) {
4972                                 src = &(*state)->dst;
4973                                 dst = &(*state)->src;
4974                         } else {
4975                                 src = &(*state)->src;
4976                                 dst = &(*state)->dst;
4977                         }
4978
4979                         if (src->wscale && dst->wscale)
4980                                 dws = dst->wscale & PF_WSCALE_MASK;
4981                         else
4982                                 dws = 0;
4983
4984                         /* Demodulate sequence number */
4985                         seq = ntohl(th.th_seq) - src->seqdiff;
4986                         if (src->seqdiff) {
4987                                 pf_change_a(&th.th_seq, icmpsum,
4988                                     htonl(seq), 0);
4989                                 copyback = 1;
4990                         }
4991
4992                         if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4993                             (!SEQ_GEQ(src->seqhi, seq) ||
4994                             !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4995                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4996                                         printf("pf: BAD ICMP %d:%d ",
4997                                             icmptype, icmpcode);
4998                                         pf_print_host(pd->src, 0, pd->af);
4999                                         printf(" -> ");
5000                                         pf_print_host(pd->dst, 0, pd->af);
5001                                         printf(" state: ");
5002                                         pf_print_state(*state);
5003                                         printf(" seq=%u\n", seq);
5004                                 }
5005                                 REASON_SET(reason, PFRES_BADSTATE);
5006                                 return (PF_DROP);
5007                         } else {
5008                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
5009                                         printf("pf: OK ICMP %d:%d ",
5010                                             icmptype, icmpcode);
5011                                         pf_print_host(pd->src, 0, pd->af);
5012                                         printf(" -> ");
5013                                         pf_print_host(pd->dst, 0, pd->af);
5014                                         printf(" state: ");
5015                                         pf_print_state(*state);
5016                                         printf(" seq=%u\n", seq);
5017                                 }
5018                         }
5019
5020                         /* translate source/destination address, if necessary */
5021                         if ((*state)->key[PF_SK_WIRE] !=
5022                             (*state)->key[PF_SK_STACK]) {
5023                                 struct pf_state_key *nk =
5024                                     (*state)->key[pd->didx];
5025
5026                                 if (PF_ANEQ(pd2.src,
5027                                     &nk->addr[pd2.sidx], pd2.af) ||
5028                                     nk->port[pd2.sidx] != th.th_sport)
5029                                         pf_change_icmp(pd2.src, &th.th_sport,
5030                                             daddr, &nk->addr[pd2.sidx],
5031                                             nk->port[pd2.sidx], NULL,
5032                                             pd2.ip_sum, icmpsum,
5033                                             pd->ip_sum, 0, pd2.af);
5034
5035                                 if (PF_ANEQ(pd2.dst,
5036                                     &nk->addr[pd2.didx], pd2.af) ||
5037                                     nk->port[pd2.didx] != th.th_dport)
5038                                         pf_change_icmp(pd2.dst, &th.th_dport,
5039                                             saddr, &nk->addr[pd2.didx],
5040                                             nk->port[pd2.didx], NULL,
5041                                             pd2.ip_sum, icmpsum,
5042                                             pd->ip_sum, 0, pd2.af);
5043                                 copyback = 1;
5044                         }
5045
5046                         if (copyback) {
5047                                 switch (pd2.af) {
5048 #ifdef INET
5049                                 case AF_INET:
5050                                         m_copyback(m, off, ICMP_MINLEN,
5051                                             (caddr_t )pd->hdr.icmp);
5052                                         m_copyback(m, ipoff2, sizeof(h2),
5053                                             (caddr_t )&h2);
5054                                         break;
5055 #endif /* INET */
5056 #ifdef INET6
5057                                 case AF_INET6:
5058                                         m_copyback(m, off,
5059                                             sizeof(struct icmp6_hdr),
5060                                             (caddr_t )pd->hdr.icmp6);
5061                                         m_copyback(m, ipoff2, sizeof(h2_6),
5062                                             (caddr_t )&h2_6);
5063                                         break;
5064 #endif /* INET6 */
5065                                 }
5066                                 m_copyback(m, off2, 8, (caddr_t)&th);
5067                         }
5068
5069                         return (PF_PASS);
5070                         break;
5071                 }
5072                 case IPPROTO_UDP: {
5073                         struct udphdr           uh;
5074
5075                         if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
5076                             NULL, reason, pd2.af)) {
5077                                 DPFPRINTF(PF_DEBUG_MISC,
5078                                     ("pf: ICMP error message too short "
5079                                     "(udp)\n"));
5080                                 return (PF_DROP);
5081                         }
5082
5083                         key.af = pd2.af;
5084                         key.proto = IPPROTO_UDP;
5085                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5086                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5087                         key.port[pd2.sidx] = uh.uh_sport;
5088                         key.port[pd2.didx] = uh.uh_dport;
5089
5090                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5091
5092                         /* translate source/destination address, if necessary */
5093                         if ((*state)->key[PF_SK_WIRE] !=
5094                             (*state)->key[PF_SK_STACK]) {
5095                                 struct pf_state_key *nk =
5096                                     (*state)->key[pd->didx];
5097
5098                                 if (PF_ANEQ(pd2.src,
5099                                     &nk->addr[pd2.sidx], pd2.af) ||
5100                                     nk->port[pd2.sidx] != uh.uh_sport)
5101                                         pf_change_icmp(pd2.src, &uh.uh_sport,
5102                                             daddr, &nk->addr[pd2.sidx],
5103                                             nk->port[pd2.sidx], &uh.uh_sum,
5104                                             pd2.ip_sum, icmpsum,
5105                                             pd->ip_sum, 1, pd2.af);
5106
5107                                 if (PF_ANEQ(pd2.dst,
5108                                     &nk->addr[pd2.didx], pd2.af) ||
5109                                     nk->port[pd2.didx] != uh.uh_dport)
5110                                         pf_change_icmp(pd2.dst, &uh.uh_dport,
5111                                             saddr, &nk->addr[pd2.didx],
5112                                             nk->port[pd2.didx], &uh.uh_sum,
5113                                             pd2.ip_sum, icmpsum,
5114                                             pd->ip_sum, 1, pd2.af);
5115
5116                                 switch (pd2.af) {
5117 #ifdef INET
5118                                 case AF_INET:
5119                                         m_copyback(m, off, ICMP_MINLEN,
5120                                             (caddr_t )pd->hdr.icmp);
5121                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5122                                         break;
5123 #endif /* INET */
5124 #ifdef INET6
5125                                 case AF_INET6:
5126                                         m_copyback(m, off,
5127                                             sizeof(struct icmp6_hdr),
5128                                             (caddr_t )pd->hdr.icmp6);
5129                                         m_copyback(m, ipoff2, sizeof(h2_6),
5130                                             (caddr_t )&h2_6);
5131                                         break;
5132 #endif /* INET6 */
5133                                 }
5134                                 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
5135                         }
5136                         return (PF_PASS);
5137                         break;
5138                 }
5139 #ifdef INET
5140                 case IPPROTO_ICMP: {
5141                         struct icmp             iih;
5142
5143                         if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
5144                             NULL, reason, pd2.af)) {
5145                                 DPFPRINTF(PF_DEBUG_MISC,
5146                                     ("pf: ICMP error message too short i"
5147                                     "(icmp)\n"));
5148                                 return (PF_DROP);
5149                         }
5150
5151                         key.af = pd2.af;
5152                         key.proto = IPPROTO_ICMP;
5153                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5154                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5155                         key.port[0] = key.port[1] = iih.icmp_id;
5156
5157                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5158
5159                         /* translate source/destination address, if necessary */
5160                         if ((*state)->key[PF_SK_WIRE] !=
5161                             (*state)->key[PF_SK_STACK]) {
5162                                 struct pf_state_key *nk =
5163                                     (*state)->key[pd->didx];
5164
5165                                 if (PF_ANEQ(pd2.src,
5166                                     &nk->addr[pd2.sidx], pd2.af) ||
5167                                     nk->port[pd2.sidx] != iih.icmp_id)
5168                                         pf_change_icmp(pd2.src, &iih.icmp_id,
5169                                             daddr, &nk->addr[pd2.sidx],
5170                                             nk->port[pd2.sidx], NULL,
5171                                             pd2.ip_sum, icmpsum,
5172                                             pd->ip_sum, 0, AF_INET);
5173
5174                                 if (PF_ANEQ(pd2.dst,
5175                                     &nk->addr[pd2.didx], pd2.af) ||
5176                                     nk->port[pd2.didx] != iih.icmp_id)
5177                                         pf_change_icmp(pd2.dst, &iih.icmp_id,
5178                                             saddr, &nk->addr[pd2.didx],
5179                                             nk->port[pd2.didx], NULL,
5180                                             pd2.ip_sum, icmpsum,
5181                                             pd->ip_sum, 0, AF_INET);
5182
5183                                 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
5184                                 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5185                                 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
5186                         }
5187                         return (PF_PASS);
5188                         break;
5189                 }
5190 #endif /* INET */
5191 #ifdef INET6
5192                 case IPPROTO_ICMPV6: {
5193                         struct icmp6_hdr        iih;
5194
5195                         if (!pf_pull_hdr(m, off2, &iih,
5196                             sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
5197                                 DPFPRINTF(PF_DEBUG_MISC,
5198                                     ("pf: ICMP error message too short "
5199                                     "(icmp6)\n"));
5200                                 return (PF_DROP);
5201                         }
5202
5203                         key.af = pd2.af;
5204                         key.proto = IPPROTO_ICMPV6;
5205                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5206                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5207                         key.port[0] = key.port[1] = iih.icmp6_id;
5208
5209                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5210
5211                         /* translate source/destination address, if necessary */
5212                         if ((*state)->key[PF_SK_WIRE] !=
5213                             (*state)->key[PF_SK_STACK]) {
5214                                 struct pf_state_key *nk =
5215                                     (*state)->key[pd->didx];
5216
5217                                 if (PF_ANEQ(pd2.src,
5218                                     &nk->addr[pd2.sidx], pd2.af) ||
5219                                     nk->port[pd2.sidx] != iih.icmp6_id)
5220                                         pf_change_icmp(pd2.src, &iih.icmp6_id,
5221                                             daddr, &nk->addr[pd2.sidx],
5222                                             nk->port[pd2.sidx], NULL,
5223                                             pd2.ip_sum, icmpsum,
5224                                             pd->ip_sum, 0, AF_INET6);
5225
5226                                 if (PF_ANEQ(pd2.dst,
5227                                     &nk->addr[pd2.didx], pd2.af) ||
5228                                     nk->port[pd2.didx] != iih.icmp6_id)
5229                                         pf_change_icmp(pd2.dst, &iih.icmp6_id,
5230                                             saddr, &nk->addr[pd2.didx],
5231                                             nk->port[pd2.didx], NULL,
5232                                             pd2.ip_sum, icmpsum,
5233                                             pd->ip_sum, 0, AF_INET6);
5234
5235                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
5236                                     (caddr_t)pd->hdr.icmp6);
5237                                 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
5238                                 m_copyback(m, off2, sizeof(struct icmp6_hdr),
5239                                     (caddr_t)&iih);
5240                         }
5241                         return (PF_PASS);
5242                         break;
5243                 }
5244 #endif /* INET6 */
5245                 default: {
5246                         key.af = pd2.af;
5247                         key.proto = pd2.proto;
5248                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5249                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5250                         key.port[0] = key.port[1] = 0;
5251
5252                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5253
5254                         /* translate source/destination address, if necessary */
5255                         if ((*state)->key[PF_SK_WIRE] !=
5256                             (*state)->key[PF_SK_STACK]) {
5257                                 struct pf_state_key *nk =
5258                                     (*state)->key[pd->didx];
5259
5260                                 if (PF_ANEQ(pd2.src,
5261                                     &nk->addr[pd2.sidx], pd2.af))
5262                                         pf_change_icmp(pd2.src, NULL, daddr,
5263                                             &nk->addr[pd2.sidx], 0, NULL,
5264                                             pd2.ip_sum, icmpsum,
5265                                             pd->ip_sum, 0, pd2.af);
5266
5267                                 if (PF_ANEQ(pd2.dst,
5268                                     &nk->addr[pd2.didx], pd2.af))
5269                                         pf_change_icmp(pd2.dst, NULL, saddr,
5270                                             &nk->addr[pd2.didx], 0, NULL,
5271                                             pd2.ip_sum, icmpsum,
5272                                             pd->ip_sum, 0, pd2.af);
5273
5274                                 switch (pd2.af) {
5275 #ifdef INET
5276                                 case AF_INET:
5277                                         m_copyback(m, off, ICMP_MINLEN,
5278                                             (caddr_t)pd->hdr.icmp);
5279                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5280                                         break;
5281 #endif /* INET */
5282 #ifdef INET6
5283                                 case AF_INET6:
5284                                         m_copyback(m, off,
5285                                             sizeof(struct icmp6_hdr),
5286                                             (caddr_t )pd->hdr.icmp6);
5287                                         m_copyback(m, ipoff2, sizeof(h2_6),
5288                                             (caddr_t )&h2_6);
5289                                         break;
5290 #endif /* INET6 */
5291                                 }
5292                         }
5293                         return (PF_PASS);
5294                         break;
5295                 }
5296                 }
5297         }
5298 }
5299
5300 static int
5301 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kkif *kif,
5302     struct mbuf *m, struct pf_pdesc *pd)
5303 {
5304         struct pf_state_peer    *src, *dst;
5305         struct pf_state_key_cmp  key;
5306
5307         bzero(&key, sizeof(key));
5308         key.af = pd->af;
5309         key.proto = pd->proto;
5310         if (direction == PF_IN) {
5311                 PF_ACPY(&key.addr[0], pd->src, key.af);
5312                 PF_ACPY(&key.addr[1], pd->dst, key.af);
5313                 key.port[0] = key.port[1] = 0;
5314         } else {
5315                 PF_ACPY(&key.addr[1], pd->src, key.af);
5316                 PF_ACPY(&key.addr[0], pd->dst, key.af);
5317                 key.port[1] = key.port[0] = 0;
5318         }
5319
5320         STATE_LOOKUP(kif, &key, direction, *state, pd);
5321
5322         if (direction == (*state)->direction) {
5323                 src = &(*state)->src;
5324                 dst = &(*state)->dst;
5325         } else {
5326                 src = &(*state)->dst;
5327                 dst = &(*state)->src;
5328         }
5329
5330         /* update states */
5331         if (src->state < PFOTHERS_SINGLE)
5332                 src->state = PFOTHERS_SINGLE;
5333         if (dst->state == PFOTHERS_SINGLE)
5334                 dst->state = PFOTHERS_MULTIPLE;
5335
5336         /* update expire time */
5337         (*state)->expire = time_uptime;
5338         if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5339                 (*state)->timeout = PFTM_OTHER_MULTIPLE;
5340         else
5341                 (*state)->timeout = PFTM_OTHER_SINGLE;
5342
5343         /* translate source/destination address, if necessary */
5344         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5345                 struct pf_state_key *nk = (*state)->key[pd->didx];
5346
5347                 KASSERT(nk, ("%s: nk is null", __func__));
5348                 KASSERT(pd, ("%s: pd is null", __func__));
5349                 KASSERT(pd->src, ("%s: pd->src is null", __func__));
5350                 KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5351                 switch (pd->af) {
5352 #ifdef INET
5353                 case AF_INET:
5354                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5355                                 pf_change_a(&pd->src->v4.s_addr,
5356                                     pd->ip_sum,
5357                                     nk->addr[pd->sidx].v4.s_addr,
5358                                     0);
5359
5360                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5361                                 pf_change_a(&pd->dst->v4.s_addr,
5362                                     pd->ip_sum,
5363                                     nk->addr[pd->didx].v4.s_addr,
5364                                     0);
5365
5366                         break;
5367 #endif /* INET */
5368 #ifdef INET6
5369                 case AF_INET6:
5370                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5371                                 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5372
5373                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5374                                 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5375 #endif /* INET6 */
5376                 }
5377         }
5378         return (PF_PASS);
5379 }
5380
5381 /*
5382  * ipoff and off are measured from the start of the mbuf chain.
5383  * h must be at "ipoff" on the mbuf chain.
5384  */
5385 void *
5386 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5387     u_short *actionp, u_short *reasonp, sa_family_t af)
5388 {
5389         switch (af) {
5390 #ifdef INET
5391         case AF_INET: {
5392                 struct ip       *h = mtod(m, struct ip *);
5393                 u_int16_t        fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5394
5395                 if (fragoff) {
5396                         if (fragoff >= len)
5397                                 ACTION_SET(actionp, PF_PASS);
5398                         else {
5399                                 ACTION_SET(actionp, PF_DROP);
5400                                 REASON_SET(reasonp, PFRES_FRAG);
5401                         }
5402                         return (NULL);
5403                 }
5404                 if (m->m_pkthdr.len < off + len ||
5405                     ntohs(h->ip_len) < off + len) {
5406                         ACTION_SET(actionp, PF_DROP);
5407                         REASON_SET(reasonp, PFRES_SHORT);
5408                         return (NULL);
5409                 }
5410                 break;
5411         }
5412 #endif /* INET */
5413 #ifdef INET6
5414         case AF_INET6: {
5415                 struct ip6_hdr  *h = mtod(m, struct ip6_hdr *);
5416
5417                 if (m->m_pkthdr.len < off + len ||
5418                     (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5419                     (unsigned)(off + len)) {
5420                         ACTION_SET(actionp, PF_DROP);
5421                         REASON_SET(reasonp, PFRES_SHORT);
5422                         return (NULL);
5423                 }
5424                 break;
5425         }
5426 #endif /* INET6 */
5427         }
5428         m_copydata(m, off, len, p);
5429         return (p);
5430 }
5431
5432 int
5433 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif,
5434     int rtableid)
5435 {
5436         struct ifnet            *ifp;
5437
5438         /*
5439          * Skip check for addresses with embedded interface scope,
5440          * as they would always match anyway.
5441          */
5442         if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
5443                 return (1);
5444
5445         if (af != AF_INET && af != AF_INET6)
5446                 return (0);
5447
5448         /* Skip checks for ipsec interfaces */
5449         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5450                 return (1);
5451
5452         ifp = (kif != NULL) ? kif->pfik_ifp : NULL;
5453
5454         switch (af) {
5455 #ifdef INET6
5456         case AF_INET6:
5457                 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE,
5458                     ifp));
5459 #endif
5460 #ifdef INET
5461         case AF_INET:
5462                 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE,
5463                     ifp));
5464 #endif
5465         }
5466
5467         return (0);
5468 }
5469
5470 #ifdef INET
5471 static void
5472 pf_route(struct mbuf **m, struct pf_krule *r, int dir, struct ifnet *oifp,
5473     struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp)
5474 {
5475         struct mbuf             *m0, *m1;
5476         struct sockaddr_in      dst;
5477         struct ip               *ip;
5478         struct ifnet            *ifp = NULL;
5479         struct pf_addr           naddr;
5480         struct pf_ksrc_node     *sn = NULL;
5481         int                      error = 0;
5482         uint16_t                 ip_len, ip_off;
5483
5484         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5485         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5486             __func__));
5487
5488         if ((pd->pf_mtag == NULL &&
5489             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5490             pd->pf_mtag->routed++ > 3) {
5491                 m0 = *m;
5492                 *m = NULL;
5493                 goto bad_locked;
5494         }
5495
5496         if (r->rt == PF_DUPTO) {
5497                 if ((pd->pf_mtag->flags & PF_DUPLICATED)) {
5498                         if (s == NULL) {
5499                                 ifp = r->rpool.cur->kif ?
5500                                     r->rpool.cur->kif->pfik_ifp : NULL;
5501                         } else {
5502                                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5503                                 PF_STATE_UNLOCK(s);
5504                         }
5505                         if (ifp == oifp) {
5506                                 /* When the 2nd interface is not skipped */
5507                                 return;
5508                         } else {
5509                                 m0 = *m;
5510                                 *m = NULL;
5511                                 goto bad;
5512                         }
5513                 } else {
5514                         pd->pf_mtag->flags |= PF_DUPLICATED;
5515                         if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) {
5516                                 if (s)
5517                                         PF_STATE_UNLOCK(s);
5518                                 return;
5519                         }
5520                 }
5521         } else {
5522                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5523                         if (s)
5524                                 PF_STATE_UNLOCK(s);
5525                         return;
5526                 }
5527                 m0 = *m;
5528         }
5529
5530         ip = mtod(m0, struct ip *);
5531
5532         bzero(&dst, sizeof(dst));
5533         dst.sin_family = AF_INET;
5534         dst.sin_len = sizeof(dst);
5535         dst.sin_addr = ip->ip_dst;
5536
5537         bzero(&naddr, sizeof(naddr));
5538
5539         if (TAILQ_EMPTY(&r->rpool.list)) {
5540                 DPFPRINTF(PF_DEBUG_URGENT,
5541                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5542                 goto bad_locked;
5543         }
5544         if (s == NULL) {
5545                 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5546                     &naddr, NULL, &sn);
5547                 if (!PF_AZERO(&naddr, AF_INET))
5548                         dst.sin_addr.s_addr = naddr.v4.s_addr;
5549                 ifp = r->rpool.cur->kif ?
5550                     r->rpool.cur->kif->pfik_ifp : NULL;
5551         } else {
5552                 if (!PF_AZERO(&s->rt_addr, AF_INET))
5553                         dst.sin_addr.s_addr =
5554                             s->rt_addr.v4.s_addr;
5555                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5556                 PF_STATE_UNLOCK(s);
5557         }
5558         if (ifp == NULL)
5559                 goto bad;
5560
5561         if (dir == PF_IN) {
5562                 if (pf_test(PF_OUT, 0, ifp, &m0, inp) != PF_PASS)
5563                         goto bad;
5564                 else if (m0 == NULL)
5565                         goto done;
5566                 if (m0->m_len < sizeof(struct ip)) {
5567                         DPFPRINTF(PF_DEBUG_URGENT,
5568                             ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5569                         goto bad;
5570                 }
5571                 ip = mtod(m0, struct ip *);
5572         }
5573
5574         if (ifp->if_flags & IFF_LOOPBACK)
5575                 m0->m_flags |= M_SKIP_FIREWALL;
5576
5577         ip_len = ntohs(ip->ip_len);
5578         ip_off = ntohs(ip->ip_off);
5579
5580         /* Copied from FreeBSD 10.0-CURRENT ip_output. */
5581         m0->m_pkthdr.csum_flags |= CSUM_IP;
5582         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5583                 m0 = mb_unmapped_to_ext(m0);
5584                 if (m0 == NULL)
5585                         goto done;
5586                 in_delayed_cksum(m0);
5587                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5588         }
5589 #if defined(SCTP) || defined(SCTP_SUPPORT)
5590         if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5591                 m0 = mb_unmapped_to_ext(m0);
5592                 if (m0 == NULL)
5593                         goto done;
5594                 sctp_delayed_cksum(m0, (uint32_t)(ip->ip_hl << 2));
5595                 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5596         }
5597 #endif
5598
5599         /*
5600          * If small enough for interface, or the interface will take
5601          * care of the fragmentation for us, we can just send directly.
5602          */
5603         if (ip_len <= ifp->if_mtu ||
5604             (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
5605                 ip->ip_sum = 0;
5606                 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5607                         ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5608                         m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5609                 }
5610                 m_clrprotoflags(m0);    /* Avoid confusing lower layers. */
5611                 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5612                 goto done;
5613         }
5614
5615         /* Balk when DF bit is set or the interface didn't support TSO. */
5616         if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5617                 error = EMSGSIZE;
5618                 KMOD_IPSTAT_INC(ips_cantfrag);
5619                 if (r->rt != PF_DUPTO) {
5620                         icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5621                             ifp->if_mtu);
5622                         goto done;
5623                 } else
5624                         goto bad;
5625         }
5626
5627         error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5628         if (error)
5629                 goto bad;
5630
5631         for (; m0; m0 = m1) {
5632                 m1 = m0->m_nextpkt;
5633                 m0->m_nextpkt = NULL;
5634                 if (error == 0) {
5635                         m_clrprotoflags(m0);
5636                         error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5637                 } else
5638                         m_freem(m0);
5639         }
5640
5641         if (error == 0)
5642                 KMOD_IPSTAT_INC(ips_fragmented);
5643
5644 done:
5645         if (r->rt != PF_DUPTO)
5646                 *m = NULL;
5647         return;
5648
5649 bad_locked:
5650         if (s)
5651                 PF_STATE_UNLOCK(s);
5652 bad:
5653         m_freem(m0);
5654         goto done;
5655 }
5656 #endif /* INET */
5657
5658 #ifdef INET6
5659 static void
5660 pf_route6(struct mbuf **m, struct pf_krule *r, int dir, struct ifnet *oifp,
5661     struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp)
5662 {
5663         struct mbuf             *m0;
5664         struct sockaddr_in6     dst;
5665         struct ip6_hdr          *ip6;
5666         struct ifnet            *ifp = NULL;
5667         struct pf_addr           naddr;
5668         struct pf_ksrc_node     *sn = NULL;
5669
5670         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5671         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5672             __func__));
5673
5674         if ((pd->pf_mtag == NULL &&
5675             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5676             pd->pf_mtag->routed++ > 3) {
5677                 m0 = *m;
5678                 *m = NULL;
5679                 goto bad_locked;
5680         }
5681
5682         if (r->rt == PF_DUPTO) {
5683                 if ((pd->pf_mtag->flags & PF_DUPLICATED)) {
5684                         if (s == NULL) {
5685                                 ifp = r->rpool.cur->kif ?
5686                                     r->rpool.cur->kif->pfik_ifp : NULL;
5687                         } else {
5688                                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5689                                 PF_STATE_UNLOCK(s);
5690                         }
5691                         if (ifp == oifp) {
5692                                 /* When the 2nd interface is not skipped */
5693                                 return;
5694                         } else {
5695                                 m0 = *m;
5696                                 *m = NULL;
5697                                 goto bad;
5698                         }
5699                 } else {
5700                         pd->pf_mtag->flags |= PF_DUPLICATED;
5701                         if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) {
5702                                 if (s)
5703                                         PF_STATE_UNLOCK(s);
5704                                 return;
5705                         }
5706                 }
5707         } else {
5708                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5709                         if (s)
5710                                 PF_STATE_UNLOCK(s);
5711                         return;
5712                 }
5713                 m0 = *m;
5714         }
5715
5716         ip6 = mtod(m0, struct ip6_hdr *);
5717
5718         bzero(&dst, sizeof(dst));
5719         dst.sin6_family = AF_INET6;
5720         dst.sin6_len = sizeof(dst);
5721         dst.sin6_addr = ip6->ip6_dst;
5722
5723         bzero(&naddr, sizeof(naddr));
5724
5725         if (TAILQ_EMPTY(&r->rpool.list)) {
5726                 DPFPRINTF(PF_DEBUG_URGENT,
5727                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5728                 goto bad_locked;
5729         }
5730         if (s == NULL) {
5731                 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5732                     &naddr, NULL, &sn);
5733                 if (!PF_AZERO(&naddr, AF_INET6))
5734                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5735                             &naddr, AF_INET6);
5736                 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5737         } else {
5738                 if (!PF_AZERO(&s->rt_addr, AF_INET6))
5739                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5740                             &s->rt_addr, AF_INET6);
5741                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5742         }
5743
5744         if (s)
5745                 PF_STATE_UNLOCK(s);
5746
5747         if (ifp == NULL)
5748                 goto bad;
5749
5750         if (dir == PF_IN) {
5751                 if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp) != PF_PASS)
5752                         goto bad;
5753                 else if (m0 == NULL)
5754                         goto done;
5755                 if (m0->m_len < sizeof(struct ip6_hdr)) {
5756                         DPFPRINTF(PF_DEBUG_URGENT,
5757                             ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5758                             __func__));
5759                         goto bad;
5760                 }
5761                 ip6 = mtod(m0, struct ip6_hdr *);
5762         }
5763
5764         if (ifp->if_flags & IFF_LOOPBACK)
5765                 m0->m_flags |= M_SKIP_FIREWALL;
5766
5767         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
5768             ~ifp->if_hwassist) {
5769                 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
5770                 m0 = mb_unmapped_to_ext(m0);
5771                 if (m0 == NULL)
5772                         goto done;
5773                 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
5774                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
5775         }
5776
5777         /*
5778          * If the packet is too large for the outgoing interface,
5779          * send back an icmp6 error.
5780          */
5781         if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5782                 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5783         if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5784                 nd6_output_ifp(ifp, ifp, m0, &dst, NULL);
5785         else {
5786                 in6_ifstat_inc(ifp, ifs6_in_toobig);
5787                 if (r->rt != PF_DUPTO)
5788                         icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5789                 else
5790                         goto bad;
5791         }
5792
5793 done:
5794         if (r->rt != PF_DUPTO)
5795                 *m = NULL;
5796         return;
5797
5798 bad_locked:
5799         if (s)
5800                 PF_STATE_UNLOCK(s);
5801 bad:
5802         m_freem(m0);
5803         goto done;
5804 }
5805 #endif /* INET6 */
5806
5807 /*
5808  * FreeBSD supports cksum offloads for the following drivers.
5809  *  em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4)
5810  *
5811  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5812  *  network driver performed cksum including pseudo header, need to verify
5813  *   csum_data
5814  * CSUM_DATA_VALID :
5815  *  network driver performed cksum, needs to additional pseudo header
5816  *  cksum computation with partial csum_data(i.e. lack of H/W support for
5817  *  pseudo header, for instance sk(4) and possibly gem(4))
5818  *
5819  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5820  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5821  * TCP/UDP layer.
5822  * Also, set csum_data to 0xffff to force cksum validation.
5823  */
5824 static int
5825 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5826 {
5827         u_int16_t sum = 0;
5828         int hw_assist = 0;
5829         struct ip *ip;
5830
5831         if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5832                 return (1);
5833         if (m->m_pkthdr.len < off + len)
5834                 return (1);
5835
5836         switch (p) {
5837         case IPPROTO_TCP:
5838                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5839                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5840                                 sum = m->m_pkthdr.csum_data;
5841                         } else {
5842                                 ip = mtod(m, struct ip *);
5843                                 sum = in_pseudo(ip->ip_src.s_addr,
5844                                 ip->ip_dst.s_addr, htonl((u_short)len +
5845                                 m->m_pkthdr.csum_data + IPPROTO_TCP));
5846                         }
5847                         sum ^= 0xffff;
5848                         ++hw_assist;
5849                 }
5850                 break;
5851         case IPPROTO_UDP:
5852                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5853                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5854                                 sum = m->m_pkthdr.csum_data;
5855                         } else {
5856                                 ip = mtod(m, struct ip *);
5857                                 sum = in_pseudo(ip->ip_src.s_addr,
5858                                 ip->ip_dst.s_addr, htonl((u_short)len +
5859                                 m->m_pkthdr.csum_data + IPPROTO_UDP));
5860                         }
5861                         sum ^= 0xffff;
5862                         ++hw_assist;
5863                 }
5864                 break;
5865         case IPPROTO_ICMP:
5866 #ifdef INET6
5867         case IPPROTO_ICMPV6:
5868 #endif /* INET6 */
5869                 break;
5870         default:
5871                 return (1);
5872         }
5873
5874         if (!hw_assist) {
5875                 switch (af) {
5876                 case AF_INET:
5877                         if (p == IPPROTO_ICMP) {
5878                                 if (m->m_len < off)
5879                                         return (1);
5880                                 m->m_data += off;
5881                                 m->m_len -= off;
5882                                 sum = in_cksum(m, len);
5883                                 m->m_data -= off;
5884                                 m->m_len += off;
5885                         } else {
5886                                 if (m->m_len < sizeof(struct ip))
5887                                         return (1);
5888                                 sum = in4_cksum(m, p, off, len);
5889                         }
5890                         break;
5891 #ifdef INET6
5892                 case AF_INET6:
5893                         if (m->m_len < sizeof(struct ip6_hdr))
5894                                 return (1);
5895                         sum = in6_cksum(m, p, off, len);
5896                         break;
5897 #endif /* INET6 */
5898                 default:
5899                         return (1);
5900                 }
5901         }
5902         if (sum) {
5903                 switch (p) {
5904                 case IPPROTO_TCP:
5905                     {
5906                         KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5907                         break;
5908                     }
5909                 case IPPROTO_UDP:
5910                     {
5911                         KMOD_UDPSTAT_INC(udps_badsum);
5912                         break;
5913                     }
5914 #ifdef INET
5915                 case IPPROTO_ICMP:
5916                     {
5917                         KMOD_ICMPSTAT_INC(icps_checksum);
5918                         break;
5919                     }
5920 #endif
5921 #ifdef INET6
5922                 case IPPROTO_ICMPV6:
5923                     {
5924                         KMOD_ICMP6STAT_INC(icp6s_checksum);
5925                         break;
5926                     }
5927 #endif /* INET6 */
5928                 }
5929                 return (1);
5930         } else {
5931                 if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5932                         m->m_pkthdr.csum_flags |=
5933                             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5934                         m->m_pkthdr.csum_data = 0xffff;
5935                 }
5936         }
5937         return (0);
5938 }
5939
5940 #ifdef INET
5941 int
5942 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5943 {
5944         struct pfi_kkif         *kif;
5945         u_short                  action, reason = 0, log = 0;
5946         struct mbuf             *m = *m0;
5947         struct ip               *h = NULL;
5948         struct m_tag            *ipfwtag;
5949         struct pf_krule         *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5950         struct pf_state         *s = NULL;
5951         struct pf_kruleset      *ruleset = NULL;
5952         struct pf_pdesc          pd;
5953         int                      off, dirndx, pqid = 0;
5954
5955         PF_RULES_RLOCK_TRACKER;
5956
5957         M_ASSERTPKTHDR(m);
5958
5959         if (!V_pf_status.running)
5960                 return (PF_PASS);
5961
5962         memset(&pd, 0, sizeof(pd));
5963
5964         kif = (struct pfi_kkif *)ifp->if_pf_kif;
5965
5966         if (kif == NULL) {
5967                 DPFPRINTF(PF_DEBUG_URGENT,
5968                     ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5969                 return (PF_DROP);
5970         }
5971         if (kif->pfik_flags & PFI_IFLAG_SKIP)
5972                 return (PF_PASS);
5973
5974         if (m->m_flags & M_SKIP_FIREWALL)
5975                 return (PF_PASS);
5976
5977         pd.pf_mtag = pf_find_mtag(m);
5978
5979         PF_RULES_RLOCK();
5980
5981         if (ip_divert_ptr != NULL &&
5982             ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5983                 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5984                 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5985                         if (pd.pf_mtag == NULL &&
5986                             ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5987                                 action = PF_DROP;
5988                                 goto done;
5989                         }
5990                         pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5991                         m_tag_delete(m, ipfwtag);
5992                 }
5993                 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5994                         m->m_flags |= M_FASTFWD_OURS;
5995                         pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5996                 }
5997         } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5998                 /* We do IP header normalization and packet reassembly here */
5999                 action = PF_DROP;
6000                 goto done;
6001         }
6002         m = *m0;        /* pf_normalize messes with m0 */
6003         h = mtod(m, struct ip *);
6004
6005         off = h->ip_hl << 2;
6006         if (off < (int)sizeof(struct ip)) {
6007                 action = PF_DROP;
6008                 REASON_SET(&reason, PFRES_SHORT);
6009                 log = 1;
6010                 goto done;
6011         }
6012
6013         pd.src = (struct pf_addr *)&h->ip_src;
6014         pd.dst = (struct pf_addr *)&h->ip_dst;
6015         pd.sport = pd.dport = NULL;
6016         pd.ip_sum = &h->ip_sum;
6017         pd.proto_sum = NULL;
6018         pd.proto = h->ip_p;
6019         pd.dir = dir;
6020         pd.sidx = (dir == PF_IN) ? 0 : 1;
6021         pd.didx = (dir == PF_IN) ? 1 : 0;
6022         pd.af = AF_INET;
6023         pd.tos = h->ip_tos & ~IPTOS_ECN_MASK;
6024         pd.tot_len = ntohs(h->ip_len);
6025
6026         /* handle fragments that didn't get reassembled by normalization */
6027         if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
6028                 action = pf_test_fragment(&r, dir, kif, m, h,
6029                     &pd, &a, &ruleset);
6030                 goto done;
6031         }
6032
6033         switch (h->ip_p) {
6034         case IPPROTO_TCP: {
6035                 struct tcphdr   th;
6036
6037                 pd.hdr.tcp = &th;
6038                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6039                     &action, &reason, AF_INET)) {
6040                         log = action != PF_PASS;
6041                         goto done;
6042                 }
6043                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6044                 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
6045                         pqid = 1;
6046                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6047                 if (action == PF_DROP)
6048                         goto done;
6049                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6050                     &reason);
6051                 if (action == PF_PASS) {
6052                         if (V_pfsync_update_state_ptr != NULL)
6053                                 V_pfsync_update_state_ptr(s);
6054                         r = s->rule.ptr;
6055                         a = s->anchor.ptr;
6056                         log = s->log;
6057                 } else if (s == NULL)
6058                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6059                             &a, &ruleset, inp);
6060                 break;
6061         }
6062
6063         case IPPROTO_UDP: {
6064                 struct udphdr   uh;
6065
6066                 pd.hdr.udp = &uh;
6067                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6068                     &action, &reason, AF_INET)) {
6069                         log = action != PF_PASS;
6070                         goto done;
6071                 }
6072                 if (uh.uh_dport == 0 ||
6073                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6074                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6075                         action = PF_DROP;
6076                         REASON_SET(&reason, PFRES_SHORT);
6077                         goto done;
6078                 }
6079                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6080                 if (action == PF_PASS) {
6081                         if (V_pfsync_update_state_ptr != NULL)
6082                                 V_pfsync_update_state_ptr(s);
6083                         r = s->rule.ptr;
6084                         a = s->anchor.ptr;
6085                         log = s->log;
6086                 } else if (s == NULL)
6087                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6088                             &a, &ruleset, inp);
6089                 break;
6090         }
6091
6092         case IPPROTO_ICMP: {
6093                 struct icmp     ih;
6094
6095                 pd.hdr.icmp = &ih;
6096                 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
6097                     &action, &reason, AF_INET)) {
6098                         log = action != PF_PASS;
6099                         goto done;
6100                 }
6101                 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
6102                     &reason);
6103                 if (action == PF_PASS) {
6104                         if (V_pfsync_update_state_ptr != NULL)
6105                                 V_pfsync_update_state_ptr(s);
6106                         r = s->rule.ptr;
6107                         a = s->anchor.ptr;
6108                         log = s->log;
6109                 } else if (s == NULL)
6110                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6111                             &a, &ruleset, inp);
6112                 break;
6113         }
6114
6115 #ifdef INET6
6116         case IPPROTO_ICMPV6: {
6117                 action = PF_DROP;
6118                 DPFPRINTF(PF_DEBUG_MISC,
6119                     ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
6120                 goto done;
6121         }
6122 #endif
6123
6124         default:
6125                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6126                 if (action == PF_PASS) {
6127                         if (V_pfsync_update_state_ptr != NULL)
6128                                 V_pfsync_update_state_ptr(s);
6129                         r = s->rule.ptr;
6130                         a = s->anchor.ptr;
6131                         log = s->log;
6132                 } else if (s == NULL)
6133                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6134                             &a, &ruleset, inp);
6135                 break;
6136         }
6137
6138 done:
6139         PF_RULES_RUNLOCK();
6140         if (action == PF_PASS && h->ip_hl > 5 &&
6141             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6142                 action = PF_DROP;
6143                 REASON_SET(&reason, PFRES_IPOPTIONS);
6144                 log = r->log;
6145                 DPFPRINTF(PF_DEBUG_MISC,
6146                     ("pf: dropping packet with ip options\n"));
6147         }
6148
6149         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6150                 action = PF_DROP;
6151                 REASON_SET(&reason, PFRES_MEMORY);
6152         }
6153         if (r->rtableid >= 0)
6154                 M_SETFIB(m, r->rtableid);
6155
6156         if (r->scrub_flags & PFSTATE_SETPRIO) {
6157                 if (pd.tos & IPTOS_LOWDELAY)
6158                         pqid = 1;
6159                 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6160                         action = PF_DROP;
6161                         REASON_SET(&reason, PFRES_MEMORY);
6162                         log = 1;
6163                         DPFPRINTF(PF_DEBUG_MISC,
6164                             ("pf: failed to allocate 802.1q mtag\n"));
6165                 }
6166         }
6167
6168 #ifdef ALTQ
6169         if (action == PF_PASS && r->qid) {
6170                 if (pd.pf_mtag == NULL &&
6171                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6172                         action = PF_DROP;
6173                         REASON_SET(&reason, PFRES_MEMORY);
6174                 } else {
6175                         if (s != NULL)
6176                                 pd.pf_mtag->qid_hash = pf_state_hash(s);
6177                         if (pqid || (pd.tos & IPTOS_LOWDELAY))
6178                                 pd.pf_mtag->qid = r->pqid;
6179                         else
6180                                 pd.pf_mtag->qid = r->qid;
6181                         /* Add hints for ecn. */
6182                         pd.pf_mtag->hdr = h;
6183                 }
6184         }
6185 #endif /* ALTQ */
6186
6187         /*
6188          * connections redirected to loopback should not match sockets
6189          * bound specifically to loopback due to security implications,
6190          * see tcp_input() and in_pcblookup_listen().
6191          */
6192         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6193             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6194             (s->nat_rule.ptr->action == PF_RDR ||
6195             s->nat_rule.ptr->action == PF_BINAT) &&
6196             IN_LOOPBACK(ntohl(pd.dst->v4.s_addr)))
6197                 m->m_flags |= M_SKIP_FIREWALL;
6198
6199         if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
6200             !PACKET_LOOPED(&pd)) {
6201                 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
6202                     sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
6203                 if (ipfwtag != NULL) {
6204                         ((struct ipfw_rule_ref *)(ipfwtag+1))->info =
6205                             ntohs(r->divert.port);
6206                         ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
6207
6208                         if (s)
6209                                 PF_STATE_UNLOCK(s);
6210
6211                         m_tag_prepend(m, ipfwtag);
6212                         if (m->m_flags & M_FASTFWD_OURS) {
6213                                 if (pd.pf_mtag == NULL &&
6214                                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6215                                         action = PF_DROP;
6216                                         REASON_SET(&reason, PFRES_MEMORY);
6217                                         log = 1;
6218                                         DPFPRINTF(PF_DEBUG_MISC,
6219                                             ("pf: failed to allocate tag\n"));
6220                                 } else {
6221                                         pd.pf_mtag->flags |=
6222                                             PF_FASTFWD_OURS_PRESENT;
6223                                         m->m_flags &= ~M_FASTFWD_OURS;
6224                                 }
6225                         }
6226                         ip_divert_ptr(*m0, dir == PF_IN);
6227                         *m0 = NULL;
6228
6229                         return (action);
6230                 } else {
6231                         /* XXX: ipfw has the same behaviour! */
6232                         action = PF_DROP;
6233                         REASON_SET(&reason, PFRES_MEMORY);
6234                         log = 1;
6235                         DPFPRINTF(PF_DEBUG_MISC,
6236                             ("pf: failed to allocate divert tag\n"));
6237                 }
6238         }
6239
6240         if (log) {
6241                 struct pf_krule *lr;
6242
6243                 if (s != NULL && s->nat_rule.ptr != NULL &&
6244                     s->nat_rule.ptr->log & PF_LOG_ALL)
6245                         lr = s->nat_rule.ptr;
6246                 else
6247                         lr = r;
6248                 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
6249                     (s == NULL));
6250         }
6251
6252         counter_u64_add(kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS],
6253             pd.tot_len);
6254         counter_u64_add(kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS],
6255             1);
6256
6257         if (action == PF_PASS || r->action == PF_DROP) {
6258                 dirndx = (dir == PF_OUT);
6259                 counter_u64_add(r->packets[dirndx], 1);
6260                 counter_u64_add(r->bytes[dirndx], pd.tot_len);
6261                 if (a != NULL) {
6262                         counter_u64_add(a->packets[dirndx], 1);
6263                         counter_u64_add(a->bytes[dirndx], pd.tot_len);
6264                 }
6265                 if (s != NULL) {
6266                         if (s->nat_rule.ptr != NULL) {
6267                                 counter_u64_add(s->nat_rule.ptr->packets[dirndx],
6268                                     1);
6269                                 counter_u64_add(s->nat_rule.ptr->bytes[dirndx],
6270                                     pd.tot_len);
6271                         }
6272                         if (s->src_node != NULL) {
6273                                 counter_u64_add(s->src_node->packets[dirndx],
6274                                     1);
6275                                 counter_u64_add(s->src_node->bytes[dirndx],
6276                                     pd.tot_len);
6277                         }
6278                         if (s->nat_src_node != NULL) {
6279                                 counter_u64_add(s->nat_src_node->packets[dirndx],
6280                                     1);
6281                                 counter_u64_add(s->nat_src_node->bytes[dirndx],
6282                                     pd.tot_len);
6283                         }
6284                         dirndx = (dir == s->direction) ? 0 : 1;
6285                         counter_u64_add(s->packets[dirndx], 1);
6286                         counter_u64_add(s->bytes[dirndx], pd.tot_len);
6287                 }
6288                 tr = r;
6289                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6290                 if (nr != NULL && r == &V_pf_default_rule)
6291                         tr = nr;
6292                 if (tr->src.addr.type == PF_ADDR_TABLE)
6293                         pfr_update_stats(tr->src.addr.p.tbl,
6294                             (s == NULL) ? pd.src :
6295                             &s->key[(s->direction == PF_IN)]->
6296                                 addr[(s->direction == PF_OUT)],
6297                             pd.af, pd.tot_len, dir == PF_OUT,
6298                             r->action == PF_PASS, tr->src.neg);
6299                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6300                         pfr_update_stats(tr->dst.addr.p.tbl,
6301                             (s == NULL) ? pd.dst :
6302                             &s->key[(s->direction == PF_IN)]->
6303                                 addr[(s->direction == PF_IN)],
6304                             pd.af, pd.tot_len, dir == PF_OUT,
6305                             r->action == PF_PASS, tr->dst.neg);
6306         }
6307
6308         switch (action) {
6309         case PF_SYNPROXY_DROP:
6310                 m_freem(*m0);
6311         case PF_DEFER:
6312                 *m0 = NULL;
6313                 action = PF_PASS;
6314                 break;
6315         case PF_DROP:
6316                 m_freem(*m0);
6317                 *m0 = NULL;
6318                 break;
6319         default:
6320                 /* pf_route() returns unlocked. */
6321                 if (r->rt) {
6322                         pf_route(m0, r, dir, kif->pfik_ifp, s, &pd, inp);
6323                         return (action);
6324                 }
6325                 break;
6326         }
6327         if (s)
6328                 PF_STATE_UNLOCK(s);
6329
6330         SDT_PROBE4(pf, ip, test, done, action, reason, r, s);
6331
6332         return (action);
6333 }
6334 #endif /* INET */
6335
6336 #ifdef INET6
6337 int
6338 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6339 {
6340         struct pfi_kkif         *kif;
6341         u_short                  action, reason = 0, log = 0;
6342         struct mbuf             *m = *m0, *n = NULL;
6343         struct m_tag            *mtag;
6344         struct ip6_hdr          *h = NULL;
6345         struct pf_krule         *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6346         struct pf_state         *s = NULL;
6347         struct pf_kruleset      *ruleset = NULL;
6348         struct pf_pdesc          pd;
6349         int                      off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0;
6350
6351         PF_RULES_RLOCK_TRACKER;
6352         M_ASSERTPKTHDR(m);
6353
6354         if (!V_pf_status.running)
6355                 return (PF_PASS);
6356
6357         memset(&pd, 0, sizeof(pd));
6358         pd.pf_mtag = pf_find_mtag(m);
6359
6360         if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6361                 return (PF_PASS);
6362
6363         kif = (struct pfi_kkif *)ifp->if_pf_kif;
6364         if (kif == NULL) {
6365                 DPFPRINTF(PF_DEBUG_URGENT,
6366                     ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6367                 return (PF_DROP);
6368         }
6369         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6370                 return (PF_PASS);
6371
6372         if (m->m_flags & M_SKIP_FIREWALL)
6373                 return (PF_PASS);
6374
6375         PF_RULES_RLOCK();
6376
6377         /* We do IP header normalization and packet reassembly here */
6378         if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6379                 action = PF_DROP;
6380                 goto done;
6381         }
6382         m = *m0;        /* pf_normalize messes with m0 */
6383         h = mtod(m, struct ip6_hdr *);
6384
6385         /*
6386          * we do not support jumbogram.  if we keep going, zero ip6_plen
6387          * will do something bad, so drop the packet for now.
6388          */
6389         if (htons(h->ip6_plen) == 0) {
6390                 action = PF_DROP;
6391                 REASON_SET(&reason, PFRES_NORM);        /*XXX*/
6392                 goto done;
6393         }
6394
6395         pd.src = (struct pf_addr *)&h->ip6_src;
6396         pd.dst = (struct pf_addr *)&h->ip6_dst;
6397         pd.sport = pd.dport = NULL;
6398         pd.ip_sum = NULL;
6399         pd.proto_sum = NULL;
6400         pd.dir = dir;
6401         pd.sidx = (dir == PF_IN) ? 0 : 1;
6402         pd.didx = (dir == PF_IN) ? 1 : 0;
6403         pd.af = AF_INET6;
6404         pd.tos = IPV6_DSCP(h);
6405         pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6406
6407         off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6408         pd.proto = h->ip6_nxt;
6409         do {
6410                 switch (pd.proto) {
6411                 case IPPROTO_FRAGMENT:
6412                         action = pf_test_fragment(&r, dir, kif, m, h,
6413                             &pd, &a, &ruleset);
6414                         if (action == PF_DROP)
6415                                 REASON_SET(&reason, PFRES_FRAG);
6416                         goto done;
6417                 case IPPROTO_ROUTING: {
6418                         struct ip6_rthdr rthdr;
6419
6420                         if (rh_cnt++) {
6421                                 DPFPRINTF(PF_DEBUG_MISC,
6422                                     ("pf: IPv6 more than one rthdr\n"));
6423                                 action = PF_DROP;
6424                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6425                                 log = 1;
6426                                 goto done;
6427                         }
6428                         if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6429                             &reason, pd.af)) {
6430                                 DPFPRINTF(PF_DEBUG_MISC,
6431                                     ("pf: IPv6 short rthdr\n"));
6432                                 action = PF_DROP;
6433                                 REASON_SET(&reason, PFRES_SHORT);
6434                                 log = 1;
6435                                 goto done;
6436                         }
6437                         if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6438                                 DPFPRINTF(PF_DEBUG_MISC,
6439                                     ("pf: IPv6 rthdr0\n"));
6440                                 action = PF_DROP;
6441                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6442                                 log = 1;
6443                                 goto done;
6444                         }
6445                         /* FALLTHROUGH */
6446                 }
6447                 case IPPROTO_AH:
6448                 case IPPROTO_HOPOPTS:
6449                 case IPPROTO_DSTOPTS: {
6450                         /* get next header and header length */
6451                         struct ip6_ext  opt6;
6452
6453                         if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6454                             NULL, &reason, pd.af)) {
6455                                 DPFPRINTF(PF_DEBUG_MISC,
6456                                     ("pf: IPv6 short opt\n"));
6457                                 action = PF_DROP;
6458                                 log = 1;
6459                                 goto done;
6460                         }
6461                         if (pd.proto == IPPROTO_AH)
6462                                 off += (opt6.ip6e_len + 2) * 4;
6463                         else
6464                                 off += (opt6.ip6e_len + 1) * 8;
6465                         pd.proto = opt6.ip6e_nxt;
6466                         /* goto the next header */
6467                         break;
6468                 }
6469                 default:
6470                         terminal++;
6471                         break;
6472                 }
6473         } while (!terminal);
6474
6475         /* if there's no routing header, use unmodified mbuf for checksumming */
6476         if (!n)
6477                 n = m;
6478
6479         switch (pd.proto) {
6480         case IPPROTO_TCP: {
6481                 struct tcphdr   th;
6482
6483                 pd.hdr.tcp = &th;
6484                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6485                     &action, &reason, AF_INET6)) {
6486                         log = action != PF_PASS;
6487                         goto done;
6488                 }
6489                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6490                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6491                 if (action == PF_DROP)
6492                         goto done;
6493                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6494                     &reason);
6495                 if (action == PF_PASS) {
6496                         if (V_pfsync_update_state_ptr != NULL)
6497                                 V_pfsync_update_state_ptr(s);
6498                         r = s->rule.ptr;
6499                         a = s->anchor.ptr;
6500                         log = s->log;
6501                 } else if (s == NULL)
6502                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6503                             &a, &ruleset, inp);
6504                 break;
6505         }
6506
6507         case IPPROTO_UDP: {
6508                 struct udphdr   uh;
6509
6510                 pd.hdr.udp = &uh;
6511                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6512                     &action, &reason, AF_INET6)) {
6513                         log = action != PF_PASS;
6514                         goto done;
6515                 }
6516                 if (uh.uh_dport == 0 ||
6517                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6518                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6519                         action = PF_DROP;
6520                         REASON_SET(&reason, PFRES_SHORT);
6521                         goto done;
6522                 }
6523                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6524                 if (action == PF_PASS) {
6525                         if (V_pfsync_update_state_ptr != NULL)
6526                                 V_pfsync_update_state_ptr(s);
6527                         r = s->rule.ptr;
6528                         a = s->anchor.ptr;
6529                         log = s->log;
6530                 } else if (s == NULL)
6531                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6532                             &a, &ruleset, inp);
6533                 break;
6534         }
6535
6536         case IPPROTO_ICMP: {
6537                 action = PF_DROP;
6538                 DPFPRINTF(PF_DEBUG_MISC,
6539                     ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6540                 goto done;
6541         }
6542
6543         case IPPROTO_ICMPV6: {
6544                 struct icmp6_hdr        ih;
6545
6546                 pd.hdr.icmp6 = &ih;
6547                 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6548                     &action, &reason, AF_INET6)) {
6549                         log = action != PF_PASS;
6550                         goto done;
6551                 }
6552                 action = pf_test_state_icmp(&s, dir, kif,
6553                     m, off, h, &pd, &reason);
6554                 if (action == PF_PASS) {
6555                         if (V_pfsync_update_state_ptr != NULL)
6556                                 V_pfsync_update_state_ptr(s);
6557                         r = s->rule.ptr;
6558                         a = s->anchor.ptr;
6559                         log = s->log;
6560                 } else if (s == NULL)
6561                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6562                             &a, &ruleset, inp);
6563                 break;
6564         }
6565
6566         default:
6567                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6568                 if (action == PF_PASS) {
6569                         if (V_pfsync_update_state_ptr != NULL)
6570                                 V_pfsync_update_state_ptr(s);
6571                         r = s->rule.ptr;
6572                         a = s->anchor.ptr;
6573                         log = s->log;
6574                 } else if (s == NULL)
6575                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6576                             &a, &ruleset, inp);
6577                 break;
6578         }
6579
6580 done:
6581         PF_RULES_RUNLOCK();
6582         if (n != m) {
6583                 m_freem(n);
6584                 n = NULL;
6585         }
6586
6587         /* handle dangerous IPv6 extension headers. */
6588         if (action == PF_PASS && rh_cnt &&
6589             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6590                 action = PF_DROP;
6591                 REASON_SET(&reason, PFRES_IPOPTIONS);
6592                 log = r->log;
6593                 DPFPRINTF(PF_DEBUG_MISC,
6594                     ("pf: dropping packet with dangerous v6 headers\n"));
6595         }
6596
6597         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6598                 action = PF_DROP;
6599                 REASON_SET(&reason, PFRES_MEMORY);
6600         }
6601         if (r->rtableid >= 0)
6602                 M_SETFIB(m, r->rtableid);
6603
6604         if (r->scrub_flags & PFSTATE_SETPRIO) {
6605                 if (pd.tos & IPTOS_LOWDELAY)
6606                         pqid = 1;
6607                 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6608                         action = PF_DROP;
6609                         REASON_SET(&reason, PFRES_MEMORY);
6610                         log = 1;
6611                         DPFPRINTF(PF_DEBUG_MISC,
6612                             ("pf: failed to allocate 802.1q mtag\n"));
6613                 }
6614         }
6615
6616 #ifdef ALTQ
6617         if (action == PF_PASS && r->qid) {
6618                 if (pd.pf_mtag == NULL &&
6619                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6620                         action = PF_DROP;
6621                         REASON_SET(&reason, PFRES_MEMORY);
6622                 } else {
6623                         if (s != NULL)
6624                                 pd.pf_mtag->qid_hash = pf_state_hash(s);
6625                         if (pd.tos & IPTOS_LOWDELAY)
6626                                 pd.pf_mtag->qid = r->pqid;
6627                         else
6628                                 pd.pf_mtag->qid = r->qid;
6629                         /* Add hints for ecn. */
6630                         pd.pf_mtag->hdr = h;
6631                 }
6632         }
6633 #endif /* ALTQ */
6634
6635         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6636             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6637             (s->nat_rule.ptr->action == PF_RDR ||
6638             s->nat_rule.ptr->action == PF_BINAT) &&
6639             IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6640                 m->m_flags |= M_SKIP_FIREWALL;
6641
6642         /* XXX: Anybody working on it?! */
6643         if (r->divert.port)
6644                 printf("pf: divert(9) is not supported for IPv6\n");
6645
6646         if (log) {
6647                 struct pf_krule *lr;
6648
6649                 if (s != NULL && s->nat_rule.ptr != NULL &&
6650                     s->nat_rule.ptr->log & PF_LOG_ALL)
6651                         lr = s->nat_rule.ptr;
6652                 else
6653                         lr = r;
6654                 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6655                     &pd, (s == NULL));
6656         }
6657
6658         counter_u64_add(kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS],
6659             pd.tot_len);
6660         counter_u64_add(kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS],
6661             1);
6662
6663         if (action == PF_PASS || r->action == PF_DROP) {
6664                 dirndx = (dir == PF_OUT);
6665                 counter_u64_add(r->packets[dirndx], 1);
6666                 counter_u64_add(r->bytes[dirndx], pd.tot_len);
6667                 if (a != NULL) {
6668                         counter_u64_add(a->packets[dirndx], 1);
6669                         counter_u64_add(a->bytes[dirndx], pd.tot_len);
6670                 }
6671                 if (s != NULL) {
6672                         if (s->nat_rule.ptr != NULL) {
6673                                 counter_u64_add(s->nat_rule.ptr->packets[dirndx],
6674                                     1);
6675                                 counter_u64_add(s->nat_rule.ptr->bytes[dirndx],
6676                                     pd.tot_len);
6677                         }
6678                         if (s->src_node != NULL) {
6679                                 counter_u64_add(s->src_node->packets[dirndx],
6680                                     1);
6681                                 counter_u64_add(s->src_node->bytes[dirndx],
6682                                     pd.tot_len);
6683                         }
6684                         if (s->nat_src_node != NULL) {
6685                                 counter_u64_add(s->nat_src_node->packets[dirndx],
6686                                     1);
6687                                 counter_u64_add(s->nat_src_node->bytes[dirndx],
6688                                     pd.tot_len);
6689                         }
6690                         dirndx = (dir == s->direction) ? 0 : 1;
6691                         counter_u64_add(s->packets[dirndx], 1);
6692                         counter_u64_add(s->bytes[dirndx], pd.tot_len);
6693                 }
6694                 tr = r;
6695                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6696                 if (nr != NULL && r == &V_pf_default_rule)
6697                         tr = nr;
6698                 if (tr->src.addr.type == PF_ADDR_TABLE)
6699                         pfr_update_stats(tr->src.addr.p.tbl,
6700                             (s == NULL) ? pd.src :
6701                             &s->key[(s->direction == PF_IN)]->addr[0],
6702                             pd.af, pd.tot_len, dir == PF_OUT,
6703                             r->action == PF_PASS, tr->src.neg);
6704                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6705                         pfr_update_stats(tr->dst.addr.p.tbl,
6706                             (s == NULL) ? pd.dst :
6707                             &s->key[(s->direction == PF_IN)]->addr[1],
6708                             pd.af, pd.tot_len, dir == PF_OUT,
6709                             r->action == PF_PASS, tr->dst.neg);
6710         }
6711
6712         switch (action) {
6713         case PF_SYNPROXY_DROP:
6714                 m_freem(*m0);
6715         case PF_DEFER:
6716                 *m0 = NULL;
6717                 action = PF_PASS;
6718                 break;
6719         case PF_DROP:
6720                 m_freem(*m0);
6721                 *m0 = NULL;
6722                 break;
6723         default:
6724                 /* pf_route6() returns unlocked. */
6725                 if (r->rt) {
6726                         pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd, inp);
6727                         return (action);
6728                 }
6729                 break;
6730         }
6731
6732         if (s)
6733                 PF_STATE_UNLOCK(s);
6734
6735         /* If reassembled packet passed, create new fragments. */
6736         if (action == PF_PASS && *m0 && (pflags & PFIL_FWD) &&
6737             (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL)
6738                 action = pf_refragment6(ifp, m0, mtag);
6739
6740         SDT_PROBE4(pf, ip, test6, done, action, reason, r, s);
6741
6742         return (action);
6743 }
6744 #endif /* INET6 */