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_inet.h"
  44 #include "opt_inet6.h"
  45 #include "opt_bpf.h"
  46 #include "opt_pf.h"
  47
  48 #include <sys/param.h>
  49 #include <sys/bus.h>
  50 #include <sys/endian.h>
  51 #include <sys/hash.h>
  52 #include <sys/interrupt.h>
  53 #include <sys/kernel.h>
  54 #include <sys/kthread.h>
  55 #include <sys/limits.h>
  56 #include <sys/mbuf.h>
  57 #include <sys/md5.h>
  58 #include <sys/random.h>
  59 #include <sys/refcount.h>
  60 #include <sys/socket.h>
  61 #include <sys/sysctl.h>
  62 #include <sys/taskqueue.h>
  63 #include <sys/ucred.h>
  64
  65 #include <net/if.h>
  66 #include <net/if_var.h>
  67 #include <net/if_types.h>
  68 #include <net/if_vlan_var.h>
  69 #include <net/route.h>
  70 #include <net/radix_mpath.h>
  71 #include <net/vnet.h>
  72
  73 #include <net/pfil.h>
  74 #include <net/pfvar.h>
  75 #include <net/if_pflog.h>
  76 #include <net/if_pfsync.h>
  77
  78 #include <netinet/in_pcb.h>
  79 #include <netinet/in_var.h>
  80 #include <netinet/in_fib.h>
  81 #include <netinet/ip.h>
  82 #include <netinet/ip_fw.h>
  83 #include <netinet/ip_icmp.h>
  84 #include <netinet/icmp_var.h>
  85 #include <netinet/ip_var.h>
  86 #include <netinet/tcp.h>
  87 #include <netinet/tcp_fsm.h>
  88 #include <netinet/tcp_seq.h>
  89 #include <netinet/tcp_timer.h>
  90 #include <netinet/tcp_var.h>
  91 #include <netinet/udp.h>
  92 #include <netinet/udp_var.h>
  93
  94 #include <netpfil/ipfw/ip_fw_private.h> /* XXX: only for DIR_IN/DIR_OUT */
  95
  96 #ifdef INET6
  97 #include <netinet/ip6.h>
  98 #include <netinet/icmp6.h>
  99 #include <netinet6/nd6.h>
 100 #include <netinet6/ip6_var.h>
 101 #include <netinet6/in6_pcb.h>
 102 #include <netinet6/in6_fib.h>
 103 #include <netinet6/scope6_var.h>
 104 #endif /* INET6 */
 105
 106 #include <machine/in_cksum.h>
 107 #include <security/mac/mac_framework.h>
 108
 109 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x
 110
 111 /*
 112  * Global variables
 113  */
 114
 115 /* state tables */
 116 VNET_DEFINE(struct pf_altqqueue,         pf_altqs[2]);
 117 VNET_DEFINE(struct pf_palist,            pf_pabuf);
 118 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_active);
 119 VNET_DEFINE(struct pf_altqqueue *,       pf_altqs_inactive);
 120 VNET_DEFINE(struct pf_kstatus,           pf_status);
 121
 122 VNET_DEFINE(u_int32_t,                   ticket_altqs_active);
 123 VNET_DEFINE(u_int32_t,                   ticket_altqs_inactive);
 124 VNET_DEFINE(int,                         altqs_inactive_open);
 125 VNET_DEFINE(u_int32_t,                   ticket_pabuf);
 126
 127 VNET_DEFINE(MD5_CTX,                     pf_tcp_secret_ctx);
 128 #define V_pf_tcp_secret_ctx              VNET(pf_tcp_secret_ctx)
 129 VNET_DEFINE(u_char,                      pf_tcp_secret[16]);
 130 #define V_pf_tcp_secret                  VNET(pf_tcp_secret)
 131 VNET_DEFINE(int,                         pf_tcp_secret_init);
 132 #define V_pf_tcp_secret_init             VNET(pf_tcp_secret_init)
 133 VNET_DEFINE(int,                         pf_tcp_iss_off);
 134 #define V_pf_tcp_iss_off                 VNET(pf_tcp_iss_off)
 135 VNET_DECLARE(int,                        pf_vnet_active);
 136 #define V_pf_vnet_active                 VNET(pf_vnet_active)
 137
 138 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx);
 139 #define V_pf_purge_idx  VNET(pf_purge_idx)
 140
 141 /*
 142  * Queue for pf_intr() sends.
 143  */
 144 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations");
 145 struct pf_send_entry {
 146         STAILQ_ENTRY(pf_send_entry)     pfse_next;
 147         struct mbuf                     *pfse_m;
 148         enum {
 149                 PFSE_IP,
 150                 PFSE_IP6,
 151                 PFSE_ICMP,
 152                 PFSE_ICMP6,
 153         }                               pfse_type;
 154         struct {
 155                 int             type;
 156                 int             code;
 157                 int             mtu;
 158         } icmpopts;
 159 };
 160
 161 STAILQ_HEAD(pf_send_head, pf_send_entry);
 162 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue);
 163 #define V_pf_sendqueue  VNET(pf_sendqueue)
 164
 165 static struct mtx pf_sendqueue_mtx;
 166 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF);
 167 #define PF_SENDQ_LOCK()         mtx_lock(&pf_sendqueue_mtx)
 168 #define PF_SENDQ_UNLOCK()       mtx_unlock(&pf_sendqueue_mtx)
 169
 170 /*
 171  * Queue for pf_overload_task() tasks.
 172  */
 173 struct pf_overload_entry {
 174         SLIST_ENTRY(pf_overload_entry)  next;
 175         struct pf_addr                  addr;
 176         sa_family_t                     af;
 177         uint8_t                         dir;
 178         struct pf_rule                  *rule;
 179 };
 180
 181 SLIST_HEAD(pf_overload_head, pf_overload_entry);
 182 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue);
 183 #define V_pf_overloadqueue      VNET(pf_overloadqueue)
 184 VNET_DEFINE_STATIC(struct task, pf_overloadtask);
 185 #define V_pf_overloadtask       VNET(pf_overloadtask)
 186
 187 static struct mtx pf_overloadqueue_mtx;
 188 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx,
 189     "pf overload/flush queue", MTX_DEF);
 190 #define PF_OVERLOADQ_LOCK()     mtx_lock(&pf_overloadqueue_mtx)
 191 #define PF_OVERLOADQ_UNLOCK()   mtx_unlock(&pf_overloadqueue_mtx)
 192
 193 VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules);
 194 struct mtx pf_unlnkdrules_mtx;
 195 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules",
 196     MTX_DEF);
 197
 198 VNET_DEFINE_STATIC(uma_zone_t,  pf_sources_z);
 199 #define V_pf_sources_z  VNET(pf_sources_z)
 200 uma_zone_t              pf_mtag_z;
 201 VNET_DEFINE(uma_zone_t,  pf_state_z);
 202 VNET_DEFINE(uma_zone_t,  pf_state_key_z);
 203
 204 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]);
 205 #define PFID_CPUBITS    8
 206 #define PFID_CPUSHIFT   (sizeof(uint64_t) * NBBY - PFID_CPUBITS)
 207 #define PFID_CPUMASK    ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT)
 208 #define PFID_MAXID      (~PFID_CPUMASK)
 209 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU);
 210
 211 static void              pf_src_tree_remove_state(struct pf_state *);
 212 static void              pf_init_threshold(struct pf_threshold *, u_int32_t,
 213                             u_int32_t);
 214 static void              pf_add_threshold(struct pf_threshold *);
 215 static int               pf_check_threshold(struct pf_threshold *);
 216
 217 static void              pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *,
 218                             u_int16_t *, u_int16_t *, struct pf_addr *,
 219                             u_int16_t, u_int8_t, sa_family_t);
 220 static int               pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *,
 221                             struct tcphdr *, struct pf_state_peer *);
 222 static void              pf_change_icmp(struct pf_addr *, u_int16_t *,
 223                             struct pf_addr *, struct pf_addr *, u_int16_t,
 224                             u_int16_t *, u_int16_t *, u_int16_t *,
 225                             u_int16_t *, u_int8_t, sa_family_t);
 226 static void              pf_send_tcp(struct mbuf *,
 227                             const struct pf_rule *, sa_family_t,
 228                             const struct pf_addr *, const struct pf_addr *,
 229                             u_int16_t, u_int16_t, u_int32_t, u_int32_t,
 230                             u_int8_t, u_int16_t, u_int16_t, u_int8_t, int,
 231                             u_int16_t, struct ifnet *);
 232 static void              pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t,
 233                             sa_family_t, struct pf_rule *);
 234 static void              pf_detach_state(struct pf_state *);
 235 static int               pf_state_key_attach(struct pf_state_key *,
 236                             struct pf_state_key *, struct pf_state *);
 237 static void              pf_state_key_detach(struct pf_state *, int);
 238 static int               pf_state_key_ctor(void *, int, void *, int);
 239 static u_int32_t         pf_tcp_iss(struct pf_pdesc *);
 240 static int               pf_test_rule(struct pf_rule **, struct pf_state **,
 241                             int, struct pfi_kif *, struct mbuf *, int,
 242                             struct pf_pdesc *, struct pf_rule **,
 243                             struct pf_ruleset **, struct inpcb *);
 244 static int               pf_create_state(struct pf_rule *, struct pf_rule *,
 245                             struct pf_rule *, struct pf_pdesc *,
 246                             struct pf_src_node *, struct pf_state_key *,
 247                             struct pf_state_key *, struct mbuf *, int,
 248                             u_int16_t, u_int16_t, int *, struct pfi_kif *,
 249                             struct pf_state **, int, u_int16_t, u_int16_t,
 250                             int);
 251 static int               pf_test_fragment(struct pf_rule **, int,
 252                             struct pfi_kif *, struct mbuf *, void *,
 253                             struct pf_pdesc *, struct pf_rule **,
 254                             struct pf_ruleset **);
 255 static int               pf_tcp_track_full(struct pf_state_peer *,
 256                             struct pf_state_peer *, struct pf_state **,
 257                             struct pfi_kif *, struct mbuf *, int,
 258                             struct pf_pdesc *, u_short *, int *);
 259 static int               pf_tcp_track_sloppy(struct pf_state_peer *,
 260                             struct pf_state_peer *, struct pf_state **,
 261                             struct pf_pdesc *, u_short *);
 262 static int               pf_test_state_tcp(struct pf_state **, int,
 263                             struct pfi_kif *, struct mbuf *, int,
 264                             void *, struct pf_pdesc *, u_short *);
 265 static int               pf_test_state_udp(struct pf_state **, int,
 266                             struct pfi_kif *, struct mbuf *, int,
 267                             void *, struct pf_pdesc *);
 268 static int               pf_test_state_icmp(struct pf_state **, int,
 269                             struct pfi_kif *, struct mbuf *, int,
 270                             void *, struct pf_pdesc *, u_short *);
 271 static int               pf_test_state_other(struct pf_state **, int,
 272                             struct pfi_kif *, struct mbuf *, struct pf_pdesc *);
 273 static u_int8_t          pf_get_wscale(struct mbuf *, int, u_int16_t,
 274                             sa_family_t);
 275 static u_int16_t         pf_get_mss(struct mbuf *, int, u_int16_t,
 276                             sa_family_t);
 277 static u_int16_t         pf_calc_mss(struct pf_addr *, sa_family_t,
 278                                 int, u_int16_t);
 279 static int               pf_check_proto_cksum(struct mbuf *, int, int,
 280                             u_int8_t, sa_family_t);
 281 static void              pf_print_state_parts(struct pf_state *,
 282                             struct pf_state_key *, struct pf_state_key *);
 283 static int               pf_addr_wrap_neq(struct pf_addr_wrap *,
 284                             struct pf_addr_wrap *);
 285 static struct pf_state  *pf_find_state(struct pfi_kif *,
 286                             struct pf_state_key_cmp *, u_int);
 287 static int               pf_src_connlimit(struct pf_state **);
 288 static void              pf_overload_task(void *v, int pending);
 289 static int               pf_insert_src_node(struct pf_src_node **,
 290                             struct pf_rule *, struct pf_addr *, sa_family_t);
 291 static u_int             pf_purge_expired_states(u_int, int);
 292 static void              pf_purge_unlinked_rules(void);
 293 static int               pf_mtag_uminit(void *, int, int);
 294 static void              pf_mtag_free(struct m_tag *);
 295 #ifdef INET
 296 static void              pf_route(struct mbuf **, struct pf_rule *, int,
 297                             struct ifnet *, struct pf_state *,
 298                             struct pf_pdesc *, struct inpcb *);
 299 #endif /* INET */
 300 #ifdef INET6
 301 static void              pf_change_a6(struct pf_addr *, u_int16_t *,
 302                             struct pf_addr *, u_int8_t);
 303 static void              pf_route6(struct mbuf **, struct pf_rule *, int,
 304                             struct ifnet *, struct pf_state *,
 305                             struct pf_pdesc *, struct inpcb *);
 306 #endif /* INET6 */
 307
 308 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len);
 309
 310 extern int pf_end_threads;
 311 extern struct proc *pf_purge_proc;
 312
 313 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]);
 314
 315 #define PACKET_LOOPED(pd)       ((pd)->pf_mtag &&                       \
 316                                  (pd)->pf_mtag->flags & PF_PACKET_LOOPED)
 317
 318 #define STATE_LOOKUP(i, k, d, s, pd)                                    \
 319         do {                                                            \
 320                 (s) = pf_find_state((i), (k), (d));                     \
 321                 if ((s) == NULL)                                        \
 322                         return (PF_DROP);                               \
 323                 if (PACKET_LOOPED(pd))                                  \
 324                         return (PF_PASS);                               \
 325                 if ((d) == PF_OUT &&                                    \
 326                     (((s)->rule.ptr->rt == PF_ROUTETO &&                \
 327                     (s)->rule.ptr->direction == PF_OUT) ||              \
 328                     ((s)->rule.ptr->rt == PF_REPLYTO &&                 \
 329                     (s)->rule.ptr->direction == PF_IN)) &&              \
 330                     (s)->rt_kif != NULL &&                              \
 331                     (s)->rt_kif != (i))                                 \
 332                         return (PF_PASS);                               \
 333         } while (0)
 334
 335 #define BOUND_IFACE(r, k) \
 336         ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all
 337
 338 #define STATE_INC_COUNTERS(s)                                           \
 339         do {                                                            \
 340                 counter_u64_add(s->rule.ptr->states_cur, 1);            \
 341                 counter_u64_add(s->rule.ptr->states_tot, 1);            \
 342                 if (s->anchor.ptr != NULL) {                            \
 343                         counter_u64_add(s->anchor.ptr->states_cur, 1);  \
 344                         counter_u64_add(s->anchor.ptr->states_tot, 1);  \
 345                 }                                                       \
 346                 if (s->nat_rule.ptr != NULL) {                          \
 347                         counter_u64_add(s->nat_rule.ptr->states_cur, 1);\
 348                         counter_u64_add(s->nat_rule.ptr->states_tot, 1);\
 349                 }                                                       \
 350         } while (0)
 351
 352 #define STATE_DEC_COUNTERS(s)                                           \
 353         do {                                                            \
 354                 if (s->nat_rule.ptr != NULL)                            \
 355                         counter_u64_add(s->nat_rule.ptr->states_cur, -1);\
 356                 if (s->anchor.ptr != NULL)                              \
 357                         counter_u64_add(s->anchor.ptr->states_cur, -1); \
 358                 counter_u64_add(s->rule.ptr->states_cur, -1);           \
 359         } while (0)
 360
 361 static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures");
 362 VNET_DEFINE(struct pf_keyhash *, pf_keyhash);
 363 VNET_DEFINE(struct pf_idhash *, pf_idhash);
 364 VNET_DEFINE(struct pf_srchash *, pf_srchash);
 365
 366 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)");
 367
 368 u_long  pf_hashmask;
 369 u_long  pf_srchashmask;
 370 static u_long   pf_hashsize;
 371 static u_long   pf_srchashsize;
 372 u_long  pf_ioctl_maxcount = 65535;
 373
 374 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN,
 375     &pf_hashsize, 0, "Size of pf(4) states hashtable");
 376 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN,
 377     &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable");
 378 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RDTUN,
 379     &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call");
 380
 381 VNET_DEFINE(void *, pf_swi_cookie);
 382
 383 VNET_DEFINE(uint32_t, pf_hashseed);
 384 #define V_pf_hashseed   VNET(pf_hashseed)
 385
 386 int
 387 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af)
 388 {
 389
 390         switch (af) {
 391 #ifdef INET
 392         case AF_INET:
 393                 if (a->addr32[0] > b->addr32[0])
 394                         return (1);
 395                 if (a->addr32[0] < b->addr32[0])
 396                         return (-1);
 397                 break;
 398 #endif /* INET */
 399 #ifdef INET6
 400         case AF_INET6:
 401                 if (a->addr32[3] > b->addr32[3])
 402                         return (1);
 403                 if (a->addr32[3] < b->addr32[3])
 404                         return (-1);
 405                 if (a->addr32[2] > b->addr32[2])
 406                         return (1);
 407                 if (a->addr32[2] < b->addr32[2])
 408                         return (-1);
 409                 if (a->addr32[1] > b->addr32[1])
 410                         return (1);
 411                 if (a->addr32[1] < b->addr32[1])
 412                         return (-1);
 413                 if (a->addr32[0] > b->addr32[0])
 414                         return (1);
 415                 if (a->addr32[0] < b->addr32[0])
 416                         return (-1);
 417                 break;
 418 #endif /* INET6 */
 419         default:
 420                 panic("%s: unknown address family %u", __func__, af);
 421         }
 422         return (0);
 423 }
 424
 425 static __inline uint32_t
 426 pf_hashkey(struct pf_state_key *sk)
 427 {
 428         uint32_t h;
 429
 430         h = murmur3_32_hash32((uint32_t *)sk,
 431             sizeof(struct pf_state_key_cmp)/sizeof(uint32_t),
 432             V_pf_hashseed);
 433
 434         return (h & pf_hashmask);
 435 }
 436
 437 static __inline uint32_t
 438 pf_hashsrc(struct pf_addr *addr, sa_family_t af)
 439 {
 440         uint32_t h;
 441
 442         switch (af) {
 443         case AF_INET:
 444                 h = murmur3_32_hash32((uint32_t *)&addr->v4,
 445                     sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed);
 446                 break;
 447         case AF_INET6:
 448                 h = murmur3_32_hash32((uint32_t *)&addr->v6,
 449                     sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed);
 450                 break;
 451         default:
 452                 panic("%s: unknown address family %u", __func__, af);
 453         }
 454
 455         return (h & pf_srchashmask);
 456 }
 457
 458 #ifdef ALTQ
 459 static int
 460 pf_state_hash(struct pf_state *s)
 461 {
 462         u_int32_t hv = (intptr_t)s / sizeof(*s);
 463
 464         hv ^= crc32(&s->src, sizeof(s->src));
 465         hv ^= crc32(&s->dst, sizeof(s->dst));
 466         if (hv == 0)
 467                 hv = 1;
 468         return (hv);
 469 }
 470 #endif
 471
 472 #ifdef INET6
 473 void
 474 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af)
 475 {
 476         switch (af) {
 477 #ifdef INET
 478         case AF_INET:
 479                 dst->addr32[0] = src->addr32[0];
 480                 break;
 481 #endif /* INET */
 482         case AF_INET6:
 483                 dst->addr32[0] = src->addr32[0];
 484                 dst->addr32[1] = src->addr32[1];
 485                 dst->addr32[2] = src->addr32[2];
 486                 dst->addr32[3] = src->addr32[3];
 487                 break;
 488         }
 489 }
 490 #endif /* INET6 */
 491
 492 static void
 493 pf_init_threshold(struct pf_threshold *threshold,
 494     u_int32_t limit, u_int32_t seconds)
 495 {
 496         threshold->limit = limit * PF_THRESHOLD_MULT;
 497         threshold->seconds = seconds;
 498         threshold->count = 0;
 499         threshold->last = time_uptime;
 500 }
 501
 502 static void
 503 pf_add_threshold(struct pf_threshold *threshold)
 504 {
 505         u_int32_t t = time_uptime, diff = t - threshold->last;
 506
 507         if (diff >= threshold->seconds)
 508                 threshold->count = 0;
 509         else
 510                 threshold->count -= threshold->count * diff /
 511                     threshold->seconds;
 512         threshold->count += PF_THRESHOLD_MULT;
 513         threshold->last = t;
 514 }
 515
 516 static int
 517 pf_check_threshold(struct pf_threshold *threshold)
 518 {
 519         return (threshold->count > threshold->limit);
 520 }
 521
 522 static int
 523 pf_src_connlimit(struct pf_state **state)
 524 {
 525         struct pf_overload_entry *pfoe;
 526         int bad = 0;
 527
 528         PF_STATE_LOCK_ASSERT(*state);
 529
 530         (*state)->src_node->conn++;
 531         (*state)->src.tcp_est = 1;
 532         pf_add_threshold(&(*state)->src_node->conn_rate);
 533
 534         if ((*state)->rule.ptr->max_src_conn &&
 535             (*state)->rule.ptr->max_src_conn <
 536             (*state)->src_node->conn) {
 537                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1);
 538                 bad++;
 539         }
 540
 541         if ((*state)->rule.ptr->max_src_conn_rate.limit &&
 542             pf_check_threshold(&(*state)->src_node->conn_rate)) {
 543                 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1);
 544                 bad++;
 545         }
 546
 547         if (!bad)
 548                 return (0);
 549
 550         /* Kill this state. */
 551         (*state)->timeout = PFTM_PURGE;
 552         (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
 553
 554         if ((*state)->rule.ptr->overload_tbl == NULL)
 555                 return (1);
 556
 557         /* Schedule overloading and flushing task. */
 558         pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT);
 559         if (pfoe == NULL)
 560                 return (1);     /* too bad :( */
 561
 562         bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr));
 563         pfoe->af = (*state)->key[PF_SK_WIRE]->af;
 564         pfoe->rule = (*state)->rule.ptr;
 565         pfoe->dir = (*state)->direction;
 566         PF_OVERLOADQ_LOCK();
 567         SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next);
 568         PF_OVERLOADQ_UNLOCK();
 569         taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask);
 570
 571         return (1);
 572 }
 573
 574 static void
 575 pf_overload_task(void *v, int pending)
 576 {
 577         struct pf_overload_head queue;
 578         struct pfr_addr p;
 579         struct pf_overload_entry *pfoe, *pfoe1;
 580         uint32_t killed = 0;
 581
 582         CURVNET_SET((struct vnet *)v);
 583
 584         PF_OVERLOADQ_LOCK();
 585         queue = V_pf_overloadqueue;
 586         SLIST_INIT(&V_pf_overloadqueue);
 587         PF_OVERLOADQ_UNLOCK();
 588
 589         bzero(&p, sizeof(p));
 590         SLIST_FOREACH(pfoe, &queue, next) {
 591                 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1);
 592                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
 593                         printf("%s: blocking address ", __func__);
 594                         pf_print_host(&pfoe->addr, 0, pfoe->af);
 595                         printf("\n");
 596                 }
 597
 598                 p.pfra_af = pfoe->af;
 599                 switch (pfoe->af) {
 600 #ifdef INET
 601                 case AF_INET:
 602                         p.pfra_net = 32;
 603                         p.pfra_ip4addr = pfoe->addr.v4;
 604                         break;
 605 #endif
 606 #ifdef INET6
 607                 case AF_INET6:
 608                         p.pfra_net = 128;
 609                         p.pfra_ip6addr = pfoe->addr.v6;
 610                         break;
 611 #endif
 612                 }
 613
 614                 PF_RULES_WLOCK();
 615                 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second);
 616                 PF_RULES_WUNLOCK();
 617         }
 618
 619         /*
 620          * Remove those entries, that don't need flushing.
 621          */
 622         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 623                 if (pfoe->rule->flush == 0) {
 624                         SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next);
 625                         free(pfoe, M_PFTEMP);
 626                 } else
 627                         counter_u64_add(
 628                             V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1);
 629
 630         /* If nothing to flush, return. */
 631         if (SLIST_EMPTY(&queue)) {
 632                 CURVNET_RESTORE();
 633                 return;
 634         }
 635
 636         for (int i = 0; i <= pf_hashmask; i++) {
 637                 struct pf_idhash *ih = &V_pf_idhash[i];
 638                 struct pf_state_key *sk;
 639                 struct pf_state *s;
 640
 641                 PF_HASHROW_LOCK(ih);
 642                 LIST_FOREACH(s, &ih->states, entry) {
 643                     sk = s->key[PF_SK_WIRE];
 644                     SLIST_FOREACH(pfoe, &queue, next)
 645                         if (sk->af == pfoe->af &&
 646                             ((pfoe->rule->flush & PF_FLUSH_GLOBAL) ||
 647                             pfoe->rule == s->rule.ptr) &&
 648                             ((pfoe->dir == PF_OUT &&
 649                             PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) ||
 650                             (pfoe->dir == PF_IN &&
 651                             PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) {
 652                                 s->timeout = PFTM_PURGE;
 653                                 s->src.state = s->dst.state = TCPS_CLOSED;
 654                                 killed++;
 655                         }
 656                 }
 657                 PF_HASHROW_UNLOCK(ih);
 658         }
 659         SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1)
 660                 free(pfoe, M_PFTEMP);
 661         if (V_pf_status.debug >= PF_DEBUG_MISC)
 662                 printf("%s: %u states killed", __func__, killed);
 663
 664         CURVNET_RESTORE();
 665 }
 666
 667 /*
 668  * Can return locked on failure, so that we can consistently
 669  * allocate and insert a new one.
 670  */
 671 struct pf_src_node *
 672 pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af,
 673         int returnlocked)
 674 {
 675         struct pf_srchash *sh;
 676         struct pf_src_node *n;
 677
 678         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1);
 679
 680         sh = &V_pf_srchash[pf_hashsrc(src, af)];
 681         PF_HASHROW_LOCK(sh);
 682         LIST_FOREACH(n, &sh->nodes, entry)
 683                 if (n->rule.ptr == rule && n->af == af &&
 684                     ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) ||
 685                     (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0)))
 686                         break;
 687         if (n != NULL) {
 688                 n->states++;
 689                 PF_HASHROW_UNLOCK(sh);
 690         } else if (returnlocked == 0)
 691                 PF_HASHROW_UNLOCK(sh);
 692
 693         return (n);
 694 }
 695
 696 static int
 697 pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule,
 698     struct pf_addr *src, sa_family_t af)
 699 {
 700
 701         KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK ||
 702             rule->rpool.opts & PF_POOL_STICKYADDR),
 703             ("%s for non-tracking rule %p", __func__, rule));
 704
 705         if (*sn == NULL)
 706                 *sn = pf_find_src_node(src, rule, af, 1);
 707
 708         if (*sn == NULL) {
 709                 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)];
 710
 711                 PF_HASHROW_ASSERT(sh);
 712
 713                 if (!rule->max_src_nodes ||
 714                     counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes)
 715                         (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO);
 716                 else
 717                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES],
 718                             1);
 719                 if ((*sn) == NULL) {
 720                         PF_HASHROW_UNLOCK(sh);
 721                         return (-1);
 722                 }
 723
 724                 pf_init_threshold(&(*sn)->conn_rate,
 725                     rule->max_src_conn_rate.limit,
 726                     rule->max_src_conn_rate.seconds);
 727
 728                 (*sn)->af = af;
 729                 (*sn)->rule.ptr = rule;
 730                 PF_ACPY(&(*sn)->addr, src, af);
 731                 LIST_INSERT_HEAD(&sh->nodes, *sn, entry);
 732                 (*sn)->creation = time_uptime;
 733                 (*sn)->ruletype = rule->action;
 734                 (*sn)->states = 1;
 735                 if ((*sn)->rule.ptr != NULL)
 736                         counter_u64_add((*sn)->rule.ptr->src_nodes, 1);
 737                 PF_HASHROW_UNLOCK(sh);
 738                 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1);
 739         } else {
 740                 if (rule->max_src_states &&
 741                     (*sn)->states >= rule->max_src_states) {
 742                         counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES],
 743                             1);
 744                         return (-1);
 745                 }
 746         }
 747         return (0);
 748 }
 749
 750 void
 751 pf_unlink_src_node(struct pf_src_node *src)
 752 {
 753
 754         PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]);
 755         LIST_REMOVE(src, entry);
 756         if (src->rule.ptr)
 757                 counter_u64_add(src->rule.ptr->src_nodes, -1);
 758 }
 759
 760 u_int
 761 pf_free_src_nodes(struct pf_src_node_list *head)
 762 {
 763         struct pf_src_node *sn, *tmp;
 764         u_int count = 0;
 765
 766         LIST_FOREACH_SAFE(sn, head, entry, tmp) {
 767                 uma_zfree(V_pf_sources_z, sn);
 768                 count++;
 769         }
 770
 771         counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count);
 772
 773         return (count);
 774 }
 775
 776 void
 777 pf_mtag_initialize()
 778 {
 779
 780         pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) +
 781             sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL,
 782             UMA_ALIGN_PTR, 0);
 783 }
 784
 785 /* Per-vnet data storage structures initialization. */
 786 void
 787 pf_initialize()
 788 {
 789         struct pf_keyhash       *kh;
 790         struct pf_idhash        *ih;
 791         struct pf_srchash       *sh;
 792         u_int i;
 793
 794         if (pf_hashsize == 0 || !powerof2(pf_hashsize))
 795                 pf_hashsize = PF_HASHSIZ;
 796         if (pf_srchashsize == 0 || !powerof2(pf_srchashsize))
 797                 pf_srchashsize = PF_SRCHASHSIZ;
 798
 799         V_pf_hashseed = arc4random();
 800
 801         /* States and state keys storage. */
 802         V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state),
 803             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 804         V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z;
 805         uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT);
 806         uma_zone_set_warning(V_pf_state_z, "PF states limit reached");
 807
 808         V_pf_state_key_z = uma_zcreate("pf state keys",
 809             sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL,
 810             UMA_ALIGN_PTR, 0);
 811
 812         V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash),
 813             M_PFHASH, M_NOWAIT | M_ZERO);
 814         V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash),
 815             M_PFHASH, M_NOWAIT | M_ZERO);
 816         if (V_pf_keyhash == NULL || V_pf_idhash == NULL) {
 817                 printf("pf: Unable to allocate memory for "
 818                     "state_hashsize %lu.\n", pf_hashsize);
 819
 820                 free(V_pf_keyhash, M_PFHASH);
 821                 free(V_pf_idhash, M_PFHASH);
 822
 823                 pf_hashsize = PF_HASHSIZ;
 824                 V_pf_keyhash = mallocarray(pf_hashsize,
 825                     sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO);
 826                 V_pf_idhash = mallocarray(pf_hashsize,
 827                     sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO);
 828         }
 829
 830         pf_hashmask = pf_hashsize - 1;
 831         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 832             i++, kh++, ih++) {
 833                 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK);
 834                 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF);
 835         }
 836
 837         /* Source nodes. */
 838         V_pf_sources_z = uma_zcreate("pf source nodes",
 839             sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
 840             0);
 841         V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z;
 842         uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT);
 843         uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached");
 844
 845         V_pf_srchash = mallocarray(pf_srchashsize,
 846             sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO);
 847         if (V_pf_srchash == NULL) {
 848                 printf("pf: Unable to allocate memory for "
 849                     "source_hashsize %lu.\n", pf_srchashsize);
 850
 851                 pf_srchashsize = PF_SRCHASHSIZ;
 852                 V_pf_srchash = mallocarray(pf_srchashsize,
 853                     sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO);
 854         }
 855
 856         pf_srchashmask = pf_srchashsize - 1;
 857         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++)
 858                 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF);
 859
 860         /* ALTQ */
 861         TAILQ_INIT(&V_pf_altqs[0]);
 862         TAILQ_INIT(&V_pf_altqs[1]);
 863         TAILQ_INIT(&V_pf_pabuf);
 864         V_pf_altqs_active = &V_pf_altqs[0];
 865         V_pf_altqs_inactive = &V_pf_altqs[1];
 866
 867         /* Send & overload+flush queues. */
 868         STAILQ_INIT(&V_pf_sendqueue);
 869         SLIST_INIT(&V_pf_overloadqueue);
 870         TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet);
 871
 872         /* Unlinked, but may be referenced rules. */
 873         TAILQ_INIT(&V_pf_unlinked_rules);
 874 }
 875
 876 void
 877 pf_mtag_cleanup()
 878 {
 879
 880         uma_zdestroy(pf_mtag_z);
 881 }
 882
 883 void
 884 pf_cleanup()
 885 {
 886         struct pf_keyhash       *kh;
 887         struct pf_idhash        *ih;
 888         struct pf_srchash       *sh;
 889         struct pf_send_entry    *pfse, *next;
 890         u_int i;
 891
 892         for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask;
 893             i++, kh++, ih++) {
 894                 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty",
 895                     __func__));
 896                 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty",
 897                     __func__));
 898                 mtx_destroy(&kh->lock);
 899                 mtx_destroy(&ih->lock);
 900         }
 901         free(V_pf_keyhash, M_PFHASH);
 902         free(V_pf_idhash, M_PFHASH);
 903
 904         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
 905                 KASSERT(LIST_EMPTY(&sh->nodes),
 906                     ("%s: source node hash not empty", __func__));
 907                 mtx_destroy(&sh->lock);
 908         }
 909         free(V_pf_srchash, M_PFHASH);
 910
 911         STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) {
 912                 m_freem(pfse->pfse_m);
 913                 free(pfse, M_PFTEMP);
 914         }
 915
 916         uma_zdestroy(V_pf_sources_z);
 917         uma_zdestroy(V_pf_state_z);
 918         uma_zdestroy(V_pf_state_key_z);
 919 }
 920
 921 static int
 922 pf_mtag_uminit(void *mem, int size, int how)
 923 {
 924         struct m_tag *t;
 925
 926         t = (struct m_tag *)mem;
 927         t->m_tag_cookie = MTAG_ABI_COMPAT;
 928         t->m_tag_id = PACKET_TAG_PF;
 929         t->m_tag_len = sizeof(struct pf_mtag);
 930         t->m_tag_free = pf_mtag_free;
 931
 932         return (0);
 933 }
 934
 935 static void
 936 pf_mtag_free(struct m_tag *t)
 937 {
 938
 939         uma_zfree(pf_mtag_z, t);
 940 }
 941
 942 struct pf_mtag *
 943 pf_get_mtag(struct mbuf *m)
 944 {
 945         struct m_tag *mtag;
 946
 947         if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL)
 948                 return ((struct pf_mtag *)(mtag + 1));
 949
 950         mtag = uma_zalloc(pf_mtag_z, M_NOWAIT);
 951         if (mtag == NULL)
 952                 return (NULL);
 953         bzero(mtag + 1, sizeof(struct pf_mtag));
 954         m_tag_prepend(m, mtag);
 955
 956         return ((struct pf_mtag *)(mtag + 1));
 957 }
 958
 959 static int
 960 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks,
 961     struct pf_state *s)
 962 {
 963         struct pf_keyhash       *khs, *khw, *kh;
 964         struct pf_state_key     *sk, *cur;
 965         struct pf_state         *si, *olds = NULL;
 966         int idx;
 967
 968         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
 969         KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__));
 970         KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__));
 971
 972         /*
 973          * We need to lock hash slots of both keys. To avoid deadlock
 974          * we always lock the slot with lower address first. Unlock order
 975          * isn't important.
 976          *
 977          * We also need to lock ID hash slot before dropping key
 978          * locks. On success we return with ID hash slot locked.
 979          */
 980
 981         if (skw == sks) {
 982                 khs = khw = &V_pf_keyhash[pf_hashkey(skw)];
 983                 PF_HASHROW_LOCK(khs);
 984         } else {
 985                 khs = &V_pf_keyhash[pf_hashkey(sks)];
 986                 khw = &V_pf_keyhash[pf_hashkey(skw)];
 987                 if (khs == khw) {
 988                         PF_HASHROW_LOCK(khs);
 989                 } else if (khs < khw) {
 990                         PF_HASHROW_LOCK(khs);
 991                         PF_HASHROW_LOCK(khw);
 992                 } else {
 993                         PF_HASHROW_LOCK(khw);
 994                         PF_HASHROW_LOCK(khs);
 995                 }
 996         }
 997
 998 #define KEYS_UNLOCK()   do {                    \
 999         if (khs != khw) {                       \
1000                 PF_HASHROW_UNLOCK(khs);         \
1001                 PF_HASHROW_UNLOCK(khw);         \
1002         } else                                  \
1003                 PF_HASHROW_UNLOCK(khs);         \
1004 } while (0)
1005
1006         /*
1007          * First run: start with wire key.
1008          */
1009         sk = skw;
1010         kh = khw;
1011         idx = PF_SK_WIRE;
1012
1013 keyattach:
1014         LIST_FOREACH(cur, &kh->keys, entry)
1015                 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0)
1016                         break;
1017
1018         if (cur != NULL) {
1019                 /* Key exists. Check for same kif, if none, add to key. */
1020                 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) {
1021                         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)];
1022
1023                         PF_HASHROW_LOCK(ih);
1024                         if (si->kif == s->kif &&
1025                             si->direction == s->direction) {
1026                                 if (sk->proto == IPPROTO_TCP &&
1027                                     si->src.state >= TCPS_FIN_WAIT_2 &&
1028                                     si->dst.state >= TCPS_FIN_WAIT_2) {
1029                                         /*
1030                                          * New state matches an old >FIN_WAIT_2
1031                                          * state. We can't drop key hash locks,
1032                                          * thus we can't unlink it properly.
1033                                          *
1034                                          * As a workaround we drop it into
1035                                          * TCPS_CLOSED state, schedule purge
1036                                          * ASAP and push it into the very end
1037                                          * of the slot TAILQ, so that it won't
1038                                          * conflict with our new state.
1039                                          */
1040                                         si->src.state = si->dst.state =
1041                                             TCPS_CLOSED;
1042                                         si->timeout = PFTM_PURGE;
1043                                         olds = si;
1044                                 } else {
1045                                         if (V_pf_status.debug >= PF_DEBUG_MISC) {
1046                                                 printf("pf: %s key attach "
1047                                                     "failed on %s: ",
1048                                                     (idx == PF_SK_WIRE) ?
1049                                                     "wire" : "stack",
1050                                                     s->kif->pfik_name);
1051                                                 pf_print_state_parts(s,
1052                                                     (idx == PF_SK_WIRE) ?
1053                                                     sk : NULL,
1054                                                     (idx == PF_SK_STACK) ?
1055                                                     sk : NULL);
1056                                                 printf(", existing: ");
1057                                                 pf_print_state_parts(si,
1058                                                     (idx == PF_SK_WIRE) ?
1059                                                     sk : NULL,
1060                                                     (idx == PF_SK_STACK) ?
1061                                                     sk : NULL);
1062                                                 printf("\n");
1063                                         }
1064                                         PF_HASHROW_UNLOCK(ih);
1065                                         KEYS_UNLOCK();
1066                                         uma_zfree(V_pf_state_key_z, sk);
1067                                         if (idx == PF_SK_STACK)
1068                                                 pf_detach_state(s);
1069                                         return (EEXIST); /* collision! */
1070                                 }
1071                         }
1072                         PF_HASHROW_UNLOCK(ih);
1073                 }
1074                 uma_zfree(V_pf_state_key_z, sk);
1075                 s->key[idx] = cur;
1076         } else {
1077                 LIST_INSERT_HEAD(&kh->keys, sk, entry);
1078                 s->key[idx] = sk;
1079         }
1080
1081 stateattach:
1082         /* List is sorted, if-bound states before floating. */
1083         if (s->kif == V_pfi_all)
1084                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]);
1085         else
1086                 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]);
1087
1088         if (olds) {
1089                 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]);
1090                 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds,
1091                     key_list[idx]);
1092                 olds = NULL;
1093         }
1094
1095         /*
1096          * Attach done. See how should we (or should not?)
1097          * attach a second key.
1098          */
1099         if (sks == skw) {
1100                 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE];
1101                 idx = PF_SK_STACK;
1102                 sks = NULL;
1103                 goto stateattach;
1104         } else if (sks != NULL) {
1105                 /*
1106                  * Continue attaching with stack key.
1107                  */
1108                 sk = sks;
1109                 kh = khs;
1110                 idx = PF_SK_STACK;
1111                 sks = NULL;
1112                 goto keyattach;
1113         }
1114
1115         PF_STATE_LOCK(s);
1116         KEYS_UNLOCK();
1117
1118         KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL,
1119             ("%s failure", __func__));
1120
1121         return (0);
1122 #undef  KEYS_UNLOCK
1123 }
1124
1125 static void
1126 pf_detach_state(struct pf_state *s)
1127 {
1128         struct pf_state_key *sks = s->key[PF_SK_STACK];
1129         struct pf_keyhash *kh;
1130
1131         if (sks != NULL) {
1132                 kh = &V_pf_keyhash[pf_hashkey(sks)];
1133                 PF_HASHROW_LOCK(kh);
1134                 if (s->key[PF_SK_STACK] != NULL)
1135                         pf_state_key_detach(s, PF_SK_STACK);
1136                 /*
1137                  * If both point to same key, then we are done.
1138                  */
1139                 if (sks == s->key[PF_SK_WIRE]) {
1140                         pf_state_key_detach(s, PF_SK_WIRE);
1141                         PF_HASHROW_UNLOCK(kh);
1142                         return;
1143                 }
1144                 PF_HASHROW_UNLOCK(kh);
1145         }
1146
1147         if (s->key[PF_SK_WIRE] != NULL) {
1148                 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])];
1149                 PF_HASHROW_LOCK(kh);
1150                 if (s->key[PF_SK_WIRE] != NULL)
1151                         pf_state_key_detach(s, PF_SK_WIRE);
1152                 PF_HASHROW_UNLOCK(kh);
1153         }
1154 }
1155
1156 static void
1157 pf_state_key_detach(struct pf_state *s, int idx)
1158 {
1159         struct pf_state_key *sk = s->key[idx];
1160 #ifdef INVARIANTS
1161         struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)];
1162
1163         PF_HASHROW_ASSERT(kh);
1164 #endif
1165         TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]);
1166         s->key[idx] = NULL;
1167
1168         if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) {
1169                 LIST_REMOVE(sk, entry);
1170                 uma_zfree(V_pf_state_key_z, sk);
1171         }
1172 }
1173
1174 static int
1175 pf_state_key_ctor(void *mem, int size, void *arg, int flags)
1176 {
1177         struct pf_state_key *sk = mem;
1178
1179         bzero(sk, sizeof(struct pf_state_key_cmp));
1180         TAILQ_INIT(&sk->states[PF_SK_WIRE]);
1181         TAILQ_INIT(&sk->states[PF_SK_STACK]);
1182
1183         return (0);
1184 }
1185
1186 struct pf_state_key *
1187 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr,
1188         struct pf_addr *daddr, u_int16_t sport, u_int16_t dport)
1189 {
1190         struct pf_state_key *sk;
1191
1192         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1193         if (sk == NULL)
1194                 return (NULL);
1195
1196         PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af);
1197         PF_ACPY(&sk->addr[pd->didx], daddr, pd->af);
1198         sk->port[pd->sidx] = sport;
1199         sk->port[pd->didx] = dport;
1200         sk->proto = pd->proto;
1201         sk->af = pd->af;
1202
1203         return (sk);
1204 }
1205
1206 struct pf_state_key *
1207 pf_state_key_clone(struct pf_state_key *orig)
1208 {
1209         struct pf_state_key *sk;
1210
1211         sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT);
1212         if (sk == NULL)
1213                 return (NULL);
1214
1215         bcopy(orig, sk, sizeof(struct pf_state_key_cmp));
1216
1217         return (sk);
1218 }
1219
1220 int
1221 pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw,
1222     struct pf_state_key *sks, struct pf_state *s)
1223 {
1224         struct pf_idhash *ih;
1225         struct pf_state *cur;
1226         int error;
1227
1228         KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]),
1229             ("%s: sks not pristine", __func__));
1230         KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]),
1231             ("%s: skw not pristine", __func__));
1232         KASSERT(s->refs == 0, ("%s: state not pristine", __func__));
1233
1234         s->kif = kif;
1235
1236         if (s->id == 0 && s->creatorid == 0) {
1237                 /* XXX: should be atomic, but probability of collision low */
1238                 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID)
1239                         V_pf_stateid[curcpu] = 1;
1240                 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT;
1241                 s->id = htobe64(s->id);
1242                 s->creatorid = V_pf_status.hostid;
1243         }
1244
1245         /* Returns with ID locked on success. */
1246         if ((error = pf_state_key_attach(skw, sks, s)) != 0)
1247                 return (error);
1248
1249         ih = &V_pf_idhash[PF_IDHASH(s)];
1250         PF_HASHROW_ASSERT(ih);
1251         LIST_FOREACH(cur, &ih->states, entry)
1252                 if (cur->id == s->id && cur->creatorid == s->creatorid)
1253                         break;
1254
1255         if (cur != NULL) {
1256                 PF_HASHROW_UNLOCK(ih);
1257                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
1258                         printf("pf: state ID collision: "
1259                             "id: %016llx creatorid: %08x\n",
1260                             (unsigned long long)be64toh(s->id),
1261                             ntohl(s->creatorid));
1262                 }
1263                 pf_detach_state(s);
1264                 return (EEXIST);
1265         }
1266         LIST_INSERT_HEAD(&ih->states, s, entry);
1267         /* One for keys, one for ID hash. */
1268         refcount_init(&s->refs, 2);
1269
1270         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1);
1271         if (pfsync_insert_state_ptr != NULL)
1272                 pfsync_insert_state_ptr(s);
1273
1274         /* Returns locked. */
1275         return (0);
1276 }
1277
1278 /*
1279  * Find state by ID: returns with locked row on success.
1280  */
1281 struct pf_state *
1282 pf_find_state_byid(uint64_t id, uint32_t creatorid)
1283 {
1284         struct pf_idhash *ih;
1285         struct pf_state *s;
1286
1287         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1288
1289         ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))];
1290
1291         PF_HASHROW_LOCK(ih);
1292         LIST_FOREACH(s, &ih->states, entry)
1293                 if (s->id == id && s->creatorid == creatorid)
1294                         break;
1295
1296         if (s == NULL)
1297                 PF_HASHROW_UNLOCK(ih);
1298
1299         return (s);
1300 }
1301
1302 /*
1303  * Find state by key.
1304  * Returns with ID hash slot locked on success.
1305  */
1306 static struct pf_state *
1307 pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir)
1308 {
1309         struct pf_keyhash       *kh;
1310         struct pf_state_key     *sk;
1311         struct pf_state         *s;
1312         int idx;
1313
1314         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1315
1316         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1317
1318         PF_HASHROW_LOCK(kh);
1319         LIST_FOREACH(sk, &kh->keys, entry)
1320                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1321                         break;
1322         if (sk == NULL) {
1323                 PF_HASHROW_UNLOCK(kh);
1324                 return (NULL);
1325         }
1326
1327         idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK);
1328
1329         /* List is sorted, if-bound states before floating ones. */
1330         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx])
1331                 if (s->kif == V_pfi_all || s->kif == kif) {
1332                         PF_STATE_LOCK(s);
1333                         PF_HASHROW_UNLOCK(kh);
1334                         if (s->timeout >= PFTM_MAX) {
1335                                 /*
1336                                  * State is either being processed by
1337                                  * pf_unlink_state() in an other thread, or
1338                                  * is scheduled for immediate expiry.
1339                                  */
1340                                 PF_STATE_UNLOCK(s);
1341                                 return (NULL);
1342                         }
1343                         return (s);
1344                 }
1345         PF_HASHROW_UNLOCK(kh);
1346
1347         return (NULL);
1348 }
1349
1350 struct pf_state *
1351 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more)
1352 {
1353         struct pf_keyhash       *kh;
1354         struct pf_state_key     *sk;
1355         struct pf_state         *s, *ret = NULL;
1356         int                      idx, inout = 0;
1357
1358         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1);
1359
1360         kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)];
1361
1362         PF_HASHROW_LOCK(kh);
1363         LIST_FOREACH(sk, &kh->keys, entry)
1364                 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0)
1365                         break;
1366         if (sk == NULL) {
1367                 PF_HASHROW_UNLOCK(kh);
1368                 return (NULL);
1369         }
1370         switch (dir) {
1371         case PF_IN:
1372                 idx = PF_SK_WIRE;
1373                 break;
1374         case PF_OUT:
1375                 idx = PF_SK_STACK;
1376                 break;
1377         case PF_INOUT:
1378                 idx = PF_SK_WIRE;
1379                 inout = 1;
1380                 break;
1381         default:
1382                 panic("%s: dir %u", __func__, dir);
1383         }
1384 second_run:
1385         TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) {
1386                 if (more == NULL) {
1387                         PF_HASHROW_UNLOCK(kh);
1388                         return (s);
1389                 }
1390
1391                 if (ret)
1392                         (*more)++;
1393                 else
1394                         ret = s;
1395         }
1396         if (inout == 1) {
1397                 inout = 0;
1398                 idx = PF_SK_STACK;
1399                 goto second_run;
1400         }
1401         PF_HASHROW_UNLOCK(kh);
1402
1403         return (ret);
1404 }
1405
1406 /* END state table stuff */
1407
1408 static void
1409 pf_send(struct pf_send_entry *pfse)
1410 {
1411
1412         PF_SENDQ_LOCK();
1413         STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next);
1414         PF_SENDQ_UNLOCK();
1415         swi_sched(V_pf_swi_cookie, 0);
1416 }
1417
1418 void
1419 pf_intr(void *v)
1420 {
1421         struct pf_send_head queue;
1422         struct pf_send_entry *pfse, *next;
1423
1424         CURVNET_SET((struct vnet *)v);
1425
1426         PF_SENDQ_LOCK();
1427         queue = V_pf_sendqueue;
1428         STAILQ_INIT(&V_pf_sendqueue);
1429         PF_SENDQ_UNLOCK();
1430
1431         STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) {
1432                 switch (pfse->pfse_type) {
1433 #ifdef INET
1434                 case PFSE_IP:
1435                         ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL);
1436                         break;
1437                 case PFSE_ICMP:
1438                         icmp_error(pfse->pfse_m, pfse->icmpopts.type,
1439                             pfse->icmpopts.code, 0, pfse->icmpopts.mtu);
1440                         break;
1441 #endif /* INET */
1442 #ifdef INET6
1443                 case PFSE_IP6:
1444                         ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL,
1445                             NULL);
1446                         break;
1447                 case PFSE_ICMP6:
1448                         icmp6_error(pfse->pfse_m, pfse->icmpopts.type,
1449                             pfse->icmpopts.code, pfse->icmpopts.mtu);
1450                         break;
1451 #endif /* INET6 */
1452                 default:
1453                         panic("%s: unknown type", __func__);
1454                 }
1455                 free(pfse, M_PFTEMP);
1456         }
1457         CURVNET_RESTORE();
1458 }
1459
1460 void
1461 pf_purge_thread(void *unused __unused)
1462 {
1463         VNET_ITERATOR_DECL(vnet_iter);
1464
1465         sx_xlock(&pf_end_lock);
1466         while (pf_end_threads == 0) {
1467                 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", hz / 10);
1468
1469                 VNET_LIST_RLOCK();
1470                 VNET_FOREACH(vnet_iter) {
1471                         CURVNET_SET(vnet_iter);
1472
1473
1474                         /* Wait until V_pf_default_rule is initialized. */
1475                         if (V_pf_vnet_active == 0) {
1476                                 CURVNET_RESTORE();
1477                                 continue;
1478                         }
1479
1480                         /*
1481                          *  Process 1/interval fraction of the state
1482                          * table every run.
1483                          */
1484                         V_pf_purge_idx =
1485                             pf_purge_expired_states(V_pf_purge_idx, pf_hashmask /
1486                             (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10));
1487
1488                         /*
1489                          * Purge other expired types every
1490                          * PFTM_INTERVAL seconds.
1491                          */
1492                         if (V_pf_purge_idx == 0) {
1493                                 /*
1494                                  * Order is important:
1495                                  * - states and src nodes reference rules
1496                                  * - states and rules reference kifs
1497                                  */
1498                                 pf_purge_expired_fragments();
1499                                 pf_purge_expired_src_nodes();
1500                                 pf_purge_unlinked_rules();
1501                                 pfi_kif_purge();
1502                         }
1503                         CURVNET_RESTORE();
1504                 }
1505                 VNET_LIST_RUNLOCK();
1506         }
1507
1508         pf_end_threads++;
1509         sx_xunlock(&pf_end_lock);
1510         kproc_exit(0);
1511 }
1512
1513 void
1514 pf_unload_vnet_purge(void)
1515 {
1516
1517         /*
1518          * To cleanse up all kifs and rules we need
1519          * two runs: first one clears reference flags,
1520          * then pf_purge_expired_states() doesn't
1521          * raise them, and then second run frees.
1522          */
1523         pf_purge_unlinked_rules();
1524         pfi_kif_purge();
1525
1526         /*
1527          * Now purge everything.
1528          */
1529         pf_purge_expired_states(0, pf_hashmask);
1530         pf_purge_fragments(UINT_MAX);
1531         pf_purge_expired_src_nodes();
1532
1533         /*
1534          * Now all kifs & rules should be unreferenced,
1535          * thus should be successfully freed.
1536          */
1537         pf_purge_unlinked_rules();
1538         pfi_kif_purge();
1539 }
1540
1541
1542 u_int32_t
1543 pf_state_expires(const struct pf_state *state)
1544 {
1545         u_int32_t       timeout;
1546         u_int32_t       start;
1547         u_int32_t       end;
1548         u_int32_t       states;
1549
1550         /* handle all PFTM_* > PFTM_MAX here */
1551         if (state->timeout == PFTM_PURGE)
1552                 return (time_uptime);
1553         KASSERT(state->timeout != PFTM_UNLINKED,
1554             ("pf_state_expires: timeout == PFTM_UNLINKED"));
1555         KASSERT((state->timeout < PFTM_MAX),
1556             ("pf_state_expires: timeout > PFTM_MAX"));
1557         timeout = state->rule.ptr->timeout[state->timeout];
1558         if (!timeout)
1559                 timeout = V_pf_default_rule.timeout[state->timeout];
1560         start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START];
1561         if (start) {
1562                 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END];
1563                 states = counter_u64_fetch(state->rule.ptr->states_cur);
1564         } else {
1565                 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START];
1566                 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END];
1567                 states = V_pf_status.states;
1568         }
1569         if (end && states > start && start < end) {
1570                 if (states < end)
1571                         return (state->expire + timeout * (end - states) /
1572                             (end - start));
1573                 else
1574                         return (time_uptime);
1575         }
1576         return (state->expire + timeout);
1577 }
1578
1579 void
1580 pf_purge_expired_src_nodes()
1581 {
1582         struct pf_src_node_list  freelist;
1583         struct pf_srchash       *sh;
1584         struct pf_src_node      *cur, *next;
1585         int i;
1586
1587         LIST_INIT(&freelist);
1588         for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) {
1589             PF_HASHROW_LOCK(sh);
1590             LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next)
1591                 if (cur->states == 0 && cur->expire <= time_uptime) {
1592                         pf_unlink_src_node(cur);
1593                         LIST_INSERT_HEAD(&freelist, cur, entry);
1594                 } else if (cur->rule.ptr != NULL)
1595                         cur->rule.ptr->rule_flag |= PFRULE_REFS;
1596             PF_HASHROW_UNLOCK(sh);
1597         }
1598
1599         pf_free_src_nodes(&freelist);
1600
1601         V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z);
1602 }
1603
1604 static void
1605 pf_src_tree_remove_state(struct pf_state *s)
1606 {
1607         struct pf_src_node *sn;
1608         struct pf_srchash *sh;
1609         uint32_t timeout;
1610
1611         timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ?
1612             s->rule.ptr->timeout[PFTM_SRC_NODE] :
1613             V_pf_default_rule.timeout[PFTM_SRC_NODE];
1614
1615         if (s->src_node != NULL) {
1616                 sn = s->src_node;
1617                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1618                 PF_HASHROW_LOCK(sh);
1619                 if (s->src.tcp_est)
1620                         --sn->conn;
1621                 if (--sn->states == 0)
1622                         sn->expire = time_uptime + timeout;
1623                 PF_HASHROW_UNLOCK(sh);
1624         }
1625         if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) {
1626                 sn = s->nat_src_node;
1627                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
1628                 PF_HASHROW_LOCK(sh);
1629                 if (--sn->states == 0)
1630                         sn->expire = time_uptime + timeout;
1631                 PF_HASHROW_UNLOCK(sh);
1632         }
1633         s->src_node = s->nat_src_node = NULL;
1634 }
1635
1636 /*
1637  * Unlink and potentilly free a state. Function may be
1638  * called with ID hash row locked, but always returns
1639  * unlocked, since it needs to go through key hash locking.
1640  */
1641 int
1642 pf_unlink_state(struct pf_state *s, u_int flags)
1643 {
1644         struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)];
1645
1646         if ((flags & PF_ENTER_LOCKED) == 0)
1647                 PF_HASHROW_LOCK(ih);
1648         else
1649                 PF_HASHROW_ASSERT(ih);
1650
1651         if (s->timeout == PFTM_UNLINKED) {
1652                 /*
1653                  * State is being processed
1654                  * by pf_unlink_state() in
1655                  * an other thread.
1656                  */
1657                 PF_HASHROW_UNLOCK(ih);
1658                 return (0);     /* XXXGL: undefined actually */
1659         }
1660
1661         if (s->src.state == PF_TCPS_PROXY_DST) {
1662                 /* XXX wire key the right one? */
1663                 pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af,
1664                     &s->key[PF_SK_WIRE]->addr[1],
1665                     &s->key[PF_SK_WIRE]->addr[0],
1666                     s->key[PF_SK_WIRE]->port[1],
1667                     s->key[PF_SK_WIRE]->port[0],
1668                     s->src.seqhi, s->src.seqlo + 1,
1669                     TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL);
1670         }
1671
1672         LIST_REMOVE(s, entry);
1673         pf_src_tree_remove_state(s);
1674
1675         if (pfsync_delete_state_ptr != NULL)
1676                 pfsync_delete_state_ptr(s);
1677
1678         STATE_DEC_COUNTERS(s);
1679
1680         s->timeout = PFTM_UNLINKED;
1681
1682         PF_HASHROW_UNLOCK(ih);
1683
1684         pf_detach_state(s);
1685         /* pf_state_insert() initialises refs to 2, so we can never release the
1686          * last reference here, only in pf_release_state(). */
1687         (void)refcount_release(&s->refs);
1688
1689         return (pf_release_state(s));
1690 }
1691
1692 void
1693 pf_free_state(struct pf_state *cur)
1694 {
1695
1696         KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur));
1697         KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__,
1698             cur->timeout));
1699
1700         pf_normalize_tcp_cleanup(cur);
1701         uma_zfree(V_pf_state_z, cur);
1702         counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1);
1703 }
1704
1705 /*
1706  * Called only from pf_purge_thread(), thus serialized.
1707  */
1708 static u_int
1709 pf_purge_expired_states(u_int i, int maxcheck)
1710 {
1711         struct pf_idhash *ih;
1712         struct pf_state *s;
1713
1714         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1715
1716         /*
1717          * Go through hash and unlink states that expire now.
1718          */
1719         while (maxcheck > 0) {
1720
1721                 ih = &V_pf_idhash[i];
1722 relock:
1723                 PF_HASHROW_LOCK(ih);
1724                 LIST_FOREACH(s, &ih->states, entry) {
1725                         if (pf_state_expires(s) <= time_uptime) {
1726                                 V_pf_status.states -=
1727                                     pf_unlink_state(s, PF_ENTER_LOCKED);
1728                                 goto relock;
1729                         }
1730                         s->rule.ptr->rule_flag |= PFRULE_REFS;
1731                         if (s->nat_rule.ptr != NULL)
1732                                 s->nat_rule.ptr->rule_flag |= PFRULE_REFS;
1733                         if (s->anchor.ptr != NULL)
1734                                 s->anchor.ptr->rule_flag |= PFRULE_REFS;
1735                         s->kif->pfik_flags |= PFI_IFLAG_REFS;
1736                         if (s->rt_kif)
1737                                 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS;
1738                 }
1739                 PF_HASHROW_UNLOCK(ih);
1740
1741                 /* Return when we hit end of hash. */
1742                 if (++i > pf_hashmask) {
1743                         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1744                         return (0);
1745                 }
1746
1747                 maxcheck--;
1748         }
1749
1750         V_pf_status.states = uma_zone_get_cur(V_pf_state_z);
1751
1752         return (i);
1753 }
1754
1755 static void
1756 pf_purge_unlinked_rules()
1757 {
1758         struct pf_rulequeue tmpq;
1759         struct pf_rule *r, *r1;
1760
1761         /*
1762          * If we have overloading task pending, then we'd
1763          * better skip purging this time. There is a tiny
1764          * probability that overloading task references
1765          * an already unlinked rule.
1766          */
1767         PF_OVERLOADQ_LOCK();
1768         if (!SLIST_EMPTY(&V_pf_overloadqueue)) {
1769                 PF_OVERLOADQ_UNLOCK();
1770                 return;
1771         }
1772         PF_OVERLOADQ_UNLOCK();
1773
1774         /*
1775          * Do naive mark-and-sweep garbage collecting of old rules.
1776          * Reference flag is raised by pf_purge_expired_states()
1777          * and pf_purge_expired_src_nodes().
1778          *
1779          * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK,
1780          * use a temporary queue.
1781          */
1782         TAILQ_INIT(&tmpq);
1783         PF_UNLNKDRULES_LOCK();
1784         TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) {
1785                 if (!(r->rule_flag & PFRULE_REFS)) {
1786                         TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries);
1787                         TAILQ_INSERT_TAIL(&tmpq, r, entries);
1788                 } else
1789                         r->rule_flag &= ~PFRULE_REFS;
1790         }
1791         PF_UNLNKDRULES_UNLOCK();
1792
1793         if (!TAILQ_EMPTY(&tmpq)) {
1794                 PF_RULES_WLOCK();
1795                 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) {
1796                         TAILQ_REMOVE(&tmpq, r, entries);
1797                         pf_free_rule(r);
1798                 }
1799                 PF_RULES_WUNLOCK();
1800         }
1801 }
1802
1803 void
1804 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af)
1805 {
1806         switch (af) {
1807 #ifdef INET
1808         case AF_INET: {
1809                 u_int32_t a = ntohl(addr->addr32[0]);
1810                 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255,
1811                     (a>>8)&255, a&255);
1812                 if (p) {
1813                         p = ntohs(p);
1814                         printf(":%u", p);
1815                 }
1816                 break;
1817         }
1818 #endif /* INET */
1819 #ifdef INET6
1820         case AF_INET6: {
1821                 u_int16_t b;
1822                 u_int8_t i, curstart, curend, maxstart, maxend;
1823                 curstart = curend = maxstart = maxend = 255;
1824                 for (i = 0; i < 8; i++) {
1825                         if (!addr->addr16[i]) {
1826                                 if (curstart == 255)
1827                                         curstart = i;
1828                                 curend = i;
1829                         } else {
1830                                 if ((curend - curstart) >
1831                                     (maxend - maxstart)) {
1832                                         maxstart = curstart;
1833                                         maxend = curend;
1834                                 }
1835                                 curstart = curend = 255;
1836                         }
1837                 }
1838                 if ((curend - curstart) >
1839                     (maxend - maxstart)) {
1840                         maxstart = curstart;
1841                         maxend = curend;
1842                 }
1843                 for (i = 0; i < 8; i++) {
1844                         if (i >= maxstart && i <= maxend) {
1845                                 if (i == 0)
1846                                         printf(":");
1847                                 if (i == maxend)
1848                                         printf(":");
1849                         } else {
1850                                 b = ntohs(addr->addr16[i]);
1851                                 printf("%x", b);
1852                                 if (i < 7)
1853                                         printf(":");
1854                         }
1855                 }
1856                 if (p) {
1857                         p = ntohs(p);
1858                         printf("[%u]", p);
1859                 }
1860                 break;
1861         }
1862 #endif /* INET6 */
1863         }
1864 }
1865
1866 void
1867 pf_print_state(struct pf_state *s)
1868 {
1869         pf_print_state_parts(s, NULL, NULL);
1870 }
1871
1872 static void
1873 pf_print_state_parts(struct pf_state *s,
1874     struct pf_state_key *skwp, struct pf_state_key *sksp)
1875 {
1876         struct pf_state_key *skw, *sks;
1877         u_int8_t proto, dir;
1878
1879         /* Do our best to fill these, but they're skipped if NULL */
1880         skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL);
1881         sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL);
1882         proto = skw ? skw->proto : (sks ? sks->proto : 0);
1883         dir = s ? s->direction : 0;
1884
1885         switch (proto) {
1886         case IPPROTO_IPV4:
1887                 printf("IPv4");
1888                 break;
1889         case IPPROTO_IPV6:
1890                 printf("IPv6");
1891                 break;
1892         case IPPROTO_TCP:
1893                 printf("TCP");
1894                 break;
1895         case IPPROTO_UDP:
1896                 printf("UDP");
1897                 break;
1898         case IPPROTO_ICMP:
1899                 printf("ICMP");
1900                 break;
1901         case IPPROTO_ICMPV6:
1902                 printf("ICMPv6");
1903                 break;
1904         default:
1905                 printf("%u", proto);
1906                 break;
1907         }
1908         switch (dir) {
1909         case PF_IN:
1910                 printf(" in");
1911                 break;
1912         case PF_OUT:
1913                 printf(" out");
1914                 break;
1915         }
1916         if (skw) {
1917                 printf(" wire: ");
1918                 pf_print_host(&skw->addr[0], skw->port[0], skw->af);
1919                 printf(" ");
1920                 pf_print_host(&skw->addr[1], skw->port[1], skw->af);
1921         }
1922         if (sks) {
1923                 printf(" stack: ");
1924                 if (sks != skw) {
1925                         pf_print_host(&sks->addr[0], sks->port[0], sks->af);
1926                         printf(" ");
1927                         pf_print_host(&sks->addr[1], sks->port[1], sks->af);
1928                 } else
1929                         printf("-");
1930         }
1931         if (s) {
1932                 if (proto == IPPROTO_TCP) {
1933                         printf(" [lo=%u high=%u win=%u modulator=%u",
1934                             s->src.seqlo, s->src.seqhi,
1935                             s->src.max_win, s->src.seqdiff);
1936                         if (s->src.wscale && s->dst.wscale)
1937                                 printf(" wscale=%u",
1938                                     s->src.wscale & PF_WSCALE_MASK);
1939                         printf("]");
1940                         printf(" [lo=%u high=%u win=%u modulator=%u",
1941                             s->dst.seqlo, s->dst.seqhi,
1942                             s->dst.max_win, s->dst.seqdiff);
1943                         if (s->src.wscale && s->dst.wscale)
1944                                 printf(" wscale=%u",
1945                                 s->dst.wscale & PF_WSCALE_MASK);
1946                         printf("]");
1947                 }
1948                 printf(" %u:%u", s->src.state, s->dst.state);
1949         }
1950 }
1951
1952 void
1953 pf_print_flags(u_int8_t f)
1954 {
1955         if (f)
1956                 printf(" ");
1957         if (f & TH_FIN)
1958                 printf("F");
1959         if (f & TH_SYN)
1960                 printf("S");
1961         if (f & TH_RST)
1962                 printf("R");
1963         if (f & TH_PUSH)
1964                 printf("P");
1965         if (f & TH_ACK)
1966                 printf("A");
1967         if (f & TH_URG)
1968                 printf("U");
1969         if (f & TH_ECE)
1970                 printf("E");
1971         if (f & TH_CWR)
1972                 printf("W");
1973 }
1974
1975 #define PF_SET_SKIP_STEPS(i)                                    \
1976         do {                                                    \
1977                 while (head[i] != cur) {                        \
1978                         head[i]->skip[i].ptr = cur;             \
1979                         head[i] = TAILQ_NEXT(head[i], entries); \
1980                 }                                               \
1981         } while (0)
1982
1983 void
1984 pf_calc_skip_steps(struct pf_rulequeue *rules)
1985 {
1986         struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT];
1987         int i;
1988
1989         cur = TAILQ_FIRST(rules);
1990         prev = cur;
1991         for (i = 0; i < PF_SKIP_COUNT; ++i)
1992                 head[i] = cur;
1993         while (cur != NULL) {
1994
1995                 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot)
1996                         PF_SET_SKIP_STEPS(PF_SKIP_IFP);
1997                 if (cur->direction != prev->direction)
1998                         PF_SET_SKIP_STEPS(PF_SKIP_DIR);
1999                 if (cur->af != prev->af)
2000                         PF_SET_SKIP_STEPS(PF_SKIP_AF);
2001                 if (cur->proto != prev->proto)
2002                         PF_SET_SKIP_STEPS(PF_SKIP_PROTO);
2003                 if (cur->src.neg != prev->src.neg ||
2004                     pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr))
2005                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR);
2006                 if (cur->src.port[0] != prev->src.port[0] ||
2007                     cur->src.port[1] != prev->src.port[1] ||
2008                     cur->src.port_op != prev->src.port_op)
2009                         PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT);
2010                 if (cur->dst.neg != prev->dst.neg ||
2011                     pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr))
2012                         PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR);
2013                 if (cur->dst.port[0] != prev->dst.port[0] ||
2014                     cur->dst.port[1] != prev->dst.port[1] ||
2015                     cur->dst.port_op != prev->dst.port_op)
2016                         PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT);
2017
2018                 prev = cur;
2019                 cur = TAILQ_NEXT(cur, entries);
2020         }
2021         for (i = 0; i < PF_SKIP_COUNT; ++i)
2022                 PF_SET_SKIP_STEPS(i);
2023 }
2024
2025 static int
2026 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2)
2027 {
2028         if (aw1->type != aw2->type)
2029                 return (1);
2030         switch (aw1->type) {
2031         case PF_ADDR_ADDRMASK:
2032         case PF_ADDR_RANGE:
2033                 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6))
2034                         return (1);
2035                 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6))
2036                         return (1);
2037                 return (0);
2038         case PF_ADDR_DYNIFTL:
2039                 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt);
2040         case PF_ADDR_NOROUTE:
2041         case PF_ADDR_URPFFAILED:
2042                 return (0);
2043         case PF_ADDR_TABLE:
2044                 return (aw1->p.tbl != aw2->p.tbl);
2045         default:
2046                 printf("invalid address type: %d\n", aw1->type);
2047                 return (1);
2048         }
2049 }
2050
2051 /**
2052  * Checksum updates are a little complicated because the checksum in the TCP/UDP
2053  * header isn't always a full checksum. In some cases (i.e. output) it's a
2054  * pseudo-header checksum, which is a partial checksum over src/dst IP
2055  * addresses, protocol number and length.
2056  *
2057  * That means we have the following cases:
2058  *  * Input or forwarding: we don't have TSO, the checksum fields are full
2059  *      checksums, we need to update the checksum whenever we change anything.
2060  *  * Output (i.e. the checksum is a pseudo-header checksum):
2061  *      x The field being updated is src/dst address or affects the length of
2062  *      the packet. We need to update the pseudo-header checksum (note that this
2063  *      checksum is not ones' complement).
2064  *      x Some other field is being modified (e.g. src/dst port numbers): We
2065  *      don't have to update anything.
2066  **/
2067 u_int16_t
2068 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp)
2069 {
2070         u_int32_t       l;
2071
2072         if (udp && !cksum)
2073                 return (0x0000);
2074         l = cksum + old - new;
2075         l = (l >> 16) + (l & 65535);
2076         l = l & 65535;
2077         if (udp && !l)
2078                 return (0xFFFF);
2079         return (l);
2080 }
2081
2082 u_int16_t
2083 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old,
2084         u_int16_t new, u_int8_t udp)
2085 {
2086         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2087                 return (cksum);
2088
2089         return (pf_cksum_fixup(cksum, old, new, udp));
2090 }
2091
2092 static void
2093 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic,
2094         u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u,
2095         sa_family_t af)
2096 {
2097         struct pf_addr  ao;
2098         u_int16_t       po = *p;
2099
2100         PF_ACPY(&ao, a, af);
2101         PF_ACPY(a, an, af);
2102
2103         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6))
2104                 *pc = ~*pc;
2105
2106         *p = pn;
2107
2108         switch (af) {
2109 #ifdef INET
2110         case AF_INET:
2111                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2112                     ao.addr16[0], an->addr16[0], 0),
2113                     ao.addr16[1], an->addr16[1], 0);
2114                 *p = pn;
2115
2116                 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc,
2117                     ao.addr16[0], an->addr16[0], u),
2118                     ao.addr16[1], an->addr16[1], u);
2119
2120                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2121                 break;
2122 #endif /* INET */
2123 #ifdef INET6
2124         case AF_INET6:
2125                 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2126                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2127                     pf_cksum_fixup(pf_cksum_fixup(*pc,
2128                     ao.addr16[0], an->addr16[0], u),
2129                     ao.addr16[1], an->addr16[1], u),
2130                     ao.addr16[2], an->addr16[2], u),
2131                     ao.addr16[3], an->addr16[3], u),
2132                     ao.addr16[4], an->addr16[4], u),
2133                     ao.addr16[5], an->addr16[5], u),
2134                     ao.addr16[6], an->addr16[6], u),
2135                     ao.addr16[7], an->addr16[7], u);
2136
2137                 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u);
2138                 break;
2139 #endif /* INET6 */
2140         }
2141
2142         if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA |
2143             CSUM_DELAY_DATA_IPV6)) {
2144                 *pc = ~*pc;
2145                 if (! *pc)
2146                         *pc = 0xffff;
2147         }
2148 }
2149
2150 /* Changes a u_int32_t.  Uses a void * so there are no align restrictions */
2151 void
2152 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u)
2153 {
2154         u_int32_t       ao;
2155
2156         memcpy(&ao, a, sizeof(ao));
2157         memcpy(a, &an, sizeof(u_int32_t));
2158         *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u),
2159             ao % 65536, an % 65536, u);
2160 }
2161
2162 void
2163 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp)
2164 {
2165         u_int32_t       ao;
2166
2167         memcpy(&ao, a, sizeof(ao));
2168         memcpy(a, &an, sizeof(u_int32_t));
2169
2170         *c = pf_proto_cksum_fixup(m,
2171             pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp),
2172             ao % 65536, an % 65536, udp);
2173 }
2174
2175 #ifdef INET6
2176 static void
2177 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u)
2178 {
2179         struct pf_addr  ao;
2180
2181         PF_ACPY(&ao, a, AF_INET6);
2182         PF_ACPY(a, an, AF_INET6);
2183
2184         *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2185             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2186             pf_cksum_fixup(pf_cksum_fixup(*c,
2187             ao.addr16[0], an->addr16[0], u),
2188             ao.addr16[1], an->addr16[1], u),
2189             ao.addr16[2], an->addr16[2], u),
2190             ao.addr16[3], an->addr16[3], u),
2191             ao.addr16[4], an->addr16[4], u),
2192             ao.addr16[5], an->addr16[5], u),
2193             ao.addr16[6], an->addr16[6], u),
2194             ao.addr16[7], an->addr16[7], u);
2195 }
2196 #endif /* INET6 */
2197
2198 static void
2199 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa,
2200     struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c,
2201     u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af)
2202 {
2203         struct pf_addr  oia, ooa;
2204
2205         PF_ACPY(&oia, ia, af);
2206         if (oa)
2207                 PF_ACPY(&ooa, oa, af);
2208
2209         /* Change inner protocol port, fix inner protocol checksum. */
2210         if (ip != NULL) {
2211                 u_int16_t       oip = *ip;
2212                 u_int32_t       opc;
2213
2214                 if (pc != NULL)
2215                         opc = *pc;
2216                 *ip = np;
2217                 if (pc != NULL)
2218                         *pc = pf_cksum_fixup(*pc, oip, *ip, u);
2219                 *ic = pf_cksum_fixup(*ic, oip, *ip, 0);
2220                 if (pc != NULL)
2221                         *ic = pf_cksum_fixup(*ic, opc, *pc, 0);
2222         }
2223         /* Change inner ip address, fix inner ip and icmp checksums. */
2224         PF_ACPY(ia, na, af);
2225         switch (af) {
2226 #ifdef INET
2227         case AF_INET: {
2228                 u_int32_t        oh2c = *h2c;
2229
2230                 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c,
2231                     oia.addr16[0], ia->addr16[0], 0),
2232                     oia.addr16[1], ia->addr16[1], 0);
2233                 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic,
2234                     oia.addr16[0], ia->addr16[0], 0),
2235                     oia.addr16[1], ia->addr16[1], 0);
2236                 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0);
2237                 break;
2238         }
2239 #endif /* INET */
2240 #ifdef INET6
2241         case AF_INET6:
2242                 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2243                     pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2244                     pf_cksum_fixup(pf_cksum_fixup(*ic,
2245                     oia.addr16[0], ia->addr16[0], u),
2246                     oia.addr16[1], ia->addr16[1], u),
2247                     oia.addr16[2], ia->addr16[2], u),
2248                     oia.addr16[3], ia->addr16[3], u),
2249                     oia.addr16[4], ia->addr16[4], u),
2250                     oia.addr16[5], ia->addr16[5], u),
2251                     oia.addr16[6], ia->addr16[6], u),
2252                     oia.addr16[7], ia->addr16[7], u);
2253                 break;
2254 #endif /* INET6 */
2255         }
2256         /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */
2257         if (oa) {
2258                 PF_ACPY(oa, na, af);
2259                 switch (af) {
2260 #ifdef INET
2261                 case AF_INET:
2262                         *hc = pf_cksum_fixup(pf_cksum_fixup(*hc,
2263                             ooa.addr16[0], oa->addr16[0], 0),
2264                             ooa.addr16[1], oa->addr16[1], 0);
2265                         break;
2266 #endif /* INET */
2267 #ifdef INET6
2268                 case AF_INET6:
2269                         *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2270                             pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup(
2271                             pf_cksum_fixup(pf_cksum_fixup(*ic,
2272                             ooa.addr16[0], oa->addr16[0], u),
2273                             ooa.addr16[1], oa->addr16[1], u),
2274                             ooa.addr16[2], oa->addr16[2], u),
2275                             ooa.addr16[3], oa->addr16[3], u),
2276                             ooa.addr16[4], oa->addr16[4], u),
2277                             ooa.addr16[5], oa->addr16[5], u),
2278                             ooa.addr16[6], oa->addr16[6], u),
2279                             ooa.addr16[7], oa->addr16[7], u);
2280                         break;
2281 #endif /* INET6 */
2282                 }
2283         }
2284 }
2285
2286
2287 /*
2288  * Need to modulate the sequence numbers in the TCP SACK option
2289  * (credits to Krzysztof Pfaff for report and patch)
2290  */
2291 static int
2292 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd,
2293     struct tcphdr *th, struct pf_state_peer *dst)
2294 {
2295         int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen;
2296         u_int8_t opts[TCP_MAXOLEN], *opt = opts;
2297         int copyback = 0, i, olen;
2298         struct sackblk sack;
2299
2300 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2)
2301         if (hlen < TCPOLEN_SACKLEN ||
2302             !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af))
2303                 return 0;
2304
2305         while (hlen >= TCPOLEN_SACKLEN) {
2306                 olen = opt[1];
2307                 switch (*opt) {
2308                 case TCPOPT_EOL:        /* FALLTHROUGH */
2309                 case TCPOPT_NOP:
2310                         opt++;
2311                         hlen--;
2312                         break;
2313                 case TCPOPT_SACK:
2314                         if (olen > hlen)
2315                                 olen = hlen;
2316                         if (olen >= TCPOLEN_SACKLEN) {
2317                                 for (i = 2; i + TCPOLEN_SACK <= olen;
2318                                     i += TCPOLEN_SACK) {
2319                                         memcpy(&sack, &opt[i], sizeof(sack));
2320                                         pf_change_proto_a(m, &sack.start, &th->th_sum,
2321                                             htonl(ntohl(sack.start) - dst->seqdiff), 0);
2322                                         pf_change_proto_a(m, &sack.end, &th->th_sum,
2323                                             htonl(ntohl(sack.end) - dst->seqdiff), 0);
2324                                         memcpy(&opt[i], &sack, sizeof(sack));
2325                                 }
2326                                 copyback = 1;
2327                         }
2328                         /* FALLTHROUGH */
2329                 default:
2330                         if (olen < 2)
2331                                 olen = 2;
2332                         hlen -= olen;
2333                         opt += olen;
2334                 }
2335         }
2336
2337         if (copyback)
2338                 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts);
2339         return (copyback);
2340 }
2341
2342 static void
2343 pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af,
2344     const struct pf_addr *saddr, const struct pf_addr *daddr,
2345     u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack,
2346     u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag,
2347     u_int16_t rtag, struct ifnet *ifp)
2348 {
2349         struct pf_send_entry *pfse;
2350         struct mbuf     *m;
2351         int              len, tlen;
2352 #ifdef INET
2353         struct ip       *h = NULL;
2354 #endif /* INET */
2355 #ifdef INET6
2356         struct ip6_hdr  *h6 = NULL;
2357 #endif /* INET6 */
2358         struct tcphdr   *th;
2359         char            *opt;
2360         struct pf_mtag  *pf_mtag;
2361
2362         len = 0;
2363         th = NULL;
2364
2365         /* maximum segment size tcp option */
2366         tlen = sizeof(struct tcphdr);
2367         if (mss)
2368                 tlen += 4;
2369
2370         switch (af) {
2371 #ifdef INET
2372         case AF_INET:
2373                 len = sizeof(struct ip) + tlen;
2374                 break;
2375 #endif /* INET */
2376 #ifdef INET6
2377         case AF_INET6:
2378                 len = sizeof(struct ip6_hdr) + tlen;
2379                 break;
2380 #endif /* INET6 */
2381         default:
2382                 panic("%s: unsupported af %d", __func__, af);
2383         }
2384
2385         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2386         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2387         if (pfse == NULL)
2388                 return;
2389         m = m_gethdr(M_NOWAIT, MT_DATA);
2390         if (m == NULL) {
2391                 free(pfse, M_PFTEMP);
2392                 return;
2393         }
2394 #ifdef MAC
2395         mac_netinet_firewall_send(m);
2396 #endif
2397         if ((pf_mtag = pf_get_mtag(m)) == NULL) {
2398                 free(pfse, M_PFTEMP);
2399                 m_freem(m);
2400                 return;
2401         }
2402         if (tag)
2403                 m->m_flags |= M_SKIP_FIREWALL;
2404         pf_mtag->tag = rtag;
2405
2406         if (r != NULL && r->rtableid >= 0)
2407                 M_SETFIB(m, r->rtableid);
2408
2409 #ifdef ALTQ
2410         if (r != NULL && r->qid) {
2411                 pf_mtag->qid = r->qid;
2412
2413                 /* add hints for ecn */
2414                 pf_mtag->hdr = mtod(m, struct ip *);
2415         }
2416 #endif /* ALTQ */
2417         m->m_data += max_linkhdr;
2418         m->m_pkthdr.len = m->m_len = len;
2419         m->m_pkthdr.rcvif = NULL;
2420         bzero(m->m_data, len);
2421         switch (af) {
2422 #ifdef INET
2423         case AF_INET:
2424                 h = mtod(m, struct ip *);
2425
2426                 /* IP header fields included in the TCP checksum */
2427                 h->ip_p = IPPROTO_TCP;
2428                 h->ip_len = htons(tlen);
2429                 h->ip_src.s_addr = saddr->v4.s_addr;
2430                 h->ip_dst.s_addr = daddr->v4.s_addr;
2431
2432                 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip));
2433                 break;
2434 #endif /* INET */
2435 #ifdef INET6
2436         case AF_INET6:
2437                 h6 = mtod(m, struct ip6_hdr *);
2438
2439                 /* IP header fields included in the TCP checksum */
2440                 h6->ip6_nxt = IPPROTO_TCP;
2441                 h6->ip6_plen = htons(tlen);
2442                 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr));
2443                 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr));
2444
2445                 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr));
2446                 break;
2447 #endif /* INET6 */
2448         }
2449
2450         /* TCP header */
2451         th->th_sport = sport;
2452         th->th_dport = dport;
2453         th->th_seq = htonl(seq);
2454         th->th_ack = htonl(ack);
2455         th->th_off = tlen >> 2;
2456         th->th_flags = flags;
2457         th->th_win = htons(win);
2458
2459         if (mss) {
2460                 opt = (char *)(th + 1);
2461                 opt[0] = TCPOPT_MAXSEG;
2462                 opt[1] = 4;
2463                 HTONS(mss);
2464                 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2);
2465         }
2466
2467         switch (af) {
2468 #ifdef INET
2469         case AF_INET:
2470                 /* TCP checksum */
2471                 th->th_sum = in_cksum(m, len);
2472
2473                 /* Finish the IP header */
2474                 h->ip_v = 4;
2475                 h->ip_hl = sizeof(*h) >> 2;
2476                 h->ip_tos = IPTOS_LOWDELAY;
2477                 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0);
2478                 h->ip_len = htons(len);
2479                 h->ip_ttl = ttl ? ttl : V_ip_defttl;
2480                 h->ip_sum = 0;
2481
2482                 pfse->pfse_type = PFSE_IP;
2483                 break;
2484 #endif /* INET */
2485 #ifdef INET6
2486         case AF_INET6:
2487                 /* TCP checksum */
2488                 th->th_sum = in6_cksum(m, IPPROTO_TCP,
2489                     sizeof(struct ip6_hdr), tlen);
2490
2491                 h6->ip6_vfc |= IPV6_VERSION;
2492                 h6->ip6_hlim = IPV6_DEFHLIM;
2493
2494                 pfse->pfse_type = PFSE_IP6;
2495                 break;
2496 #endif /* INET6 */
2497         }
2498         pfse->pfse_m = m;
2499         pf_send(pfse);
2500 }
2501
2502 static void
2503 pf_return(struct pf_rule *r, struct pf_rule *nr, struct pf_pdesc *pd,
2504     struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th,
2505     struct pfi_kif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen,
2506     u_short *reason)
2507 {
2508         struct pf_addr  * const saddr = pd->src;
2509         struct pf_addr  * const daddr = pd->dst;
2510         sa_family_t      af = pd->af;
2511
2512         /* undo NAT changes, if they have taken place */
2513         if (nr != NULL) {
2514                 PF_ACPY(saddr, &sk->addr[pd->sidx], af);
2515                 PF_ACPY(daddr, &sk->addr[pd->didx], af);
2516                 if (pd->sport)
2517                         *pd->sport = sk->port[pd->sidx];
2518                 if (pd->dport)
2519                         *pd->dport = sk->port[pd->didx];
2520                 if (pd->proto_sum)
2521                         *pd->proto_sum = bproto_sum;
2522                 if (pd->ip_sum)
2523                         *pd->ip_sum = bip_sum;
2524                 m_copyback(m, off, hdrlen, pd->hdr.any);
2525         }
2526         if (pd->proto == IPPROTO_TCP &&
2527             ((r->rule_flag & PFRULE_RETURNRST) ||
2528             (r->rule_flag & PFRULE_RETURN)) &&
2529             !(th->th_flags & TH_RST)) {
2530                 u_int32_t        ack = ntohl(th->th_seq) + pd->p_len;
2531                 int              len = 0;
2532 #ifdef INET
2533                 struct ip       *h4;
2534 #endif
2535 #ifdef INET6
2536                 struct ip6_hdr  *h6;
2537 #endif
2538
2539                 switch (af) {
2540 #ifdef INET
2541                 case AF_INET:
2542                         h4 = mtod(m, struct ip *);
2543                         len = ntohs(h4->ip_len) - off;
2544                         break;
2545 #endif
2546 #ifdef INET6
2547                 case AF_INET6:
2548                         h6 = mtod(m, struct ip6_hdr *);
2549                         len = ntohs(h6->ip6_plen) - (off - sizeof(*h6));
2550                         break;
2551 #endif
2552                 }
2553
2554                 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af))
2555                         REASON_SET(reason, PFRES_PROTCKSUM);
2556                 else {
2557                         if (th->th_flags & TH_SYN)
2558                                 ack++;
2559                         if (th->th_flags & TH_FIN)
2560                                 ack++;
2561                         pf_send_tcp(m, r, af, pd->dst,
2562                                 pd->src, th->th_dport, th->th_sport,
2563                                 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0,
2564                                 r->return_ttl, 1, 0, kif->pfik_ifp);
2565                 }
2566         } else if (pd->proto != IPPROTO_ICMP && af == AF_INET &&
2567                 r->return_icmp)
2568                 pf_send_icmp(m, r->return_icmp >> 8,
2569                         r->return_icmp & 255, af, r);
2570         else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 &&
2571                 r->return_icmp6)
2572                 pf_send_icmp(m, r->return_icmp6 >> 8,
2573                         r->return_icmp6 & 255, af, r);
2574 }
2575
2576
2577 static int
2578 pf_ieee8021q_setpcp(struct mbuf *m, u_int8_t prio)
2579 {
2580         struct m_tag *mtag;
2581
2582         KASSERT(prio <= PF_PRIO_MAX,
2583             ("%s with invalid pcp", __func__));
2584
2585         mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_OUT, NULL);
2586         if (mtag == NULL) {
2587                 mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_OUT,
2588                     sizeof(uint8_t), M_NOWAIT);
2589                 if (mtag == NULL)
2590                         return (ENOMEM);
2591                 m_tag_prepend(m, mtag);
2592         }
2593
2594         *(uint8_t *)(mtag + 1) = prio;
2595         return (0);
2596 }
2597
2598 static int
2599 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m)
2600 {
2601         struct m_tag *mtag;
2602         u_int8_t mpcp;
2603
2604         mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL);
2605         if (mtag == NULL)
2606                 return (0);
2607
2608         if (prio == PF_PRIO_ZERO)
2609                 prio = 0;
2610
2611         mpcp = *(uint8_t *)(mtag + 1);
2612
2613         return (mpcp == prio);
2614 }
2615
2616 static void
2617 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af,
2618     struct pf_rule *r)
2619 {
2620         struct pf_send_entry *pfse;
2621         struct mbuf *m0;
2622         struct pf_mtag *pf_mtag;
2623
2624         /* Allocate outgoing queue entry, mbuf and mbuf tag. */
2625         pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT);
2626         if (pfse == NULL)
2627                 return;
2628
2629         if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) {
2630                 free(pfse, M_PFTEMP);
2631                 return;
2632         }
2633
2634         if ((pf_mtag = pf_get_mtag(m0)) == NULL) {
2635                 free(pfse, M_PFTEMP);
2636                 return;
2637         }
2638         /* XXX: revisit */
2639         m0->m_flags |= M_SKIP_FIREWALL;
2640
2641         if (r->rtableid >= 0)
2642                 M_SETFIB(m0, r->rtableid);
2643
2644 #ifdef ALTQ
2645         if (r->qid) {
2646                 pf_mtag->qid = r->qid;
2647                 /* add hints for ecn */
2648                 pf_mtag->hdr = mtod(m0, struct ip *);
2649         }
2650 #endif /* ALTQ */
2651
2652         switch (af) {
2653 #ifdef INET
2654         case AF_INET:
2655                 pfse->pfse_type = PFSE_ICMP;
2656                 break;
2657 #endif /* INET */
2658 #ifdef INET6
2659         case AF_INET6:
2660                 pfse->pfse_type = PFSE_ICMP6;
2661                 break;
2662 #endif /* INET6 */
2663         }
2664         pfse->pfse_m = m0;
2665         pfse->icmpopts.type = type;
2666         pfse->icmpopts.code = code;
2667         pf_send(pfse);
2668 }
2669
2670 /*
2671  * Return 1 if the addresses a and b match (with mask m), otherwise return 0.
2672  * If n is 0, they match if they are equal. If n is != 0, they match if they
2673  * are different.
2674  */
2675 int
2676 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m,
2677     struct pf_addr *b, sa_family_t af)
2678 {
2679         int     match = 0;
2680
2681         switch (af) {
2682 #ifdef INET
2683         case AF_INET:
2684                 if ((a->addr32[0] & m->addr32[0]) ==
2685                     (b->addr32[0] & m->addr32[0]))
2686                         match++;
2687                 break;
2688 #endif /* INET */
2689 #ifdef INET6
2690         case AF_INET6:
2691                 if (((a->addr32[0] & m->addr32[0]) ==
2692                      (b->addr32[0] & m->addr32[0])) &&
2693                     ((a->addr32[1] & m->addr32[1]) ==
2694                      (b->addr32[1] & m->addr32[1])) &&
2695                     ((a->addr32[2] & m->addr32[2]) ==
2696                      (b->addr32[2] & m->addr32[2])) &&
2697                     ((a->addr32[3] & m->addr32[3]) ==
2698                      (b->addr32[3] & m->addr32[3])))
2699                         match++;
2700                 break;
2701 #endif /* INET6 */
2702         }
2703         if (match) {
2704                 if (n)
2705                         return (0);
2706                 else
2707                         return (1);
2708         } else {
2709                 if (n)
2710                         return (1);
2711                 else
2712                         return (0);
2713         }
2714 }
2715
2716 /*
2717  * Return 1 if b <= a <= e, otherwise return 0.
2718  */
2719 int
2720 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e,
2721     struct pf_addr *a, sa_family_t af)
2722 {
2723         switch (af) {
2724 #ifdef INET
2725         case AF_INET:
2726                 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) ||
2727                     (ntohl(a->addr32[0]) > ntohl(e->addr32[0])))
2728                         return (0);
2729                 break;
2730 #endif /* INET */
2731 #ifdef INET6
2732         case AF_INET6: {
2733                 int     i;
2734
2735                 /* check a >= b */
2736                 for (i = 0; i < 4; ++i)
2737                         if (ntohl(a->addr32[i]) > ntohl(b->addr32[i]))
2738                                 break;
2739                         else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i]))
2740                                 return (0);
2741                 /* check a <= e */
2742                 for (i = 0; i < 4; ++i)
2743                         if (ntohl(a->addr32[i]) < ntohl(e->addr32[i]))
2744                                 break;
2745                         else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i]))
2746                                 return (0);
2747                 break;
2748         }
2749 #endif /* INET6 */
2750         }
2751         return (1);
2752 }
2753
2754 static int
2755 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p)
2756 {
2757         switch (op) {
2758         case PF_OP_IRG:
2759                 return ((p > a1) && (p < a2));
2760         case PF_OP_XRG:
2761                 return ((p < a1) || (p > a2));
2762         case PF_OP_RRG:
2763                 return ((p >= a1) && (p <= a2));
2764         case PF_OP_EQ:
2765                 return (p == a1);
2766         case PF_OP_NE:
2767                 return (p != a1);
2768         case PF_OP_LT:
2769                 return (p < a1);
2770         case PF_OP_LE:
2771                 return (p <= a1);
2772         case PF_OP_GT:
2773                 return (p > a1);
2774         case PF_OP_GE:
2775                 return (p >= a1);
2776         }
2777         return (0); /* never reached */
2778 }
2779
2780 int
2781 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p)
2782 {
2783         NTOHS(a1);
2784         NTOHS(a2);
2785         NTOHS(p);
2786         return (pf_match(op, a1, a2, p));
2787 }
2788
2789 static int
2790 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u)
2791 {
2792         if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2793                 return (0);
2794         return (pf_match(op, a1, a2, u));
2795 }
2796
2797 static int
2798 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g)
2799 {
2800         if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE)
2801                 return (0);
2802         return (pf_match(op, a1, a2, g));
2803 }
2804
2805 int
2806 pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag)
2807 {
2808         if (*tag == -1)
2809                 *tag = mtag;
2810
2811         return ((!r->match_tag_not && r->match_tag == *tag) ||
2812             (r->match_tag_not && r->match_tag != *tag));
2813 }
2814
2815 int
2816 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag)
2817 {
2818
2819         KASSERT(tag > 0, ("%s: tag %d", __func__, tag));
2820
2821         if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL))
2822                 return (ENOMEM);
2823
2824         pd->pf_mtag->tag = tag;
2825
2826         return (0);
2827 }
2828
2829 #define PF_ANCHOR_STACKSIZE     32
2830 struct pf_anchor_stackframe {
2831         struct pf_ruleset       *rs;
2832         struct pf_rule          *r;     /* XXX: + match bit */
2833         struct pf_anchor        *child;
2834 };
2835
2836 /*
2837  * XXX: We rely on malloc(9) returning pointer aligned addresses.
2838  */
2839 #define PF_ANCHORSTACK_MATCH    0x00000001
2840 #define PF_ANCHORSTACK_MASK     (PF_ANCHORSTACK_MATCH)
2841
2842 #define PF_ANCHOR_MATCH(f)      ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH)
2843 #define PF_ANCHOR_RULE(f)       (struct pf_rule *)                      \
2844                                 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK)
2845 #define PF_ANCHOR_SET_MATCH(f)  do { (f)->r = (void *)                  \
2846                                 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH);  \
2847 } while (0)
2848
2849 void
2850 pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth,
2851     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2852     int *match)
2853 {
2854         struct pf_anchor_stackframe     *f;
2855
2856         PF_RULES_RASSERT();
2857
2858         if (match)
2859                 *match = 0;
2860         if (*depth >= PF_ANCHOR_STACKSIZE) {
2861                 printf("%s: anchor stack overflow on %s\n",
2862                     __func__, (*r)->anchor->name);
2863                 *r = TAILQ_NEXT(*r, entries);
2864                 return;
2865         } else if (*depth == 0 && a != NULL)
2866                 *a = *r;
2867         f = stack + (*depth)++;
2868         f->rs = *rs;
2869         f->r = *r;
2870         if ((*r)->anchor_wildcard) {
2871                 struct pf_anchor_node *parent = &(*r)->anchor->children;
2872
2873                 if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) {
2874                         *r = NULL;
2875                         return;
2876                 }
2877                 *rs = &f->child->ruleset;
2878         } else {
2879                 f->child = NULL;
2880                 *rs = &(*r)->anchor->ruleset;
2881         }
2882         *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2883 }
2884
2885 int
2886 pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth,
2887     struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a,
2888     int *match)
2889 {
2890         struct pf_anchor_stackframe     *f;
2891         struct pf_rule *fr;
2892         int quick = 0;
2893
2894         PF_RULES_RASSERT();
2895
2896         do {
2897                 if (*depth <= 0)
2898                         break;
2899                 f = stack + *depth - 1;
2900                 fr = PF_ANCHOR_RULE(f);
2901                 if (f->child != NULL) {
2902                         struct pf_anchor_node *parent;
2903
2904                         /*
2905                          * This block traverses through
2906                          * a wildcard anchor.
2907                          */
2908                         parent = &fr->anchor->children;
2909                         if (match != NULL && *match) {
2910                                 /*
2911                                  * If any of "*" matched, then
2912                                  * "foo/ *" matched, mark frame
2913                                  * appropriately.
2914                                  */
2915                                 PF_ANCHOR_SET_MATCH(f);
2916                                 *match = 0;
2917                         }
2918                         f->child = RB_NEXT(pf_anchor_node, parent, f->child);
2919                         if (f->child != NULL) {
2920                                 *rs = &f->child->ruleset;
2921                                 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr);
2922                                 if (*r == NULL)
2923                                         continue;
2924                                 else
2925                                         break;
2926                         }
2927                 }
2928                 (*depth)--;
2929                 if (*depth == 0 && a != NULL)
2930                         *a = NULL;
2931                 *rs = f->rs;
2932                 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match))
2933                         quick = fr->quick;
2934                 *r = TAILQ_NEXT(fr, entries);
2935         } while (*r == NULL);
2936
2937         return (quick);
2938 }
2939
2940 #ifdef INET6
2941 void
2942 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr,
2943     struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af)
2944 {
2945         switch (af) {
2946 #ifdef INET
2947         case AF_INET:
2948                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2949                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2950                 break;
2951 #endif /* INET */
2952         case AF_INET6:
2953                 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) |
2954                 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]);
2955                 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) |
2956                 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]);
2957                 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) |
2958                 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]);
2959                 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) |
2960                 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]);
2961                 break;
2962         }
2963 }
2964
2965 void
2966 pf_addr_inc(struct pf_addr *addr, sa_family_t af)
2967 {
2968         switch (af) {
2969 #ifdef INET
2970         case AF_INET:
2971                 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1);
2972                 break;
2973 #endif /* INET */
2974         case AF_INET6:
2975                 if (addr->addr32[3] == 0xffffffff) {
2976                         addr->addr32[3] = 0;
2977                         if (addr->addr32[2] == 0xffffffff) {
2978                                 addr->addr32[2] = 0;
2979                                 if (addr->addr32[1] == 0xffffffff) {
2980                                         addr->addr32[1] = 0;
2981                                         addr->addr32[0] =
2982                                             htonl(ntohl(addr->addr32[0]) + 1);
2983                                 } else
2984                                         addr->addr32[1] =
2985                                             htonl(ntohl(addr->addr32[1]) + 1);
2986                         } else
2987                                 addr->addr32[2] =
2988                                     htonl(ntohl(addr->addr32[2]) + 1);
2989                 } else
2990                         addr->addr32[3] =
2991                             htonl(ntohl(addr->addr32[3]) + 1);
2992                 break;
2993         }
2994 }
2995 #endif /* INET6 */
2996
2997 int
2998 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m)
2999 {
3000         struct pf_addr          *saddr, *daddr;
3001         u_int16_t                sport, dport;
3002         struct inpcbinfo        *pi;
3003         struct inpcb            *inp;
3004
3005         pd->lookup.uid = UID_MAX;
3006         pd->lookup.gid = GID_MAX;
3007
3008         switch (pd->proto) {
3009         case IPPROTO_TCP:
3010                 if (pd->hdr.tcp == NULL)
3011                         return (-1);
3012                 sport = pd->hdr.tcp->th_sport;
3013                 dport = pd->hdr.tcp->th_dport;
3014                 pi = &V_tcbinfo;
3015                 break;
3016         case IPPROTO_UDP:
3017                 if (pd->hdr.udp == NULL)
3018                         return (-1);
3019                 sport = pd->hdr.udp->uh_sport;
3020                 dport = pd->hdr.udp->uh_dport;
3021                 pi = &V_udbinfo;
3022                 break;
3023         default:
3024                 return (-1);
3025         }
3026         if (direction == PF_IN) {
3027                 saddr = pd->src;
3028                 daddr = pd->dst;
3029         } else {
3030                 u_int16_t       p;
3031
3032                 p = sport;
3033                 sport = dport;
3034                 dport = p;
3035                 saddr = pd->dst;
3036                 daddr = pd->src;
3037         }
3038         switch (pd->af) {
3039 #ifdef INET
3040         case AF_INET:
3041                 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4,
3042                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
3043                 if (inp == NULL) {
3044                         inp = in_pcblookup_mbuf(pi, saddr->v4, sport,
3045                            daddr->v4, dport, INPLOOKUP_WILDCARD |
3046                            INPLOOKUP_RLOCKPCB, NULL, m);
3047                         if (inp == NULL)
3048                                 return (-1);
3049                 }
3050                 break;
3051 #endif /* INET */
3052 #ifdef INET6
3053         case AF_INET6:
3054                 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6,
3055                     dport, INPLOOKUP_RLOCKPCB, NULL, m);
3056                 if (inp == NULL) {
3057                         inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport,
3058                             &daddr->v6, dport, INPLOOKUP_WILDCARD |
3059                             INPLOOKUP_RLOCKPCB, NULL, m);
3060                         if (inp == NULL)
3061                                 return (-1);
3062                 }
3063                 break;
3064 #endif /* INET6 */
3065
3066         default:
3067                 return (-1);
3068         }
3069         INP_RLOCK_ASSERT(inp);
3070         pd->lookup.uid = inp->inp_cred->cr_uid;
3071         pd->lookup.gid = inp->inp_cred->cr_groups[0];
3072         INP_RUNLOCK(inp);
3073
3074         return (1);
3075 }
3076
3077 static u_int8_t
3078 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3079 {
3080         int              hlen;
3081         u_int8_t         hdr[60];
3082         u_int8_t        *opt, optlen;
3083         u_int8_t         wscale = 0;
3084
3085         hlen = th_off << 2;             /* hlen <= sizeof(hdr) */
3086         if (hlen <= sizeof(struct tcphdr))
3087                 return (0);
3088         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3089                 return (0);
3090         opt = hdr + sizeof(struct tcphdr);
3091         hlen -= sizeof(struct tcphdr);
3092         while (hlen >= 3) {
3093                 switch (*opt) {
3094                 case TCPOPT_EOL:
3095                 case TCPOPT_NOP:
3096                         ++opt;
3097                         --hlen;
3098                         break;
3099                 case TCPOPT_WINDOW:
3100                         wscale = opt[2];
3101                         if (wscale > TCP_MAX_WINSHIFT)
3102                                 wscale = TCP_MAX_WINSHIFT;
3103                         wscale |= PF_WSCALE_FLAG;
3104                         /* FALLTHROUGH */
3105                 default:
3106                         optlen = opt[1];
3107                         if (optlen < 2)
3108                                 optlen = 2;
3109                         hlen -= optlen;
3110                         opt += optlen;
3111                         break;
3112                 }
3113         }
3114         return (wscale);
3115 }
3116
3117 static u_int16_t
3118 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af)
3119 {
3120         int              hlen;
3121         u_int8_t         hdr[60];
3122         u_int8_t        *opt, optlen;
3123         u_int16_t        mss = V_tcp_mssdflt;
3124
3125         hlen = th_off << 2;     /* hlen <= sizeof(hdr) */
3126         if (hlen <= sizeof(struct tcphdr))
3127                 return (0);
3128         if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af))
3129                 return (0);
3130         opt = hdr + sizeof(struct tcphdr);
3131         hlen -= sizeof(struct tcphdr);
3132         while (hlen >= TCPOLEN_MAXSEG) {
3133                 switch (*opt) {
3134                 case TCPOPT_EOL:
3135                 case TCPOPT_NOP:
3136                         ++opt;
3137                         --hlen;
3138                         break;
3139                 case TCPOPT_MAXSEG:
3140                         bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2);
3141                         NTOHS(mss);
3142                         /* FALLTHROUGH */
3143                 default:
3144                         optlen = opt[1];
3145                         if (optlen < 2)
3146                                 optlen = 2;
3147                         hlen -= optlen;
3148                         opt += optlen;
3149                         break;
3150                 }
3151         }
3152         return (mss);
3153 }
3154
3155 static u_int16_t
3156 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer)
3157 {
3158 #ifdef INET
3159         struct nhop4_basic      nh4;
3160 #endif /* INET */
3161 #ifdef INET6
3162         struct nhop6_basic      nh6;
3163         struct in6_addr         dst6;
3164         uint32_t                scopeid;
3165 #endif /* INET6 */
3166         int                      hlen = 0;
3167         uint16_t                 mss = 0;
3168
3169         switch (af) {
3170 #ifdef INET
3171         case AF_INET:
3172                 hlen = sizeof(struct ip);
3173                 if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) == 0)
3174                         mss = nh4.nh_mtu - hlen - sizeof(struct tcphdr);
3175                 break;
3176 #endif /* INET */
3177 #ifdef INET6
3178         case AF_INET6:
3179                 hlen = sizeof(struct ip6_hdr);
3180                 in6_splitscope(&addr->v6, &dst6, &scopeid);
3181                 if (fib6_lookup_nh_basic(rtableid, &dst6, scopeid, 0,0,&nh6)==0)
3182                         mss = nh6.nh_mtu - hlen - sizeof(struct tcphdr);
3183                 break;
3184 #endif /* INET6 */
3185         }
3186
3187         mss = max(V_tcp_mssdflt, mss);
3188         mss = min(mss, offer);
3189         mss = max(mss, 64);             /* sanity - at least max opt space */
3190         return (mss);
3191 }
3192
3193 static u_int32_t
3194 pf_tcp_iss(struct pf_pdesc *pd)
3195 {
3196         MD5_CTX ctx;
3197         u_int32_t digest[4];
3198
3199         if (V_pf_tcp_secret_init == 0) {
3200                 read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret));
3201                 MD5Init(&V_pf_tcp_secret_ctx);
3202                 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret,
3203                     sizeof(V_pf_tcp_secret));
3204                 V_pf_tcp_secret_init = 1;
3205         }
3206
3207         ctx = V_pf_tcp_secret_ctx;
3208
3209         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short));
3210         MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short));
3211         if (pd->af == AF_INET6) {
3212                 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr));
3213                 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr));
3214         } else {
3215                 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr));
3216                 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr));
3217         }
3218         MD5Final((u_char *)digest, &ctx);
3219         V_pf_tcp_iss_off += 4096;
3220 #define ISN_RANDOM_INCREMENT (4096 - 1)
3221         return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) +
3222             V_pf_tcp_iss_off);
3223 #undef  ISN_RANDOM_INCREMENT
3224 }
3225
3226 static int
3227 pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction,
3228     struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd,
3229     struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp)
3230 {
3231         struct pf_rule          *nr = NULL;
3232         struct pf_addr          * const saddr = pd->src;
3233         struct pf_addr          * const daddr = pd->dst;
3234         sa_family_t              af = pd->af;
3235         struct pf_rule          *r, *a = NULL;
3236         struct pf_ruleset       *ruleset = NULL;
3237         struct pf_src_node      *nsn = NULL;
3238         struct tcphdr           *th = pd->hdr.tcp;
3239         struct pf_state_key     *sk = NULL, *nk = NULL;
3240         u_short                  reason;
3241         int                      rewrite = 0, hdrlen = 0;
3242         int                      tag = -1, rtableid = -1;
3243         int                      asd = 0;
3244         int                      match = 0;
3245         int                      state_icmp = 0;
3246         u_int16_t                sport = 0, dport = 0;
3247         u_int16_t                bproto_sum = 0, bip_sum = 0;
3248         u_int8_t                 icmptype = 0, icmpcode = 0;
3249         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3250
3251         PF_RULES_RASSERT();
3252
3253         if (inp != NULL) {
3254                 INP_LOCK_ASSERT(inp);
3255                 pd->lookup.uid = inp->inp_cred->cr_uid;
3256                 pd->lookup.gid = inp->inp_cred->cr_groups[0];
3257                 pd->lookup.done = 1;
3258         }
3259
3260         switch (pd->proto) {
3261         case IPPROTO_TCP:
3262                 sport = th->th_sport;
3263                 dport = th->th_dport;
3264                 hdrlen = sizeof(*th);
3265                 break;
3266         case IPPROTO_UDP:
3267                 sport = pd->hdr.udp->uh_sport;
3268                 dport = pd->hdr.udp->uh_dport;
3269                 hdrlen = sizeof(*pd->hdr.udp);
3270                 break;
3271 #ifdef INET
3272         case IPPROTO_ICMP:
3273                 if (pd->af != AF_INET)
3274                         break;
3275                 sport = dport = pd->hdr.icmp->icmp_id;
3276                 hdrlen = sizeof(*pd->hdr.icmp);
3277                 icmptype = pd->hdr.icmp->icmp_type;
3278                 icmpcode = pd->hdr.icmp->icmp_code;
3279
3280                 if (icmptype == ICMP_UNREACH ||
3281                     icmptype == ICMP_SOURCEQUENCH ||
3282                     icmptype == ICMP_REDIRECT ||
3283                     icmptype == ICMP_TIMXCEED ||
3284                     icmptype == ICMP_PARAMPROB)
3285                         state_icmp++;
3286                 break;
3287 #endif /* INET */
3288 #ifdef INET6
3289         case IPPROTO_ICMPV6:
3290                 if (af != AF_INET6)
3291                         break;
3292                 sport = dport = pd->hdr.icmp6->icmp6_id;
3293                 hdrlen = sizeof(*pd->hdr.icmp6);
3294                 icmptype = pd->hdr.icmp6->icmp6_type;
3295                 icmpcode = pd->hdr.icmp6->icmp6_code;
3296
3297                 if (icmptype == ICMP6_DST_UNREACH ||
3298                     icmptype == ICMP6_PACKET_TOO_BIG ||
3299                     icmptype == ICMP6_TIME_EXCEEDED ||
3300                     icmptype == ICMP6_PARAM_PROB)
3301                         state_icmp++;
3302                 break;
3303 #endif /* INET6 */
3304         default:
3305                 sport = dport = hdrlen = 0;
3306                 break;
3307         }
3308
3309         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3310
3311         /* check packet for BINAT/NAT/RDR */
3312         if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk,
3313             &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) {
3314                 KASSERT(sk != NULL, ("%s: null sk", __func__));
3315                 KASSERT(nk != NULL, ("%s: null nk", __func__));
3316
3317                 if (pd->ip_sum)
3318                         bip_sum = *pd->ip_sum;
3319
3320                 switch (pd->proto) {
3321                 case IPPROTO_TCP:
3322                         bproto_sum = th->th_sum;
3323                         pd->proto_sum = &th->th_sum;
3324
3325                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3326                             nk->port[pd->sidx] != sport) {
3327                                 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum,
3328                                     &th->th_sum, &nk->addr[pd->sidx],
3329                                     nk->port[pd->sidx], 0, af);
3330                                 pd->sport = &th->th_sport;
3331                                 sport = th->th_sport;
3332                         }
3333
3334                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3335                             nk->port[pd->didx] != dport) {
3336                                 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum,
3337                                     &th->th_sum, &nk->addr[pd->didx],
3338                                     nk->port[pd->didx], 0, af);
3339                                 dport = th->th_dport;
3340                                 pd->dport = &th->th_dport;
3341                         }
3342                         rewrite++;
3343                         break;
3344                 case IPPROTO_UDP:
3345                         bproto_sum = pd->hdr.udp->uh_sum;
3346                         pd->proto_sum = &pd->hdr.udp->uh_sum;
3347
3348                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) ||
3349                             nk->port[pd->sidx] != sport) {
3350                                 pf_change_ap(m, saddr, &pd->hdr.udp->uh_sport,
3351                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3352                                     &nk->addr[pd->sidx],
3353                                     nk->port[pd->sidx], 1, af);
3354                                 sport = pd->hdr.udp->uh_sport;
3355                                 pd->sport = &pd->hdr.udp->uh_sport;
3356                         }
3357
3358                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) ||
3359                             nk->port[pd->didx] != dport) {
3360                                 pf_change_ap(m, daddr, &pd->hdr.udp->uh_dport,
3361                                     pd->ip_sum, &pd->hdr.udp->uh_sum,
3362                                     &nk->addr[pd->didx],
3363                                     nk->port[pd->didx], 1, af);
3364                                 dport = pd->hdr.udp->uh_dport;
3365                                 pd->dport = &pd->hdr.udp->uh_dport;
3366                         }
3367                         rewrite++;
3368                         break;
3369 #ifdef INET
3370                 case IPPROTO_ICMP:
3371                         nk->port[0] = nk->port[1];
3372                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET))
3373                                 pf_change_a(&saddr->v4.s_addr, pd->ip_sum,
3374                                     nk->addr[pd->sidx].v4.s_addr, 0);
3375
3376                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET))
3377                                 pf_change_a(&daddr->v4.s_addr, pd->ip_sum,
3378                                     nk->addr[pd->didx].v4.s_addr, 0);
3379
3380                         if (nk->port[1] != pd->hdr.icmp->icmp_id) {
3381                                 pd->hdr.icmp->icmp_cksum = pf_cksum_fixup(
3382                                     pd->hdr.icmp->icmp_cksum, sport,
3383                                     nk->port[1], 0);
3384                                 pd->hdr.icmp->icmp_id = nk->port[1];
3385                                 pd->sport = &pd->hdr.icmp->icmp_id;
3386                         }
3387                         m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
3388                         break;
3389 #endif /* INET */
3390 #ifdef INET6
3391                 case IPPROTO_ICMPV6:
3392                         nk->port[0] = nk->port[1];
3393                         if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6))
3394                                 pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum,
3395                                     &nk->addr[pd->sidx], 0);
3396
3397                         if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6))
3398                                 pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum,
3399                                     &nk->addr[pd->didx], 0);
3400                         rewrite++;
3401                         break;
3402 #endif /* INET */
3403                 default:
3404                         switch (af) {
3405 #ifdef INET
3406                         case AF_INET:
3407                                 if (PF_ANEQ(saddr,
3408                                     &nk->addr[pd->sidx], AF_INET))
3409                                         pf_change_a(&saddr->v4.s_addr,
3410                                             pd->ip_sum,
3411                                             nk->addr[pd->sidx].v4.s_addr, 0);
3412
3413                                 if (PF_ANEQ(daddr,
3414                                     &nk->addr[pd->didx], AF_INET))
3415                                         pf_change_a(&daddr->v4.s_addr,
3416                                             pd->ip_sum,
3417                                             nk->addr[pd->didx].v4.s_addr, 0);
3418                                 break;
3419 #endif /* INET */
3420 #ifdef INET6
3421                         case AF_INET6:
3422                                 if (PF_ANEQ(saddr,
3423                                     &nk->addr[pd->sidx], AF_INET6))
3424                                         PF_ACPY(saddr, &nk->addr[pd->sidx], af);
3425
3426                                 if (PF_ANEQ(daddr,
3427                                     &nk->addr[pd->didx], AF_INET6))
3428                                         PF_ACPY(saddr, &nk->addr[pd->didx], af);
3429                                 break;
3430 #endif /* INET */
3431                         }
3432                         break;
3433                 }
3434                 if (nr->natpass)
3435                         r = NULL;
3436                 pd->nat_rule = nr;
3437         }
3438
3439         while (r != NULL) {
3440                 r->evaluations++;
3441                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3442                         r = r->skip[PF_SKIP_IFP].ptr;
3443                 else if (r->direction && r->direction != direction)
3444                         r = r->skip[PF_SKIP_DIR].ptr;
3445                 else if (r->af && r->af != af)
3446                         r = r->skip[PF_SKIP_AF].ptr;
3447                 else if (r->proto && r->proto != pd->proto)
3448                         r = r->skip[PF_SKIP_PROTO].ptr;
3449                 else if (PF_MISMATCHAW(&r->src.addr, saddr, af,
3450                     r->src.neg, kif, M_GETFIB(m)))
3451                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3452                 /* tcp/udp only. port_op always 0 in other cases */
3453                 else if (r->src.port_op && !pf_match_port(r->src.port_op,
3454                     r->src.port[0], r->src.port[1], sport))
3455                         r = r->skip[PF_SKIP_SRC_PORT].ptr;
3456                 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af,
3457                     r->dst.neg, NULL, M_GETFIB(m)))
3458                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3459                 /* tcp/udp only. port_op always 0 in other cases */
3460                 else if (r->dst.port_op && !pf_match_port(r->dst.port_op,
3461                     r->dst.port[0], r->dst.port[1], dport))
3462                         r = r->skip[PF_SKIP_DST_PORT].ptr;
3463                 /* icmp only. type always 0 in other cases */
3464                 else if (r->type && r->type != icmptype + 1)
3465                         r = TAILQ_NEXT(r, entries);
3466                 /* icmp only. type always 0 in other cases */
3467                 else if (r->code && r->code != icmpcode + 1)
3468                         r = TAILQ_NEXT(r, entries);
3469                 else if (r->tos && !(r->tos == pd->tos))
3470                         r = TAILQ_NEXT(r, entries);
3471                 else if (r->rule_flag & PFRULE_FRAGMENT)
3472                         r = TAILQ_NEXT(r, entries);
3473                 else if (pd->proto == IPPROTO_TCP &&
3474                     (r->flagset & th->th_flags) != r->flags)
3475                         r = TAILQ_NEXT(r, entries);
3476                 /* tcp/udp only. uid.op always 0 in other cases */
3477                 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done =
3478                     pf_socket_lookup(direction, pd, m), 1)) &&
3479                     !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1],
3480                     pd->lookup.uid))
3481                         r = TAILQ_NEXT(r, entries);
3482                 /* tcp/udp only. gid.op always 0 in other cases */
3483                 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done =
3484                     pf_socket_lookup(direction, pd, m), 1)) &&
3485                     !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1],
3486                     pd->lookup.gid))
3487                         r = TAILQ_NEXT(r, entries);
3488                 else if (r->prio &&
3489                     !pf_match_ieee8021q_pcp(r->prio, m))
3490                         r = TAILQ_NEXT(r, entries);
3491                 else if (r->prob &&
3492                     r->prob <= arc4random())
3493                         r = TAILQ_NEXT(r, entries);
3494                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3495                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3496                         r = TAILQ_NEXT(r, entries);
3497                 else if (r->os_fingerprint != PF_OSFP_ANY &&
3498                     (pd->proto != IPPROTO_TCP || !pf_osfp_match(
3499                     pf_osfp_fingerprint(pd, m, off, th),
3500                     r->os_fingerprint)))
3501                         r = TAILQ_NEXT(r, entries);
3502                 else {
3503                         if (r->tag)
3504                                 tag = r->tag;
3505                         if (r->rtableid >= 0)
3506                                 rtableid = r->rtableid;
3507                         if (r->anchor == NULL) {
3508                                 match = 1;
3509                                 *rm = r;
3510                                 *am = a;
3511                                 *rsm = ruleset;
3512                                 if ((*rm)->quick)
3513                                         break;
3514                                 r = TAILQ_NEXT(r, entries);
3515                         } else
3516                                 pf_step_into_anchor(anchor_stack, &asd,
3517                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3518                                     &match);
3519                 }
3520                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3521                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3522                         break;
3523         }
3524         r = *rm;
3525         a = *am;
3526         ruleset = *rsm;
3527
3528         REASON_SET(&reason, PFRES_MATCH);
3529
3530         if (r->log || (nr != NULL && nr->log)) {
3531                 if (rewrite)
3532                         m_copyback(m, off, hdrlen, pd->hdr.any);
3533                 PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a,
3534                     ruleset, pd, 1);
3535         }
3536
3537         if ((r->action == PF_DROP) &&
3538             ((r->rule_flag & PFRULE_RETURNRST) ||
3539             (r->rule_flag & PFRULE_RETURNICMP) ||
3540             (r->rule_flag & PFRULE_RETURN))) {
3541                 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum,
3542                     bip_sum, hdrlen, &reason);
3543         }
3544
3545         if (r->action == PF_DROP)
3546                 goto cleanup;
3547
3548         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3549                 REASON_SET(&reason, PFRES_MEMORY);
3550                 goto cleanup;
3551         }
3552         if (rtableid >= 0)
3553                 M_SETFIB(m, rtableid);
3554
3555         if (!state_icmp && (r->keep_state || nr != NULL ||
3556             (pd->flags & PFDESC_TCP_NORM))) {
3557                 int action;
3558                 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off,
3559                     sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum,
3560                     hdrlen);
3561                 if (action != PF_PASS) {
3562                         if (action == PF_DROP &&
3563                             (r->rule_flag & PFRULE_RETURN))
3564                                 pf_return(r, nr, pd, sk, off, m, th, kif,
3565                                     bproto_sum, bip_sum, hdrlen, &reason);
3566                         return (action);
3567                 }
3568         } else {
3569                 if (sk != NULL)
3570                         uma_zfree(V_pf_state_key_z, sk);
3571                 if (nk != NULL)
3572                         uma_zfree(V_pf_state_key_z, nk);
3573         }
3574
3575         /* copy back packet headers if we performed NAT operations */
3576         if (rewrite)
3577                 m_copyback(m, off, hdrlen, pd->hdr.any);
3578
3579         if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) &&
3580             direction == PF_OUT &&
3581             pfsync_defer_ptr != NULL && pfsync_defer_ptr(*sm, m))
3582                 /*
3583                  * We want the state created, but we dont
3584                  * want to send this in case a partner
3585                  * firewall has to know about it to allow
3586                  * replies through it.
3587                  */
3588                 return (PF_DEFER);
3589
3590         return (PF_PASS);
3591
3592 cleanup:
3593         if (sk != NULL)
3594                 uma_zfree(V_pf_state_key_z, sk);
3595         if (nk != NULL)
3596                 uma_zfree(V_pf_state_key_z, nk);
3597         return (PF_DROP);
3598 }
3599
3600 static int
3601 pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a,
3602     struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk,
3603     struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport,
3604     u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm,
3605     int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen)
3606 {
3607         struct pf_state         *s = NULL;
3608         struct pf_src_node      *sn = NULL;
3609         struct tcphdr           *th = pd->hdr.tcp;
3610         u_int16_t                mss = V_tcp_mssdflt;
3611         u_short                  reason;
3612
3613         /* check maximums */
3614         if (r->max_states &&
3615             (counter_u64_fetch(r->states_cur) >= r->max_states)) {
3616                 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1);
3617                 REASON_SET(&reason, PFRES_MAXSTATES);
3618                 goto csfailed;
3619         }
3620         /* src node for filter rule */
3621         if ((r->rule_flag & PFRULE_SRCTRACK ||
3622             r->rpool.opts & PF_POOL_STICKYADDR) &&
3623             pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) {
3624                 REASON_SET(&reason, PFRES_SRCLIMIT);
3625                 goto csfailed;
3626         }
3627         /* src node for translation rule */
3628         if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) &&
3629             pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) {
3630                 REASON_SET(&reason, PFRES_SRCLIMIT);
3631                 goto csfailed;
3632         }
3633         s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO);
3634         if (s == NULL) {
3635                 REASON_SET(&reason, PFRES_MEMORY);
3636                 goto csfailed;
3637         }
3638         s->rule.ptr = r;
3639         s->nat_rule.ptr = nr;
3640         s->anchor.ptr = a;
3641         STATE_INC_COUNTERS(s);
3642         if (r->allow_opts)
3643                 s->state_flags |= PFSTATE_ALLOWOPTS;
3644         if (r->rule_flag & PFRULE_STATESLOPPY)
3645                 s->state_flags |= PFSTATE_SLOPPY;
3646         s->log = r->log & PF_LOG_ALL;
3647         s->sync_state = PFSYNC_S_NONE;
3648         if (nr != NULL)
3649                 s->log |= nr->log & PF_LOG_ALL;
3650         switch (pd->proto) {
3651         case IPPROTO_TCP:
3652                 s->src.seqlo = ntohl(th->th_seq);
3653                 s->src.seqhi = s->src.seqlo + pd->p_len + 1;
3654                 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN &&
3655                     r->keep_state == PF_STATE_MODULATE) {
3656                         /* Generate sequence number modulator */
3657                         if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) ==
3658                             0)
3659                                 s->src.seqdiff = 1;
3660                         pf_change_proto_a(m, &th->th_seq, &th->th_sum,
3661                             htonl(s->src.seqlo + s->src.seqdiff), 0);
3662                         *rewrite = 1;
3663                 } else
3664                         s->src.seqdiff = 0;
3665                 if (th->th_flags & TH_SYN) {
3666                         s->src.seqhi++;
3667                         s->src.wscale = pf_get_wscale(m, off,
3668                             th->th_off, pd->af);
3669                 }
3670                 s->src.max_win = MAX(ntohs(th->th_win), 1);
3671                 if (s->src.wscale & PF_WSCALE_MASK) {
3672                         /* Remove scale factor from initial window */
3673                         int win = s->src.max_win;
3674                         win += 1 << (s->src.wscale & PF_WSCALE_MASK);
3675                         s->src.max_win = (win - 1) >>
3676                             (s->src.wscale & PF_WSCALE_MASK);
3677                 }
3678                 if (th->th_flags & TH_FIN)
3679                         s->src.seqhi++;
3680                 s->dst.seqhi = 1;
3681                 s->dst.max_win = 1;
3682                 s->src.state = TCPS_SYN_SENT;
3683                 s->dst.state = TCPS_CLOSED;
3684                 s->timeout = PFTM_TCP_FIRST_PACKET;
3685                 break;
3686         case IPPROTO_UDP:
3687                 s->src.state = PFUDPS_SINGLE;
3688                 s->dst.state = PFUDPS_NO_TRAFFIC;
3689                 s->timeout = PFTM_UDP_FIRST_PACKET;
3690                 break;
3691         case IPPROTO_ICMP:
3692 #ifdef INET6
3693         case IPPROTO_ICMPV6:
3694 #endif
3695                 s->timeout = PFTM_ICMP_FIRST_PACKET;
3696                 break;
3697         default:
3698                 s->src.state = PFOTHERS_SINGLE;
3699                 s->dst.state = PFOTHERS_NO_TRAFFIC;
3700                 s->timeout = PFTM_OTHER_FIRST_PACKET;
3701         }
3702
3703         if (r->rt) {
3704                 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) {
3705                         REASON_SET(&reason, PFRES_MAPFAILED);
3706                         pf_src_tree_remove_state(s);
3707                         STATE_DEC_COUNTERS(s);
3708                         uma_zfree(V_pf_state_z, s);
3709                         goto csfailed;
3710                 }
3711                 s->rt_kif = r->rpool.cur->kif;
3712         }
3713
3714         s->creation = time_uptime;
3715         s->expire = time_uptime;
3716
3717         if (sn != NULL)
3718                 s->src_node = sn;
3719         if (nsn != NULL) {
3720                 /* XXX We only modify one side for now. */
3721                 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af);
3722                 s->nat_src_node = nsn;
3723         }
3724         if (pd->proto == IPPROTO_TCP) {
3725                 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m,
3726                     off, pd, th, &s->src, &s->dst)) {
3727                         REASON_SET(&reason, PFRES_MEMORY);
3728                         pf_src_tree_remove_state(s);
3729                         STATE_DEC_COUNTERS(s);
3730                         uma_zfree(V_pf_state_z, s);
3731                         return (PF_DROP);
3732                 }
3733                 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub &&
3734                     pf_normalize_tcp_stateful(m, off, pd, &reason, th, s,
3735                     &s->src, &s->dst, rewrite)) {
3736                         /* This really shouldn't happen!!! */
3737                         DPFPRINTF(PF_DEBUG_URGENT,
3738                             ("pf_normalize_tcp_stateful failed on first pkt"));
3739                         pf_normalize_tcp_cleanup(s);
3740                         pf_src_tree_remove_state(s);
3741                         STATE_DEC_COUNTERS(s);
3742                         uma_zfree(V_pf_state_z, s);
3743                         return (PF_DROP);
3744                 }
3745         }
3746         s->direction = pd->dir;
3747
3748         /*
3749          * sk/nk could already been setup by pf_get_translation().
3750          */
3751         if (nr == NULL) {
3752                 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p",
3753                     __func__, nr, sk, nk));
3754                 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport);
3755                 if (sk == NULL)
3756                         goto csfailed;
3757                 nk = sk;
3758         } else
3759                 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p",
3760                     __func__, nr, sk, nk));
3761
3762         /* Swap sk/nk for PF_OUT. */
3763         if (pf_state_insert(BOUND_IFACE(r, kif),
3764             (pd->dir == PF_IN) ? sk : nk,
3765             (pd->dir == PF_IN) ? nk : sk, s)) {
3766                 if (pd->proto == IPPROTO_TCP)
3767                         pf_normalize_tcp_cleanup(s);
3768                 REASON_SET(&reason, PFRES_STATEINS);
3769                 pf_src_tree_remove_state(s);
3770                 STATE_DEC_COUNTERS(s);
3771                 uma_zfree(V_pf_state_z, s);
3772                 return (PF_DROP);
3773         } else
3774                 *sm = s;
3775
3776         if (tag > 0)
3777                 s->tag = tag;
3778         if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) ==
3779             TH_SYN && r->keep_state == PF_STATE_SYNPROXY) {
3780                 s->src.state = PF_TCPS_PROXY_SRC;
3781                 /* undo NAT changes, if they have taken place */
3782                 if (nr != NULL) {
3783                         struct pf_state_key *skt = s->key[PF_SK_WIRE];
3784                         if (pd->dir == PF_OUT)
3785                                 skt = s->key[PF_SK_STACK];
3786                         PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af);
3787                         PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af);
3788                         if (pd->sport)
3789                                 *pd->sport = skt->port[pd->sidx];
3790                         if (pd->dport)
3791                                 *pd->dport = skt->port[pd->didx];
3792                         if (pd->proto_sum)
3793                                 *pd->proto_sum = bproto_sum;
3794                         if (pd->ip_sum)
3795                                 *pd->ip_sum = bip_sum;
3796                         m_copyback(m, off, hdrlen, pd->hdr.any);
3797                 }
3798                 s->src.seqhi = htonl(arc4random());
3799                 /* Find mss option */
3800                 int rtid = M_GETFIB(m);
3801                 mss = pf_get_mss(m, off, th->th_off, pd->af);
3802                 mss = pf_calc_mss(pd->src, pd->af, rtid, mss);
3803                 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss);
3804                 s->src.mss = mss;
3805                 pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport,
3806                     th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1,
3807                     TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL);
3808                 REASON_SET(&reason, PFRES_SYNPROXY);
3809                 return (PF_SYNPROXY_DROP);
3810         }
3811
3812         return (PF_PASS);
3813
3814 csfailed:
3815         if (sk != NULL)
3816                 uma_zfree(V_pf_state_key_z, sk);
3817         if (nk != NULL)
3818                 uma_zfree(V_pf_state_key_z, nk);
3819
3820         if (sn != NULL) {
3821                 struct pf_srchash *sh;
3822
3823                 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)];
3824                 PF_HASHROW_LOCK(sh);
3825                 if (--sn->states == 0 && sn->expire == 0) {
3826                         pf_unlink_src_node(sn);
3827                         uma_zfree(V_pf_sources_z, sn);
3828                         counter_u64_add(
3829                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3830                 }
3831                 PF_HASHROW_UNLOCK(sh);
3832         }
3833
3834         if (nsn != sn && nsn != NULL) {
3835                 struct pf_srchash *sh;
3836
3837                 sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)];
3838                 PF_HASHROW_LOCK(sh);
3839                 if (--nsn->states == 0 && nsn->expire == 0) {
3840                         pf_unlink_src_node(nsn);
3841                         uma_zfree(V_pf_sources_z, nsn);
3842                         counter_u64_add(
3843                             V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1);
3844                 }
3845                 PF_HASHROW_UNLOCK(sh);
3846         }
3847
3848         return (PF_DROP);
3849 }
3850
3851 static int
3852 pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif,
3853     struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am,
3854     struct pf_ruleset **rsm)
3855 {
3856         struct pf_rule          *r, *a = NULL;
3857         struct pf_ruleset       *ruleset = NULL;
3858         sa_family_t              af = pd->af;
3859         u_short                  reason;
3860         int                      tag = -1;
3861         int                      asd = 0;
3862         int                      match = 0;
3863         struct pf_anchor_stackframe     anchor_stack[PF_ANCHOR_STACKSIZE];
3864
3865         PF_RULES_RASSERT();
3866
3867         r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr);
3868         while (r != NULL) {
3869                 r->evaluations++;
3870                 if (pfi_kif_match(r->kif, kif) == r->ifnot)
3871                         r = r->skip[PF_SKIP_IFP].ptr;
3872                 else if (r->direction && r->direction != direction)
3873                         r = r->skip[PF_SKIP_DIR].ptr;
3874                 else if (r->af && r->af != af)
3875                         r = r->skip[PF_SKIP_AF].ptr;
3876                 else if (r->proto && r->proto != pd->proto)
3877                         r = r->skip[PF_SKIP_PROTO].ptr;
3878                 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af,
3879                     r->src.neg, kif, M_GETFIB(m)))
3880                         r = r->skip[PF_SKIP_SRC_ADDR].ptr;
3881                 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af,
3882                     r->dst.neg, NULL, M_GETFIB(m)))
3883                         r = r->skip[PF_SKIP_DST_ADDR].ptr;
3884                 else if (r->tos && !(r->tos == pd->tos))
3885                         r = TAILQ_NEXT(r, entries);
3886                 else if (r->os_fingerprint != PF_OSFP_ANY)
3887                         r = TAILQ_NEXT(r, entries);
3888                 else if (pd->proto == IPPROTO_UDP &&
3889                     (r->src.port_op || r->dst.port_op))
3890                         r = TAILQ_NEXT(r, entries);
3891                 else if (pd->proto == IPPROTO_TCP &&
3892                     (r->src.port_op || r->dst.port_op || r->flagset))
3893                         r = TAILQ_NEXT(r, entries);
3894                 else if ((pd->proto == IPPROTO_ICMP ||
3895                     pd->proto == IPPROTO_ICMPV6) &&
3896                     (r->type || r->code))
3897                         r = TAILQ_NEXT(r, entries);
3898                 else if (r->prio &&
3899                     !pf_match_ieee8021q_pcp(r->prio, m))
3900                         r = TAILQ_NEXT(r, entries);
3901                 else if (r->prob && r->prob <=
3902                     (arc4random() % (UINT_MAX - 1) + 1))
3903                         r = TAILQ_NEXT(r, entries);
3904                 else if (r->match_tag && !pf_match_tag(m, r, &tag,
3905                     pd->pf_mtag ? pd->pf_mtag->tag : 0))
3906                         r = TAILQ_NEXT(r, entries);
3907                 else {
3908                         if (r->anchor == NULL) {
3909                                 match = 1;
3910                                 *rm = r;
3911                                 *am = a;
3912                                 *rsm = ruleset;
3913                                 if ((*rm)->quick)
3914                                         break;
3915                                 r = TAILQ_NEXT(r, entries);
3916                         } else
3917                                 pf_step_into_anchor(anchor_stack, &asd,
3918                                     &ruleset, PF_RULESET_FILTER, &r, &a,
3919                                     &match);
3920                 }
3921                 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd,
3922                     &ruleset, PF_RULESET_FILTER, &r, &a, &match))
3923                         break;
3924         }
3925         r = *rm;
3926         a = *am;
3927         ruleset = *rsm;
3928
3929         REASON_SET(&reason, PFRES_MATCH);
3930
3931         if (r->log)
3932                 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd,
3933                     1);
3934
3935         if (r->action != PF_PASS)
3936                 return (PF_DROP);
3937
3938         if (tag > 0 && pf_tag_packet(m, pd, tag)) {
3939                 REASON_SET(&reason, PFRES_MEMORY);
3940                 return (PF_DROP);
3941         }
3942
3943         return (PF_PASS);
3944 }
3945
3946 static int
3947 pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst,
3948         struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off,
3949         struct pf_pdesc *pd, u_short *reason, int *copyback)
3950 {
3951         struct tcphdr           *th = pd->hdr.tcp;
3952         u_int16_t                win = ntohs(th->th_win);
3953         u_int32_t                ack, end, seq, orig_seq;
3954         u_int8_t                 sws, dws;
3955         int                      ackskew;
3956
3957         if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) {
3958                 sws = src->wscale & PF_WSCALE_MASK;
3959                 dws = dst->wscale & PF_WSCALE_MASK;
3960         } else
3961                 sws = dws = 0;
3962
3963         /*
3964          * Sequence tracking algorithm from Guido van Rooij's paper:
3965          *   http://www.madison-gurkha.com/publications/tcp_filtering/
3966          *      tcp_filtering.ps
3967          */
3968
3969         orig_seq = seq = ntohl(th->th_seq);
3970         if (src->seqlo == 0) {
3971                 /* First packet from this end. Set its state */
3972
3973                 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) &&
3974                     src->scrub == NULL) {
3975                         if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) {
3976                                 REASON_SET(reason, PFRES_MEMORY);
3977                                 return (PF_DROP);
3978                         }
3979                 }
3980
3981                 /* Deferred generation of sequence number modulator */
3982                 if (dst->seqdiff && !src->seqdiff) {
3983                         /* use random iss for the TCP server */
3984                         while ((src->seqdiff = arc4random() - seq) == 0)
3985                                 ;
3986                         ack = ntohl(th->th_ack) - dst->seqdiff;
3987                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
3988                             src->seqdiff), 0);
3989                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
3990                         *copyback = 1;
3991                 } else {
3992                         ack = ntohl(th->th_ack);
3993                 }
3994
3995                 end = seq + pd->p_len;
3996                 if (th->th_flags & TH_SYN) {
3997                         end++;
3998                         if (dst->wscale & PF_WSCALE_FLAG) {
3999                                 src->wscale = pf_get_wscale(m, off, th->th_off,
4000                                     pd->af);
4001                                 if (src->wscale & PF_WSCALE_FLAG) {
4002                                         /* Remove scale factor from initial
4003                                          * window */
4004                                         sws = src->wscale & PF_WSCALE_MASK;
4005                                         win = ((u_int32_t)win + (1 << sws) - 1)
4006                                             >> sws;
4007                                         dws = dst->wscale & PF_WSCALE_MASK;
4008                                 } else {
4009                                         /* fixup other window */
4010                                         dst->max_win <<= dst->wscale &
4011                                             PF_WSCALE_MASK;
4012                                         /* in case of a retrans SYN|ACK */
4013                                         dst->wscale = 0;
4014                                 }
4015                         }
4016                 }
4017                 if (th->th_flags & TH_FIN)
4018                         end++;
4019
4020                 src->seqlo = seq;
4021                 if (src->state < TCPS_SYN_SENT)
4022                         src->state = TCPS_SYN_SENT;
4023
4024                 /*
4025                  * May need to slide the window (seqhi may have been set by
4026                  * the crappy stack check or if we picked up the connection
4027                  * after establishment)
4028                  */
4029                 if (src->seqhi == 1 ||
4030                     SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi))
4031                         src->seqhi = end + MAX(1, dst->max_win << dws);
4032                 if (win > src->max_win)
4033                         src->max_win = win;
4034
4035         } else {
4036                 ack = ntohl(th->th_ack) - dst->seqdiff;
4037                 if (src->seqdiff) {
4038                         /* Modulate sequence numbers */
4039                         pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq +
4040                             src->seqdiff), 0);
4041                         pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0);
4042                         *copyback = 1;
4043                 }
4044                 end = seq + pd->p_len;
4045                 if (th->th_flags & TH_SYN)
4046                         end++;
4047                 if (th->th_flags & TH_FIN)
4048                         end++;
4049         }
4050
4051         if ((th->th_flags & TH_ACK) == 0) {
4052                 /* Let it pass through the ack skew check */
4053                 ack = dst->seqlo;
4054         } else if ((ack == 0 &&
4055             (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) ||
4056             /* broken tcp stacks do not set ack */
4057             (dst->state < TCPS_SYN_SENT)) {
4058                 /*
4059                  * Many stacks (ours included) will set the ACK number in an
4060                  * FIN|ACK if the SYN times out -- no sequence to ACK.
4061                  */
4062                 ack = dst->seqlo;
4063         }
4064
4065         if (seq == end) {
4066                 /* Ease sequencing restrictions on no data packets */
4067                 seq = src->seqlo;
4068                 end = seq;
4069         }
4070
4071         ackskew = dst->seqlo - ack;
4072
4073
4074         /*
4075          * Need to demodulate the sequence numbers in any TCP SACK options
4076          * (Selective ACK). We could optionally validate the SACK values
4077          * against the current ACK window, either forwards or backwards, but
4078          * I'm not confident that SACK has been implemented properly
4079          * everywhere. It wouldn't surprise me if several stacks accidentally
4080          * SACK too far backwards of previously ACKed data. There really aren't
4081          * any security implications of bad SACKing unless the target stack
4082          * doesn't validate the option length correctly. Someone trying to
4083          * spoof into a TCP connection won't bother blindly sending SACK
4084          * options anyway.
4085          */
4086         if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) {
4087                 if (pf_modulate_sack(m, off, pd, th, dst))
4088                         *copyback = 1;
4089         }
4090
4091
4092 #define MAXACKWINDOW (0xffff + 1500)    /* 1500 is an arbitrary fudge factor */
4093         if (SEQ_GEQ(src->seqhi, end) &&
4094             /* Last octet inside other's window space */
4095             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) &&
4096             /* Retrans: not more than one window back */
4097             (ackskew >= -MAXACKWINDOW) &&
4098             /* Acking not more than one reassembled fragment backwards */
4099             (ackskew <= (MAXACKWINDOW << sws)) &&
4100             /* Acking not more than one window forward */
4101             ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo ||
4102             (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) ||
4103             (pd->flags & PFDESC_IP_REAS) == 0)) {
4104             /* Require an exact/+1 sequence match on resets when possible */
4105
4106                 if (dst->scrub || src->scrub) {
4107                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4108                             *state, src, dst, copyback))
4109                                 return (PF_DROP);
4110                 }
4111
4112                 /* update max window */
4113                 if (src->max_win < win)
4114                         src->max_win = win;
4115                 /* synchronize sequencing */
4116                 if (SEQ_GT(end, src->seqlo))
4117                         src->seqlo = end;
4118                 /* slide the window of what the other end can send */
4119                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4120                         dst->seqhi = ack + MAX((win << sws), 1);
4121
4122
4123                 /* update states */
4124                 if (th->th_flags & TH_SYN)
4125                         if (src->state < TCPS_SYN_SENT)
4126                                 src->state = TCPS_SYN_SENT;
4127                 if (th->th_flags & TH_FIN)
4128                         if (src->state < TCPS_CLOSING)
4129                                 src->state = TCPS_CLOSING;
4130                 if (th->th_flags & TH_ACK) {
4131                         if (dst->state == TCPS_SYN_SENT) {
4132                                 dst->state = TCPS_ESTABLISHED;
4133                                 if (src->state == TCPS_ESTABLISHED &&
4134                                     (*state)->src_node != NULL &&
4135                                     pf_src_connlimit(state)) {
4136                                         REASON_SET(reason, PFRES_SRCLIMIT);
4137                                         return (PF_DROP);
4138                                 }
4139                         } else if (dst->state == TCPS_CLOSING)
4140                                 dst->state = TCPS_FIN_WAIT_2;
4141                 }
4142                 if (th->th_flags & TH_RST)
4143                         src->state = dst->state = TCPS_TIME_WAIT;
4144
4145                 /* update expire time */
4146                 (*state)->expire = time_uptime;
4147                 if (src->state >= TCPS_FIN_WAIT_2 &&
4148                     dst->state >= TCPS_FIN_WAIT_2)
4149                         (*state)->timeout = PFTM_TCP_CLOSED;
4150                 else if (src->state >= TCPS_CLOSING &&
4151                     dst->state >= TCPS_CLOSING)
4152                         (*state)->timeout = PFTM_TCP_FIN_WAIT;
4153                 else if (src->state < TCPS_ESTABLISHED ||
4154                     dst->state < TCPS_ESTABLISHED)
4155                         (*state)->timeout = PFTM_TCP_OPENING;
4156                 else if (src->state >= TCPS_CLOSING ||
4157                     dst->state >= TCPS_CLOSING)
4158                         (*state)->timeout = PFTM_TCP_CLOSING;
4159                 else
4160                         (*state)->timeout = PFTM_TCP_ESTABLISHED;
4161
4162                 /* Fall through to PASS packet */
4163
4164         } else if ((dst->state < TCPS_SYN_SENT ||
4165                 dst->state >= TCPS_FIN_WAIT_2 ||
4166                 src->state >= TCPS_FIN_WAIT_2) &&
4167             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) &&
4168             /* Within a window forward of the originating packet */
4169             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
4170             /* Within a window backward of the originating packet */
4171
4172                 /*
4173                  * This currently handles three situations:
4174                  *  1) Stupid stacks will shotgun SYNs before their peer
4175                  *     replies.
4176                  *  2) When PF catches an already established stream (the
4177                  *     firewall rebooted, the state table was flushed, routes
4178                  *     changed...)
4179                  *  3) Packets get funky immediately after the connection
4180                  *     closes (this should catch Solaris spurious ACK|FINs
4181                  *     that web servers like to spew after a close)
4182                  *
4183                  * This must be a little more careful than the above code
4184                  * since packet floods will also be caught here. We don't
4185                  * update the TTL here to mitigate the damage of a packet
4186                  * flood and so the same code can handle awkward establishment
4187                  * and a loosened connection close.
4188                  * In the establishment case, a correct peer response will
4189                  * validate the connection, go through the normal state code
4190                  * and keep updating the state TTL.
4191                  */
4192
4193                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4194                         printf("pf: loose state match: ");
4195                         pf_print_state(*state);
4196                         pf_print_flags(th->th_flags);
4197                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4198                             "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack,
4199                             pd->p_len, ackskew, (unsigned long long)(*state)->packets[0],
4200                             (unsigned long long)(*state)->packets[1],
4201                             pd->dir == PF_IN ? "in" : "out",
4202                             pd->dir == (*state)->direction ? "fwd" : "rev");
4203                 }
4204
4205                 if (dst->scrub || src->scrub) {
4206                         if (pf_normalize_tcp_stateful(m, off, pd, reason, th,
4207                             *state, src, dst, copyback))
4208                                 return (PF_DROP);
4209                 }
4210
4211                 /* update max window */
4212                 if (src->max_win < win)
4213                         src->max_win = win;
4214                 /* synchronize sequencing */
4215                 if (SEQ_GT(end, src->seqlo))
4216                         src->seqlo = end;
4217                 /* slide the window of what the other end can send */
4218                 if (SEQ_GEQ(ack + (win << sws), dst->seqhi))
4219                         dst->seqhi = ack + MAX((win << sws), 1);
4220
4221                 /*
4222                  * Cannot set dst->seqhi here since this could be a shotgunned
4223                  * SYN and not an already established connection.
4224                  */
4225
4226                 if (th->th_flags & TH_FIN)
4227                         if (src->state < TCPS_CLOSING)
4228                                 src->state = TCPS_CLOSING;
4229                 if (th->th_flags & TH_RST)
4230                         src->state = dst->state = TCPS_TIME_WAIT;
4231
4232                 /* Fall through to PASS packet */
4233
4234         } else {
4235                 if ((*state)->dst.state == TCPS_SYN_SENT &&
4236                     (*state)->src.state == TCPS_SYN_SENT) {
4237                         /* Send RST for state mismatches during handshake */
4238                         if (!(th->th_flags & TH_RST))
4239                                 pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4240                                     pd->dst, pd->src, th->th_dport,
4241                                     th->th_sport, ntohl(th->th_ack), 0,
4242                                     TH_RST, 0, 0,
4243                                     (*state)->rule.ptr->return_ttl, 1, 0,
4244                                     kif->pfik_ifp);
4245                         src->seqlo = 0;
4246                         src->seqhi = 1;
4247                         src->max_win = 1;
4248                 } else if (V_pf_status.debug >= PF_DEBUG_MISC) {
4249                         printf("pf: BAD state: ");
4250                         pf_print_state(*state);
4251                         pf_print_flags(th->th_flags);
4252                         printf(" seq=%u (%u) ack=%u len=%u ackskew=%d "
4253                             "pkts=%llu:%llu dir=%s,%s\n",
4254                             seq, orig_seq, ack, pd->p_len, ackskew,
4255                             (unsigned long long)(*state)->packets[0],
4256                             (unsigned long long)(*state)->packets[1],
4257                             pd->dir == PF_IN ? "in" : "out",
4258                             pd->dir == (*state)->direction ? "fwd" : "rev");
4259                         printf("pf: State failure on: %c %c %c %c | %c %c\n",
4260                             SEQ_GEQ(src->seqhi, end) ? ' ' : '1',
4261                             SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ?
4262                             ' ': '2',
4263                             (ackskew >= -MAXACKWINDOW) ? ' ' : '3',
4264                             (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4',
4265                             SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5',
4266                             SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6');
4267                 }
4268                 REASON_SET(reason, PFRES_BADSTATE);
4269                 return (PF_DROP);
4270         }
4271
4272         return (PF_PASS);
4273 }
4274
4275 static int
4276 pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst,
4277         struct pf_state **state, struct pf_pdesc *pd, u_short *reason)
4278 {
4279         struct tcphdr           *th = pd->hdr.tcp;
4280
4281         if (th->th_flags & TH_SYN)
4282                 if (src->state < TCPS_SYN_SENT)
4283                         src->state = TCPS_SYN_SENT;
4284         if (th->th_flags & TH_FIN)
4285                 if (src->state < TCPS_CLOSING)
4286                         src->state = TCPS_CLOSING;
4287         if (th->th_flags & TH_ACK) {
4288                 if (dst->state == TCPS_SYN_SENT) {
4289                         dst->state = TCPS_ESTABLISHED;
4290                         if (src->state == TCPS_ESTABLISHED &&
4291                             (*state)->src_node != NULL &&
4292                             pf_src_connlimit(state)) {
4293                                 REASON_SET(reason, PFRES_SRCLIMIT);
4294                                 return (PF_DROP);
4295                         }
4296                 } else if (dst->state == TCPS_CLOSING) {
4297                         dst->state = TCPS_FIN_WAIT_2;
4298                 } else if (src->state == TCPS_SYN_SENT &&
4299                     dst->state < TCPS_SYN_SENT) {
4300                         /*
4301                          * Handle a special sloppy case where we only see one
4302                          * half of the connection. If there is a ACK after
4303                          * the initial SYN without ever seeing a packet from
4304                          * the destination, set the connection to established.
4305                          */
4306                         dst->state = src->state = TCPS_ESTABLISHED;
4307                         if ((*state)->src_node != NULL &&
4308                             pf_src_connlimit(state)) {
4309                                 REASON_SET(reason, PFRES_SRCLIMIT);
4310                                 return (PF_DROP);
4311                         }
4312                 } else if (src->state == TCPS_CLOSING &&
4313                     dst->state == TCPS_ESTABLISHED &&
4314                     dst->seqlo == 0) {
4315                         /*
4316                          * Handle the closing of half connections where we
4317                          * don't see the full bidirectional FIN/ACK+ACK
4318                          * handshake.
4319                          */
4320                         dst->state = TCPS_CLOSING;
4321                 }
4322         }
4323         if (th->th_flags & TH_RST)
4324                 src->state = dst->state = TCPS_TIME_WAIT;
4325
4326         /* update expire time */
4327         (*state)->expire = time_uptime;
4328         if (src->state >= TCPS_FIN_WAIT_2 &&
4329             dst->state >= TCPS_FIN_WAIT_2)
4330                 (*state)->timeout = PFTM_TCP_CLOSED;
4331         else if (src->state >= TCPS_CLOSING &&
4332             dst->state >= TCPS_CLOSING)
4333                 (*state)->timeout = PFTM_TCP_FIN_WAIT;
4334         else if (src->state < TCPS_ESTABLISHED ||
4335             dst->state < TCPS_ESTABLISHED)
4336                 (*state)->timeout = PFTM_TCP_OPENING;
4337         else if (src->state >= TCPS_CLOSING ||
4338             dst->state >= TCPS_CLOSING)
4339                 (*state)->timeout = PFTM_TCP_CLOSING;
4340         else
4341                 (*state)->timeout = PFTM_TCP_ESTABLISHED;
4342
4343         return (PF_PASS);
4344 }
4345
4346 static int
4347 pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif,
4348     struct mbuf *m, int off, void *h, struct pf_pdesc *pd,
4349     u_short *reason)
4350 {
4351         struct pf_state_key_cmp  key;
4352         struct tcphdr           *th = pd->hdr.tcp;
4353         int                      copyback = 0;
4354         struct pf_state_peer    *src, *dst;
4355         struct pf_state_key     *sk;
4356
4357         bzero(&key, sizeof(key));
4358         key.af = pd->af;
4359         key.proto = IPPROTO_TCP;
4360         if (direction == PF_IN) {       /* wire side, straight */
4361                 PF_ACPY(&key.addr[0], pd->src, key.af);
4362                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4363                 key.port[0] = th->th_sport;
4364                 key.port[1] = th->th_dport;
4365         } else {                        /* stack side, reverse */
4366                 PF_ACPY(&key.addr[1], pd->src, key.af);
4367                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4368                 key.port[1] = th->th_sport;
4369                 key.port[0] = th->th_dport;
4370         }
4371
4372         STATE_LOOKUP(kif, &key, direction, *state, pd);
4373
4374         if (direction == (*state)->direction) {
4375                 src = &(*state)->src;
4376                 dst = &(*state)->dst;
4377         } else {
4378                 src = &(*state)->dst;
4379                 dst = &(*state)->src;
4380         }
4381
4382         sk = (*state)->key[pd->didx];
4383
4384         if ((*state)->src.state == PF_TCPS_PROXY_SRC) {
4385                 if (direction != (*state)->direction) {
4386                         REASON_SET(reason, PFRES_SYNPROXY);
4387                         return (PF_SYNPROXY_DROP);
4388                 }
4389                 if (th->th_flags & TH_SYN) {
4390                         if (ntohl(th->th_seq) != (*state)->src.seqlo) {
4391                                 REASON_SET(reason, PFRES_SYNPROXY);
4392                                 return (PF_DROP);
4393                         }
4394                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4395                             pd->src, th->th_dport, th->th_sport,
4396                             (*state)->src.seqhi, ntohl(th->th_seq) + 1,
4397                             TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL);
4398                         REASON_SET(reason, PFRES_SYNPROXY);
4399                         return (PF_SYNPROXY_DROP);
4400                 } else if (!(th->th_flags & TH_ACK) ||
4401                     (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4402                     (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4403                         REASON_SET(reason, PFRES_SYNPROXY);
4404                         return (PF_DROP);
4405                 } else if ((*state)->src_node != NULL &&
4406                     pf_src_connlimit(state)) {
4407                         REASON_SET(reason, PFRES_SRCLIMIT);
4408                         return (PF_DROP);
4409                 } else
4410                         (*state)->src.state = PF_TCPS_PROXY_DST;
4411         }
4412         if ((*state)->src.state == PF_TCPS_PROXY_DST) {
4413                 if (direction == (*state)->direction) {
4414                         if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) ||
4415                             (ntohl(th->th_ack) != (*state)->src.seqhi + 1) ||
4416                             (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) {
4417                                 REASON_SET(reason, PFRES_SYNPROXY);
4418                                 return (PF_DROP);
4419                         }
4420                         (*state)->src.max_win = MAX(ntohs(th->th_win), 1);
4421                         if ((*state)->dst.seqhi == 1)
4422                                 (*state)->dst.seqhi = htonl(arc4random());
4423                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4424                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4425                             sk->port[pd->sidx], sk->port[pd->didx],
4426                             (*state)->dst.seqhi, 0, TH_SYN, 0,
4427                             (*state)->src.mss, 0, 0, (*state)->tag, NULL);
4428                         REASON_SET(reason, PFRES_SYNPROXY);
4429                         return (PF_SYNPROXY_DROP);
4430                 } else if (((th->th_flags & (TH_SYN|TH_ACK)) !=
4431                     (TH_SYN|TH_ACK)) ||
4432                     (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) {
4433                         REASON_SET(reason, PFRES_SYNPROXY);
4434                         return (PF_DROP);
4435                 } else {
4436                         (*state)->dst.max_win = MAX(ntohs(th->th_win), 1);
4437                         (*state)->dst.seqlo = ntohl(th->th_seq);
4438                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst,
4439                             pd->src, th->th_dport, th->th_sport,
4440                             ntohl(th->th_ack), ntohl(th->th_seq) + 1,
4441                             TH_ACK, (*state)->src.max_win, 0, 0, 0,
4442                             (*state)->tag, NULL);
4443                         pf_send_tcp(NULL, (*state)->rule.ptr, pd->af,
4444                             &sk->addr[pd->sidx], &sk->addr[pd->didx],
4445                             sk->port[pd->sidx], sk->port[pd->didx],
4446                             (*state)->src.seqhi + 1, (*state)->src.seqlo + 1,
4447                             TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL);
4448                         (*state)->src.seqdiff = (*state)->dst.seqhi -
4449                             (*state)->src.seqlo;
4450                         (*state)->dst.seqdiff = (*state)->src.seqhi -
4451                             (*state)->dst.seqlo;
4452                         (*state)->src.seqhi = (*state)->src.seqlo +
4453                             (*state)->dst.max_win;
4454                         (*state)->dst.seqhi = (*state)->dst.seqlo +
4455                             (*state)->src.max_win;
4456                         (*state)->src.wscale = (*state)->dst.wscale = 0;
4457                         (*state)->src.state = (*state)->dst.state =
4458                             TCPS_ESTABLISHED;
4459                         REASON_SET(reason, PFRES_SYNPROXY);
4460                         return (PF_SYNPROXY_DROP);
4461                 }
4462         }
4463
4464         if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) &&
4465             dst->state >= TCPS_FIN_WAIT_2 &&
4466             src->state >= TCPS_FIN_WAIT_2) {
4467                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4468                         printf("pf: state reuse ");
4469                         pf_print_state(*state);
4470                         pf_print_flags(th->th_flags);
4471                         printf("\n");
4472                 }
4473                 /* XXX make sure it's the same direction ?? */
4474                 (*state)->src.state = (*state)->dst.state = TCPS_CLOSED;
4475                 pf_unlink_state(*state, PF_ENTER_LOCKED);
4476                 *state = NULL;
4477                 return (PF_DROP);
4478         }
4479
4480         if ((*state)->state_flags & PFSTATE_SLOPPY) {
4481                 if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP)
4482                         return (PF_DROP);
4483         } else {
4484                 if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason,
4485                     &copyback) == PF_DROP)
4486                         return (PF_DROP);
4487         }
4488
4489         /* translate source/destination address, if necessary */
4490         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4491                 struct pf_state_key *nk = (*state)->key[pd->didx];
4492
4493                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4494                     nk->port[pd->sidx] != th->th_sport)
4495                         pf_change_ap(m, pd->src, &th->th_sport,
4496                             pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx],
4497                             nk->port[pd->sidx], 0, pd->af);
4498
4499                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4500                     nk->port[pd->didx] != th->th_dport)
4501                         pf_change_ap(m, pd->dst, &th->th_dport,
4502                             pd->ip_sum, &th->th_sum, &nk->addr[pd->didx],
4503                             nk->port[pd->didx], 0, pd->af);
4504                 copyback = 1;
4505         }
4506
4507         /* Copyback sequence modulation or stateful scrub changes if needed */
4508         if (copyback)
4509                 m_copyback(m, off, sizeof(*th), (caddr_t)th);
4510
4511         return (PF_PASS);
4512 }
4513
4514 static int
4515 pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif,
4516     struct mbuf *m, int off, void *h, struct pf_pdesc *pd)
4517 {
4518         struct pf_state_peer    *src, *dst;
4519         struct pf_state_key_cmp  key;
4520         struct udphdr           *uh = pd->hdr.udp;
4521
4522         bzero(&key, sizeof(key));
4523         key.af = pd->af;
4524         key.proto = IPPROTO_UDP;
4525         if (direction == PF_IN) {       /* wire side, straight */
4526                 PF_ACPY(&key.addr[0], pd->src, key.af);
4527                 PF_ACPY(&key.addr[1], pd->dst, key.af);
4528                 key.port[0] = uh->uh_sport;
4529                 key.port[1] = uh->uh_dport;
4530         } else {                        /* stack side, reverse */
4531                 PF_ACPY(&key.addr[1], pd->src, key.af);
4532                 PF_ACPY(&key.addr[0], pd->dst, key.af);
4533                 key.port[1] = uh->uh_sport;
4534                 key.port[0] = uh->uh_dport;
4535         }
4536
4537         STATE_LOOKUP(kif, &key, direction, *state, pd);
4538
4539         if (direction == (*state)->direction) {
4540                 src = &(*state)->src;
4541                 dst = &(*state)->dst;
4542         } else {
4543                 src = &(*state)->dst;
4544                 dst = &(*state)->src;
4545         }
4546
4547         /* update states */
4548         if (src->state < PFUDPS_SINGLE)
4549                 src->state = PFUDPS_SINGLE;
4550         if (dst->state == PFUDPS_SINGLE)
4551                 dst->state = PFUDPS_MULTIPLE;
4552
4553         /* update expire time */
4554         (*state)->expire = time_uptime;
4555         if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE)
4556                 (*state)->timeout = PFTM_UDP_MULTIPLE;
4557         else
4558                 (*state)->timeout = PFTM_UDP_SINGLE;
4559
4560         /* translate source/destination address, if necessary */
4561         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4562                 struct pf_state_key *nk = (*state)->key[pd->didx];
4563
4564                 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) ||
4565                     nk->port[pd->sidx] != uh->uh_sport)
4566                         pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum,
4567                             &uh->uh_sum, &nk->addr[pd->sidx],
4568                             nk->port[pd->sidx], 1, pd->af);
4569
4570                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) ||
4571                     nk->port[pd->didx] != uh->uh_dport)
4572                         pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum,
4573                             &uh->uh_sum, &nk->addr[pd->didx],
4574                             nk->port[pd->didx], 1, pd->af);
4575                 m_copyback(m, off, sizeof(*uh), (caddr_t)uh);
4576         }
4577
4578         return (PF_PASS);
4579 }
4580
4581 static int
4582 pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif,
4583     struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason)
4584 {
4585         struct pf_addr  *saddr = pd->src, *daddr = pd->dst;
4586         u_int16_t        icmpid = 0, *icmpsum;
4587         u_int8_t         icmptype;
4588         int              state_icmp = 0;
4589         struct pf_state_key_cmp key;
4590
4591         bzero(&key, sizeof(key));
4592         switch (pd->proto) {
4593 #ifdef INET
4594         case IPPROTO_ICMP:
4595                 icmptype = pd->hdr.icmp->icmp_type;
4596                 icmpid = pd->hdr.icmp->icmp_id;
4597                 icmpsum = &pd->hdr.icmp->icmp_cksum;
4598
4599                 if (icmptype == ICMP_UNREACH ||
4600                     icmptype == ICMP_SOURCEQUENCH ||
4601                     icmptype == ICMP_REDIRECT ||
4602                     icmptype == ICMP_TIMXCEED ||
4603                     icmptype == ICMP_PARAMPROB)
4604                         state_icmp++;
4605                 break;
4606 #endif /* INET */
4607 #ifdef INET6
4608         case IPPROTO_ICMPV6:
4609                 icmptype = pd->hdr.icmp6->icmp6_type;
4610                 icmpid = pd->hdr.icmp6->icmp6_id;
4611                 icmpsum = &pd->hdr.icmp6->icmp6_cksum;
4612
4613                 if (icmptype == ICMP6_DST_UNREACH ||
4614                     icmptype == ICMP6_PACKET_TOO_BIG ||
4615                     icmptype == ICMP6_TIME_EXCEEDED ||
4616                     icmptype == ICMP6_PARAM_PROB)
4617                         state_icmp++;
4618                 break;
4619 #endif /* INET6 */
4620         }
4621
4622         if (!state_icmp) {
4623
4624                 /*
4625                  * ICMP query/reply message not related to a TCP/UDP packet.
4626                  * Search for an ICMP state.
4627                  */
4628                 key.af = pd->af;
4629                 key.proto = pd->proto;
4630                 key.port[0] = key.port[1] = icmpid;
4631                 if (direction == PF_IN) {       /* wire side, straight */
4632                         PF_ACPY(&key.addr[0], pd->src, key.af);
4633                         PF_ACPY(&key.addr[1], pd->dst, key.af);
4634                 } else {                        /* stack side, reverse */
4635                         PF_ACPY(&key.addr[1], pd->src, key.af);
4636                         PF_ACPY(&key.addr[0], pd->dst, key.af);
4637                 }
4638
4639                 STATE_LOOKUP(kif, &key, direction, *state, pd);
4640
4641                 (*state)->expire = time_uptime;
4642                 (*state)->timeout = PFTM_ICMP_ERROR_REPLY;
4643
4644                 /* translate source/destination address, if necessary */
4645                 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
4646                         struct pf_state_key *nk = (*state)->key[pd->didx];
4647
4648                         switch (pd->af) {
4649 #ifdef INET
4650                         case AF_INET:
4651                                 if (PF_ANEQ(pd->src,
4652                                     &nk->addr[pd->sidx], AF_INET))
4653                                         pf_change_a(&saddr->v4.s_addr,
4654                                             pd->ip_sum,
4655                                             nk->addr[pd->sidx].v4.s_addr, 0);
4656
4657                                 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx],
4658                                     AF_INET))
4659                                         pf_change_a(&daddr->v4.s_addr,
4660                                             pd->ip_sum,
4661                                             nk->addr[pd->didx].v4.s_addr, 0);
4662
4663                                 if (nk->port[0] !=
4664                                     pd->hdr.icmp->icmp_id) {
4665                                         pd->hdr.icmp->icmp_cksum =
4666                                             pf_cksum_fixup(
4667                                             pd->hdr.icmp->icmp_cksum, icmpid,
4668                                             nk->port[pd->sidx], 0);
4669                                         pd->hdr.icmp->icmp_id =
4670                                             nk->port[pd->sidx];
4671                                 }
4672
4673                                 m_copyback(m, off, ICMP_MINLEN,
4674                                     (caddr_t )pd->hdr.icmp);
4675                                 break;
4676 #endif /* INET */
4677 #ifdef INET6
4678                         case AF_INET6:
4679                                 if (PF_ANEQ(pd->src,
4680                                     &nk->addr[pd->sidx], AF_INET6))
4681                                         pf_change_a6(saddr,
4682                                             &pd->hdr.icmp6->icmp6_cksum,
4683                                             &nk->addr[pd->sidx], 0);
4684
4685                                 if (PF_ANEQ(pd->dst,
4686                                     &nk->addr[pd->didx], AF_INET6))
4687                                         pf_change_a6(daddr,
4688                                             &pd->hdr.icmp6->icmp6_cksum,
4689                                             &nk->addr[pd->didx], 0);
4690
4691                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
4692                                     (caddr_t )pd->hdr.icmp6);
4693                                 break;
4694 #endif /* INET6 */
4695                         }
4696                 }
4697                 return (PF_PASS);
4698
4699         } else {
4700                 /*
4701                  * ICMP error message in response to a TCP/UDP packet.
4702                  * Extract the inner TCP/UDP header and search for that state.
4703                  */
4704
4705                 struct pf_pdesc pd2;
4706                 bzero(&pd2, sizeof pd2);
4707 #ifdef INET
4708                 struct ip       h2;
4709 #endif /* INET */
4710 #ifdef INET6
4711                 struct ip6_hdr  h2_6;
4712                 int             terminal = 0;
4713 #endif /* INET6 */
4714                 int             ipoff2 = 0;
4715                 int             off2 = 0;
4716
4717                 pd2.af = pd->af;
4718                 /* Payload packet is from the opposite direction. */
4719                 pd2.sidx = (direction == PF_IN) ? 1 : 0;
4720                 pd2.didx = (direction == PF_IN) ? 0 : 1;
4721                 switch (pd->af) {
4722 #ifdef INET
4723                 case AF_INET:
4724                         /* offset of h2 in mbuf chain */
4725                         ipoff2 = off + ICMP_MINLEN;
4726
4727                         if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2),
4728                             NULL, reason, pd2.af)) {
4729                                 DPFPRINTF(PF_DEBUG_MISC,
4730                                     ("pf: ICMP error message too short "
4731                                     "(ip)\n"));
4732                                 return (PF_DROP);
4733                         }
4734                         /*
4735                          * ICMP error messages don't refer to non-first
4736                          * fragments
4737                          */
4738                         if (h2.ip_off & htons(IP_OFFMASK)) {
4739                                 REASON_SET(reason, PFRES_FRAG);
4740                                 return (PF_DROP);
4741                         }
4742
4743                         /* offset of protocol header that follows h2 */
4744                         off2 = ipoff2 + (h2.ip_hl << 2);
4745
4746                         pd2.proto = h2.ip_p;
4747                         pd2.src = (struct pf_addr *)&h2.ip_src;
4748                         pd2.dst = (struct pf_addr *)&h2.ip_dst;
4749                         pd2.ip_sum = &h2.ip_sum;
4750                         break;
4751 #endif /* INET */
4752 #ifdef INET6
4753                 case AF_INET6:
4754                         ipoff2 = off + sizeof(struct icmp6_hdr);
4755
4756                         if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6),
4757                             NULL, reason, pd2.af)) {
4758                                 DPFPRINTF(PF_DEBUG_MISC,
4759                                     ("pf: ICMP error message too short "
4760                                     "(ip6)\n"));
4761                                 return (PF_DROP);
4762                         }
4763                         pd2.proto = h2_6.ip6_nxt;
4764                         pd2.src = (struct pf_addr *)&h2_6.ip6_src;
4765                         pd2.dst = (struct pf_addr *)&h2_6.ip6_dst;
4766                         pd2.ip_sum = NULL;
4767                         off2 = ipoff2 + sizeof(h2_6);
4768                         do {
4769                                 switch (pd2.proto) {
4770                                 case IPPROTO_FRAGMENT:
4771                                         /*
4772                                          * ICMPv6 error messages for
4773                                          * non-first fragments
4774                                          */
4775                                         REASON_SET(reason, PFRES_FRAG);
4776                                         return (PF_DROP);
4777                                 case IPPROTO_AH:
4778                                 case IPPROTO_HOPOPTS:
4779                                 case IPPROTO_ROUTING:
4780                                 case IPPROTO_DSTOPTS: {
4781                                         /* get next header and header length */
4782                                         struct ip6_ext opt6;
4783
4784                                         if (!pf_pull_hdr(m, off2, &opt6,
4785                                             sizeof(opt6), NULL, reason,
4786                                             pd2.af)) {
4787                                                 DPFPRINTF(PF_DEBUG_MISC,
4788                                                     ("pf: ICMPv6 short opt\n"));
4789                                                 return (PF_DROP);
4790                                         }
4791                                         if (pd2.proto == IPPROTO_AH)
4792                                                 off2 += (opt6.ip6e_len + 2) * 4;
4793                                         else
4794                                                 off2 += (opt6.ip6e_len + 1) * 8;
4795                                         pd2.proto = opt6.ip6e_nxt;
4796                                         /* goto the next header */
4797                                         break;
4798                                 }
4799                                 default:
4800                                         terminal++;
4801                                         break;
4802                                 }
4803                         } while (!terminal);
4804                         break;
4805 #endif /* INET6 */
4806                 }
4807
4808                 switch (pd2.proto) {
4809                 case IPPROTO_TCP: {
4810                         struct tcphdr            th;
4811                         u_int32_t                seq;
4812                         struct pf_state_peer    *src, *dst;
4813                         u_int8_t                 dws;
4814                         int                      copyback = 0;
4815
4816                         /*
4817                          * Only the first 8 bytes of the TCP header can be
4818                          * expected. Don't access any TCP header fields after
4819                          * th_seq, an ackskew test is not possible.
4820                          */
4821                         if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason,
4822                             pd2.af)) {
4823                                 DPFPRINTF(PF_DEBUG_MISC,
4824                                     ("pf: ICMP error message too short "
4825                                     "(tcp)\n"));
4826                                 return (PF_DROP);
4827                         }
4828
4829                         key.af = pd2.af;
4830                         key.proto = IPPROTO_TCP;
4831                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4832                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4833                         key.port[pd2.sidx] = th.th_sport;
4834                         key.port[pd2.didx] = th.th_dport;
4835
4836                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4837
4838                         if (direction == (*state)->direction) {
4839                                 src = &(*state)->dst;
4840                                 dst = &(*state)->src;
4841                         } else {
4842                                 src = &(*state)->src;
4843                                 dst = &(*state)->dst;
4844                         }
4845
4846                         if (src->wscale && dst->wscale)
4847                                 dws = dst->wscale & PF_WSCALE_MASK;
4848                         else
4849                                 dws = 0;
4850
4851                         /* Demodulate sequence number */
4852                         seq = ntohl(th.th_seq) - src->seqdiff;
4853                         if (src->seqdiff) {
4854                                 pf_change_a(&th.th_seq, icmpsum,
4855                                     htonl(seq), 0);
4856                                 copyback = 1;
4857                         }
4858
4859                         if (!((*state)->state_flags & PFSTATE_SLOPPY) &&
4860                             (!SEQ_GEQ(src->seqhi, seq) ||
4861                             !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) {
4862                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4863                                         printf("pf: BAD ICMP %d:%d ",
4864                                             icmptype, pd->hdr.icmp->icmp_code);
4865                                         pf_print_host(pd->src, 0, pd->af);
4866                                         printf(" -> ");
4867                                         pf_print_host(pd->dst, 0, pd->af);
4868                                         printf(" state: ");
4869                                         pf_print_state(*state);
4870                                         printf(" seq=%u\n", seq);
4871                                 }
4872                                 REASON_SET(reason, PFRES_BADSTATE);
4873                                 return (PF_DROP);
4874                         } else {
4875                                 if (V_pf_status.debug >= PF_DEBUG_MISC) {
4876                                         printf("pf: OK ICMP %d:%d ",
4877                                             icmptype, pd->hdr.icmp->icmp_code);
4878                                         pf_print_host(pd->src, 0, pd->af);
4879                                         printf(" -> ");
4880                                         pf_print_host(pd->dst, 0, pd->af);
4881                                         printf(" state: ");
4882                                         pf_print_state(*state);
4883                                         printf(" seq=%u\n", seq);
4884                                 }
4885                         }
4886
4887                         /* translate source/destination address, if necessary */
4888                         if ((*state)->key[PF_SK_WIRE] !=
4889                             (*state)->key[PF_SK_STACK]) {
4890                                 struct pf_state_key *nk =
4891                                     (*state)->key[pd->didx];
4892
4893                                 if (PF_ANEQ(pd2.src,
4894                                     &nk->addr[pd2.sidx], pd2.af) ||
4895                                     nk->port[pd2.sidx] != th.th_sport)
4896                                         pf_change_icmp(pd2.src, &th.th_sport,
4897                                             daddr, &nk->addr[pd2.sidx],
4898                                             nk->port[pd2.sidx], NULL,
4899                                             pd2.ip_sum, icmpsum,
4900                                             pd->ip_sum, 0, pd2.af);
4901
4902                                 if (PF_ANEQ(pd2.dst,
4903                                     &nk->addr[pd2.didx], pd2.af) ||
4904                                     nk->port[pd2.didx] != th.th_dport)
4905                                         pf_change_icmp(pd2.dst, &th.th_dport,
4906                                             saddr, &nk->addr[pd2.didx],
4907                                             nk->port[pd2.didx], NULL,
4908                                             pd2.ip_sum, icmpsum,
4909                                             pd->ip_sum, 0, pd2.af);
4910                                 copyback = 1;
4911                         }
4912
4913                         if (copyback) {
4914                                 switch (pd2.af) {
4915 #ifdef INET
4916                                 case AF_INET:
4917                                         m_copyback(m, off, ICMP_MINLEN,
4918                                             (caddr_t )pd->hdr.icmp);
4919                                         m_copyback(m, ipoff2, sizeof(h2),
4920                                             (caddr_t )&h2);
4921                                         break;
4922 #endif /* INET */
4923 #ifdef INET6
4924                                 case AF_INET6:
4925                                         m_copyback(m, off,
4926                                             sizeof(struct icmp6_hdr),
4927                                             (caddr_t )pd->hdr.icmp6);
4928                                         m_copyback(m, ipoff2, sizeof(h2_6),
4929                                             (caddr_t )&h2_6);
4930                                         break;
4931 #endif /* INET6 */
4932                                 }
4933                                 m_copyback(m, off2, 8, (caddr_t)&th);
4934                         }
4935
4936                         return (PF_PASS);
4937                         break;
4938                 }
4939                 case IPPROTO_UDP: {
4940                         struct udphdr           uh;
4941
4942                         if (!pf_pull_hdr(m, off2, &uh, sizeof(uh),
4943                             NULL, reason, pd2.af)) {
4944                                 DPFPRINTF(PF_DEBUG_MISC,
4945                                     ("pf: ICMP error message too short "
4946                                     "(udp)\n"));
4947                                 return (PF_DROP);
4948                         }
4949
4950                         key.af = pd2.af;
4951                         key.proto = IPPROTO_UDP;
4952                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
4953                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
4954                         key.port[pd2.sidx] = uh.uh_sport;
4955                         key.port[pd2.didx] = uh.uh_dport;
4956
4957                         STATE_LOOKUP(kif, &key, direction, *state, pd);
4958
4959                         /* translate source/destination address, if necessary */
4960                         if ((*state)->key[PF_SK_WIRE] !=
4961                             (*state)->key[PF_SK_STACK]) {
4962                                 struct pf_state_key *nk =
4963                                     (*state)->key[pd->didx];
4964
4965                                 if (PF_ANEQ(pd2.src,
4966                                     &nk->addr[pd2.sidx], pd2.af) ||
4967                                     nk->port[pd2.sidx] != uh.uh_sport)
4968                                         pf_change_icmp(pd2.src, &uh.uh_sport,
4969                                             daddr, &nk->addr[pd2.sidx],
4970                                             nk->port[pd2.sidx], &uh.uh_sum,
4971                                             pd2.ip_sum, icmpsum,
4972                                             pd->ip_sum, 1, pd2.af);
4973
4974                                 if (PF_ANEQ(pd2.dst,
4975                                     &nk->addr[pd2.didx], pd2.af) ||
4976                                     nk->port[pd2.didx] != uh.uh_dport)
4977                                         pf_change_icmp(pd2.dst, &uh.uh_dport,
4978                                             saddr, &nk->addr[pd2.didx],
4979                                             nk->port[pd2.didx], &uh.uh_sum,
4980                                             pd2.ip_sum, icmpsum,
4981                                             pd->ip_sum, 1, pd2.af);
4982
4983                                 switch (pd2.af) {
4984 #ifdef INET
4985                                 case AF_INET:
4986                                         m_copyback(m, off, ICMP_MINLEN,
4987                                             (caddr_t )pd->hdr.icmp);
4988                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
4989                                         break;
4990 #endif /* INET */
4991 #ifdef INET6
4992                                 case AF_INET6:
4993                                         m_copyback(m, off,
4994                                             sizeof(struct icmp6_hdr),
4995                                             (caddr_t )pd->hdr.icmp6);
4996                                         m_copyback(m, ipoff2, sizeof(h2_6),
4997                                             (caddr_t )&h2_6);
4998                                         break;
4999 #endif /* INET6 */
5000                                 }
5001                                 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh);
5002                         }
5003                         return (PF_PASS);
5004                         break;
5005                 }
5006 #ifdef INET
5007                 case IPPROTO_ICMP: {
5008                         struct icmp             iih;
5009
5010                         if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN,
5011                             NULL, reason, pd2.af)) {
5012                                 DPFPRINTF(PF_DEBUG_MISC,
5013                                     ("pf: ICMP error message too short i"
5014                                     "(icmp)\n"));
5015                                 return (PF_DROP);
5016                         }
5017
5018                         key.af = pd2.af;
5019                         key.proto = IPPROTO_ICMP;
5020                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5021                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5022                         key.port[0] = key.port[1] = iih.icmp_id;
5023
5024                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5025
5026                         /* translate source/destination address, if necessary */
5027                         if ((*state)->key[PF_SK_WIRE] !=
5028                             (*state)->key[PF_SK_STACK]) {
5029                                 struct pf_state_key *nk =
5030                                     (*state)->key[pd->didx];
5031
5032                                 if (PF_ANEQ(pd2.src,
5033                                     &nk->addr[pd2.sidx], pd2.af) ||
5034                                     nk->port[pd2.sidx] != iih.icmp_id)
5035                                         pf_change_icmp(pd2.src, &iih.icmp_id,
5036                                             daddr, &nk->addr[pd2.sidx],
5037                                             nk->port[pd2.sidx], NULL,
5038                                             pd2.ip_sum, icmpsum,
5039                                             pd->ip_sum, 0, AF_INET);
5040
5041                                 if (PF_ANEQ(pd2.dst,
5042                                     &nk->addr[pd2.didx], pd2.af) ||
5043                                     nk->port[pd2.didx] != iih.icmp_id)
5044                                         pf_change_icmp(pd2.dst, &iih.icmp_id,
5045                                             saddr, &nk->addr[pd2.didx],
5046                                             nk->port[pd2.didx], NULL,
5047                                             pd2.ip_sum, icmpsum,
5048                                             pd->ip_sum, 0, AF_INET);
5049
5050                                 m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp);
5051                                 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5052                                 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih);
5053                         }
5054                         return (PF_PASS);
5055                         break;
5056                 }
5057 #endif /* INET */
5058 #ifdef INET6
5059                 case IPPROTO_ICMPV6: {
5060                         struct icmp6_hdr        iih;
5061
5062                         if (!pf_pull_hdr(m, off2, &iih,
5063                             sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) {
5064                                 DPFPRINTF(PF_DEBUG_MISC,
5065                                     ("pf: ICMP error message too short "
5066                                     "(icmp6)\n"));
5067                                 return (PF_DROP);
5068                         }
5069
5070                         key.af = pd2.af;
5071                         key.proto = IPPROTO_ICMPV6;
5072                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5073                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5074                         key.port[0] = key.port[1] = iih.icmp6_id;
5075
5076                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5077
5078                         /* translate source/destination address, if necessary */
5079                         if ((*state)->key[PF_SK_WIRE] !=
5080                             (*state)->key[PF_SK_STACK]) {
5081                                 struct pf_state_key *nk =
5082                                     (*state)->key[pd->didx];
5083
5084                                 if (PF_ANEQ(pd2.src,
5085                                     &nk->addr[pd2.sidx], pd2.af) ||
5086                                     nk->port[pd2.sidx] != iih.icmp6_id)
5087                                         pf_change_icmp(pd2.src, &iih.icmp6_id,
5088                                             daddr, &nk->addr[pd2.sidx],
5089                                             nk->port[pd2.sidx], NULL,
5090                                             pd2.ip_sum, icmpsum,
5091                                             pd->ip_sum, 0, AF_INET6);
5092
5093                                 if (PF_ANEQ(pd2.dst,
5094                                     &nk->addr[pd2.didx], pd2.af) ||
5095                                     nk->port[pd2.didx] != iih.icmp6_id)
5096                                         pf_change_icmp(pd2.dst, &iih.icmp6_id,
5097                                             saddr, &nk->addr[pd2.didx],
5098                                             nk->port[pd2.didx], NULL,
5099                                             pd2.ip_sum, icmpsum,
5100                                             pd->ip_sum, 0, AF_INET6);
5101
5102                                 m_copyback(m, off, sizeof(struct icmp6_hdr),
5103                                     (caddr_t)pd->hdr.icmp6);
5104                                 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6);
5105                                 m_copyback(m, off2, sizeof(struct icmp6_hdr),
5106                                     (caddr_t)&iih);
5107                         }
5108                         return (PF_PASS);
5109                         break;
5110                 }
5111 #endif /* INET6 */
5112                 default: {
5113                         key.af = pd2.af;
5114                         key.proto = pd2.proto;
5115                         PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af);
5116                         PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af);
5117                         key.port[0] = key.port[1] = 0;
5118
5119                         STATE_LOOKUP(kif, &key, direction, *state, pd);
5120
5121                         /* translate source/destination address, if necessary */
5122                         if ((*state)->key[PF_SK_WIRE] !=
5123                             (*state)->key[PF_SK_STACK]) {
5124                                 struct pf_state_key *nk =
5125                                     (*state)->key[pd->didx];
5126
5127                                 if (PF_ANEQ(pd2.src,
5128                                     &nk->addr[pd2.sidx], pd2.af))
5129                                         pf_change_icmp(pd2.src, NULL, daddr,
5130                                             &nk->addr[pd2.sidx], 0, NULL,
5131                                             pd2.ip_sum, icmpsum,
5132                                             pd->ip_sum, 0, pd2.af);
5133
5134                                 if (PF_ANEQ(pd2.dst,
5135                                     &nk->addr[pd2.didx], pd2.af))
5136                                         pf_change_icmp(pd2.dst, NULL, saddr,
5137                                             &nk->addr[pd2.didx], 0, NULL,
5138                                             pd2.ip_sum, icmpsum,
5139                                             pd->ip_sum, 0, pd2.af);
5140
5141                                 switch (pd2.af) {
5142 #ifdef INET
5143                                 case AF_INET:
5144                                         m_copyback(m, off, ICMP_MINLEN,
5145                                             (caddr_t)pd->hdr.icmp);
5146                                         m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2);
5147                                         break;
5148 #endif /* INET */
5149 #ifdef INET6
5150                                 case AF_INET6:
5151                                         m_copyback(m, off,
5152                                             sizeof(struct icmp6_hdr),
5153                                             (caddr_t )pd->hdr.icmp6);
5154                                         m_copyback(m, ipoff2, sizeof(h2_6),
5155                                             (caddr_t )&h2_6);
5156                                         break;
5157 #endif /* INET6 */
5158                                 }
5159                         }
5160                         return (PF_PASS);
5161                         break;
5162                 }
5163                 }
5164         }
5165 }
5166
5167 static int
5168 pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif,
5169     struct mbuf *m, struct pf_pdesc *pd)
5170 {
5171         struct pf_state_peer    *src, *dst;
5172         struct pf_state_key_cmp  key;
5173
5174         bzero(&key, sizeof(key));
5175         key.af = pd->af;
5176         key.proto = pd->proto;
5177         if (direction == PF_IN) {
5178                 PF_ACPY(&key.addr[0], pd->src, key.af);
5179                 PF_ACPY(&key.addr[1], pd->dst, key.af);
5180                 key.port[0] = key.port[1] = 0;
5181         } else {
5182                 PF_ACPY(&key.addr[1], pd->src, key.af);
5183                 PF_ACPY(&key.addr[0], pd->dst, key.af);
5184                 key.port[1] = key.port[0] = 0;
5185         }
5186
5187         STATE_LOOKUP(kif, &key, direction, *state, pd);
5188
5189         if (direction == (*state)->direction) {
5190                 src = &(*state)->src;
5191                 dst = &(*state)->dst;
5192         } else {
5193                 src = &(*state)->dst;
5194                 dst = &(*state)->src;
5195         }
5196
5197         /* update states */
5198         if (src->state < PFOTHERS_SINGLE)
5199                 src->state = PFOTHERS_SINGLE;
5200         if (dst->state == PFOTHERS_SINGLE)
5201                 dst->state = PFOTHERS_MULTIPLE;
5202
5203         /* update expire time */
5204         (*state)->expire = time_uptime;
5205         if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE)
5206                 (*state)->timeout = PFTM_OTHER_MULTIPLE;
5207         else
5208                 (*state)->timeout = PFTM_OTHER_SINGLE;
5209
5210         /* translate source/destination address, if necessary */
5211         if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) {
5212                 struct pf_state_key *nk = (*state)->key[pd->didx];
5213
5214                 KASSERT(nk, ("%s: nk is null", __func__));
5215                 KASSERT(pd, ("%s: pd is null", __func__));
5216                 KASSERT(pd->src, ("%s: pd->src is null", __func__));
5217                 KASSERT(pd->dst, ("%s: pd->dst is null", __func__));
5218                 switch (pd->af) {
5219 #ifdef INET
5220                 case AF_INET:
5221                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5222                                 pf_change_a(&pd->src->v4.s_addr,
5223                                     pd->ip_sum,
5224                                     nk->addr[pd->sidx].v4.s_addr,
5225                                     0);
5226
5227
5228                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5229                                 pf_change_a(&pd->dst->v4.s_addr,
5230                                     pd->ip_sum,
5231                                     nk->addr[pd->didx].v4.s_addr,
5232                                     0);
5233
5234                                 break;
5235 #endif /* INET */
5236 #ifdef INET6
5237                 case AF_INET6:
5238                         if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET))
5239                                 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af);
5240
5241                         if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET))
5242                                 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af);
5243 #endif /* INET6 */
5244                 }
5245         }
5246         return (PF_PASS);
5247 }
5248
5249 /*
5250  * ipoff and off are measured from the start of the mbuf chain.
5251  * h must be at "ipoff" on the mbuf chain.
5252  */
5253 void *
5254 pf_pull_hdr(struct mbuf *m, int off, void *p, int len,
5255     u_short *actionp, u_short *reasonp, sa_family_t af)
5256 {
5257         switch (af) {
5258 #ifdef INET
5259         case AF_INET: {
5260                 struct ip       *h = mtod(m, struct ip *);
5261                 u_int16_t        fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3;
5262
5263                 if (fragoff) {
5264                         if (fragoff >= len)
5265                                 ACTION_SET(actionp, PF_PASS);
5266                         else {
5267                                 ACTION_SET(actionp, PF_DROP);
5268                                 REASON_SET(reasonp, PFRES_FRAG);
5269                         }
5270                         return (NULL);
5271                 }
5272                 if (m->m_pkthdr.len < off + len ||
5273                     ntohs(h->ip_len) < off + len) {
5274                         ACTION_SET(actionp, PF_DROP);
5275                         REASON_SET(reasonp, PFRES_SHORT);
5276                         return (NULL);
5277                 }
5278                 break;
5279         }
5280 #endif /* INET */
5281 #ifdef INET6
5282         case AF_INET6: {
5283                 struct ip6_hdr  *h = mtod(m, struct ip6_hdr *);
5284
5285                 if (m->m_pkthdr.len < off + len ||
5286                     (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) <
5287                     (unsigned)(off + len)) {
5288                         ACTION_SET(actionp, PF_DROP);
5289                         REASON_SET(reasonp, PFRES_SHORT);
5290                         return (NULL);
5291                 }
5292                 break;
5293         }
5294 #endif /* INET6 */
5295         }
5296         m_copydata(m, off, len, p);
5297         return (p);
5298 }
5299
5300 #ifdef RADIX_MPATH
5301 static int
5302 pf_routable_oldmpath(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5303     int rtableid)
5304 {
5305         struct radix_node_head  *rnh;
5306         struct sockaddr_in      *dst;
5307         int                      ret = 1;
5308         int                      check_mpath;
5309 #ifdef INET6
5310         struct sockaddr_in6     *dst6;
5311         struct route_in6         ro;
5312 #else
5313         struct route             ro;
5314 #endif
5315         struct radix_node       *rn;
5316         struct rtentry          *rt;
5317         struct ifnet            *ifp;
5318
5319         check_mpath = 0;
5320         /* XXX: stick to table 0 for now */
5321         rnh = rt_tables_get_rnh(0, af);
5322         if (rnh != NULL && rn_mpath_capable(rnh))
5323                 check_mpath = 1;
5324         bzero(&ro, sizeof(ro));
5325         switch (af) {
5326         case AF_INET:
5327                 dst = satosin(&ro.ro_dst);
5328                 dst->sin_family = AF_INET;
5329                 dst->sin_len = sizeof(*dst);
5330                 dst->sin_addr = addr->v4;
5331                 break;
5332 #ifdef INET6
5333         case AF_INET6:
5334                 /*
5335                  * Skip check for addresses with embedded interface scope,
5336                  * as they would always match anyway.
5337                  */
5338                 if (IN6_IS_SCOPE_EMBED(&addr->v6))
5339                         goto out;
5340                 dst6 = (struct sockaddr_in6 *)&ro.ro_dst;
5341                 dst6->sin6_family = AF_INET6;
5342                 dst6->sin6_len = sizeof(*dst6);
5343                 dst6->sin6_addr = addr->v6;
5344                 break;
5345 #endif /* INET6 */
5346         default:
5347                 return (0);
5348         }
5349
5350         /* Skip checks for ipsec interfaces */
5351         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5352                 goto out;
5353
5354         switch (af) {
5355 #ifdef INET6
5356         case AF_INET6:
5357                 in6_rtalloc_ign(&ro, 0, rtableid);
5358                 break;
5359 #endif
5360 #ifdef INET
5361         case AF_INET:
5362                 in_rtalloc_ign((struct route *)&ro, 0, rtableid);
5363                 break;
5364 #endif
5365         }
5366
5367         if (ro.ro_rt != NULL) {
5368                 /* No interface given, this is a no-route check */
5369                 if (kif == NULL)
5370                         goto out;
5371
5372                 if (kif->pfik_ifp == NULL) {
5373                         ret = 0;
5374                         goto out;
5375                 }
5376
5377                 /* Perform uRPF check if passed input interface */
5378                 ret = 0;
5379                 rn = (struct radix_node *)ro.ro_rt;
5380                 do {
5381                         rt = (struct rtentry *)rn;
5382                         ifp = rt->rt_ifp;
5383
5384                         if (kif->pfik_ifp == ifp)
5385                                 ret = 1;
5386                         rn = rn_mpath_next(rn);
5387                 } while (check_mpath == 1 && rn != NULL && ret == 0);
5388         } else
5389                 ret = 0;
5390 out:
5391         if (ro.ro_rt != NULL)
5392                 RTFREE(ro.ro_rt);
5393         return (ret);
5394 }
5395 #endif
5396
5397 int
5398 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif,
5399     int rtableid)
5400 {
5401 #ifdef INET
5402         struct nhop4_basic      nh4;
5403 #endif
5404 #ifdef INET6
5405         struct nhop6_basic      nh6;
5406 #endif
5407         struct ifnet            *ifp;
5408 #ifdef RADIX_MPATH
5409         struct radix_node_head  *rnh;
5410
5411         /* XXX: stick to table 0 for now */
5412         rnh = rt_tables_get_rnh(0, af);
5413         if (rnh != NULL && rn_mpath_capable(rnh))
5414                 return (pf_routable_oldmpath(addr, af, kif, rtableid));
5415 #endif
5416         /*
5417          * Skip check for addresses with embedded interface scope,
5418          * as they would always match anyway.
5419          */
5420         if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6))
5421                 return (1);
5422
5423         if (af != AF_INET && af != AF_INET6)
5424                 return (0);
5425
5426         /* Skip checks for ipsec interfaces */
5427         if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC)
5428                 return (1);
5429
5430         ifp = NULL;
5431
5432         switch (af) {
5433 #ifdef INET6
5434         case AF_INET6:
5435                 if (fib6_lookup_nh_basic(rtableid, &addr->v6, 0, 0, 0, &nh6)!=0)
5436                         return (0);
5437                 ifp = nh6.nh_ifp;
5438                 break;
5439 #endif
5440 #ifdef INET
5441         case AF_INET:
5442                 if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) != 0)
5443                         return (0);
5444                 ifp = nh4.nh_ifp;
5445                 break;
5446 #endif
5447         }
5448
5449         /* No interface given, this is a no-route check */
5450         if (kif == NULL)
5451                 return (1);
5452
5453         if (kif->pfik_ifp == NULL)
5454                 return (0);
5455
5456         /* Perform uRPF check if passed input interface */
5457         if (kif->pfik_ifp == ifp)
5458                 return (1);
5459         return (0);
5460 }
5461
5462 #ifdef INET
5463 static void
5464 pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5465     struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp)
5466 {
5467         struct mbuf             *m0, *m1;
5468         struct sockaddr_in      dst;
5469         struct ip               *ip;
5470         struct ifnet            *ifp = NULL;
5471         struct pf_addr           naddr;
5472         struct pf_src_node      *sn = NULL;
5473         int                      error = 0;
5474         uint16_t                 ip_len, ip_off;
5475
5476         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5477         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5478             __func__));
5479
5480         if ((pd->pf_mtag == NULL &&
5481             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5482             pd->pf_mtag->routed++ > 3) {
5483                 m0 = *m;
5484                 *m = NULL;
5485                 goto bad_locked;
5486         }
5487
5488         if (r->rt == PF_DUPTO) {
5489                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5490                         if (s)
5491                                 PF_STATE_UNLOCK(s);
5492                         return;
5493                 }
5494         } else {
5495                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5496                         if (s)
5497                                 PF_STATE_UNLOCK(s);
5498                         return;
5499                 }
5500                 m0 = *m;
5501         }
5502
5503         ip = mtod(m0, struct ip *);
5504
5505         bzero(&dst, sizeof(dst));
5506         dst.sin_family = AF_INET;
5507         dst.sin_len = sizeof(dst);
5508         dst.sin_addr = ip->ip_dst;
5509
5510         if (TAILQ_EMPTY(&r->rpool.list)) {
5511                 DPFPRINTF(PF_DEBUG_URGENT,
5512                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5513                 goto bad_locked;
5514         }
5515         if (s == NULL) {
5516                 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src,
5517                     &naddr, NULL, &sn);
5518                 if (!PF_AZERO(&naddr, AF_INET))
5519                         dst.sin_addr.s_addr = naddr.v4.s_addr;
5520                 ifp = r->rpool.cur->kif ?
5521                     r->rpool.cur->kif->pfik_ifp : NULL;
5522         } else {
5523                 if (!PF_AZERO(&s->rt_addr, AF_INET))
5524                         dst.sin_addr.s_addr =
5525                             s->rt_addr.v4.s_addr;
5526                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5527                 PF_STATE_UNLOCK(s);
5528         }
5529         if (ifp == NULL)
5530                 goto bad;
5531
5532         if (oifp != ifp) {
5533                 if (pf_test(PF_OUT, 0, ifp, &m0, inp) != PF_PASS)
5534                         goto bad;
5535                 else if (m0 == NULL)
5536                         goto done;
5537                 if (m0->m_len < sizeof(struct ip)) {
5538                         DPFPRINTF(PF_DEBUG_URGENT,
5539                             ("%s: m0->m_len < sizeof(struct ip)\n", __func__));
5540                         goto bad;
5541                 }
5542                 ip = mtod(m0, struct ip *);
5543         }
5544
5545         if (ifp->if_flags & IFF_LOOPBACK)
5546                 m0->m_flags |= M_SKIP_FIREWALL;
5547
5548         ip_len = ntohs(ip->ip_len);
5549         ip_off = ntohs(ip->ip_off);
5550
5551         /* Copied from FreeBSD 10.0-CURRENT ip_output. */
5552         m0->m_pkthdr.csum_flags |= CSUM_IP;
5553         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
5554                 in_delayed_cksum(m0);
5555                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
5556         }
5557 #ifdef SCTP
5558         if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
5559                 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
5560                 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
5561         }
5562 #endif
5563
5564         /*
5565          * If small enough for interface, or the interface will take
5566          * care of the fragmentation for us, we can just send directly.
5567          */
5568         if (ip_len <= ifp->if_mtu ||
5569             (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
5570                 ip->ip_sum = 0;
5571                 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
5572                         ip->ip_sum = in_cksum(m0, ip->ip_hl << 2);
5573                         m0->m_pkthdr.csum_flags &= ~CSUM_IP;
5574                 }
5575                 m_clrprotoflags(m0);    /* Avoid confusing lower layers. */
5576                 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5577                 goto done;
5578         }
5579
5580         /* Balk when DF bit is set or the interface didn't support TSO. */
5581         if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) {
5582                 error = EMSGSIZE;
5583                 KMOD_IPSTAT_INC(ips_cantfrag);
5584                 if (r->rt != PF_DUPTO) {
5585                         icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0,
5586                             ifp->if_mtu);
5587                         goto done;
5588                 } else
5589                         goto bad;
5590         }
5591
5592         error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist);
5593         if (error)
5594                 goto bad;
5595
5596         for (; m0; m0 = m1) {
5597                 m1 = m0->m_nextpkt;
5598                 m0->m_nextpkt = NULL;
5599                 if (error == 0) {
5600                         m_clrprotoflags(m0);
5601                         error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL);
5602                 } else
5603                         m_freem(m0);
5604         }
5605
5606         if (error == 0)
5607                 KMOD_IPSTAT_INC(ips_fragmented);
5608
5609 done:
5610         if (r->rt != PF_DUPTO)
5611                 *m = NULL;
5612         return;
5613
5614 bad_locked:
5615         if (s)
5616                 PF_STATE_UNLOCK(s);
5617 bad:
5618         m_freem(m0);
5619         goto done;
5620 }
5621 #endif /* INET */
5622
5623 #ifdef INET6
5624 static void
5625 pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp,
5626     struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp)
5627 {
5628         struct mbuf             *m0;
5629         struct sockaddr_in6     dst;
5630         struct ip6_hdr          *ip6;
5631         struct ifnet            *ifp = NULL;
5632         struct pf_addr           naddr;
5633         struct pf_src_node      *sn = NULL;
5634
5635         KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__));
5636         KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction",
5637             __func__));
5638
5639         if ((pd->pf_mtag == NULL &&
5640             ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) ||
5641             pd->pf_mtag->routed++ > 3) {
5642                 m0 = *m;
5643                 *m = NULL;
5644                 goto bad_locked;
5645         }
5646
5647         if (r->rt == PF_DUPTO) {
5648                 if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) {
5649                         if (s)
5650                                 PF_STATE_UNLOCK(s);
5651                         return;
5652                 }
5653         } else {
5654                 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) {
5655                         if (s)
5656                                 PF_STATE_UNLOCK(s);
5657                         return;
5658                 }
5659                 m0 = *m;
5660         }
5661
5662         ip6 = mtod(m0, struct ip6_hdr *);
5663
5664         bzero(&dst, sizeof(dst));
5665         dst.sin6_family = AF_INET6;
5666         dst.sin6_len = sizeof(dst);
5667         dst.sin6_addr = ip6->ip6_dst;
5668
5669         if (TAILQ_EMPTY(&r->rpool.list)) {
5670                 DPFPRINTF(PF_DEBUG_URGENT,
5671                     ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__));
5672                 goto bad_locked;
5673         }
5674         if (s == NULL) {
5675                 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src,
5676                     &naddr, NULL, &sn);
5677                 if (!PF_AZERO(&naddr, AF_INET6))
5678                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5679                             &naddr, AF_INET6);
5680                 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL;
5681         } else {
5682                 if (!PF_AZERO(&s->rt_addr, AF_INET6))
5683                         PF_ACPY((struct pf_addr *)&dst.sin6_addr,
5684                             &s->rt_addr, AF_INET6);
5685                 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL;
5686         }
5687
5688         if (s)
5689                 PF_STATE_UNLOCK(s);
5690
5691         if (ifp == NULL)
5692                 goto bad;
5693
5694         if (oifp != ifp) {
5695                 if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp) != PF_PASS)
5696                         goto bad;
5697                 else if (m0 == NULL)
5698                         goto done;
5699                 if (m0->m_len < sizeof(struct ip6_hdr)) {
5700                         DPFPRINTF(PF_DEBUG_URGENT,
5701                             ("%s: m0->m_len < sizeof(struct ip6_hdr)\n",
5702                             __func__));
5703                         goto bad;
5704                 }
5705                 ip6 = mtod(m0, struct ip6_hdr *);
5706         }
5707
5708         if (ifp->if_flags & IFF_LOOPBACK)
5709                 m0->m_flags |= M_SKIP_FIREWALL;
5710
5711         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 &
5712             ~ifp->if_hwassist) {
5713                 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6);
5714                 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr));
5715                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
5716         }
5717
5718         /*
5719          * If the packet is too large for the outgoing interface,
5720          * send back an icmp6 error.
5721          */
5722         if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr))
5723                 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index);
5724         if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu)
5725                 nd6_output_ifp(ifp, ifp, m0, &dst, NULL);
5726         else {
5727                 in6_ifstat_inc(ifp, ifs6_in_toobig);
5728                 if (r->rt != PF_DUPTO)
5729                         icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu);
5730                 else
5731                         goto bad;
5732         }
5733
5734 done:
5735         if (r->rt != PF_DUPTO)
5736                 *m = NULL;
5737         return;
5738
5739 bad_locked:
5740         if (s)
5741                 PF_STATE_UNLOCK(s);
5742 bad:
5743         m_freem(m0);
5744         goto done;
5745 }
5746 #endif /* INET6 */
5747
5748 /*
5749  * FreeBSD supports cksum offloads for the following drivers.
5750  *  em(4), fxp(4), lge(4), ndis(4), nge(4), re(4), ti(4), txp(4), xl(4)
5751  *
5752  * CSUM_DATA_VALID | CSUM_PSEUDO_HDR :
5753  *  network driver performed cksum including pseudo header, need to verify
5754  *   csum_data
5755  * CSUM_DATA_VALID :
5756  *  network driver performed cksum, needs to additional pseudo header
5757  *  cksum computation with partial csum_data(i.e. lack of H/W support for
5758  *  pseudo header, for instance hme(4), sk(4) and possibly gem(4))
5759  *
5760  * After validating the cksum of packet, set both flag CSUM_DATA_VALID and
5761  * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper
5762  * TCP/UDP layer.
5763  * Also, set csum_data to 0xffff to force cksum validation.
5764  */
5765 static int
5766 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af)
5767 {
5768         u_int16_t sum = 0;
5769         int hw_assist = 0;
5770         struct ip *ip;
5771
5772         if (off < sizeof(struct ip) || len < sizeof(struct udphdr))
5773                 return (1);
5774         if (m->m_pkthdr.len < off + len)
5775                 return (1);
5776
5777         switch (p) {
5778         case IPPROTO_TCP:
5779                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5780                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5781                                 sum = m->m_pkthdr.csum_data;
5782                         } else {
5783                                 ip = mtod(m, struct ip *);
5784                                 sum = in_pseudo(ip->ip_src.s_addr,
5785                                 ip->ip_dst.s_addr, htonl((u_short)len +
5786                                 m->m_pkthdr.csum_data + IPPROTO_TCP));
5787                         }
5788                         sum ^= 0xffff;
5789                         ++hw_assist;
5790                 }
5791                 break;
5792         case IPPROTO_UDP:
5793                 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
5794                         if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) {
5795                                 sum = m->m_pkthdr.csum_data;
5796                         } else {
5797                                 ip = mtod(m, struct ip *);
5798                                 sum = in_pseudo(ip->ip_src.s_addr,
5799                                 ip->ip_dst.s_addr, htonl((u_short)len +
5800                                 m->m_pkthdr.csum_data + IPPROTO_UDP));
5801                         }
5802                         sum ^= 0xffff;
5803                         ++hw_assist;
5804                 }
5805                 break;
5806         case IPPROTO_ICMP:
5807 #ifdef INET6
5808         case IPPROTO_ICMPV6:
5809 #endif /* INET6 */
5810                 break;
5811         default:
5812                 return (1);
5813         }
5814
5815         if (!hw_assist) {
5816                 switch (af) {
5817                 case AF_INET:
5818                         if (p == IPPROTO_ICMP) {
5819                                 if (m->m_len < off)
5820                                         return (1);
5821                                 m->m_data += off;
5822                                 m->m_len -= off;
5823                                 sum = in_cksum(m, len);
5824                                 m->m_data -= off;
5825                                 m->m_len += off;
5826                         } else {
5827                                 if (m->m_len < sizeof(struct ip))
5828                                         return (1);
5829                                 sum = in4_cksum(m, p, off, len);
5830                         }
5831                         break;
5832 #ifdef INET6
5833                 case AF_INET6:
5834                         if (m->m_len < sizeof(struct ip6_hdr))
5835                                 return (1);
5836                         sum = in6_cksum(m, p, off, len);
5837                         break;
5838 #endif /* INET6 */
5839                 default:
5840                         return (1);
5841                 }
5842         }
5843         if (sum) {
5844                 switch (p) {
5845                 case IPPROTO_TCP:
5846                     {
5847                         KMOD_TCPSTAT_INC(tcps_rcvbadsum);
5848                         break;
5849                     }
5850                 case IPPROTO_UDP:
5851                     {
5852                         KMOD_UDPSTAT_INC(udps_badsum);
5853                         break;
5854                     }
5855 #ifdef INET
5856                 case IPPROTO_ICMP:
5857                     {
5858                         KMOD_ICMPSTAT_INC(icps_checksum);
5859                         break;
5860                     }
5861 #endif
5862 #ifdef INET6
5863                 case IPPROTO_ICMPV6:
5864                     {
5865                         KMOD_ICMP6STAT_INC(icp6s_checksum);
5866                         break;
5867                     }
5868 #endif /* INET6 */
5869                 }
5870                 return (1);
5871         } else {
5872                 if (p == IPPROTO_TCP || p == IPPROTO_UDP) {
5873                         m->m_pkthdr.csum_flags |=
5874                             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
5875                         m->m_pkthdr.csum_data = 0xffff;
5876                 }
5877         }
5878         return (0);
5879 }
5880
5881
5882 #ifdef INET
5883 int
5884 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
5885 {
5886         struct pfi_kif          *kif;
5887         u_short                  action, reason = 0, log = 0;
5888         struct mbuf             *m = *m0;
5889         struct ip               *h = NULL;
5890         struct m_tag            *ipfwtag;
5891         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
5892         struct pf_state         *s = NULL;
5893         struct pf_ruleset       *ruleset = NULL;
5894         struct pf_pdesc          pd;
5895         int                      off, dirndx, pqid = 0;
5896
5897         PF_RULES_RLOCK_TRACKER;
5898
5899         M_ASSERTPKTHDR(m);
5900
5901         if (!V_pf_status.running)
5902                 return (PF_PASS);
5903
5904         memset(&pd, 0, sizeof(pd));
5905
5906         kif = (struct pfi_kif *)ifp->if_pf_kif;
5907
5908         if (kif == NULL) {
5909                 DPFPRINTF(PF_DEBUG_URGENT,
5910                     ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname));
5911                 return (PF_DROP);
5912         }
5913         if (kif->pfik_flags & PFI_IFLAG_SKIP)
5914                 return (PF_PASS);
5915
5916         if (m->m_flags & M_SKIP_FIREWALL)
5917                 return (PF_PASS);
5918
5919         pd.pf_mtag = pf_find_mtag(m);
5920
5921         PF_RULES_RLOCK();
5922
5923         if (ip_divert_ptr != NULL &&
5924             ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) {
5925                 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1);
5926                 if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) {
5927                         if (pd.pf_mtag == NULL &&
5928                             ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
5929                                 action = PF_DROP;
5930                                 goto done;
5931                         }
5932                         pd.pf_mtag->flags |= PF_PACKET_LOOPED;
5933                         m_tag_delete(m, ipfwtag);
5934                 }
5935                 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) {
5936                         m->m_flags |= M_FASTFWD_OURS;
5937                         pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT;
5938                 }
5939         } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) {
5940                 /* We do IP header normalization and packet reassembly here */
5941                 action = PF_DROP;
5942                 goto done;
5943         }
5944         m = *m0;        /* pf_normalize messes with m0 */
5945         h = mtod(m, struct ip *);
5946
5947         off = h->ip_hl << 2;
5948         if (off < (int)sizeof(struct ip)) {
5949                 action = PF_DROP;
5950                 REASON_SET(&reason, PFRES_SHORT);
5951                 log = 1;
5952                 goto done;
5953         }
5954
5955         pd.src = (struct pf_addr *)&h->ip_src;
5956         pd.dst = (struct pf_addr *)&h->ip_dst;
5957         pd.sport = pd.dport = NULL;
5958         pd.ip_sum = &h->ip_sum;
5959         pd.proto_sum = NULL;
5960         pd.proto = h->ip_p;
5961         pd.dir = dir;
5962         pd.sidx = (dir == PF_IN) ? 0 : 1;
5963         pd.didx = (dir == PF_IN) ? 1 : 0;
5964         pd.af = AF_INET;
5965         pd.tos = h->ip_tos & ~IPTOS_ECN_MASK;
5966         pd.tot_len = ntohs(h->ip_len);
5967
5968         /* handle fragments that didn't get reassembled by normalization */
5969         if (h->ip_off & htons(IP_MF | IP_OFFMASK)) {
5970                 action = pf_test_fragment(&r, dir, kif, m, h,
5971                     &pd, &a, &ruleset);
5972                 goto done;
5973         }
5974
5975         switch (h->ip_p) {
5976
5977         case IPPROTO_TCP: {
5978                 struct tcphdr   th;
5979
5980                 pd.hdr.tcp = &th;
5981                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
5982                     &action, &reason, AF_INET)) {
5983                         log = action != PF_PASS;
5984                         goto done;
5985                 }
5986                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
5987                 if ((th.th_flags & TH_ACK) && pd.p_len == 0)
5988                         pqid = 1;
5989                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
5990                 if (action == PF_DROP)
5991                         goto done;
5992                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
5993                     &reason);
5994                 if (action == PF_PASS) {
5995                         if (pfsync_update_state_ptr != NULL)
5996                                 pfsync_update_state_ptr(s);
5997                         r = s->rule.ptr;
5998                         a = s->anchor.ptr;
5999                         log = s->log;
6000                 } else if (s == NULL)
6001                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6002                             &a, &ruleset, inp);
6003                 break;
6004         }
6005
6006         case IPPROTO_UDP: {
6007                 struct udphdr   uh;
6008
6009                 pd.hdr.udp = &uh;
6010                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6011                     &action, &reason, AF_INET)) {
6012                         log = action != PF_PASS;
6013                         goto done;
6014                 }
6015                 if (uh.uh_dport == 0 ||
6016                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6017                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6018                         action = PF_DROP;
6019                         REASON_SET(&reason, PFRES_SHORT);
6020                         goto done;
6021                 }
6022                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6023                 if (action == PF_PASS) {
6024                         if (pfsync_update_state_ptr != NULL)
6025                                 pfsync_update_state_ptr(s);
6026                         r = s->rule.ptr;
6027                         a = s->anchor.ptr;
6028                         log = s->log;
6029                 } else if (s == NULL)
6030                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6031                             &a, &ruleset, inp);
6032                 break;
6033         }
6034
6035         case IPPROTO_ICMP: {
6036                 struct icmp     ih;
6037
6038                 pd.hdr.icmp = &ih;
6039                 if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN,
6040                     &action, &reason, AF_INET)) {
6041                         log = action != PF_PASS;
6042                         goto done;
6043                 }
6044                 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd,
6045                     &reason);
6046                 if (action == PF_PASS) {
6047                         if (pfsync_update_state_ptr != NULL)
6048                                 pfsync_update_state_ptr(s);
6049                         r = s->rule.ptr;
6050                         a = s->anchor.ptr;
6051                         log = s->log;
6052                 } else if (s == NULL)
6053                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6054                             &a, &ruleset, inp);
6055                 break;
6056         }
6057
6058 #ifdef INET6
6059         case IPPROTO_ICMPV6: {
6060                 action = PF_DROP;
6061                 DPFPRINTF(PF_DEBUG_MISC,
6062                     ("pf: dropping IPv4 packet with ICMPv6 payload\n"));
6063                 goto done;
6064         }
6065 #endif
6066
6067         default:
6068                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6069                 if (action == PF_PASS) {
6070                         if (pfsync_update_state_ptr != NULL)
6071                                 pfsync_update_state_ptr(s);
6072                         r = s->rule.ptr;
6073                         a = s->anchor.ptr;
6074                         log = s->log;
6075                 } else if (s == NULL)
6076                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6077                             &a, &ruleset, inp);
6078                 break;
6079         }
6080
6081 done:
6082         PF_RULES_RUNLOCK();
6083         if (action == PF_PASS && h->ip_hl > 5 &&
6084             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6085                 action = PF_DROP;
6086                 REASON_SET(&reason, PFRES_IPOPTIONS);
6087                 log = r->log;
6088                 DPFPRINTF(PF_DEBUG_MISC,
6089                     ("pf: dropping packet with ip options\n"));
6090         }
6091
6092         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6093                 action = PF_DROP;
6094                 REASON_SET(&reason, PFRES_MEMORY);
6095         }
6096         if (r->rtableid >= 0)
6097                 M_SETFIB(m, r->rtableid);
6098
6099         if (r->scrub_flags & PFSTATE_SETPRIO) {
6100                 if (pd.tos & IPTOS_LOWDELAY)
6101                         pqid = 1;
6102                 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6103                         action = PF_DROP;
6104                         REASON_SET(&reason, PFRES_MEMORY);
6105                         log = 1;
6106                         DPFPRINTF(PF_DEBUG_MISC,
6107                             ("pf: failed to allocate 802.1q mtag\n"));
6108                 }
6109         }
6110
6111 #ifdef ALTQ
6112         if (action == PF_PASS && r->qid) {
6113                 if (pd.pf_mtag == NULL &&
6114                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6115                         action = PF_DROP;
6116                         REASON_SET(&reason, PFRES_MEMORY);
6117                 } else {
6118                         if (s != NULL)
6119                                 pd.pf_mtag->qid_hash = pf_state_hash(s);
6120                         if (pqid || (pd.tos & IPTOS_LOWDELAY))
6121                                 pd.pf_mtag->qid = r->pqid;
6122                         else
6123                                 pd.pf_mtag->qid = r->qid;
6124                         /* Add hints for ecn. */
6125                         pd.pf_mtag->hdr = h;
6126                 }
6127
6128         }
6129 #endif /* ALTQ */
6130
6131         /*
6132          * connections redirected to loopback should not match sockets
6133          * bound specifically to loopback due to security implications,
6134          * see tcp_input() and in_pcblookup_listen().
6135          */
6136         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6137             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6138             (s->nat_rule.ptr->action == PF_RDR ||
6139             s->nat_rule.ptr->action == PF_BINAT) &&
6140             (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET)
6141                 m->m_flags |= M_SKIP_FIREWALL;
6142
6143         if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL &&
6144             !PACKET_LOOPED(&pd)) {
6145
6146                 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0,
6147                     sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO);
6148                 if (ipfwtag != NULL) {
6149                         ((struct ipfw_rule_ref *)(ipfwtag+1))->info =
6150                             ntohs(r->divert.port);
6151                         ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir;
6152
6153                         if (s)
6154                                 PF_STATE_UNLOCK(s);
6155
6156                         m_tag_prepend(m, ipfwtag);
6157                         if (m->m_flags & M_FASTFWD_OURS) {
6158                                 if (pd.pf_mtag == NULL &&
6159                                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6160                                         action = PF_DROP;
6161                                         REASON_SET(&reason, PFRES_MEMORY);
6162                                         log = 1;
6163                                         DPFPRINTF(PF_DEBUG_MISC,
6164                                             ("pf: failed to allocate tag\n"));
6165                                 } else {
6166                                         pd.pf_mtag->flags |=
6167                                             PF_FASTFWD_OURS_PRESENT;
6168                                         m->m_flags &= ~M_FASTFWD_OURS;
6169                                 }
6170                         }
6171                         ip_divert_ptr(*m0, dir ==  PF_IN ? DIR_IN : DIR_OUT);
6172                         *m0 = NULL;
6173
6174                         return (action);
6175                 } else {
6176                         /* XXX: ipfw has the same behaviour! */
6177                         action = PF_DROP;
6178                         REASON_SET(&reason, PFRES_MEMORY);
6179                         log = 1;
6180                         DPFPRINTF(PF_DEBUG_MISC,
6181                             ("pf: failed to allocate divert tag\n"));
6182                 }
6183         }
6184
6185         if (log) {
6186                 struct pf_rule *lr;
6187
6188                 if (s != NULL && s->nat_rule.ptr != NULL &&
6189                     s->nat_rule.ptr->log & PF_LOG_ALL)
6190                         lr = s->nat_rule.ptr;
6191                 else
6192                         lr = r;
6193                 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd,
6194                     (s == NULL));
6195         }
6196
6197         kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6198         kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++;
6199
6200         if (action == PF_PASS || r->action == PF_DROP) {
6201                 dirndx = (dir == PF_OUT);
6202                 r->packets[dirndx]++;
6203                 r->bytes[dirndx] += pd.tot_len;
6204                 if (a != NULL) {
6205                         a->packets[dirndx]++;
6206                         a->bytes[dirndx] += pd.tot_len;
6207                 }
6208                 if (s != NULL) {
6209                         if (s->nat_rule.ptr != NULL) {
6210                                 s->nat_rule.ptr->packets[dirndx]++;
6211                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6212                         }
6213                         if (s->src_node != NULL) {
6214                                 s->src_node->packets[dirndx]++;
6215                                 s->src_node->bytes[dirndx] += pd.tot_len;
6216                         }
6217                         if (s->nat_src_node != NULL) {
6218                                 s->nat_src_node->packets[dirndx]++;
6219                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6220                         }
6221                         dirndx = (dir == s->direction) ? 0 : 1;
6222                         s->packets[dirndx]++;
6223                         s->bytes[dirndx] += pd.tot_len;
6224                 }
6225                 tr = r;
6226                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6227                 if (nr != NULL && r == &V_pf_default_rule)
6228                         tr = nr;
6229                 if (tr->src.addr.type == PF_ADDR_TABLE)
6230                         pfr_update_stats(tr->src.addr.p.tbl,
6231                             (s == NULL) ? pd.src :
6232                             &s->key[(s->direction == PF_IN)]->
6233                                 addr[(s->direction == PF_OUT)],
6234                             pd.af, pd.tot_len, dir == PF_OUT,
6235                             r->action == PF_PASS, tr->src.neg);
6236                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6237                         pfr_update_stats(tr->dst.addr.p.tbl,
6238                             (s == NULL) ? pd.dst :
6239                             &s->key[(s->direction == PF_IN)]->
6240                                 addr[(s->direction == PF_IN)],
6241                             pd.af, pd.tot_len, dir == PF_OUT,
6242                             r->action == PF_PASS, tr->dst.neg);
6243         }
6244
6245         switch (action) {
6246         case PF_SYNPROXY_DROP:
6247                 m_freem(*m0);
6248         case PF_DEFER:
6249                 *m0 = NULL;
6250                 action = PF_PASS;
6251                 break;
6252         case PF_DROP:
6253                 m_freem(*m0);
6254                 *m0 = NULL;
6255                 break;
6256         default:
6257                 /* pf_route() returns unlocked. */
6258                 if (r->rt) {
6259                         pf_route(m0, r, dir, kif->pfik_ifp, s, &pd, inp);
6260                         return (action);
6261                 }
6262                 break;
6263         }
6264         if (s)
6265                 PF_STATE_UNLOCK(s);
6266
6267         return (action);
6268 }
6269 #endif /* INET */
6270
6271 #ifdef INET6
6272 int
6273 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp)
6274 {
6275         struct pfi_kif          *kif;
6276         u_short                  action, reason = 0, log = 0;
6277         struct mbuf             *m = *m0, *n = NULL;
6278         struct m_tag            *mtag;
6279         struct ip6_hdr          *h = NULL;
6280         struct pf_rule          *a = NULL, *r = &V_pf_default_rule, *tr, *nr;
6281         struct pf_state         *s = NULL;
6282         struct pf_ruleset       *ruleset = NULL;
6283         struct pf_pdesc          pd;
6284         int                      off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0;
6285
6286         PF_RULES_RLOCK_TRACKER;
6287         M_ASSERTPKTHDR(m);
6288
6289         if (!V_pf_status.running)
6290                 return (PF_PASS);
6291
6292         memset(&pd, 0, sizeof(pd));
6293         pd.pf_mtag = pf_find_mtag(m);
6294
6295         if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED)
6296                 return (PF_PASS);
6297
6298         kif = (struct pfi_kif *)ifp->if_pf_kif;
6299         if (kif == NULL) {
6300                 DPFPRINTF(PF_DEBUG_URGENT,
6301                     ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname));
6302                 return (PF_DROP);
6303         }
6304         if (kif->pfik_flags & PFI_IFLAG_SKIP)
6305                 return (PF_PASS);
6306
6307         if (m->m_flags & M_SKIP_FIREWALL)
6308                 return (PF_PASS);
6309
6310         PF_RULES_RLOCK();
6311
6312         /* We do IP header normalization and packet reassembly here */
6313         if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) {
6314                 action = PF_DROP;
6315                 goto done;
6316         }
6317         m = *m0;        /* pf_normalize messes with m0 */
6318         h = mtod(m, struct ip6_hdr *);
6319
6320 #if 1
6321         /*
6322          * we do not support jumbogram yet.  if we keep going, zero ip6_plen
6323          * will do something bad, so drop the packet for now.
6324          */
6325         if (htons(h->ip6_plen) == 0) {
6326                 action = PF_DROP;
6327                 REASON_SET(&reason, PFRES_NORM);        /*XXX*/
6328                 goto done;
6329         }
6330 #endif
6331
6332         pd.src = (struct pf_addr *)&h->ip6_src;
6333         pd.dst = (struct pf_addr *)&h->ip6_dst;
6334         pd.sport = pd.dport = NULL;
6335         pd.ip_sum = NULL;
6336         pd.proto_sum = NULL;
6337         pd.dir = dir;
6338         pd.sidx = (dir == PF_IN) ? 0 : 1;
6339         pd.didx = (dir == PF_IN) ? 1 : 0;
6340         pd.af = AF_INET6;
6341         pd.tos = 0;
6342         pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr);
6343
6344         off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr);
6345         pd.proto = h->ip6_nxt;
6346         do {
6347                 switch (pd.proto) {
6348                 case IPPROTO_FRAGMENT:
6349                         action = pf_test_fragment(&r, dir, kif, m, h,
6350                             &pd, &a, &ruleset);
6351                         if (action == PF_DROP)
6352                                 REASON_SET(&reason, PFRES_FRAG);
6353                         goto done;
6354                 case IPPROTO_ROUTING: {
6355                         struct ip6_rthdr rthdr;
6356
6357                         if (rh_cnt++) {
6358                                 DPFPRINTF(PF_DEBUG_MISC,
6359                                     ("pf: IPv6 more than one rthdr\n"));
6360                                 action = PF_DROP;
6361                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6362                                 log = 1;
6363                                 goto done;
6364                         }
6365                         if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL,
6366                             &reason, pd.af)) {
6367                                 DPFPRINTF(PF_DEBUG_MISC,
6368                                     ("pf: IPv6 short rthdr\n"));
6369                                 action = PF_DROP;
6370                                 REASON_SET(&reason, PFRES_SHORT);
6371                                 log = 1;
6372                                 goto done;
6373                         }
6374                         if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) {
6375                                 DPFPRINTF(PF_DEBUG_MISC,
6376                                     ("pf: IPv6 rthdr0\n"));
6377                                 action = PF_DROP;
6378                                 REASON_SET(&reason, PFRES_IPOPTIONS);
6379                                 log = 1;
6380                                 goto done;
6381                         }
6382                         /* FALLTHROUGH */
6383                 }
6384                 case IPPROTO_AH:
6385                 case IPPROTO_HOPOPTS:
6386                 case IPPROTO_DSTOPTS: {
6387                         /* get next header and header length */
6388                         struct ip6_ext  opt6;
6389
6390                         if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6),
6391                             NULL, &reason, pd.af)) {
6392                                 DPFPRINTF(PF_DEBUG_MISC,
6393                                     ("pf: IPv6 short opt\n"));
6394                                 action = PF_DROP;
6395                                 log = 1;
6396                                 goto done;
6397                         }
6398                         if (pd.proto == IPPROTO_AH)
6399                                 off += (opt6.ip6e_len + 2) * 4;
6400                         else
6401                                 off += (opt6.ip6e_len + 1) * 8;
6402                         pd.proto = opt6.ip6e_nxt;
6403                         /* goto the next header */
6404                         break;
6405                 }
6406                 default:
6407                         terminal++;
6408                         break;
6409                 }
6410         } while (!terminal);
6411
6412         /* if there's no routing header, use unmodified mbuf for checksumming */
6413         if (!n)
6414                 n = m;
6415
6416         switch (pd.proto) {
6417
6418         case IPPROTO_TCP: {
6419                 struct tcphdr   th;
6420
6421                 pd.hdr.tcp = &th;
6422                 if (!pf_pull_hdr(m, off, &th, sizeof(th),
6423                     &action, &reason, AF_INET6)) {
6424                         log = action != PF_PASS;
6425                         goto done;
6426                 }
6427                 pd.p_len = pd.tot_len - off - (th.th_off << 2);
6428                 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd);
6429                 if (action == PF_DROP)
6430                         goto done;
6431                 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd,
6432                     &reason);
6433                 if (action == PF_PASS) {
6434                         if (pfsync_update_state_ptr != NULL)
6435                                 pfsync_update_state_ptr(s);
6436                         r = s->rule.ptr;
6437                         a = s->anchor.ptr;
6438                         log = s->log;
6439                 } else if (s == NULL)
6440                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6441                             &a, &ruleset, inp);
6442                 break;
6443         }
6444
6445         case IPPROTO_UDP: {
6446                 struct udphdr   uh;
6447
6448                 pd.hdr.udp = &uh;
6449                 if (!pf_pull_hdr(m, off, &uh, sizeof(uh),
6450                     &action, &reason, AF_INET6)) {
6451                         log = action != PF_PASS;
6452                         goto done;
6453                 }
6454                 if (uh.uh_dport == 0 ||
6455                     ntohs(uh.uh_ulen) > m->m_pkthdr.len - off ||
6456                     ntohs(uh.uh_ulen) < sizeof(struct udphdr)) {
6457                         action = PF_DROP;
6458                         REASON_SET(&reason, PFRES_SHORT);
6459                         goto done;
6460                 }
6461                 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd);
6462                 if (action == PF_PASS) {
6463                         if (pfsync_update_state_ptr != NULL)
6464                                 pfsync_update_state_ptr(s);
6465                         r = s->rule.ptr;
6466                         a = s->anchor.ptr;
6467                         log = s->log;
6468                 } else if (s == NULL)
6469                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6470                             &a, &ruleset, inp);
6471                 break;
6472         }
6473
6474         case IPPROTO_ICMP: {
6475                 action = PF_DROP;
6476                 DPFPRINTF(PF_DEBUG_MISC,
6477                     ("pf: dropping IPv6 packet with ICMPv4 payload\n"));
6478                 goto done;
6479         }
6480
6481         case IPPROTO_ICMPV6: {
6482                 struct icmp6_hdr        ih;
6483
6484                 pd.hdr.icmp6 = &ih;
6485                 if (!pf_pull_hdr(m, off, &ih, sizeof(ih),
6486                     &action, &reason, AF_INET6)) {
6487                         log = action != PF_PASS;
6488                         goto done;
6489                 }
6490                 action = pf_test_state_icmp(&s, dir, kif,
6491                     m, off, h, &pd, &reason);
6492                 if (action == PF_PASS) {
6493                         if (pfsync_update_state_ptr != NULL)
6494                                 pfsync_update_state_ptr(s);
6495                         r = s->rule.ptr;
6496                         a = s->anchor.ptr;
6497                         log = s->log;
6498                 } else if (s == NULL)
6499                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6500                             &a, &ruleset, inp);
6501                 break;
6502         }
6503
6504         default:
6505                 action = pf_test_state_other(&s, dir, kif, m, &pd);
6506                 if (action == PF_PASS) {
6507                         if (pfsync_update_state_ptr != NULL)
6508                                 pfsync_update_state_ptr(s);
6509                         r = s->rule.ptr;
6510                         a = s->anchor.ptr;
6511                         log = s->log;
6512                 } else if (s == NULL)
6513                         action = pf_test_rule(&r, &s, dir, kif, m, off, &pd,
6514                             &a, &ruleset, inp);
6515                 break;
6516         }
6517
6518 done:
6519         PF_RULES_RUNLOCK();
6520         if (n != m) {
6521                 m_freem(n);
6522                 n = NULL;
6523         }
6524
6525         /* handle dangerous IPv6 extension headers. */
6526         if (action == PF_PASS && rh_cnt &&
6527             !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) {
6528                 action = PF_DROP;
6529                 REASON_SET(&reason, PFRES_IPOPTIONS);
6530                 log = r->log;
6531                 DPFPRINTF(PF_DEBUG_MISC,
6532                     ("pf: dropping packet with dangerous v6 headers\n"));
6533         }
6534
6535         if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) {
6536                 action = PF_DROP;
6537                 REASON_SET(&reason, PFRES_MEMORY);
6538         }
6539         if (r->rtableid >= 0)
6540                 M_SETFIB(m, r->rtableid);
6541
6542         if (r->scrub_flags & PFSTATE_SETPRIO) {
6543                 if (pd.tos & IPTOS_LOWDELAY)
6544                         pqid = 1;
6545                 if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) {
6546                         action = PF_DROP;
6547                         REASON_SET(&reason, PFRES_MEMORY);
6548                         log = 1;
6549                         DPFPRINTF(PF_DEBUG_MISC,
6550                             ("pf: failed to allocate 802.1q mtag\n"));
6551                 }
6552         }
6553
6554 #ifdef ALTQ
6555         if (action == PF_PASS && r->qid) {
6556                 if (pd.pf_mtag == NULL &&
6557                     ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) {
6558                         action = PF_DROP;
6559                         REASON_SET(&reason, PFRES_MEMORY);
6560                 } else {
6561                         if (s != NULL)
6562                                 pd.pf_mtag->qid_hash = pf_state_hash(s);
6563                         if (pd.tos & IPTOS_LOWDELAY)
6564                                 pd.pf_mtag->qid = r->pqid;
6565                         else
6566                                 pd.pf_mtag->qid = r->qid;
6567                         /* Add hints for ecn. */
6568                         pd.pf_mtag->hdr = h;
6569                 }
6570         }
6571 #endif /* ALTQ */
6572
6573         if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP ||
6574             pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL &&
6575             (s->nat_rule.ptr->action == PF_RDR ||
6576             s->nat_rule.ptr->action == PF_BINAT) &&
6577             IN6_IS_ADDR_LOOPBACK(&pd.dst->v6))
6578                 m->m_flags |= M_SKIP_FIREWALL;
6579
6580         /* XXX: Anybody working on it?! */
6581         if (r->divert.port)
6582                 printf("pf: divert(9) is not supported for IPv6\n");
6583
6584         if (log) {
6585                 struct pf_rule *lr;
6586
6587                 if (s != NULL && s->nat_rule.ptr != NULL &&
6588                     s->nat_rule.ptr->log & PF_LOG_ALL)
6589                         lr = s->nat_rule.ptr;
6590                 else
6591                         lr = r;
6592                 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset,
6593                     &pd, (s == NULL));
6594         }
6595
6596         kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len;
6597         kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++;
6598
6599         if (action == PF_PASS || r->action == PF_DROP) {
6600                 dirndx = (dir == PF_OUT);
6601                 r->packets[dirndx]++;
6602                 r->bytes[dirndx] += pd.tot_len;
6603                 if (a != NULL) {
6604                         a->packets[dirndx]++;
6605                         a->bytes[dirndx] += pd.tot_len;
6606                 }
6607                 if (s != NULL) {
6608                         if (s->nat_rule.ptr != NULL) {
6609                                 s->nat_rule.ptr->packets[dirndx]++;
6610                                 s->nat_rule.ptr->bytes[dirndx] += pd.tot_len;
6611                         }
6612                         if (s->src_node != NULL) {
6613                                 s->src_node->packets[dirndx]++;
6614                                 s->src_node->bytes[dirndx] += pd.tot_len;
6615                         }
6616                         if (s->nat_src_node != NULL) {
6617                                 s->nat_src_node->packets[dirndx]++;
6618                                 s->nat_src_node->bytes[dirndx] += pd.tot_len;
6619                         }
6620                         dirndx = (dir == s->direction) ? 0 : 1;
6621                         s->packets[dirndx]++;
6622                         s->bytes[dirndx] += pd.tot_len;
6623                 }
6624                 tr = r;
6625                 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule;
6626                 if (nr != NULL && r == &V_pf_default_rule)
6627                         tr = nr;
6628                 if (tr->src.addr.type == PF_ADDR_TABLE)
6629                         pfr_update_stats(tr->src.addr.p.tbl,
6630                             (s == NULL) ? pd.src :
6631                             &s->key[(s->direction == PF_IN)]->addr[0],
6632                             pd.af, pd.tot_len, dir == PF_OUT,
6633                             r->action == PF_PASS, tr->src.neg);
6634                 if (tr->dst.addr.type == PF_ADDR_TABLE)
6635                         pfr_update_stats(tr->dst.addr.p.tbl,
6636                             (s == NULL) ? pd.dst :
6637                             &s->key[(s->direction == PF_IN)]->addr[1],
6638                             pd.af, pd.tot_len, dir == PF_OUT,
6639                             r->action == PF_PASS, tr->dst.neg);
6640         }
6641
6642         switch (action) {
6643         case PF_SYNPROXY_DROP:
6644                 m_freem(*m0);
6645         case PF_DEFER:
6646                 *m0 = NULL;
6647                 action = PF_PASS;
6648                 break;
6649         case PF_DROP:
6650                 m_freem(*m0);
6651                 *m0 = NULL;
6652                 break;
6653         default:
6654                 /* pf_route6() returns unlocked. */
6655                 if (r->rt) {
6656                         pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd, inp);
6657                         return (action);
6658                 }
6659                 break;
6660         }
6661
6662         if (s)
6663                 PF_STATE_UNLOCK(s);
6664
6665         /* If reassembled packet passed, create new fragments. */
6666         if (action == PF_PASS && *m0 && (pflags & PFIL_FWD) &&
6667             (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL)
6668                 action = pf_refragment6(ifp, m0, mtag);
6669
6670         return (action);
6671 }
6672 #endif /* INET6 */